Real Analysis Via Sequences An Series. Charles Little, Teo Kee, Bruce Van Brunt

Undergraduate Texts in Mathematics
CharlesH.C.Little
KeeL.Teo
BrucevanBrunt
Real Analysis
via Sequences
and Series
Series Editors:
Sheldon Axler
San Francisco State University, San Francisco, CA, USA
Kenneth Ribet
University of California, Berkeley, CA, USA
Advisory Board:
Colin Adams, Williams College

David A. Cox, Amherst College
Pamela Gorkin, Bucknell University
Roger E. Howe, Yale University
Michael Orrison, Harvey Mudd College
Jill Pipher, Brown University
Fadil Santosa, University of Minnesota
Undergraduate Texts in Mathematics are generally aimed at third- and fourth-

year undergraduate mathematics students at North American universities. These
texts strive to provide students and teachers with new perspectives and novel
approaches. The books include motivation that guides the reader to an appreciation
of interrelations among different aspects of the subject. They feature examples that
illustrate key concepts as well as exercises that strengthen understanding.
More information about this series at http://www.springer.com/series/666

Charles H.C. Little Kee L. Teo Bruce van Brunt
Real Analysis via Sequences

and Series
123
Charles H.C. Little Kee L. Teo
Research Fellow and former Research Fellow and former
Professor of Mathematics Associate Professor of Mathenatics
Institute of Fundamental Sciences Institute of Fundamental Sciences
Massey University Massey University
Palmerston North, New Zealand Palmerston North, New Zealand
Bruce van Brunt

Associate Professor of Mathematics
Institute of Fundamental Sciences
Massey University
Palmerston North, New Zealand
ISSN 0172-6056 ISSN 2197-5604 (electronic)

ISBN 978-1-4939-2650-3 ISBN 978-1-4939-2651-0 (eBook)
DOI 10.1007/978-1-4939-2651-0
Library of Congress Control Number: 2015935731
Mathematics Subject Classification (2010): 26-01
Springer New York Heidelberg Dordrecht London

Springer Science+Business Media New York 2015
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.
Printed on acid-free paper
Springer Science+Business Media LLC New York is part of Springer Science+Business Media (www.
springer.com)
Preface
This book is a text on real analysis for students with a basic knowledge of calculus
of a single variable. There are many fine works on analysis, and one must ask
what advantages a new book brings. This one contains the standard material for
a first course in analysis, but our treatment differs from many other accounts in
that concepts such as continuity, differentiation, and integration are approached via
sequences. The main analytical concept is thus the convergence of a sequence, and
this idea is extended to define infinite series and limits of functions. This approach
not only has the merit of simplicity but also places the student in a position to
appreciate and understand more sophisticated concepts such as completeness that
play a central part in more advanced fields such as functional analysis.
The theory of sequences and series forms the backbone of this book. Much of
the material in the book is devoted to this theory and, in contrast to many other
texts, infinite series are treated early. The appearance of series in Chap. 3 has the
advantages that it provides many straightforward applications of the results for
sequences given in Chap. 2 and permits the introduction of the elementary tran-
scendental functions as infinite series. The disadvantage is that certain convergence
tests such as the integral test must be postponed until the improper integral is defined
in Chap. 7. The Cauchy condensation test is used in Chap. 3 to tackle convergence
problems where the integral test is normally applied. Although much of the material
in Chap. 2 is standard, there are some unusual features such as the treatment of
harmonic, geometric, and arithmetic means and the sequential definition of the
exponential function. In Chap. 3 we present results, such as the KummerJensen
test, Dirichlets test, and Riemanns theorem on the rearrangement of series, that are
often postponed or not treated in a first course in analysis.
Limits of functions are introduced in Chap. 4 through the use of convergent
sequences, and this concept is then used in Chaps. 5 and 6 to introduce continuity
and differentiation. As with any analysis book, results such as the intermediate-value
theorem and the mean-value theorem can be found in these chapters, but there are
also some other features. For instance, the logarithm is introduced in Chap. 5 and
then used to prove Gausss test for infinite series. In Chap. 6 we present a discrete
v
vi Preface
version of lHpitals rule that is seldom found in analysis texts. We conclude this
chapter with a short account of the differentiation of power series using differential
equations to motivate the discussion.
The Riemann integral is presented in Chap. 7. In the framework of elementary
analysis this integral is perhaps more accessible than, say, the Lebesgue integral,
and it is still an important concept. This chapter features a number of items beyond
the normal fare. In particular, a proof of Walliss formula followed by Stirlings
formula, and a proof that and e are irrational, appear here. In addition, there is
also a short section on numerical integration that further illustrates the definition of
the Riemann integral and applications of results such as the mean-value theorem.
Chapter 8 consists of a short account of Taylor series. Much of the analytical
apparatus for this topic is established earlier in the book so that, aside from Taylors
theorem, the chapter really covers mostly the mechanics of determining Taylor
series. There is an extensive theory on this topic, and it is difficult to limit oneself so
severely to these basic ideas. Here, despite the books emphasis on series, the authors
eschew topics such as the theorems of Abel and Tauber and, more importantly, the
question of which functions have a Taylor series. A full appreciation of this theory
requires complex analysis, which takes us too far afield.
The student encounters Newtons method in a first calculus course as an
application of differentiation. This method is based on constructing a sequence,
motivated geometrically, that converges (hopefully) to the solution of a given
equation. The emphasis in this first encounter is on the mechanics of the method
and choosing a sensible initial guess. In Chap. 9 we look at this method in the
wider context of the fixed-point problem. Fixed-point problems provide a practical
application of the theory of sequences. The sequence produced by Newtons method
is already familiar to the student, and the theory shows how problems such as error
estimates and convergence can be resolved.
The final chapter deals with sequences of functions and uniform convergence. By
this stage the reader is familiar with the example of power series, but those series
have particularly nice properties not shared generally by other series of functions.
The material is motivated by a problem in differential equations followed by various
examples that illustrate the need for more structure. This chapter forms a short
introduction to the field and is meant to prime the reader for more advanced topics
in analysis.
An introductory course in analysis is often the first time a student is exposed to
the rigor of mathematics. Upon reflection, some students might even view such a
course as a rite of passage into mathematics, for it is here that they are taught the
need for proofs, careful language, and precise arguments. There are few shortcuts
to mastering the subject, but there are certain things a book can do to mitigate
difficulties and keep the student interested in the material. In this book we strive to
motivate definitions, results, and proofs and present examples that illustrate the new
material. These examples are generally the simplest available that fully illuminate
the material. Where possible, we also provide examples that show why certain
conditions are needed. A simple counterexample is an exceedingly valuable tool
Preface vii
for understanding and remembering a result that is laden with technical conditions.
There are exercises at the end of most sections. Needless to say, it is here that the
student begins to fully understand the material.
The authors appreciate the encouragement and support of their wives. They also
thank Fiona Richmond for her help in preparing the figures. The work has also
benefited from the thoughtful comments and suggestions of the reviewers.
Palmerston North, New Zealand Charles H.C. Little

Kee L. Teo
Bruce van Brunt
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Ordered Pairs, Relations, and Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Induction and Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5 Finite Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2 Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.1 Definitions and Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Convergence of Sequences. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3 Algebra of Limits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.4 Subsequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.5 The Sandwich Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.6 The Cauchy Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.7 Monotonic Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.8 Unbounded Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3 Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.2 Definition of a Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.3 Elementary Properties of Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.4 The Comparison Test. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
3.5 Cauchys Condensation Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.6 The Limit Comparison Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
3.7 The Ratio Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
3.8 The Root Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
3.9 The KummerJensen Test. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
3.10 Alternating Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
3.11 Dirichlets Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
3.12 Absolute and Conditional Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
3.13 Rearrangements of Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
3.14 Products of Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
ix
x Contents
3.15 Introduction to Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

3.16 The Exponential, Sine, and Cosine Functions . . . . . . . . . . . . . . . . . . . . . . 175
4 Limits of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
4.2 Definition and Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
4.3 Basic Properties of Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
4.4 Algebra of Limits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
4.5 One-Sided Limits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
4.6 Infinite Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
5 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
5.1 Definition and Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
5.2 One-Sided Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
5.3 Continuity over an Interval . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
5.4 The Logarithm Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
5.5 Uniformly Continuous Functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
6 Differentiability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
6.1 Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
6.2 Differentiation Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
6.3 The Mean-Value Theorem for Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . 257
6.4 Periodicity of Sine and Cosine. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
6.5 LHpitals Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
6.6 A Discrete Version of lHpitals Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
6.7 Differentiation of Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
7 The Riemann Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
7.1 Area Under a Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
7.2 Upper and Lower Integrals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
7.3 The Riemann Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
7.4 Basic Properties of Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
7.5 The Fundamental Theorem of Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
7.6 The CauchySchwarz Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
7.7 Numerical Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
7.8 Improper Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
7.9 Integral Test for Convergence of a Series . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
8 Taylor Polynomials and Taylor Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
8.1 Taylors Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
8.2 Taylor Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
8.3 Some Shortcuts for Computing Taylor Series . . . . . . . . . . . . . . . . . . . . . . 416
9 The Fixed-Point Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
9.1 The Fixed-Point Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
9.2 Existence of Fixed Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
9.3 Fixed-Point Iteration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
Contents xi
10 Sequences of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
10.2 Uniform Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
10.3 Properties of Uniformly Convergent Sequences . . . . . . . . . . . . . . . . . . . . 452
10.4 Infinite Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
Chapter 1
Introduction
Classification: 03E15
1.1 Sets
To study analysis successfully, the reader must be conversant with some of the
basic concepts of mathematics. Foremost among these is the idea of a set. For our
purposes a set may be thought of as a collection of objects. This statement is too
imprecise to be regarded as a definition, and in fact it leads to logical difficulties,
but it does convey a mental image of a set that is satisfactory for our purposes. The
reader who wishes to delve into the nature of this concept more deeply is referred
to [10].
Sets are important in that they can be used to construct a host of mathematical
concepts. In fact, every mathematical object studied in this book can be constructed
from sets. We therefore begin this introductory chapter with some basic properties of
sets. Proofs are omitted because most of the properties are evident and their proofs
are straightforward.
First, the objects in a set X are called its elements or members. They are said
to be contained in X and to belong to X . If an object x is contained in X , then we
write x 2 X ; otherwise we write x X .
We shall assume the existence of a set with no elements. This set is denoted by ;,
and it is unique. It is said to be empty.
If X and Y are sets such that every element of X is also an element of Y , then
we say that X is a subset of Y and that it is included in Y . In this case we write
X Y ; otherwise we write X 6 Y . Note that ; X for every set X . The reasoning
is that since ; has no elements at all, it certainly has no elements that are not in X .
Therefore we can safely say, without fear of contradiction, that each of its elements
Springer Science+Business Media New York 2015 1

C.H.C. Little et al., Real Analysis via Sequences and Series, Undergraduate
Texts in Mathematics, DOI 10.1007/978-1-4939-2651-0_1
2 1 Introduction
does belong to X . In particular, ; is a subset of itself, and in fact it is its only subset.
Observe also that every set includes itself. Moreover, if X , Y , Z are sets such that
X Y and Y Z, then X Z. In this case we write X Y Z.
If X and Y are sets such that X Y X , then X and Y contain exactly the
same elements. In this case we say that these sets are equal, and we write X D Y .
Otherwise X and Y are distinct, and we write X Y . Any set is equal to itself,
and if X D Y , then Y D X . Furthermore, if X; Y; Z are sets such that X D Y and
Y D Z, then X D Z. In this case we write X D Y D Z. Equal sets are treated as
identical since they contain the same elements.
The collection of all subsets of a given set X is another set, called the power
set of X . It is denoted by P.X /. For example, P.;/ is a set having ; as its only
element. This set is denoted by f;g. Moreover, P.f;g/ is a set containing only the
elements ; and f;g and is denoted by f;; f;gg.
If X is any set, we may replace the elements of X by other objects and thereby
construct a new set. For example, we may replace the unique element ; of the set
f;g by any object Y . We then have a new set whose only element is Y . This set is
denoted by fY g. Similarly, if we replace the elements ; and f;g of the set f;; f;gg
by objects Y and Z, respectively, then we obtain a new set whose only elements
are Y and Z. This set is denoted by fY; Zg. The notation may be extended to an
arbitrary number of objects.
If X is a set and P is a property that may be satisfied by some elements of X ,
then we can construct a subset of X whose elements are precisely the members of
X that do satisfy P . This set is denoted by fx 2 X j P g. For example, let X and
Y be sets, and let P be the property that x 2 Y , where x 2 X . Then fx 2 X j P g
is the set whose elements are the objects that are in both X and Y . This set is called
the intersection of X and Y and is denoted by X \ Y . If this intersection happens
to be empty, then the sets X and Y are disjoint. If S is a collection of sets (in other
words, a set whose elements are themselves sets), then the sets in S are said to be
mutually disjoint if the sets A and B are disjoint whenever A 2 S and B 2 S.
On the other hand, if P is the property that x Y , then fx 2 X j P g is the
set whose elements are the members of X that are not in Y . This set is denoted by
X Y and is called the complement of Y with respect to X .
Let S be a collection of sets. Then we may construct another set whose elements
are the objects that belong to at least one member of S. This set is called the union
of S. For example, if S D fX; Y g for some sets X and Y , then the union of S is the
set of all objects that are in X or Y . In particular, it contains all the objects that are
in both of those sets. It is denoted by X [ Y .
We may define the intersection of S as the set of all objects in the union of S that
belong to every set in S. If S D fX; Y g for some sets X and Y , then the intersection
of S is the set of all objects that are in both X and Y . Thus it is equal to X \ Y .
1.2 Ordered Pairs, Relations, and Functions 3
1.2 Ordered Pairs, Relations, and Functions
As we said before, sets can be used to construct a large number of mathematical

concepts. In this section we show how to construct ordered pairs, relations, and
functions from sets, but once again the reader is referred to [10] for the details.
If x and y are any objects, then the ordered pair .x; y/ is defined as the set
ffxg; fx; ygg. We refer to x and y as its components, x being the first component
and y the second. The important observation to be made here is that the definition
does not treat x and y similarly (if in fact they are distinct objects). Instead, we
are given a way of distinguishing them: y is a member of just one of the two sets
in .x; y/, but x belongs to both sets. From this observation it is easy to deduce
that the ordered pairs .x; y/ and .a; b/ are equal if and only if x D a and y D b. In
other words, for equality to hold it is not sufficient for the sets fx; yg and fa; bg to be
equal. Their elements must also be listed in the same order. This is the only property
of ordered pairs that is important to remember. Once it is grasped, the definition may
be forgotten. The definition can be extended to ordered triples by defining
.x; y; z/ D ..x; y/; z/
for all objects x; y; z. Thus .x; y; z/ technically is an ordered pair whose first
component is itself an ordered pair. This notation may be extended to an arbitrary
number of objects, as we shall see later.
If X and Y are sets, we may construct a set X Y whose elements are the ordered
pairs .x; y/ such that x 2 X and y 2 Y . This set is called the Cartesian product
of X and Y . A subset of X Y is called a relation from X to Y . Thus a relation is
just a set of ordered pairs. A relation from X to X is sometimes called a relation on
X . An example is the relation R on X such that .x; y/ 2 R if and only if x D y.
This relation is called the equality relation. Similarly, we may define the inclusion
relation by specifying that it contains the ordered pair .x; y/ if and only if x and y
are sets such that x y.
If R is a relation and x and y are objects, we often write xRy to indicate that
.x; y/ 2 R. For instance, we are already accustomed to using D to denote the
equality relation and writing x D y instead of .x; y/ 2 D. We also write x 6 R y
instead of .x; y/ R.
If R and S are relations and x; y; z are objects, then we write xRySz if xRy and
ySz. This notation may be extended to arbitrarily long chains of relations.
Some kinds of relations are of particular importance, and we discuss them now.
If R is a relation on a set X , then R is reflexive if xRx for each x 2 X . Examples
include the equality and inclusion relations. The relation R is symmetric if yRx
whenever x and y are members of X satisfying xRy. Equality has this property, but
inclusion does not. We also say that R is transitive if xRy whenever there is a z 2 X
for which xRzRy. Equality and inclusion both exhibit this property. A relation with
all three of these properties is an equivalence relation. Equality is such a relation,
but inclusion is not.
4 1 Introduction
Equivalence relations are closely linked to partitions. A partition of a set X is a

set of nonempty, mutually disjoint subsets of X whose union is X . The elements of
a partition are sometimes called its cells. It can be shown that with any equivalence
relation R on a set X there is an associated unique partition P of X with the
property that elements x and y of X belong to the same cell of P if and only if xRy.
Conversely, any partition P of X has associated with it a unique equivalence relation
R on X constructed in the same way: xRy if and only if x and y belong to the same
cell of P . The cells of P are called the equivalence classes of R. For each x in X ,
we denote by x the unique equivalence class to which it belongs. Thus x D y
if and only if y is an element of X that belongs to the equivalence class x.
There is yet another kind of relation that is of paramount importance. A relation
f from a set X to a set Y is called a function from X into Y if for each x 2 X there
is a unique y 2 Y for which .x; y/ 2 f . We usually write y as f .x/. We say that f
maps x to y and that y is the image of x under f and corresponds to x under f .
We think of the function f as providing a rule for associating with each x in X a
unique corresponding element y of Y . It is this aspect of the concept that is usually
important for our purposes, but it is of interest to see how to construct the idea of a
function from sets, as we have done.
Given sets X and Y , we sometimes write f W X ! Y to indicate that f is a
function from X into Y . The set X is called the domain of the function f . The
subset of Y consisting of the images of the elements of X is the range of f . In this
book the range of f will usually be a set of real or complex numbers, in which case
we describe the function as real-valued or complex-valued, respectively. The range
of f is not necessarily the whole of Y : Some elements of Y might not correspond
to any element of X . The domain of f is denoted by Df and the range of f by Rf .
If Rf D Y , then f is described as surjective and called a surjection from X onto
Y . In the case of a surjection, each member of Y does correspond to an element of
X , but this element need not be unique.
Suppose on the other hand that each element of Rf does correspond to a unique
element of X . Then f is injective and an injection from X into Y . Thus f is
injective if and only if w D x whenever f .w/ D f .x/: Distinct elements of X
must be mapped to distinct elements of Y . An injective function is also described as
one-to-one.
Perhaps f is both injective and a surjection from X onto Y . Then f is a
bijection from X onto Y and described as bijective. In this case each member of Y
corresponds to a unique element of X . Thus there exists a function g from Y into X
such that g.f .x// D x for all x 2 X . This function is called the inverse of f and
is denoted by f 1 . It is a bijection from Y onto X . Note that f 1 .f .x// D x for
all x 2 X . Moreover f .f 1 .y// D y for each y 2 Y , and .f 1 /1 D f:
A function f with domain X is said to be constant (on X ) if f .w/ D f .x/
for each w and x in X . The range of a constant function with nonempty domain
therefore consists of a single element.
Now let f be a function from a set X into a set Y and g a function from Y into
a set Z. Then g.f .x// is defined for each x 2 X . Letting h.x/ D g.f .x// for each
1.3 Induction and Inequalities 5
such x, we see that h is a function from X into Z. We call it the composition of g

and f and denote it by g f . Thus
.g f /.x/ D g.f .x//
for each x 2 X . For instance, if f is a bijection from X onto Y , it follows that

.f 1 f /.x/ D x for each x 2 X and .f f 1 /.y/ D y for each y 2 Y . It is
easily checked that a composition of injections is injective and that a composition
of surjections is surjective. It follows that a composition of bijections is bijective.
If f is a function and X Df , then we write
f .X / D ff .x/ 2 Rf j x 2 X g:
Thus f .Df / D Rf .
If f is a real-valued function whose domain is a set of real numbers, then the
graph of f is the set of all points .x; f .x// in the Cartesian plane, where x 2 Df .
It is also the graph of the equation f .x/ D y.
1.3 Induction and Inequalities
Natural numbers are those used to count. The set f1; 2; : : :g of natural numbers is
denoted by N. We assume familiarity with the basic properties of these numbers
and simply highlight those that are of particular importance for our development
of analysis. The details of the development of natural numbers, integers, rational
numbers, and real numbers in terms of sets are given in [10] and will not be repeated
here.
The main properties of N that we require are that it contains the number 1 and
that every natural number has a successor in N. The successor of a natural number
n is denoted by n C 1. For example, the successor of 1 is 2 D 1 C 1 and that of 2
is 3 D 2 C 1. This notion of a successor for each natural number enables us to list
the natural numbers in order, beginning with 1. In fact, it can be shown that each
natural number n 1 is the successor of a unique natural number n 1. If Y is a
set of natural numbers such that 1 2 Y and n C 1 2 Y for each n 2 Y , then Y D N.
This observation can be applied to yield an important technique, called induc-
tion, for proving theorems about natural numbers. In this application, Y is the set
of all natural numbers for which the desired result is true. First, prove the desired
theorem for the natural number 1. (In other words, prove that 1 2 Y .) Then assume
that it is true for a particular natural number n (so that n 2 Y ) and prove it for n C 1
under this assumption. The conclusion that Y D N then shows that the theorem is
indeed true for all natural numbers. This idea is perhaps most easily visualized as
follows. Imagine a line of dominoes standing on end and close together so that the
line begins with a particular domino and extends indefinitely to the right of that first
domino. Knock the first domino onto the second. Then it is easy to see that all the
6 1 Introduction
dominoes will fall over, because the first domino will fall and every other domino
has one before it that will eventually knock it over. This picture captures the essence
of induction.
Before giving examples of induction at work, let us make some observations
about the pattern of a proof by induction. One approach to the use of induction to
prove a theorem about a natural number n, as we have seen, is to prove the theorem
for the natural number 1, then assume it for a particular natural number n (that is,
for a particular integer n 1), and finally prove it for n C 1. The assumption that
the theorem holds for a particular n is commonly called the inductive hypothesis.
We can in fact strengthen it by assuming that the theorem holds for all natural
numbers less than n as well. Equivalently, one could prove the theorem first for 1,
then assume as an inductive hypothesis that it holds for a particular integer n 1 (or
for all natural numbers less than n) where n 2, and finally prove it for n under
this assumption. Another observation is that the process of induction need not begin
with the natural number 1. In fact, it should begin with the smallest integer for which
the theorem is to be proved. In other words, suppose we wish to prove a theorem for
all integers n a for some fixed integer a. We start an inductive proof in this case
by proving the theorem for a. Then there are two equivalent ways to continue. One
is to assume as an inductive hypothesis that the theorem holds for some particular
integer n a (or for all integers k such that a k n) and then prove it for n C 1.
The alternative is to assume that it holds for n 1 for a particular integer n > a (or
for all integers k such that a k < n) and then prove it for n under this assumption.
Because of its importance, we now state the principle of induction formally as a
theorem. We denote the set of all integers by Z, so that
Z D f: : : ; 2; 1; 0; 1; 2; : : :g:
Theorem 1.3.1. Let Y Z and a 2 Y . Suppose also that y C 1 2 Y whenever

y 2 Y . Then Y contains every integer greater than a.
In applications of Theorem 1.3.1 to prove a given assertion about integers, Y is
the set of all integers for which the assertion in question is true.
We now offer an example of an inductive proof.
Example 1.3.1. We shall prove by induction that if x is a nonnegative real number,
then
n.n 1/ 2
.1 C x/n 1 C nx C x (1.1)
2
for all integers n 0. It is easy to see that equality holds for n D 0. Assume that
inequality (1.1) holds for a particular integer n 0. Then
.1 C x/nC1 D .1 C x/n .1 C x/

n.n 1/ 2
1 C nx C x .1 C x/
2

n.n 1/ n.n 1/ 3
D 1 C .n C 1/x C n C x2 C x
2 2
2n C n2 n 2
1 C .n C 1/x C x
2
n.n C 1/ 2
D 1 C .n C 1/x C x ;
2
since n.n 1/x 3 =2 0. The proof by induction is now complete.
It follows that
.1 C x/n 1 C nx (1.2)
and
n.n 1/ 2
.1 C x/n > x (1.3)
2
hold for every nonnegative real number x and nonnegative integer n. Inequality (1.2)
is known as Bernoullis inequality. Both of these inequalities will be found to be
useful later. 4
Example 1.3.1 involved an inequality. It is assumed that the reader is conversant
with inequalities and can work with them comfortably. One of the salient points to
remember is that multiplication of both sides of an inequality by a negative number
causes a change in the direction of the inequality. For instance, if x < y, then
x > y. Similarly, if both sides of an inequality have the same sign, then taking
the reciprocals of both sides again induces a change in the direction of the inequality.
Thus 1=x > 1=y if either x < y < 0 or 0 < x < y.
Induction can be used to make definitions as well as to prove theorems. Let us
illustrate this point by defining finite sums inductively. Let m and n be integers with
n > m. If am ; amC1 ; : : : ; an are numbers, we define
X
m
aj D am
j Dm
and
X
n X
n1
aj D aj C an :
j Dm j Dm
Often we write
X
n
aj D am C amC1 C C an :
j Dm
8 1 Introduction
This quantity is called the sum of am ; amC1 ; : : : ; an , and those numbers are its
terms. We also define
X
n
aj D 0
j Dm
if m > n, and we regard this expression as a sum with no terms.

We may define the product of the same numbers in a similar way. Thus
Y
m
aj D am
j Dm
and
Y
n Y
n1
aj D an aj
j Dm j Dm
for all n > m. We often write

Y
n
aj D am amC1 : : : an :
j Dm
This number is the product of am ; amC1 ; : : : ; an , and those numbers are its factors.
If m > n, then we define
Y
n
aj D 1
j Dm
and regard this expression as a product with no factors.

Similarly, P
we obtain
S analogous
T expressions for unions and intersections of sets
by replacing with and , respectively.
Induction may also be used to define exponentiation for powers that are natural
numbers: Given a real number a, set a1 D a and if an has been defined for a specific
natural number n, let
anC1 D an a:
We may also define a0 D 1, though this definition is normally made only if a 0.

Furthermore we define
1
an D
an
if n 2 N and a 0.
One can easily prove by induction that
.ab/n D an b n (1.4)
for any real numbers a and b and natural number n, and similarly that
a n an
D (1.5)
b bn
if b 0. If ab 0, then these two equations extend to the case where n is any

integer. For each fixed m 2 N, it is also easy to prove by induction on n that
am an D amCn (1.6)
for all n 2 N and then that
.am /n D amn (1.7)
for all n 2 N. If a 0, then these two rules may also be extended to the case where
m and n are any integers.
In order to extend these ideas, we need the following definition.
Definition 1.3.1. If n is a nonnegative integer and a0 ; a1 ; : : : ; an are constants, then
the function given by
X
n
aj x j (1.8)
j D0
for all numbers x is called a polynomial. Its degree is n if an 0. The numbers

a0 ; a1 ; : : : ; an are its coefficients. If
X
n
aj c j D 0;
j D0
then c is called a root of the polynomial (1.8).

It is shown in [10] that every nonnegative numberp a has a unique nonnegative
square root. This square root is denoted by a1=2 or a. Thus we have .a1=2 /2 D a.
It is also shown in [10] that every polynomial of odd degree whose coefficients
are real has a real root. For example, if m is an odd positive integer and a is a
real number, then there is a real number c satisfying the equation c m D a. The
uniqueness of c follows from the fact that if x and y are real numbers such that
x < y, then x m < y m since m is odd. (This fact is obvious if x 0 and y 0,
for then x m 0, y m 0, and at least one of x m and y m is nonzero. If 0 < x < y,
then use induction to prove it for all positive integers m. If x < y < 0, then note
10 1 Introduction
that 0 < y < x and apply the previous case with m odd.) We write a1=m D c. In
the case where m D 1, this definition is consistent with the equation a1 D a. In any
case, we have .a1=m /m D a, and so we refer to a1=m as the mth root of a. Note that
01=m D 0 and that a1=m has the same sign as a if a 0.
An arbitrary positive integer can be written in the form 2k m, where k and m are
nonnegative integers and m is odd. If n D 2k m, then we can show by induction
on k that each a 0 has a unique nonnegative nth root. Indeed, we have already
observed this fact if k D 0. Suppose that k > 0 and that a has a unique nonnegative
rth root c, where r D 2k1 m D n=2. Then
2 r 1 2r 1 n
1
aDc D
r
c 2 D c2 D c2 :
On the other hand, if a D b n D b 2r D .b 2 /r , wherepb 0, then b 2 D c by the

uniqueness of c, and so b D c 1=2 . We conclude that c is the unique nonnegative
nth root of a. We write it as a1=n . Hence
1 n
a n D a: (1.9)
Moreover 01=n D 0 and 11=n D 1.

An alternative proof of the existence of the nonnegative nth root of a will be
given in Example 5.3.5.
Let 0 a < b, and for some positive integer n let c D a1=n and d D b 1=n . If
c d , then we should have the contradiction that a D c n d n D b. We conclude
that c < d . For instance, if a > 1, then a1=n > 1.
Let a 0, let r and s be positive integers, and let c D a1=r . Then
1 rs 1 s r
cs D cs D c r D a;
so that
1 1
1 1s
a rs D c s D a r :
Next, let a 0, let n 2 N, and let m be a nonnegative integer. We then define
m
1 m
a n D an :
This definition generalizes Eq. (1.9) and is consistent with the equation a1 D a in
the case where m D 1 or n D 1. It is also consistent with the equation a0 D 1 in the
case where m D 0. If m D jk and n D jl for some positive integers j; k; l, then
1 jk 1 j !k
jk 1 j 1 k k
a D a jl
jl D al D al D al ;
a result that is consistent with the equation jk=.j l/ D k= l. If a > 0, then we define
m 1 1 1 m
a n D m D
1 m
D an :
a n
an
We proceed to the generalization of Eqs. (1.4)(1.7). Let a 0, let m be an

integer, and let n 2 N. Since
m n 1 m n 1 nm 1 n m
an D an D an D an D am ;
it follows that
1 m
1 m
.am / n D a n D a n :
Thus if p and q are also integers and q > 0, then

! 1 m !p
m pq 1 m q1 p 1 q 1 mp mp
an D an D an D a nq D a nq :
This result generalizes Eq. (1.7). Equation (1.6) also generalizes, since
p qm pn 1 qm 1 pn 1 qmCpn qmCpn m p
D a qn D a n C q :
m
a n a q D a qn a qn D a qn a qn D a qn
Now suppose that b is also a nonnegative real number. As

1 n 1 n 1 1 n
ab D a n bn D anbn ;
it then follows that

1 1 1
.ab/ n D a n b n :
Therefore
m 1 1 1 1 m m
.ab/ n D ..ab/m / n D .am b m / n D .am / n .b m / n D a n b n ;
a result that generalizes Eq. (1.4). Thus if b > 0, then

a mn m m m m
m
an
1 1 m
D .ab / n D a .bn / n Da b n n D m ;
b bn
and so Eq. (1.5) generalizes as well.
12 1 Introduction
We have now defined the real number ax for every number a > 0 and every
rational power x. Later we shall extend this definition to the case where x is any real
number. In fact, similar ideas may be used to define wz for any complex numbers w
and z provided that if w is real, then it is positive.
As another example of an inductive definition, we may define the factorial n of
a nonnegative integer n by writing 0 D 1 and
n D n.n 1/
for all n 2 N.
Given objects x1 ; x2 ; : : : ; xn , where n > 1, we may define the ordered set
.x1 ; x2 ; : : : ; xn /
by induction: The ordered pair .x1 ; x2 / has already been defined, and for each n > 2
we let
.x1 ; x2 ; : : : ; xn / D ..x1 ; x2 ; : : : ; xn1 /; xn /:
Let X1 ; X2 ; : : : ; Xn be sets, where n > 1. We can also define the Cartesian

product X1 X2 Xn of X1 ; X2 ; : : : ; Xn by induction: The definition has
already been made for n D 2, and for each n > 2 define
X1 X2 Xn D .X1 X2 Xn1 / Xn :
The result is the collection of all ordered sets .x1 ; x2 ; : : : ; xn / such that xj 2 Xj for
each j . If Xj D X for each j , then we write
X n D X1 X2 Xn :
The absolute value of a real number x plays a prominent role in analysis. Denoted
by jxj, it is defined as x if x 0 and x otherwise. Thus it is always the case that
jxj 0 and that
p
jxj D x 2 D j xj:
For every a > 0 it is also easy to see that jxj < a if and only if a < x < a. From
these inequalities it follows that jxj > a if and only if x > a or x < a, and that
jx yj < a if and only if y a < x < y C a, where y is another real number. In
addition, observe that jxj x and jxj x for all real x. Thus jxj C jyj x C y
and jxj C jyj .x C y/, so that
jx C yj jxj C jyj:
This result is known as the triangle inequality.

Inequalities can be used to define intervals on the real line. If a and b are real
numbers with a < b, then we write
.a; b/ D fx j a < x < bg;

a; b D fx j a x bg;
.a; b D fx j a < x bg;
a; b/ D fx j a x < bg:
These sets are called intervals. The first is open and the second closed; the others
are half-open. It should be clear from the context whether the notation .a; b/
specifies an open interval or an ordered pair. The numbers a and b are called the
ends of each of these intervals. In addition, we write
.a; 1/ D fx j a < xg;

a; 1/ D fx j a xg;
.1; a/ D fx j x < ag;
.1; a D fx j x ag:
These sets are also reckoned as intervals, and a is regarded as an end of each.
Let f be a real-valued function whose domain includes an interval I . Then f
is said to be increasing on I if f .x1 / < f .x2 / whenever x1 and x2 are numbers
in I such that x1 < x2 . If, on the other hand, f .x1 / f .x2 / for each such x1
and x2 , then f is nondecreasing on I . We define functions that are decreasing or
nonincreasing on I similarly. All these functions are said to be monotonic on I ,
and those that are increasing on I or decreasing on I are strictly monotonic on I .
Exercises 1.1.
1. Show that if a < b, then
a a C .1 /b b
for all such that 0 1.

2. Prove that
jxyj D jxjjyj
for all real numbers x and y. If y 0, prove also that

x
D jxj :
y jyj
14 1 Introduction
3. Use the triangle inequality to prove that
jjxj jyjj jx yj;
where x and y are real numbers.

4. Solve the following equations and inequalities, where x is real:
(a) j3x 2j D j5x C 4j; (d) j3x C 4j 2;
(b) jx C 4j D 4x; (e) j3x C 2j < 3jxj.
(c) j2x 3j < 4;
5. Prove the following by induction for all positive integers n and real numbers
a; a1 ; a2 ; : : : ; an :
jaQj D jaj . Q
n n
(a)
n n
(b) j D1 aj D j D1 jaj j.
P P

(c) nj D1 aj nj D1 jaj j.
p p
6. Prove that 2 is irrational by writing 2 D a=b, where a and b are positive
integers with no common factor, and obtaining a contradiction by showing that a
and b must both be even.
1.4 Complex Numbers
Throughout this book we will assume that any numbers we are working with are
complex unless an indication to the contrary is given by either the context or an
explicit statement. For example, the numbers a0 ; a1 ; : : : ; an ; x in the definition of a
polynomial (Definition 1.3.1) need not be real. As complex numbers might not be
as familiar to the reader as real numbers, we define them here and prove some basic
properties.
The equation
x 2 D 1 (1.10)
has no real solution. We seek to extend the real number system to include a number
i such that i 2 D 1 while preserving familiar operations such as addition and
multiplication. If we put x Ciy D 0, where x and y are real numbers, then x D iy,
so that x 2 D i 2 y 2 D y 2 and hence x D y D 0. Now if
x C iy D a C ib; (1.11)
where a and b are also real, then
x a C i.y b/ D 0:
1.4 Complex Numbers 15
The argument above shows that x a D y b D 0 and we conclude that Eq. (1.11)
holds if and only if x D a and y D b. Consequently x C iy may be identified with
the ordered pair .x; y/ of real numbers.
We therefore define a complex number as an ordered pair of real numbers.
Hence every complex number can be visualized as a point in the Cartesian plane. If
z D .x; y/, where x and y are real, then we define Re .z/ D x, and Im .z/ D y, and
we refer to x and y as the real and imaginary parts, respectively, of z. We define
addition and multiplication by the rules
.u; v/ C .x; y/ D .u C x; v C y/
and
.u; v/.x; y/ D .ux vy; uy C vx/;
respectively, where u; v; x; y are all real. These rules are motivated by the
calculations
.u C iv/ C .x C iy/ D u C x C i.v C y/
and
.u C iv/.x C iy/ D ux vy C i.uy C vx/:
Thus
.u; 0/ C .x; 0/ D .u C x; 0/
and
.u; 0/.x; 0/ D .ux; 0/:
Therefore we may identify .x; 0/ with the real number x. In particular, it follows
that .0; 0/ D 0 and .1; 0/ D 1. Note also that
Re .w C z/ D Re .w/ C Re .z/
and
Im .w C z/ D Im .w/ C Im .z/
for any two complex numbers w and z. It is also worth observing that the addition of
complex numbers may be interpreted geometrically as vector addition. In Chap. 6
we will give a geometric interpretation of the multiplication of complex numbers.
16 1 Introduction
The required solution i to Eq. (1.10) is identified with the ordered pair .0; 1/.
According to our definition of multiplication, it has the desired property:
i 2 D .0; 1/.0; 1/ D .1; 0/ D 1:
Note also that if x and y are real, then the definitions of addition and multiplication
do indeed give
x C iy D .x; 0/ C .0; 1/.y; 0/ D .x; 0/ C .0; y/ D .x; y/:
These results show that the algebraic manipulation of complex numbers is identical
to that of real numbers with the additional rule that i 2 D 1. In particular, we have
the laws that
a C b D b C a; (1.12)
a C .b C c/ D .a C b/ C c; (1.13)
and
a.b C c/ D ab C ac (1.14)
for all complex numbers a; b; c. We summarise Eqs. (1.12) and (1.13) by asserting
that the addition of complex numbers is commutative and associative, respectively.
Similarly, multiplication of complex numbers is commutative and associative.
Equation (1.14) asserts that multiplication is distributive over addition.
We also define
.x C iy/ D x iy:
Then subtraction is defined by the equation
w z D w C .z/
for all complex numbers w and z.

p a number i that gives
We have now extended the real number system to include
a solution to the equation x 2 D 1. More generally, i c gives a solution to the
equation x 2 D c, where c 0. In fact, it can be shown (see [10]) that every
nonconstant polynomial with complex coefficients has at least one complex root.
This result is known as the fundamental theorem of algebra. For example, suppose
that
az2 C bz C c D 0;
where a; b; c; z are all complex numbers and a 0. As the left-hand side of this
equation is a polynomial of degree 2, the equation is said to be quadratic. It can be
solved for z by the following procedure. Since a 0, we have
bz c
0 D z2 C C
a a

b 2 b2 c
D zC 2C
2a 4a a
2
b b 2 4ac
D zC ;
2a 4a2
so that

b 2 b 2 4ac
zC D :
2a 4a2
Thus
b 1 p 2
zC D b 4ac;
2a 2a
and we conclude that
p
b b 2 4ac
zD :
2a
The expression b 2 4ac is called the discriminant of the polynomial az2 C bz C c.

If a; b; c are real, then the polynomial has just two real roots if its discriminant is
positive, just one if its discriminant is 0, and none otherwise.
We proceed to the exponentiation of complex numbers. As in the case of real
numbers, we define z1 D z for every complex number z, and if zn has been defined
for a specific natural number n, we write znC1 D zn z. We also define z0 D 1 if
z 0.
Next, let
z D x C iy 0;
where x and y are real. Then
.x C iy/.x iy/ D x 2 C y 2 0;
and so
1 1 x iy x iy
D D D 2 :
z x C iy .x C iy/.x iy/ x C y2
18 1 Introduction
Thus if we define
x iy
z1 D ;
x2 C y2
then we have zz1 D 1. For every positive integer n we now define zn D .z1 /n if
z 0. This definition agrees with the equation z1 D z in the case where n D 1. We
also define w=z D wz1 for every complex number w.
We turn our attention now to two numbers that are associated with a given
complex number. The first is the conjugate of a complex number z D x C iy, where
x and y are real. The conjugate of z is defined as x iy and is denoted by zN. Thus
zN D z, and z D zN if and only if z is real. Geometrically, the function that maps z to zN
is a reflection about the x-axis.
If we also have w D u C iv, where u and v are real, then
w C z D u C x C i.v C y/ D u C x i.v C y/ D u iv C x iy D wN C zN:
Since
z D x iy D x C iy D .x iy/ D Nz;
it also follows that
w z D w C .z/ D w
N C z D wN C .Nz/ D wN zN:
Moreover
wz D ux vy C i.uy C vx/ D ux vy i.uy C vx/;
and so
N zN D .u iv/.x iy/ D ux vy C i.uy vx/ D wz:

w
If z 0, then
x C iy
1=z D D 1=Nz;
x2 C y2
and it follows that

wN
w=z D wz1 D wz
N 1 D w
N 1=z D :
zN
Note also that
z C zN D 2x D 2Re .z/
and, similarly,
z zN D 2i Im .z/I
hence
z C zN
Re .z/ D
2
and
z zN
Im .z/ D :
2i
A further observation is that
zNz D .x C iy/.x iy/ D x 2 C y 2 :
We now define
p
jzj D x2 C y2:
This number is called the modulus of z. For example, ji j D 1. We note that
zNz D jzj2 ;
p
jzj x 2 D jxj D jRe .z/j;
and, similarly,
p
jzj y 2 D jIm .z/j:
Moreover if z D x, then
p
jzj D x 2 D jxj:
This observation shows that jzj is a generalization of the absolute value of a real
number. It follows that
z C zN D 2Re .z/ 2jRe .z/j 2jzj:
If w is as defined in the previous paragraph, then

p
jz wj D jx C iy u ivj D jx u C i.y v/j D .x u/2 C .y v/2 I
20 1 Introduction
hence
jz wj D jw zj
and we perceive jz wj geometrically as the distance between z and w. In particular,

jzj is the distance between z and the origin. Note also that
jz cj D r;
where c is a complex number and r 0 is the equation of a circle with center c and
radius r.
We also have jzj 0 for all z, and jzj D 0 if and only if z D 0. In addition,
jzj D j zj D jNzj:
Since
jwzj2 D wz wz D wzwN
N z D wwzN
N z D jwj2 jzj2 D .jwjjzj/2 ;
we deduce that
jwzj D jwjjzj:
If z 0, it follows that

wz w
jwj D D jzj;
z z
and so

w
D jwj :
z jzj
The triangle inequality
jw C zj jwj C jzj
also holds, as can be inferred from the calculation
jw C zj2 D .w C z/w C z
D .w C z/.w
N C zN/
D wwN C wNz C zwN C zNz
D jwj2 C wNz C wNz C jzj2
jwj2 C 2jwNzj C jzj2
D jwj2 C 2jwjjNzj C jzj2

D jwj2 C 2jwjjzj C jzj2
D .jwj C jzj/2 :
Thus
jwj D jw z C zj jw zj C jzj;
so that
jw zj jwj jzj:
But we also have
jw zj D jz wj jzj jwj D .jwj jzj/;
and so we conclude that
jw zj jjwj jzjj :
Finally, the triangle inequality also shows that
jzj jxj C jiyj D jxj C jyj D jRe .z/j C jIm .z/j:
The notion of an inequality for real numbers does not extend to complex
numbers. We have x 2 0 for all real x, but i 2 < 0. It is meaningless to write
w < z if either w or z is not real. The inequality jzj < a is equivalent to a < z < a
if and only if a and z are both real.
Throughout the book we will denote by Z, Q, R, and C the sets of integers,
rational numbers, real numbers, and complex numbers, respectively.
Let f and g be functions whose domains are subsets of C. For all z 2 Df \ Dg ,
we define
.f C g/.z/ D f .z/ C g.z/;
.f g/.z/ D f .z/ g.z/;
and
.fg/.z/ D f .z/g.z/:
22 1 Introduction
We also define

f f .z/
.z/ D
g g.z/
for all z 2 Df \ Dg such that g.z/ 0. Thus f C g, f g, fg, and f =g are all
functions. In addition, if c 2 C, then we define cf to be the function such that
.cf /.z/ D cf .z/
for all z 2 Df .
Exercises 1.2.
1. Express in the form x C iy, where x and y are real,
(a) 3 C 4i C .1 i /.1 C i /;
(b) .2 5i /2 ;
12i
(c) 3C4i .
2. Compute i n for every integer n.
3. For every complex number z, find the real and imaginary parts of the following
expressions:
zC2
(a) z2i ;
(b) iz.
4. Let z D x C iy and w D a C ib, where x; y; a; b are real. Suppose that z2 D w.
Show that
a C jwj
x2 D
2
and
a C jwj
y2 D :
2
Deduce that
z D . C i /;
where
r
a C jwj
D ;
2
r
a C jwj
D ;
2
1.5 Finite Sums 23
D 1 if b 0 and D 1 if b < 0. Hence, conclude that the square root of a

complex number is real if and only if the complex number is real and positive.
5. Solve the following equations:
(a) z2 D 1 i ;
(b) z4 D i .
6. Show that
jzj jzj
;
ju C vj jjuj jvjj
where u; v; z are complex numbers and juj jvj.

7. If z and w are complex numbers, show that
jz C wj D jzj C jwj
if and only if z D w for some real number .

8. Recall that the function d.z; w/ D jz wj measures the distance between the
points representing the complex numbers z and w. Prove the inequality
d.z; w/ d.z; v/ C d.v; w/;
where w; v; z are complex numbers. More generally, let z1 ; z2 ; : : : ; zn be

complex numbers. Show that
d.z1 ; zn / d.z1 ; z2 / C d.z2 ; z3 / C : : : C d.zn1 ; zn /:
9. Show that
jd.z; v/ d.v; w/j d.z; w/
for complex numbers v; w; z.

10. Give a condition for
d.z; w/ D d.z; v/ C d.v; w/;
where v; w; z are complex numbers.
1.5 Finite Sums
Our development of analysis is based on the concepts of sequences and series. In

order to be able to deal with series, we need to be familiar with the properties
of finite sums. This section is therefore devoted to the development of their basic
properties.
24 1 Introduction
Let m and n be integers with n m. Recall that
X
n
aj D am C amC1 C C an ;
j Dm
where am ; amC1 ; : : : ; an are numbers. Observe that j is a dummy variable. In other

words, the sum is independent of j , so that if k is another letter, then
X
n X
n
ak D aj : (1.15)
kDm j Dm
(Sometimes j is referred to as an index.) More generally, let r be any integer and

put k D j C r. Then j D k r. Moreover k D m C r when j D m, and k D n C r
when j D n, and so we may write
X
n X
nCr
aj D akr :
j Dm kDmCr
Using Eq. (1.15), we therefore obtain the following result.

Proposition 1.5.1. If m; n; r are integers with m n and aj is a number for each
integer j such that m j n, then
X
nCr X
n
aj r D aj :
j DmCr j Dm
Moreover the associativity of addition shows that if m; n and r are integers such
that m r < n, then
X
n X
r X
n
aj D aj C aj :
j Dm j Dm j DrC1
The next theorem is a basic property of finite sums.

Theorem 1.5.2. Let m; n be integers with m n, and let aj and bj be numbers for
each integer j for which m j n. For all numbers s; t , we have
X
n X
n X
n
.saj C tbj / D s aj C t bj : (1.16)
j Dm j Dm j Dm
1.5 Finite Sums 25
Proof. We fix m and use induction on n. For n D m we have
X
m
.saj C tbj / D sam C tbm
j Dm
X
m X
m
Ds aj C t bj :
j Dm j Dm
Now suppose that Eq. (1.16) holds for some integer n m. Then
X
nC1 X
n
.saj C tbj / D .saj C tbj / C sanC1 C tbnC1
j Dm j Dm
X
n X
n
Ds aj C t bj C sanC1 C tbnC1
j Dm j Dm
0 1 0 1
X
n X
n
D s@ aj C anC1 A C t @ bj C bnC1 A
j Dm j Dm
X
nC1 X
nC1
Ds aj C t bj ;
j Dm j Dm
and the proof by induction is complete. t

u
For example, by taking t D 0, we obtain the distributive law:
X
n X
n
saj D s aj : (1.17)
j Dm j Dm
Similarly, putting s D t D 1, we find that
X
n X
n X
n
.aj C bj / D aj C bj ;
j Dm j Dm j Dm
and by setting s D 1 and t D 1, we have
X
n X
n X
n
.aj bj / D aj bj :
j Dm j Dm j Dm
26 1 Introduction
Thus if m n and am ; amC1 ; : : : ; anC1 are numbers, then
X
n X
n X
n
.aj C1 aj / D aj C1 aj
j Dm j Dm j Dm
X
nC1 X
n
D aj aj
j DmC1 j Dm
0 1
X
n X
n
D aj C anC1 @am C aj A
j DmC1 j DmC1
D anC1 am ;
a result known as the telescoping property. We state it as a theorem.

Theorem 1.5.3. Let m; n be integers such that m n, and let aj be a number for
each j such that m j n C 1. Then
X
n
.aj C1 aj / D anC1 am :
j Dm
This theorem in fact is intuitively clear, since the sum can be written as
.anC1 an / C .an an1 / C C .amC1 am /
and cancellation yields anC1 am . Note also that
X
n X
n
.aj aj C1 / D .aj C1 aj / D .anC1 am / D am anC1 :
j Dm j Dm
As an example, if aj D j for each j , we obtain
X
n X
n
1D .j C 1 j / D n C 1 m
j Dm j Dm
by the telescoping property. In particular, if m D 1 and n is a positive integer, then
X
n
1 D n C 1 1 D n;
j D1
as expected, because we are simply adding n copies of the number 1.

1.5 Finite Sums 27
For a less trivial application of the telescoping property, let us evaluate
X
n
.2j C 1/:
j D1
Note first that
.j C 1/2 j 2 D j 2 C 2j C 1 j 2
D 2j C 1:
In the telescoping property we therefore take aj D j 2 for each j and thereby obtain
X
n X
n
.2j C 1/ D ..j C 1/2 j 2 /
j D1 j D1
D .n C 1/2 1
D n2 C 2n:
But we also have
X
n X
n X
n
.2j C 1/ D 2 jC 1
j D1 j D1 j D1
X
n
D2 j C n;
j D1
and so
X
n X
n
2 j D .2j C 1/ n
j D1 j D1
D n2 C 2n n
D n2 C n:
Hence we obtain the following theorem.

Theorem 1.5.4. For every positive integer n
X
n
n.n C 1/
j D :
j D1
2
28 1 Introduction
A similar argument can be used to show that
X
n
n.n C 1/.2n C 1/
j2 D :
j D1
6
The next theorem also illustrates the telescoping property in action.

Theorem 1.5.5. If n is a nonnegative integer and a b, then
X
n
anC1 b nC1 b nC1 anC1
aj b nj D D ;
j D0
ab ba
where the convention that 00 D 1 is used.

Proof. By distributivity and the telescoping property, we have
X
n X
n
.a b/ aj b nj D .aj C1 b nj aj b nj C1 /
j D0 j D0
D anC1 b nC1 ;
and the result follows. t

u
Putting b D 1, we draw the following conclusion.
Corollary 1.5.6. If a 1, then
X
n
anC1 1 1 anC1
aj D D ;
j D0
a1 1a
where 00 D 1.
There is one further theorem concerning finite sums that we include in this
section. It is called the binomial theorem and gives a formula for .a C b/n for every
nonnegative integer n. It furnishes another example of a proof by induction.
First we introduce some new notation. If n and r are integers such that 0 r n,
then we define
!
n n
D
r r.n r/
n.n 1/ .n r C 1/
D :
r
1.5 Finite Sums 29
For example,
!
n n
D D 1:
0 0n
Similarly,
!
n
D 1:
n
Because of their role in the binomial theorem, these numbers are often called the
binomial coefficients. They satisfy the following lemma.
Lemma 1.5.7. If n and r are positive integers and r n, then
! ! !
nC1 n n
D C :
r r r 1
Proof. By direct calculation, we have

! !
n n n n
C D C
r r 1 r.n r/ .r 1/.n r C 1/
n.n r C 1 C r/
D
r.n r C 1/
.n C 1/
D
r.n r C 1/
!
nC1
D :
r
t
u
Theorem 1.5.8 (Binomial Theorem). Let a and b be numbers and n a nonnegative
integer. Then
!
X n
n j nj
.a C b/ D
n
a b ; (1.18)
j D0
j
where the convention that 00 D 1 is used.

30 1 Introduction
Proof. Both sides are equal to 1 if n D 0. Assume that the theorem holds for some
integer n 0. Then
.a C b/nC1 D .a C b/.a C b/n

!
X n
n j nj
D .a C b/ a b
j D0
j
! !
X n
n j C1 nj X n j nj C1
n
D a b C a b
j D0
j j D0
j
! !
X
nC1
n X n
n j nj C1
j nj C1
D a b C a b
j D1
j 1 j D0
j
! !
X n
n X n
n j nC1j
D j nC1j
a b Ca nC1
Cb nC1
C a b
j D1
j 1 j D1
j
! !!
X n
n n
D b nC1 C C aj b nC1j C anC1
j D1
j 1 j
!
X n
n C 1 j nC1j
Db nC1
C a b C anC1
j D1
j
!
X
nC1
n C 1 j nC1j
D a b :
j D0
j
The result follows by induction. t

u
The sum on the right-hand side of Eq. (1.18) is often referred to as the binomial
expansion of .a C b/n . Note that the result of Example 1.3.1 can easily be deduced
by considering three terms of the binomial expansion with a D 1 and b D x.
Exercises 1.3.
1. Use induction to prove the following formulas for all positive integers n:
Pn
(a) j D n.nC1/ ;
Pjn D1 2 2
n.nC1/.2nC1/
(b) j D ;
Pjn D1 1 6
(c) j D1 D n
;
Pn j.jjC1/ nC1
(d) j D1 .j C1/ D 1 .nC1/
1
;
Pn P 2
n
j D1 j D
3
(e) j D1 j ;
1.5 Finite Sums 31
Pn n.nC1/.2nC1/.3n2 C3n1/
(f) j D1 j4 D 30
;
Pn1 xnxn C.n1/x nC1
(g) j D1 jxj D 1x 2
for every real x 1.
2. Let n 2 N and let a1 ; a2 ; : : : ; an be real numbers such that 0 aj 1 for all j .
Prove that
Y
n X
n
.1 aj / 1 aj :
j D1 j D1
3. Discover and prove a theorem about the relative sizes of 3n and n, where n is a
positive integer.
4. A function f W R ! R is said to be convex if for all nonnegative real numbers 1
and 2 with sum equal to 1 we have
f .1 x1 C 2 x2 / 1 f .x1 / C 2 f .x2 /
for each x1 2 R and x2 2 R. Prove that

0 1
X n X
n
f @ j xj A j f .xj /
j D1 j D1
whenever n is an integer greater than 1 and 1 ; 2 ; : : : ; n are nonnegative real

numbers with sum 1. This result is known as Jensens inequality. More generally,
show that
0 1
Xn X n Xn X n
f @ .j xj / k A .j f .xj // k :
j D1 kD1 j D1 kD1
5. Use the telescoping property to prove the following identities for all n 2 N:
Pn 1
j D1 j.j C2/ D 4 2 nC1 C nC2 ;
1 3 1 1
(a)
Pn p
(b) p 1
p D 1 C n C 1;
Pjn D1 j C j C1
(c) j D1 j j D .n C 1/ 1;
Pn 6j 3nC1
(d) j D1 .3j C1 2j C1 /.3j 2j / D 3 3nC1 2nC1 ;
Pn 1
j D1 j 2 C4j C3 D 12 2 nC2 nC3 ;
1 5 1 1
(e)
Pn
(f) j D1
p 1 p D 1 pnC1
1
;
Pn .j C1/ j
j Cj j C1
2 C1 D 2 2n.nC1/C2 ;
1 1
(g) j D1 4
Pn j Cj j
j D1 4j 4 C1 D 4 8n2 C8nC4 .
1 1
(h)
32 1 Introduction
6. Let d and a1 be numbers. Define aj for each integer j > 1 by the equation
aj C1 D aj C d , and suppose that aj 0 for all j 2 N. Find the sum
X
n
1
j D1
aj aj C1
for all n 2 N.
7. Prove that
X
n
j .j 2 C j C 1/ D n.n C 1/2 1
j D1
for all n 2 N.
8. Evaluate the sum
X
n
1
j 24
j D3
for all integers n 3.

9. Suppose that jnan j M for all n 0. Show that
1 X
n
Sn n D jaj M
n C 1 j D1
Pn Pn
for all n, where Sn D j D0 aj and n D j D0 Sj for all n.
Chapter 2
Sequences
Classification: 40A05
Analysis is based on the notion of a limit, a concept that can be defined in terms of
sequences. Moreover, elementary functions, such as trigonometric, exponential, and
logarithm functions and many algebraic functions, can be approximated by using
sequences. With modern computers, such approximations can be made accurate
enough for most practical purposes.
2.1 Definitions and Examples
A sequence is a function whose domain is the set of all integers greater than or equal
to some fixed integer. More formally, we have the following definition.
Definition 2.1.1. Let a be a fixed integer and A the set of all integers greater than
or equal to a. A function s from A into a set F is called a sequence in F. The images
of the members of A are called the terms of the sequence. They are ordered by
the ordering of A itself so that, for example, s.a/ is the first term of the sequence,
s.a C 1/ is the second, and so forth.
Throughout this book F will denote one of the three fields Q; R; or C. The
sequence will then be said to be rational, real, or complex, respectively. Results
established for complex sequences will therefore be valid for rational and real
sequences as they are special cases.
Given a sequence s W A ! F, we usually write sn instead of s.n/, where n 2 A.
Moreover we denote the sequence by fsn gn2A or simply fsn g if A is either clear from
the context or immaterial. If A is the set of all integers n a, then we also write the
sequence as fsn gna . Occasionally, the first few terms of the sequence may be listed
in order, so that we write sa ; saC1 ; : : :, for instance.

34 2 Sequences
We now present some examples of sequences.

Example 2.1.1 (Sign Sequence). The sequence given by f.1/nC1 gn1 is
1; 1; 1; 1; : : : :
Example 2.1.2 (Harmonic Sequence). The sequence given by f1=ngn1 is
1 1 1
1; ; ; ; : : : :
2 3 4
4
Example 2.1.3 (Complex Harmonic Sequence). The sequence given by fi n =ngn1 is
1 i 1 i 1 i 1
i; ; ; ; ; ; ; ; : : : :
2 3 4 5 6 7 8
4
Example 2.1.4 (Geometric Sequence). If we temporarily adopt the convention that
00 D 1, then for all complex numbers a and r, the sequence given by farn gn0 is
a; ar; ar2 ; ar3 ; : : : :
Example 2.1.5 (Arithmetic Sequence). For each complex a and d the sequence
given by fa C ndgn0 is
a; a C d; a C 2d; a C 3d; : : : :
Example 2.1.6. The sequence given by f.1 C 1=n/n gn1 is
9 64 625
2; ; ; ;::::
4 27 256
We shall return to this sequence later. 4
Often, particularly in computer applications, a sequence is defined inductively
by specifying the first few terms and then defining the remaining terms by means
of the preceding ones. For example, the geometric sequence farn g can be defined
2.1 Definitions and Examples 35
inductively by writing s0 D a and sk D rsk1 for all k > 0. Likewise, the arithmetic
sequence fa Cndg can be defined by setting s0 D a and sk D sk1 Cd for all k > 0.
Example 2.1.7. A sequence that is often used in botany is the Fibonacci sequence
(see [6]). It is specified by putting F0 D F1 D 1 and
Fk D Fk1 C Fk2
for all k > 1. The sequence so defined is
1; 1; 2; 3; 5; 8; 13; 21; : : : :
We show by induction that

0 p !nC1 p !nC1 1
1 @ 1C 5 1 5 A
Fn D p
5 2 2
p
pn 0. First, for convenience let us define D .1 C
for all 5/=2 and D
.1 5/=2. These are the solutions of the equation
x 2 x 1 D 0; (2.1)
p
so that 2 D C 1 and 2 D C 1. Since D 5, the required formula holds p
for n D 0. It also holds for n D 1, since 2 2 D C 1 . C 1/ D D 5.
Assuming that n > 1 and that the result holds for all positive integers less than n, it
follows that
Fn D Fn1 C Fn2
1
D p . n n C n1 n1 /
5
1
D p . n1 . C 1/ n1 . C 1//
5
1
D p . nC1 nC1 /;
5
as required. It is somewhat surprising that the formula yields an integer. The number
is known as the golden ratio. 4
Example 2.1.8. Fix a positive integer k. Define s1 D 1 and
sn1 C k
sn D
sn1 C 1
36 2 Sequences
for all n > 1. The resulting sequence is
k C 1 3k C 1 k 2 C 6k C 1
1; ; ; ;::::
2 kC3 4k C 4
We
p will see later (Example 2.3.2) that its terms give a good approximation for
k when n is large. For k D 2, the sequence was familiar to the ancient Greeks,
having appeared in Chapter 31 of the first volume of the manuscript Expositio rerum
mathematicarum ad legendum Platonem utilium by Theon of Smyrna in 130 AD.
The reader is invited to show by induction that
p p
p .1 C k/n C .1 k/n
sn D k p p (2.2)
.1 C k/n .1 k/n
for all n. 4
Remark. The reader may wonder how the expressions for the general term were
obtained in the two preceding examples. One method using matrices is given in [15].
Here is another way of obtaining the formula for the Fibonacci sequence. Let us
write
Fn D A n C B n
for some constants A; B; ; such that 0 and 0. For the equation
Fn D Fn1 C Fn2
to hold, we need to ensure that
A n C B n D A n1 C B n1 C A n2 C B n2 ;
that is,
A n2 . 2 1/ C B n2 . 2 1/ D 0:
For this purpose it is sufficient to take and to be the solutions of Eq. (2.1). This
observation gives the values for and in Example 2.1.7. Next, the initial condition
that F0 D F1 D 1 yields the equations
ACB D1
and
A C B D 1:
2.2 Convergence of Sequences 37
The solution gives

p
1 51
BD D p D p ;
2 5 5
so that

AD1B D p :
5
The desired expression for Fn is now obtained by substitution.

Exercises 2.1.
1. Prove Eq. (2.2).
2.2 Convergence of Sequences
We are often concerned with the limiting behavior of a sequence fsn g as n becomes
large. The process of enlarging n indefinitely is indicated by the notation n ! 1.
In what follows the sequences will be assumed to be complex unless an indication
to the contrary is given.
Definition 2.2.1. Let fsn g be a sequence and L a number. We say that fsn g
converges to L, and that L is the limit of fsn g, if for each " > 0 there exists an
integer N such that
jsn Lj < "
whenever n N . If fsn g converges to L, then we also say that sn approaches L as

n approaches infinity, and we write
lim sn D L
n!1
or sn ! L as n ! 1.
A sequence is said to be convergent if there exists a number to which it
converges. A sequence is said to diverge, and to be divergent, if it does not
converge.
In Definition 2.2.1 the integer N may of course depend on ". It is also clear that
the inequality n N may be replaced by the corresponding strict inequality, for
n N if and only if n > N 1.
Given L and " > 0, we may define the set of all z for which jz Lj < "
as the "-neighborhood of L. This set is denoted by N" .L/. Roughly speaking,
the definition of the convergence of a sequence to L says that for each pre-
scribed "-neighborhood of L, the terms of the sequence will eventually enter the
38 2 Sequences
neighborhood and remain there. In other words, the terms of the sequence become
arbitrarily close to L as n increases.
Note that the inequality jsn Lj < " is equivalent in R to
L " < sn < L C ":
A sequence that converges to 0 is said to be null.

Example 2.2.1. For each number a the constant sequence fag converges to a.
Proof. Define sn D a for each integer n 2 N. For every " > 0 and every n > 0 we
have jsn aj D ja aj D 0 < ", as desired.
4
Example 2.2.2. Let us show that
in
lim D 0:
n!1 n
Given any " > 0, we need to find an integer N such that

n
i
0 < "
n
for all n N . Since ji j D 1, we need 1=n < ", and so we choose any N > 1=" > 0.
For all n N we deduce that
1 1
< ";
n N
and the desired result follows. 4
Remark. The only property of i used in this proof is that ji j D 1. Hence the
argument can also be used to show that
1
lim D 0:
n!1 n
Example 2.2.3. We show that the sequence f.1/n g diverges. This result is intu-
itively clear, for the values of .1/n oscillate between 1 and 1 as n increases. The
distance between these numbers is 2, and so every number L must be at a distance
of at least 1 from one or the other of them. But if L were the limit of our sequence,
then the terms would become arbitrarily close to L as n increases. We infer that no
number can be the limit of the sequence, and the sequence therefore diverges.
We can also cast this intuitive argument in terms of neighborhoods. For every
number L we can choose " so small that the "-neighborhood of L does not contain
both 1 and 1. Therefore there is no term in the sequence beyond which the values
in the sequence remain in the "-neighborhood.
We now transform this intuition into a formal argument. Suppose the sequence
were to have a limit L, and choose any " such that 0 < " < 1. There would be an
integer N such that j.1/n Lj < " whenever n N , no matter whether n is even
or odd. It would follow that
j1 Lj < "
and
j1 C Lj D j 1 Lj < ":
Using the triangle inequality, we would therefore reach the contradiction that
2 D j1 C L C 1 Lj j1 C Lj C j1 Lj < 2" < 2:
4
Example 2.2.4. For every complex number z such that jzj < 1, we will show that
lim zn D 0:
n!1
If z D 0, then the result follows from Example 2.2.1 with a D 0, and so we

assume that z 0. Choose " > 0. We must find an integer N such that jzn j < " for
all n N . Since 0 < jzj < 1, we have 1=jzj > 1. Let us write
1
D1Cp
jzj
for some p > 0. By Example 1.3.1,
.1 C p/n 1 C np
for all positive integers n. Choose N large enough so that
1
1 C Np > :
"
Then, for all n N , we have
n
1 1 1
D D .1 C p/n 1 C np 1 C Np > > 0;
jzn j jzj "
so that jzn j < " for each such n, as required. 4

40 2 Sequences
p p p
Remark. Since 1 < 5 < 3, we have 0 > 1 5 > 2. Thus j1 5j < 2, and
so

1 p5

< 1:
2
Applying the result of the last example, we conclude that

p !n
1 5
lim D 0:
n!1 2
This observation motivates us to approximate Fn in Example 2.1.7 by

p !nC1
1 1C 5
p :
5 2
In fact, this approximation turns out to be accurate to within one decimal place for
all n 4.
Since the convergence of a sequence fsn g depends only on the behavior of the
terms sn where n is large, we would expect intuitively that the first few terms are
immaterial. This expectation is encapsulated in the following lemma.
Lemma 2.2.1. Let k be a positive integer. Then a sequence fsn g approaches a
number L as n approaches infinity if and only if fsnCk g also approaches L.
Proof. Suppose first that sn ! L as n ! 1 and choose " > 0. Since
lim sn D L;
n!1
there exists N such that jsn Lj < " whenever n N . But for each n N we
have n C k > n N , and so jsnCk Lj < ". Therefore
lim snCk D L;
n!1
as required.
Conversely, suppose snCk ! L as n ! 1. For every " > 0 there exists N such
that jsnCk Lj < " for all n N . Choose n N C k. Then jsn Lj < ", and we
conclude that sn ! L as n ! 1.
In view of this lemma, every theorem or definition concerning limits that
postulates sn to satisfy a specified property for all n will remain valid if in fact
sn satisfies the specified property only for all n greater than some fixed integer.
We show next that if a sequence converges, then the limit is unique. Certainly,
our intuition leads us to expect this result. The reasoning is akin to that used in
Example 2.2.3. If the sequence fsn g were to converge to two distinct numbers L1
and L2 , which are a distance jL1 L2 j apart, then each term sn would be at a distance
of at least jL1 L2 j=2 from one or the other of them. Therefore sn could not become
arbitrarily close to both L1 and L2 as n increases. In terms of neighborhoods, we
choose " so small that the "-neighborhoods of L1 and L2 are disjoint. It is therefore
impossible that as n increases the terms of the sequence enter both neighborhoods
and remain there. The proof of the uniqueness of the limit just formalizes this
intuitive argument.
Theorem 2.2.2. A sequence has at most one limit.
Proof. Let fsn g be a sequence with limits L1 and L2 and choose " > 0. Since sn
approaches L1 , there is an integer N1 such that
jsn L1 j < "
for all n N1 . Likewise, there is an integer N2 such that
jsn L2 j < "
for all n N2 . For every n maxfN1 ; N2 g, both of the preceding inequalities hold.
For such n we therefore have
jL1 L2 j D jL1 sn C sn L2 j jsn L1 j C jsn L2 j < 2":
But " is an arbitrary positive number. If L1 L2 , then we could choose
jL1 L2 j
"D > 0;
2
thereby obtaining the contradiction that jL1 L2 j < jL1 L2 j. We deduce that
L1 D L2 , as required.
The next proposition gives another way of expressing the definition of a limit of
a sequence. It is often more convenient to use
Proposition 2.2.3. Let fsn g be a sequence, L a number, and c a positive number.
Then limn!1 sn D L if and only if for each " > 0 there exists an integer N such
that
jsn Lj < c"
for all n N .
Proof. Since c" > 0, it is immediate from the definition of the limit that the stated
condition holds if limn!1 sn D L. Let us suppose therefore that for each " > 0
42 2 Sequences
there exists an integer N such that jsn Lj < c" for all n N . Choose " > 0.
Since "=c > 0, there exists N such that
"
jsn Lj < c D"
c
for all n N . Hence limn!1 sn D L. t
u
Sometimes we wish to consider more than one sequence, in which case the
following proposition is sometimes useful.
Proposition 2.2.4. Let fsn g and ftn g be sequences with respective limits K and
L. Then for each " > 0 there is an integer N such that both jsn Kj < " and
jtn Lj < " for all n N .
Proof. Given " > 0, there exists N1 such that jsn Kj < " for all n N1 .
Similarly, there exists N2 such that jtn Lj < " for all n N2 . If we now take
N D maxfN1 ; N2 g and choose n N , then n N1 and n N2 , so that both
required inequalities follow. t
u
Remark. This result can clearly be generalized to handle situations where more than
two sequences are under consideration.
Exercises 2.2.
1. Show from the definition that the following sequences are divergent:

(a) .1/n C n1 ;
n n
o
(b) 1C.1/
2
;
n
(c) i C 2n .
1
2. Show that
n
1Ci
lim D 0:
n!1 2
3. Suppose that xn > 0 for all n 2 N and that limn!1 xn D 0. Show that the set
S D fxn j n 2 Ng
contains a maximum member. Must S always contain a minimum member?

4. Let
x C xn
sn .x/ D
1 C xn
for all n 2 N and x 2 R f1g. Find each real number x for which the sequence
fsn g is convergent and find the limit of the sequence.
2.3 Algebra of Limits 43
5. Test the convergence of the real sequence

an b n
;
an C b n
where a and b are numbers such that an C b n 0 for all n.

6. Suppose that
zn L
lim D 0;
n!1 zn CL
where L 2 C and zn 2 C fLg for all n. Show that
jzn j jLj
lim D0
n!1 jzn j C jLj
and limn!1 zn D L.
7. Let w be a complex number such that jwj 1 and let
wn
zn D
1 C w2n
for all n. Show that the sequence fzn g converges to 0. Is this statement still true
if jwj D 1?
2.3 Algebra of Limits
Using the definition of convergence to determine whether a sequence converges, and

if so to what limit, can be a tedious process. However, if a given sequence is a sum,
difference, product, or quotient of other sequences, then the behavior of the given
sequence can be investigated by studying simpler sequences. Our next objective is
to see how this simplification is effected. We begin with some results concerning
the size of terms in a convergent sequence.
A sequence fsn g is said to be bounded if there exists M such that jsn j < M for
all n.
Theorem 2.3.1. Every convergent sequence is bounded.
Proof. Let fsn g be a convergent sequence with limit L. For each " > 0 there is an
integer N such that jsn Lj < " for all n N . For every such n we have
jjsn j jLjj jsn Lj < ";
so that
jLj " < jsn j < jLj C ":

44 2 Sequences
Setting
M D maxfjs1 j; js2 j; : : : ; jsN 1 j; jLj C "g;
we find that jsn j M for all n. Hence fsn g is bounded. t

u
The preceding proof shows that if
lim sn D L;
n!1
then
lim jsn j D jLj:

n!1
On the other hand, we observe from Example 2.2.3 that the converse does not
necessarily hold. Nevertheless, if limn!1 jsn j D 0, then for each " > 0 there exists
an integer N such that jsn j < " for all n N , and we conclude that limn!1 sn D 0:
We summarize these observations in the following proposition.
Proposition 2.3.2. 1. If limn!1 sn D L, then limn!1 jsn j D jLj.
2. If limn!1 jsn j D 0, then limn!1 sn D 0.
Proposition 2.3.3. Suppose fsn g is a real sequence such that
lim sn D L > c
n!1
for some number c. Then there exist numbers N and k > c such that sn > k for all
n N.
Proof. Choose " such that 0 < " < L c. There exists N such that jsn Lj < "
for all n N . For each such n we have
L " < sn < L C ":
The conclusion of the theorem is therefore satisfied by k D L " > c: t

u
Similarly, one can establish the following result.
Proposition 2.3.4. Suppose fsn g is a real sequence such that
lim sn D L < c:
n!1
Then there exist numbers N and k < c such that sn < k for all n N .
Also, if fsn g is a real sequence such that
lim sn D L > 0;
n!1
then we may take c D L=2 in Proposition 2.3.3 and infer the existence of a number
N such that sn > L=2 for all n N . A corresponding statement holds if L < 0.
Thus we have the following result.
Proposition 2.3.5. Let fsn g be a real sequence such that
lim sn D L 0:
n!1
1. If L > 0, then there is an integer N such that sn > L=2 for all n N .
2. If L < 0, then there is an integer N such that sn < L=2 for all n N .
Corollary 2.3.6. Let fsn g be a (real or complex) sequence such that
lim sn D L 0:
n!1
Then there is an integer N such that jsn j > jLj=2 for all n N .
Proof. From the hypothesis it follows that
lim jsn j D jLj > 0:

n!1
An appeal to Proposition 2.3.5(1) completes the proof. t

u
We are now ready to establish the sum, product, and quotient rules for limits.
Theorem 2.3.7. Let fsn g and ftn g be sequences, and suppose that limn!1 sn D K
and limn!1 tn D L. Then
1.
lim .sn C tn / D K C LI
n!1
2.
lim sn tn D KLI
n!1
3.
sn K
lim D
n!1 tn L
if L 0 and tn 0 for all n.

Proof. 1. Choose " > 0. Using Proposition 2.2.4, we find an integer N such that
both
jsn Kj < "

46 2 Sequences
and
jtn Lj < "
for all n N . For every such n we deduce that
j.sn C tn / .K C L/j D j.sn K/ C .tn L/j

jsn Kj C jtn Lj
< 2";
and the result follows from Proposition 2.2.3.

2. Choosing " > 0, we find an integer N with the property stated in the proof of
part (1). Moreover, the convergent sequence fsn g is bounded, and so there exists
M such that jsn j < M for all n. For all n N we therefore have
jsn tn KLj D jsn tn Lsn C Lsn KLj
jsn jjtn Lj C jLjjsn Kj
< M " C jLj"
D .M C jLj/";
and again the result follows from Proposition 2.2.3.
3. By Corollary 2.3.6 there exists an integer N1 such that jtn j > jLj=2 > 0 for all
n N1 . Thus
1 2
< :
jtn j jLj
We prove next that
1 1
lim D :
n!1 tn L
To this end, choose " > 0. There is an integer N2 such that jtn Lj < " for all
n N2 . For every n maxfN1 ; N2 g we therefore have

1
1 D L tn
t L Ltn
n
jtn Lj
D
jLjjtn j
1 2
< "
jLj jLj
2
D ";
jLj2
as required.
We conclude the proof by using part (2) to deduce that
sn 1 K
lim D lim sn lim D :
n!1 tn n!1 n!1 tn L
t
u
Note that the sequences fsn C tn g, fsn tn g, and fsn =tn g may converge even when
neither fsn g nor ftn g does so. For instance, f.1/n g and f.1/n1 g both diverge,
but f.1/n C .1/n1 g, f.1/n .1/n1 g, and f.1/n =.1/n1 g are all constant
sequences and therefore converge.
Before presenting some examples, we note three corollaries.
Corollary 2.3.8. Using the notation of the theorem, we have
lim .sn tn / D K L:
n!1
Proof. From parts (1) and (2) of the theorem, we infer that
lim .sn tn / D lim .sn C .1/tn /

n!1 n!1
D K C .1/L
D K L:
t
u
Corollary 2.3.9. Let sn ; tn ; K; L be as in the statement of the theorem.
1. If sn tn for all n, then K L.
2. If sn tn for all n, then K L.
Proof. 1. By Corollary 2.3.8 we have
lim .sn tn / D K L:
n!1
If K L < 0, then, by Proposition 2.3.5(2), there would be an integer N such

that
K L
sn tn < <0
2
for all n N . This contradiction shows that K L 0.
2. The proof of part (2) is similar.
t
u
Putting ftn g equal to a constant sequence fcg yields the following corollary.
48 2 Sequences
Corollary 2.3.10. Let fsn g be a real sequence converging to K, and let c be a real
constant.
1. If sn c for all n, then K c.
2. If sn c for all n, then K c.
Remark. It should not be thought that if sn > c for all n then K > c. For instance,
1=n ! 0 as n ! 1, but 1=n > 0 for all n > 0. Similarly, if sn < c for all n, then
we can conclude only that K c.
Example 2.3.1. Let
4n.4n C 3/
sn D
.4n C 1/.4n C 2/
for each n 2 N. Dividing both numerator and denominator by n2 , we obtain

4 4 C n3
sn D :
4 C n1 4 C n2
As
1
lim D 0;
n!1 n
we may apply Theorem 2.3.7 to find that
4.4 C 0/
lim sn D D 1:
n!1 .4 C 0/.4 C 0/
4
Example 2.3.2. For every positive integer k we have
p p p p
j1 kj D k1< k C 1 D j1 C kj;
and so

1 pk

p < 1:
1 C k
Using the result of Example 2.2.4, we therefore find that

p !n
1 k
lim p D 0:
n!1 1C k
Applying an argument similar to that of theppreceding example, we conclude that

the sequence in Example 2.1.8 converges to k. 4
Example 2.3.3. Let a and r be real numbers and suppose that jrj < 1. Let
X
n
1 r nC1
sn D rj D
j D0
1r
for all n. By Example 2.2.4 and Theorem 2.3.7, we deduce that
1
lim sn D :
n!1 1r
4
Moreover if p denotes a polynomial and fsn g is a sequence converging to L, then
Theorem 2.3.7 shows that
lim p.sn / D p.L/:

n!1
If p and q are polynomials such that q.L/ 0 and q.sn / 0 for all n, then it also
follows that
p.sn / p.L/
lim D :
n!1 q.sn / q.L/
We show next that the study of a complex sequence can be reduced to the study
of two real sequences.
Let fsn g be a complex sequence and suppose that sn D an C ibn for all n, where
each an and bn is real. Then the sequences fan g and fbn g are called the real and
imaginary parts, respectively, of fsn g. They satisfy the following theorem.
Theorem 2.3.11. A complex sequence converges to a number L if and only if its
real and imaginary parts converge, respectively, to the real and imaginary parts
of L.
Proof. Let sn D an C ibn for all n, where each an and bn is real. If the sequences
fan g and fbn g converge to A and B, respectively, then fsn g converges to A C iB, by
Theorem 2.3.7.
Conversely, suppose that
lim sn D L D A C iB;
n!1
where A and B are real. We must show that fan g and fbn g converge to A and B,
respectively. Choose " > 0. There exists N such that jsn Lj < " for all n N .
Recalling that jRe .z/j jzj for all z 2 C, we find that
jan Aj jsn Lj < "
for all n N . Similarly, jbn Bj < " for all n N , and the theorem follows. t
u
50 2 Sequences
Corollary 2.3.12. If a complex sequence fsn g converges to a number L, then fsn g

converges to L.
The following algebraic result is also frequently used in calculations.
Theorem 2.3.13. Let fsn g be a sequence of nonnegative real numbers converging
to L. Then
lim s 1=m D L1=m

n!1 n
for every positive integer m.

Proof. Note that L 0 by Corollary 2.3.10.
Suppose first that L D 0. For every " > 0 there exists N such that sn D jsn j < "m
1=m 1=m
for all n N . Thus sn < " for all such n, and it follows that sn ! 0 as n ! 1.
Suppose therefore that L > 0, and choose " > 0. There is an integer N such
1=m
that jsn Lj < " for all n N . For each such n define an D sn > 0, and let
b D L1=m > 0. Then for all n N it follows from Theorem 1.5.5 that
X
m1
anm b m D .an b/ anj b mj 1
j D0
D .an b/Kn
even if an D b, where
X
m1
Kn D anj b mj 1
j D0
X
m1
D b m1 C anj b mj 1 :
j D1
Now Kn b m1 > 0 since an b mj 1 > 0 for each j > 0. Therefore
j
jsn1=m L1=m j D jan bj

janm b m j
D
Kn
jsn Lj

b m1
"
< m1 ;
b
as required. t
u
Remark. If sn > 0 for all n and L > 0, then the theorem also holds for each negative
integer m, for if m < 0, then m > 0 and
1 1
sn1=m D 1=m
! D L1=m :
sn L1=m
Example 2.3.4. Using Theorem 2.3.13, we find that

p p nC1n
lim n C 1 n D lim p p
n!1 n!1 nC1C n
1
D lim p p
n!1 nC1C n
q
1
n
D lim q
n!1
1C 1
n
C1
0
D p
1C0C1
D 0:
4
Example 2.3.5. For all integers n > 1 we have
1
1
D n !0
n1 1 1
n
as n ! 1. Hence
1
lim p D0
n!1 n1
by Theorem 2.3.13. 4
We conclude this section by showing that taking an arithmetic mean does not
alter the limit. In other words, the limit of a convergent sequence is also the limit of
the arithmetic mean of the terms of the sequence. This result is due to Cauchy. We
will give a corresponding result for geometric means later.
Theorem 2.3.14. If limn!1 sn D L, then
1X
n
lim sj D L:
n!1 n
j D1
52 2 Sequences
Proof. Putting tn D sn L for all n 2 N, we find that

0 1
1X 1X 1 @X X 1X
n n n n n
sj D .tj C L/ D tj C LA D tj C L:
n j D1 n j D1 n j D1 j D1
n j D1
It therefore suffices to show that
1X
n
lim tj D 0:
n!1 n
j D1
Since
lim tn D lim .sn L/ D 0;

n!1 n!1
for each " > 0 there is a positive integer N such that jtn j < " for each n N . For
all n N we also have

n
1 X 1X
n

tj jtj j
n n j D1
j D1
0 1
N 1
1 @X X n
D jtj j C jtj jA :
n j D1 j DN
Since 1=n ! 0, we may choose n so large that

N 1
1X
jtj j < ":
n j D1
For large enough n it follows that

X N 1
1 n 1X 1 X
n
t jt j C jtj j
n j
n j D1
j
n j DN
j D1
1 X
n
< "C "
n j DN
n .N 1/
D "C "
n
2";
and the proof is complete. t

u
2.4 Subsequences 53
Exercises 2.3.
1. Find the limits
n p of othe following sequences:
np o
n1 nC.nC1/i
(a) p ; (g) ;
n
nC1
o
q nC3
1 n
p p
(b) nC3i
n
C 1Ci
; (h) n C k n where k 0;
in n n
p o
(c) nC2i
; (i) 2n
nC2
C inC1n ;
nCi n np o
(d) nC1
, (j) n2 C 1 n ;
n o p p p
i n
(e) 1
4
C 2
; (k) . n C 1 n/ n ;
n o n p o
.2n1/
(f) .2nC1/
; (l) n .n C a/.n C b/ .
2. Suppose that limn!1 xn D L > 0. Show that there is a number N such that
9L < 10xn < 11L
for all n N . Obtain a similar result for L < 0.
2.4 Subsequences
Roughly speaking, a subsequence of a sequence is obtained by discarding some

terms. Recall from the definition of an increasing function that a real sequence fsn g
is increasing if sn < snC1 for all n. In order to construct a subsequence ftn g of a
sequence fsn g, we use an increasing sequence of positive integers to pick out the
terms of fsn g that are to appear in ftn g. Thus we define a sequence ftn g to be a
subsequence of a sequence fsn g if there is an increasing sequence fkn g of positive
integers such that tn D skn for all n.
Example 2.4.1. Let fsn g be the sequence f1=n2 g. Its terms are
1 1 1 1 1 1 1
1; ; ; ; ; ; ; ; : : : :
4 9 16 25 36 49 64
Then fs2n g is the subsequence
1 1 1 1
; ; ; ;:::;
4 16 36 64
whereas fs2n1 g is the subsequence
1 1 1
1; ; ; ; : : : :
9 25 49
4
54 2 Sequences
We show next that subsequences enjoy the same convergence behavior as the
parent sequence.
Theorem 2.4.1. If a sequence converges to a number L, then so does each of its
subsequences.
Proof. Given that fsn g is a sequence that converges to L, for every " > 0 there
exists an integer N such that jsn Lj < " for all n N . If fskn g is a subsequence
of fsn g, then jskn Lj < " whenever kn N . Since fkn g is an increasing sequence
in N, we have kn kN N for all n N . For all such n it therefore follows that
jskn Lj < ", and the proof is complete. t
u
This theorem provides a useful test for divergence of a sequence.
Corollary 2.4.2. Any sequence possessing subsequences that converge to distinct
limits must be divergent.
Example 2.4.2. Let sn D i n for all n. Since s4n D 1 and s4nC1 D i for all n, the
sequence fsn g has subsequences converging to distinct limits and hence diverges.
4
Example 2.4.3. The sequence

1
C .1/n
n
has subsequences

1
C1
2n
and

1
1
2n C 1
that converge to 1 and 1, respectively. The sequence is therefore divergent. 4

Theorem 2.4.3. Let fsn g be a sequence. If the subsequences fs2n g and fs2nC1 g
converge to the same number L, then so does fsn g.
Proof. Choose " > 0. There is an integer N1 such that
js2n Lj < "
for all n N1 . Similarly, there is an N2 such that
js2nC1 Lj < "

2.5 The Sandwich Theorem 55
for all n N2 . Now let
N D maxf2N1 ; 2N2 C 1g:
For all n N it follows that
jsn Lj < ":
t
u
Exercises 2.4.
1. For all n 2 N let
(
n
if n is divisible by 3;
xn D n C1
2
n
12 otherwise.
Show that limn!1 xn D 0.

2. Show that the sequence

n1 n
i
nC1
is divergent.
3. Test the convergence of the following sequences:
n n
o
(a) 1C.1/
2Ci n
;

(b) .1/ 1 C n1 .
n
2.5 The Sandwich Theorem
Sometimes a real sequence is flanked by two sequences that are known to converge
to the same limit. It is natural to expect that the given sequence also converges to
that limit. In this section we confirm that expectation. We begin with the following
special case.
Lemma 2.5.1. Let fan g and fbn g be real sequences such that 0 an bn for all
n. If limn!1 bn D 0, then limn!1 an D 0.
Proof. Choose " > 0. There is an integer N such that jbn j < " for all n N . For
each such n it follows that jan j jbn j < ". t
u
56 2 Sequences
Theorem 2.5.2 (Sandwich Theorem). Let fan g, fbn g, and fcn g be real sequences
such that an bn cn for all n. If
lim an D lim cn D L;
n!1 n!1
then
lim bn D L:
n!1
Proof. The hypotheses show that
0 bn an cn an
for all n, and that
lim .cn an / D L L D 0:
n!1
The previous lemma therefore shows that
lim .bn an / D 0:
n!1
Thus
lim bn D lim .bn an C an /

n!1 n!1
D lim .bn an / C lim an

n!1 n!1
D 0CL
D L:
t
u
Example 2.5.1. We shall show that
lim n1=n D 1:
n!1
Let
sn D n1=n 1
for all n > 0. It suffices to prove that the sequence fsn g is null. Note that sn 0 for
all positive n. Therefore
n.n 1/ 2
n D .sn C 1/n > sn
2
for all n > 0, by Eq. (1.3). Thus
2
sn2 <
n1
for all n > 1, and so
p
2
0 sn 0.

Let us begin with the case where b 1. Fix an integer n b. Then
1 b 1=n n1=n :
The required result follows in this case by using the previous example and the
sandwich theorem.
For b such that 0 1. Hence
1=n
1 1
lim D lim D 1:
n!1 b 1=n n!1 b
Therefore the desired result follows in this case as well. 4

Example 2.5.3. Let 0 < x < 1. We know from Example 2.2.4 that
lim x n D 0:
n!1
We now prove this result using the sandwich theorem.

58 2 Sequences
Put
1x
pD > 0:
x
Then xp D 1 x, so that x.1 C p/ D 1; hence
1
xD :
1Cp
Using inequality (1.2), we conclude that
1 1 1
0 < xn D < :
.1 C p/ n 1 C np np
Since
1
lim D 0;
n!1 np
the desired result follows by the sandwich theorem.

If z is a complex number such that jzj < 1, it follows that
lim jzn j D lim jzjn D 0:

n!1 n!1
Therefore
lim zn D 0
n!1
by Proposition 2.3.2(2). 4
Example 2.5.4. Given real numbers x1 ; x2 ; : : : ; xk , define
0 11=n
X
k
sn D @ jxj jn A
j D1
for all positive integers n. It is easy to show that the sequence fsn g converges to
M D maxfjx1 j; jx2 j; : : : ; jxk jg:
Note first that
X
k
Mn jxj jn kM n :
j D1
The result now follows from the sandwich theorem applied to the inequalities
M D .M n /1=n sn .kM n /1=n D k 1=n M
upon noting that
lim k 1=n M D M:
n!1
4
We conclude this section with more applications of the sandwich theorem.
Theorem 2.5.3. Suppose that fsn g and ftn g are sequences such that limn!1 sn D 0
and ftn g is bounded. Then
lim sn tn D 0:
n!1
Proof. Since ftn g is bounded, there exists M such that jtn j < M for all n. Hence
0 jsn tn j M jsn j ! 0 as n ! 1. Therefore jsn tn j ! 0 as n ! 1, by the
sandwich theorem, and the result follows. t
u
Theorem 2.5.4. Let fsn g be a sequence of nonzero real numbers and suppose that

snC1
lim D L < 1:
n!1 s
n
Then
lim sn D 0:
n!1
Proof. Taking
1L
"D > 0;
2
we find that there is an integer N such that

snC1
L < 1 L
s 2
n
for each n N . For each such n it follows that
jsnC1 j 1L 1CL

<LC D ;
jsn j 2 2
60 2 Sequences
and so
1CL
0 < jsnC1 j < jsn j:
2
An inductive argument therefore shows that
k
1CL
0 < jsN Ck j < jsN j
2
for every positive integer k. Now

k
1CL
lim D0
k!1 2
as .1 C L/=2 < 1. The sandwich theorem therefore implies that
lim jsN Ck j D 0;
k!1
and the result in question follows immediately. t

u
Example 2.5.5. Let sn D n =c for all positive integers n, where p is any rational
p n
number and c > 1. Then

snC1
0 <
sn
.n C 1/p c n
D p
c nC1 n
p
1 nC1
D
c n

1 1 p
D 1C
c n
1
!
c
as n ! 1. As 1=c < 1, the sequence fsn g converges to 0. 4

Example 2.5.6. Let b > 0 and sn D b n =n for all nonnegative integers n. Then
snC1 b nC1 n b
0< D n D :
sn .n C 1/ b nC1
As the sequence fb=.n C 1/g converges to 0, so does fsn g. 4

Remark. The condition that

snC1

s <1
n
for all n does not suffice to ensure that limn!1 sn D 0. For example, let
n
sn D
n1
for all n > 1. Then
1
lim sn D lim D 1;
n!1 n!1 1 1
n
but
snC1 nC1 n1 n2 1

D D < 1:
sn n n n2
The sandwich theorem can be used to prove that the limit of a convergent
sequence is also the limit of the geometric mean of the terms of the sequence. First
we establish the following lemma.
Lemma 2.5.5. Let fsn g be a sequence of positive numbers. For all integers n > 0
define
X
n
1
Hn D n ;
s
j D1 j
0 11=n
Y
n
Gn D @ sj A ;
j D1
and
1X
n
An D sj :
n j D1
Then Hn Gn An for all n, with equality holding for a given n if and only if
s1 D s 2 D : : : D s n .
Proof. Certainly, G1 D A1 . In order to prove that Gn An for a given n > 1,
we use induction on the number k of subscripts j n for which sj An . If
k D 0, then sj D An for all j n. In this case Gn D .Ann /1=n D An , as desired.
Assume therefore that k > 0 and that the result holds whenever fewer than k of the
62 2 Sequences
first n terms of a sequence are different from the arithmetic mean of those terms.
Since k > 0, we may assume without loss of generality that s1 < An , the argument
being similar if s1 > An . As An is the average of s1 ; s2 ; : : : ; sn , we may therefore
assume, again without losing generality, that s2 > An . The n numbers An ; s1 C
s2 An ; s3 ; s4 ; : : : ; sn have the same average An as s1 ; s2 ; : : : ; sn since they have the
same sum, but fewer than k of them are different from An . It therefore follows from
the inductive hypothesis that
.An .s1 C s2 An /s3 s4 : : : sn /1=n An :
But
An .s1 C s2 An / s1 s2 D An s1 C An s2 A2n s1 s2
D .An s1 /.s2 An /
> 0;
and so
s1 s2 < An .s1 C s2 An /:
Therefore
Gn D .s1 s2 : : : sn /1=n
< .An .s1 C s2 An /s3 s4 : : : sn /1=n
An ;
as required.
It follows that
0 11=n
Yn
1 X n
0<@ A 1 1
;
s
j D1 j
n s
j D1 j
equality holding if and only if s1 ; s2 ; : : : ; sn are equal. Hence, taking reciprocals,

0 11=n 0 0 111=n 0 11=n
X
n Y n Y n Yn
1 1
n 1 @ A D @1 @1 s j AA D @ sj A ;
j D1
s j j D1
s j j D1 j D1
and we have proved that Hn Gn . Once again, equality holds for n if and only if
s1 ; s2 ; : : : ; sn are equal. t
u
The proof that Gn An was suggested to us by J. Hudson and is a modification

of an argument due to Ehlers. The numbers Hn , Gn , and An are, respectively, the
harmonic, geometric, and arithmetic means of the first n terms of the sequence.
Theorem 2.5.6. Let fsn g be a sequence of positive numbers that converges to some
number L > 0. Then
0 11=n
Y
n
lim @ sj A D L:
n!1
j D1
Proof. With the notation of Lemma 2.5.5, Theorem 2.3.14 shows that An ! L as
n ! 1. Moreover
1 1
lim D ;
n!1 sn L
and so the same theorem shows that
1X 1
n
1 1
D !
Hn n j D1 sj L
as n ! 1; hence limn!1 Hn D L. By the sandwich theorem it follows that

0 11=n
Y
n
lim @ sj A D L:
n!1
j D1
t
u
Exercises 2.5.
1. Test the sequence

2n in
C n
n 2
for convergence.
1=n
n C inwn
is convergent if jwj < 1.

3. Suppose that jan j bn for all n and that
lim bn D 0:
n!1
64 2 Sequences
Show that
lim an D 0:
n!1

n 1=n
2
converges.
f.2n C 3n /1=n g
converges to 3.
6. Show that
1 X 1=j
n
lim j D 1:
n!1 n
j D1
7. Suppose that xn > 0 for all n 0 and that

xnC1
lim D L > 0:
n!1 xn
Show that
lim xn1=n D L:
n!1
Is the converse true? (Hint: Use Theorem 2.5.6.)

8. For all k 2 N show that
!1=n
nk kk
lim D ;
n!1 n .k 1/k1
where 00 D 1. (Use question 7.)

9. Test the convergence of the following sequences:
nP o
n n
(a) j D0 n2 Cj ;
nP o
n j
(b) j D1 n2 Cj ;
nP o
n n2
(c) j D0 n3 C2nCj .
2.6 The Cauchy Principle 65
2.6 The Cauchy Principle
Sometimes it is possible to establish that a sequence is convergent without actually

finding its limit. We shall show that a sequence is convergent if and only if its terms
are ultimately close to one another even when they are not consecutive. This idea
leads to the concept of a Cauchy sequence.
A sequence fsn g is called a Cauchy sequence if for every " > 0 there is an
integer N such that
jsn sm j < "
whenever m N and n N . Of course, " may be replaced in this definition by c"

for any c > 0.
Cauchy sequences are characterized in the next theorem.
Theorem 2.6.1. A sequence fsn g is Cauchy if and only if for all " > 0 there exists
an integer N such that
jsn sN j < "
for all n N .
Proof. The necessity follows immediately by taking m D N in the definition. To
prove the sufficiency of the stated condition, choose " > 0. By hypothesis there is
jsn sN j < "
for all n N . Choosing m N and n N , we have
jsn sm j D jsn sN C sN sm j
jsn sN j C jsm sN j
< 2":
Therefore fsn g is a Cauchy sequence by definition. t

u
We now establish some properties of Cauchy sequences.
Theorem 2.6.2. Every Cauchy sequence is bounded.
Proof. Let fsn g be a Cauchy sequence and choose " > 0. By Theorem 2.6.1 there is
jsn sN j < "

66 2 Sequences
for all n N . The proof is now completed by the argument of Theorem 2.3.1,
replacing L by sN . t
u
Since the terms of a Cauchy sequence are ultimately close to one another, it is
intuitively clear that if some subsequence of the sequence converges to a number L,
then the whole sequence must converge to L.
Theorem 2.6.3. If fsn g is a Cauchy sequence with a subsequence that converges to
L, then fsn g also converges to L.
Proof. Let fskn g be a subsequence that converges to L and choose " > 0. There
exist N1 and N2 such that
jskn Lj < "
for each n N1 and
jsn sm j < "
whenever m N2 and n N2 . Take
N D maxfN1 ; N2 g:
Then for each n N we have n N1 and kn n N2 since fkn g is an increasing

sequence of positive integers. Hence
jsn Lj D jsn skn C skn Lj

jsn skn j C jskn Lj
< 2";
and so
lim sn D L:
n!1
t
u
Theorem 2.6.4. Every convergent sequence is Cauchy.
Proof. Let fsn g be a sequence that converges to L and replace sN by L in the proof
of Theorem 2.6.1. t
u
But is it true that every Cauchy sequence converges? It is possible to find a
rational Cauchy sequence that converges to an irrational number. We have already
p
seen this phenomenon in Example 2.3.2. (It is shown in Exercises 1.1 that 2 is
irrational.) However, every real Cauchy sequence does converge to a real number. In
fact, one way of defining real numbers is via Cauchy sequences of rational numbers,
as is done in [10]. The first step is to define two sequences fsn g and ftn g to be
equivalent if the sequence fsn tn g is null. It can be shown that this notion defines
an equivalence relation on the set of all rational Cauchy sequences, and the real
numbers are defined as the corresponding equivalence classes. The convergence of
real Cauchy sequences and the fact that every real number is the limit of a rational
Cauchy sequence can be established as consequences. The property that all real
Cauchy sequences converge to real numbers is referred to as the completeness of
the real number system.
Let x and y be real numbers such that x < y. From the definition of the
real number field we see that .x C y/=2 is a limit of a Cauchy sequence of
rational numbers. Hence the interval
p .x;
p y/ contains at least one rational number.
Consequently, the interval .x= 2; y= 2/ contains a rational number r. We may
assume that r 0: If x < 0 <py, then replace y by 0. Thus the interval .x; y/
contains the irrational number r 2. We conclude that every open interval .x; y/,
where x < y, contains at least one rational number and at least one irrational
number. This condition is described as the density property of the real number
system.
The completeness property of real numbers is equivalent to another property
known as the supremum property. In order to explain it, we need some additional
definitions. A set S of real numbers is said to be bounded above if there exists a
real number b such that s b for all s 2 S . Any number b with this property is
called an upper bound of S . For example, 0 is an upper bound for the set of all
negative real numbers. Any positive number is also an upper bound for that set. An
upper bound b of S is the least upper bound or supremum of S if b c for every
upper bound c of S . The least upper bound of S , if it exists, is unique, for if b and
c are both least upper bounds of S , then b c and c b. It is denoted by sup S .
For example, if S is the set of all negative numbers, then sup S D 0. Sets that are
bounded below, lower bounds of such sets, and greatest lower bounds or infima
are defined analogously. The greatest lower bound of a set S , if it exists, is unique
and is denoted by inf S . If S is the range of a bounded sequence fsn g, then its least
upper bound is also denoted by supfsn g and its greatest lower bound by inffsn g.
The supremum property alluded to above asserts that every nonempty set of real
numbers that is bounded above has a least upper bound. The proof of the equivalence
of the completeness and supremum properties is our next goal. Before tackling it,
however, we insert an example of a Cauchy sequence.
Example 2.6.1. Let a and b be complex numbers. Define s0 D a, s1 D b, and
sn C sn1
snC1 D
2
for all n > 0. If a D b, then sn D a for all n. Suppose a b. It is not hard to show
by induction that
ba
snC1 sn D .1/n (2.3)
2n
68 2 Sequences
for all n 0. For every m and n such that n > m, the use of the telescoping property
combined with the triangle inequality shows that

X
n1

jsn sm j D .sj C1 sj /
j Dm
X
n1
jsj C1 sj j
j Dm
Xn1
jb aj
D
j Dm
2j
jb aj X 1
n1
D
2m j Dm 2j m
jb aj
< ;
2m1
since
X
n1
1 X
nm1
1 1 2nm
1
1
D D < 1 D 2:
j Dm
2j m j D0
2j 1 12 2
Example 2.2.4 shows that jb aj=2m1 ! 0 as m ! 1, and so for each " > 0 we
may choose N such that
jb aj
< ":
2N 1
For all m N and n > m N , it follows that
jb aj jb aj
jsn sm j < N 1 < ":
2m1 2
Therefore fsn g is a Cauchy sequence. By the completeness property it converges to

some number L.
In order to find L, one might attempt to make use of the recurrence relation that
defines the sequence, but such an attempt leads to the equation
LCL
LD D L;
2
which does not give any information about L. Instead, we use induction to show
from Eq. (2.3) that for all k 0,
X
k1
ba
s2k D a C
j D0
22j C1
ba X 1
k1
D aC
2 j D0 4j
!
ba 1 4k
1
D aC
2 1 14

ba 4 1
D aC 1 k
2 3 4
2.b a/
!aC
3
a C 2b
D
3
as k ! 1. By Theorem 2.4.1 it follows that
a C 2b
lim sn D :
n!1 3
4
Theorem 2.6.5. The completeness property for real numbers implies the supremum
property.
Proof. Let S be a nonempty set of real numbers that is bounded above. We must
show that S has a least upper bound. If an element s 2 S happens to be an upper
bound of S , then s is in fact the least upper bound of S . We may therefore assume
that no member of S is an upper bound of S .
Since S is nonempty and bounded above, we may choose a0 2 S and an upper
bound b0 of S . By assumption, a0 is not an upper bound of S . If .a0 C b0 /=2
is an upper bound of S , then let a1 D a0 and b1 D .a0 C b0 /=2; otherwise let
a1 D .a0 C b0 /=2 and b1 D b0 . In both cases b1 is an upper bound of S , but a1 is
not. Proceeding inductively, we obtain a nested sequence
a0 ; b0 a1 ; b1 a2 ; b2 : : :
of closed intervals such that no an is an upper bound of S , but each bn is.

We shall establish that the sequence fbn g is Cauchy. First, an inductive argument
shows that
b0 a0
bn an D
2n
70 2 Sequences
and
b0 a0
0 bn bnC1
2nC1
for all n 0. For each positive integer p it therefore follows from the telescoping
property that
jbnCp bn j D bn bnCp
X
p1
D .bnCj bnCj C1 /
j D0
X
p1
b0 a0

j D0
2nCj C1
b0 a0 X 1
p1
D nC1
2 j D0
2j
b0 a0
< ;
2n
since
X
p1
1 1 21p
D < 2:
j D0
2j 1 12
Example 2.2.4 shows that 1=2n ! 0 as n ! 1. Given " > 0, we may therefore
choose N such that
b0 a0
< ":
2N
For all n N and all positive integers p it follows that
b0 a0 b0 a0
jbnCp bn j < n
< ";
2 2N
and we infer that fbn g is indeed a Cauchy sequence. A similar argument demon-
strates that fan g is also a Cauchy sequence.
By the completeness principle fan g and fbn g converge to some numbers a and b,
respectively. For all n we have an anC1 < bnC1 bn , and so
an a b bn ;
by Corollary 2.3.10. Therefore bn an b a. If b > a, then there is an n such

that
b0 a0
bn an D b. Since bn ! b, we
may use Proposition 2.3.4 to choose n such that bn < s, in contradiction to the fact
that bn is an upper bound of S . Thus s b for all s 2 S , and so b is indeed an upper
bound of S .
Next, choose t t , we may choose n such that t < an . As an
is not an upper bound of S , neither is t . Thus b is the least upper bound of S , and
the proof is complete. t
u
Corollary 2.6.6. Any nonempty set of real numbers that is bounded below has a
greatest lower bound.
Proof. Let S be a nonempty set that is bounded below, and apply the supremum
property to the set fs j s 2 S g. t
u
We now know that the completeness property implies the supremum property.
We prove next that the converse is also true. Throughout this discussion we
therefore assume that the supremum property holds. First we show that, under this
assumption, bounded real sequences always possess convergent subsequences.
Theorem 2.6.7. Every bounded sequence of real numbers has a convergent
subsequence.
Proof. Let fsn g be a bounded sequence of real numbers. If its range is a finite set,
then some term b is repeated infinitely many times. In other words, there are positive
integers k1 ; k2 ; : : : such that fkn g is an increasing sequence and skn D b for all n.
Hence limn!1 skn D b.
Suppose therefore that the range Rs of fsn g is infinite. Since the sequence is
bounded, we may choose m and M such that m < sn < M for all n. Let V be the
set of all x 2 m; M / such that Rs \ .x; M / is infinite. Certainly, m 2 V , so that
V ;. But V is also bounded, and so the supremum property implies that V has a
least upper bound b M . We distinguish two cases.
Case 1: Suppose b D M . Since b 1 is not an upper bound of V , there is a
v1 2 V such that b 1 < v1 < M . Thus Rs \ .v1 ; M / is infinite, and so we may
choose k1 such that sk1 2 .v1 ; M /: Similarly, b 1=2 is not an upper bound of V ,
and so there exists v2 2 V such that b 1=2 < v2 < M . Moreover Rs \ .v2 ; M /
is infinite, so that there exists k2 > k1 such that sk2 2 .v2 ; M /. Continuing
72 2 Sequences
inductively, we construct an increasing sequence fkn g of positive integers such

that skn 2 .b 1=n; M / for all n > 0. Thus
1
b < skn < M D b:
n
By the sandwich theorem it follows that
lim skn D b:
n!1
Case 2: If b < M , then we may choose N such that
1
b<bC < M:
N
Thus
1
b<bC <M
N Cn
for all n 0. Furthermore, as b 1=.N C n/ is not an upper bound of V , there

exists un 2 V such that
1
b < un b:
N Cn
Since b is an upper bound of V , we have b C 1=.N C n/ V , and so

1
Rs \ b C ;M
N Cn
is finite since
1
mb<bC < M:
N Cn
But Rs \ .un ; M / is infinite since un 2 V , and as b 1=.N C n/ < un , we infer

that

1 1
Rs \ b ;b C
N Cn N Cn
is infinite. Putting n D 0, we may therefore choose k0 so that
1 1
b < sk0 0 there exists kn > kn1 such that
1 1
b < skn < b C :
N Cn N Cn
It follows from the sandwich theorem that
lim skn D b:
n!1
t
u
Corollary 2.6.8. Every bounded sequence of complex numbers has a convergent
subsequence.
Proof. Let fzn g be a bounded sequence of complex numbers. Then, for each n,
it follows that zn D xn C iyn for some real numbers xn and yn . Using the facts
that jRe .z/j jzj and jIm .z/j jzj for each complex number z, we deduce that
fxn g and fyn g are also bounded. By Theorem 2.6.7 fxn g has a subsequence, fxkn g,
that converges to some number a. Similarly, fykn g has a subsequence, fykln g, that
converges to some number b. The subsequence fxkln g of fxkn g also converges to a,
by Theorem 2.4.1. Hence fzkln g converges to a C ib. t
u
We have proved that if the supremum property holds, then every bounded
sequence contains a convergent subsequence. Since every Cauchy sequence is
bounded, a Cauchy sequence fsn g therefore has a convergent subsequence. It follows
by Theorem 2.6.3 that fsn g converges as well. In other words, the supremum
property implies completeness. For real numbers these two properties therefore
imply each other. Henceforth we assume they both hold, referring the reader to [10]
for a proof.
We now have the following theorem.
Theorem 2.6.9 (Cauchy Principle). A real sequence is convergent if and only if it
is a Cauchy sequence.
In fact, by considering separately the real and imaginary parts of the terms
of a complex sequence, one can show that this principle holds even for complex
sequences. The details are left as an exercise.
Roughly speaking, the Cauchy principle tells us that convergent sequences are
those whose terms are getting closer and closer. In particular, for a sequence to
converge it is necessary that the distance between successive terms diminishes. In
fact, if fsn g is convergent, then
lim .snC1 sn / D 0:
n!1
However, the converse is not true, as we see in the next example.

Example 2.6.2. Let
X n
1
sn D
j D1
j
for all n. We show that fsn g is divergent.

74 2 Sequences
Suppose that limn!1 sn D s. Then
X n
1 1X1
n
s
tn D D !
j D1
2j 2 j D1 j 2
as n ! 1, and
X
n
1 s s
un D D s2n tn ! s D
j D1
2j 1 2 2
as n ! 1. These results cannot both hold: Since

1 1
>0
2j 1 2j
for all j > 0, we have
1
lim .un tn / > u1 t1 D > 0:
n!1 2
Therefore fsn g is indeed divergent. However,
1
snC1 sn D !0
nC1
as n ! 1. 4
Example 2.6.3. The sequence fz g, where z is a complex number, is convergent if
n
and only if jzj < 1 or z D 1.

Proof. We have shown in Example 2.2.4 that if jzj < 1, then the sequence is
convergent. It is certainly so if z D 1.
Conversely, suppose that jzj 1 and z 1. Then
jznC1 zn j D jzjn jz 1j jz 1j > 0:
Therefore znC1 zn does not approach 0 as n ! 1. We conclude that the sequence

is divergent. 4
A sequence is injective if and only if its terms are distinct. The following theorem
concerns the existence of an injective convergent sequence.
Theorem 2.6.10 (Bolzano, Weierstrass). Every bounded infinite set of numbers
contains an injective convergent sequence.
Proof. Any infinite set S of numbers certainly contains an injective sequence. If S

is bounded, then this sequence is also bounded and therefore contains a convergent
subsequence that is necessarily injective. t
u
A number b is a limit point or an accumulation point of a set S if there is
an injective sequence fsn g in S that converges to b. It follows from the Bolzano
Weierstrass theorem that every bounded infinite set of numbers has a limit point.
The following theorem gives an alternative definition of a limit point. For every
" > 0 we define
N" .b/ D N" .b/ fbg:
Theorem 2.6.11. Let S be a set of numbers and b a number. The following two
statements are equivalent:
1. b is a limit point of S ;
2. for each " > 0,
S \ N" .b/ ;:
Proof. Let b be a limit point of S . Then there exists an injective sequence fsn g in S
converging to b. Choose " > 0. There exists N such that
jsn bj < "
for all n N . As fsn g is injective, there exists n N such that sn b. Thus

sn 2 N" .b/, as required.
Conversely, suppose (2) holds. Letting "0 D 1, by hypothesis we may choose
s0 2 S \ N"0 .b/. Thus 0 < js0 bj < 1. Continuing by induction, suppose that k is
a positive integer and that s0 ; s1 ; : : : ; sk1 are distinct real numbers satisfying
1
0 < jsk1 bj <
2k1
and
jsk1 bj < jsj bj
for each nonnegative integer j < k 1. Let
jsk1 bj
"k D > 0:
2
By hypothesis, there exists sk 2 S \ N"k .b/. Thus
jsk bj < "k < jsk1 bj < jsj bj

76 2 Sequences
for each j < k 1. Therefore jsk bj < jsj bj for each j < k, so that sk sj
for each such j . Moreover
jsk1 bj 1
0 < jsk bj < < k:
2 2
We have now defined an injective sequence fsn g in S such that
1
0 < jsn bj <
2n
for each n. By the sandwich theorem, the sequence fsn bg converges to 0, and so
lim sn D b:
n!1
t
u
A subset of C is closed if it contains all its limit points.
Example 2.6.4. Let a and b be real numbers with a < b. We can confirm that
the interval I D a; b satisfies this definition, thereby establishing agreement of
this terminology with that introduced earlier in connection with intervals. Indeed,
choose z D .x; y/ 2 C I . If y 0, then
N" .z/ \ I D ; (2.4)
for each positive " < jyj. Suppose therefore that z is real. If z < a, then Eq. (2.4)
holds for every positive " < a z. If z > b, then apply the same argument with
0 < " < z b. We conclude from Theorem 2.6.11 that I is closed.
A similar argument may be applied to show that the rectangle R D a; b c; d
is a closed set in R2 D C. Here a 0 at least one of the following cases arises:
x D b C r; x D a r; y D d C r; y D c r:
Choose such an r as small as possible, and let " D r=2 > 0. Then N" .z/ \ R D ;,
and so z is not a limit point of R. 4
Exercises 2.6.
1. Show from the definition that the sequence f1=n2 g is Cauchy.
2. Show that if for each " > 0 there exists N such that jxn xN j < " whenever
n > N , then fxn g is a Cauchy sequence.
3.(a) Let fxn g be a sequence and suppose there exists a constant r 2 .0; 1/ such
that
jxnC1 xn j r n
for all n. Show that fxn g is Cauchy.

2.7 Monotonic Sequences 77
(b) Give an example of a divergent sequence fyn g with the property that
lim .ynC1 yn / D 0:
n!1
4. Show that if
jxnC2 xnC1 j rjxnC1 xn j
for some r 2 .0; 1/ and all n, then fxn g is Cauchy.

5. Show that if
lim supfjxj xk j j j > k ng D 0;

n!1
then fxn g is Cauchy.

6. Let xn 0 for all n and suppose that limn!1 xn 0. Prove that
inffjxn j j n 2 Ng > 0:
7. Show that a sequence fxn g is Cauchy if and only if for each " > 0 there exists N
such that
jxn xm j
<"
1 C jxn xm j
whenever n > m N .
2.7 Monotonic Sequences
We now study an important family of sequences whose convergence can often be

determined without any knowledge of their limits, namely, monotonic sequences.
We begin with some examples that illustrate techniques for establishing the
monotonicity of sequences. Note that a sequence fsn g is increasing if and only
if fsn g is decreasing. Similarly, fsn g is nondecreasing if and only if fsn g is
nonincreasing.
Example 2.7.1. Let us show that the sequence fsn g is decreasing, where
3n C 1
sn D
2n 3
for all n 2.
78 2 Sequences
One approach is to show that snC1 sn < 0 for all n 2. This result follows
from the calculation
3.n C 1/ C 1 3n C 1
snC1 sn D
2.n C 1/ 3 2n 3
3n C 4 3n C 1
D
2n 1 2n 3
11
D
.2n 1/.2n 3/
< 0:
An alternative is to demonstrate that snC1 =sn < 1 for all n 2:
snC1 3n C 4 2n 3
D
sn 2n 1 3n C 1
6n2 n 12
D
6n2 n 1
< 1:
4
Example 2.7.2. We show by induction that the sequence fsn g is increasing, where
s1 D 1 and
p
sn D sn1 C 1
for all n > 1. p

Certainly, s2 D 2 > 1 D s1 . Suppose sn > sn1 for some n 2. Then
p p
snC1 D sn C 1 > sn1 C 1 D sn ;
as required. 4
Our next goal is to show that a monotonic sequence is convergent if and only if
it is bounded. We need the following notions concerning unbounded sequences.
Definition 2.7.1. Let fsn g be a real sequence. We write
lim sn D 1
n!1
if for each M there exists N such that sn > M for all n N . We also say that sn
approaches infinity as n approaches infinity, and we write sn ! 1 as n ! 1.
Likewise, we write
lim sn D 1
n!1
if for each M there exists N such that sn < M for all n N . In this case we say
that sn approaches minus infinity as n approaches infinity, and we write sn ! 1
as n ! 1.
Remark 1. Clearly, we may assume that M > 0 in the former definition and that
M < 0 in the latter. We may also replace the inequalities sn > M and sn < M by
sn > cM and sn < cM, respectively, for all c 0.
Remark 2. Let tn D sn for all n. Then tn < M if and only if sn > M .
Consequently sn ! 1 as n ! 1 if and only if tn ! 1 as n ! 1.
Theorem 2.7.1. Let fsn g be a nondecreasing sequence. Then
lim sn D supfsn g
n!1
if fsn g is bounded, and
lim sn D 1
n!1
otherwise.
Proof. Suppose first that fsn g is bounded. Then its range has a least upper bound L.
We must show that sn ! L.
Choose " > 0. Then L " is not an upper bound of fsn g. Therefore there exists
N such that sN > L ". As fsn g is nondecreasing, sn sN for all n N . Thus
sn > L " for all such n. Hence
L " < sn < L C "
for all n N , and we conclude that
lim sn D L:
n!1
Finally, suppose the sequence is not bounded. Since s1 is a lower bound, fsn g
must not be bounded above. Thus for each M > 0 there exists N such that sN > M .
As fsn g is nondecreasing, we have sn sN > M for all n N , and it follows that
lim sn D 1:
n!1
t
u
80 2 Sequences
The next result can be proved in a similar way or by applying the previous result
to the sequence fsn g.
Theorem 2.7.2. Let fsn g be a nonincreasing sequence. Then
lim sn D inffsn g
n!1
if fsn g is bounded, and
lim sn D 1
n!1
otherwise.
Example 2.7.3. In Example 2.7.1 we saw that the sequence fsn g is decreasing,
where
3n C 1
sn D
2n 3
for all n 2. This sequence is bounded below by 0 and therefore converges to its
greatest lower bound. In fact, its limit is easily calculated:
3C 1
3
sn D n
!
2 3
n
2
as n ! 1. 4
Example 2.7.4. In Example 2.7.2 we saw that the sequence fsn g is increasing,
where s1 D 1 and
p
sn D sn1 C 1 (2.5)
for all n > 1. It is also easy to see by induction that the sequence is bounded above
by 2 and therefore converges to some number L. In order to find L, we begin by
taking limits of both sides of Eq. (2.5), thereby obtaining
p
LD L C 1:
Hence
L2 L 1 D 0;
and so
p
1 5
LD :
2
But sn > 0 for all n, and so we conclude that

p
1C 5
lim sn D :
n!1 2
4
Example 2.7.5. In Example 2.1.8 we studied the sequence fsn g, where s1 D 1 and
sn1 C k
sn D (2.6)
sn1 C 1
for all n > 1. Here kpis a positive integer. We saw in Example 2.3.2 that this
sequence converges to k. We now present a different proof. Note that the sequence
is the constant sequence f1g
p if k D 1, and so we assume that k > 1. Moreover
sn > 0 for all n, and s1 < k.
Suppose that
p sn1 C k
k < sn D
sn1 C 1
for some n > 1. Then

p p
k sn1 C k < sn1 C k;
so that
p p p p
. k 1/sn1 < k k D k. k 1/I
hence
p
sn1 < k:
p p p p
Similarly, if sn < pk, then sn1 > k. Since s1 < k, it follows that sn < k if
subsequence fs2n g is bounded below by
n is odd and sn > k if n is even. Thus the p
p
k, whereas fs2nC1 g is bounded above by k. Moreover it is easy to see from the
recurrence relation (2.6) that
sn Ck
snC1 C k sn C1
Ck .k C 1/sn C 2k
snC2 D D sn Ck
D (2.7)
snC1 C 1 sn C1
C1 2sn C k C 1
for all n, and so
.k C 1/sn C 2k 2.k sn2 /

snC2 sn D sn D :
2sn C k C 1 2sn C k C 1
82 2 Sequences
p p
Thus snC2 > sn if sn < k, whereas snC2 < sn if sn > k. Hence fs2n g is
a decreasing subsequence and fs2nC1 g is increasing. Both of these subsequences
therefore converge.
Let L be the limit of the subsequence fs2n g. From Eq. (2.7) we see that
.k C 1/L C 2k
LD ;
2L C k C 1
whence
2L2 C .k C 1/L D .k C 1/L C 2k;

p p
so that L D k. But L 0, and so L p D k. The same argument shows that
thepsubsequence fs2nC1 g also converges to k. Since both subsequences converge
to k, so does the sequence fsn g, by Theorem 2.4.3. 4
In the next example we use a sequence to obtain an algorithm that was used
in Babylon around 1700 BC for finding the square root of a positive number. It is
sometimes known as the divide-and-average method and can also be derived from
Newtons method, which we shall study in Sect. 9.3.
Example 2.7.6. Let k and a1 be positive numbers, and for each n 2 N define

1 k a2 C k
anC1 D an C D n :
2 an 2an
p
We shall show that an ! k as n ! 1.
Certainly, an > 0 for all n. Moreover
an2 2anC1 an C k D 0: (2.8)
This equation must have a real solution for an , and so the discriminant
2
4anC1 4k
2
must be nonnegative. Hence anC1 k, and since anC1 > 0, it follows that
p
anC1 k (2.9)
for all n.
The sequence (excluding the first term) is nonincreasing: For all n > 1, we have
an2 C k k an2
anC1 an D an D 0:
2an 2an
This nonincreasing sequence, which is bounded below, therefore converges to some

p L 0. From Eq. (2.8) we see that L D k. Since L 0, it follows that
2
number
L D k. 4
p
Remark. Define hn D k=an > 0 for all n 2 N. Since an ! k as n ! 1, the
same is true of hn . Moreover
an C hn
anC1 D
2
and
k 2k 2 2
hnC1 D D D D :
anC1 an C k
an
an
k
C 1
an
1
hn
C 1
an
Thus anC1 and hnC1 are the arithmetic and harmonic means, respectively, of an and
hn , so that hnC1 anC1 by Lemmap2.5.5. Note p also that an hn D k for all n. Thus
the geometric mean of an and hn is an hn D k, and both fan g and fhn g converge
to this number. The convergence of fhn g can also be confirmed by the observations
that it is a nondecreasing sequence (because fan g is nonincreasing) and bounded
above by maxfk=a1 ; a2 g: We have h1 D k=a1 and hn an a2 for all n > 1.
We next consider two sequences defined by arithmetic and geometric means.
These rapidly converging sequences were first studied by Lagrange and Gauss
independently in the eighteenth century and by Borchardt in 1888 in relation to
the computation of elliptic integrals. For a good discussion of this approach, see [5].
Example 2.7.7. Let a0 and g0 be positive numbers, and for all n 0 let
an C gn
anC1 D
2
and
p
gnC1 D an gn :
If g0 D a0 , then a1 D g1 D a0 and it follows by induction that an D gn D a0 for

all n. We assume therefore that a0 g0 . From Lemma 2.5.5 we have an gn > 0
for all positive n. Hence
an anC1 gnC1 gn
for all n > 0. Thus fan g is a nonincreasing sequence bounded below by minfa0 ; g1 g
and fgn g is a nondecreasing sequence bounded above by maxfg0 ; a1 g. Both
sequences therefore converge.
84 2 Sequences
Now for all n > 0 we have
0 anC1 gnC1
anC1 gn
an gn
D :
2
By induction we deduce that
0 an gn
a1 g1
: (2.10)
2n1
From the sandwich theorem we conclude that
lim an D lim gn :
n!1 n!1
This common limit is denoted by agm.a0 ; g0 / and called the arithmeticgeometric

mean of a0 and g0 . It is positive, since
lim an g1 > 0:
n!1
It cannot be written in closed form

p butmay be expressed in terms of a certain elliptic
integral. The number 1=agm 2; 1 is known as Gausss constant. Its value is
approximately 0.8346268.
Although inequality (2.10) provides a means of estimating the rate of conver-
gence of the sequences, a better estimate may be obtained by comparing the values
of an gn and anC1 gnC1 . For each n define
n D an gn
p p p p
D an gn an C gn :
Then
nC1 D anC1 gnC1

an C gn p
D an gn
2
p p 2
an gn
D ;
2
so that
p p 2 p p 2
n2 D an C gn an gn
p p 2
D 2 an C gn nC1 :
Therefore nC1 0 if n 0, and since 0 0, we conclude inductively that

n 0 for all n. Thus
nC1 1
2
D p p 2 ;
n 2 an C gn
so that
nC1 1
lim2
D :
n!1 8agm.a0 ; g0 /
n
We leave it to the reader to show that the sequences defined by the harmonic and
geometric means also converge to the same limit. These sequences are given by
2
hnC1 D 1
hn
C 1
gn
and
p
gnC1 D hn gn ;
respectively, for some given positive numbers h0 and g0 , respectively. Their

common limit is denoted by hgm.h0 ; g0 /. In fact,
1
hgm.h0 ; g0 / D :
1
agm ; 1
h0 g0
4
Recall that an empty product (a product with no factors) is defined to be 1. We
use this convention in the next example.
Example 2.7.8. Let

1 n nC1 n
sn D 1 C D
n n
for all n > 0. The sequence fsn g has been presented earlier as Example 2.1.6. We
shall show that it is increasing and bounded above.
86 2 Sequences
For all n > 0 the binomial theorem shows that

!
X n
n 1 j
sn D
j D0
j n
X n
n.n 1/ .n j C 1/
D
j D0
j nj
X n
1 1 2 j 1
D 1 1 1
j D0
j n n n
X n
1

j D0
j
X n
1
1C
j D1
2 1
j
X
n1
1
D 1C
j D0
2j
1 2n
1
D 1C
1 12
< 1C2
D 3:
Hence the sequence fsn g is bounded above by 3. Moreover for all n > 0 we have
X
nC1
1 1 2 j 1
snC1 D 1 1 1
j D0
j nC1 nC1 nC1
X n
1 1 2 j 1
> 1 1 1
j D0
j nC1 nC1 nC1
X n
1 1 2 j 1
1 1 1
j D0
j n n n
D sn ;
and so fsn g is an increasing bounded sequence. Therefore it converges. Its limit

is denoted by e. Since sn s1 D 2 for all n and the sequence is increasing and
bounded above by 3, we find that 2 < e < 3. Indeed we have s2 > s1 D 2, so that
e > 2. Moreover for all n 3 we have
1 X 1
n1
1
sn 1 C 1 C C C
2 6 j D3 2j
X
n1
1 1 1
D 1C j
C
j D0
2 4 6
1
< 1C2
12
35
D :
12
Thus e 35=12 < 3:

More meticulous calculations show that e is approximately 2.718. 4
We now consider the more general sequence
n

x n o nCx n
1C D ;
n n
where x 2 R. For each x 0 we have

x
x n 1 n 1 mx 1 m
1C D 1C n D 1C D 1C ;
n x
m m
where m D n=x. In the case where m is a positive integer, we have

1 m
e D lim 1C :
m!1 m
It therefore seems reasonable to define

x n
e x D lim 1C
n!1 n
for all real x, but first we must show that the limit exists.
Proposition 2.7.3. For all x 2 R the sequence
n x n o
fsn g D 1C
n
is convergent. Moreover if x 0, then the sequence is increasing for large

enough n.
88 2 Sequences
Proof. The sequence converges to 1 if x D 0. Suppose therefore that x 0. Letting

n > jxj, define a1 D 1 and
x
ak D 1 C
n
for each k 2 f2; 3; : : : ; n C 1g. Thus ak 1 D a1 for each k > 1. Moreover
n > jxj x, so that x=n > 1; hence ak > 0 for each k. By Lemma 2.5.5 we
have
x n nC1
1
1
1C D .a1 a2 anC1 / nC1
n
1 x
< 1Cn 1C
nC1 n
x
D 1C I
nC1
hence
nC1
x n x
1C < 1C :
n nC1
We conclude that the sequence fsn g is increasing for n > jxj.

It therefore suffices to show that fsn g is bounded. Choose an integer M jxj.
Then the sequence

M n
1C
n
is increasing for n > M > 0. As nM n, we therefore have

x n x n

1C D 1 C
n n

jxj n
1C
n

M n
1C
n

M nM
1C
nM
M
1 n
D 1C
n
 jxj, it
follows that e x > 0 for all x 2 R. Note also that e 0 D 1 and e 1 D e.
Example 2.7.9. For all positive integers n define
X
n
xj
tn D ;
j D0
j
where x > 0. We shall show that the sequence ftn g converges to e x .

For each n 2 N define
x n
sn D 1 C :
n
A simple modification of the calculation in Example 2.7.8 gives
X n
xj 1 2 j 1
sn D 1 1 1
j D0
j n n n
X
n
xj

j D0
j
D tn :
Moreover ftn g is certainly an increasing sequence. We shall prove it convergent

by showing that it is bounded above by e x . Let m and n be positive integers with
n > m. Then

x n X x j
n
1 2 j 1
1C D 1 1 1
n j D0
j n n n
X m
xj 1 2 j 1
> 1 1 1 :
j D0
j n n n
Hence
x n

e x D lim 1C
n!1 n
Xx
m j
1

2

j 1

lim 1 1 1
n!1
j D0
j n n n
X
m
xj
D
j D0
j
D tm :
90 2 Sequences
Thus we have sm tm e x for all positive integers m, and so it follows from the
sandwich theorem that
lim tn D e x :
n!1
4
Since ftn g is an increasing sequence when x > 0, we find that
ex > 1 C x > 1
for all positive x.

It follows from the definition that e 0 D 1. The result of Example 2.7.9 therefore
holds also when x D 0 if we adopt the convention that 00 D 1. In fact, it holds for
all real x. In order to prove this result, we need the following lemma.
Lemma 2.7.4. Let a1 ; a2 ; : : : ; an be nonzero numbers that are greater than or equal
to 1 and have equal sign. Then
Y
n X
n
.1 C aj / 1 C aj :
j D1 j D1
Proof. The result certainly holds if n D 1. Suppose therefore that n > 1 and that
the lemma holds for sets of fewer than n numbers satisfying the hypotheses. The
hypotheses imply that 1 C a1 0 and a1 aj > 0 for all j > 1. Therefore, using the
inductive hypothesis, we find that
Y
n Y
n
.1 C aj / D .1 C a1 / .1 C aj /
j D1 j D2
0 1
X
n
.1 C a1 / @1 C aj A
j D2
X
n X
n
D 1 C a1 C aj C a1 aj
j D2 j D2
X
n
> 1C aj ;
j D1
as required. t
u
Corollary 2.7.5. For each x 1 and each nonnegative integer n, we have
.1 C x/n 1 C nx:
Proof. We have already seen in Example 1.3.1 that this inequality holds for all
x 0. Suppose that 1 x < 0. The result holds for n D 0 if we take 00 D 1. For
all n > 0 it follows immediately from the lemma by taking aj D x for all j n.
t
u
Theorem 2.7.6. For all positive integers n define
X
n
xj
tn D ;
j D0
j
where x 2 R. Then the sequence ftn g converges to e x .

Proof. Letting
x n
sn D 1 C
n
for all n 2 N, using Lemma 2.7.4 and Theorem 1.5.4, and recalling that the empty
product is defined as 1, we have

X j
n xj X n
x 1 2 j 1
jtn sn j D 1 1 1

j D0 j j D0 j n n n

X j
n 1 2 j 1 x
D 1 1 1 1
j D2 n n n j
X n j 1
Y ! j
k jxj
1 1
j D2
n j
kD1
j 1
!!
X
n X k jxjj
1 1
j D2
n j
kD1
j 1
!
X
n X k jxjj
D
j D2
n j
kD1
X n
j.j 1/ jxjj
D
j D2
2n j
x 2 X jxjj 2
n
D
2n j D2 .j 2/
x 2 X jxjj
n2
D
2n j D0 j
92 2 Sequences
for all n 2. We deduce that
lim jtn sn j D 0;
n!1
for it follows from Example 2.7.9 that the sequence

8 9
<Xn2
jxj j=
: j ;
j D0
converges. Therefore
lim .tn sn / D 0
n!1
and so
lim tn D lim .tn sn C sn / D lim sn D e x :

n!1 n!1 n!1
t
u
We can also show that e is irrational. First let x D 1 and let m and n be positive
integers with n > m. Then e tm > e tn 0 and
X
n
1
tn tm D
j DmC1
j
1 Xn
1
D
.m C 1/ j DmC1 .m C 2/.m C 3/ j
1 X
nm1
1
D
.m C 1/ j D0 .m C 2/.m C 3/ .m C j C 1/
1 X
nm1
1

.m C 1/ j D0 .m C 1/j
1 1
<
.m C 1/ 1 mC1
1
1 mC1
D
.m C 1/ m
1
D :
mm
Thus
1
e tm D e tn C tn tm < e tn C ;
mm
and taking the limit as n ! 1 yields
1
0 < e tm : (2.11)
mm
If e were rational, then we could write e D p=q for some positive integers p
and q. Moreover e is not an integer since 2 < e < 3, and so q > 1. Putting m D q
in (2.11), we obtain
1
0 < q.e tq / < 1:
q
But
qe D p.q 1/
by assumption, and
X q
q
qtq D ;
j D0
j
which is also an integer. We therefore have the contradiction that q.e tq / is an

integer between 0 and 1.
Example 2.7.10. Let sn D n=nn for all n > 0. Then
snC1 .n C 1/ nn
D
sn .n C 1/nC1 n
n
n
D
nC1
1
D n
1 C n1
1
!
e
as n ! 1. Since e > 1, it follows from Theorem 2.5.4 that the sequence fsn g
converges to 0. 4
94 2 Sequences
Example 2.7.11. Since

n
nC1
lim D e;
n!1 n
Theorem 2.5.6 implies that

0 1
Yn j 1=n
j C 1
e D lim @ A :
n!1
j D1
j
But it is easy to show by induction that
Y n
j C1 j .n C 1/n
D
j D1
j n
for all positive integers n, and so

1=n
.n C 1/n
e D lim
n!1 n
nC1
D lim :
n!1 .n/1=n
Thus n is close to .n C 1/n =e n for large values of n. 4

We now show that our definition of e x satisfies at least one of the usual laws
governing exponentiation. First we prove the following useful lemma.
Lemma 2.7.7. For every x 2 R,
x n
1 C 2 D 1:
lim
n!1 n
p
Proof. For every integer n > jxj we have n2 > jxj x, so that x=n2 > 1.
Applying Corollary 2.7.5 and writing m D n2 , we therefore find that
x x n x m 1=n
1C 1C 2 D 1C e 1=n :
n n m
Using Example 2.5.2, we have
x
lim 1C D 1 D lim e 1=n :
n!1 n n!1
The required result now follows from the sandwich theorem. t

u
Corollary 2.7.8. For every real x we have
1
e x D :
ex
Proof. This result follows from the lemma by taking limits of both sides of the
identity
n
x n x n x2
1C 1 D 1 2 :
n n n
t
u
Thus e x e x D 1. This equation is a special case of the following law governing
exponents. Its proof is an adaptation of an argument due to Kemeny [9].
Theorem 2.7.9. For each x; y 2 R we have
e x e y D e xCy :
Proof. The result clearly holds if x D 0 or y D 0, since e 0 D 1. We may therefore

assume that xy 0.
Let z D x y and, for all positive integers n, define
n n n
nCx nCy nCz
sn D
n n n
n
n3 C n2 .x C y C z/ C n.xy C xz C yz/ C xyz
D
n3

xy C xz C yz xyz n
D 1C C 3 :
n2 n
Note that the sequence fsn g converges to e x e y e z , and therefore so does the
subsequence fs2n1 g.
Suppose that xyz > 0. Then
xyz xyz
;
.2n 1/ 2 .2n 1/3
and so
2n1
xy C xz C yz xy C xz C yz C xyz 2n1
1C < s2n1 1 C ;
.2n 1/2 .2n 1/2
96 2 Sequences
since 2n 1 is odd. Therefore Lemma 2.7.7 and the sandwich theorem show that
lim s2n1 D 1:
n!1
Consequently,
e x e y e z D 1;
and so
1
ex ey D D e z D e xCy :
ez
Thus we have confirmed the theorem if xyz > 0.
We now distinguish three cases.
Case 1: Suppose x < 0 and y < 0. Then z > 0, so that xyz > 0. We conclude
that
e x e y D e xCy
in this case.
Case 2: If x > 0 and y > 0, then by case 1 we have
e x e y D e .xCy/ :
Thus
1 1
D ;
ex ey e xCy
and the result follows.
Case 3: If xy < 0, then we may assume without loss of generality that x < 0 < y.
Case 3.1: If x < y, then z < 0, so that xyz > 0.
Case 3.2: If x D y, then
e x e y D e x e x D 1 D e 0 D e xCy :
Case 3.3: If x > y, then z > 0. It therefore follows from case 1 that
e z e y D e x ;
and so
e x e y D e z D e xCy :
t
u
The function given by e x for all real x is called the exponential function and
often denoted by exp. Here are some of its properties.
Theorem 2.7.10. Let fan g be a sequence of real numbers.
1. If limn!1 an D 1, then
lim e an D 1:
n!1
2. If limn!1 an D a, then
lim e an D e a :
n!1
3. If an 0 for all n and limn!1 an D 0, then
e an 1
lim D 1:
n!1 an
Proof. 1. We have already seen that
ex > 1 C x
for all x > 0. Choose M > 1. Since limn!1 an D 1, there exists N such that
an > M 1 > 0 for each integer n N . Thus
e an > 1 C an > M
for all such n, and the result follows.

2. We first prove the result for a D 0, in which case we need to show that
lim e an D 1:
n!1
Since fan g is convergent, it is bounded. Let jan j < M for all n. For all positive
integers m and n we have

m j m j
X an X an
1 D

j D0 j j D1 j
X
m
jan jj

j D1
j
X
m
jan jj 1
D jan j
j D1
j
98 2 Sequences
X m
M j 1
jan j
j D1
.j 1/
X
m1
Mj
D jan j :
j D0
j
Taking limits as m ! 1, we therefore conclude from Proposition 2.3.2(1) that
0 je an 1j jan je M :
Thus the sandwich theorem yields
lim je an 1j D 0;
n!1
whence
lim .e an 1/ D 0
n!1
and the desired result follows.

In the general case, where limn!1 an D a 0, we use the result just
established:
lim e an D lim e a .e an a / D e a lim e an a D e a :

n!1 n!1 n!1
3. As fan g converges, there exists M such that jan j < M for all n. Thus for all
integers m > 1 and n > 0 we have
0 1

1 X m
a
j
1 X m
a
j

@ n
A
1 1 D
n
1
a
n j D0 j an j D1 j

m j 1
X an
D 1
j D1 j

m1
X j
an
D 1
j D0 .j C 1/

m1
X a
j

D
n

j D1 .j C 1/
X
m1
jan jj 1
jan j
j D1
.j C 1/
X
m1
M j 1
< jan j
j D1
.j 1/
X
m2
Mj
D jan j :
j D0
j
We now complete the proof by an argument similar to that used in part (2): Taking
limits as m ! 1 yields
an
e 1

0 1 jan je M ;
a n
from which the desired result is obtained by taking limits as n ! 1.

t
u
Corollary 2.7.11. If fan g is a null sequence of nonzero terms, then
e xCan e x
lim D ex :
n!1 an
Proof. Part (3) of the theorem shows that
e xCan e x e x .e an 1/
lim D lim
n!1 an n!1 an
D ex :
t
u
Theorem 2.7.12. The exponential function is increasing.
Proof. If y > x, then y x > 0, so that e yx > 1; hence
e y D e yx e x > e x
as e x > 0. t
u
We have already shown that every bounded sequence has a convergent subse-
quence. We conclude this section by demonstrating that such a subsequence can be
chosen to be monotonic.
100 2 Sequences
A positive integer m is called a peak index for a real sequence fsn g if sn sm

for all n m. For example, consider the sequence f1 C .1/n =ng. The number 1 is
not a peak index since s1 D 0 and s2 D 3=2. Since
1 3
sn 1 C
n 2
for all n 2, we see that 2 is a peak index. Similarly, 3 is not a peak index but 4 is.
Theorem 2.7.13. Every real sequence has a monotonic subsequence.
Proof. Let P be the set of peak indices for a real sequence fsn g.
Case 1: If P is finite, it must have an upper bound, N . We shall find an increasing
subsequence fskn g of fsn g. Let k1 D N C 1. Then k1 is not a peak index, and so
there exists k2 > k1 such that sk2 > sk1 . For some integer n > 1 we may now
assume the existence of integers k1 ; k2 ; : : : ; kn such that kj < kl and skj < skl
whenever j < l. Since kn > N , kn is not a peak index and so there exists knC1 >
kn such that sknC1 > skn . This observation completes the inductive definition of
the increasing subsequence fskn g of fsn g.
Case 2: If P is infinite, then for each positive integer n we may define kn to be
the nth peak index. It follows that sknC1 skn for all n 2 N, so that fskn g is a
nonincreasing subsequence of fsn g.
t
u
Exercises 2.7.
1. Let fan g and fbn g be sequences of positive numbers, and for all positive integers
m and n define
X
m
pn D aj nj
j D0
and
X
m
qn D b j nj :
j D0
(a) Show that

pn am
lim D :
n!1 qn bm
(b) Show that if bm > am , then there exists N such that pn =qn < 1 for all
n N.
(c) Show that if bm < am , then there exists N such that pn =qn > 1 for all
n N.
2. Show that nthe followingo sequences

n 2areomonotonic for large enough n:
.n1/.nC2/ n 5 5n

(a) .nC1/.n3/
; (b) nC1
; (c) n
.
3. Show that a monotonic sequence is convergent if it has a convergent
subsequence.
4. Show that the following sequences are nondecreasing and bounded above, and
find their limits: p
(a) x0 D 0, xnC1 D 3xn4C1 ; (c) x0 D 1, xnC1 D 3xn C 1;
p
(b) x0 D 0, xnC1 D 2xn3C4 ; (d) x0 D 1, xnC1 D 2xn C 7.
5. Consider a sequence given by
xn2 C 3
xnC1 D
4
for all n > 0.
(a) Show that the sequence is nonincreasing if 1 < x1 < 3.
(b) Show that the sequence is nondecreasing if 0 x1 < 1 or x1 > 3.
(c) For each of the sequences in (a) or (b), find its limit if it converges.
(d) Study the convergence of the sequence for the following cases: x1 D 1,
x1 D 3, x1 < 0.
p
6. Let x1 D b and xnC1 D axn for all n > 0, where a > 0 and b > 0. Show that
fxn g converges and find its limit.
7. Let x1 D a and xnC1 D a C xn2 for all n > 0, where a 0. Discuss the
convergence of the sequence fxn g.
8. Let fan g be increasing and fbn g decreasing, and suppose that
1
0 bn an
2n
p that fan g and fbn g converge to the same limit.

for all n. Show
9. Let xnC1 D a C xn for all n > 0, where a > 0. Discuss the convergence of
the sequence for each x1 > 0.
10. Let x1 D c > 0 and
6.1 C xn /
xnC1 D
7 C xn
for all n > 0. Discuss the convergence of the sequence. (Consider the cases
c 2 and 0 < c < 2 separately.)
11. Consider a sequence given by
3xn C 1
xnC1 D
xn C 3
for all n > 0. Study the convergence of the sequence for the cases x1 1,
1 < x1 < 1, x1 D 1, and x1 > 1.
102 2 Sequences
12. Let a 1, 0 < x1 < a2 , and

p
xnC1 D a a 2 xn
for all n > 0.

(a) Show that 0 < xn < a2 for all n and that fxn g is nonincreasing.
(b) Show that fxn g converges to 0.
(c) Show that
xn
lim D 2a:
n!1 xnC1
13. For all n 0 let
X
2n
1
xn D :
k
kDnC1
Show that fxn g is nondecreasing and converges to a limit between 1=2 and 1.
14. Let x1 D 1 and
4 C 3xn
xnC1 D
3 C 2xn
for all n > 0. Show that the sequence fxn g is nondecreasing and bounded above
by 3=2, andp find its limit.
15. Let x1 D 2 and
q
p
xnC1 D 2C xn
for all n > 0. Prove that the sequence converges.

16. Show that the sequences

2n .n 1/n
.2n/
and

22n1 .n 1/n
.2n/
are monotonic
p andpfind their limits.
17. Let k x1 < 3 k, where k > 0, and for each n > 0 define

1 k
xnC1 D xn C :
2 xn
Show that
p !2n
p p x1 k
xnC1 k 2 k p
2 k
p
and hence that the sequence converges to k.
18. Let h0 and g0 be positive numbers and let a0 D 1= h0 and b0 D 1=g0 . For all
n 0 define
2
hnC1 D ;
1
hn
C 1
gn
p
gnC1 D hn gn ;
an C bn
anC1 D ;
2
and
p
bnC1 D an bn :
Show that hn D 1=an and gn D 1=bn for all n and hence that
1
lim hn D lim gn D :
n!1 n!1 1
agm ; 1
h0 g0
19. Let x1 D 1=2 and y1 D 1, and define

p
xnC1 D xn yn
and
2
ynC1 D 1
xnC1
C 1
yn
for all n 1. Prove that xn < xnC1 < ynC1 < yn for all n and deduce that both
sequences fxn g and fyn g converge to the same limit L, where 1=2 < L < 1.
20. Let x1 > 0 and y1 > 0, and define
x n C yn
xnC1 D
2
and
p
ynC1 D xnC1 yn
104 2 Sequences
for all n 1. Prove that the sequences fxn g and fyn g are monotonic and
converge to the same limit.
( nC1 )
1
1C
n
is nonincreasing and bounded below, and find its limit.
2.8 Unbounded Sequences
In the previous section we saw that a monotonic sequence is convergent if and only
if it is bounded. Specifically, if the sequence fsn g is nondecreasing and unbounded,
then limn!1 sn D 1, and if it is nonincreasing and unbounded, then limn!1 sn D
1. However it may be that a divergent sequence is bounded. An example is
furnished by the sequence f.1/n g. It may also be that limn!1 sn D 1, but
fsn g is not monotonic.
Example 2.8.1. Let
sn D n C 2.1/n
for all n. Then
s2n D 2n C 2;
s2nC1 D 2n C 1 2 D 2n 1;
and
s2nC2 D 2n C 2 C 2 D 2n C 4:
Hence fsn g is not monotonic. But for each n we have
sn n 2;
and so
lim sn D 1:
n!1
4
The next proposition is easy to establish.
2.8 Unbounded Sequences 105
Proposition 2.8.1. Suppose fsn g and ftn g are sequences and there exists N such
that tn sn for all n N .
1. If sn ! 1 as n ! 1, then tn ! 1 as n ! 1.
2. If tn ! 1 as n ! 1, then sn ! 1 as n ! 1.
Proof. The first part is immediate from Definition 2.7.1 and the fact that if, given
any number M , there exists N1 such that sn > M for all n N1 , then tn sn > M
for all n maxfN; N1 g. The proof of the second part is similar. t
u
At this point the reader may be wondering to what extent the symbols 1 and 1
may be treated as if they were numbers. We now attempt to answer that question by
studying the properties of limits of sums and products of unbounded sequences.
For sums we have the following theorem.
Theorem 2.8.2. Let fsn g and ftn g be sequences and L a number. If sn ! 1 and
either tn ! 1 or tn ! L as n ! 1, then sn C tn ! 1.
Proof. Choose a number M , and suppose first that tn ! 1. There exists N such
that sn > M and tn > M for all n N . Then sn C tn > 2M for all such n, as
required.
If tn ! L, then there exists N1 such that jtn Lj < 1 for all n N1 . For each
such n it follows that
tn > L 1:
Moreover there exists N such that
sn > M L C 1
for all n N . For all n maxfN; N1 g we deduce that sn C tn > M . t

u
Thus Theorem 2.3.7(1) may be extended if we write 1 C 1 D 1 C L D 1.
Because of the commutativity of addition of real numbers, we also write
L C 1 D 1. A corresponding theorem may be proved in which sn approaches
1 and tn approaches L or 1, and so we may also write 1 C .1/ D
1 C L D L C .1/ D 1. Extending the rule that x y D x C .y/,
we simplify the left-hand side of this equation to 1 1. Similarly, L C .1/
may be simplified to L 1. On the other hand, we cannot ascribe any meaning to
1 1. We may be tempted to set it equal to 0. However, suppose sn D 2n and
tn D n for all n. Then sn ! 1 and tn ! 1 as n ! 1, but the sequence fsn tn g
does not approach 0 as n ! 1 since sn tn D 2n n D n ! 1.
Let us move on to products.
Theorem 2.8.3. Let fsn g and ftn g be sequences and L a positive number.
If sn ! 1 and either tn ! 1 or tn ! L as n ! 1, then sn tn ! 1.
Proof. Choose M > 0, and suppose that tn ! 1. There exists N such that sn > M
and tn > 1 for all n N . For all such n we deduce that sn tn > M , as required.
106 2 Sequences
Suppose on the other hand that tn ! L. There exists N1 such that tn > L=2
for all n N1 . There also exists N such that sn > 2M=L for all n N . For all
n maxfN; N1 g it therefore follows that
2M L
sn tn > D M:
L 2
t
u
Again we use this theorem as a pretext for extending Theorem 2.3.7, this time by
writing 11 D 1L D L1 D 1, where L > 0. A corresponding result may be
proved in which sn ! 1: Simply apply the theorem to the sequence fsn g. Thus
we also write .1/ 1 D 1 .1/ D .1/ L D L .1/ D 1. Similarly,
replacing tn by tn gives the additional equations 1 .L/ D .L/ 1 D 1
and .1/ .1/ D .1/ .L/ D .L/ .1/ D 1.
On the other hand, we cannot ascribe meanings to 10 or .1/0. For example,
let k 0 and for all n > 0 define sn D kn and tn D 1=n. Then sn approaches 1 or
1 and tn approaches 0, but sn tn D k.
We prepare ourselves for the incorporation of division into this framework by
proving the following theorem.
Theorem 2.8.4. Let fsn g be a sequence of positive numbers.
1. We have sn ! 0 as n ! 1 if and only if 1=sn ! 1 as n ! 1.
2. Similarly, sn ! 1 as n ! 1 if and only if 1=sn ! 0 as n ! 1.
Proof. 1. Suppose first that sn ! 0 as n ! 1, and choose M > 0. There exists N
such that sn < 1=M for all n N . Thus 1=sn > M for all such n, and we have
proved that 1=sn ! 1 as n ! 1.
Conversely, suppose that 1=sn ! 1 as n ! 1, and choose " > 0. There
exists N such that 1=sn > 1=" for all n N . Hence sn < " for all such n, as
required.
2. Apply part (1) to the sequence f1=sn g.
t
u
The following corollary is immediate.
Corollary 2.8.5. Let fsn g be a sequence of nonzero terms. Then jsn j ! 1 as n !
1 if and only if 1=jsn j ! 0 as n ! 1.
Example 2.8.2. Let fan g be a sequence of real numbers, and suppose that
lim an D 1:
n!1
Then
lim .an / D 1:
n!1
2.8 Unbounded Sequences 107
It therefore follows from Theorems 2.7.10(1) and 2.8.4(2) that
1
lim e an D lim D 0:
n!1 n!1 e an
4
Example 2.8.3. Let sn D c for all positive integers n, where c is a fixed real
n
number. By Example 2.6.3 the sequence fsn g is convergent if and only if jcj < 1 or
c D 1. In fact,
lim sn D 1
n!1
if c > 1, for in that case we have 0 < 1=c < 1 so that
1
lim D 0:
n!1 c n
4
Example 2.8.4. Let fsn g be a sequence of nonzero real numbers. Theorem 2.5.4
shows that if

snC1
lim D L < 1;
n!1 sn
then
lim sn D 0:
n!1
We turn now to the case where L > 1. By Proposition 2.3.3 there exist numbers
k > 1 and N such that

snC1

s >k
n
for all n N . Hence
jsnC1 j > kjsn j
for all n N . By induction it follows that
jsN Cp j > k p jsN j

108 2 Sequences
for all positive integers p. Since k > 1, we deduce from the previous example that
k p ! 1 as p ! 1, and so
lim jsN Cp j D 1
p!1
by Proposition 2.8.1. It follows that jsn j ! 1 as n ! 1. 4

Theorem 2.8.4(2) motivates the equation 1=1 D 0. A similar theorem may be
proved in which sn approaches 1, where fsn g is a sequence of negative terms, and
so we also write 1=.1/ D 0. However, we cannot use part (1) to justify writing
1=0 D 1 because of the requirement that fsn g be a sequence of positive terms.
Had fsn g been a sequence of negative terms, we would have been equally tempted
to write 1=0 D 1! By extending the rule that x=y D x.1=y/ whenever y 0,
we may write L=1 D L=.1/ D 0 for each number L. Furthermore, if L > 0,
then we may write 1=L D 1 .1=L/ D 1 and, similarly, .1/=L D 1.
Likewise we write 1=.L/ D 1 and .1/=.L/ D 1. Note that no meaning
is ascribed to such forms as 0=0 or 1=1. This question will be explored later.
Exercises 2.8.
1=n
1. Suppose that the sequence fxn g is increasing and that xn > 1 for all n > 0.
Show that
lim xn D 1:
n!1
(Hint: Prove it by contradiction.)

2. Show that if 0 < a < xn and limn!1 yn D 1, then limn!1 xn yn D 1. Does
the result remain true if we replace the inequalities by xn > 0?
3. Find the following limits:
(a) limn!1 .n2 n3 /;
n
(b) limn!1 2n5C3n .
4. Let fxn g and fyn g be sequences such that
lim xn D lim yn D 1:
n!1 n!1
Give examples to show that
lim .xn yn /
n!1
may be any number or 1.

Chapter 3
Series
3.1 Introduction
The theory of sequences can be combined with the familiar notion of a finite sum
to produce the theory of infinite series. The concept of a series is an attempt
to encapsulate the idea of a sum of infinitely many real or complex numbers.
Applications of series appear in many areas of pure and applied mathematics, and
the study of their properties forms a major part of analysis.
The idea of a series disturbed the ancients. In the fifth century BC, for example,
Zeno argued that it is impossible to walk from one place to another. For the walker
must first travel half the distance, then half the remaining distance, then half the
distance left after that, and so forth. The journey can never be completed, because
after each stage there is still some distance to go. The inference is that motion is
impossible!
Where did Zeno go wrong? He argued that in the first stage of the walk 1/2 of
the total distance must be covered, in the next stage 1/4 of the total distance, in the
third stage 1/8 of the distance, and so on. Zeno was thus attempting to add infinitely
many numbers, and concluded that the sum would be infinite. The absurdity of his
conclusion suggests that the sum of infinitely many numbers should not necessarily
be infinite. But at this juncture it is not really clear precisely what is meant by the
sum of infinitely many numbers. For us to be able to make progress, this notion
must be clarified. Specifically, to resolve Zenos paradox, we need a definition that
enables us to conclude that
1 1 1
C C C D 1: (3.1)
2 4 8

110 3 Series
This problem is resolved through what is called the sequence of partial sums. In
other words, we construct a sequence whose first term is the first of the infinite set
of numbers to be added up, whose second term is the sum of the first two numbers to
be added up, and so forth. In general, the sum of the first n numbers in our infinite
set gives the nth term of the sequence of partial sums. The infinite sum, which is
called a series, is then defined as the limit of the sequence of partial sums, provided
of course that the limit exists.
For example, let us returnP to Zenos paradox. The sum of the first n terms on the
left-hand side of Eq. (3.1) is nj D1 1=2j . In this case the sequence of partial sums is
P
therefore f nj D1 1=2j gn1 . We now show that this sequence indeed converges to 1.
Setting a D 1=2 in Corollary 1.5.6, we have
X n
1 X
n1
1
j
D j C1
j D1
2 j D0
2
1X 1
n1
D
2 j D0 2j
!
1 1 2n
1
D
2 1 12
1
D 1 :
2n
P
Thus the sequence f nj D1 1=2j g converges to 1, as desired.
The use of series troubled many mathematicians in the 18th century. The
construction of an acceptable, rigorous theory took many decades and involved such
mathematicians as Weierstrass, Bolzano, Fourier, Cauchy, Dirichlet, Riemann, and
Dedekind. The theory did not reach its current form until the end of the 19th century.
3.2 Definition of a Series
Let us now formalize the concept of a series. Let fzn gn0 be a sequence of real or
complex numbers, and for each n let
X
n
Sn D zj :
j D0
The sequence fSn g is called a series and is denoted by

1
X
zj (3.2)
j D0
3.2 Definition of a Series 111
or
z0 C z1 C :
We refer to z0 ; z1 ; : : : as the terms of the series. We consider them to be in the

order z0 ; z1 ; : : :, and so we may refer to zj as the .j C 1/th term, for each j 0.
A series is real if all its terms are real. For every series, whether real or not, the
numbers S0 ; S1 ; : : : are the partial sums. More particularly, for each n we may
describe Sn as the partial sum corresponding to zn , or the .n C 1/th partial sum. If
fSn g converges to some number S , then we write
1
X
zj D S:
j D0
In other words,
1
X X
n
zj D lim zj :
n!1
j D0 j D0
As in the case of finite sums, we note that the index j is a dummy variable in the
expression (3.2). Thus, if k is another index, then
1
X 1
X
zj D zk :
j D0 kD0
Motivated by Proposition 1.5.1, we also write

1
X 1
X
zj D zj Cm
j Dm j D0
for each integer m for which zm ; zmC1 ; : : : are defined. It is also evident that if m r,
then
1
X X
r 1
X
zj D zj C zj :
j Dm j Dm j DrC1
In writing this equation we admit the possibility that both series diverge. However,
if one of them converges, then so does the other. When testing these series for
convergence, it therefore suffices to test just one of them.
It is seldom easy to determine the number, if any, to which a given series
converges. However, we do have the following theorem, which follows easily from
our earlier work on finite sums.
112 3 Series
Theorem 3.2.1. If z is a complex number such that jzj < 1, then

1
X 1
zj D ;
j D0
1z
where 00 D 1, but if jzj 1, then the series diverges.

Proof. Let n be a nonnegative integer. If z D 1, then
X
n X
n
zj D 1 D n C 1;
j D0 j D0
and so the series diverges. Suppose therefore that z 1. Then

X
n
1 znC1
zj D ;
j D0
1z
by Corollary 1.5.6. Using the results of Examples 2.2.4 and 2.6.3, we find that the
series converges to 1=.1 z/ if jzj < 1 but diverges if jzj 1. t
u
The series considered in Theorem 3.2.1 is said to be geometric.
Example 3.2.1. The repeating decimal 0:22 : : : can be written as
1
X
0:2 C 0:02 C 0:002 C : : : D 2 aj
j D1
0 1
1
X
D 2@ aj 1A ;
j D0
where a D 0:1 D 1=10. By Theorem 3.2.1 this series converges to

!
1 10 2
2 1 D2 1 D :
1 10
1 9 9
4
The telescoping property can also sometimes be used to deduce the number to
which a given series converges. Specifically, we have the following result.
Theorem 3.2.2. The series
1
X
.zj C1 zj /
j D0
converges if and only if the sequence fzn g converges, and in that case
1
X
.zj C1 zj / D lim zn z0 :
n!1
j D0
Proof. By the telescoping property,
X
n
.zj C1 zj / D znC1 z0
j D0
for every n, and the result follows immediately. t

u
A series of the type contemplated in Theorem 3.2.2 is said to be telescoping or
to telescope. Note that
1
X 1
X
.zj zj C1 / D .zj C1 zj / D z0 lim zn
n!1
j D0 j D0
if the series converges.

1 1 1
D
j.j C 1/ j j C1
for all j > 0, the series

1
X 1
j D1
j.j C 1/
telescopes. It converges to 1 0 D 1. 4
Example 3.2.3. Let a and b be complex numbers, and let z0 D a, z1 D b, and
zn1 C zn2
zn D
2
for all n 2. In Example 2.6.1 we showed that this sequence is Cauchy and
converges to .a C 2b/=3. We now give another way to find its limit.
Note first that
1
zn zn1 D .zn1 zn2 /
2
114 3 Series
for all n 2. By induction it follows that

n1
1
zn zn1 D .b a/;
2
for all n 1, so that
X1 X1
1 n ba 2.b a/
.zn zn1 / D .b a/ D 1 D :
nD1 nD0
2 1 2 3
But the series on the left-hand side is telescoping, and so
2.b a/
lim zn z0 D ;
n!1 3
whence
2.b a/ a C 2b
lim zn D a C D :
n!1 3 3
4
P1
Suppose a series j D1 zj converges to some number S . Then, by definition, the
sequence fSn g also converges to S , where
X
n
Sn D zj
j D1
for each n > 0. So does the subsequence fStn g, where ftn g is any increasing
sequence of positive integers. (See Theorem 2.4.1.) Now
X
tn X
n1 tX
kC1
Stn D zj D zj ;
j D1 kD0 j Dtk C1
where we define t0 D 0. Thus
1
X 1
X X
tkC1
zj D zj : (3.3)
j D1 kD0 j Dtk C1
This result shows that the terms of a convergent series may be grouped together by
means of parentheses without affecting the convergence of the series. In Eq. (3.3)
we group together first the terms
z1 ; z2 ; : : : ; zt1 ;
followed by the terms
zt1 C1 ; zt1 C2 ; : : : ; zt2 ;
and so forth, giving
.z1 C z2 C C zt1 / C .zt1 C1 C zt1 C2 C C zt2 / C :
On the other hand, if some terms are already grouped together, then removal of the
parentheses used to group them may change a convergent series into a divergent
one. For example,
1
X 1
X
.1 1/ D 0 D 0;
j D0 j D0
but the series obtained by removing the parentheses is

1
X
.1/j ;
j D0
which diverges because its partial sums alternate between 1 and 0. However, if the
series with the parentheses removed does converge to a number S , then the original
series must also converge to S , as we have just seen.
There is one important circumstance under which the removal of parentheses
does not alter the convergence of the series: let us suppose that aj 0 for each j .
Assuming that the series on the right-hand side of (3.3) converges when zj D aj
for all j > 0, by definition the sequence fStn g also converges to some number L,
where
X
n
Sn D aj
j D1
for each n > 0. Hence fStn g is bounded above. Now the sequences fSn g and fStn g
are nondecreasing, since aj 0 for each j . Therefore both sequences are bounded
above by L because for each n there exists an integer k such that n tk , and so
Sn Stk L. Being nondecreasing, P the sequence fSn g converges to some number
M L. In other words, the series 1 j D1 aj converges to M . But we also have
L M because Stk Sn M for all n tk . Thus we have proved the following
fact.
Theorem 3.2.3. Suppose that aj 0 for each j . If the series
1
X X
tkC1
aj
kD0 j Dtk C1
P1
converges to some number L, then so does j D1 aj .
116 3 Series
Exercises 3.1. 1. Find the limits of the following series:

P1 2 P1 1
(a) (e) p p
j D0 3j j D0 j C j C1
P1 j 1 P1 4
(b) j D2 j (f) j D3 j 2 4
P1 1 P1 2
(c) j D1 j.j C2/ (g) j D0 j 2 C4j C3
P1 2j C1 P1 6j
(d) j D1 j 2 .j C1/2 (h) j D0 3j C1 2j C1 C3j 2j .
2. Show that if limn!1 xn D 1, then
X1
1 1

j D0
xj xj C1
converges.
3. A decimal number is said to be repeating if there is a finite sequence of digits
that is repeated indefinitely.
(a) Express the repeating decimal 1:2323 : : : as a fraction.
(b) Show that every repeating decimal number is rational.
X n
1
Sn D :
j D1
j
Show that
1
S2n Sn
2
and hence that fSn g is not Cauchy.
5. Let fxP
n g be a sequence of positive terms, and suppose that fSn g diverges, where
Sn D nj D0 xj for all n.
(a) Show that
8 9
<X n
xj =
: S2 ;
j D0 j
is Cauchy and hence converges.

(b) Is the above true for
8 9
<X n
xj =

: S ;
j D0 j
3.3 Elementary Properties of Series 117
6. Show that
1
X 1
;
j D1
j.j C 1/ .j C k/
where k 2 N, is telescoping and hence find the sum.

7. Find the sum
X1
1 1
:
j D0
j Ci C1 j Ci
P1
8. Let fxn g be a sequence of positive terms. Show that j D0 xj converges if and
P
only if the sequence f nj D0 xj g is bounded.
3.3 Elementary Properties of Series
Since the behavior of a series is determined by its sequence of partial sums, many
theorems about series can be derived from the analogous theorems about sequences.
For instance, our next theorem follows immediately from Theorem 1.5.2.
P P1
Theorem 3.3.1. Let 1 j D0 wj and j D0 zj be convergent series. Then, for all
numbers s and t ,
1
X 1
X 1
X
.swj C t zj / D s wj C t; zj : (3.4)
j D0 j D0 j D0
Note that the convergence of the series on the left-hand side of Eq. (3.4)
constitutes part of the conclusion of the theorem. However, as for sequences, the
series on the left-hand side may converge while the series on the right do not. For
instance, for s D t D 1 the series on the right-hand side both diverge if wj D 1 and
zj D 1 for all j , but the series on the left-hand side converges to 0.
The following theorem also is immediate from Theorem 2.3.11 and the definition
of series in terms of sequences of partial sums.
P P1
Theorem 3.3.2. A series 1 j D0 zj converges if and only if the series j D0 Re .zj /
P1
and j D0 Im .zj / both converge.
The following result gives a necessary condition for a series to be convergent.
P
Theorem 3.3.3. If the series 1 j D0 zj converges, then limn!1 zn D 0.
Proof. For each n 0 let

X
n
Sn D zj :
j D0
118 3 Series
Then zn D Sn Sn1 for each n > 0. If

1
X
zj D S;
j D0
then
lim Sn1 D lim Sn D S;

n!1 n!1
so that
lim zn D lim .Sn Sn1 / D S S D 0:

n!1 n!1
t
u
The contrapositive of the theorem above is particularly useful in establishing the
divergence of a series. It is known as the nth-term test.
CorollaryP3.3.4 (nth-Term Test). If limn!1 zn does not exist or is nonzero, then
the series 1j D0 zj diverges.
P
For example, if jzj 1, then the series 1 j
j D0 z must diverge since
lim jzn j D lim jzjn 0:

n!1 n!1
This argument gives an alternative proof of part of Theorem 3.2.1.

P convergence of the sequence fzn g to 0 does not imply convergence of
However,
the series 1j D0 zj . This point is illustrated in the following example. The series in
this example is called the harmonic series.
Example 3.3.1. The series
X1
1
j D1
j
diverges. A proof has been given in Example 2.6.2. We now present an alternative
proof. P
Let Sn D nj D1 1=j for all positive integers n. We use induction to show that
nC1
S2n >
2
for all nonnegative integers n. It will then follow that fSn g diverges, as it has a
divergent subsequence. Since S1 D 1 > 1=2, the required inequality certainly holds
for n D 0. We may therefore assume that n > 0 and that the result holds for n 1.
Then
3.3 Elementary Properties of Series 119
n
X2
1
S2n D
j D1
j
n1
2X n
1 X
2
1
D C
j D1
j j
j D2n1 C1
n
X
2
1
S2n1 C
2n
j D2n1 C1
n
n 1 X
2
> C n 1
2 2
j D2n1 C1
n 2n 2n1
D C
2 2n
nC1
D :
2
4
Remark. For every positive integer n we have
n
X2
1
S2n D
j D1
j
n
X2
1
D 1C
j D2
j
kC1
X
n1 2X
1
D 1C
j
kD0 j D2k C1
kC1
X
n1 2X
1
< 1C
2k
kD0 j D2k C1
kC1
X
n1
1
2X
D 1C 1
2k
kD0 j D2k C1
X
n1 k
2
D 1C
2k
kD0
120 3 Series
X
n1
D 1C 1
kD0
D 1 C n:
Thus the divergence of the harmonic series is extraordinarily slow.

P P1
Exercises 3.2. 1. Can 1 j D0 .xj C yj / converge when at least one of j D0 xj and
P1
j D0 yj diverges?
P
2. Show that if 1 j D0 zj converges, where zj 0 for all j , then
X1
1
z
j D0 j
diverges.
3. Is
1
X
.1/j 21=j
j D1
convergent?P
4. Show that 1 j D1 aj converges if and only if a D 0.
5. Prove that
1
X j2
j D0
7j 2C 11
diverges.
6. Evaluate the following sums:
P
(a) 1 .1/j C2j
j D0 3j I
P1
(b) j D1 52j j.j1C1/ :
P1
7. Let fxn g be a decreasing sequence of positive terms and suppose that j D0 xj
converges. Show that
lim nxn D 0:
n!1
This result is known as Pringsheims theorem. [Hint: Define m D n=2 for even
n and m D .n C 1/=2 for odd n. Then
Sn Sm nxn :
3.4 The Comparison Test 121
8. Show that if an C b 0 for all nonnegative integers n, then

1
X 1
j D0
aj C b
diverges. (Use question 7.)
3.4 The Comparison Test
Often it is difficult or impossible to compute the partial sums for a series, and so it
is necessary to have some tests for convergence that do not depend on knowledge
of the partial sums. Many such tests require the series to contain only nonnegative
terms (which of course must be real). We devote the next few sections to such tests.
One of them is the comparison test, in which we compare the terms of a given series
with the corresponding terms of a series whose behavior is already known.
P P1
Theorem 3.4.1 (Comparison Test). Let 1 j D0 aj and j D0 bj be two series of
nonnegative terms, and suppose that aj bj for all j greater than or equal to
some nonnegative integer N . If the latter series converges, then so does the former.
P
Proof. We may assume that N D 0, since the series 1 j D0 bj converges if and
P1 P1 P1
only if j DN bj does so, and similarly for j D0 aj . Since the series j D0 bj
P
converges, so does the sequence f nj D0 bj g of partial sums. This is a nondecreasing
sequence since bj 0 for all j . If its limit is L then, since aj bj for all j , it
follows that
X
n X
n
aj bj L
j D0 j D0
P
for each n. Thus the nondecreasing sequence f nj D0 aj g is bounded above and
P1
hence converges. We conclude that the series j D0 aj converges. t
u
P1 P1
Corollary 3.4.2. Let j D0 aj and j D0 bj be series of nonnegative terms, and
suppose that aj bj whenever j N 0. If the former series diverges, then so
does the latter.
Remark. If a series of nonnegative terms converges, then the corresponding
sequence of partial sums is convergent and therefore bounded. Conversely, if
the sequence of partial sums is bounded, then, being nondecreasing, it converges
and so the series is convergent.
Example 3.4.1. Let us test the series
X1
1
jj
:
j D1
3
122 3 Series
P1aj D 1=.3 j / and bj D 1=3 for each j 1. Then aj bj for each j . But
j j
Set
j D0 bj is the convergent geometric series
1 j
X 1
:
j D0
3
P1 P1
Thus j D1 bj converges, and therefore so does j D1 aj by the comparison
test. 4
Exercises 3.3. 1. Show that
1
X 1
p
j D1
ej j
converges.
2. Show that
X1
1
p
j D1
j
diverges. P
3. Let fxn g be a sequence of positive terms and suppose that 1 j D0 xj converges.
P1 2
Show that j D0 xj converges.
P
4. Let fxn g be a sequence of positive terms. Show that 1 j D0 xj converges if and
only if
1
X xj
j D0
1 C xj
converges. P
5. Show that if 1 2
j D0 xj converges, then so does
1
X jxj j
:
j D1
j
6. Suppose that 0 < anC1 an for all n, where < 1. Show that
1
X a0
aj :
j D0
1
P1
7. Let fxn g be a sequence of positive terms. Show that if j D0 xj converges, then
so does
3.5 Cauchys Condensation Test 123
1
X p
xj xj C1 :
j D0
(Hint: Use Lemma 2.5.5.)

8. Let fxP
n g be a sequence of positive terms, and suppose that fSn g diverges, where
Sn D nj D0 xj for all n. Show that
X1
xj
m
S
j D0 j
converges if and only if m > 1. (See question 5 at the end of Sect. 3.2.)
9. Let fan g be an increasing sequence of positive terms. Show that
X1
aj
1
j D0
aj C1
converges if and only if fan g is bounded. (Apply question 5 with
xn D anC1 an :/
10. Show that if m > 0 and p > 0, then

0 1
X
n X
n
jm jp
lim @ A D 1:
n!1
j D1
1 C j mC1 j D1
1 C j pC1
P
11. Let fyn g be a bounded sequence of positive terms. Suppose that 1 x is
P1 j D0 j
a convergent series of nonnegative terms. Show that the series j D0 xj yj is
convergent.
3.5 Cauchys Condensation Test

P1
Given the series j D0 aj , the series
1
X
2k a2k
kD0
is called the corresponding condensed series. We may write it as
a1 C .a2 C a2 / C .a4 C a4 C a4 C a4 / C :
124 3 Series
The importance of the condensed series is revealed by the following theorem. It is

remarkable in that it enables us to settle the convergence of a series by considering
only a very thin sample of its terms.
Theorem 3.5.1 (Cauchys Condensation Test).PIf fan g is a nonincreasing
sequence of nonnegative terms, then the series 1 j D0 aj converges if and only
if its condensed series does so.
P
Proof. It suffices to check the convergence of the series 1 j D1 aj and its condensed
series. We may write the sum of the first 2nC1 1 terms of the former series as
a1 C .a2 C a3 / C .a4 C a5 C a6 C a7 / C : : : C .a2n C a2n C1 C : : : C a2nC1 1 /:
Thus
X1
2nC1 1
kC1
X
n 2X
aj D aj :
j D1 kD0 j D2k
Since aj a2k whenever 2k j < 2kC1 , we find that
X1
2kC1 X1
2kC1
aj a2k
j D2k j D2k
X1
2kC1
D a2k 1
j D2k
D 2k a2k :
It therefore follows from the comparison test and Theorem 3.2.3, because of the
assumption
P that aj 0 for all j , that if the condensed series converges, then so
does 1 j D1 aj .
On the other hand, for all n > 0 we can also write
n
X
2
aj D a1 C a2 C .a3 C a4 / C : : : C .a2n1 C1 C a2n1 C2 C : : : C a2n /
j D1
k
X
n X
2
D a1 C aj :
kD1 j D2k1 C1
As aj a2k whenever 2k1 < j 2k , it follows that

k k
X
2 X
2
aj a2k
j D2k1 C1 j D2k1 C1
3.5 Cauchys Condensation Test 125
D 2k1 a2k
1 k
D 2 a2k ;
2
and so
k
X
n X
n X
2
2 a2k 2
k
aj :
kD1 kD1 j D2k1 C1
P1
If the series j D1 aj converges, then so does
1
X X
2k
2aj ;
kD1 j D2k1 C1
and consequently the comparison test shows that the condensed series also con-
verges. t
u
Remark. We observe from the proof above that if
1
X
aj D S
j D1
and
1
X
2j a2j D T;
j D0
then S T 2S .
The next theorem, which generalizes Example 3.3.1, is an application of the
condensation test.
Theorem 3.5.2. For every rational number p, the series
X1
1
j p
j D1
converges if p > 1 and diverges otherwise.

Proof. If p 0, then 1=j p 1 for all positive integers j . Thus the sequence
f1=np g does not converge to 0 and the series diverges by the nth-term test.
For each p > 0 the sequence f1=np g is decreasing and Cauchys condensation
test can be applied. The condensed series is
X1 X1 X1 j
j 1 1 1
2 jp D D :
j D0
2 j D0
2j.p1/ j D0
2p1
126 3 Series
This is a geometric series and converges if and only if 1=2p1 < 1. Hence the
condensed series, and therefore the given series, converges if and only if p 1 > 0,
as required. t
u
The series given in Theorem 3.5.2 is generally referred to as the p-series.
Exercises 3.4. 1. Using the knowledge gained so far, discuss the convergence of
the following series:
P1
(a) p1
j D2 j j 1 ;
P1 1
(b) j D0 aj2 Cb , where a > 0 and b > 0;
P1 1
(c) j D0 j .
Let fan g be a nonincreasing sequence of nonnegative terms. Prove that the series
2. P
1
j D0 aj converges if and only if the series
1
X
3j a3j
j D0
converges. (This result is in fact true if we replace 3 by any integer greater than
2 [12].) P
3. (a) The following result is a special case of a theorem in [1]. Let 1 j D1 aj be a
series of positive terms. Suppose that the set

maxfa2n ; a2nC1 g
n 2 N
an
is bounded below and above by l and L, respectively. Then the series
converges if L < 1=2 and diverges if l > 1=2.
Fill in the details in the following brief sketch of the proof. Suppose that
L < 1=2. As in Cauchys condensation test,
X1
2nC1 X
n
S2nC1 1 D aj D a1 C Tk ;
j D1 kD1
where
X1
2kC1
Tk D aj
j D2k
k 1
2X
D .a2j C a2j C1 /
j D2k1
2LTk1
.2L/k a1 :
3.6 The Limit Comparison Test 127
Deduce that the sequence fS2nC1 1 g converges, and hence that fSn g does so.
Similarly, prove the theorem for the case where l > 1=2.
(b) Use the theorem of part (a) to prove the convergence of the following
series:
P
i. 1 1
j D1 p for all rational p > 1;
P1 j .2j 1/
ii. j D1 22j 1 .j 1/.j C1/ .
[Hint: If an denotes the nth term of the series, then

a2n 1 1 n1
1 :
an 2 4n
4. Test the convergence of the series

1
X .2j 1/
2j 1 .j 1/j
:
j D1
2
[Hint: If an denotes the nth term of the series, then an > 1=.2n/.]
3.6 The Limit Comparison Test
Our next test for convergence is often easier to apply than the comparison test. First,
however, we require some new notation.
Definition 3.6.1. Let fan g and fbn g be sequences of positive terms. Then an and bn
are said to be of the same order of magnitude if there is a positive number L such
that
an
lim D L:
n!1 bn
In this case we write an Lbn . We say that an is of a lesser order of magnitude

than bn , and write an << bn , if
an
lim D 0:
n!1 bn
Finally, an is of a greater order of magnitude than bn if

an
lim D 1:
n!1 bn
We then write an >> bn .

128 3 Series
The basic intuitive idea is that if an << bn , then bn increases with n much faster
(or decreases much more slowly) than does an . Theorem 2.8.4 shows that an << bn
if and only if bn >> an . Moreover if fcn g is another sequence of positive terms
and an << bn << cn , then an << cn . This result follows immediately from the
observation that
an an bn
D :
cn b n cn
Observe also that if an Lbn , then bn an =L.

The following example is worth noting.
Example 3.6.1. If p is a rational number and c > 1, then
np << c n << n << nn :
Proof. These results are immediate from Examples 2.5.5, 2.5.6, and 2.7.10. 4
We are now ready for our next convergence test, which is known as the limit
comparison test.
P P1
Theorem 3.6.1 (Limit Comparison Test). Let 1 j D0 aj and j D0 bj be series of
positive terms.
1. If an Lbn for some
P L > 0, then both series converge
P1or both diverge.
2. If an << bn and 1 b converges, then so does a .
Pj1D0 j P1j D0 j
3. If an >> bn and j D0 bj diverges, then so does j D0 aj .
Proof. 1. We are given that
an
lim D L > 0:
n!1 bn
Proposition 2.3.5 shows the existence of a number N1 such that an =bn > L=2
for all n N1 . Similarly, there exists N2 such that an =bn < L C L=2 D 3L=2
for all n N2 . Take N D maxfN1 ; N2 g and choose n N . Then
L an 3L
< < I
2 bn 2
hence
3Lbn
an < :
2
P1
We now apply the comparison test. Suppose j D0 bj is convergent. By
Theorem 3.3.1 the series
1
X 3Lbj
j D0
2
P
also converges, whence 1 j D0 aj converges by the comparison test. Similarly, if
P1 P1
a
j D0 j converges, then so does b
j D0 j .
2. Suppose
an
lim D 0:
n!1 bn
Then there exists N such that

an
<1
bn
for all n N . HenceP an < bn for all nP N , and we conclude from the
comparison test that 1 a
j D0 j converges if 1
j D0 bj does so.
3. This statement is equivalent to the previous one. u
t
Example 3.6.2. Test the series
1
X 1
j D0
.j C 1/.j C 2/
for convergence.
Solution. Let
1
an D
.n C 1/.n C 2/
for all n > 0. We show that an has the same order of magnitude as 1=n2 .
Accordingly, we put
1
bn D
n2
for all n > 0. Then
an n2
D !1
bn .n C 1/.n C 2/
P
as n ! 1. Thus an bn . As 1 j D1 bj converges by Theorem 3.5.2, so does
P1
j D0 aj by the limit comparison test. 4
130 3 Series

1
X p p
. j C 1 j /2
j D0
for convergence.
Solution. For all n 0 let
p p
an D . n C 1 n/2
p p p p !2
. n C 1 n/. n C 1 C n/
D p p
nC1C n
1
D p p ;
. n C 1 C n/2
and let
1
bn D
n
for all n > 0. Since
p p
bn . n C 1 C n/2
D
an n
p
2n C 1 C 2 n.n C 1/
D
n
r
1 1
D 2C C2 1C
n n
!4
P
as n ! 1, it follows that an bn =4. As 1 j D1 bj is the divergent harmonic series,
P1
j D0 aj diverges by the limit comparison test. 4
1
X 1
1
:
j D1 j 1C j
Solution. Taking
1
an D
1C n1
n
and
1
bn D
n
for all n > 0, we have
1
bn n1C n
D D n1=n ! 1
an n
P P1
as n ! 1. Since 1 j D1 bj diverges, so does j D1 aj .
The reader should compare this result with that of Theorem 3.5.2. 4
Exercises 3.5. 1. Test the convergence of each of the following series:
P1 2j C1
(a) j D0 5j j ;
P1 p p
j C1 j
(b) j D1 j
;
P1 e 1=j
(c) j D1 j 2 ;
P1 1p
(d) p ;
j D3 j 3C j C3
P1 1=j p
(e) j D1 .e 1/ for every rational p [use Theorem 2.7.10(3)];
P1
(f) j D1 .e
1=j
1/p for every rational p;
P1 p
j D0 . j C 1
(g) 2 j /;
P1 p p
j C1 j
(h) p ;
j D1 j.j C1/
P1 1
(i) j D1 j 21=j .
2. Find all integers t such that
1
X 1
p
j D1
j 1Ct t
is convergent.
3. Let a > 0 and b > 0. Find all rational p such that
1
X 1
j D0
.aj C b/p
is convergent.
132 3 Series

1
X 1
;
j D0
aj C bj C c
2
where a; b; c are all positive.

5. Prove that
lim .n/1=n D 1:
n!1
3.7 The Ratio Test
In order to use the comparison or limit comparison test effectively, we need a supply
of series whose convergence or divergence has already been established. We shall
therefore develop another test which involves only the terms of the series being
tested. Its proof uses the comparison test.
P
Theorem 3.7.1. Let 1 j D0 aj be a series of positive terms and suppose there exist
numbers r and N such that
anC1
r <1
an
for all n N . Then the series converges. On the other hand, if

anC1
1
an
for all n N , then the series diverges.
Proof. In the first case we have anC1 ran for all n N . It follows by induction
that
aN Cj r j aN
P
for each positive integer j . Now the series 1 j
j D0 aN r converges since 0 P
< r < 1,
P1 1
and so j D0 aN Cj converges also, by the comparison test. Hence j D0 aj
converges.
In the second case we have anC1 an for all n N and the series diverges by
the nth-term test. t
u
3.7 The Ratio Test 133
P1
Corollary 3.7.2 (Ratio Test). Let j D0 aj be a series of positive terms and let
anC1
lim DL
n!1 an
for some number L. Then the series converges if L < 1 and diverges if L > 1.
Proof. If L < 1, then by Proposition 2.3.4 there exist numbers N1 and r < 1 such
that
anC1
<r
an
for all n N1 . The result therefore follows immediately from Theorem 3.7.1 in this
case.
On the other hand, suppose L > 1. There exists N2 such that anC1 =an > 1 for
all n N2 , and again the result follows from Theorem 3.7.1. t
u
Remark. No conclusion can be drawn P when L DP11in the ratio test. For example,
this is the case for both the series 1
j D1 1=j and j D1 1=j 2
, but the former series
diverges whereas the latter converges.
X1
.j C 3/2j C1
:
j D0
3j
Solution. Putting
.n C 3/2nC1
an D
3n
for all n 0, we have
anC1 .n C 4/2nC2 3n
D
an 3nC1 .n C 3/2nC1
2.n C 4/
D
3.n C 3/
2
!
3
as n ! 1. Hence the series converges by the ratio test. 4
134 3 Series
X1
x 2j
;
j D0
.2j /
where x 0.
Solution. With
x 2n
an D
.2n/
for all n, we have
anC1 x 2nC2 .2n/

D 2n
an .2n C 2/ x
x2
D
.2n C 2/.2n C 1/
! 0;
and so the series converges.

A similar argument shows that
1
X x 2j C1
j D0
.2j C 1/
also converges for all x > 0. 4

Example 3.7.3. The results of Example 3.6.1 can also be achieved by using the ratio
test. For instance, let us show that n << nn . If we set an D n=nn for all n > 0,
then the calculation in Example 2.7.10 shows that
anC1 1 1
D ! < 1:
1 n
an 1C n e
Hence the series
X1
j
j D1
jj
3.7 The Ratio Test 135
converges by the ratio test. We deduce that
n
lim D 0;
n!1 nn
as required. 4
P1
Exercises 3.6. 1. Let j D0 aj be a series of positive terms. Show that if anC1
an for all n, then the series diverges.
2. Test the convergence of the following series:
P1 j ej P1 .2j /
(a) j D1 j j ; (d) j D1 j j ;
P1 j 2 P1 .j /2
(b) j D1 j ; (e) j D0 .2j / ;
P1 2j j 2 P1 .j /2 3j
(c) j D1 j ; (f) j D0 .2j / .
3. (a) Suppose
P1 n > 0, and anC1 =an bnC1 =bn for all n. Show that
that an > 0, bP
1
j D0 a j converges if j D0 bj does. (Hint: Show that the sequence fan =bn g
is decreasing and use question 11 in the exercises at the end of Sect. 3.4.)
(b) Use part (a) to test the convergence of the following series:
P1 j j
i. j D1 jej j ;
P1 jj
ii. j D1 e j j .
Note that the ratio test yields no conclusion in these examples.

1
X j x j
j D1
jj
for all x > 0.

5. The result in question 3 at the end of Sect. 3.5 is known as the second ratio test.
Here is a special case that can be proved by using Cauchys condensation test.
Let fan g be a decreasing sequence
Pof positive terms. By applying the ratio test to
the condensed series, show that 1 j D0 aj converges if
a2n 1
lim <
n!1 an 2
and diverges if
a2nC1 1
lim > :
n!1 an 2
136 3 Series
3.8 The Root Test
A somewhat different application of the comparison test gives us another test called
the root test. Again, it is applicable only to series of nonnegative terms. Its proof is
similar to that of the ratio test.
P
Theorem 3.8.1. Let 1 j D0 aj be a series of nonnegative terms, and suppose that
1=n
there exist numbers r and N > 0 such that an r < 1 for all n N . Then the
series converges.
P1 The hypothesis implies that an r for all n N . Hence P

n
Proof. the series
1 j
a
j D0 j converges by comparison with the convergent geometric series j D0 r :
t
u
P1
Corollary 3.8.2 (Root Test). Let j D0 aj be a series of nonnegative terms and let
lim an1=n D L
n!1
for some number L. Then the series converges if L < 1 and diverges if L > 1.
1=n
Proof. If L < 1, then there exist numbers N1 > 0 and r < 1 such that an < r for
all n N1 . In this case the result follows immediately from Theorem 3.8.1.
1=n
On the other hand, suppose L > 1. There exists N2 such that an > 1 for all
n N2 . For each such n it follows that an > 1. Hence fan g cannot converge to 0
and the given series diverges. t
u
Remark. P1test, no2 inference may be drawn if L D 1. Consider the
P As in the ratio
series 1 j D1 1=j and j D1 1=j as in the remark following the introduction of the
ratio test. Using the fact that n1=n ! 1 as n ! 1 (Example 2.5.1), we see that
L D 1 for both series, but one series diverges and the other converges.
1
X j 2
j
:
j D1
j C1
Solution. Setting
n2
n
an D
nC1
3.8 The Root Test 137
for all n > 0, we obtain

n
n 1 1
an1=n D D ! < 1:
1 n
nC1 1C n e
Hence the series converges by the root test. 4

1=n
Remark. If an < 1 for all n > 0, it does not necessarily follow that the series
P 1
j D0 aj converges. Indeed, the harmonic series provides an example of a divergent
series where this inequality holds for all n > 1.
We now show that if the ratio test is applicable, then so is the root test. In fact,
the root test is stronger.
Theorem 3.8.3. Let fan g be a sequence of positive terms. If
anC1
lim D L;
n!1 an
where L may be a number or 1, then
lim an1=n D L:
n!1
Proof. Since an > 0 for all n, it follows that if L is a number then it must be
nonnegative.
Case 1: Suppose L > 0, and choose " 2 .0; L/. There exists N1 such that
anC1
L"< <LC"
an
for all n N1 . Thus
.L "/an < anC1 < .L C "/an (3.5)
for all such n. In particular,
.L "/aN1 < aN1 C1 < .L C "/aN1 :
Suppose that
.L "/m aN1 < aN1 Cm < .L C "/m aN1 (3.6)
for some m 2 N. Then, using inequality (3.5) and the fact that L " > 0, we
find that
.L"/mC1 aN1 < .L"/aN1 Cm < aN1 CmC1 < .LC"/aN1 Cm < .LC"/mC1 aN1 :
138 3 Series
Therefore inequality (3.6) holds for all positive integers m, by induction. Any
integer n > N1 can be written as n D N1 C m, where m D n N1 > 0, and so
inequality (3.6) can be rewritten as
aN1 aN1
.L "/n < an < .L C "/n :
.L "/N1 .L C "/N1
Thus,
1=n 1=n
aN1 aN1
.L "/ < an1=n < .L C "/ (3.7)
.L "/N1 .L C "/N1
for all n > N1 , since L " > 0. Denoting the left- and right-hand sides of
inequality (3.7) by sn and tn , respectively, and using the result of Example 2.5.2,
we find that sn ! L " and tn ! L C " as n ! 1. Therefore there exist N2
and N3 such that
" < sn L C " < "
for all n N2 and
" < tn L " < "
for all n N3 . For all n > maxfN1 ; N2 ; N3 g it follows that
L 2" < sn < an1=n < tn < L C 2";
and so
lim an1=n D L;
n!1
as required.
Case 2: Suppose L D 0. For every " > 0 there exists N such that
anC1
<"
an
whenever n N . For all such n we therefore have
anC1 < "an :
Arguing as in case 1, we see by induction that
aN Cm < "m aN
3.8 The Root Test 139
for all positive integers m, and therefore that

aN
an < "nN aN D "n
"N
for all n > N . Hence
a 1=n
N
an1=n < "
"N
for all such n. Denoting the right-hand side of this inequality by sn , we see that
sn ! " as n ! 1. We then argue as in case 1 to show that
0 < an1=n < sn < 2"
for large enough n, and the result follows in this case also.
Case 3: Suppose finally that
anC1
lim D 1;
n!1 an
and choose a number M . There exists N such that

anC1
>M C1
an
for all n N . Thus
anC1 > .M C 1/an
for all such n. Arguing as in the previous cases, we see by induction that
aN
an > .M C 1/n
.M C 1/N
for all n > N and hence that

1=n
aN
an1=n > .M C 1/ :
.M C 1/N
If we denote the right-hand side of this inequality by sn , then sn ! M C 1 as

n ! 1. For large enough n we therefore have
an1=n > sn > M;

u
140 3 Series
Example 3.8.2. Consider the series

1
X 1
:
j D0
2j C.1/j
Letting
1
an D
2nC.1/n
for all n, we find that
(
1
if n is even;
an D 2nC1
1
2n1
if n is odd.
Thus, if n is even so that n C 1 is odd, then

anC1 2nC1
D n D 2;
an 2
but if n is odd, then

anC1 2n1 1
D nC2 D :
an 2 8
Therefore limn!1 anC1 =an does not exist. However,
8
< 11=n if n is even;
22
an D
1=n
: 21=n
2
if n is odd.
Hence
1
lim an1=n D < 1;
n!1 2
so that the series converges by the root test.
Note that the comparison test could also have been used, since an 1=2n1 for
every n. 4
Exercises 3.7. 1. Prove that
n
lim D e:
n!1 .n/1=n
(Hint: Apply Theorem 3.8.3.)

3.9 The KummerJensen Test 141
2. Apply the root test to investigate the convergence of the series

1
X ej
:
j D0
j

P1 j j P1 xj
(a) j D1 2j 1 ; (e) j D1 j C1 j 2 for every x > 0;
j
P1 P1 j 2 C1
j 2
(b) j D2 .j
1=j
1/j ; (f) j D0 j 2 Cj C1
;
P1 P1 j 2
3j=2
(c) j D0 2j ; (g) j D1 1 1
j
;
P1 P1 jj
(d) j D1 j 2 e j for every real ; (h) j D1 2j 2 .
4. Consider the series

X1
1 1 j
1 :
j D2
j j
(a) Show that the root test is not applicable.

(b) Test the convergence of the series. (Hint: You may find Proposition 2.7.3
useful.)
5. Let 0 < < < 1, and for each nonnegative integer n define
(
n if n is odd;
an D
n if n is even.
P
Determine the convergence of 1 j D0 aj . (Note that the ratio and root tests both
fail.)
3.9 The KummerJensen Test
The ratio test is an application of the comparison test. In fact, it is a special case of
a more general application due to Kummer and Jensen.
P
Theorem 3.9.1 (KummerJensen Test). Let 1 j D0 aj be a series of positive terms
and fbn g a sequence of positive terms. Let
anC1
cn D bn bnC1
an
for all n.
142 3 Series
P
1. If limn!1 cn > 0, then 1 j D0 aj converges.
P
2. If 1 1=b diverges and there exists N such that cn 0 for all n N , then
P1 j D0 j
j D0 aj diverges.
Proof. 1. Suppose that
lim cn D L > 0
n!1
and choose r such that 0 < r < L. Then there exists N > 0 such that
anC1
cn D bn bnC1 > r
an
for all n N . Hence
an bn anC1 bnC1 > ran (3.8)
for each such n.

Define
X
n
Sn D aj
j D0
for all n 0. Since aj > 0 for all j , the sequence fSn g of partial sums is
increasing. It therefore suffices to show that it is bounded above. Using (3.8) and
the telescoping property, for all integers m N we have
X
m
r.Sm SN 1 / D r an
nDN
X
m
< .an bn anC1 bnC1 /
nDN
D aN bN amC1 bmC1
< aN bN :
Hence
aN bN
Sm < C SN 1
r
for all such m, and the sequence fSn g is indeed bounded above, as desired.
2. By hypothesis, we have
an bn anC1 bnC1 0
for all n N . Hence the sequence fan bn g is nondecreasing when n N , and

so an bn aN bN for all such n. Thus
aN bN
an ;
bn
P1 P1
and the divergence of j D0 1=bj implies that of j D0 aj by the comparison
test. t
u
The ratio test is obtained immediately by putting bn D 1 for all n. In view of
the limit comparison test, the divergence of the harmonic series shows that another
possibility is to have bn D n 1 for all n. We then obtain
anC1
cn D n 1 n
an

anC1
D n 1 1
an
for all n, and so we deduce the following result, due to Raabe.

P
Corollary 3.9.2 (Raabes Test). Let 1 j D0 aj be a series of positive terms and
suppose that

anC1
lim n 1 DL
n!1 an
for some number L. Then the series converges if L > 1 and diverges if L < 1.
In fact, the KummerJensen test shows that in order to establish divergence, it
suffices to find an N such that

anC1
n 1 1
an
whenever n N . Such an N certainly exists if L < 1.
1
X m
.2j 1/
j D1
22j 1 .j 1/j
for convergence, where m 2 N.

144 3 Series
Solution. Writing
m
.2n 1/
an D
22n1 .n 1/n
for all n 2 N, we compute

m
anC1 .2n C 1/ 22n1 .n 1/n
D
an 22nC1 n.n C 1/ .2n 1/
m
2n C 1
D :
2n C 2
Thus
anC1
lim D 1;
n!1 an
so that the ratio test is inconclusive.

Let us try Raabes test. We have
m
anC1 1
n 1 D n 1 1
an 2n C 2
0 0 ! 11
m Xm
m .1/ j
D n @1 @1 C AA
2n C 2 j D2 j .2n C 2/j
!
nm X m
m .1/j
D n
2n C 2 j D2
j .2n C 2/j
m
!
2
as n ! 1. Raabes test therefore shows that the series diverges for m D 1 and
converges for m > 2.
We now apply the KummerJensen test to the case m D 2, where Raabes test is
not applicable. Taking bn D n for all n, we find that
anC1 .2n C 1/2 .n C 1/

bn bnC1 D n
an 4.n C 1/2
1
D
4.n C 1/
< 0:
Testing n=1 an , an >0
no
an 0?
divergent
yes or uncertain
an contains n!, yes an+1

Find lim an = L
cn or nn ? n
and try ratio test
L> 1 L< 1 L does not exist or L= 1
no
an +1
divergent convergent an 1?
yes
yes
an is a ratio of Try limit
comparison test divergent by
polynomials? nth term test
no
no
Try comparison test or Try root test. yes

Finished
Conclusion?
no
yes
Finished
Conclusion?
no
Try the Kummer-

Jensen test
Fig. 3.1 Testing a series of positive terms for convergence
Recalling that the harmonic series diverges, we conclude by the KummerJensen

test that the given series diverges. 4
Figure 3.1 suggests a procedure for testing a series of positive terms for
convergence.
146 3 Series
Exercises 3.8. 1. Test the following series for convergence:

P1 j5 P1 1
(a) j D1 2j ; (j) j D0 22j C.1/j ;
P1 100j P1 j
(b) j D0 j
; (k) j D1 e j 2 ;
P1 j C5 P1 .2j /
(c) j D1 j 2 ; (l) j D0 22j .j /2 ;
P1 1 P1 .2j 1/
(d) j D0 .2j C1/ ; (m) j D1 22j .j /2 ;
P1 pj C1 P1 .2j 1/
(e) j D0 j 2 C1 ; (n) j D1 22j 1 .j 1/.j C1/ ;
j
P1 1C j1 P1 23j 1 .j /3
(f) j D1 j
; (o) j D0 .j C1/.2j C1/ ;
Qj
P1 p
3j C5 j C7 P1 kD1 .kCa1/
(g) j D0 j 3 C3j 3=2 C5 ; (p) j D1 Qj .kCb1/ for a > 0 and b > 0;
kD1
Qj
P1 1C2C:::Cj P1 2
kD1 .3k1/ .3kC1/
2
(h) j D1 j3
; (q) j D1 Qj
4
;
kD1 .3k/
P1 1 P1 .C1/.Cj 1/
(i) j D1 j 3C.1/j ; (r) j D1 j
for every real .
2. For each real define
!

D1
0
and
!
. 1/ . n C 1/
D
n n
for each n 2 N. Show that the series

1
!
X
.1/j
j D1
j 1
converges
P if 0 and diverges if < 0.
3. Let 1 j D0 aj be a series of positive terms and let fbn g be a sequence of positive
terms. For all n define
an
cn D bn bnC1
anC1
and

an
rn D n 1 :
anC1
3.10 Alternating Series 147
Prove the following variants of the KummerJensen test and Raabes test:
(a) If cP
n > t > 0 for all n and some fixed t , then the series converges.
(b) If 1 j D0 1=bj diverges and cn 0 for all n, then the series diverges.
(c) If rn > t > 1 for all n and some fixed t , then the series converges.
(d) If rn 1 for all n, then the series diverges.
3.10 Alternating Series
In the previous several sections we dealt with series of nonnegative terms. Clearly,
the results can be applied also if the terms are all negative. If neither of these
conditions obtains, the series may be difficult to handle. However, there is a
convenient test, due to Leibniz, that can be applied to what is known as an alternating
series.
Definition 3.10.1. The series
1
X
.1/j bj (3.9)
j D0
is alternating if each bj is positive.

Theorem 3.10.1 (Leibnizs Test). The alternating series (3.9) is convergent if fbn g
is a nonincreasing sequence of positive terms converging to 0.
Proof. For each integer n 0 let
X
n
Sn D .1/j bj :
j D0
We need to show that the sequence fSn g converges. We achieve this result by proving
that the subsequences fS2n g and fS2nC1 g both converge to some number S and then
appealing to Theorem 2.4.3.
First we have
X
2nC1
S2nC1 D .1/j bj
j D0
X
2n
D b0 C .1/j bj b2nC1
j D1
X
n
D b0 C .b2j b2j 1 / b2nC1 :
j D1
148 3 Series
Note that b2nC1 > 0 and, since fbn g is nonincreasing, b2j b2j 1 for all j . Hence,
the sequence fS2nC1 g is bounded above by b0 . It is nondecreasing, because for all
n > 0 we have
S2nC1 S2n1 D b2n b2nC1 0:
We conclude that fS2nC1 g converges to some number S . It follows that
lim S2n D lim .S2nC1 b2nC1 /

n!1 n!1
D S 0
D S;

u
1
X j
.1/j :
j D1
.j C 1/2
Setting
n
bn D
.n C 1/2
for all n > 0, we observe that the sequence fbn g converges to 0. To show that it is
nonincreasing, note first that
nC1 n
bnC1 bn D
.n C 2/ 2 .n C 1/2
.n C 1/3 n.n C 2/2
D :
.n C 1/2 .n C 2/2
As the denominator of this expression is positive, in order to determine the sign of

bnC1 bn it suffices to inspect the numerator. Since
.n C 1/3 n.n C 2/2 D n2 n C 1 < 0
for all n > 0, we confirm that the sequence fbn g is in fact decreasing. Hence the
given series converges by Leibnizs test. 4
Example 3.10.2. Let an D i n =n for all n > 0. Then Re .an / is equal to 0 if n is
odd, to 1=n if n is divisible by 4, and to 1=n otherwise. Thus
1
X 1
X
ij .1/k
Re D ;
j D1
j 2k
kD1
which converges by Leibnizs test. Similarly,

1
X 1
X
ij .1/k1
Im D ;
j D1
j 2k 1
kD1
P1
which also converges. Hence the series j D1 i j =j converges by Theorem 3.3.2.
The conjugate series
1
X .i /j
j D1
j
also converges, by the same argument. 4

Our next example shows that the condition that the sequence fbn g of Leibnizs
test be nonincreasing cannot be dropped.
P
Example 3.10.3. Consider the series 1 j D1 .1/
j C1
bj , where
1
b2j 1 D
j C1
and
1
b2j D
.j C 1/2
for all j > 0. The sequence fbn g converges to 0, but it is not monotonic.
If the series were to converge, then we could obtain another convergent series by
grouping its terms in any way. However, by grouping the terms in pairs, we obtain
1
X 1
X
1 1 j
D ;
j D1
j C 1 .j C 1/2 j D1
.j C 1/2
and this series is divergent by the limit comparison test applied to the harmonic
series. 4
Often it is difficult to determine to what number S a convergent series converges,
but in the situation where Leibnizs test is applicable it is possible to approximate S .
Indeed, we see in the following theorem that if we attempt to approximate S by
taking the sum of the first few terms of the series, then the error is bounded above
by the absolute value of the first term omitted from the sum.
150 3 Series
Theorem 3.10.2. Let fbn g be a nonincreasing sequence of positive terms, and let
1
X
.1/j bj D S:
j D0
Then
jS Sn j bnC1
for each n 0, where
X
n
Sn D .1/j bj :
j D0
Proof. We saw in the proof of Theorem 3.10.1 that the sequence fS2nC1 g is
nondecreasing. Similarly, fS2n g is nonincreasing, since
S2nC2 S2n D b2nC2 b2nC1 0
for all n. Therefore S is an upper bound for fS2nC1 g and a lower bound for fS2n g.
Hence
S2n1 S2nC1 S S2n (3.10)
for all n > 0, so that
0 S S2n1
S2n S2n1
D b2n ;
as desired. The last two inequalities of (3.10) hold even if n D 0. Therefore
0 S2n S
S2n S2nC1
D .b2nC1 /
D b2nC1
for all n 0, and the proof is complete. t

u
Remark. The theorem shows that
Sn bnC1 S Sn C bnC1
for all n 0. Moreover it is easily seen that the inequalities are strict if the sequence
fbn g is decreasing. In particular, since we then have b0 > b1 , it follows that S > 0.
Example 3.10.4. Consider
1
X .1/j
:
j D1
j2
This series converges to some number S by Leibnizs test, since the sequence f1=n2 g
is decreasing and converges to 0. We approximate S by the partial sum
X
n
.1/j
Sn D
j D1
j2
for some n > 0. If we require accuracy to within a given positive number ", then
according to Theorem 3.10.2, we need to choose n so that
1
< ":
.n C 1/2
Thus we must have

1
.n C 1/2 > ;
"
and to ensure that this inequality holds, we take
1
n > p 1:
"
The exact value of the sum requires ideas that are beyond the scope of this book. 4
Exercises 3.9. 1. Determine the convergence of the following series:
P
(a) 1
2
j C1 j
j D1 .1/ j 3 C1
;
P1 C1
(b) j D1 .1/j C1 j.j C1/ ;
2j
P
(c) 1 j D0 .1/
j C1 j 5
7j C3
;
P1
(d) j D2 .1/j C1 .j 1=j 1/;
P
(e) 1 j D1 .1/
j C1 1=j
.a 1/, where a > 0;
j C1
P1 j C1
(f) j D1 .1/ 1C j
1
e ;
P1 .1/j i j
(g) j D1 pj .
152 3 Series
2. Show that the series

X1
.1/j C1
SD p
j D2
j j
converges. Letting Sn be the nth partial sum of the series, find an upper bound
for jS15 S j.
3. Let fan g be a decreasing null sequence. Show that
1
X a1 C a2 C : : : C aj
.1/j C1
j D1
j
converges.
P P1
4. If 1 j D0 aj converges, does it follow that
j
j D0 .1/ aj converges? Give a
proof or a counterexample.
5. Determine the convergence of
1
X .1/j C1
:
j D0
j C 2 C .1/j
[Hint:
1 j C 2 .1/j
D :
j C 2 C .1/j .j C 2/2 1
3.11 Dirichlets Test
All the tests we have learned so far help us to determine the convergence of series
whose terms are all positive or alternate in sign. In this section we provide a useful
test that does not require these assumptions. The proof relies on Abels partial
summation identity.
Lemma 3.11.1 (Abels Partial Summation Identity). Let fun g and fvn g be two
sequences of complex numbers, and let fUn g be a sequence such that Un Un1 D un
for all n 2 N. Then
X
n X
n1
uj vj D Un vn Um vmC1 Uj .vj C1 vj / (3.11)
j DmC1 j DmC1
whenever m < n.
3.11 Dirichlets Test 153
Proof. Since Uj Uj 1 D uj , we have
X
n X
n

uj vj D vj Uj Uj 1
j DmC1 j DmC1
X
n X
n
D vj Uj vj Uj 1
j DmC1 j DmC1
X
n1 X
n
D vj Uj C vn Un vmC1 Um vj Uj 1
j DmC1 j DmC2
X
n1 X
n1
D vn Un vmC1 Um C vj Uj vj C1 Uj
j DmC1 j DmC1
X
n1
D Un vn Um vmC1 Uj .vj C1 vj /:
j DmC1
t
u
The next result is due to Shiu [14].
Theorem 3.11.2. Let fun g and fvn g be sequences of complex numbers and suppose
that
P
1. 1 j D0 jvj C1 vj j converges,
2. fvn g converges to 0, and
3. there is a constant K such that

X
n
uj K

j D0
for all n 0.
P
Then 1 j D0 uj vj is convergent.
Proof. Choose " > 0. The convergent sequence

8 9
<Xn =
jvj C1 vj j
: ;
j D0
154 3 Series
is Cauchy. Therefore by hypotheses (1) and (2) there exists N such that for all
n N we have
X
n
jvj C1 vj j < "
j DN C1
and
jvn j < ":
For each n 0 let
X
n
Sn D uj vj
j D0
and
X
n
Un D uj :
j D0
Thus jUn j K for all n 0, by hypothesis. For all integers m and n such that
n > m N , Abels identity therefore gives

X
n

jSn Sm j D uj vj
j DmC1
X
n1
jUn vn j C jUm vmC1 j C jUj jjvj C1 vj j
j DmC1
X
n1
Kjvn j C KjvmC1 j C K jvj C1 vj j
j DmC1
< 3K";
as it may be assumed that K > 0. The desired conclusion follows from Cauchys
criterion. t
u
Corollary 3.11.3 (Dirichlets Test). Let fun g be a complex sequence and let fvn g
be a real sequence. Suppose that
1. fvn g is monotonic and converges to 0, and
3.11 Dirichlets Test 155
2. there is a constant K such that

X
n
u
j K

j D0
for all n 0.
P
Then 1 j D0 uj vj is convergent.
Proof. According to Theorem 3.11.2, it is enough to show that

1
X
jvj C1 vj j
j D0
converges. Choose " > 0. By condition (1), there exists N such that
jvn j < "
for all n N . For each such n, define
X
n
Tn D jvj C1 vj j:
j D0
We must show that fTn g is convergent.

Since fvj g is monotonic, either fvj C1 vj g is a sequence of nonnegative terms
or it is a sequence of nonpositive terms. Therefore for all integers m and n such that
n > m N , it follows that
X
n
jTn Tm j D jvj C1 vj j
j DmC1

X
n
D .vj C1 vj /
j DmC1
D jvnC1 vmC1 j
jvnC1 j C jvmC1 j
< 2":
The desired conclusion thus follows from Cauchys criterion. t

u
Remark. We obtain Leibnizs test by taking un D .1/ for all n in Dirichlets test.
n
156 3 Series
Example 3.11.1. Let jzj D 1 and z 1. We show that
X1
zj
j D1
jp
is convergent for every rational p > 0.

Let us take vn D 1=np and un D zn for all n. Then

n
X j 1 znC1 1 C jzjnC1 2
D
z 1 z j1 zj D j1 zj
j D0
for all n. Therefore the series converges by Dirichlets test. 4

Exercises 3.10. 1. (a) Let fvn g be a monotonic sequence, and let fun g be a
sequence for which there is a constant K such that

n
X
uj K

j D0
for all n 0. Prove that

X
n
uj vj 2K.jvmC1 j C jvn j/

j DmC1
for all positive integers m and n such that m < n.

(b) Use part (a) to prove Dirichlets test. P
2. Suppose that fvn g is a monotonic bounded sequence and that 1j D0 uj converges.
P1
Prove that j D0 uj vj converges. (This result is known as Abels test.)
3.12 Absolute and Conditional Convergence
We have seen the advantages of considering series whose terms are all nonnegative.
Given an arbitrary series, we may obtain a series of nonnegative terms by replacing
each term with its absolute value. If the resulting series converges, then so does the
original series. That is the content of our next theorem. We begin its proof with a
useful lemma.
P1
Lemma 3.12.1. Let fzn g be a sequence of complex numbers. Then j D0 jzj j
P1 P1
converges if and only if j D0 jRe .zj /j and j D0 jIm .zj /j converge.
3.12 Absolute and Conditional Convergence 157
Proof. Recalling that jRe .zn /j jzn j and

P jIm .zn /j jzn j for all n, we find that the
lemma holds by the comparison test if 1 j D0 jzj j converges. The converse follows
from the fact that
jzn j jRe .zn /j C jIm .zn /j
for all n, by the triangle inequality. t

u
P1
Theorem 3.12.2. Let fzn g be a sequence (not necessarily real). If j D0 jzj j
P
converges, then so does 1
j D0 zj .
Proof. In view of Lemma 3.12.1, we may assume that each zn is real. Thus we write
xn D zn for each n. Certainly,
0 xn C jxn j 2jxn j
P1
for each n. Given that j D0 jxj j converges, so does
1
X
.xj C jxj j/;
j D0
by the comparison test. As

1
X 1
X 1
X
xj D .xj C jxj j/ jxj j
j D0 j D0 j D0
by Theorem 3.3.1, the result follows. t

u
Remark. The converse of the theorem above is not always true. For example, the
alternating series
1
X .1/j
j D1
j
converges by Leibnizs test, but the harmonic series diverges.

P P1
A series 1 j D0 zj is said to be absolutely convergent if j D0 jzj j converges.
Thus every absolutely convergent series does in fact converge, according to
Theorem 3.12.2. A series that is convergent but not absolutely is said to be
conditionally convergent.
Example 3.12.1. The series
X1
ij
j D1
j2
158 3 Series
is absolutely convergent since the series

1 j
X 1
i X 1
D
j2 j2
j D1 j D1
converges. 4
1
X j
.1/j :
j D1
2j
Solution. Note that the series
X1
j
j D1
2j
converges by the ratio test, since
n C 1 2n nC1 1
nC1
D ! < 1:
2 n 2n 2
Hence the given series converges absolutely. 4
The work in this example suggests that the ratio test can be extended to series
whose terms are not necessarily positive or even real. In fact, we can prove the
following theorem.
P
Theorem 3.12.3 (Generalized Ratio Test). Let 1 j D0 zj be a series (not neces-
sarily real) and suppose that

znC1
lim DL
n!1 zn
for some number L. Then the series converges absolutely if L < 1 but diverges
if L > 1.
P
Proof. If L < 1, then the series 1 j D0 jzj j converges by the ratio test, and so
the given series is absolutely convergent. If L > 1, then, arguing as in the proof
of the ratio test, we see that the sequence fjzn jg cannot converge to 0. Hence the
sequence fzn g cannot converge to 0 either. Thus the given series fails the nth term
test and therefore cannot converge. t
u
There is also a generalized root test, which has a similar proof.
3.12 Absolute and Conditional Convergence 159
P1
Theorem 3.12.4 (Generalized Root Test). Let j D0 zj be a series and let
lim jzn j1=n D L

n!1
for some number L. Then the series converges absolutely if L < 1 but diverges if
L > 1.
Proof. If L < 1, then the series is absolutely convergent by the root test. If L > 1,
then, as in the proof of the root test, the sequence fzn g cannot converge to 0. t
u
P1 2
Theorem 3.12.5. Let fxn g and fyn g be real sequences, and suppose that j D0 xj
P P1
and 1 2
j D0 yj converge. Then the series j D0 xj yj is absolutely convergent.
Proof. Note first that
.jxj j jyj j/2 0
for all j . Hence

1 2
.x C yj2 / jxj yj j:
2 j
The result now follows from the comparison test. t
u
Exercises 3.11. 1. Use Cauchys principle to prove that if a complex series is
absolutely convergent,
P then it is convergent. P1 2
2. (a) Show that if 1 j D0 xj converges absolutely, then so does j D0 xj .
(b) Is the converse true?
P P1 2
(c) Is it true that if 1 j D0 xP
j converges, then so does j D0 xj ?
P1
3. Show that if j D0 zj and 1 w are absolutely convergent complex series,
P Pj1D0 j
then so are 1 z
j D0 j jw and j D0 .z j Cw j / for all complex numbers and .
P 1
4. Show that if j D0 zj is an absolutely convergent complex series, then

X
1 X 1
z jzj j:
j
j D0 j D0
5. Test the following series for absolute convergence and conditional conver-
gence:
P1 .1/j C1
(a) p
j D2 j j ;
P1 j
j C1 j C2
(b) j D0 .1/ 3j 1
;
P1 j
j D0 .j C 1/ for all real x 2;
2 x
(c) xC2
P1 j C1 .j i/ 2
(d) j D0 .1/ 2j
;
P1 i j
(e) j D1 j 3=2 .
160 3 Series
3.13 Rearrangements of Series
The notion of a conditionally convergent series was introduced in the previous

section. Part of the reason for this terminology is an unpleasant and perhaps
counterintuitive property: The convergence turns out to depend on the order in
which the terms of the series are written. In fact, we have the following remarkable
theorem, due to Riemann.
Theorem 3.13.1 (Riemann). For each conditionally convergent real series and
given number S , the terms of the series may be rearranged to yield a series that
converges to S . There is also a rearrangement of the terms so that the resulting
series diverges.
Before proving Theorem 3.13.1, we make some important observations. Given a
real sequence fan g, let us define two new sequences fPn g and fQn g, where
jan j C an
Pn D (3.12)
2
and
jan j an
Qn D (3.13)
2
for all n. Thus, if an 0, then Pn D an D jan j and Qn D 0, but ifP an < 0,

1
then Pn D 0 and Qn D an D jan j. We can therefore think of j D1 Pj
P1
and j D1 Qj as the series composed of the nonnegative and negative terms,
P P
respectively, of the series 1 a . It is easy to see that if 1j D1 aj is conditionally
P1 j D1 j P1
convergent, then both j D1 Pj and j D1 Qj diverge. Indeed, we have
jan j D 2Pn an
P
for all n, and so if 1 j D1 Pj were to converge, then it follows from Theorem 3.3.1
P1
that j D1 jaj j would converge. This conclusion would contradict the conditional
P P1
convergence of 1 j D1 aj . The proof that j D1 Qj diverges is similar.
Note also that if n > 1 and x1 ; x2 ; : : : ; xn are numbers such that
X
n1
xj S;
j D1
then
X
n X
n1
xj S D xn C xj S xn : (3.14)
j D1 j D1
3.13 Rearrangements of Series 161
Similarly, if
X
n1
xj S;
j D1
then
X
n X
n1
S xj D S xj xn xn : (3.15)
j D1 j D1
The proof of the theorem involves

P1 some technical details, but the idea is simple.
From the terms of the series j D1 aj , we construct a series converging to S
as follows. First, we take just enough positive terms to obtain a partial sum that
exceeds S . Next, we throw in just enough negative terms to produce a partial sum
below S . Then we add some more positive terms, just enough to bring the partial
sum above S once more. We continue in this fashion, so that the partial sums
oscillate about S . The resulting series converges to S .
Proof of Theorem
P 3.13.1. Let us first try to use the terms of a conditionally conver-
gent series 1 j D1 aj to construct a series that converges to S . For all n > 0 define
Pn and Qn as in (3.12) and (3.13), respectively. Let M1 be the smallest positive
integer such that
X
M1
Pj > S:
j D1
P
Certainly M1 exists: The sequence f nj D1 Pj g is nondecreasing and therefore
cannot be bounded above since it diverges. Next let N1 be the smallest positive
integer such that
X
M1 X
N1
Pj Qj < S I
j D1 j D1
P1
N1 also exists since j D1 Qj diverges. In fact, the choice of N1 shows that
X
M1 X
N 1 1
Pj Qj S I
j D1 j D1
hence
0 1
X
M1 X
N1
0<S @ Pj Qj A .QN1 / D QN1
j D1 j D1
by (3.15).
162 3 Series
Continuing by induction, suppose that n > 0 and that
M1 ; M2 ; : : : ; Mn ; N1 ; N2 ; : : : ; Nn
have been defined so that
0 < S Tn QNn ; (3.16)
where
0 1
X
n1 X
MkC1
X
NkC1
Tn D @ Pj Qj A
kD0 j DMk C1 j DNk C1
and M0 D N0 D 0. Continue by letting MnC1 be the smallest integer greater than

Mn such that
X
MnC1
Tn C Pj > S
j DMn C1
and letting NnC1 be the smallest integer greater than Nn such that
X
MnC1
X
NnC1
TnC1 D Tn C Pj Qj < S I
j DMn C1 j DNn C1
both MnC1 and NnC1 exist. The choice of NnC1 implies that
NnC1 1
X
MnC1
X
Tn C Pj Qj S;
j DMn C1 j DNn C1
so that
0 < S TnC1 QNnC1 ;
by (3.15).
We must show that the resulting series converges to S . This goal is achieved by
studying the partial sum Tn and those between Tn and TnC1 . From the choice of
MnC1 and inequalities (3.16) it follows that
X
m
0 S Tn Pj QNn (3.17)
j DMn C1
3.13 Rearrangements of Series 163
for each m such that Mn < m < MnC1 , since Pj 0 for all j . Moreover the choice
of MnC1 also shows that
X
MnC1
0 < Tn C Pj S PMnC1 ; (3.18)
j DMn C1
by (3.14), whence
X
MnC1
X
m
0 Tn C Pj Qj S PMnC1 (3.19)
j DMn C1 j DNn C1
for all m such that Nn < m < NnC1 . From (3.163.19) we find that the partial sums
in question all differ from S by an amount no greater than
maxfPMnC1 ; QNn g:
P
But the sequence fan g converges to 0 by the convergence of 1 j D1 aj . Hence fjan jg
also converges to 0 and therefore so do its subsequences fPMnC1 g and fQNn g, as
required. P
Its easier to construct a divergent series from the terms of the series 1 j D1 aj .
Defining Pn and Qn as before, let K1 be the smallest positive integer such that
X
K1
Pj Q1 > 1:
j D1
Suppose that K1 ; K2 ; : : : ; Kn have been defined for some positive integer n so that
0 1
X
n1 X
KkC1
@ Pj QkC1 A > n;
kD0 j DKk C1
where K0 D 0. Let KnC1 be the smallest integer greater than Kn such that
0 1
X
n X
KkC1
@ Pj QkC1 A > n C 1:
kD0 j DKk C1
The series defined inductively by this procedure diverges. t

u
By contrast, Dirichlet proved that an absolutely convergent series converges to
the same number regardless of the order in which the terms are written. In order to
make the discussion rigorous, let us first introduce the following definition.
164 3 Series
Definition 3.13.1. Let fkn g be a sequence of nonnegative integers in which each

P Then the sequence fakn g is a rearrange-
nonnegative integer appears exactly once.
ment of a sequence fan g and the series 1 j D0 akj is a rearrangement of the series
P1
a
j D0 j .
Theorem 3.13.2 (Dirichlet). All rearrangements of an absolutely convergent

series are absolutely convergent and converge to the same number.
P1
Proof. Let j D0 wj be a rearrangement of an absolutely convergent series
P1
j D0 z j . For all n 2 N write
X
n
Sn D zj ;
j D0
X
n
Sn0 D jzj j;
j D0
X
n
Tn D wj ;
j D0
and
X
n
Tn0 D jwj j:
j D0
Every convergent
P1 sequence is bounded, and so Pthere exists M such that Sn0 M
1
for all n. As j D0 jwj j is a rearrangement of j D0 jzj j, we also have Tn0 M for
P
all n. Hence 1 j D0 wj is absolutely convergent.
Let
1
X
SD zj
j D0
and
1
X
T D wj :
j D0
It remains only to show that S D T . Choose " > 0. We can find N such that
jS Sn j < "
and
Sm0 Sn0 < "

3.14 Products of Series 165
whenever m > n N . Fix n N and choose p large enough so that
jTp T j < "
and the sum Tp includes z0 ; z1 ; : : : ; zn among its terms. Now choose m > n so large
that every term of Tp is also a term of Sm . Thus Sm Tp is a summation whose
terms form a subset of fznC1 ; znC2 ; : : : ; zm g, so that
jSm Tp j Sm0 Sn0

< ":
Hence
jS T j jS Sm j C jSm Tp j C jTp T j
< 3":
Since " is arbitrary, we conclude that S D T , as required. t

u
3.14 Products of Series
In order to understand products of series, we must first carefully consider the

process of multiplying finite sums. Suppose that am ; amC1 ; : : : ; an ; bp ; bpC1 ; : : : ; bq
are numbers and we wish to form the product
0 10 1
X
n X
q
@ aj A @ bk A : (3.20)
j Dm kDp
In fact, this product may be written without the parentheses,

P as we now show. Using
the distributive law (1.17) twice, first with s D nj Dm aj and then with s D bk ,
gives
0 10 1 0 1
X
n X
q
X
q
X
n
@ aj A @ bk A D @ aj A bk
j Dm kDp kDp j Dm
0 1
X
q
X
n
D @ aj bk A
kDp j Dm
X
q
X
n
D aj bk :
kDp j Dm
166 3 Series
a0 b0 a0 b1 a0 b2 a0 bq1 a0 bq


.. .. .. .. .. ..
. . . . . .
an1 b0 an1 b1 an1 b2 an1 bq1 an1 bq

an b0 an b1 an b2 an bq1 an bq
Pn Pq
Fig. 3.2 The terms of . j Dm aj /. kDp bk /
As multiplication is commutative, we conclude that

0 10 1
X
q
X
n X
q
X
n
aj bk D @ bk A @ aj A
kDp j Dm kDp j Dm
X
n X
q
D aj bk :
j Dm kDp
It now follows that

0 10 1
X
n X
q
X
n X
q
@ aj A @ bk A D aj bk
j Dm kDp j Dm kDp
0 1
X
n X
q
D @aj bk A
j Dm kDp
X
n X q
D aj bk ;
j Dm kDp
where the penultimate line was obtained by another application of distributivity.

In order to understand better how to evaluate the expression (3.20), let us begin by
considering the case where m D p D 0. Basically, the product (3.20) is calculated
by adding up the products of the form aj bk for all relevant j and k. In other words,
we simply add all the entries in Fig. 3.2. In the calculation above, the terms aj bk
are ordered so that all those containing a0 are listed first, then all those containing
a1 , and so forth. In other words, the entries ofPthe figure are added row by row. The
q
sum of the entries in the .j C 1/th row is aj kD0 bk .
3.14 Products of Series 167
However, there is another way of performing this calculation which is of some

importance. Let us assume first that n D q, so that Fig. 3.2 gives a square array.
Instead of ordering the terms in Fig. 3.2 row by row, we may order them as indicated
by the arrows. In other words, we begin with the term a0 b0 in the top left corner,
followed by the terms in the diagonal from a0 b1 to a1 b0 , then those in the diagonal
from a0 b2 to a2 b0 , and so on, up to and including the diagonal that contains a0 bq D
a0 bn , the entry in the top right corner. As the array is assumed to be square, this
diagonal also contains an b0 . We then continue with the diagonal from a1 bn to an b1 ,
then that from a2 bn to an b2 , and so on. The last diagonal consists only of the term
an bn . In each diagonal the sum of the subscripts is a constant between 0 and 2n
(inclusive), and this constant is different from the corresponding constant for each
other diagonal. For the diagonal where this constant is t , the sum of the terms in the
diagonal is
X
t
as bts ;
sD0
provided we agree that as D 0 whenever s > n and bts D 0 whenever t s > n.

With this understanding, we conclude that
X
n X
n X
2n X
t
aj bk D as bts :
j D0 kD0 tD0 sD0
If n > q, then we may set bk D 0 for all k > q and apply the result of the
calculation above. Similarly, if n < q, then we set aj D 0 for all j > n. In both
cases we reach the conclusion that
X
n X
q
XX
nCq
t
aj bk D as bts ;
j D0 kD0 tD0 sD0
since as bts D 0 whenever t D s C .t s/ > n C q. In general,

X
n X
q
X
nm X
qp
aj bk D aj Cm bkCp
j Dm kDp j D0 kD0
X
nmCqp
X
t
D asCm btsCp :
tD0 sD0
Setting v D s C m and u D t C m C p, we therefore obtain
X
n X
q
X X
nCq
up
aj bk D av buv ;
j Dm kDp uDmCp vDm
since t C p s D u m s D u v.
168 3 Series
Using Eq. (1.15), we may summarize the result of this calculation in the
following theorem.
Theorem 3.14.1. For integers m; n; p; q such that m n and p q, we have
p
X
n X
q
X jX
nCq
aj bk D ak bj k ;
j Dm kDp j DmCp kDm
where al D 0 for all l > n and bl D 0 for all l > q. In particular,
X
n X
q
XX
nCq j
aj bk D ak bj k ;
j D0 kD0 j D0 kD0
for all nonnegative integers n and q.

We now turn our attention to the multiplication of absolutely convergent series
1
X
AD aj
j D0
and
1
X
BD bj :
j D0
Thus we need to find the sum of all products of the form aj bk for nonnegative
integers j and k. This sum may be written as the series
1 j 1
!
X X
j
X
aj bk C al bj (3.21)
j D0 kD0 lD0
D a0 b0 C a1 b0 C a1 b1 C a0 b1 C a2 b0 C a2 b1 C a2 b2 C a0 b2 C a1 b2 C :
For each nonnegative integer n let An ; Bn ; Sn be the nth partial sums of A, B, and
the series (3.21), respectively. Thus,
X
n1
An D aj
j D0
and
X
n1
Bn D bj :
j D0
3.15 Introduction to Power Series 169
The product An Bn has n2 terms, and it follows that
Sn2 D An Bn :
For each nonnegative integer n let Tn be the sum of the absolute values of the
first n terms of series (3.21). Since A and B are absolutely convergent, we find that
T n T n2
X
n1 X
n1
D jaj j jbj j
j D0 j D0
1
X 1
X
jaj j jbj j:
j D0 j D0
Thus the series (3.21) converges absolutely. Since fSn2 g is a subsequence of fSn g,
we find that
1 j 1
!
X Xj
X
aj bk C al bj D lim Sn
n!1
j D0 kD0 lD0
D lim Sn2
n!1
D lim An Bn
n!1
D AB:
Because of the absolute convergence of this series, Dirichlets theorem shows that
every rearrangement must also converge absolutely to AB. We have now proved the
following theorem.
P1 P1
Theorem 3.14.2. If j D0 aj and j D0 bj converge absolutely to A and B,
respectively, then
1 X
X j
ak bj k D AB
j D0 kD0
and the convergence is absolute.
3.15 Introduction to Power Series
A power series in a variable z is defined as a series of the form

1
X
cj .z c/j ; (3.22)
j D0
170 3 Series
where c; c0 ; c1 ; : : : are constants, not necessarily real. Throughout this discussion

we shall take 00 to be 1. Thus the first term of the power series is c0 . We refer to c
as the center of the power series.
Each term of a power series is therefore a function of z. Series of functions played
a major role in the development of analysis in the nineteenth century. Their study
was a driving force behind the development of a satisfactory standard of rigor. The
general theory of series of functions is by no means simple.
Our first task is to find all z for which the power series converges. Often the ratio
test is very useful. Let us try some examples.
Example 3.15.1. Discuss the convergence of the series
1
X jx j
;
j D1
.j C 1/2
where x is a real variable.

Solution. Certainly the series converges if x D 0. Suppose x 0. Setting
nx n
an D
.n C 1/2
for all n > 0, we find that

anC1 .n C 1/jxjnC1 .n C 1/2

a D .n C 2/2

njxjn
n
.n C 1/3
D jxj
n.n C 2/2
! jxj
as n ! 1. Hence the series converges absolutely if jxj < 1. If jxj > 1, then there
exists N such that janC1 j > jan j for all n N . Thus the series fails the nth-term test
and therefore diverges. (If x is not real, it is also immediate from these arguments
that the series converges absolutely if jxj < 1 and diverges if jxj > 1.)
It remains to check the cases where x D 1. With x D 1, we obtain the
alternating series
X1
.1/j j
;
j D1
.j C 1/2
whose convergence has already been confirmed in Example 3.10.1.

Suppose therefore that x D 1. Then the series becomes
1
X j
:
j D1
.j C 1/2
We use the limit comparison test, comparing this series with the harmonic series.
Putting an D n=.n C 1/2 and bn D 1=n for all n > 0, we find that
an n2
D !1
bn .n C 1/2
as n ! 1. Hence an bn and our series diverges by the limit comparison test.
We conclude that the given series converges if and only if 1 x < 1. The
convergence is absolute if jxj < 1 but not if x D 1. 4
1
X
j j .z 1/j :
j D0
Solution. The series converges if z D 1. Suppose z 1. Then for all n > 0 we

have
n
.n C 1/nC1 .z 1/nC1
D .n C 1/ n C 1 jz 1j
nn .z 1/n n
> .n C 1/jz 1j
!1
as n ! 1. Hence the series converges if and only if z D 1. 4

1 j
X z
:
j D1
j
The series converges if z D 0. If z 0, then

nC1
z n njzj

n C 1 zn D n C 1 ! jzj

as n ! 1. Hence the series converges absolutely by the ratio test if jzj < 1 and
diverges by the nth-term test if jzj > 1. We have already seen in Example 3.11.1
that the series converges if jzj D 1 and z 1. If z D 1, then the series is harmonic
and therefore diverges. 4
Example 3.15.4. We have already seen in Theorem 2.7.6 that the series
1
X xj
j D0
j
converges to e x for all real x. 4

172 3 Series
Our next theorem shows that there are three possibilities: The power series (3.22)
converges absolutely for all z; it converges only when z D c; or there exists a
positive number r such that the power series converges absolutely if jz cj < r
and diverges if jz cj > r. The number r in the third case is called the radius of
convergence of the power series. We write r D 0 in the second case, and in the
first case we say that the power series has an infinite radius of convergence. Thus
every power series in a real variable has associated with it an interval on which it
converges (provided we consider each real number to constitute an interval and the
set R of all real numbers also to be an interval). This interval is called the interval
of convergence of the power series. In our four examples above, the respective
intervals of convergence are 1; 1/, f1g, 1; 1/, and R. Similarly, a power series
in a complex variable has associated with it a circle of convergence in the interior
of which the power series is absolutely convergent and in the exterior of which it
diverges.
But we are getting ahead of ourselves. The result referred to in the previous
paragraph still needs to be proved. First, though, observe that we can simplify the
expression (3.22) by substituting w D z c. This change of variable shows that
there is no loss of generality in assuming that c D 0.
P
Theorem 3.15.1. For the power series 1 j
j D0 aj z one of the following possibilities
must hold:
1. the series converges only when z D 0;
2. the series is absolutely convergent for all z;
3. there exists r > 0 such that the series converges absolutely whenever jzj < r and
diverges whenever jzj > r.
Proof. Certainly the power series converges when z D 0. We may also choose a
particular w 0 for which it converges, for if no such w exists, then possibility (1)
obtains.
We show first that
1
X
aj zj (3.23)
j D0
converges absolutely for each z such that jzj < jwj. Since
1
X
aj wj
j D0
converges, it follows that
lim jan wn j D 0:
n!1
Thus the sequence fjan wn jg is bounded above by some number M . Therefore

n z n
z
jan zn j D jan wn j n M :
w w
But jz=wj < 1, and so the series

1
X z j

M
j D0
w
converges. Hence
1
X
jaj zj j
j D0
also converges, by the comparison test.

One conclusion to be drawn is that if the series (3.23) converges for all z, then it
converges absolutely for all z, because for every z there exists w, of modulus greater
than jzj, for which the series converges. In this case possibility (2) holds. We may
therefore suppose that there exists z0 for which (3.23) diverges. The series must then
diverge for every w0 such that jw0 j > jz0 j. Therefore the set of moduli of all z such
that (3.23) converges is bounded above by jz0 j, and since it contains 0, it must have
a supremum, r. The series therefore converges absolutely for all z such that jzj < r
(because there exists w for which jzj < jwj r and for which the series converges)
and diverges for all z such that jzj > r by the definition of r. t
u
Example 3.15.5. Find the radius of convergence for the power series
1
X 2 C .i /j
zj :
j D1
j2
Solution. Set
2 C .i /n n
an D z
n2
for each n > 0. If jzj 1, then jan j 3=n2 . Since

X1
3
j D1
j2
converges, the given series is absolutely convergent by the comparison test. Suppose
therefore that jzj > 1. Since
174 3 Series
jzjn
jan j
n2
for all n > 0, it follows from Example 2.5.5 that the sequence fan g does
not converge to 0 and the given series therefore diverges. Hence the radius of
convergence is 1.
Note that the ratio test could not have been applied in this example, since

anC1
lim
n!1 an
does not exist. 4

We conclude this section with some simple observations about the addition and
multiplication
P1 power series. Given numbers s and t and convergent power series
ofP
j 1 j
j D0 aj z and j D0 bj z , Theorem 3.3.1 shows that
1
X 1
X 1
X
s aj zj C t bj z j D .saj zj C t bj zj /
j D0 j D0 j D0
1
X
D .saj C t bj /zj :
j D0
Similarly, Theorem 3.14.2 gives
1
X 1
X 1 X
X j
aj zj bj z j D ak zk bj k zj k
j D0 j D0 j D0 kD0
1 X
X j
D ak bj k zj :
j D0 kD0
P1
Exercises 3.12. 1. Consider the power series j D0 cj .z c/j , where cj 0 for
all j .
(a) If

cnC1
lim D L;
n!1 c
n
prove that the radius of convergence of the series is 1=L (take 1=L D 1 if
L D 0 and 1=L D 0 if L D 1).
3.16 The Exponential, Sine, and Cosine Functions 175
(b) If
lim jcn j1=n D L;

n!1
prove that the radius of convergence of the series is 1=L.

2. Let the radii of convergence of
1
X 1
X 1
X
aj zj ; bj z j ; .aj C bj /zj
j D0 j D0 j D0
be R1 ; R2 ; R, respectively.
(a) Show that if R1 R2 , then R D minfR1 ; R2 g.
(b) Give an example with the property that R > R1 D R2 .
P
3. Can the power series 1 j D0 cj .z 2/ converge at 0 but diverge at 3?
j
4. For each of the following series find all real x for which the series converges:
P1 j xj
P1 1 j
j D0 22j 1 C 32j x ;
1
(a) j D1 .1/ j ; (f)
P1 j j
P1 j j
(b) j D1 j 2 x ; (g) j D0 j j x ;
P1 j x 2j P 1 j .xC4/
j
(c) j D0 .1/ .2j / ; (h) j D1 .1/ j 2j
;
P1 .j /2 j P1 j j
(d) j D0 .2j / x ; (i) j D1 .j C1/.j C2/ x ;
P1 x j P1 x j
(e) p p
j D1 j ; (j) j D1 j j .
5. For each of the following series find all complex z for which the series converges:
P1 j 2 P1 .zCi/j
(a) j D0 z ; (e) j D0 e j ;
P1 .i1/j j P1 j 2j
j D1 .1/ j .z i / ;
j
(b) j D1 j
z ; (f)
P1 j P1 z2j
j D0 2j .z i / ;
j
(c) (g) j D0 .3/j ;
P1
j D0 .3 C .1/ /z .
j j
(d)
P
6. Find the radius of convergence of the series 1 j D0 aj z , where a2j D 1 and
j
a2j C1 D 2 for all j 0.
3.16 The Exponential, Sine, and Cosine Functions
We now use the work of the previous section to generalize our earlier results
concerning the exponential function. Recall that, for real x,
1
X xj
ex D ;
j D0
j
176 3 Series
by Theorem 2.7.6. More generally, for every complex z let us define
X1 j
z
ez D ;
j D0
j
noting that the series converges absolutely by the ratio test, since
nC1
z n z

.n C 1/ zn n C 1 ! 0:
D
With this definition it is easily seen that parts (2) and (3) of Theorem 2.7.10 are valid
also for complex sequences. In addition, we define exp.z/ D e z for all z 2 C, and
refer to exp as the exponential function.
Because of the absolute convergence of the series for e z , we may apply
Theorem 3.14.2 to prove that
0 10 1
1
X X1 j
wj z
ewez D @ A@ A
j D0
j j D0
j
X1 X j
wk zj k
D
j D0
k.j k/
kD0
1
!
X 1 X j
j
D wk zj k
j D0
j k
kD0
1
X .w C z/j
D
j D0
j
D e wCz
for all complex numbers w and z. This result generalizes Theorem 2.7.9. It follows
by an easy induction that
.e z /n D e nz
for every positive integer n.

Now suppose that z D ix for some real x. Then
1 j j
X i x
e ix D
j D0
j
X1 2j 2j C1
j x j x
D .1/ C i.1/ : (3.24)
j D0
.2j / .2j C 1/
Let cos.x/ D Re .e ix / and sin.x/ D Im .e ix /. It follows from Theorem 2.3.11 that

1
X x 2j
cos.x/ D .1/j
j D0
.2j /
and
1
X x 2j C1
sin.x/ D .1/j
j D0
.2j C 1/
for all real x, and the convergence of these power series is absolute. (See also
Example 3.7.2.) Thus
e ix D cos.x/ C i sin.x/
for all real x. We refer to cos and sin as the cosine and sine functions, respectively.
They will be given a geometric interpretation later.
We may generalize the definitions of the sine and cosine functions to complex
numbers. Thus
X1
j z
2j
z2j C1
e D
iz
.1/ C i.1/ j
j D0
.2j / .2j C 1/
D cos.z/ C i sin.z/;
where
1
X z2j
cos.z/ D .1/j
j D0
.2j /
and
1
X z2j C1
sin.z/ D .1/j
j D0
.2j C 1/
for all complex numbers z. Both series are absolutely convergent.

It is clear from the definition that sin.0/ D 0, but cos.0/ D 1 by the convention
that 00 D 1. It is also immediate that
sin.z/ D sin.z/
and
cos.z/ D cos.z/
178 3 Series
for all z. We define
sin.z/
tan.z/ D
cos.z/
for all z for which cos.z/ 0. The function so defined is called the tangent function.
Given a real- or complex-valued function f , let g be the function defined by
1
g.z/ D
f .z/
for all z such that f .z/ is defined and nonzero. The function g is called the
reciprocal of f . The sine, cosine, and tangent functions and their reciprocals are
said to be trigonometric. The reciprocals of the sine, cosine, and tangent functions
are called the cosecant, secant, and cotangent functions, respectively. They are
denoted, respectively, by csc, sec, and cot. The parentheses around the arguments of
the trigonometric functions are usually omitted.
Since
e iz D cos z C i sin z
for all z, we also have
e iz D cos z i sin z:
Thus e iz C e iz D 2 cos z, and so

e iz C e iz e 2iz C 1
cos z D D :
2 2e iz
Similarly,
e iz e iz e 2iz 1
sin z D D :
2i 2ieiz
For instance, if x is real, then
e x C e x
cos ix D
2
and
e x e x i.e x e x /
sin ix D D ;
2i 2
since i 2 D 1. Moreover
.e 2iw 1/.e 2iz C 1/ .e 2iw C 1/.e 2iz 1/

sin w cos z C cos w sin z D iw iz
C
4ie e 4ieiw e iz
e 2i.wCz/ C e 2iw e 2iz 1 C e 2i.wCz/ e 2iw C e 2iz 1
D
4iei.wCz/
e 2i.wCz/ 1
D
2iei.wCz/
D sin.w C z/ (3.25)
and
.e 2iw C 1/.e 2iz C 1/ .e 2iw 1/.e 2iz 1/

cos w cos z sin w sin z D
4e iw e iz 4e iw e iz
e 2i.wCz/ C e 2iw C e 2iz C 1 C e 2i.wCz/ e 2iw e 2iz C 1
D
4e i.wCz/
e 2i.wCz/ C 1
D
2e i.wCz/
D cos.w C z/ (3.26)
for all complex numbers w and z. Thus
sin 2z D sin z cos z C cos z sin z D 2 sin z cos z
and, similarly,
cos 2z D cos2 z sin2 z:
It also follows by induction that if sin z D 0, then sin nz D 0 for every n 2 N, for if
this equation holds for a particular n 2 N, then
sin..n C 1/z/ D sin.nz C z/

D sin nz cos z C cos nz sin z
D 0:
Furthermore
1 D cos 0 D cos.z z/ D cos z cos.z/ sin z sin.z/ D cos2 z C sin2 z

180 3 Series
for all z. If x is a real number, we infer that
j sin xj 1
and
j cos xj 1:
In any case we also have
cos 2z D cos2 z .1 cos2 z/ D 2 cos2 z 1I
hence
1 C cos 2z
cos2 z D :
2
In addition,
p
je iz j D cos2 z C sin2 z D 1:
Thus e iz 0, and it follows that e z 0 for each z 2 C. Notice, however, that
je x e x j
j sin ixj D :
2
We therefore infer from Theorem 2.7.10(1) that the function j sin zj is unbounded.
Similarly, j cos zj is unbounded.
Note that if z D x C iy, where x and y are real, then
sin z D sin.x C iy/

D sin x cos iy C cos x sin iy
1 y i
D .e C e y / sin x C .e y e y / cos x:
2 2
Similarly,
1 y i
cos z D .e C e y / cos x .e y e y / sin x:
2 2
If w and z are numbers such that tan w tan z D 1, then cos w cos z 0 and
sin w sin z
D 1;
cos w cos z
so that
sin w sin z D cos w cos z;
whence
cos.w C z/ D cos w cos z sin w sin z D 0:
Conversely, if cos.w C z/ D 0 cos w cos z, then tan w tan z D 1. If neither

cos w cos z nor cos.w C z/ is equal to 0, then
sin.w C z/
tan.w C z/ D
cos.w C z/
sin w cos z C cos w sin z
D
cos w cos z sin w sin z
sin w cos zCcos w sin z
cos w cos z
D cos w cos zsin w sin z
cos w cos z
tan w C tan z
D :
1 tan w tan z
In particular, if tan z is defined and j tan zj 1, then
2 tan z
tan 2z D :
1 tan2 z
Note also that
sin.z/
tan.z/ D D tan z:
cos.z/
Proposition 3.16.1. 1. If x 0, then
sin x x: (3.27)
2. If x 2 R, then
x2 x2 x4
1 cos x 1 C : (3.28)
2 2 24
Proof. 1. We may assume that x < 1 since sin x 1 for all real x. Then the
sequence fx 2nC1 =.2n C 1/g is decreasing, and we may invoke inequality (3.10)
in the proof of Theorem 3.10.2, with n D 0, to conclude that sin x x.
2. We may assume that x > 0, since cos.x/ D cos x for all x and cos 0 D 1.
182 3 Series
In regard to the first inequality, we may assume also that x < 2, as cos x 1
for all real x. Then the sequence fx 2n =.2n/gn1 is decreasing, since
x 2nC2 .2n/ x2 1
2n D < <1
.2n C 1/ x .2n C 2/.2n C 1/ 3
when n 1 and 0 < x < 2. Consequently,

1
X x 2j
cos x 1 D .1/j
j D1
.2j /
1
X x 2j
D .1/j 1
j D1
.2j /
1
X x 2j C2
D .1/j
j D0
.2j C 2/
x2
;
2
again using inequality (3.10) with n D 0. The first of the required inequalities
follows. p
For the second inequality we may assume that x < 2 3, since cos x 1.
The sequence fx 2n =.2n/gn1 is still decreasing, and we may appeal to
Theorem 3.10.2 to conclude that
x2 x4
0 cos x 1 C :
2 24
t
u
For ease of reference, in the following theorem we summarize some of the more
important results we have proved.
Theorem 3.16.2. Let w and z be complex numbers. Then
iz
cos z D e Ce
iz
1. 2
,
e iz e iz
2. sin z D 2i ,
3. sin.w C z/ D sin w cos z C cos w sin z,
4. cos.w C z/ D cos w cos z sin w sin z,
5. tan.w C z/ D 1tan tan wCtan z
w tan z
if cos w cos z cos.w C z/ 0,
6. sin 2z D 2 sin z cos z,
7. cos 2z D cos2 z sin2 z,
8. tan 2z D 1tan 2 z if cos z cos 2z 0,
2 tan z
9. cos2 z C sin2 z D 1,
10. cos2 z D 1Ccos
2
2z
,
11. je j D 1.
iz
In addition, if x is real, then j sin xj 1 and j cos xj 1.


sin n 1

np np
for all n 2 N and p 2 Q, the comparison test shows that the series
1
X sin j
(3.29)
j D1
jp
is absolutely convergent if p > 1. We now prove that it is conditionally convergent

when 0 0 and each z 1 such that jzj D 1. In particular, if
z D e i D cos 1 C i sin 1;
then jzj D 1 and z 1 since cos 1 1. [If cos 1 D 1, then

p
j sin 1j D 1 cos2 1 D 0:
However, sin 1 > 0 by the remark following Theorem 3.10.2, as the sequence

1
.2n C 1/
is decreasing.] Therefore the series (3.30) converges for this z, and since
zn D e i n D cos n C i sin n;
it follows by Theorem 3.3.2 that the series (3.29) converges.

184 3 Series
It remains to show that this series is not absolutely convergent if 0 0 let us write
j sin nj an C bn
D ;
n 2
where
j sin nj j sin.n 1/j

an D
n
and
j sin nj C j sin.n 1/j

bn D :
n
P P1
It suffices to show that 1 j D1 aj converges and j D1 bj diverges.
In regard to the former series, define
un D j sin nj j sin.n 1/j
and
1
vn D
n
for each n > 0. Then fvn g is a decreasing null sequence. Moreover

X X
n n
u D .j sin j j j sin.j 1/j/
j
j D1 j D1
D j sin nj
1
for all n > 0, by the telescoping property. We conclude from Dirichlets test that
P 1
j D1 aj converges.
P1
In order to verify the divergence of j D1 bj , note first that for all real numbers
and we have
j sin. /j D j sin cos cos sin j

j sin jj cos j C j cos jj sin j
j sin j C j sin j:
Therefore
sin 1 D j sin 1j D j sin.n .n 1//j j sin nj C j sin.n 1/j:
Consequently,
sin 1
bn >0
n
P1
for all n > 0, and the series j D1 bj diverges by comparison with the harmonic
series. 4
1
X sin jx
j D1
j
for any x 2 R. It converges to 0 if sin.x=2/ D 0. Suppose therefore that

sin.x=2/ 0. The trigonometric identity (3.26) implies that
x x x
cos C jx D cos cos jx sin sin jx
2 2 2
and
x x x
cos jx D cos cos jx C sin sin jxI
2 2 2
hence
x x x
2 sin sin jx D cos jx cos C jx :
2 2 2
186 3 Series
Since
x
1
cos jx D cos j x
2 2

1
D cos j x
2

1
D cos .j 1/ C x ;
2
it follows that
X n x x
cos jx cos C jx
j D1
2 2
X n
1 1
D cos .j 1/ C x cos jC x
j D1
2 2

x 1
D cos cos nC x I
2 2
hence
X
n
cos x2 cos..n C 12 /x/
sin jx D
j D1
2 sin x2
for all n 2 N, since sin.x=2/ 0. Consequently,

X
n cos x2 C cos n C 12 x

sin jx
2 sin x2
j D1
1
x :
sin
2
Dirichlets test therefore implies that the series converges. (In fact, the previous
example shows that it is conditionally convergent if x D 1.) 4
Example 3.16.3. Let fan g be a null sequence of real numbers. We show that
sin an
lim D 1:
n!1 an
From the definition we have
X 1 2j
sin an an
D .1/j D 1 C an Rn
an j D0
.2j C 1/
for all n, where

1
X 2j 1
an
Rn D .1/j :
j D1
.2j C 1/
Since fan g is convergent, there is an M such that jan j M for all n. Hence

1

X 2j 1
an
jRn j
.2j C 1/
j D1
X1
M 2j 1

j D1
.2j 1/
X1
M 2j C1
D
j D0
.2j C 1/
< eM :
Thus
sin an
lim D 1 C lim an Rn D 1:
n!1 an n!1
4
Exercises 3.13. 1. If u and v are any complex numbers, prove that
uCv uv
sin u sin v D 2 cos sin
2 2
and
uCv uv
cos u cos v D 2 sin sin :
2 2
Hence show that
2 sin sin D cos. C / cos. /

188 3 Series
and
2 cos sin D sin. C / sin. /
for all real numbers and .

2. Use the inequality

X
n 1
sin j ;
1
j D1 sin 2
for all n 2 N, to verify the convergence of series (3.29) for all rational p > 0.
3. Prove that if fan g is a decreasing null sequence, then
1
X
aj sin j
j D1
and
1
X
aj cos j
j D1
converge.
4. Let fan g be a null sequence of real numbers. Show that
lim cos an D 1
n!1
and
tan an
lim D 1:
n!1 an
5. The functions given by
e x e x
sinh x D
2
and
e x C e x
cosh x D ;
2
for each x, are called the hyperbolic sine and hyperbolic cosine functions,
respectively.
(a) Show that
cosh2 x sinh2 x D 1
for all x.
(b) Let z D x C iy, where x and y are real.
i. Prove the following:
A. sin z D sin x cosh y C i sinh y cos x.
B. j sinh yj j sin zj j cosh yj.
C. j sin zj2 D sin2 x C sinh2 y.
ii. For every c 2 R, show that the function f .z/ D sin z maps the line y D c
to an ellipse and the line x D c to a hyperbola.
6. Express sin i and cos i in the form x C iy where x and y are real.
7. Find the sum
1
X 1 3
sin cos j :
j D0
2j 2
(Hint: Rewrite it as a telescoping series.)

8. Determine the convergence of
X1
1
1 cos :
j D1
j
(Hint: Write each term in terms of the sine function.)

9. Show that j cos zj is unbounded, where z 2 C.
Chapter 4
Limits of Functions
The concept of a limit of a function is central to the study of mathematical analysis.

It generalizes the notion of the limit of a sequence (a function whose domain is a
set of integers). Indeed, the former can be defined in terms of the latter. Unless an
indication to the contrary is given or it is evident that a restriction to real numbers
is required, the domain and range of a given function are assumed to be sets of
complex numbers.
4.1 Introduction
We start by looking at an example. Consider the real functions f; g; h given by
2x 2 2x
f .x/ D g.x/ D h.x/ D
x1
for all x 1, g.1/ D 1, and h.1/ D 2 (see Fig. 4.1). Thus Df D R f1g and
Dg D Dh D R. For each x 1,
2x.x 1/
f .x/ D g.x/ D h.x/ D D 2x:
x1
Let us first contemplate the function f . As x becomes close to 1, f .x/ becomes

close to 2 provided that x 1. More formally, we say that the limit of f , as x
approaches 1, is 2. In this case we write
lim f .x/ D 2:
x!1

192 4 Limits of Functions
f ( x) g( x) h( x)
2 2 2
1 1 1
x x x
1 1 1
Fig. 4.1 Graphs of f , g and h
Likewise limx!1 g.x/ D 2 and limx!1 h.x/ D 2; the values of these functions at 1
are immaterial.
Roughly speaking, we can ensure that the distance between f .x/ and 2 is as
small as we please by choosing x close enough to 1 but distinct from 1. In other
words, given any sequence fsn g in R f1g that converges to 1, the sequence ff .sn /g
converges to 2.
4.2 Definition and Examples
We now define the concept formally.

Definition 4.2.1. Let f be a function, c an accumulation point of Df , and L a
number. We write
lim f .z/ D L (4.1)

z!c
if the sequence ff .sn /g converges to L for every sequence fsn g in Df fcg

converging to c. We call L a limit of f at c. We also say that f .z/ approaches
L as z approaches c, and we write f .z/ ! L as z ! c.
Remark 1. The number c may or may not be a member of Df .
Remark 2. The symbol z in Definition 4.2.1 represents a dummy variable. Thus if
limz!c f .z/ D L, then we also have limw!c f .w/ D L for any other symbol w.
Remark 3. The assumption that c is an accumulation point of Df guarantees the
existence of a sequence in Df fcg that converges to c.
We have seen that a sequence has at most one limit. This observation leads
immediately to the following theorem.
Theorem 4.2.1. A function has at most one limit at a given accumulation point of
its domain.
4.2 Definition and Examples 193
Example 4.2.1. It follows from Theorem 2.7.10(2) that
lim e x D e a
x!a
for all a 2 R. Similarly, Theorem 2.7.10(3) shows that
eh 1 4
lim D 1:
h!0 h
Example 4.2.2. Let f .x/ D i 1=x for each x in the range of the sequence f1=ng.
Because this sequence is null, we see that 0 is an accumulation point of Df . We
now demonstrate that
lim f .x/
x!0
does not exist. For each positive integer n take
1
sn D
4n
and
1
tn D :
4n C 2
Thus
lim sn D lim tn D 0:
n!1 n!1
However,
i 1=sn D i 4n D 1n D 1
and
i 1=tn D i 2.2nC1/ D .1/2nC1 D 1 1:
Therefore the desired limit does not exist. 4

Proposition 4.2.2. Suppose that f .z/ D k for all z 2 Df . Then for every
accumulation point c of Df we have
lim f .z/ D k:
z!c
Proof. Let fsn g be any sequence in Df fcg converging to c. Since ff .sn /g is the
constant sequence fkg, it converges to k. t
u
The next two propositions can be proved in a similar way. In both propositions
we take c to be an accumulation point of Df .
Proposition 4.2.3. Suppose that f .z/ D z for all z 2 Df . Then
lim f .z/ D c:
z!c
Proposition 4.2.4. If f .z/ D jzj for all z 2 Df , then
lim f .z/ D jcj:

z!c
From Example 3.16.3 we obtain the following result.

Proposition 4.2.5.
sin x
lim D 1:
x!0 x
4.3 Basic Properties of Limits
We now present an equivalent formulation of the definition of a limit. Recall that if

c is an accumulation point of Df , then there is an injective sequence fsn g in Df that
converges to c.
Theorem 4.3.1. Let f be a function, c an accumulation point of Df , and L a
number. The following statements are equivalent:
1.
lim f .z/ D LI
z!c
2. for every " > 0 there exists > 0 such that
jf .z/ Lj < "
for each z 2 Df satisfying 0 < jz cj < .

Proof. Suppose that condition (2) holds. Let fsn g be a sequence in Df fcg
converging to c. Choose " > 0. There exists > 0 such that
jf .z/ Lj < "

4.3 Basic Properties of Limits 195
for each z 2 Df satisfying 0 < jz cj < . Moreover there exists N such that
jsn cj <
for all n N . For all such n we have sn 2 Df and sn c by hypothesis, and so
jf .sn / Lj < ":
Hence ff .sn /g converges to L, and we conclude that condition (1) holds.

Conversely, suppose that condition (2) does not hold. Then, for some " > 0,
every > 0 has the property that there exists z 2 Df satisfying 0 < jz cj < and
jf .z/ Lj ". In particular, for each positive integer n we can choose sn 2 Df
such that
1
0 < jsn cj <
n
and jf .sn / Lj ". The sequence fsn g evidently converges to c, by Lemma 2.5.1.
However,
lim f .sn / LI
n!1
otherwise there would exist N such that jf .sn /Lj < " for all n N . We conclude
that condition (1) does not hold. t
u
Remark 1. If we use the notation
Nr .a/ D fz j jz aj < rg
and
Nr .a/ D Nr .a/ fag;
where a is any number and r is a positive (real) number, then Theorem 4.3.1 shows
that
lim f .z/ D L
z!c
if and only if for every " > 0 there exists > 0 such that
f .N .c/ \ Df / N" .L/:
The two parts of Fig. 4.2 depict Nr .a/ in R and C, respectively.

a
r
r
( )
a
Fig. 4.2 Nr .a/ in R and C
Remark 2. We can clearly assume that " < M and < M for some constant M .
And, as in Proposition 2.2.3, we can replace " by k", where k is a positive constant.
Remark 3. If x and c are real numbers, we often write the inequality jx cj < as
c < x < c C :
Example 4.3.1. Let us prove from Theorem 4.3.1 that
lim .z2 C 1/ D 2:
z!1
Proof. Choose " > 0. We need to find a corresponding > 0 such that
jz2 C 1 2j D jz2 1j < k";
for some constant k > 0, whenever 0 < jz 1j < .

Now
jz2 1j D j.z C 1/.z 1/j

D jz C 1jjz 1j
D jz 1 C 2jjz 1j
.jz 1j C 2/jz 1j
< . C 2/
whenever jz 1j < . Thus, if 1, then jz2 1j < 3, and if in addition ",

then it follows that jz2 1j < 3". Hence we can guarantee that jz2 1j < 3" if we
take
D minf1; "g:
4.3 Basic Properties of Limits 197
We conclude that
lim .z2 C 1/ D 2: 4
z!1
Example 4.3.2. Now let us prove that
1
lim D 1:
z!1 z
Proof. Choose " > 0. We need to find > 0 such that

1
1 < k"; (4.2)
z
for some constant k > 0, whenever z 0 and 0 < jz 1j < . If z 0, then

1
1 D j1 zj <
z jzj jzj
whenever
> jz 1j jjzj 1j 1 jzj:
Taking 1=2, we therefore have
1
jzj > 1 ;
2
so that
1
< 2:
jzj
If in addition ", then

1
1 < 2 2":
z
Thus we can guarantee that inequality (4.2) holds with k D 2 if we take

1
D min ;" :
2
We conclude that
1
lim D 1:
z!1 z 4
Motivated by our study of sequences, we say that a function f satisfies the

Cauchy condition at an accumulation point c of Df if, for every " > 0, there
exists > 0 such that
jf .z/ f .w/j < "
for all z and w in N .c/ \ Df .

Theorem 4.3.2. Let c be an accumulation point of the domain of a function f .
Then limz!c f .z/ D L for some number L if and only if f satisfies the Cauchy
condition at c.
Proof. Suppose first that
lim f .z/ D L:
z!c
Choose " > 0. There exists > 0 such that jf .z/ Lj < " for all z 2 Df satisfying
0 < jz cj < . Choose z and w in N .c/ \ Df . Then jf .z/ Lj < " and
jf .w/ Lj < ". Hence
jf .z/ f .w/j jf .z/ Lj C jL f .w/j < 2":
We conclude that f satisfies the Cauchy condition at c.

Conversely, suppose that f satisfies the Cauchy condition at c. Choose " > 0.
There exists > 0 such that
jf .z/ f .w/j < " (4.3)
for all z and w in N .c/ \ Df .

Since c is an accumulation point of Df , we can find an injective sequence fzn g
in Df that converges to c. Choose N such that
0 < jzn cj <
for every n N . Thus zn 2 N .c/ \ Df for each such n, and so
jf .zn / f .zm /j < "
whenever m N and n N . Consequently, the sequence ff .zn /g is Cauchy and

therefore converges to some number L.
We conclude that there exists M1 such that jf .zn / Lj < " for all n M1 .
Now let fwn g be any sequence in Df fcg converging to c. There exists M2
maxfN; M1 g such that 0 < jwn cj < for all n M2 . Choose n M2 N .
Then we also have 0 < jzn cj < , and so jf .wn / f .zn /j < " by the Cauchy
criterion. Therefore
jf .wn / Lj jf .wn / f .zn /j C jf .zn / Lj < 2";
since n M1 , and we deduce that ff .wn /g also converges to L.

Hence
lim f .z/ D L;
z!c
by definition. t
u
Exercises 4.1. 1. Let f be a function, c an accumulation point of Df , and L a
number. Suppose there exists a real K such that
jf .z/ Lj Kjz cj
for all z 2 C. Show that limz!c f .z/ D L.

2. Let > 0 and limx!0 f .x/ D L. Show that limx!0 f .x/ D L.
3. Suppose that limx!c f .x/ D L for some numbers c and L. Show that the
function f is bounded on some neighborhood
p of c.
4. Use Theorem 4.3.1 to prove that limx!1 x D 1.
5. Show that limz!0 .z=z/ does not exist for any z 2 C.
4.4 Algebra of Limits
We shall now write down some theorems that are useful for calculating limits. Most
of them follow from the corresponding theorems for sequences. In each theorem
we shall assume that the limit is being evaluated at an accumulation point for the
domain of the relevant function.
Theorem 4.4.1. Let f and g be functions, let K and L be numbers, and let c be an
accumulation point of Df \ Dg . Suppose that
lim f .z/ D K
z!c
and
lim g.z/ D L:
z!c
Then
1.
lim .f .z/ C g.z// D K C L;

z!c
2.
lim f .z/g.z/ D KL;

z!c
3.
f .z/ K
lim D
z!c g.z/ L
if L 0 and there exists a neighborhood N .c/ such that g.z/ 0 for all
z 2 N .c/ \ Dg .
Proof. 1. It follows from the hypothesis that the sequence ff .sn /g converges to K
for every sequence fsn g in Df fcg converging to c. Similarly, fg.tn /g converges
to L for every sequence ftn g in Dg fcg converging to c.
Choose a sequence fsn g in
Df Cg fcg D .Df \ Dg / fcg
converging to c. Then, for all n,
.f C g/.sn / D f .sn / C g.sn / ! K C L
as n ! 1, and the result follows.

2. The proofs of parts (2) and (3) are similar. t
u
Remark 1. Sometimes the limits in the left-hand sides in the conclusion of the
theorem exist even if limz!c f .z/ or limz!c g.z/ do not. For example, let f .z/ D
1=z and g.z/ D 1=z for all z 0. Neither of these functions has a limit at 0, yet
their sum and quotient do.
Remark 2. The insistence on c being an accumulation point of Df \ Dg is
inserted in order to guarantee that the limitsp in question are well defined.p(See
Theorem 4.2.1.) For instance, if f .x/ D x for all x 0 and g.x/ D x
for all x 0, then 0 is not an accumulation point of Df \ Dg since Df \ Dg D f0g.
In this case the desired limits are undefined.
Corollary 4.4.2. Let n be a positive integer. Using the notation of the theorem, we
have
1.
lim f n .z/ D Ln ;
z!c
2.
1 1
lim D n
z!c f n .z/ L
if L 0 and there exists a neighborhood N .c/ such that f .z/ 0 for all
z 2 N .c/ \ Df .
Using parts (1) and (2) of the theorem, we also obtain the following corollary.
Corollary 4.4.3. Let
X
n
p.z/ D ak zk
kD0
for all z, where n; a0 ; a1 ; : : : ; an are constants. Then
lim p.z/ D p.c/:

z!c
The following theorem is an analog of the sandwich theorem for sequences.

Theorem 4.4.4. Let f; g; h be functions and let c be an accumulation point of the
set Df \ Dg \ Dh . Suppose there exists > 0 such that
f .x/ g.x/ h.x/
for each x 2 Df \ Dg \ Dh for which 0 < jx cj < . If
lim f .x/ D lim h.x/ D L;

x!c x!c
then
lim g.x/ D L:
x!c
Proof. Let fsn g be a sequence in Df fcg converging to c. There exists N such that
0 < jsn cj < for all n N . For every such n we have
f .sn / g.sn / h.sn /:
The hypotheses imply that f .sn / ! L and h.sn / ! L as n ! 1, and the result
now follows from the sandwich theorem. t
u

1
0 x sin jxj
x
for all x 0, and limx!0 jxj D 0, the sandwich theorem for functions
(Theorem 4.4.4) shows that
1
lim x sin D 0:
x!0 x
Similarly,
1
lim x cos D 0:
x!0 x 4
The next theorem is also analogous to a theorem on sequences from which it
follows immediately.
Theorem 4.4.5. If f is a function such that m f .x/ M and limx!c f .x/ D
L, then m L M .
Theorem 4.4.6. Let f be a function and let L and c be numbers such that L is
real. Suppose
lim f .z/ D L 0:
z!c
1. If L > 0, then there exists > 0 such that f .z/ > L=2 for all z 2 N .c/ \ Df .
2. If L < 0, then there exists > 0 such that f .z/ < L=2 for all z 2 N .c/ \ Df .
Proof. 1. There exists > 0 such that
L
jf .z/ Lj <
2
for all z 2 Df for which 0 < jz cj < . For all such z it follows that
L L
f .z/ > L D :
2 2
2. The proof of part (2) is similar. t
u
Theorem 4.4.7. Let u and v be real-valued functions. Define
f .z/ D u.z/ C iv.z/
for each z 2 Du \ Dv . Let c be an accumulation point of Du \ Dv . Then
lim f .z/ D A C iB (4.4)

z!c
if and only if limz!c u.z/ D A and limz!c v.z/ D B.

4.5 One-Sided Limits 203
Proof. If limz!c u.z/ D A and limz!c v.z/ D B, then the result follows from
Theorem 4.4.1. Conversely, suppose that Eq. (4.4) holds. Considering the real and
imaginary parts of f .z/ A iB, we see that
ju.z/ Aj jf .z/ .A C iB/j
and
jv.z/ Bj jf .z/ .A C iB/j:
The proof now follows by an argument similar to the proof of Theorem 2.3.11. t
u
Exercises 4.2. 1. Find the following limits if they exist, giving reasons for any that
do not: p
x1
(a) limx!0 tansinxx
x
; (f) limx!1 x1 ;
(b) limx!0 sin x
sin x
for 0; (g) limx!0 sin cx
x
for c 2 R;
x n c n
(c) limx!c xc for n 2 N and c 2 R; (h) cos x1
limx!0 x ;
limz!0 .z1/z 1 ;
2 2 C1
(d) (i) limz!i zzi ;
e ix 1
(e) limx!0 x sin x1 C cos x1 ; (j) limx!0 x .
2. If z D x C iy, where x and y are real, find limz!0 .x 2 =z/.
4.5 One-Sided Limits
In this section we restrict the domain of our function to a subset of R. For instance,
sometimes a function does not have a limit at a point a, but it would if the domain
were restricted to a set of numbers greater than a.
Example 4.5.1. Suppose that
(
1 if x > a;
h.x/ D
0 if x a.
The function h is known as a Heaviside function. If we restrict x to the interval

a; 1/, we obtain a function f given by
(
1 if x > a;
f .x/ D
0 if x D a.
Then
lim f .x/ D 1:
x!a
We say that the limit of h at a from the right is 1, and write
lim h.x/ D 1:
x!aC
Likewise the limit of h at a from the left is 0. We write
lim h.x/ D 0:
x!a
Notice that
lim h.x/ lim h.x/:

x!aC x!a
We shall see later that this observation implies that limx!a h.x/ does not exist. We
can also prove this fact by considering the sequences fsn g and ftn g, where
1
sn D a C
n
and
1
tn D a
n
for all n > 0. Then
lim sn D lim tn D a;
n!1 n!1
but h.sn / D 1 and h.tn / D 0 for all n. Therefore limx!a h.x/ does not exist. 4
Example 4.5.2. Let bxc be the largest integer less than or equal to a real number x.
The function f given by f .x/ D bxc for all x is called the floor of x (see Fig. 4.3).
For every n 2 N we have
lim bxc D n
x!nC
and
lim bxc D n 1: 4
x!n
Example 4.5.3. It is easy to show from the definition that

p
lim x D 0:
x!0C
4.5 One-Sided Limits 205
Fig. 4.3 Graph of the floor

function
1
1
x
1 2
It is also true that

p
lim x D 0;
x!0
p
even though x is not defined for x < 0. 4
Let us now define the notion of a one-sided limit formally.
Definition 4.5.1. Let f be a function of a real variable. Let a be an accumulation
point of fx 2 Df j x < ag. Then
lim f .x/ D L
x!a
if the sequence ff .sn /g converges to L for each sequence fsn g in fx 2 Df j x < ag

that converges to a.
The notation
lim f .x/ D L
x!aC
can be defined in a similar way.

The following theorem is analogous to Theorem 4.3.1 and can be proved in a
similar manner.
Theorem 4.5.1. Let f be a function of a real variable and let a and L be
numbers.
1. If a is an accumulation point of fx 2 Df j x < ag, then
lim f .x/ D L
x!a
if and only if for every " > 0 there exists > 0 such that jf .x/ Lj < " for all
x 2 Df satisfying 0 < a x < .
2. If a is an accumulation point of fx 2 Df j x > ag, then
lim f .x/ D L
x!aC
if and only if for every " > 0 there exists > 0 such that jf .x/ Lj < " for all
x 2 Df satisfying 0 < x a < .
Note that 0 < a x < if and only if a < x < a and that 0 < x a < if
and only if a < x < a C .
Theorem 4.5.2. Let f be a function and L a number. Let a be an accumulation
point of fx 2 Df j x > ag and fx 2 Df j x < ag. Then
lim f .x/ D L (4.5)

x!a
if and only if limx!aC f .x/ and limx!a f .x/ exist and
lim f .x/ D lim f .x/ D L: (4.6)

x!aC x!a
Proof. Suppose first that Eq. (4.5) holds, and choose " > 0. There exists > 0 such
that
jf .x/ Lj < " (4.7)
for all x 2 Df such that 0 < jx aj < . In particular, inequality (4.7) holds for all
x 2 Df such that 0 < x a < . Therefore
lim f .x/ D L:
x!aC
lim f .x/ D L:
x!a
Conversely, suppose that Eq. (4.6) holds, and choose " > 0. There exist 1 > 0
and 2 > 0 such that inequality (4.7) holds for all x 2 Df satisfying either 0 <
x a < 1 or 0 < a x < 2 . Thus inequality (4.7) holds for all x 2 Df such that
0 < jx aj < minf1 ; 2 g:
Consequently, Eq. (4.5) holds. t

u
The work in Sect. 4.4 also holds for one-sided limits.
Example 4.5.4. Let f be the function given by
f .x/ D e 1=x
4.6 Infinite Limits 207
for all x 0. Then limx!0C f .x/ does not exist, for if we set sn D 1=n for all
n 2 N, then sn ! 0 as n ! 1 but
f .sn / D e n ! 1
as n ! 1.
On the other hand, for each x > 0 we have x < e x . Thus if t < 0, then
1
0< < e 1=t ;
t
whence
0 < e 1=t < t:
It therefore follows from the sandwich theorem that
lim e 1=t D 0: 4
t!0
Exercises 4.3. 1. Find the following limits if they exist, giving reasons for any that
do not: p
2 C2x3
(a) limx!1 x jx1j ; (e) limx!0C xpx x ;
p
sin x
(b) limx!1 p1x ; (f) limx!0C ;
1x x
1 .xC1/1=3 1
(c) limx!0C 1C21=x
; (g) limx!0C x
;
1 1
(d) limx!0 1C21=x
; (h) limx!1 bxc1
.
4.6 Infinite Limits
Some functions increase or decrease without bound as their arguments approach a

given number. We now define this concept formally.
Definition 4.6.1. Let f be a function and c an accumulation point of Df . We write
lim f .x/ D 1
x!c
if f .sn / ! 1 as n ! 1 for every sequence fsn g in Df fcg converging to c.

The equation
lim f .x/ D 1
x!c
may be defined similarly.

Thus
lim f .x/ D 1
x!c
if and only if for every N there exists > 0 such that f .x/ > N for all x 2 Df
satisfying 0 < jx cj < . Similarly,
lim f .x/ D 1
x!c
if and only if this condition holds with the inequality f .x/ > N replaced
by f .x/<N .
We now discuss the behavior of a function f as its argument approaches 1.
Definition 4.6.2. Let f be a function and L a number. Suppose that Df is not
bounded above. We write
lim f .x/ D L
x!1
if f .sn / ! L as n ! 1 for every sequence fsn g in Df such that sn ! 1 as

n ! 1.
The notation
lim f .x/ D L
x!1
may be defined in a similar manner.

Note that if Df is not bounded above, then
lim f .x/ D L
x!1
if and only if for every " > 0 there exists M such that jf .x/Lj < " for all x 2 Df
satisfying x > M . Moreover
lim f .x/ D L
x!1
if and only if this condition holds with the inequality x > M replaced by x < M ,
provided that Df is not bounded below (though it may be bounded above).
Of course, we also have such notations as
lim f .x/ D 1 (4.8)

x!1
and
lim f .x/ D 1;

x!1
and so forth. We leave it to the reader to list the possibilities and to formulate
appropriate definitions for them. For example, Eq. (4.8) holds if and only if Df
is not bounded above and for every sequence fsn g in Df such that
lim sn D 1;
n!1
we have
lim f .sn / D 1:
n!1
Thus it follows from Theorem 2.7.10(1) that
lim e x D 1:
x!1
Similarly,
lim e x D 0;
x!1
by Example 2.8.2.
Example 4.6.1. Prove that
1
lim D 0:
x!1 x
Proof. Let fsn g be a sequence of positive numbers such that limn!1 sn D 1. Then
1
lim D 0;
n!1 sn
by Theorem 2.8.4(2). The result follows. 4

Example 4.6.2. We have seen in Example 2.7.8 that

1 n
lim 1 C D e:
n!1 n
We now show that limx!1 f .x/ D e, where

x
1
f .x/ D 1 C
x
for all x 2 Q 1; 0. (Later we shall extend this result to the set R 1; 0.)
First let fsn g be any sequence in Q such that sn 1 for all n and
lim sn D 1:
n!1
For each k let nk D bsk c. Then
0 < nk sk < nk C 1:
Thus
1 1 1
< ;
nk C 1 sk nk
and so
nk
1 1 sk 1 nk C1
1C < 1C < 1C :
nk C 1 sk nk
Letting nk ! 1 and using Lemma 2.2.1, we obtain
nk nk C1 1
1 1 1
1C D 1C 1C !e1De
nk C 1 nk C 1 nk C 1
and
nk C1
1 1 nk 1
1C D 1C 1C ! e 1 D e:
nk nk nk
Hence
sk
1
lim 1C D e:
k!1 sk
We conclude that

1 x
lim 1C D e:
x!1 x
In order to show that

1 x
lim 1C D e;
x!1 x
choose a sequence fxn g in Q such that
lim xn D 1
n!1
and xn < 1 for all n. Let yn D xn for all n. Then
lim yn D 1
n!1
and yn > 1 for all n, and so

1 xn 1 yn
1C D 1
xn yn

yn 1 yn
D
yn
yn
yn
D
yn 1
yn
1
D 1C
yn 1
yn 1
1 1
D 1C 1C
yn 1 yn 1
!e
as n ! 1. 4
Theorem 4.6.1. Let f and g be functions, let L be a number, and let a be an
accumulation point of both fx 2 Df j x > ag and fx 2 Dg j x > ag. Suppose that
lim f .x/ D L > 0

x!aC
and
lim g.x/ D 0:
x!aC
Suppose also that there exists > 0 such that g.x/ > 0 for all x 2 Dg satisfying
0 < x a < . Then
f .x/
lim D 1:
x!aC g.x/
Proof. Choose N > 0. The first limit given shows the existence of 1 > 0 such that
L
f .x/ >
2
for all x 2 Df satisfying 0 < x a < 1 . It also follows from the second hypothesis
that there exists 2 > 0 such that
L
jg.x/j <
2N
for all x 2 Dg for which 0 < x a < 2 .

Now let 3 D minf; 1 ; 2 g, so that 3 > 0, and choose x 2 Df \ Dg such that

0 < x a < 3 . Then
L
f .x/ > > 0:
2
Moreover
L
0 < g.x/ D jg.x/j < ;
2N
so that
1 2N
> > 0:
g.x/ L
Hence
f .x/ L 2N
> D N;
g.x/ 2 L

u
Example 4.6.3. It follows immediately from Theorem 4.6.1 that
1
lim D 1: 4
x!0C x
Theorem 4.6.2. Let f and g be functions and let a be an accumulation point of

Df g . If
lim f .x/ D L;
x!1
for some number L, and
lim g.x/ D 1;
x!a
then
lim f .g.x// D L:
x!a
Proof. Note that a is an accumulation point of Dg , since Df g Dg .

Choose " > 0. By hypothesis, there exists M such that jf .x/ Lj < " for all
x 2 Df satisfying x > M . Similarly, there exists > 0 such that g.x/ > M for all
x 2 Dg for which 0 < jx aj < . Choose x 2 Df g such that 0 < jx aj < .
Then x 2 Dg , so that g.x/ > M , and g.x/ 2 Df . Therefore
jf .g.x// Lj < ";

u
Clearly, a and L may each be replaced by 1 or 1. Moreover the limits as x
approaches a may be replaced with one-sided limits. Thus Example 4.6.3 yields the
following corollary.
Corollary 4.6.3. If limx!1 f .x/ D L, then

1
lim f D L:
x!0C x
Corollary 4.6.4. If limx!a g.x/ D 1, where a is an accumulation point of
fx 2 Dg j g.x/ 0g;
then
1
lim D 0:
x!a g.x/
Proof. Define f .x/ D 1=x for all x 0. Then limx!1 f .x/ D 0, by

Example 4.6.1. It therefore follows from Theorem 4.6.2 that
1
lim D lim f .g.x// D 0: t
u
x!a g.x/ x!a
Once again a may be replaced by 1 or 1 and the limits may be replaced by

one-sided limits.
Example 4.6.4. From Example 4.6.3 and Theorem 4.6.2 we have
lim e 1=x D 1:
x!0C
Consequently
1
lim D 0;
x!0C e 1=x C 1
using Corollary 4.6.4. On the other hand, Example 4.5.4 shows that
1 1 4
lim D D 1:
x!0 e 1=x C1 0C1
Example 4.6.5. We show that limx!0 f .x/ D e, where
f .x/ D .1 C x/1=x
for all nonzero rational numbers x > 1. Since

1 x
lim 1 C De
x!1 x
by Example 4.6.2, it follows from Corollary 4.6.3 that
lim .1 C x/1=x D e:
x!0C
lim .1 C x/1=x D e:
x!0
The result now follows from Theorem 4.5.2. 4

Exercises 4.4. 1. Find the following limits: p
xx
(a) limx!1 cosx2x ; (e) limx!1 p
xCx
;
p x 1
(b) limx!1 ; (f) limx!0 pjxj ;
p4x 2 C5xC1
1 x
(c) limx!1 . x CxC
2 1 1/; (g) limx!1 1 C x
;
p
1x
(d) limx!1 x sin x1 ; (h) limx!1 x1
.
2. Give examples of functions f and g to show that if
lim f .x/ D lim g.x/ D 0

x!0 x!0
then
f .x/
lim
x!0 g.x/
may be any real number or 1 or may not exist.

Chapter 5
Continuity
5.1 Definition and Examples
We now come to a most important concept in analysis, one that has many
applications, notably in optimization. Roughly speaking, we are talking about those
functions that have no gaps in their graphs. Such functions are said to be continuous.
For example, the functions f and g, such that
2x 2 2x
f .x/ D g.x/ D
x1
for all x 1 and g.1/ D 1, are not continuous at 1 since the graphs of both
functions have a gap where x D 1. On the other hand, the function h, such that
h.x/ D f .x/ for all x 1 and h.1/ D 2, is continuous (see Fig. 5.1).
The function given by
k.x/ D e 1=x ;
for all x 0, is not continuous at 0, for Example 4.5.4 shows that limx!0C e 1=x D
1 but limx!0 e 1=x D 0, and so the graph of k will have a gap no matter how we
define k.0/ (see Fig. 5.2).
Similarly, the function given by
1
;
e 1=xC1

216 5 Continuity
f (x) g(x) h(x)
2 2 2
1 1 1
x x x
1 1 1
Fig. 5.1 Graphs of f , g and h
Fig. 5.2 Graph of e 1=x k(x)
for all x 0, is not continuous at 0 no matter how we define its value at 0 because
Example 4.6.4 shows that
1
lim D0
x!0C e 1=x C 1
but
1
lim D1
x!0 e 1=xC1
(see Fig. 5.3).

A function f fails to be continuous at a certain point if its limit at that point is not
equal to its value there. In particular, f is not continuous at a given point if either
it or its limit is not defined at that point. We now make the concept of continuity
precise.
Definition 5.1.1. A function f is continuous at an accumulation point c of Df if
lim f .z/ D f .c/:

z!c
In view of the fact that jf .z/ f .c/j D 0 when z D c, the following theorem is
immediate from Theorem 4.3.1.
Fig. 5.3 Graph of

1=.e 1=x C 1/
1
1
2
Theorem 5.1.1. A function f is continuous at an accumulation point c of Df if

and only if for all " > 0 there exists > 0 such that
jf .z/ f .c/j < "
for each z 2 Df satisfying jz cj < .

Of course, the notion of continuity can also be formulated in terms of sequences.
Theorem 5.1.2. Let f be a function and c an accumulation point of Df . Then f
is continuous at c if and only if ff .sn /g converges to f .c/ for every sequence fsn g
in Df converging to c.
The next theorem follows immediately from Theorem 4.4.7.
Theorem 5.1.3. Let u and v be real-valued functions, and define
f .z/ D u.z/ C iv.z/
for each z 2 Du \ Dv . Then f is continuous at an accumulation point c of Du \ Dv

if and only if u and v are continuous at c.
It is an immediate consequence of Propositions 4.2.2 and 4.2.3 that constant
functions and the identity function (the function that maps each number to itself)
are continuous everywhere, as is the absolute value function, by Proposition 4.2.4.
It follows from Theorem 4.4.1 that the sum or product of functions f and g that
are continuous at a point c is also continuous at c if c is an accumulation point
of Df \ Dg . Thus polynomial functions are continuous everywhere. The quotient
f =g is also continuous at c provided that in addition g.x/ 0 for all x in some
neighborhood of c. Theorem 2.3.13 shows that for each m 2 N the function given
by x 1=m for all x 0 is continuous at each such x.
Sometimes a function is not continuous but its product with another function
is continuous. For example, the functions given by e 1=x and e 1=x , for all x 0,
are not continuous at 0. However, e 1=x e 1=x gives the constant function 1 for all
x 0, and if we define its value at 0 to be 1, then the resulting function is continuous
everywhere.
218 5 Continuity
Theorem 5.1.4. Let f and g be functions and let c and L be numbers. Suppose that
c is an accumulation point of Df g , that limz!c g.z/ D L, and that f is continuous
at L. Then

lim f .g.z// D f .L/ D f lim g.z/ D lim f .g.z//:
z!c z!c g.z/!L
Proof. As c is an accumulation point of Df g and Df g Dg , it is also an

accumulation point of Dg .
Choose " > 0. There exists 1 > 0 such that
jf .z/ f .L/j < "
for all z 2 Df satisfying jz Lj < 1 . Similarly, there exists > 0 such that
jg.z/ Lj < 1
for all z 2 Dg satisfying 0 < jz cj < . For each such z we therefore have
jf .g.z// f .L/j < "
if g.z/ 2 Df , and the first equation follows. The proof is completed with the
observation that f .L/ is equal to the two rightmost expressions, respectively, by
substitution and the continuity hypothesized for f . t
u
Remark. Clearly, the limits as z approaches c may be replaced by one-sided limits.
Moreover the accumulation point c of Df g may be replaced by 1 or 1.
Corollary 5.1.5. If g is continuous at an accumulation point c of Df g and f is
continuous at g.c/, then f g is continuous at c.
Proof. Apply the theorem with L D g.c/. t
u
Corollary 5.1.6. Let f be a function and L and c numbers. Suppose that c is an
accumulation point of Df . If limz!c f .z/ D L, then
lim jf .z/j D jLj:

z!c
Corollary 5.1.7. Let f be a real function and L and a real numbers. Suppose that
a is an accumulation point of Df and
lim f .x/ D L:
x!a
Then, for every positive integer m,
lim .f .x//1=m D L1=m

x!a
if for each > 0 there exists x 2 N .a/ \ Df such that f .x/ 0.

The next result is a corollary of Theorem 4.4.6.
Theorem 5.1.8. Let f be a real-valued function that is continuous at a number c.
1. If f .c/ > 0, then there exists > 0 such that f .z/ > 0 for all z 2 N .c/ \ Df .
2. If f .c/ < 0, then there exists > 0 such that f .z/ < 0 for all z 2 N .c/ \ Df .
Example 4.2.1 establishes the continuity of the exponential function.
Theorem 5.1.9. The function e x is continuous at all x.
We now give a famous example of a discontinuous function that is due to
Dirichlet.
Example 5.1.1. Let
(
1 if x is rational;
f .x/ D
0 otherwise.
We show that f is not continuous anywhere. First let c be a rational number. By the
density theorem, for each n 2 N we can find an irrational number

1 1
sn 2 c ; c C :
n n
Thus f .sn / D 0, and so
lim f .sn / D 0 f .c/;

n!1
yet the sequence fsn g converges to c. Consequently, f is not continuous at c.

The case where c is an irrational number can be handled in a similar manner. 4
The following function is a modification of the Dirichlet function. It is known as
Thomaes function or the popcorn function.
Example 5.1.2. Let f W .0; 1/ ! R be defined by
(
0 if x is irrational;
f .x/ D 1
n
if x D mn , where m and n are relatively prime positive integers.
We show that f is continuous at irrational points but not at rational points.

220 5 Continuity
Let c be a rational number in .0; 1/. As in the previous example, we can construct
a sequence fsn g of irrational numbers in .0; 1/ such that sn ! c. Hence f .sn / D 0
for all n. But f .c/ 0; hence f is not continuous at c.
Now let c be an irrational number in .0; 1/. Choose any " > 0 and let

m 1
BD 2 .0; 1/ m 2 N; n 2 N; " :
n n
Then B is finite, as m < n 1=", and c B. Choose > 0 such that N .c/
.0; 1/ and
N .c/ \ B D ;:
Hence f .x/ < " for all x 2 N .c/. Consequently,
lim f .x/ D 0 D f .c/:

x!c
We conclude that f is continuous at c. 4

However, we can prove the following theorem. The proof is due to Volterra
(1881).
Theorem 5.1.10. There is no function f W .0; 1/ ! R that is continuous at all
rational numbers in .0; 1/ but not at any irrational number.
Proof. Let gW .0; 1/ ! R be defined by
(
g.x/ D 1
n
if x D m
n
, where m and n are relatively prime positive integers.
We have shown in Example 5.1.2 that g is discontinuous on A D Q \ .0; 1/ and

continuous on B D .0; 1/ A.
Suppose there is a function f W .0; 1/ ! R that is continuous on A but
discontinuous on B. Let c 2 A. Then f is continuous at c. Hence there exists
> 0 such that .c ; c C / .0; 1/ and
1
jf .x/ f .c/j <
4
for all x 2 .c ; c C/. We may assume that 1=4. Choose a1 ; b1 2 .c ; c C/

such that a1 < b1 ; hence b1 a1 < 1=2. Then for all x; y 2 .a1 ; b1 / we have
1 1 1
jf .x/ f .y/j jf .x/ f .c/j C jf .y/ f .c/j < C D : (5.1)
4 4 2
Now g is continuous at some point in .a1 ; b1 /. Therefore by the preceding argument

with .0; 1/ replaced by .a1 ; b1 /, there exist a10 ; b10 2 .a1 ; b1 / such that
1
jg.x/ g.y/j < (5.2)
2
for all x; y 2 .a10 ; b10 /.

In summary, inequalities (5.1) and (5.2) both hold for all x; y 2 .a10 ; b10 /.
Moreover .a10 ; b10 / contains a point at which f is continuous and a point at which g is
continuous. Therefore we may repeat the argument inductively to obtain a sequence
of nested intervals
.a10 ; b10 /
.a20 ; b20 /
: : :
in which each interval .an0 ; bn0 / has length less than 1=2n and satisfies the condition
that
1
jf .x/ f .y/j <
2n
and
1
jg.x/ g.y/j <
2n
whenever x; y 2 .an0 ; bn0 /. Moreover the sequence fan0 g is increasing and bounded
above. Therefore it converges. Similarly, fbn0 g converges. Since
1
0 < bn0 an0 <
2n
for all n, it follows from the sandwich theorem that fan0 g and fbn0 g both converge to
some number d that belongs to every interval in the sequence. Therefore for every
n 2 N and every x 2 .an0 ; bn0 /, it follows that
1
jf .x/ f .d /j <
2n
and
1
jg.x/ g.d /j < :
2n
It is now easy to show that f is continuous at d . Choose " > 0. Since 1=2n ! 0
as n ! 1, there exists N such that 1=2n < " for all n N . Choose x such that
0
jx d j < minfd aN ; bN0 d g:
222 5 Continuity
0
Then x 2 .aN ; bN0 /, and so
1
jf .x/ f .d /j < < ";
2N
as required. A similar argument shows that g is continuous at d . Hence we reach
the contradiction that d 2 A \ B D ;. t
u
Exercises 5.1.
1. Study the continuity of each of the following functions:
(
jxj
if x 0;
(a) f .x/ D x
0 if x D 0.
(
x 2
if x 1;
(b) f .x/ D
2 x if x > 1.
( ix
i.e 1/
if x 0;
(c) f .x/ D x
1 if x D 0.
(
x if x is rational;
(d) f .x/ D
0 if x is irrational.
2. Find c for which the following functions are continuous:
(
x sin x1 if x 0;
(a) f .x/ D
c if x D 0.
(
x x C 1 if x 1;
2
(b) f .x/ D
cx2 C 1 if x > 1.
3. Let f be continuous on an interval I . Suppose that f .r/ D r 2 for every rational
number r in I . Prove that f .x/ D x 2 for all x 2 I .
4. Let f and g be continuous functions. Show that the functions f _ g and f ^ g
are also continuous, where
.f _ g/.x/ D maxff .x/; g.x/g
and
.f ^ g/.x/ D minff .x/; g.x/g
for all x.
(Hint:
a C b C ja bj
maxfa; bg D
2
5.2 One-Sided Continuity 223
and
a C b ja bj
minfa; bg D :/
2
5. A subset S of C is open if for every c 2 S there exists > 0 such that N .c/
S . Let f W C ! C be continuous. Show that if S is open in C, then
fz 2 C j f .z/ 2 S g
is open.
5.2 One-Sided Continuity
The concept of one-sided continuity is analogous to that of a one-sided limit.

Definition 5.2.1. A function f is continuous on the right at an accumulation point
c of f x 2 Df j x > c g if
lim f .x/ D f .c/:

x!c C
The definition of a function that is continuous on the left at c is analogous.
Example 5.2.1. If
1
f .x/ D
e 1=x C1
for all x 0, then we have already seen that
lim f .x/ D 0
x!0C
and
lim f .x/ D 1:
x!0
Therefore, if we were to define f .0/ D 0, then f would be continuous on the right

at 0, and if we were to define f .0/ D 1, then f would be continuous on the left
at 0. 4
Example 5.2.2. Let f be the function given by
p
f .x/ D x
224 5 Continuity
for all x 0. Then f is continuous on the right at 0. It is not continuous on the left
at 0 since it is undefined at each x < 0. Nevertheless f is continuous at 0 according
to our definition. 4
5.3 Continuity over an Interval
Definition 5.3.1. A function is continuous over a given set if it is continuous at all

points in the set.
We shall show that if a function is continuous over a closed set, then it reaches a
maximum and a minimum somewhere in that set.
In many applications the domain of a function is assumed to be closed and
bounded. A closed bounded set is said to be compact.
Theorem 5.3.1. Let f W X ! C be a continuous function, where X is a compact
subset of C. Then f .X / is compact.
Proof. Suppose that f .X / is not bounded. Choose w 2 f .X /. For every n 2 N, we
can find wn 2 f .X / such that
jwn wj > n:
There exists zn 2 X such that f .zn / D wn . As X is bounded, Theorem 2.6.10 shows

that fzn g contains an injective convergent subsequence fzkn g. Let
z D lim zkn :
n!1
Since X is closed, we have z 2 X .

The continuity of f shows that
lim wkn D lim f .zkn / D f .z/;

n!1 n!1
by the remark following Theorem 5.1.4. Thus there exists N such that
jwkn f .z/j < 1
for all n N . For each such n it follows that
n kn
< jwkn wj
jwkn f .z/j C jf .z/ wj
< 1 C jf .z/ wj:
5.3 Continuity over an Interval 225
Fig. 5.4 Graph of f in f(x)

Example 5.3.1
1
1
x
0 1
1
However this conclusion is impossible, as the right-hand side of the inequality above
is a constant. Consequently f .X / is bounded.
In order to show that f .X / is closed, choose a limit point w of f .X /. There exists
a sequence fwn g in f .X / such that limn!1 wn D w. For each n there exists zn 2 X
such that f .zn / D wn . As before, fzn g contains a subsequence fzkn g converging to a
number z 2 X . Theorem 2.4.1 and the continuity of f show that
w D lim wkn D f .z/:

n!1
Thus w 2 f .X /, and we conclude that f .X / is closed. t

u
Corollary 5.3.2 (Maximum- and Minimum-Value Theorem). Let f W X ! R be
a continuous function, where X is a compact set. Then f .X / contains a maximum
and a minimum value.
Proof. The result follows from the theorem since the supremum and infimum of
f .X / are necessarily limit points. u
t
In particular, a function that is continuous on a closed interval reaches a
maximum and a minimum value in that interval. If a function is not continuous
on a closed interval, then it may or may not have a maximum or a minimum on that
interval.
Example 5.3.1. Let
(
x if 1 < x < 1;
f .x/ D
0 otherwise.
Here f is not continuous on the closed interval 1; 1 although it is continuous on

the open interval .1; 1/. In this case f has neither a maximum nor a minimum on
1; 1, since there is no largest number less than 1 and no smallest number greater
than 1. Note that neither 1 nor 1 is a value of the function (see Fig. 5.4). 4
Example 5.3.2. Let

(
x if 1 < x < 1;
f .x/ D
2 if jxj 1.
226 5 Continuity

Example 5.3.2
2
1
x
0 1

Example 5.3.3
1
1
x
0 1
Here f is not continuous on the closed interval 1; 1, though it is continuous on
the open interval .1; 1/. It has no minimum on 1; 1, but it has a maximum of 2,
attained at 1 (see Fig. 5.5). 4
It is not only discontinuities at endpoints of intervals that may invalidate the
conclusion of the theorem.
Example 5.3.3. Let
(
x if x 0;
f .x/ D
1x if x > 0.
This function is not continuous on 1; 1 as it is not continuous at 0. It has a

minimum of 1 on 1; 1, attained at 1, but no maximum (see Fig. 5.6). 4
Our next theorem asserts the existence of a zero of a continuous function where
the values of the function are not all of the same sign.
Theorem 5.3.3 (Bolzano). Suppose that f W a; b ! R is a continuous function
such that f .a/f .b/ < 0. Then there exists 2 .a; b/ such that f ./ D 0.
Proof. We assume that f .a/ < 0 and f .b/ > 0, as the proof in the other case is
similar.
Suppose there is no 2 .a; b/ such that f ./ D 0. Let a0 D a and b0 D b.

Suppose now that aj and bj have been defined for some nonnegative integer j ,
that they are both in a; b, that f .aj / < 0, that f .bj / > 0, and that bj aj D
.b a/=2j . If

aj C bj
f > 0;
2
then let aj C1 D aj and bj C1 D .aj C bj /=2; otherwise define aj C1 D .aj C

bj /=2 and bj C1 D bj . Then fan g is a nondecreasing sequence in a; b, fbn g is a
nonincreasing sequence in a; b, and for all n we have f .an / < 0, f .bn / > 0 and
ba
bn an D :
2n
As the sequences fan g and fbn g are bounded and monotonic, they both converge.
Moreover their limits are equal, since bn an ! 0. Let
lim an D lim bn D :
n!1 n!1
Since f is continuous, we have
0 lim f .an / D f ./ D lim f .bn / 0:

n!1 n!1
Hence f ./ D 0. Moreover 2 .a; b/ since f .a/f .b/ 0. t

u
Corollary 5.3.4 (Intermediate-Value Theorem). Let f W a; b ! R be a contin-
uous function such that f .a/ f .b/. For each number k between f .a/ and f .b/
there exists 2 .a; b/ such that f ./ D k.
Proof. Apply Bolzanos theorem to the function
g.x/ D f .x/ k
for all x 2 a; b. t

u
Example 5.3.4. We show that the equation
e x D 4x
has a solution between 0 and 1. Let
f .x/ D e x 4x
for all x. Then f .0/ D 1 > 0 and f .1/ D e 4 < 0. By Bolzanos theorem, the
equation f .x/ D 0 has a solution between 0 and 1, and the result follows. 4
228 5 Continuity
Example 5.3.5. If a > 0 and n 2 N, then we can use the intermediate-value theorem
to prove the existence of a positive nth root of a. Choose c > maxf1; ag, and let
f .x/ D x n for all x 2 R. This function is continuous on 0; c and
f .0/ D 0 < c n D f .c/:
As 0 < a < c < c n , the intermediate-value theorem confirms the existence of > 0
such that n D f ./ D a. Thus is an nth root of a. 4
A fixed point for a function f is a number x 2 Df such that f .x/ D x.
Corollary 5.3.5 (Fixed-Point Theorem). Let f W a; b ! a; b be a continuous
function. Then f has a fixed point.
Proof. If either f .a/ D a or f .b/ D b, then f has a fixed point. The remaining
possibility is that f .a/ a > 0 and f .b/ b < 0. Let
g.x/ D f .x/ x
for all x 2 a; b. Then g is continuous on a; b, g.a/ > 0 and g.b/ < 0. Therefore,
by Bolzanos theorem, there exists 2 .a; b/ such that f ./ D 0. The result
follows. t
u
By combining the maximum-value theorem, the minimum-value theorem, and
the intermediate-value theorem, we obtain the following corollary.
Corollary 5.3.6. If f W a; b ! R is continuous and m and M are its minimum and
maximum values, respectively, then
f .a; b/ D m; M :
Theorem 5.3.7. A continuous real function is injective if and only if it is strictly

monotonic.
Proof. Clearly, every strictly monotonic function is injective.
Let f be a continuous injective function that is not strictly monotonic. We
shall complete the proof by finding a contradiction. Since f is injective but not
monotonic, there exist ; ; 2 Df such that < < and either
f ./ > maxff ./; f . /g
or
f ./ < minff ./; f . /g:
We may suppose that the former inequality obtains, the argument in the other case
being similar. Choose k such that
maxff ./; f . /g < k < f ./:
Then the intermediate-value theorem establishes the existence of c and d such that
< c < < d < and f .c/ D k D f .d /, contradicting the assumption that f
is injective. t
u
Theorem 5.3.8. Let f W a; b ! R be an increasing (respectively, decreasing) con-
tinuous function. Then f 1 exists and is also increasing (respectively, decreasing)
and continuous.
Proof. We confine our attention to the case where f is increasing, as the argu-
ment in the other case is analogous. Our previous results (Corollary 5.3.6 and
Theorem 5.3.7) show that f 1 exists with domain m; M , where m D f .a/ and
M D f .b/. Now choose y1 ; y2 2 m; M , where y1 < y2 . Then y1 D f .x1 / and
y2 D f .x2 / for some x1 ; x2 2 a; b. Thus x1 D f 1 .y1 / and x2 D f 1 .y2 /: If
x1 x2 , then y1 D f .x1 / f .x2 / D y2 , a contradiction. Hence f 1 is increasing.
Let c 2 m; M and choose " > 0. We want to find > 0 such that
jf 1 .y/ f 1 .c/j < " (5.3)
for all y 2 N .c/ \ m; M . If we set x D f 1 .y/ and D f 1 .c/, then

inequality (5.3) becomes
jx j < ";
which is equivalent to
" < x < C ": (5.4)
As f is increasing, inequality (5.4) is equivalent to
f . "/ < f .x/ < f . C "/
and therefore to
f . "/ f ./ < f .x/ f ./ < f . C "/ f ./:
Since
f .x/ f ./ D y c
whenever y 2 m; M , the proof is completed by choosing
D minff ./ f . "/; f . C "/ f ./g W

230 5 Continuity
For each y 2 m; M such that jy cj < we have
f . "/ f ./ < 0 y c < f . C "/ f ./
if y c 0, whereas if y c < 0, then
c y < f ./ f . "/;
so that
f . "/ f ./ < y c < 0 < f . C "/ f ./: t

u
Exercises 5.2.
1. Let f W a; b ! R be continuous. Show that
M D maxfjf .x/j j x 2 a; bg
exists. Show also that for every " > 0 there is an interval ; , included in a; b,
such that
jf .x/j > M "
for all x 2 ; .
2. Let f; gW a; b ! R be continuous functions. Suppose that f .a/ < g.a/ and
f .b/ > g.b/. Show that there exists c 2 .a; b/ such that f .c/ D g.c/.
3. Let f W .a; b/ ! R be continuous. Show that for any c1 ; c2 ; : : : ; cn 2 .a; b/ there
exists c 2 .a; b/ such that
1X
n
f .c/ D f .cj /:
n j D1
4. Let p be a nonzero polynomial. Show that e x D jp.x/j has a real solution.

5. Let f W a; 1/ ! R be continuous and suppose that limx!1 f .x/ exists and is
finite. Show that f is bounded on a; 1/.
6. Let f be continuous on R and suppose that
lim f .x/ D lim f .x/ D 0:

x!1 x!1
Show that f attains a maximum value or a minimum value.

7. Let f be a function and c a number. We say that f .c/ is a local maximum
(respectively, local minimum) of f if there exists > 0 such that f .c/ f .x/
[respectively, f .c/ f .x/] for all x 2 N .c/.
5.4 The Logarithm Function 231
Let f W a; b ! R be continuous. Suppose that f ./ and f ./ are local

maxima of f , where < . Show that there exists c 2 .; / such that f .c/ is
a local minimum.
8. Let f W a; b ! R be continuous. Show that if f does not have a local maximum
or a local minimum, then it must be monotonic on a; b.
5.4 The Logarithm Function
Recall that the exponential function is increasing (Theorem 2.7.12) and continuous
everywhere (Theorem 5.1.9). Therefore it has a continuous increasing inverse. This
inverse is called the logarithm function and is denoted by log. (The reader is
cautioned that some authors use a different notation.) The logarithm of a number
x is often written as log x rather than log.x/. Since limx!1 e x D 1 and
limx!1 e x D 0, the continuity of the exponential function, together with the
intermediate-value theorem, shows that log x is defined for all x > 0. Theorem 5.3.8
shows that it is increasing and continuous on .0; 1/.
As the logarithm and exponential functions are inverses, log x D y if and only if
e y D x. Therefore
e log x D x
and
log e y D y:
In particular, log e D 1, so that e x D e x log e . We use this observation to motivate the

following definition for every a > 0 and every real x:
ax D e x log a : (5.5)
In particular,
a0 D e 0 D 1
and
a1 D e log a D a:
We also define 0x D 0 for all x > 0.

If a; x; y are real numbers with a > 0, then
ax ay D e x log a e y log a D e x log aCy log a D e .xCy/ log a D axCy :

232 5 Continuity
Thus, for every number w we have
1 D a0 D aww D aw aw :
Therefore
1
aw D :
aw
Applying this result with a and w replaced by e and y log a, respectively, we find
that
ax e x log a
y
D y log a D e x log a e y log a D e .xy/ log a D axy :
a e
Moreover Eq. (5.5) shows that
log ax D x log a
for all a > 0 and all real x. If y is also a real number, it follows that
log.ax /y D y log ax D xy log a D log axy ;
and we deduce that
.ax /y D axy :
In addition,
log 1 D log a0 D 0 log a D 0:
Hence
1x D e x log 1 D e 0 D 1
for all real x.

Note that
axC1 D ax a1 D ax a:
Since we also have a0 D 1, it follows by induction that our definition of ax agrees

with our previous understanding in the case where x is a nonnegative integer. As
ax D 1=ax , the same can be said if x is a negative integer. Note also that if m and
n are integers and n > 0, then .a1=n /m D am=n . In particular,
.a1=n /n D a1 D a:
Thus our definition also agrees with our previous understanding in the case where x
is rational.
Now suppose that a > 0 and b > 0. Since
e log aClog b D e log a e log b D ab;
it follows that
log ab D log a C log b:
One consequence is that
ax b x D e x log a e x log b D e x.log aClog b/ D e x log ab D .ab/x
for every real x. Similarly,

a
log D log a log b;
b
since
e log a a
e log alog b D e log a e log b D log b
D :
e b
Hence
ax e x log a a x
x log ab
D D e x.log alog b/
D e D :
bx e x log b b
We also observe that the result of Example 4.6.2 holds for all real x 1; 0.
Indeed, the argument used in that example extends to the case of such an x. The
result of Example 3.6.1 may similarly be extended to all real p.
Since log 1 D 0 and log is increasing and continuous on .0; 1/, we see that
log x is negative for all x 2 .0; 1/ and positive for all x > 1. Moreover if we choose
M > 0 and x > e M , then
log x > log e M D M:
Hence
lim log x D 1:
x!1
Also, log x < M for all x such that 0 < x < e M , and so
lim log x D 1:

x!0C
234 5 Continuity
Example 5.4.1. By continuity we see that
log n
lim D lim log n1=n D log lim n1=n D log 1 D 0:
n!1 n n!1 n!1
In other words, log n << n. 4

Example 5.4.2. We show that
1
X 1
j D2
j log j
is divergent. This series is known as Abels series.

We observe that f1=.n log n/g is a decreasing sequence of positive terms. Hence
Cauchys condensation test is applicable. The condensed series is
1
X 1
X 1
1
2k k k
D
2 log 2 k log 2
kD1 kD1
1
1 X1
D ;
log 2 k
kD1
which is divergent. Hence Abels series is also divergent. 4

Logarithms are used in the proof of a test, due to Gauss, for the convergence of a
series.
P
Theorem 5.4.1 (Gausss Test). Let 1 j D1 aj be a series of positive terms. Suppose
there exist a bounded sequence fsn g and a constant c such that
anC1 c sn
D1 C 2 (5.6)
an n n
for all n > 0. Then the series is convergent if c > 1 and divergent if c 1.
Proof. Case 1: Suppose c 1. From Eq. (5.6) we see that

anC1 sn
n 1 Dc ;
an n
and the expression on the right-hand side approaches c as n approaches infinity

since fsn g is bounded. The required result therefore follows from Raabes test.
Case 2: Suppose c D 1. We apply the Kummer-Jensen test to the sequence
f.n 1/ log.n 1/g:

We already know that Abels series diverges, and so it remains only to investigate
the limiting behavior of

1 sn
cn D .n 1/ log.n 1/ 1 C 2 n log n
n n
sn
D .n 1/ log.n 1/ n 1 C log n
n
sn log n
D .n 1/.log.n 1/ log n/ :
n
Since fsn g is bounded, we have
sn log n
lim D 0:
n!1 n
(See Theorem 2.5.3 and recall that log n << n.) Moreover, using the analog of
Theorem 5.1.4 with c replaced by 1, we obtain
lim cn D lim .n 1/.log.n 1/ log n/

n!1 n!1
n1
n1
D lim log
n!1 n

n 1 n
D log lim 1
n!1 n 1 n
1

D log 1 e
D 1:
As the result of this calculation is negative, there exists N such that cn < 0 for
all n N . Consequently, the theorem follows from the Kummer-Jensen test.
t
u
1
X 2
.2j 1/
j D1
22j 1 .j 1/j
for convergence.
Solution. As in Example 3.9.1 we write
2
.2n 1/
an D
22n1 .n 1/n
236 5 Continuity
for all n > 0. Then

2
anC1 2n C 1
D
an 2n C 2

1 1 5n2 C 4n
D 1 C 2 ;
n n 4n2 C 8n C 4
the last step being obtained upon division of 4n2 C 4n C 1 by 4n2 C 8n C 4 to give
a quotient of 1 1=n and a remainder of 5 C 4=n. Since the sequence

5n2 C 4n
4n2 C 8n C 4
converges (to 5=4), it is bounded. The divergence of the given series now follows
from Gausss test with c D 1.
4
Exercises 5.3.
Let an D loglog n n for all n > 1. Show that an D nlog log n and hence that
1. P
1
j D2 1=aj converges.
2. Determine the convergence of the following series:
P1
j D1 .log.j C 1/ log j / where p > 0;
p
(a)
P1 log j
(b) j D2 j ;
P1 1
(c) j D1 j log 1 C 1
j
;
P1
j D2 j logp j where p 2 R;
1
(d)
P1 1
(e) j D2 j log j log log j .
(Hint: Try Cauchys condensation test.)

3. Find the interval of convergence of the power series
1
X log j
.x 1/j :
j D2
.j C 1/2
4. Let fan g be a sequence of positive terms and fsn g a bounded sequence. Let c be
a constant such that
an c sn
D1C C 2
anC1 n n
P1
for all n > 0. Prove that the series j D1 aj converges if c > 1 and diverges if
c 1.
5.5 Uniformly Continuous Functions 237
P1
5. (Bertrands test) Let j D1 aj be a series of positive terms, and for each n 2 N
let

anC1
n D n 1 1 log n:
an
(a) Show that if limn!1 n > 1, then the series converges.

(b) Show that if limn!1 n < 1, then the series diverges.
(Hint: Take
bn D .n 1/ log.n 1/
in the Kummer-Jensen test.)
5.5 Uniformly Continuous Functions
Let f be a continuous function and let c be an accumulation point of Df . Then for

each " > 0 there exists > 0 such that
jf .z/ f .c/j < "
for all z 2 N .c/ \ Df . It may be that depends on c. If not, then we say that f is
uniformly continuous.
Definition 5.5.1. A function f is uniformly continuous if for each " > 0 there
jf .z1 / f .z2 /j < "
for all z1 ; z2 2 Df satisfying jz1 z2 j < .
Remark. Thus a uniformly continuous function must be continuous at every point

in its domain.
We can prove the following sequential characterisation of uniform continuity.
Proposition 5.5.1. A function f is uniformly continuous if and only if for each pair
of sequences fsn g and ftn g in Df such that
lim .sn tn / D 0;
n!1
238 5 Continuity
we have
lim .f .sn / f .tn // D 0:

n!1
Proof. Suppose there exist sequences fsn g and ftn g in Df such that fsn tn g is null
but ff .sn / f .tn /g is not. Some " > 0 therefore has the property that for every M
there exists n M satisfying
jf .sn / f .tn /j "; (5.7)
but on the other hand, for every > 0 there exists N such that
jsn tn j <
for all n N . Some such n must satisfy inequality (5.7), and so f cannot be
uniformly continuous.
Conversely, suppose f is not uniformly continuous. Then for some " > 0 and all
> 0 there exist s; t 2 Df such that js t j < and jf .s/ f .t /j ". Thus for
each n 2 N there exist sn ; tn 2 Df such that
1
jsn tn j <
n
and
jf .sn / f .tn /j ":
The sandwich theorem shows that the sequence fsn tn g is null, but ff .sn / f .tn /g
is evidently not. t
u
Example 5.5.1. Consider the function f W .0; 2 ! R defined by
1
f .x/ D :
x
Let " D 1=2 and, for each n 2 N, define sn D 1=n and tn D 2=n. Then
1
jsn tn j D !0
n
as n ! 1, but
n
jf .sn / f .tn /j D "
2
for all n 2 N. We conclude that f is not uniformly continuous. 4

We also deduce the following theorem.

Theorem 5.5.2. Every function that is continuous over a compact set is uniformly
continuous.
Proof. Suppose that a function f is continuous over a compact set X but not
uniformly. Then we can find " > 0 with the property that for each n 2 N there
exist sn and tn in X satisfying
lim jsn tn j D 0
n!1
and
jf .sn / f .tn /j ": (5.8)
Since fsn g is bounded, by Corollary 2.6.8 it has a subsequence fskn g that converges
to some number s. If skm D s for some number m, then s 2 X . In the remaining
case s is a limit point of X by Theorem 2.6.11, and once again s 2 X since X is
closed. Now
jtkn sj jtkn skn j C jskn sj ! 0
as n ! 1. Hence
lim tkn D s:
n!1
As f is continuous at s 2 X , we obtain
lim f .skn / D lim f .tkn / D f .s/:

n!1 n!1
Therefore
jf .skn / f .tkn /j jf .skn / f .s/j C jf .tkn / f .s/j ! 0
as n ! 1. We now have a contradiction to inequality (5.8). t

u
Theorem 5.5.3. A continuous function f is uniformly continuous on an open
interval .a; b/ if and only if limx!aC f .x/ and limx!b f .x/ exist and are finite.
Proof. Suppose that f is uniformly continuous on .a; b/. Choose " > 0. There
jf .x/ f .y/j < " (5.9)

for all x and y in .a; b/ satisfying jx yj < . Choose x and y in N=2 .a/ \ .a; b/.
Then
240 5 Continuity

jx yj jx aj C ja yj < C D ;
2 2
so that inequality (5.9) holds. Thus f satisfies Cauchys condition at a, and we

conclude from Theorem 4.3.2 that limx!aC f .x/ exists and is finite. The limit as
x ! b is dealt with in a similar manner.
Conversely, suppose that both limits exist and are finite. Then f can be extended
to a continuous function on a; b that is uniformly continuous by Theorem 5.5.2.
Hence f is also uniformly continuous. t
u
Example 5.5.2. Let f .x/ D e 1=x for all x 0. Then limx!0C f .x/ D 0. Hence
f is uniformly continuous on .0; b for each b > 0. 4
A function is uniformly continuous over a closed interval if and only if it is
continuous over that interval. Thus if functions f and g are uniformly continuous
over an interval a; b, then so are f C g, fg, and f for all 2 R. In view of
Theorem 5.5.3, this result is also true if we replace the interval by .a; b/. However,
it is in general not true if we replace the interval by a; 1/.
Example 5.5.3. Let f .x/ D g.x/ D x for all x. Then, clearly, f and g are
uniformly continuous over 0; 1/. However, we show that the function givenp by
uniformly continuous over that interval. For all n 0 let sn D n C 1
x 2 is not p
and tn D n. Then
p p
sn tn D nC1 n
p p p p
. n C 1 n/. n C 1 C n/
D p p
nC1C n
1
D p p
nC1C n
! 0:
However,
sn2 tn2 D 1
for all n. Hence according to Proposition 5.5.1, the function x 2 is not uniformly
continuous over 0; 1/. 4
The family of uniformly continuous functions includes an important subfamily,
which we define below.
Definition 5.5.2. A function f is said to be Lipschitz continuous if there exists a
positive constant M such that
jf .z1 / f .z2 /j M jz1 z2 j
for all z1 and z2 in Df .

Example 5.5.4. The modulus function is Lipschitz continuous since
jjz1 j jz2 jj jz1 z2 j
for all z1 ; z2 2 C. 4
It follows from Proposition 5.5.1 and the sandwich theorem that every function
that is Lipschitz continuous is also uniformly continuous. However, the converse is
not always true.
Example 5.5.5. The function f given by
p
f .x/ D x
is continuous and therefore uniformly continuous over 0; 1. We show that it is not
Lipschitz continuous. Suppose there exists M > 0 such that
jf .x1 / f .x2 /j M jx1 x2 j
for all x1 ; x2 2 0; 1. In particular,
jf .x/ f .0/j M jxj
for every x 2 .0; 1. Thus

p
x Mx;
p
so that 1 M x. However, this result is contradicted for x such that 0 < x <
1=M 2 . 4
Exercises 5.4.
1. Show that the following functions are uniformly continuous:
p
(a) x, where x 2 0; 1/;
(b) x sin x1 , where x 2 .0; 1/.
2. Show that the following functions are not uniformly continuous:
(a) log x, where x 2 .0; 1/;
(b) e x , where x 2 0; 1/.
3. Suppose that f is uniformly continuous on .a; b and on b; c/. Show that f is
uniformly continuous on .a; c/.
4. Let f be continuous on a; 1/ and suppose that limx!1 f .x/ exists and is finite.
Show that f is uniformly continuous on a; 1/.
Chapter 6
Differentiability
The notion of a derivative is motivated by studying two kinds of problems: finding

instantaneous velocities and determining slopes of tangents to curves. Here we
shall not dwell on these problems. Rather, we undertake a study of the general
mathematical properties of derivatives. We assume all functions to be of a complex
variable and complex-valued unless an indication to the contrary is given.
6.1 Derivatives
Definition 6.1.1. Let f be a function and c an accumulation point of Df . Define
f .z/ f .c/
f 0 .c/ D lim :
z!c zc
If the limit exists and is a number, then f is said to be differentiable at c and f 0 .c/
is the derivative of f at c.
Remark 1. We sometimes write
d
f .z/ D f 0 .z/
dz
if f is differentiable at z.
Remark 2. In the case where f is a real-valued function of a real variable x, the
quotient

244 6 Differentiability
f .x/ f .c/
xc
gives the slope of the line joining the points .c; f .c// and .x; f .x// on the graph
of f . We may interpret f 0 .c/ geometrically as the slope of the graph of f at the
point .c; f .c//.
Example 6.1.1. Let f .z/ D az C b for all z 2 C, where a; b 2 C. Then, for each
c 2 C and z c,
f .z/ f .c/ .az C b/ .ac C b/

D
zc zc
a.z c/
D
zc
D a:
Hence f 0 .c/ D a. 4
0
In particular, if f .z/ D b for all z 2 C, then f .c/ D 0 for all c. If f .z/ D z for
all z 2 C, then f 0 .c/ D 1 for all c.
Example 6.1.2. Let f .z/ D z2 for all z 2 C. Then, for each c and z c,
f .z/ f .c/ z2 c 2
D
zc zc
D zCc
! 2c
as z ! c. Hence f 0 .c/ D 2c. 4

Definition 6.1.1 may be rewritten using the limit in the next theorem.
Theorem 6.1.1. Let f be a function that is differentiable at a number c. Then
f .c C h/ f .c/
f 0 .c/ D lim :
h!0 h
Proof. Let k.z/ D c C z for all z. Then limz!0 k.z/ D c. Define
f .z/ f .c/
g.z/ D
zc
for all z 2 Df fcg, and let g.c/ D f 0 .c/. Then 0 is an accumulation point of Dgk
and g is continuous at c. Notice also that
f .c C z/ f .c/
g.k.z// D :
z
6.1 Derivatives 245
It therefore follows from Theorem 5.1.4 that

f .c C z/ f .c/
f 0 .c/ D lim g.z/ D lim g.k.z// D lim ;
z!c k.z/!c z!0 z
as required. t
u
Example 6.1.3. Recall that exp.z/ D e for all complex z. We have seen that The-
z
orem 2.7.10 holds also for complex numbers. Therefore so does Corollary 2.7.11,
and we can use the latter, together with Theorem 6.1.1, to show that
e cCh e c
exp0 .c/ D lim D ec
h!0 h
for all c. In other words, the exponential function is its own derivative. 4
The following theorem is clear from the sequential formulation of limits.
Theorem 6.1.2. Let f be a function. If there exist sequences fsn g and ftn g in Df
such that
lim sn D lim tn D c
n!1 n!1
and
f .sn / f .c/ f .tn / f .c/

lim lim ;
n!1 sn c n!1 tn c
then f 0 .c/ does not exist.

Example 6.1.4. Consider the function f .z/ D z, defined on C. We show that f 0 .0/
does not exist. For each n 2 N, take sn D 1=n and tn D i=n. Clearly,
n!1 n!1
Now
f .sn / f .0/
D1
sn 0
and
f .tn / f .0/
D 1:
tn 0
Hence
f .sn / f .0/ f .tn / f .0/

lim lim :
n!1 sn 0 n!1 tn 0
Therefore f 0 .0/ does not exist. 4

Example 6.1.5. Consider the real function

(
x2 if x 0;
f .x/ D
x if x > 0:
We show that f 0 .0/ does not exist. For each x 0 let
f .x/ f .0/
Q.x/ D
x0
f .x/
D
x
(
x if x 0;
D
1 if x > 0:
Thus
lim Q.x/ D 0
x!0
and
lim Q.x/ D 1:
x!0C
Hence limx!0 Q.x/ does not exist. In other words, f 0 .0/ does not exist. Note that
the function is continuous but its graph has a corner at 0 (see Fig. 6.1). 4
We now show that if f is differentiable at c, then f is continuous at c.
Examples 6.1.4 and 6.1.5 show that the converse is not always true.
Theorem 6.1.3. If a function f is differentiable at a number c, then f is continuous
at c.
Fig. 6.1 Graph of f f(x)

in Example 6.1.5
x
0
6.1 Derivatives 247
Proof. Let f be differentiable at c. Then, for each z 2 Df fcg,
f .z/ f .c/
f .z/ f .c/ D .z c/:
zc
As z approaches c, the right-hand side approaches f 0 .c/ 0 D 0. Hence
lim f .z/ D f .c/:

z!c
t
u
Our next example shows that, even for a real function f , the existence of
lim f 0 .x/
x!c
does not guarantee the existence of f 0 .c/.

Example 6.1.6. Let
(
x C 1 if x 0;
f .x/ D
x if x > 0.
Then f 0 .x/ D 1 for all x 0. Thus
lim f 0 .x/ D 1:
x!0
But f is not continuous at 0, since limx!0 f .x/ D 1 and limx!0C f .x/ D 0.

Therefore, by Theorem 6.1.3, f 0 .0/ does not exist. 4
Our next theorem gives an idea of the behavior of a function in the vicinity of a
point where its derivative exists. The result is stronger than Theorem 6.1.3.
Theorem 6.1.4. Let f be differentiable at c. Then there exist > 0 and M > 0
such that
jf .z/ f .c/j < M jz cj
for all z 2 N .c/ \ Df .
Proof. There exists > 0 such that

f .z/ f .c/
f .c/ < 1
0
zc
for all z 2 N .c/ \ Df . Hence

f .z/ f .c/
jf 0 .c/j < 1:
zc
Setting M D jf 0 .c/j C 1 > 0, we obtain
jf .z/ f .c/j < M jz cj
for all z 2 N .c/ \ Df . t

u
The condition given in the conclusion of Theorem 6.1.4 is called the Lipschitz
condition at c. Functions that satisfy it are continuous at c but not necessarily
differentiable there.
Example 6.1.7. The absolute value function jxj satisfies the Lipschitz condition at 0
but can be shown to be not differentiable there. Hence the converse of Theorem 6.1.4
is in general not true. 4
We end this section by defining f .0/ D f , where f is a function, and if f .n/ has
been defined for some nonnegative integer n, then f .nC1/ D .f .n/ /0 . We call f .n/
the nth derivative of f . If f is a function of a variable z, then its nth derivative is
dn
sometimes written as dz n f .z/. These higher-order derivatives will be used later to
study approximations of functions by polynomials.

Exercises 6.1.
1. Find the derivatives of the following functions:
(a) x1 , where x 0;
(b) x n , where n 2 N;
(c) xjxj.
2. Show that the absolute p
value function is not differentiable at 0.
3. Show that the function x, where x 0, is differentiable at all x > 0.
4. Use the properties of the sine and cosine functions to prove that sin0 x D cos x
and cos0 x D sin x for all x.
5. Let
(
x 2 if x 2 Q;
f .x/ D
0 if x Q.
Show that f is differentiable only at 0.

6.
(
x2 if x < 0;
f .x/ D
x 3
if x 0.
Find f 0 .x/ and f 00 .x/ for all x.

7. Let f .x/ D jxj3 for all x. Find f 0 .x/ and f 00 .x/ for all x and show that f 000 .0/
does not exist.
6.2 Differentiation Formulas 249
6.2 Differentiation Formulas
We now present some results that are helpful in finding derivatives of functions.
Theorem 6.2.1. If f and g are functions and c is an accumulation point of Df \
Dg , where f and g are differentiable, then
1.
.f C g/0 .c/ D f 0 .c/ C g 0 .c/;
2.
.fg/0 .c/ D f .c/g 0 .c/ C f 0 .c/g.c/;
3.
0
f f 0 .c/g.c/ f .c/g 0 .c/
.c/ D
g g 2 .c/
if there is a neighborhood N .c/ such that g.z/ 0 for all z 2 N .c/.

Proof. 1. We compute
.f C g/.z/ .f C g/.c/
.f C g/0 .c/ D lim
z!c zc

f .z/ f .c/ g.z/ g.c/
D lim C
z!c zc zc
D f 0 .c/ C g 0 .c/:
2. Since
.fg/.z/ .fg/.c/ D f .z/g.z/ f .c/g.c/

D f .z/g.z/ f .z/g.c/ C f .z/g.c/ f .c/g.c/
D f .z/.g.z/ g.c// C .f .z/ f .c//g.c/;
we have
g.z/ g.c/ f .z/ f .c/

.fg/0 .c/ D lim f .z/ lim C lim g.c/
z!c z!c zc z!c zc
D f .c/g 0 .c/ C f 0 .c/g.c/
because f is continuous at c.
3. We deal first with the function 1=g, recalling that g is continuous at c:

0 1
1
1 g.z/ g.c/
.c/ D lim
g z!c zc

1 g.c/ g.z/
D lim
z!c zc g.z/g.c/
1 g.c/ g.z/
D lim lim
z!c g.z/g.c/ z!c zc
g 0 .c/
D :
g 2 .c/
An application of part (2) therefore gives

0 0
f 1
.c/ D f .c/
g g
0
01 1
D f .c/ C f .c/ .c/
g.c/ g
f 0 .c/ f .c/g 0 .c/
D
g.c/ g 2 .c/
f 0 .c/g.c/ f .c/g 0 .c/
D :
g 2 .c/
t
u
Remark. Again, the hypothesis that c be an accumulation point of Df \ Dg is
essential (see Remark 2 after Theorem 4.4.1).
Corollary 6.2.2. Let f be a function and a and c numbers. If f is differentiable
at c, then
.af /0 .c/ D af 0 .c/:
Proof. Let g.z/ D a for all z. Thus g 0 .c/ D 0 for all c. Hence
.af /0 .c/ D af 0 .c/ C 0 f .c/ D af 0 .c/;
by Theorem 6.2.1(2). t
u
Example 6.2.1. Let f .z/ D zn for all z, where n is a positive integer. We show by
induction that
f 0 .z/ D nzn1
for all z, where 00 D 1. Certainly, f 0 .z/ D 1 for all z if n D 1. Assume that n > 1
and that the result holds with n replaced by n 1. Then zn D z zn1 , and so
f 0 .z/ D z.n 1/zn2 C 1 zn1

D nzn1
for all z, by Theorem 6.2.1(2), as required.

Suppose now that z 0. Then the result just proved holds also when
n D 0. In fact, we can extend it to the case where n is a negative integer, using
Theorem 6.2.1(3) and the fact that zn D 1=zn . Thus if f .z/ D zn for all z 0 and
n is a negative integer, then n > 0, and so
.nzn1 /
f 0 .z/ D D nzn1
z2n
for all z 0. 4
Our next theorem, known as the chain rule, deals with compositions of functions.
Theorem 6.2.3 (Chain Rule). Suppose that f and g are functions such that both
g 0 .c/ and f 0 .g.c// exist, where c 2 Df g . Suppose also that c is an accumulation
point of Df g . Then
.f g/0 .c/ D f 0 .g.c//g 0 .c/:
Proof. Define
g.z/ g.c/
u.z/ D g 0 .c/
zc
for all z 2 Dg fcg. Thus
lim u.z/ D g 0 .c/ g 0 .c/ D 0:

z!c
Define u.c/ D 0. Then u is continuous at c.

Similarly, let b D g.c/ and define
(
f .z/f .b/
f 0 .b/ if z 2 Df fbg;
v.z/ D zb
0 if z D b.
Then v is continuous at b.
From the definitions of u.z/ and v.z/, we have
g.z/ b D .z c/.g 0 .c/ C u.z//

for all z 2 Dg (even for z D c) and
f .z/ f .b/ D .z b/.f 0 .b/ C v.z//
for all z 2 Df . Hence
f .g.z// f .g.c// D f .g.z// f .b/

D .g.z/ b/.f 0 .b/ C v.g.z///
D .z c/.g 0 .c/ C u.z//.f 0 .b/ C v.g.z///
for all z 2 Df g , so that
.f g/0 .c/ D lim .g 0 .c/ C u.z//.f 0 .b/ C v.g.z///

z!c
D g 0 .c/f 0 .b/;
for v g is continuous at c by Corollary 5.1.5 since g is continuous at c and v is

continuous at b D g.c/. u
t
e iz e iz
sin z D
2i
for all z 2 C, we have
ieiz C ieiz e iz C e iz

sin0 z D D D cos z;
2i 2
by Theorems 6.2.1 and 6.2.3. Similarly,
ieiz ieiz i 2 .e iz e iz /

cos0 z D D D sin z:
2 2i
Thus the sine and cosine functions are continuous everywhere.
Recall also that sec x D 1= cos x and tan x D .sin x/=.cos x/ whenever
cos x 0. We therefore have
cos2 x C sin2 x 1
tan0 x D 2
D D sec2 x:
cos x cos2 x
It follows that the tangent function is continuous wherever it is defined. Note also
that
1
sec0 x D . sin x/ D sec x tan x
cos2 x
for all x such that cos x 0. 4
Theorem 6.2.4 (Inverse Function Theorem). Let f W a; b ! R be an increasing

differentiable function. Let c be a number in a; b such that f 0 .c/ 0. Then f 1
is differentiable at f .c/ and
1 0 1
f .f .c// D 0 .c/
:
f
Proof. Note first that f is continuous, by Theorem 6.1.3. Therefore f 1 exists and
is continuous and increasing on its domain, by Theorem 5.3.8. Let g D f 1 . Thus
Dg D f .a/; f .b/.
According to Theorem 6.1.1, we must show that
g.f .c/ C k/ c 1
lim D 0 ;
k!0 k f .c/
since g.f .c// D c. It is therefore enough to prove that
k
f 0 .c/ D lim :
k!0 g.f .c/ C k/ c
For each k such that f .c/ C k 2 Dg , define
h.k/ D g.f .c/ C k/ c:
Thus k 2 Dh if and only if f .a/ f .c/ C k f .b/. Consequently,
Dh D f .a/ f .c/; f .b/ f .c/:
Therefore 0 2 Dh since f is increasing, and
h.0/ D g.f .c// c D c c D 0:
Furthermore,
f 1 .f .c/ C k/ D c C h.k/I
hence
f .c/ C k D f .c C h.k//;
so that
k D f .c C h.k// f .c/:
Thus it suffices to show that
f .c C h.k// f .c/
f 0 .c/ D lim :
k!0 h.k/
But
f .c C h.k// f .c/
f 0 .c/ D lim :
h.k/!0 h.k/
We therefore introduce the function j defined by j.0/ D f 0 .c/ and
f .c C x/ f .c/
j.x/ D
x
for all x 2 a c; b c f0g, so that
f 0 .c/ D lim j.h.k//:

h.k/!0
We now check that the hypotheses of Theorem 5.1.4 are satisfied by the functions
j and h. As h is evidently continuous on its domain, we see that
lim h.k/ D h.0/ D 0:

k!0
Moreover j is continuous at 0 by definition. Finally, we have k 2 Dj h if and only

if k 2 Dh and h.k/ 2 a c; b c. But if k 2 Dh , then h.k/ 2 a c; b c since
Rg D a; b. We conclude that Dj h D Dh and therefore that 0 is an accumulation
point of Dj h . We can thus apply Theorem 5.1.4 to show that
f 0 .c/ D lim j.h.k//;

k!0
as required. t
u
The inverse function theorem also holds for functions of a complex variable with
continuous derivatives (see [11]).
Since the logarithm and exponential functions are inverses, we have
x D exp.log x/ (6.1)
for all x > 0. Moreover the logarithm function is differentiable at all x > 0
by Theorem 6.2.4, since the exponential function is nonzero, differentiable, and
increasing everywhere. Differentiation of Eq. (6.1) therefore yields
1 D exp.log x/ log0 x D x log0 x;

and so
1
log0 x D
x
for all x > 0. This calculation provides another confirmation that the logarithm
function is continuous at all x > 0.
Let us now define f .x/ D x a D e a log x > 0 for all x > 0, where a is any real
number. Then
a a
f 0 .x/ D e a log x D x a D axa1 :
x x
Similarly, let g.x/ D ax D e x log a for all x, where a > 0. Then
g 0 .x/ D e x log a log a D ax log a:
Exercises 6.2.
1. Show that log0 x D 1=x for all x > 0 by evaluating
log.x C h/ log x
lim :
h!0 h
2. Let f W a; b ! R be differentiable at c 2 .a; b/. Let fan g and fbn g be sequences

such that
a < an < c < bn < b
for all positive integers n. If
lim an D lim bn D c;
n!1 n!1
prove that
f .bn / f .an /
f 0 .c/ D lim :
n!1 bn an
[Hint: Use the fact that
f .bn / f .an /
f 0 .c/
bn an

bn c f .bn / f .c/ 0 an c f .an / f .c/ 0
D f .c/ f .c/
bn an bn c bn an an c
for all n 2 N.]

3. Let f and g be functions having nth derivatives. Prove that

!
X n
n
.fg/ .n/
D f .nj / g .j / :
j D0
j
(This result is known as Leibnizs rule.)

4. Let F D f1 f2 fn and Fj D F=fk whenever fj .x/ 0. Show by induction
that we then have
X
n
F 0 .x/ D Fj .x/fj0 .x/:
j D1
5. Let
(
x if x 2 Q;
f .x/ D
x if x Q.
Show that .f f /.x/ D x for all x. What can you say about the chain rule?
6. Let f be as in Exercise 6.5 and let g D f . Show that .fg/.x/ D x 2 for all x.
What can you say about the product rule?
7. Let m and n be positive integers. An m n matrix is defined as an array of
numbers arranged in m rows and n columns. The array is usually
enclosed in
p q p q
parentheses. The determinant of the 2 2 matrix is defined by the
r s r s
equation

p q

r s D ps rq:
Let .a; b/ be an open interval, and for all x 2 .a; b/ define

f1 .x/ f2 .x/

F .x/ D ;
g1 .x/ g2 .x/
where f1 ; f2 ; g1 ; g2 are functions that are differentiable on .a; b/. Show that
0
f .x/ f 0 .x/ f1 .x/ f2 .x/
F 0 .x/ D 1 2 C
g1 .x/ g2 .x/ g10 .x/ g20 .x/
for all x 2 .a; b/.

6.3 The Mean-Value Theorem for Derivatives 257
6.3 The Mean-Value Theorem for Derivatives
The first theorem to be discussed in this section is important in its own right but is
even more important as the basis for a number of other theorems that are among the
most useful in the theory of functions.
Theorem 6.3.1 (Rolle). Let f W a; b ! R be continuous on a; b and differ-
entiable on .a; b/, and suppose that f .a/ D f .b/. Then f 0 ./ D 0 for some
2 .a; b/.
Proof. If f is a constant function, then f 0 .x/ D 0 for all x 2 a; b. Hence we
assume that f .x1 / f .a/ for some x1 2 .a; b/. We may also assume that
f .x1 / > f .a/;
as the argument for the other case is similar.

By the maximum-value theorem, there exists 2 a; b such that f .x/ f ./
for all x 2 a; b. Since
f ./ f .x1 / > f .a/ D f .b/;
we have fa; bg. Thus 2 .a; b/.

We claim that f 0 ./ D 0. Define
f .x/ f ./
Q.x/ D
x
for all x 2 a; b fg. Then
lim Q.x/ D f 0 ./:

x!
As f .x/ f ./ for all x 2 a; b, it follows that f .x/ f ./ 0 for all such x.
Hence Q.x/ 0 for all x 2 a; / and Q.x/ 0 for all x 2 .; b. Thus
lim Q.x/ 0
x!
and
lim Q.x/ 0I
x! C
consequently, limx! Q.x/ D 0, as required. t

u
Remark 1. The number in Rolles theorem need not be unique. For example, let
f .x/ D x 3 x for all x 2 1; 1. Then f satisfies the hypotheses of the theorem.
However, f 0 .x/ D 3x 2 1 for all x 2 .1; 1/, so that

0 1 0 1
f p D f p D 0:
3 3
Remark 2. The conditions in Rolles theorem cannot be relaxed. For instance, for
all x 2 0; 1 let
f .x/ D x;
g.x/ D x bxc;
and
h.x/ D j2x 1j:
The function f is continuous and differentiable everywhere, yet f 0 .x/ D 1 0 for

all x. Note that f .0/ f .1/. The function g satisfies g.x/ D x for all x 2 0; 1/
and g.1/ D 1 1 D 0 D g.0/. It is differentiable on .0; 1/, but g 0 .x/ D 1 0 for
all x 2 .0; 1/. Note that g is not continuous at 1. Finally, h is continuous everywhere
and h.0/ D h.1/ D 1. Its derivative is 2 0 at all x 2 .0; 1=2/ and 2 0 at all
x 2 .1=2; 1/. It is not differentiable at 1=2.
Remark 3. The converse of Rolles theorem is not, in general, true. In fact, the
conclusion of the theorem does not imply any of its hypotheses. Take, for example,
the function f such that
f .x/ D x 2 bx 2 c
for each x 2 1; 3=2. Here we have f .3=2/ D 9=4 2 D 1=4 0 D f .1/.
Moreover f is not continuous (and therefore not differentiable) at 1. Nevertheless,
f .x/ D x 2 for all x 2 .1; 1/, and so f 0 .0/ D 0. Note that 0 2 .1; 3=2/.
By maneuvering the x-axis, we can generalize Rolles theorem to one of the most
fundamental theorems of real analysisthe mean-value theorem.
Theorem 6.3.2 (Mean-Value Theorem). Let f be a function of a real variable
and suppose that f is continuous on a closed interval a; b and differentiable on
.a; b/. Then there is a number 2 .a; b/ such that
f .b/ f .a/
f 0 ./ D : (6.2)
ba
Discussion: Notice that the right-hand side of Eq. (6.2) is the slope of the chord
of the graph of f joining the points .a; f .a// and .b; f .b//. The theorem asserts
that some tangent to the graph between these points is parallel to the chord. We may
rotate and translate the chord, if necessary, until it is superimposed on the x-axis.
The mean-value theorem then becomes identical to Rolles theorem.
Proof. The equation of the chord joining .a; f .a// and .b; f .b// is
f .b/ f .a/
y f .a/ D .x a/:
ba
Thus
f .b/ f .a/
y D f .a/ C .x a/:
ba
Define
f .b/ f .a/
g.x/ D f .x/ y D f .x/ f .a/ .x a/
ba
for each x 2 a; b. Then g is continuous on a; b and differentiable on .a; b/.

Moreover g.a/ D 0 and
g.b/ D f .b/ f .a/ .f .b/ f .a// D 0:
Thus Rolles theorem may be applied to find a number 2 .a; b/ such that
g 0 ./ D 0. But
f .b/ f .a/
g 0 ./ D f 0 ./ :
ba
The result follows. t
u
Example 6.3.1. If f .x/ D e x for all x, then f 0 .x/ D e x for all x. We can use this
result and the mean-value theorem to show that
ex 1 C x
for all real x. Indeed, equality holds if x D 0. Suppose that x > 0. By the mean-
value theorem, there exists 2 .0; x/ such that
e x e 0 D e .x 0/:
Since > 0, we conclude that
e x 1 D e x > x;

The case where x < 0 is handled in a similar way. 4
We now establish some theorems that are plausible intuitively but difficult to
prove without invoking the mean-value theorem.
Theorem 6.3.3. Let f W a; b ! R be a function that is continuous on a; b and
differentiable on .a; b/. If f 0 .x/ D 0 for all x 2 .a; b/, then f is a constant
function.
Proof. Suppose f is not a constant function. Then there exist c and d in a; b such
that f .c/ f .d /. Suppose without loss of generality that c < d . By the mean-
value theorem, there exists 2 .c; d / such that
f .d / f .c/
D f 0 ./ D 0:
d c
Thus we reach the contradiction that f .d / D f .c/. t

u
Corollary 6.3.4. Let f; gW a; b ! R be functions that are continuous on a; b
and differentiable on .a; b/. Suppose that f 0 .x/ D g 0 .x/ for all x 2 .a; b/. Then
there is a constant c such that
f .x/ D g.x/ C c
for all x 2 a; b.

Proof. Apply Theorem 6.3.3 to the function f g. t
u
The next two theorems relate the sign of the derivative of a function to the
monotonicity of the function.
differentiable on .a; b/. If f 0 .x/ > 0 for all x 2 .a; b/, then f is increasing on
a; b.
Proof. Choose c and d in a; b such that c < d . The function f satisfies the
hypotheses of the mean-value theorem on c; d . Therefore there exists 2 .c; d /
such that
f .d / f .c/
D f 0 ./ > 0:
d c
Consequently, f .d / > f .c/, since d c > 0. We conclude that f is increasing on

a; b. t
u
differentiable on .a; b/. If f 0 .x/ < 0 for all x 2 .a; b/, then f is decreasing on
a; b.
Proof. Apply Theorem 6.3.5 to the function f . t
u
Corollary 6.3.7. If f 0 is continuous at some c 2 .a; b/ and f 0 .c/ 0, then f is

strictly monotonic in some neighborhood of c.
Proof. Suppose first that f 0 .c/ > 0. Since f 0 is continuous at c, Theorem 5.1.8
confirms the existence of > 0 such that f 0 .x/ > 0 for all x 2 N .c/. By
Theorem 6.3.5, f is increasing on N .c/.
The argument is similar if f 0 .c/ < 0. t
u
Example 6.3.2. Let f .x/ D x a for all x > 0, where a 2 R. Then f 0 .x/ D axa1
for all x > 0. Thus f is increasing if a > 0 and decreasing if a < 0. 4
Example 6.3.3. Let f .x/ D ax for all x 2 R, where a > 0. Then f 0 .x/ D ax log a
for all x, so that f is increasing if a > 1 and decreasing if 0 < a < 1. 4
Example 6.3.4. Let
f .x/ D ax2 C bx C c
for all real x, where a; b; c are real and a > 0. Then
f 0 .x/ D 2ax C b
for all x, and so f 0 .x/ D 0 if and only if x D b=.2a/. Moreover f 0 .x/ < 0 for
all x < b=.2a/ and f 0 .x/ > 0 for all x > b=.2a/, since a > 0. It follows from
Theorems 6.3.6 and 6.3.5 that f is decreasing on .1; b=.2a/ and increasing
on b=.2a/; 1/. Therefore, by Theorem 5.3.7, the equation f .x/ D 0 can have at
most two real solutions, one in the former interval and one in the latter. In fact, since

bx c
f .x/ D a x 2 C C
a a
2 !
b b2 c
Da xC 2C ;
2a 4a a
we have
lim f .x/ D lim f .x/ D 1;

x!1 x!1
and so there are two real solutions if f .b=.2a// < 0, none if f .b=.2a// > 0,
and just the solution x D b=.2a/ if f .b=.2a// D 0. Note that

b b2 b2
f Da Cc
2a 4a2 2a
4ac b 2
D :
4a
As we saw in Sect. 1.4, the number b 2 4ac is called the discriminant of f .x/. Since
a > 0, the equation has two distinct real solutions if the discriminant is positive,
just one if D 0, and none if < 0. We also observe that the equation has two
distinct real solutions if and only if there exists a number such that f ./ < 0.
If a < 0, similar arguments show that once again the equation has two distinct
real solutions if > 0, just one if D 0, and none if < 0. In this case,
however, two distinct real solutions exist if and only if there is a number such that
f ./ > 0. 4
Definition 6.3.1. Let f be a real-valued function. The value of f at a number
c 2 Df is called a local maximum if there exists a neighborhood N .c/ such
that f .z/ f .c/ for all z 2 N .c/. A local minimum of f is defined analogously.
A number c is an extremal point or extremum of f if f .c/ is a local maximum or
local minimum, and f .c/ is then an extremal value of f .
Our next result gives a sufficient condition for the existence of an extremum.
Theorem 6.3.8 (First Derivative Test). Let f W a; b ! R be a continuous
function and let c 2 .a; b/. Suppose there exists > 0 such that f is differentiable
at all x 2 N .c/.
1. If f 0 .x/ 0 whenever c < x < c and f 0 .x/ 0 whenever c < x < c C ,
then f has a local maximum at c.
2. If f 0 .x/ 0 whenever c < x < c and f 0 .x/ 0 whenever c < x < c C ,
then f has a local minimum at c.
Proof.
1. The mean-value theorem shows that for each x 2 .c ; c/ there exists 2 .x; c/
such that
f .c/ f .x/ D f 0 ./.c x/ 0:
Hence f .c/ f .x/. Likewise, f .c/ f .x/ for each x 2 .c; c C /. Therefore
f .c/ is a local maximum.
2. The proof of part (2) is similar. t
u
Example 6.3.5. This theorem shows that the function jxj, for all x 2 R, has a local
minimum at 0. Note that this function is not differentiable at 0. 4
The following theorem is often of assistance in locating local maxima and
minima.
Theorem 6.3.9. Let f W a; b ! R and c 2 .a; b/. If f .c/ is an extremal value of
f and f 0 .c/ exists, then f 0 .c/ D 0.
Theorem 6.3.9 is an immediate consequence of the following lemma, which
asserts that an extremal point of a function f is a turning point of the graph of f .
Lemma 6.3.10. Let f W a; b ! R and c 2 .a; b/.

1. If f 0 .c/ > 0, then there exists > 0 such that f .x/ < f .c/ for all x 2 Df for
which c < x < c and f .x/ > f .c/ for all x 2 Df for which c < x < c C .
2. If f 0 .c/ < 0, then there exists > 0 such that f .x/ > f .c/ for all x 2 Df for
which c < x < c and f .x/ < f .c/ for all x 2 Df for which c < x < c C .
Proof. If f 0 .c/ > 0, then by Theorem 4.4.6 there exists > 0 such that
f .x/ f .c/
>0
xc
for all x 2 N .c/. Thus f .x/ > f .c/ if x > c, but f .x/ < f .c/ if x < c.
The case where f 0 .c/ < 0 follows by considering the function f . t
u
A number c such that f 0 .c/ D 0 is sometimes called a critical point of the
function f .
Our next theorem shows that derivatives satisfy the conclusion of the
intermediate-value theorem even though they may not be continuous. First, we
establish a special case.
Lemma 6.3.11. If f W a; b ! R is differentiable and f 0 .a/f 0 .b/ < 0, then there
exists 2 .a; b/ such that f 0 ./ D 0.
Proof. We may assume that f 0 .a/ > 0 [and therefore that f 0 .b/ < 0] as the
argument is similar if f 0 .a/ < 0. The differentiable function f is continuous and
therefore has a maximum value at some 2 a; b. By Lemma 6.3.10, there exists
> 0 such that f .x/ > f .a/ for each x such that a < x < a C . Therefore a.
Similarly, b, so that 2 .a; b/. Finally, f 0 ./ D 0 by Theorem 6.3.9. t
u
Theorem 6.3.12 (Darboux). If f W a; b ! R is differentiable and f 0 .a/
f 0 .b/, then for each v between f 0 .a/ and f 0 .b/ there exists 2 .a; b/ such that
f 0 ./ D v.
Proof. Apply Lemma 6.3.11 to the function g such that
g.x/ D f .x/ vx
for all x 2 a; b. t

u
0
Remark. If the derivative f is continuous, then Darbouxs theorem follows
immediately from the intermediate-value theorem.
Corollary 6.3.13. Let f be continuous on a; b and differentiable on .a; b/, and
suppose that f 0 .x/ 0 for all x 2 .a; b/. Then f is strictly monotonic on a; b.
Proof. If there exist x and y such that a < x < y 0 for all x 2 .a; b/ or f 0 .x/ < 0 for all x 2 .a; b/. The conclusion
is immediate in both cases, by Theorem 6.3.5 or Theorem 6.3.6, respectively. t
u
Remark. The condition hypothesized for the continuity of f in this corollary may
be relaxed provided that a corresponding change is made to the conclusion. For
example, suppose that the interval a; b is replaced by a; b/, and choose y such
that a < y < b. Then f is continuous on a; y and therefore strictly monotonic on
that interval. As y is any number in .a; b/, we conclude that f is strictly monotonic
on a; b/.
Exercises 6.3.
1.(a) Show that
x2 x2 x3
x < log.1 C x/ < x C
2 2 3
for all x > 0.
(b) Let
n2
1 C n1
sn D
en
for each n > 0. Show that
1 1 1
< log sn < C :
2 2 3n
(c) Compute limn!1 sn .
2.(a) Show that
2x
< log.1 C x/
xC2
for any x > 0. (Hint: Try using Theorem 6.3.5 or 6.3.6.)

(b) Show that
lim n.a1=n 1/ D log a:

n!1
3. Prove that
x2
log.1 C x/ < x
2.1 C x/
if x > 0 and that the inequality is reversed if 1 < x < 0.

4. For all x > 0 let

1 x
f .x/ D 1 C
x
and

1 x
g.x/ D 1 :
x
Show that f is increasing on .0; 1/ and g is decreasing on .1; 1/. [Hint: Study
the signs of the derivatives of log f .x/ and log g.x/.]
5. Use the mean-value theorem to show that the following functions are uniformly
continuous on 0; 1/:
(a) cos kx for every real k.
(b) log.1 C x/.
6. Use the mean-value theorem to show that
1 p p 1
p < nC1 n< p
2 nC1 n
for every n 2 N.
7. Show that the equation
x 4 C 4x C c D 0
has at most two real roots for every real c and exactly two if c < 0.
8. Show that x 5 C 7x 2 D 0 has exactly one real root.
9. Suppose that f is differentiable on R and has two real roots. Show that f 0 has
at least one root.
10. Let p.x/ be a polynomial of degree n 2. Suppose that the equation p.x/ D 0
has n real roots (which may be repeated). Show that p 0 .x/ D 0 has n 1 real
roots.
11. Let f be a function such that f 0 is continuous on a; b and differentiable on
.a; b/, and suppose that
f .a/ D f .b/ D f 0 .a/ D 0:
Show that there exists 2 .a; b/ such that f 00 ./ D 0.

12. Suppose that f 00 exists and is bounded on .a; b/. Show by the mean-value
theorem that f 0 is also bounded on .a; b/ p and hence that f is uniformly
continuous on .a; b/. [Note that the function x is uniformly continuous and
differentiable on .0; 1/, but its derivative is not bounded on .0; 1/.]
13. Let f be a function that is continuous on 0; 1 and differentiable on .0; 1/, and
suppose that f .0/ D f .1/ D 0. Show that there exists c 2 .0; 1/ for which
f 0 .c/ D f .c/. [Hint: Apply the mean-value theorem to f .x/=e x .]
14. Let f be a differentiable function. Suppose that jf 0 .x/j M for all x 2 Df .
Show that pf is Lipschitz continuous.
15. Show that x 2 C 1 and sin x are Lipschitz continuous on R.
6.4 Periodicity of Sine and Cosine
The sine and cosine functions may be used to define another important mathematical
constant. First, we derive the following proposition.
Proposition 6.4.1. There exists
2 .1; 2/ such that cos
D 0:
Proof. Proposition 3.16.1 shows that
x2
cos x 1
2
for every real x. Hence
1 1
cos 1 1 D > 0:
2 2
Similarly,
x2 x4
cos x 1 C ;
2 24
so that
4 16 1
cos 2 1 C D < 0:
2 24 3
An appeal to continuity and the intermediate-value theorem completes the proof. u
t
For the sine function we have the following result.
Proposition 6.4.2. If x 2 .0; 2/, then sin x > 0.
Proof. The proposition follows from the facts that
1
X x 2j C1
sin x D .1/j
j D0
.2j C 1/
6.4 Periodicity of Sine and Cosine 267
X1
x 4j C1 x 4j C3
D
j D0
.4j C 1/ .4j C 3/
1
X
x 4j C1 x2
D 1
j D0
.4j C 1/ .4j C 3/.4j C 2/
for all real x and
.4j C 3/.4j C 2/ 6 > x 2
for all x 2 .0; 2/. t

u
Thus
cos0 x D sin x < 0
for all x 2 .0; 2/, so that cos is decreasing on the interval 0; 2. We infer that the
of Proposition 6.4.1 must be unique. Since sin

> 0, it follows that
p
sin
D 1 cos2
D 1:
Furthermore,
sin 2
D 2 sin
cos
D 0:
We also have the following proposition.

Proposition 6.4.3. If x 2 .0; 2
/, then sin x > 0.
Proof. Since
2 .1; 2/, we already know that sin x > 0 for all x 2 .0;
by
Proposition 6.4.2. It remains to consider the case where
< x < 2
. Let y D x
;
hence 0 < y <
. Moreover
sin x D sin.y C
/ D sin y cos
C cos y sin
D cos y:
We therefore need to show that cos y > 0. But this result is immediate from the
facts that cos
D 0, 0 < y <
, and cos is decreasing on 0;
. u
t
We now define D 2
. This is the new mathematical constant whose
introduction was foreshadowed earlier. Our results therefore show that 2 < < 4,
cos.=2/ D 0, sin.=2/ D 1, and sin D 0. In fact, we see from Proposition 6.4.3
that is the smallest positive number whose sine is 0. Moreover it follows from
Proposition 6.4.3 and the formula cos0 x D sin x that cos is decreasing on 0; .
In addition,

sin x C D sin x cos C cos x sin D cos x (6.3)
2 2 2
and

cos x C D cos x cos sin x sin D sin x (6.4)
2 2 2
for all x. Therefore

sin.x C / D sin x C C D cos x C D sin x:
2 2 2
It follows that sin x < 0 for all x 2 .; 2/ and hence that cos is increasing on
; 2. Furthermore,

cos.x C / D sin x C D cos x
2
and
sin.x C 2/ D sin.x C / D sin x
for all x. Similarly,
cos.x C 2/ D cos x:
Since cos 0 D 1 and cos D cos 0 D 1, an appeal to the continuity of the
cosine and the intermediate-value theorem shows that the cosine, restricted to the
interval 0; , is a bijection between that interval and the interval 1; 1. Note
also that cos.3=2/ D cos.=2/ D 0. Since the cosine is decreasing on 0;
but increasing on ; 2, we see that cos x < 0 if =2 < x < 3=2. Therefore
cos x > 0 if =2 < x < =2. We deduce that the sine function is increasing on
=2; =2 and decreasing on =2; 3=2. Since sin.=2/ D 1 and sin.=2/ D
sin.=2/ D 1, it follows that the sine, restricted to the interval =2; =2, is
a bijection between that interval and the interval 1; 1.
A function f defined for all x and not constant is said to be periodic if there
f .x C / D f .x/
for all x. The smallest such is called the period of f . We have now shown that
the sine and cosine functions are both periodic with period 2.
We have observed that sin is continuous everywhere, and it is also increasing
on =2; =2. Consequently, if its domain were restricted to that interval, then
the resulting function would have an inverse. This inverse is called the inverse sine
function and is denoted by arcsin. Its domain is 1; 1, since sin.=2/ D 1 and
sin.=2/ D 1. As cos x 0 for all x 2 .=2; =2/, we see from the inverse
function theorem (Theorem 6.2.4) that arcsin is differentiable on .1; 1/. In fact, if
y 2 .1; 1/, then y D sin x for some x 2 .=2; =2/, and
1 1 1
arcsin0 .y/ D D Dp
sin0 x cos x 1 y2
since cos x > 0.

A similar argument shows that if the domain of cos were restricted to the interval
0; , then the resulting function would have an inverse. This inverse is called the
inverse cosine function and denoted by arccos. It is also differentiable on .1; 1/.
If y 2 .1; 1/, then y D cos x for some x 2 .0; /, and
1 1
arccos0 .y/ D D p :
sin x 1 y2
Since tan x D sin x= cos x for all x for which cos x 0, we have
cos2 x C sin2 x
tan0 x D D 1 C tan2 x > 0
cos2 x
for all such x. We infer that tan is continuous and increasing on each interval over
which it is defined. An example of such an interval is .=2; =2/. Since
sin x
lim tan x D lim D1
x! 2
x! 2 cos x
and, similarly,
lim tan x D 1;

x! 2 C
it follows that for all y there exists a unique x 2 .=2; =2/ for which tan x D y.
We write x D arctan.y/ and refer to x as the inverse tangent of y. For instance,
arctan.0/ D 0. This function is differentiable, by the inverse function theorem. Since
tan arctan.y/ D y;
differentiation yields
1 D .1 C tan2 arctan.y// arctan0 .y/ D .1 C y 2 / arctan0 .y/;

so that
1
arctan0 .y/ D :
1 C y2
The parentheses around the arguments of the inverse sine, inverse cosine, and
inverse tangent functions are usually omitted.
Let D arcsin.x/, where x 2 1; 1. Then sin D x and =2
=2. Hence =2 =2 and sin. / D sin D x, so that D
arcsin x. We deduce that
arcsin.x/ D arcsin x:
On the other hand, let D arccos.x/, where x 2 1; 1. Then cos D x
and 0 . Hence 0 and
cos. / D cos. / D cos D x;
so that D arccos x. Therefore
arccos.x/ D arccos x:
Example 6.4.1. We now show that
1
lim cos
x!0 x
does not exist. For each positive integer n take
1
sn D
2n
and
2
tn D :
.2n C 1/
Then
n!1 n!1
However,
1
cos D cos 2n D 1
sn
and
1
cos D cos n C D 0 1:
tn 2
Therefore the limit in question does not exist. In fact, this argument shows that
1
lim cos
x!0C x
does not exist. Neither does limx!0 cos.1=x/, by a similar argument. 4

In the next example, we use the sine and cosine functions to show that the
derivative of a differentiable function might not be continuous.
Example 6.4.2. Let f be the real function defined by
(
x 2 sin x1 if x 0;
f .x/ D
0 if x D 0.
By the product and chain rules, f is differentiable at all x 0. Moreover for all
x 0 we have

f .x/ f .0/ f .x/ 1

0
x0 D x D x sin x jxj:
It therefore follows from the sandwich theorem that
f .x/ f .0/
f 0 .0/ D lim D 0:
x!0 x0
Hence f is differentiable at all x.

On the other hand, for all x 0 we have

1 1 1
f 0 .x/ D 2x sin C x 2 2 cos
x x x
1 1
D 2x sin cos :
x x
As
1
lim 2x sin D0
x!0 x
by Example 4.4.1 but limx!0 cos.1=x/ does not exist, we conclude that
limx!0 f 0 .x/ does not exist. Therefore f 0 is not continuous at 0. 4
Our next example shows that a function may have a local extremum at some
point c, yet the values of its derivative might not have equal sign throughout the left
or right half of any neighborhood of c.
Example 6.4.3. Let
(
2x 4 C x 4 sin x1 if x 0;
f .x/ D
0 if x D 0:
Since

1 1
2x C x sin D x 2 C sin
4 4 4
x4 > 0
x x
for all x 0, we find that f .0/ D 0 is a local minimum.

Now f is differentiable at all x 0. In fact,
1 1
f 0 .x/ D 8x 3 C 4x 3 sin x 2 cos
x x
for all x 0. Thus for all integers n 2 we have

1 1 1 4 n
f0 D D < 0;
2n n3 3 4n2 2 4n3 3
whereas

0 1 12
f D 3 > 0:
2n C
2 2n C 2
Since
1
!0
2n
and
1
!0
2n C
2
as n ! 0, we find that each neighborhood N .0/ of 0 contains positive numbers a

and b such that f 0 .a/f 0 .b/ < 0. We can also find negative numbers in N .0/ with
the same property, because

0 1
f <0
2n
and

0 1 8 1
f D C
.2n C 1/ .2n C 1/3 3 .2n C 1/2 2
8 C .2n C 1/
D
.2n C 1/3 3
>0
for all integers n > 1. 4

We now give an example showing that f 0 .c/ may be nonzero, but f is neither
increasing nor decreasing on any interval containing c.
Example 6.4.4. If
(
x 2 sin x1 C x
for x 0;
f .x/ D 2
0 for x D 0,
then
1 1 1
f 0 .x/ D 2x sin cos C
x x 2
for all x 0. In order to compute f 0 .0/, for each x 0 define
f .x/ f .0/
Q.x/ D
x0
f .x/
D
x
1 1
D x sin C :
x 2
Thus
1
f 0 .0/ D lim Q.x/ D :
x!0 2
But for each n 2 Z f0g we have

1 1
f0 D
2n 2
and

0 1 3
f D :
.2n C 1/ 2
Thus f is neither increasing nor decreasing on any interval containing 0. 4

The mean-value theorem may be used to sharpen inequality (3.27).
Example 6.4.5. We shall show that sin x < x for all x > 0. This inequality is
certainly true if x > 1, and so we may suppose that 0 < x 1. As the sine function
is differentiable everywhere, we may apply the mean-value theorem to it on the
interval 0; x to find 2 .0; x/ such that
sin x sin 0 D .x 0/ cos :
Therefore
sin x D x cos < x;
for cos < 1 since 0 < < x 1 < 2. 4

The next example presents a result known as Jordans inequality.
Example 6.4.6. For all x 2 =2; =2 f0g we shall show that
2 sin x
:
x
Define
f .x/ D sin x 2x
for all such x. Then
f 0 .x/ D cos x 2
and
f 00 .x/ D sin x:
The last equation shows that f 0 is decreasing on 0; =2. Since f 0 .0/ D 2 > 0
and f 0 .=2/ D 2 < 0, it follows from the intermediate-value theorem that there
is a unique 2 .0; =2/ such that f 0 ./ D 0. Thus f 0 .x/ > 0 for all x 2 0; /
and f 0 .x/ < 0 for all x 2 .; =2, so that f is increasing on 0; and decreasing
on ; =2. As f .0/ D f .=2/ D 0, we conclude that f .x/ > 0 for all x 2
.0; =2/. Thus
sin x > 2x
for all such x and the desired inequality follows in this case upon division by
x > 0.
If x 2 .=2; 0/, then x 2 .0; =2/. The previous result therefore shows that
sin.x/ > 2x;
whence sin x < 2x and the desired result again holds.

Clearly, equality holds if x D =2.
Jordans inequality spawns several other results. For instance, let 0 < x < =2.
Then

0< x < :
2 2
Moreover

sin x D sin x D cos x
2 2
by Eq. (6.4). Jordans inequality therefore shows that

2 sin 2 x cos x
D :

2
x 2
x
Thus
2 2x
cos x x D1
2
for all x 2 .0; =2/. This inequality is due to Kober. Since
2x
1 cos x ;

Kobers inequality may be rewritten as
1 cos x 2
:
x
Note that equality holds for x D =2.

Further information about these inequalities may be found in [18]. 4
As an illustration of the computation of a sine or a cosine of a real number, we
offer the following example.
Example 6.4.7.
1
sin Dp :
4 2
Proof. First, observe that

1 D sin D 2 sin cos :
2 4 4
Since 0 < =4 < =2, we also know that cos.=4/ > 0. Hence
r
1
D sin cos D sin 1 sin2 ;
2 4 4 4 4
so that
1
D sin2 1 sin2 D sin2 sin4 :
4 4 4 4 4
Thus
2
0 D 4 sin4 4 sin2 C 1 D 2 sin2 1 ;
4 4 4
whence
1
sin2 D
4 2
4
Hence
r
1 1
cos D 1 Dp :
4 2 2
Note that cos x D 0 if and only if x D .2k C 1/=2 for some integer k. For all
other values of x we have
sin x
tan x D :
cos x
For instance,
tan 0 D 0
and
1 p
tan D p 2 D 1:
4 2
Theorem 6.4.4. If cos x D cos y and sin x D sin y, then x D y C 2k for some
integer k.
Proof. First we reduce the problem to values of x and y in 0; 2/. There exists an
integer l such that
2l x < 2.l C 1/ D 2l C 2:
Let x1 D x 2l; hence 0 x1 < 2. Similarly, let m be the integer such that
2m y < 2.m C 1/
and let y1 D y 2m, so that 0 y1 < 2. Note also that cos x1 D cos y1 and
sin x1 D sin y1 .
Suppose that sin x1 D sin y1 D 0. Then fx1 ; y1 g f0; g. Since cos 0 cos ,
it follows that x1 D y1 .
Suppose sin x1 > 0. Then x1 and y1 are both in .0; /. Since cos is decreasing
on that interval and cos x1 D cos y1 , we deduce that x1 D y1 .
Suppose sin x1 < 0. Now fx1 ; y1 g .; 2/. Since cos is increasing on .; 2/
and cos x1 D cos y1 , we again have x1 D y1 .
Thus x1 D y1 in every case. Hence
x 2l D y 2m;
and so
x D y C 2l 2m D y C 2.l m/;
as required. t
u
We now give a geometric interpretation of the sine and cosine functions for real
numbers. The idea is to establish a bijection between the interval 0; 2/ and the
unit circle
x 2 C y 2 D 1;
the circle that is centered at the origin and has radius 1. Given a point .x; y/ on the
circle, we observe that jxj 1. Therefore there is a unique number 2 0; for
which cos D x. Moreover
y 2 D 1 x 2 D 1 cos2 D sin2 ;
Fig. 6.2 sin ' D y and y

cos ' D x
1
P

x
1 0 x Q 1
so that y D sin . We define f .x; y/ D if y D sin . Otherwise y D sin

0. In this case 0, x D cos./ D cos.2/, and y D sin./ D sin.2/,
and we define f .x; y/ D 2 . In each case we find that f .x; y/ is the unique
(see Theorem 6.4.4) number in 0; 2/ with cosine x and sine y. The function f is
a surjection onto 0; 2/: Given ' such that 0 ' < 2, if we put x D cos ' and
y D sin ', then f .x; y/ D ' and x 2 C y 2 D 1. We see that it is also injective, for
if f .u; v/ D f .x; y/ D ', then u D cos ' D x and similarly v D y.
The bijection f manifests itself geometrically as the measurement of an angle by
means of a number in the interval 0; 2/. Thus for every point P D .x; y/ on the
circle x 2 C y 2 D 1 there is a unique angle ' 2 0; 2/, measured from the positive
x-axis in the counterclockwise sense to the line joining the origin O to P , such that
x D cos ' and y D sin '. If 0 < ' < =2, then x > 0 and y > 0. In this case let Q
be the foot of the perpendicular from P to the positive x-axis. Then sin ' and cos '
are the lengths of PQ and OQ, respectively (see Fig. 6.2). If ' D =2, then P is on
the positive y-axis and ' is a right angle.
If we now multiply all coordinates by some factor r > 0, then the unit circle is
replaced by a circle C of radius r but still centered at the origin. Its equation is
x2 C y2 D r 2:
If 0 < ' < =2, then the triangle OPQ is replaced by a right triangle with
hypotenuse of length r extending from the origin to a point .x; y/ on C . The
remaining vertex of the triangle is the point .x; 0/ on the positive x-axis. The angle
from the positive x-axis to the hypotenuse, measured in the counterclockwise sense,
is still ' > 0. The length of the side coincident with the x-axis is x D r cos ', and
Fig. 6.3 sin ' D s=r and

cos ' D t =r
r s
that of the side parallel to the y-axis is y D r sin '. We therefore perceive that, for
an angle ' < =2 in a given right triangle, sin ' is the ratio of the length s of the
side opposite ' to the length r of the hypotenuse. Similarly, cos ' D t =r, where t
is the length of the side adjacent to the right angle and ' (see Fig. 6.3). In fact, the
triangle may be positioned anywhere in the plane and oriented in any manner, so that
the angle ' may be measured, geometrically, from any line. However, since ' > 0,
we use the convention that the angle is measured in the counterclockwise sense.
Negative numbers may similarly be perceived as angles measured in the clockwise
sense. Thus the numbers ' and ' measure the same angle but in opposite senses.
The numbers ' and ' C 2k, for each integer k, also measure the same angle.
For each ' 2 0; 2/ we define D ' 2 if < ' < 2 and D ' otherwise.
Then 2 .; , cos D x=r, and sin D y=r. Moreover 2 .=2; =2/ if
and only if x > 0, whereas 2 .0; / if and only if y > 0. If x D 0, then D =2
if y > 0 and D =2 if y < 0. On the other hand, for y D 0 we have D 0 if
x > 0 and D if x < 0.
We now give a geometric interpretation of the multiplication of complex num-
bers. Suppose that z D x C iy, where x and y are real and jzj D 1. Geometrically,
z is therefore a point on the unit circle centered at the origin. There exists a unique
number 2 .; , which we will call the argument of z, such that x D cos and
y D sin . Thus
z D cos C i sin D e i :
For example,
e i D cos C i sin D 1:
Note also that
z D e i.C2k/
for every integer k, and that
z D cos i sin D cos. / C i sin./ D e i :
Now let us multiply two complex numbers w and z with modulus 1 and respective
arguments and :
wz D e i e i D e i.C/ :
Thus the multiplication of two complex numbers on the unit circle centered at the
origin manifests itself geometrically as a rotation about the origin. We easily obtain
a corresponding result for division:
w e i
D i D e i./ :
z e
Furthermore, since
.e i /n D e in
for every integer n, we obtain the following theorem, which is due to de Moivre.
Theorem 6.4.5 (de Moivre). For each integer n and each real ,
.cos C i sin /n D cos n C i sin n:
More generally, for every z 0 we have

z
D jzj D 1;
jzj jzj
and so there is a unique number 2 .; for which

z
D e i :
jzj
Hence
z D rei ;
where r D jzj. This expression is called the polar form of z. The number is the
argument of z and is written as arg z.
There are various formulas for arg z. Let z D .x; y/ 0, where x and y are real.
If 2 0; =2, then sin D y=r, so that
y
arg z D D arcsin
r
in this case. If 2 .=2; , then 0 < =2 and

y
sin. / D sin. / D sin D :
r
Hence
y
D arcsin ;
r
so that
y
arg z D D arcsin :
r
If 2 =2; 0/, then 0 < =2 and sin. / D y=r, whence

y y
D arcsin D arcsin ;
r r
so that
y
arg z D arcsin :
r
If 2 .; =2/, then 0 < C < =2 and sin. C / D y=r, so that
y y
arg z D arcsin D arcsin :
r r
Note that if z is real, then arg z D 0 if z > 0 and arg z D if z < 0.
We have already observed that
z D jzje i D jzj.cos C i sin /
for all z 0, where D arg z. If we also have
z D jzj.cos C i sin /
for some , then
jzj.cos C i sin / D jzj.cos C i sin /:
Hence cos D cos and sin D sin , and so
D C 2k D arg z C 2k
for some integer k.

If w is another nonzero complex number, then
wz D jwzje i arg wz D jwzj.cos C i sin /;
where D arg wz. If arg w D

, then we can also write
wz D jwje i
jzje i
D jwzje i.
C/
D jwzj.cos.
C / C i sin.
C //:
Hence
cos D cos.
C /
and
sin D sin.
C /:
We conclude that D
C C 2k for some integer k. In other words,
arg wz D arg w C arg z C 2k:

w
arg D arg w arg z C 2k
z
for some integer k.
There are many expressions for . Several can be obtained by using the argument
of a complex number. For instance, starting with the equation
.2 C i /.3 C i / D .2; 1/.3; 1/ D .5; 5/ D 5.1; 1/; (6.5)
observe that the components of the complex numbers .1; 1/; .2; 1/; .3; 1/ are all
positive. Therefore the arguments of these numbers are all in the interval .0; =2/.
p of the number on the right-hand side is equal to arg.1; 1/,
In fact, the argument
which is arcsin.1= 2/ D =4. Letting
1
D arg.2; 1/ D arcsin p ;
5
p
we find that sin D 1= 5, so that
r
1 2
cos D 1 Dp
5 5
and therefore tan D 1=2. Thus arg.2; 1/ D arctan 1=2. Similarly, since
1
arg.3; 1/ D arcsin p
10
p
we have cos arg.3; 1/ D 3= 10, and we deduce that arg.3; 1/ D arctan 1=3. As
arg.2; 1/ C arg.3; 1/ < ;
it follows that
1 1
arg..2; 1/.3; 1// D arctan C arctan :
2 3
Hence
1 1
arctan C arctan D ;
2 3 4
so that

1 1
D 4 arctan C arctan :
2 3
The power series expansion for arctan x, which we will derive later, therefore
provides a means of estimating . More efficient formulas with which to begin can
be found in the exercises. The approximate value of is 3:14159.
In contrast to the injective nature of the exponential function for real variables,
for complex arguments we have the following theorem.
Theorem 6.4.6. For complex w and z we have e w D e z if and only if z D w C 2k i
for some integer k.
Proof. If z D w C 2k i , then, since
e 2k i D cos 2k C i sin 2k D 1;
we have
e z D e wC2k i D e w e 2k i D e w :
To prove the converse, suppose first that e z D e 0 D 1, where z D x C iy for some

real numbers x and y. Then
1 D e z D e xCiy D e x e iy D e x .cos y C i sin y/;

so that e x cos y D 1 and e x sin y D 0. Since e x 0, it follows that sin y D 0, and

so y D n for some integer n. Hence
1 D e x cos n D .1/n e x ;
and since e x > 0, it follows that n is even and e x D 1. The former condition shows
that n D 2k for some integer k, and from the latter we have x D 0, so that
z D iy D n i D 2k i;
as required.
The general case now follows easily, for if e w D e z , then e zw D 1, so that
z w D 2k i
for some integer k. Hence z D w C 2k i . t

u
For all integers n we have cos 2n D 1 and sin n D 0. Thus fcos nxg is
convergent if x D 2k for some k 2 Z, and fsin nxg is convergent if x D k.
We now show that these are the only cases where the sequences are convergent and
x is real.
Theorem 6.4.7. 1. The sequence fcos nxg is convergent if and only if x D 2k for
some k 2 Z.
2. The sequence fsin nxg is convergent if and only if x D k for some k 2 Z.
Proof. Suppose that x 2k for every integer k. Clearly, fcos nxg is divergent if
x D .2k C 1/, for in that case we have cos nx D 1 if n is even and cos nx D 1 if
n is odd. We therefore assume further that x k for every k 2 Z.
Now
cos.n C 1/x cos nx D cos nx cos x sin nx sin x cos nx
for all n. If fcos nxg were convergent, then the left-hand side would converge to 0.
Therefore fsin nxg would converge as well (sin x 0 since x k for each
integer k), and therefore so would fe nxi g since
e nxi D cos nx C i sin nx:
We now find a contradiction by showing that the difference between consecutive

terms of this sequence does not approach 0 as n ! 1. First,
je .nC1/xi e nxi j D je nxi .e xi 1/j D je xi 1j:

But
e xi D cos x C i sin x 1;
for cos x 1 since x 2k for each integer k. Therefore je xi 1j is a positive

constant.
By Cauchys principle, we have now reached the contradiction that the sequence
fe nxi g does not converge. Part (1) of the theorem follows. The proof of part (2) is
similar, using the identity
sin.n C 1/x sin nx D sin nx cos x C cos nx sin x sin nx
for all n. t
u
We conclude this section with the observation that the mean-value theorem does
not hold in general for functions of a complex variable. For example, let f .z/ D e iz
for all z 2 C. If a D 0 and b D 2, then
f .b/ f .a/ e 2 i e 0 11

D D D 0;
ba 2 2
but f 0 .z/ D ieiz 0 for every z.

Exercises 6.4.
1. Use the equation
e ix C e ix
cos x D ;
2
for all x, to express cos 3x in terms of cos x.
2. Use the result of Example 6.4.5 to show that
x2
cos x > 1
2
for all x 0.
3. Show that
1 tan x2

x x
for all x such that 0 < x =2.

4. Prove that the following strengthening of Kobers inequality holds for all x such
that 0 < x =2:
1 cos x tan x2 2
:
x x
[Hint: 1 cos x D 2 sin2 .x=2/.]
5. For each real and , show that
sin x cos. C x/ C cos x sin. C x/
is a constant function. Deduce the addition rules for the sine and cosine
functions from this fact.
6. Differentiate the identity

sin x D cos x
2
to deduce that

cos x D sin x:
2
7. Show that cos x C x sin x is increasing on 0; =2 and hence deduce that
cos x C x sin x > 1
for all x 2 .0; =2.

8. Let and r be real numbers and suppose that jrj < 1.
(a) Show that
P
i. 1 r j cos j D 1r cos
;
Pj1D0 j 12r cos Cr 2
ii. j D1 r cos j D r sin
12r cos Cr 2
.
P
(Hint: Write z D re and investigate 1
i j
j D0 z .)
P P
(b) Find the sums nj D0 cos j and nj D1 sin j. Are they convergent?
(Hint: See Theorem 6.4.7.)
9. Prove that
1
X cos j
j D1
j
converges if and only if is not an integer multiple of 2.

10.(a) Establish the following identities:

i. cos j 12 D cos j cos 2 sin j sin 2 ;

ii. cos j 12 cos j C 12 D 2 sin j sin 2 .
(b) Use part (a) to find the sum
X
n
Sn D sin j:
j D1
Show also that the sequence fSn g is bounded.

11. Show that
sin jzj
lim
x!0 z
does not exist.

12. Show that
p
3
cos D :
6 2
13. Compute .5 i /4 .1 C i / and hence show that
1 1
D 4 arctan arctan :
4 5 239
This formula is due to Machin.
14. Compute
.4 C i /3 .20 C i /
1Ci
and hence show that

1 1 1
D 3 arctan C arctan C arctan :
4 4 20 1985
15. Show that sin z D 0 if and only if z is an integer multiple of .
16. Solve the following equations:
(a) cos z D 0.
(b) cos z D 1.
17. Let n be a positive integer.
(a) Express sin2n x in terms of cos kx, where k 2 Z.
(b) Express sin2nC1 x in terms of sin kx, where k 2 Z.
6.5 LHpitals Rule
The concept of a derivative may be applied to the calculation of limits that do not
succumb to the methods of Chap. 4. First we need a generalized version of the
mean-value theorem. This generalization is due to Cauchy.
Consider two functions f and g that satisfy the hypotheses of the mean-value
theorem on some interval a; b. Then there exist numbers f and g in .a; b/ such
that
f .b/ f .a/
D f 0 .f /
ba
and
g.b/ g.a/
D g 0 .g /:
ba
Thus
f .b/ f .a/ f 0 .f /

D 0 (6.6)
g.b/ g.a/ g .g /
if g.b/ g.a/ and g 0 .g / 0. It is natural to wonder whether there is a number

2 .a; b/ such that Eq. (6.6) holds with f D g D . Cauchys mean-value
formula satisfies our curiosity in this regard.
Theorem 6.5.1 (Cauchys Mean-Value Formula). Let f and g be functions that
are continuous on a closed interval a; b and differentiable on .a; b/. Suppose also
that g 0 .x/ 0 for all x 2 .a; b/. Then there exists 2 .a; b/ such that
f .b/ f .a/ f 0 ./

D 0 :
g.b/ g.a/ g ./
Proof. Notice first that g.b/ g.a/ 0, for if g.a/ D g.b/, then Rolles theorem
would reveal the existence of a number c 2 .a; b/ for which g 0 .c/ D 0. Such a c
cannot exist, however, by hypothesis.
Now define a function F such that
F .x/ D .f .b/ f .a//g.x/ .g.b/ g.a//f .x/

D f .b/g.x/ f .a/g.x/ f .x/g.b/ C f .x/g.a/
for all x 2 a; b. This function is continuous on a; b and differentiable on .a; b/.
Since
F .a/ D f .b/g.a/ f .a/g.b/ D F .b/;

6.5 LHpitals Rule 289
it satisfies all the hypotheses of Rolles theorem. Therefore there is a number 2

.a; b/ such that
0 D F 0 ./ D .f .b/ f .a//g 0 ./ .g.b/ g.a//f 0 ./:
Hence
.f .b/ f .a//g 0 ./ D .g.b/ g.a//f 0 ./;
and the result follows upon division by the nonzero number .g.b/ g.a//g 0 ./. u
t
Remark. The mean-value theorem is the special case where g.x/ D x for all x 2
a; b.
Corollary 6.5.2. Let f and g be functions that are differentiable on an open
interval .a; b/ and suppose that g 0 .x/ 0 for all x 2 .a; b/. Suppose also that
f 0 .x/
lim DL
x!aC g 0 .x/
for some real number L. Then for every " > 0 there exists 2 .0; b a/ such that

f .y/ f .x/

g.y/ g.x/ L < "
for all x and y satisfying a < x < y < a C .

Proof. Choose " > 0. By hypothesis we have .a; b/ Df 0 =g0 and
f 0 .x/
lim D L:
x!aC g 0 .x/
Therefore there exists > 0 such that

0
f .y/

g 0 .y/ L < " (6.7)
for each y 2 .a; b/ satisfying a < y < a C . For each such y we therefore have
f 0 .c/
<LC" (6.8)
g 0 .c/
for all c 2 .a; y/. We may assume that < b a.

Now fix x and y such that a < x < y < a C . Functions f and g satisfy
the hypotheses of Cauchys mean-value formula on x; y. Therefore there exists
2 .x; y/ such that
f .y/ f .x/ f 0 ./

D 0 :
g.y/ g.x/ g ./
As 2 .a; y/, it follows from inequality (6.8) that
f .y/ f .x/
< L C ":
g.y/ g.x/
f .y/ f .x/
> L ";
g.y/ g.x/
and the conclusion follows. t

u
Remark. Corresponding results may be obtained by replacing L with 1. For
instance, if
f 0 .x/
lim D 1;
x!aC g 0 .x/
then for each number M there exists 2 .0; b a/ such that
f .y/ f .x/
>M
g.y/ g.x/
for all x and y satisfying a < x < y < a C .

We now discuss several closely related theorems known collectively as
lHpitals rule. Suppose first that f and g are functions of a real variable x
and that they both approach 0 as x approaches some value (finite or infinite). The
limit, which may be one- or two-sided, of f .x/=g.x/ is then described as being of
the indeterminate form 0=0. LHpitals rule can sometimes be applied to find this
limit. We deal first with the case of one-sided limits.
Theorem 6.5.3. Let f and g be functions that are differentiable on an open interval
.a; b/ and suppose that g 0 .x/ 0 for all x 2 .a; b/. Suppose also that

x!aC x!aC
and
f 0 .x/
lim DL
x!aC g 0 .x/
for some real number L. Then
f .x/
lim D L:
x!aC g.x/
Proof. Choose " > 0. By Corollary 6.5.2 there exists 2 .0; b a/ such that
f .x/ f .y/
L"< <LC" (6.9)
g.x/ g.y/
for all x and y satisfying a < y < x < a C . Thus > 0. Choose x such that
0 < x a < and g.x/ 0. Then
a < x < a C < a C b a D b;
so that x 2 Df =g . Moreover inequality (6.9) holds for each y 2 .a; x/. Taking limits
as y ! aC , we find that
f .x/
L" L C ";
g.x/
so that

f .x/

g.x/ L " < 2"

u
Remark. A similar argument shows that L may be replaced by 1 or 1. Similar
results also hold for limits as x ! b , and therefore for two-sided limits as well.
Example 6.5.1. Evaluate
sin x
lim :
x!0 x
Solution. Since sin x ! 0 as x ! 0, we have
sin x
lim D lim cos x D 1
x!0 x x!0
by lHpitals rule.
4
ex 1
lim :
x!0 sin x
Solution. Note first that
lim .e x 1/ D 0 D lim sin x:

x!0 x!0
Therefore lHpitals rule may be applied. We conclude that
ex 1 ex
lim D lim D 1:
x!0 sin x x!0 cos x
4
Example 6.5.3. LHpitals rule is not applicable to the evaluation of
log x
lim
x!0C x
because
lim log x D 1:

x!0C
An attempt to apply lHpitals rule would yield a wrong answer, since

1
1
lim x
D lim D 1;
x!0C 1 x!0C x
yet for each x 2 .0; 1/ we find that log x is negative and x positive. 4
sin x tan x
lim :
x!0 x2
Solution. Since
lim .sin x tan x/ D 0 D lim x 2 ;

x!0 x!0
we may apply lHpitals rule. Therefore
sin x tan x cos x sec2 x

lim D lim :
x!0 x2 x!0 2x
As
lim .cos x sec2 x/ D 0 D lim 2x;

x!0 x!0
we must apply lHpitals rule once again. We conclude that
sin x tan x cos x sec2 x

lim 2
D lim
x!0 x x!0 2x
sin x 2 sec2 x tan x
D lim
x!0 2
D 0:
4
The next theorem also concerns the indeterminate form 0=0, but the limit is taken
as x ! 1.
Theorem 6.5.4. Let f and g be functions that are differentiable at all x > a for
some positive number a, and suppose that g 0 .x/ 0 for all x > a. Suppose also
that

x!1 x!1
and
f 0 .x/
lim D L;
x!1 g 0 .x/
where L may be any real number, 1, or 1. Then
f .x/
lim D L:
x!1 g.x/
Proof. For each x > a > 0 we may define t D 1=x. Thus 0 < t < 1=a. For
each t satisfying these inequalities, define F .t / D f .1=t / and G.t / D g.1=t /. By
Corollary 4.6.3 we then have

1
lim F .t / D lim f D lim f .x/ D 0
t!0 C t!0 C t x!1
and

1
lim G.t / D lim g D lim g.x/ D 0:
t!0C t!0C t x!1
Moreover F and G are differentiable on .0; 1=a/, and for each t 2 .0; 1=a/ we have
1=t > a, so that

1 1
G .t / D 2 g 0
0
0
t t
by hypothesis. Therefore Theorem 6.5.3 is applicable. We conclude that

f .x/ f 1t
lim D lim 1
x!1 g.x/ t!0C g
t
F .t /
D lim
t!0C G.t /
F 0 .t /
D lim
t!0C G 0 .t /
1
t12 f 0
D lim 1t
t!0C t12 g 0 t
f 0 .x/
D lim 0
x!1 g .x/
D L:
t
u
Remark. A similar proof establishes a corresponding theorem in which x ! 1.
Sometimes an application of these theorems makes matters worse.
Example 6.5.5. Suppose we wish to evaluate
lim x log x:
x!0C
We may convert this limit to the indeterminate form 0=0 by writing

x
x log x D 1
log x
for all x > 0 such that x 1, because
1
lim D0
x!0C log x
since log x ! 1 as x ! 0C . An application of lHpitals rule then shows that
x 1 x
lim 1
D lim D lim 1
:
x!0C x!0C 1 log12 x x!0C
log x x log2 x
Although the limit on the right-hand side is still of the indeterminate form 0=0, it is
more complicated than the limit on the left-hand side. The appeal to Theorem 6.5.3
has therefore failed to solve the problem. However, we can also write
log x
x log x D 1
;
x
for all x > 0, and note that
lim log x D 1
x!0C
and
1
lim D 1:
x!0C x
4
The observation at the end of the example above motivates a study of other
indeterminate forms. We prepare for the resulting theorem with a helpful lemma
concerning limits.
Lemma 6.5.5. Let f be a function defined on an interval .a; b/ and let L be a
number. Suppose that for each q1 > L there exists c1 2 .a; b/ such that f .x/ < q1
whenever x 2 .a; c1 /. Suppose similarly that for each q2 < L there exists c2 2
.a; b/ such that q2 < f .x/ whenever x 2 .a; c2 /. Then
lim f .x/ D L:
x!aC
Proof. Choose " > 0. Then L " < L < L C ". Consequently, there exist numbers
c1 and c2 in .a; b/ such that f .x/ < L C " whenever x 2 .a; c1 / and L " < f .x/
whenever x 2 .a; c2 /. If we choose x such that
0 < x a < minfc1 a; c2 ag;
then a < x < c1 and a < x < c2 , so that
L " < f .x/ < L C ":
Hence jf .x/ Lj < ", and the result follows. t

u
Theorem 6.5.6. Let f and g be functions that are differentiable on an open interval
.a; b/ and suppose that g 0 .x/ 0 for all x 2 .a; b/. Suppose also that
lim g.x/ D 1
x!aC
and
f 0 .x/
lim DL
x!aC g 0 .x/
for some real number L. Then
f .x/
lim D L:
x!aC g.x/
Discussion: One obvious way to attempt a proof of this theorem is to try to show
that limx!aC f .x/ D 1 and then to convert the limit to the indeterminate form 0=0
by writing f .x/=g.x/ as
1
g.x/
1
:
f .x/
Using Theorem 6.5.3 together with the chain rule, we should then obtain
1
f .x/ g.x/
lim D lim 1
x!aC g.x/ x!aC
f .x/
0
gg2.x/
.x/
D lim 0
x!aC ff 2.x/
.x/
f 2 .x/ g 0 .x/
D lim 2
lim 0
x!aC g .x/ x!aC f .x/

f .x/ 2 1
D lim :
x!aC g.x/ L
We now encounter several problems, the most obvious of which is that L might
be 0. Furthermore, limx!aC f .x/=g.x/ might not exist. Even if it were to exist, it
would have to be nonzero in order for us to be able to deduce the required result
as a consequence of the calculation above. The following proof circumvents these
difficulties.
Proof. Choose q1 > L. The density property shows that there is a q satisfying
L < q < q1 . By Corollary 6.5.2 there exists 2 .0; b a/ such that

f .x/ f .y/

g.x/ g.y/ L < q L
for all x and y satisfying a < x < y < a C . Fix a y for which
a < y < a C < b:

It follows that
f .x/ f .y/
<LCqLDq (6.10)
g.x/ g.y/
for all x 2 .a; y/.

Since
lim g.x/ D 1;
x!aC
there exists 1 > 0 such that
g.x/ > maxf0; g.y/g (6.11)
for all x 2 .a; b/ satisfying 0 < x a < 1 . Let
c D minfy; a C 1 g > a:
Thus c 2 .a; b/, since y < b.

Now choose x 2 .a; c/. Then a < x < c y, so that inequality (6.10) holds.
But we also have
0 < x a < c a 1 ;
so that g.x/ g.y/ > 0 by inequality (6.11). Therefore
f .x/ f .y/ < q.g.x/ g.y//;
and so
f .x/ < f .y/ C qg.x/ qg.y/:
Moreover g.x/ > 0. We conclude that
f .x/
< h.x/
g.x/
for all x 2 .a; c/, where
f .y/ g.y/
h.x/ D Cqq
g.x/ g.x/
for all such x.

Now
lim h.x/ D q
x!aC
since
1
lim D0
x!aC g.x/
by Corollary 4.6.4. Therefore there exists 2 > 0 such that
jh.x/ qj < q1 q (6.12)
for all x 2 .a; c/ satisfying 0 < x a < 2 . Define
c1 D minfc; a C 2 g > a:
Thus c1 2 .a; b/. Choose x 2 .a; c1 /. Then x 2 .a; c/ and
0 < x a < c 1 a 2 :
Therefore inequality (6.12) holds, so that
h.x/ < q C q1 q D q1 :
We conclude that
f .x/
< h.x/ < q1 :
g.x/
In summary, we have now shown that for each q1 > L there exists c1 2 .a; b/
such that
f .x/
< q1
g.x/
whenever x 2 .a; c1 /. A similar argument shows that for each q2 < L there exists
c2 2 .a; b/ such that
f .x/
q2 <
g.x/
whenever x 2 .a; c2 /. An appeal to Lemma 6.5.5 therefore completes the proof. u

t
lim x log x:
x!0C
Solution. We have
log x
lim x log x D lim 1
:
x!0C x!0C
x
Since limx!0C 1=x D 1, it follows from Theorem 6.5.6 that
1
lim x log x D lim x
D lim x D 0:
x!0C x!0C x12 x!0C
4
Theorem 6.5.6 may be extended to limits as x approaches a , a, 1, or 1.
Remark. If lim f 0 .x/=g 0 .x/ does not exist, then we cannot draw any conclusion
about lim f .x/=g.x/.
Example 6.5.7. If f .x/ D sin x and g.x/ D x for all x, then f 0 .x/ D cos x and
g 0 .x/ D 1, so that limx!1 f 0 .x/=g 0 .x/ does not exist. 4
When applying lHpitals rule to find limx!a f .x/=g.x/, we must make sure
that all the hypotheses are satisfied, as we may get spurious results otherwise.
For instance, if g 0 has a zero in each neighborhood of a, then we must not apply
lHpitals rule. Corresponding remarks hold if x approaches aC , a , 1, or 1. In
a case where f 0 .x/ D .x/h.x/, g 0 .x/ D .x/k.x/, and .x/ does not approach
a limit but h.x/=k.x/ does, then we must resist the temptation to cancel .x/ in
f 0 .x/=g 0 .x/. In 1879, Stolz gave an example to illustrate this point. In 1956, Boas
constructed infinitely many examples, including the one constructed by Stolz. We
present an example here.
Example 6.5.8. For all x let
sin 2x C 2x
f .x/ D ;
4
and let ' be any function such that the functions '.sin x/ and ' 0 .sin x/ are
positive and bounded for all x. For instance, the exponential function satisfies these
conditions. Let
g.x/ D f .x/'.sin x/
for all x. Thus
lim g.x/ D lim f .x/ D 1:

x!1 x!1
Moreover g.x/ > 0 for all x > 1=2.

Now let us attempt to apply lHpitals rule to compute limx!1 f .x/=g.x/.
First we check that the hypotheses are satisfied. For all x we have
cos 2x C 1
f 0 .x/ D D cos2 x
2
and
g 0 .x/ D '.sin x/ cos2 x C f .x/' 0 .sin x/ cos x

D .'.sin x/ cos x C f .x/' 0 .sin x// cos x:
Thus g 0 .x/ D 0 whenever cos x D 0. Consequently, for every a there exist values
of x > a such that g 0 .x/ D 0. We conclude that lHpitals rule cannot in fact be
applied.
Note that cos x is a common factor of f 0 .x/ and g 0 .x/. Canceling this factor, we
obtain
f 0 .x/ cos x
D
g 0 .x/ '.sin x/ cos x C f .x/' 0 .sin x/
for 1all x such that g 0 .x/ 0. The properties hypothesized for ' and the fact that
limx!1 f .x/ D 1 therefore show that
f 0 .x/
lim D 0:
x!1 g 0 .x/
However,
f .x/ 1
D 1
g.x/ '.sin x/
for all x > 1=2, and this quotient does not approach 0 as x ! 1. 4
It is easy to use lHpitals rule to prove that log x << x:
log x 1
lim D lim D 0:
x!1 x x!1 x
In fact, it can be proved by induction that logk x << x for every positive integer k,
for if we assume that
logk x
lim D0
x!1 x
for some fixed k, then, using lHpitals rule, we obtain
logkC1 x .k C 1/ logk x
lim D lim D 0:
x!1 x x!1 x
Exercises 6.5.
1. Show that the conclusion of Theorem 6.5.1 does not hold for the functions
f .x/ D 4x 3 C 6x 2 12x
and
g.x/ D 3x 4 C 4x 3 6x 2
on the interval 0; 1. Which of the conditions of the theorem is not satisfied in
this case?
2. Show that Theorem 6.5.6 may not be applied to evaluate
x C cos x
lim
x!1 x C sin x
directly.
3. Let
f .x/ D x C cos x sin x
and
g.x/ D e sin x f .x/
for all x. Show that lHpitals rule does not apply to the evaluation of
f .x/
lim :
x!1 g.x/
Does the limit exist?

4. Show that lHpitals rule does not apply to the evaluation of
2x C sin 2x
lim :
x!1 x sin x C cos x
Does the limit exist?

5. Evaluate the following limits:
e x 1
(a) limx!0 log.xC1/
;
x log x
(b) limx!1 x 2 C1 ;
x1
(c) limx!1C ex1 ;
log sin x
(d) limx!0C log x ;
(e) limx!1 x.log.x C 2/ log x/;
(f) limx!1 x.a1=x 1/, where a > 0.
(g) limx!.1/C .x C 1/1=3 log.x C 1/;
x
(h) limx!0C xp;
(i) limx!1 . x2 C x x/;
(j) limx!.=2/ x 2 tan x;
x
(k) limx!1 xC1x2
.
6. The determinant

a11 a12 a13

a21 a22 a23

a a a
31 32 33
of the 3 3 matrix
0 1
a11 a12 a13
@ a21 a22 a23 A
a31 a32 a33
is defined by the equation

a11 a12 a13

a21 a22 a23 D a11 a22 a23 a12 a21 a23 C a13 a21 a22 :
a32 a33 a31 a33 a31 a32
a a a
31 32 33
Let f; gW a; b ! R. Apply Rolles theorem with

1 1 1

F .x/ D f .a/ f .x/ f .b/
g.a/ g.x/ g.b/
to prove Cauchys mean-value formula. (See question 7 of the exercises for

Sect. 6.2.)
7. Let f and g be functions that are continuous on a; b and differentiable on
.a; b/. Then, as in the proof of Cauchys mean-value formula, there exists 2
.a; b/ satisfying the equation
.f .b/ f .a//g 0 ./ D .g.b/ g.a//f 0 ./: (6.13)

Show that Cauchys mean-value formula does not apply in the case where
f .x/ D 3x 4 2x 3 x 2 C 1
and
g.x/ D 4x 3 3x 2 2x
for all x 2 0; 1 but that there exists 2 .0; 1/ satisfying Eq. (6.13).
8. [17] Let 1 ; 2 ; 3 be real numbers with sum equal to 1. Let f1 ; f2 ; f3 be
functions that are continuous on a; b and differentiable on .a; b/, and suppose
that fk .a/ fk .b/ for each k 2 f1; 2; 3g. Show that there exists 2 .a; b/ for
which
1 2 3
f10 ./ C f20 ./ C f 0 ./ D 0:
f1 .b/ f1 .a/ f2 .b/ f2 .a/ f3 .b/ f2 .a/ 3
[Hint: Apply Rolles theorem with
F .x/ D 1 .f2 .b/ f2 .a//.f3 .b/ f3 .a//.f1 .x/ f1 .a//

C2 .f1 .b/ f1 .a//.f3 .b/ f3 .a//.f2 .x/ f2 .a//
C3 .f1 .b/ f1 .a//.f2 .b/ f2 .a//.f3 .x/ f3 .a//:
Obtain an expression for f10 ./ by taking 1 D 1, and derive Cauchys mean-
value formula by taking 1 D 1, 2 D 1, and 3 D 0.
The result can be generalized to an arbitrary number of functions. Use it to
show that the equation
x ex 1
f .x/ D 3x C cos C C D0
2 2 e1 .x C 1/ log 2
has at least one solution in .0; 1/. [Note that f .0/ and f .1/ are both positive
and so the conclusion is not an obvious consequence of the intermediate-value
theorem.]
9. [2] Let a and b > a be real numbers and suppose that f and g are
functions that are continuous on a; b and differentiable on .a; b/. Suppose
also that g 0 .x/ 0 for all x 2 .a; b/. Show that if f 0 .x/=g 0 .x/ is increasing
(respectively, decreasing), then so are
f .x/ f .a/
g.x/ g.a/
and
f .x/ f .b/
:
g.x/ g.b/
[Hint: Note that Corollary 6.3.13 implies that either g 0 .x/ < 0 for all x 2 .a; b/
or g 0 .x/ > 0 for all x 2 .a; b/. Assuming that f 0 .x/=g 0 .x/ is increasing, show
that the derivative of
f .x/ f .a/
g.x/ g.a/
is nonnegative.]
10. This problem concerns the following question. Suppose that f and g are
functions such that
f .x/
lim (6.14)
x!aC g.x/
is of the indeterminate form 0=0, where a is a real number. It follows from

Theorem 6.5.3 that if
f 0 .x/
lim (6.15)
x!aC g 0 .x/
exists, then so does the limit (6.14). The question is whether we can extend this
observation to say that if f 0 .x/=g 0 .x/ exists throughout some interval .a; aC/,
where > 0, but the limit (6.15) does not exist, then the limit (6.14) also does
not exist.
(a) Show that the answer to the question posed is no by considering the
functions defined by
1
f .x/ D x 2 sin ;
x
for all x 0, and g.x/ D sin x. Use the facts that
1
lim x sin D0
x!0C x
(see Example 4.4.1),
1
lim cos
x!0C x
does not exist (Example 6.4.1) and
sin x
lim D1
x!0C x
6.6 A Discrete Version of lHpitals Rule 305
(Example 6.5.1) in order to demonstrate that f 0 .x/=g 0 .x/ exists throughout

the interval .0; =2/ and that
f 0 .x/
lim
x!0C g 0 .x/
does not exist but
f .x/
lim D 0:
x!0C g.x/
(b) i. Show that if f 0 .x/=g 0 .x/ exists throughout some interval .a; a C/, where
> 0, then for each x 2 .a; a C / there exists 2 .a; x/ such that
f .x/ f 0 ./
D 0 :
g.x/ g ./
ii. Why is it therefore not valid to conclude that the answer to the question
posed is yes by using the following argument?
If the limit (6.15) does not exist, then
f 0 ./
lim
x!aC g 0 ./
does not exist since a < < x for each x. Because
f .x/ f 0 ./
D 0 ;
g.x/ g ./
it therefore follows that the limit (6.14) also does not exist.
6.6 A Discrete Version of lHpitals Rule
LHpitals rule is a powerful tool for calculating limits of indeterminate forms.

However, in applications one may encounter an indeterminate form f .x/=g.x/
whose limit exists even though functions f and g are not differentiable or the
limit of f 0 .x/=g 0 .x/ does not exist. In 1988, Huang [8] dealt with this problem
by proving a discrete version of lHpitals rule.
Let f be a function defined at all x > a for some number a. For each x > a and
h > a x let us define
h f .x/ D f .x C h/ f .x/:
Then
h f .x/
f 0 .x/ D lim :
h!0 h
Huangs result shows that, under certain circumstances,
f .x/ h f .x/
lim D lim :
x!1 g.x/ x!1 h g.x/
Theorem 6.6.1. Let f and g be functions defined at all x > a for some number a.
Suppose that
1.
lim f .x/ D lim g.x/ D 0;

x!1 x!1
and
2. for some h > 0,
(a) either h g.x/ > 0 for all x > a or h g.x/ < 0 for all x > a, and
(b)
h f .x/
lim D L;
x!1 h g.x/
where L may be any real number, 1, or 1.

Then
f .x/
lim D L:
x!1 g.x/
Proof. Let us assume that h g.x/ > 0 for all x > a; the argument in the other case
is similar.
Case 1: Suppose that L is a real number. Choose " > 0. By assumption there
exists N > a such that
f .x C h/ f .x/
L"< <LC"
g.x C h/ g.x/
for all x N . Fix x N . Then for all positive integers k we have

x C .k 1/h N , and so
f .x C kh/ f .x C .k 1/h/
L"< < L C ":
g.x C kh/ g.x C .k 1/h/
Since
g.x C kh/ g.x C .k 1/h/ > 0;
it follows that
.L "/.g.x C kh/ g.x C .k 1/h//

< f .x C kh/ f .x C .k 1/h/
< .L C "/.g.x C kh/ g.x C .k 1/h//:
Summation from k D 1 to k D n, where n is a positive integer, gives
.L "/.g.x C nh/ g.x// < f .x C nh/ f .x/ < .L C "/.g.x C nh/ g.x//;
by the telescoping property. Taking limits as n ! 1, we obtain
.L "/.g.x// f .x/ .L C "/.g.x//;
by Theorem 4.6.2. Our initial assumption shows that the sequence fg.x C nh/g
is increasing. Since
lim g.x C nh/ D 0;

n!1
it follows that g.x/ < 0. Thus
f .x/
L" L C ";
g.x/
from which the required result follows.

Case 2: If L D 1, then for each M > 0 there exists N > a such that
f .x C h/ f .x/ > M.g.x C h/ g.x//
whenever x N . Fix such an x. Using the method employed in the previous

case, we find that
f .x C nh/ f .x/ > M.g.x C nh/ g.x//
for all positive integers n. By letting n ! 1, we see that
f .x/
M;
g.x/

Case 3: The case where L D 1 can be dealt with by applying the argument of
case 2 to the functions f and g.
t
u
By applying the proof of this theorem to sequences, we obtain the following
result.
Corollary 6.6.2. Let fan g and fbn g be sequences that converge to 0. Let h be a
positive integer for which there exists a real number N such that the sign of bnCh bn
is constant for all n N . Suppose that
anCh an
lim D L;
n!1 bnCh bn

an
lim D L:
n!1 bn
Example 6.6.1. For each x > 1 define
sin 2x
f .x/ D
bxc
and
p p 1
g.x/ D xC1 xDp p :
xC1C x
Then
lim f .x/ D lim g.x/ D 0:

x!1 x!1
We wish to evaluate limx!1 f .x/=g.x/. Evidently, lHpitals rule is not

applicable. Taking h D 1 in Theorem 6.6.1, we obtain
1 g.x/ D g.x C 1/ g.x/

p p
D xC2 x
2
D p p
xC2C x
>0
and
1 f .x/ D f .x C 1/ f .x/
sin 2.x C 1/ sin 2x
D
bx C 1c bxc

1 1
D sin 2x
bxc C 1 bxc
sin 2x
D ;
bxc.bxc C 1/
so that
p p
1 f .x/ . x C 2 C x/ sin 2x
D I
1 g.x/ 2bxc.bxc C 1/
hence
p
1 f .x/
< 2 xC2 !0
g.x/ 2.x 1/2
1
as x ! 1. We conclude that
f .x/
lim D 0:
x!1 g.x/
4
We now establish another version of the theorem.
Theorem 6.6.3. Let f and g be functions defined at all x > a for some number
a, and suppose that they are bounded on every finite subinterval of .a; 1/. Suppose
also that
1.
lim g.x/ D 1;
x!1
and
2. for some h > 0,
(a) either h g.x/ > 0 for all x > a or h g.x/ < 0 for all x > a, and
(b)
h f .x/
lim D L;
x!1 h g.x/
where L may be any real number, 1, or 1.

Then
f .x/
lim D L:
x!1 g.x/
Proof. Let us assume that h g.x/ > 0 for all x > a, as the argument in the other
case is similar. Since limx!1 g.x/ D 1, we may also assume that g.x/ > 0 for
all x > a.
Case 1: Suppose L is a real number. Choose " > 0. Arguing as in the proof of
Theorem 6.6.1, we deduce the existence of N > a such that

f .x C nh/ f .x/

g.x C nh/ g.x/ L < " (6.16)
for all x N and all positive integers n.

Choose x N C h, and write x D N C h. Let j D bc and r D N C . j /h.
Thus j > 0, x D r C jh, and r 2 N; N C h/. Furthermore,
g.x/ g.r/ D g.r C jh/ g.r/ > 0;
so that
g.r/
0< < 1I
g.x/
hence

1 g.r/ < 1;
g.x/
and so

f .x/ f .r/ Lg.r/ f .x/ f .r/
L
g.x/ L g.x/ C
g.x/ g.r/
because
f .x/ f .x/ Lg.x/

L D
g.x/ g.x/
f .r/ Lg.r/ C f .x/ f .r/ Lg.x/ C Lg.r/
D
g.x/

f .r/ Lg.r/ g.x/ g.r/ f .x/ f .r/ L.g.x/ g.r//
D C
g.x/ g.x/ g.x/ g.r/

f .r/ Lg.r/ g.r/ f .x/ f .r/
D C 1 L :
g.x/ g.x/ g.x/ g.r/
Note that

f .x/ f .r/ f .r C jh/ f .r/

g.x/ g.r/ L D g.r C jh/ g.r/ L < "
by inequality (6.16). Furthermore, since f and g are bounded on N; N C h/ and

limx!1 g.x/ D 1, there exists N1 N such that

f .r/ Lg.r/
<"
g.x/
for all x N1 . For all such x we therefore have

f .x/

g.x/ L < 2";
and so
f .x/
lim D L:
x!1 g.x/
Case 2: Suppose that L D 1.

We first show that h f .x/ > 0 for sufficiently large x. Choose M > 0. There
exists N > a such that
f .x C h/ f .x/
>M
g.x C h/ g.x/
for all x N . Hence
h f .x/ D f .x C h/ f .x/ > M.g.x C h/ g.x// > 0
for all x N , as claimed.

As in Theorem 6.6.1, we see that
f .x C nh/ f .x/
>M
g.x C nh/ g.x/
for all x N and all positive integers n. Choose x N C h, and write x D r C jh,
where r 2 N; N C h/ and j is a positive integer. Thus
f .x/ f .r/ f .r C jh/ f .r/

D > M;
g.x/ g.r/ g.r C jh/ g.r/
so that
f .x/ > M.g.x/ g.r// C f .r/:
Hence
lim f .x/ D 1;
x!1
and so we may assume also that f .x/ > 0 for all x > a.
Finally,
h g.x/ 1
lim D lim h f .x/
x!1 h f .x/ x!1
h g.x/
and
h f .x/
lim D 1:
x!1 h g.x/
Consequently,
h g.x/
lim D 0:
x!1 h f .x/
It therefore follows from case 1 that
g.x/
lim D 0;
x!1 f .x/
whence
f .x/ 1
lim D lim g.x/ D 1;
x!1 g.x/ x!1
f .x/
because f .x/=g.x/ > 0 for all x > a.

Case 3: The case where L D 1 can be handled by the same argument with f
replaced by f .
t
u
We immediately obtain the corresponding result for sequences.
Corollary 6.6.4. Let fan g and fbn g be sequences, where limn!1 bn D 1. Let h
be a positive integer for which there exists a real number N such that the sign of
bnCh bn is constant for all n N . Suppose that
anCh an
lim D L;
n!1 bnCh bn

an
lim D L:
n!1 bn
By taking h D 1, we obtain the following result, which is due to Stolz.

Corollary 6.6.5. Let fan g be a sequence and fbn g an increasing sequence such that
limn!1 bn D 1. If
anC1 an
lim DL
n!1 bnC1 bn
for some real L, then

an
lim D L:
n!1 bn
Example 6.6.2. Let x1 D 1=2 and
xnC1 D xn .1 xn /
for all n 1. It is easy to see by induction that 0 < xn < 1 for all n, and so
xnC1 < xn for all n. Thus the sequence fxn g is decreasing. It therefore converges to
some number L. From the recurrence relation we obtain
L D L.1 L/
by taking limits, and we infer that L D 0. Thus the sequence f1=xn g is increasing
and approaches infinity as n does so. Noting first that
xn xnC1 D xn xn .1 xn / D xn2
for all n, we conclude from Stolzs theorem, with an D n and bn D 1=xn for all n,
that
n
lim nxn D lim
n!1 n!1 1
xn
.n C 1/ n
D lim
n!1 1
xnC1
x1n
xn xnC1
D lim
n!1 xn xnC1
xnC1
D lim
n!1 xn
D lim .1 xn /
n!1
D 1:
4
Corollary 6.6.6. Let a be a positive number and let k be a function that is defined
and positive at all x > a and bounded on every finite subinterval of .a; 1/. Suppose
that
k.x C 1/
lim D L;
x!1 k.x/
where L may be any positive number or 1. Then
lim .k.x//1=x D L:
x!1
Proof. Let f .x/ D log k.x/ and g.x/ D x for all x > a. Then
f .x C 1/ f .x/ k.x C 1/
D log k.x C 1/ log k.x/ D log ! log L
g.x C 1/ g.x/ k.x/
as x ! 1. It therefore follows from Theorem 6.6.3 that
log k.x/
lim D log L:
x!1 x
Hence
log k.x/
lim .k.x//1=x D lim exp log.k.x//1=x D lim exp D exp log L D L:
x!1 x!1 x!1 x
t
u
As a consequence of Corollary 6.6.6, we obtain the corresponding result for
sequences.
Corollary 6.6.7. If fan g is a sequence of positive terms and
anC1
lim D L;
n!1 an
then
lim an1=n D L:
n!1
Exercises 6.6.
1. Use the work of this section to show that
0 1
Xn X
n
lim @ .2j 1/2 j 2 A D 4:
n!1
j D1 j D1
2. Use the work of this section to show that
1 X
n
1
lim kk D
n!1 nkC1 j D1
kC1
for all positive integers k.

3. Use Theorem 6.6.3 to find
sin x C x
lim :
x!1 x
4. For each x 0 let k.x/ be the integer such that
x D y.x/ C 2k.x/
for some number y.x/ 2 0; 2/, and let
g.x/ D sin y.x/ C k.x/:
Show that
x
lim D 2:
x!1 g.x/
5. Use the work in this section to find

log x
lim
x!1 x
and
x
lim :
x!1 ex
6. If fan g and fbn g are sequences such that
lim an D lim bn D 0
n!1 n!1
and
an
lim DL
n!1 bn
for some number L, show that

0 1
X
n X
n
lim @ aj bj A D L:
n!1
j D1 j D1
7. Prove that if
lim .xn xnC1 / D L;

n!1
then
xn
lim D L:
n!1 n
6.7 Differentiation of Power Series
We move on to consider derivatives of power series. Let us define

1
X
f .z/ D aj .z c/j
j D0
for each z within the circle of convergence of the power series on the right-hand
side. If we differentiate the power series term by term, we obtain a new series,
1
X
jaj .z c/j 1 ;
j D1
which is said to be the corresponding derived series. It is natural to ask whether the
derived series has the same circle of convergence and, if so, whether the function it
defines is f 0 . It turns out that the answers to these questions are affirmative within
the interior of the circle of convergence.
First we investigate the radius of convergence.
Theorem 6.7.1. A power series and its derived series have the same radius of
convergence.
6.7 Differentiation of Power Series 317
Proof. It suffices to consider power series with center 0. Let r1 and r2 be the radii
of convergence of the power series
1
X
aj zj
j D0
and its derived series, respectively. For each positive integer n we have
jan zn j jnan zn j D jzjjnan zn1 j:
Thus if the derived series converges, then so does the given series by the comparison
test. We deduce that r1 r2 .
It remains to show that r2 r1 . This inequality certainly holds if r1 D 0. Assume
therefore that r1 > 0. Choose z and r such that 0 < jzj < r < r1 . Then
1
X
jaj r j j
j D0
is convergent, and so
0 D lim jan jr n D r lim jan jr n1 :

n!1 n!1
As it therefore converges to 0, the sequence fjan jr n1 g is bounded above by some

number M . Thus
z n1

jnan zn1 j D njan jr n1
r
z n1

Mn :
r
Now the series
X1 z j 1

j
j D1
r
converges by the ratio test, since
.n C 1/jzjn r n1 n C 1 jzj

n
n1
D
r njzj n r
and this quantity approaches jzj=r < 1 as n ! 1. Therefore the derived series
converges at jzj by the comparison test. We conclude that r2 r1 , as required. u
t
P1
By applying the theorem k times, we see that the series j D0 aj zj has the same
radius of convergence as
1
X 1
X j
j.j 1/ .j k C 1/aj zj k D aj zj k :
.j k/
j Dk j Dk
The theorem cannot be applied to deduce information about convergence on

the circle of convergence or, in the real case, at the end points of the interval of
convergence. For instance, the real series
1
X .1/j 1
xj
j D1
j
is convergent when x D 1, by Leibnizs test. Therefore the radius of convergence,

r, is at least 1. The derived series,
1
X
.1/j x j ;
j D0
diverges when x D 1. Thus, r D 1. However, the two series exhibit different

behaviour in regard to convergence at at least one endpoint of the interval of
convergence.
Theorem 6.7.2. Let
1
X
f .z/ D aj zj
j D0
for all z such that jzj < r, where r is the radius of convergence of the power series
and is nonzero. Then f is differentiable at all z such that jzj < r, and
1
X
f 0 .z/ D jaj zj 1
j D1
for all such z.

Proof. Choose w such that jwj < r. Also, choose t such that jwj < t < r, and z
such that jzj < t and z w. Define
1
f .z/ f .w/ X
D jaj wj 1
zw j D1
1 1
1 X X
D aj .zj wj / jaj wj 1
z w j D1 j D1
1
X
zj wj j 1
D aj jw :
j D2
zw
The challenge is to show that ! 0 as z ! w. It will be met by finding an upper

bound for j j.
We begin by investigating the expression in parentheses. Noting that
j 1
X zj wj
zk wj k1 D
zw
kD0
for all j > 0, by Theorem 1.5.5, and that
j 1 j 1
X X
j 1 j 1
w Dw 1 D j wj 1
kD0 kD0
for all j > 0, we find that
j 1 j 1
zj wj X X
j wj 1 D zk wj k1 wj 1
zw
kD0 kD0
j 1
X
D .zk wj k1 wj 1 /
kD0
j 1
X
D wj k1 .zk wk /
kD1
j 1
X X
k1
D wj k1 .z w/ zm wkm1
kD1 mD0
j 1 k1
X X
D .z w/ zm wj m2 :
kD1 mD0
Since t > maxfjzj; jwjg, for every nonnegative s and u we have
jzs wu j D jzjs jwju

tstu
D t sCu :
Thus
j j

aj z w j wj 1 D jaj j z w j wj 1
j j
zw zw

Xj 1 k1
X

D jaj j .z w/ zm wj m2

kD1 mD0
j 1 k1
X X
jaj jjz wj jzm wj m2 j
kD1 mD0
j 1 k1
X X
jaj jjz wj t j 2
kD1 mD0
j 1 k1
X X
D jaj jjz wjt j 2 1
kD1 mD0
j 1
X
D jaj jjz wjt j 2 k
kD1
j.j 1/
D jz wjjaj t j 2 j
2
for all j > 1, since
j 1
X j.j 1/
kD
2
kD1
P1
by Theorem 1.5.4. As 0 < t < r, the series j D0 jaj t j j converges, and it follows
by two applications of Theorem 6.7.1 that
1
X
jj.j 1/aj t j 2 j
j D2
also converges. Writing

1
X
SD jj.j 1/aj t j 2 j
j D2
1
X
D j.j 1/jaj t j 2 j;
j D2
we obtain
X1 j

aj z w j wj 1
j
j j
j D2
zw
1
jz wj X
j.j 1/jaj t j 2 j
2 j D2
jz wjS
D :
2
Hence ! 0 as z ! w, as required. t
u
P1 j
Corollary 6.7.3. Let r be the radius of convergence of the series j D0 aj z . Then
the series
1
X zj C1
aj
j D0
j C1
converges for all z such that jzj < r.

Proof. This result is immediate from the theorem, for the first series is the derivative
of the second. t
u
The application of Theorem 6.7.2 is sometimes referred to as term-by-term
differentiation and that of Corollary 6.7.3 as term-by-term integration. We will study
the general concept of integration in the next chapter.
Theorem 6.7.2 implies that the limit of a power series is continuous within the
interior of the interval of convergence.
Example 6.7.1. Theorem 3.2.1 shows that
1
X 1
.1/j zj D
j D0
1Cz
for all z such that jzj < 1, since .1/j zj D .z/j . For all such z differentiation
gives
X1
1
D .1/j j zj 1
.1 C z/2 j D1
1
X
D .1/j C1 .j C 1/zj ;
j D0
so that
X 1
1
D .1/j .j C 1/zj
.1 C z/ 2
j D0
for all z such that jzj < 1.

For instance, take z D 1=3. Then
X1
j C1 1 9
j
D 2 D :
j D0
3 2 4
3
Hence
X1 X1
j j C1 1 9 1 3
j
D j
j
D 2 D :
j D1
3 j D0
3 3 4 3
4
4
X1 j
z
exp.z/ D e z D
j D0
j
for all z, it follows that

1
X 1
X X zj 1
0 j zj 1 zj 1
exp .z/ D D D D e z D exp.z/;
j D1
j j D1
.j 1/ j D0
j
as we saw earlier. 4
It is possible to express the logarithm function as a power series. Indeed, we have
the following theorem.
Theorem 6.7.4. For all x 2 .1; 1/
1
X xj
log.1 C x/ D .1/j C1 :
j D1
j
Proof. Note first that the power series is absolutely convergent, by Example 3.15.3.
Its derivative is
1
X X 1 X 1
jxj 1 1
.1/j C1 D .x/j 1 D .x/j D ;
j D1
j j D1 j D0
1Cx
by Theorem 3.2.1. This expression is also the derivative of log.1 C x/, and so there
exists c such that
1
X xj
.1/j C1 D log.1 C x/ C c:
j D1
j
Substituting x D 0 gives
0 D log 1 C c D c
and the theorem follows. t

u
If x D 1, then the series above diverges since the harmonic series does so. Thus
the radius of convergence is 1. If x D 1, then the series is alternating and converges.
We would naturally expect it to converge to log 2. We will confirm this fact later.
Since j xj D jxj < 1 for all x 2 .1; 1/, we also have
1
X X1
.x/j xj
log.1 x/ D .1/j C1 D :
j D1
j j D1
j
We proceed to extend this result. For all x 2 1; 1/ let
1Cx
t .x/ D : (6.17)
1x
Note that t .x/ > 0 for all x 2 .1; 1/ and that t is continuous at all x 2 1; 1/.
Since t .1/ D 0 and
1Cx
lim D 1;
x!1 1x
the range of t must be the set of all nonnegative real numbers. Furthermore, for all
x 2 .1; 1/,
1Cx
log t .x/ D log
1x
D log.1 C x/ log.1 x/
1
X 1
X xj
xj
D .1/j C1 C
j D1
j j D1
j
1
X x 2kC1
D2 : (6.18)
2k C 1
kD0
But if we write t instead of t .x/ for convenience, Eq. (6.17) shows that
t tx D 1 C x;
whence
t 1 D x.t C 1/:
Since t > 0, it follows that
t 1
xD ;
t C1
and so we obtain the following theorem.

Theorem 6.7.5. For all t > 0
1
X 2kC1
1 t 1
log t D 2 : (6.19)
2k C 1 t C1
kD0
For instance,
1
X 1
log 2 D 2 : (6.20)
.2k C 1/32kC1
kD0
Equation (6.20) may be used to approximate log 2 to a high level of accuracy. Even
if we use only five terms of the series, we obtain 0:6931 : : :, which is correct to two
decimal places.
Theorem 6.7.5 can be used to find a quick way of approximating log t for each
positive t . First find an integer k such that
2k t < 2kC1 :
Thus
t
1 < 2;
2k
so that Theorem 6.7.4 can be used to approximate log.t =2k /. But
t
log D log t k log 2;
2k
and so log t may be calculated from the formula
t
log t D log C k log 2:
2k
This method is particularly effective for large values of t .

The logarithm function features in the definition of an important mathematical
constant. Note first that for every j > 0 the mean-value theorem yields the existence
of a 2 .j; j C 1/ such that
log.j C 1/ log j
D log0 :
j C1j
In other words,
1
log.j C 1/ log j D :

Since 0 < j < < j C 1, it follows that
1 1
< log.j C 1/ log j < (6.21)
j C1 j
for all j > 0. Thus for all n > 1 we have
X
n1
1 X
n1 X
n1
1
< .log.j C 1/ log j / < : (6.22)
j D1
j C 1 j D1 j D1
j
Now let
X n
1 X
n1
1
Sn D D
j D1
j j D0
j C1
for all n > 0. Then Eq. (6.22) yields
Sn 1 < log n < Sn1
for all n > 1, where the telescoping property was used to evaluate the middle
summation. Thus
1 1
log n C < Sn1 C D Sn < 1 C log n: (6.23)
n n
Let
an D Sn log n
for all n 2 N. We shall prove the sequence fan g convergent. First, since
1
SnC1 Sn D ;
nC1
we have
1
anC1 an D log.n C 1/ C log n;
nC1
and it follows that anC1 < an because
1
log.n C 1/ > C log n
nC1
by Eq. (6.21). Thus fan g is a decreasing sequence. It is bounded below by 0:

Inequalities (6.23) give
1
an D Sn log n > > 0:
n
Hence fan g converges to some number , which is called Eulers constant. Thus
0 1
Xn
1
D lim an D lim @ log nA :
n!1 n!1
j D1
j
From inequalities (6.23) we have
1
< Sn log n < 1;
n
so that 0 1. In fact, is approximately 0:577.
We can use these ideas to obtain a convenient power series expansion for log 2.
Define
"n D Sn log n
for all n > 0. Then "n ! 0 as n ! 1, and
Sn D log n C C "n :
Now we put
X
n
.1/j C1
Tn D
j D1
j
for all n > 0. Thus
X
2n
.1/j C1
T2n D
j D1
j
X2n
1 X n
1
D 2
j D1
j j D1
2j
D S2n Sn
D log 2n C C "2n log n "n
D log 2 C "2n "n
! log 2
as n ! 1, so that
1
X .1/j C1
log 2 D :
j D1
j
We have now extended Theorem 6.7.4 to the case x D 1. However, this result does
not provide an efficient method for approximating log 2. For instance, by taking 20
terms of the series we obtain the approximation 0:6669, rounded to four decimal
places. This result is much worse than that reached by taking only five terms of the
series given in Eq. (6.20).
The next theorem gives a power series expansion, due to Gregory, for arctan x
that is valid for all x such that jxj < 1.
Theorem 6.7.6. For all x such that jxj < 1,
1
X x 2j C1
arctan x D .1/j :
j D0
2j C 1
Proof. It follows from Example 3.15.3 that the power series is absolutely conver-
gent. Its derivative is
1
X 1
X
.1/j x 2j D .x 2 /j
j D0 j D0
1
D
1 C x2
D arctan0 x
for all x such that jxj < 1. By Corollary 6.3.4 there is a constant c such that
1
X x 2j C1
.1/j D arctan x C c
j D0
2j C 1
for all such x. Since arctan 0 D 0, we have c D 0 and the theorem follows. t
u
We finish this section with an example of an application of the differentiation of
power series to the solution of a certain type of differential equation. Such equations
arise, for instance, in wave mechanics.
Example 6.7.3. The differential equation
y 00 2xy0 C 2y D 0; (6.24)
where 0, is known as Hermites equation of order . We aim to find a power

series solution.
Let
1
X
y.x/ D aj x j
j D0
be a solution. From Eq. (6.24) we obtain

1
X 1
X 1
X
j 2 j 1
0D j.j 1/aj x 2x jaj x C 2 aj x j
j D2 j D1 j D0
1
X 1
X 1
X
D .j C 2/.j C 1/aj C2 x j 2 jaj x j C 2 aj x j
j D0 j D0 j D0
1
X
D ..j C 2/.j C 1/aj C2 2jaj C 2aj /x j
j D0
for all x for which the series converges. Therefore
.j C 2/.j C 1/aj C2 C 2. j /aj D 0
for each j , so that
2. j /
aj C2 D aj :
.j C 2/.j C 1/
Note that y.0/ D a0 . Similarly, since

1
X
y 0 .x/ D jaj x j 1 ;
j D1
we have y 0 .0/ D a1 . These two equations are called the initial conditions for the
differential equation. The series for y is uniquely determined if a0 and a1 are known.
In fact, an inductive argument shows that
2j . 2/. 4/ . 2j C 2/
a2j D .1/j a0
.2j /
and
2j . 1/. 3/. 5/ . 2j C 1/
a2j C1 D .1/j a1
.2j C 1/
for all j 0. Recalling the convention that the empty product is 1, we therefore
obtain
1 Qj 1 1 Qj 1
X X
2j kD0 . 2k/ 2j 2j kD0 . 2k 1/
y.x/ D a0 .1/ j
x C a1 .1/ j
x 2j C1 :
.2j / .2j C 1/
j D0 j D0
In the case where is a positive integer, the first series is a polynomial if is

even and the second series is a polynomial if is odd. The solution is therefore a
polynomial of degree if either is even and .a0 ; a1 / D .1; 0/ or is odd and
.a0 ; a1 / D .0; 1/. In either case it is called the Hermite polynomial of degree
and is denoted by H . For example, for all x we have
H1 .x/ D x;
H2 .x/ D 1 2x 2 ;
2x 3
H3 .x/ D x ;
3
4x 4
H4 .x/ D 1 4x 2 C ;
3
4x 3 4x 5
H5 .x/ D x C ;
3 15
8x 6
H6 .x/ D 1 6x 2 C 4x 4 :
15
There are other ways of defining the Hermite polynomials. For instance, it can
be shown [16] that they satisfy the recursive relation
HnC1 .x/ D 2xH n .x/ 2nH n1 .x/
for all n > 1 and all x.

It is not hard to see that the solution converges absolutely for all real x. This
fact also follows from the following beautiful result, which is due to Fuchs [13].
Consider the differential equation
y 00 C p.x/y 0 C q.x/y D 0;
where y.0/ and y 0 .0/ are given. Let r > 0. If p and q can be represented by
power series on the interval .r; r/, then the differential equation has a unique
power series solution and this solution also converges on .r; r/. In other words,
the radius of convergence of the solution is at least as big as the minimum of the
radii of convergence of p and q. 4
Exercises 6.7.
1. Verify the formulas for the derivatives of the sine and cosine functions by
differentiating their power series.
2. Let
X1
z3j
f .z/ D :
j D0
.3j /
(a) Find the radius of convergence of the series.

(b) Show that
f .z/ C f 0 .z/ C f 00 .z/ D e z
for all z within the circle of convergence.

3. Suppose that the radius of convergence of
1
X
f .z/ D aj zj
j D0
is r > 0. Show that

1
X .n C j /
f .n/ .z/ D anCj zj
j D0
n
for all nonnegative integers n and all z such that jzj < r.
4. Suppose that
1
X
f .z/ D 1 C aj zj
j D1
is the series solution of the differential equation f 0 .z/ D f .z/. Show that
an
anC1 D
nC1
for all n 2 N and hence Pthat f .z/ D e z for all z.

1
5. Suppose that f .x/ D j D0 aj x j is the series solution of f 0 .x/ D 1 C x 2 and
that f .0/ D 0. Show that
P f .x/ Dj tan x for all x for which cos x 0.
6. Suppose that f .x/ D 1 a
j D0 j z is the series solution of f 00
.z/ D f .z/ and
that f .0/ D 0 and f 0 .0/
P D 1. Show that f .z/ D sin z for all z.
7. Suppose that f .x/ D 1 00
j D0 aj z is the series solution of f .z/ D f .z/ and
j
0
that f .0/ D 1 and f .0/ D 0. Show that f .z/ D cos z for all z.
8. Is the solution of Hermites equation satisfying the following conditions a
polynomial:
D 3; a0 D 1; a1 D 0
P1
9. (Airys equation) Suppose that f .x/ D j D0 aj x j is the series solution of
f 00 .x/ xf .x/ D 0:
First, show that a2 D 0 and

an1
anC2 D
.n C 1/.n C 2/
for all n 2 N. Hence for all k 2 N, show that

a0
a3k D ;
2 3 5 6 .3k 1/3k
a1
a3kC1 D ;
3 4 6 7 3k.3k C 1/
a3kC2 D 0:
P1
10. (Legendres equation) Suppose that j D0 aj x j is the series solution of
.1 x 2 /f 00 .x/ 2xf 0 .x/ C . C 1/f .x/ D 0:

Substitute the power series for f .x/ into the differential equation to obtain a
series p.x/. Show that the constant term of the latter series satisfies the equation
2a2 C . C 1/a0 D 0;
that the coefficient of x satisfies
6a3 C .2 C . C 1//a1 D 0;
and that, for all n > 1, the coefficient of x n satisfies
.n C 2/.n C 1/anC2 C .n.n 1/ 2n C . C 1//an D 0:
Writing
p.x/ D a0 p1 .x/ C a1 p2 .x/;
show that the series p.x/ converges absolutely whenever jxj < 1.
When n is a nonnegative integer, either p1 or p2 is a polynomial. Show that
if n D 0, then p1 .x/ D 1 and
1 1Cx
p2 .x/ D log
2 1x
and that if n D 1, then p2 .x/ D x and
1 1Cx
p1 .x/ D 1 log :
2 1x
11. Use the fact that

0 1
Xn
1
lim @ log nA D
n!1
j D1
j
to show that
1 X1
n
lim D 1:
n!1 log n j
j D1
Derive this result also from Stolzs theorem.

Chapter 7
The Riemann Integral
In this chapter we use the idea of an area to motivate a concept called an integral of a
function, and we show that the process of finding an integral of a function is closely
related to that of obtaining an antiderivative of the function, that is, a function whose
derivative is the given function. Some techniques for finding integrals are derived
and the use of integrals for testing the convergence of certain types of series are
discussed. All functions under consideration are assumed to be real-valued functions
of one or more real variables.
7.1 Area Under a Curve
Suppose we wish to approximate the area bounded by the curve y D f .x/ and the
lines x D a, x D b, and y D 0 (see Fig. 7.1. For the sake of clarity, the figure is
drawn for a function f such that f .x/ 0 for all x 2 a; b.)
We approximate the area using small rectangular regions. Let us first introduce a
few definitions and notation.
Definition 7.1.1. Let a; b be a closed interval. By a partition of a; b we mean a
(finite) sequence x0 ; x1 ; : : : ; xn of numbers such that
a D x0 < x1 < : : : < xn D b:
(More formally, x0 D a, xn D b, and xj C1 > xj for all j < n.)

This partition P is denoted by .x0 ; x1 ; : : : ; xn /. The numbers x0 ; x1 ; : : : ; xn are
called the division points of P . The set of division points of P is denoted by PO .
We say that PO determines P . If .x0 ; x1 ; : : : ; xn / is a partition P of a; b, we define

334 7 The Riemann Integral
y = f (x)
x
a b
Fig. 7.1 Area under the graph of f between a and b
the norm or mesh of P to be the maximum of the quantities xj C1 xj , where

0 j < n. This number is denoted by .P /. Each interval xj ; xj C1 is called
a subinterval of P , and the elements of .xj ; xj C1 / are called interior points of
xj ; xj C1 .
Recall that a function f mapping a; b into R is bounded if there exist m and M
such that m f .x/ M for all x 2 a; b. If .x0 ; x1 ; : : : ; xn / is a partition of a; b
and f is bounded on a; b, then f is also bounded on each subinterval xj ; xj C1
of a; b, where j < n. For each j < n we set
Mj .f / D supff .x/ j x 2 xj ; xj C1 g
and
mj .f / D infff .x/ j x 2 xj ; xj C1 g:
If f and g are functions that are defined and bounded on a; b and f .x/ g.x/
for all x 2 a; b, then it is clear that Mj .f / Mj .g/ and mj .f / mj .g/ for
each j . For the rest of this chapter, unless we state otherwise, we assume that f
is a function that is defined and bounded on a closed interval a; b. Given such a
function f and a partition .x0 ; x1 ; : : : ; xn / of a; b, we shall also take Mj .f / and
mj .f / to be defined as above.
Remark. There might not exist c 2 xj ; xj C1 such that f .c/ D Mj .f /, and
similarly for mj .f /. For example, let f W 0; 2 ! R be defined by
8
< x bxc if x 2 0; 1
f .x/ D
:
bxc x if x 2 .1; 2
7.1 Area Under a Curve 335
Fig. 7.2 Graph of x bxc f (x)

for x 2 0; 1 and bxc x for
x 2 .1; 2
1
x
1 2
(see Fig. 7.2). Let P be the partition .0; 1; 2/ of 0; 2. Then M0 .f / D 1 and
m1 .f / D 1. It is clear that there is no c 2 0; 1 such that f .c/ D 1 and no
c 2 1; 2 such that f .c/ D 1. However, if f is continuous on a; b, then there is
always a number c 2 xj ; xj C1 such that f .c/ D Mj .f / (see Corollary 5.3.2).
A corresponding statement holds for mj .f /. The number c also exists if f is
nondecreasing or nonincreasing on a; b. For instance, if f is nondecreasing on
a; b, then Mj .f / D f .xj C1 / and mj .f / D f .xj /.
Let .x0 ; x1 ; : : : ; xn / be a partition P of some closed interval J a; b. Then we
define
X
n1
U.P; f / D Mj .f /.xj C1 xj /:
j D0
The number U.P; f / is called the upper (Riemann) sum of f over J relative to P .
An upper sum of f over the whole of a; b can be thought of geometrically as the
shaded area in Fig. 7.3. This sum is an approximation, from above, for the area A
bounded by the curve y D f .x/ and the lines x D a, x D b, and y D 0. It is clear
that if c 2 .a; b/ and P1 and P2 are partitions of a; c and c; b, respectively, then
U.P; f / D U.P1 ; f / C U.P2 ; f /;
where P is the partition of a; b for which PO D PO1 [ PO2 .

The sum
X
n1
L.P; f / D mj .f /.xj C1 xj /
j D0
y f(x)
x
x0 x1 x2 xn
a b
Fig. 7.3 Illustration of an upper Riemann sum
y f(x)
x0 x1 x2 xn x
a b
Fig. 7.4 Illustration of a lower Riemann sum
is called the lower (Riemann) sum of f over J relative to the partition P of J . If

J D a; b, then this lower sum gives the shaded area in Fig. 7.4 and constitutes an
approximation for A from below. Note also that
L.P; f / D L.P1 ; f / C L.P2 ; f /;
where P1 and P2 are partitions of a; c and c; b, respectively, for some c 2 .a; b/,
and P is the partition of a; b for which PO D PO1 [ PO2 .
We also have
X
n1 X
n1
U.P; f / L.P; f / D Mj .f /.xj C1 xj / mj .f /.xj C1 xj /
j D0 j D0
X
n1
D .Mj .f / mj .f //.xj C1 xj /:
j D0
In addition, if f and g are functions that are defined and bounded on J and satisfy
f .x/ g.x/ for all x 2 J , then L.P; f / L.P; g/ and U.P; f / U.P; g/.
Example 7.1.1. If P is the partition .a; b/ of a; b determined by fa; bg, then
L.P; f / D m.b a/
and
U.P; f / D M.b a/;
where M D supff .x/ j x 2 a; bg and m D infff .x/ j x 2 a; bg. 4

The sum
X
n1
S.P; f / D f .cj /.xj C1 xj /; (7.1)
j D0
where cj 2 xj ; xj C1 for all j < n, is called a Riemann sum of f over J

relative to P and the intermediate points c0 ; c1 ; : : : ; cn1 . It gives the shaded area in
Fig. 7.5 if J D a; b. Observe that the notation does not indicate the dependence
of the Riemann sum S.P; f / on the intermediate points. Whenever we need to
make this dependence explicit, then we write S.P; f; c/ instead of S.P; f /, where
c D .c0 ; c1 ; : : : ; cn1 /.
Example 7.1.2. If f .x/ D k for all x 2 a; b, then
X
n1
S.P; f / D k.xj C1 xj /
j D0
D k.b a/;
by the telescoping property. In this case we also have
U.P; f / D L.P; f / D S.P; f /:
4
y = f(x)
x
x0 c0 x1 c1 x2 c2 cn 1 xn
a b
Fig. 7.5 Illustration of a Riemann sum
If f and g are functions defined and bounded on J , and k and l are constants,
then
X
n1
S.P; kf C lg/ D .kf .cj / C lg.cj //.xj 1 xj /
j D0
X
n1 X
n1
Dk f .cj /.xj C1 xj / C l g.cj /.xj C1 xj /
j D0 j D0
D kS.P; f / C lS.P; g/:
It is also clear that
L.P; f / S.P; f / U.P; f /: (7.2)
Intuitively, as the partition becomes finer, we expect the Riemann sum to

provide a better approximation for the required area. However, it is not always the
case that if .P1 / .P2 /; we necessarily obtain a better approximation for the
area by using P1 rather than P2 .
Definition 7.1.2. We say that a partition Q of a; b is a refinement of a partition
P if PO Q;
O that is, every division point of P is a division point of Q.
Let P and Q be partitions of a; b. We define P [ Q to be the partition
determined by the set PO [ QO of division points.
It is clear that if P and Q are partitions of a; b, then P [ Q is a refinement of
both P and Q and that if Q is a refinement of P , then .Q/ .P /.
Theorem 7.1.1. Suppose P and Q are partitions of a; b and that Q is a

refinement of P . Then for each function f defined and bounded on a; b we have
L.P; f / L.Q; f /
and
U.P; f / U.Q; f /:
Proof. To establish the first inequality, it suffices to prove the result for the case
where Q contains exactly one more point than P . An easy induction then gives us
the general case.
Suppose QO D PO [ fyg, where P D .x0 ; x1 ; : : : ; xn / and xj < y < xj C1 for
some integer j < n. Let
m01 D infff .x/ j x 2 xj ; yg
and
m02 D infff .x/ j x 2 y; xj C1 g:
Clearly, m01 mj .f / and m02 mj .f /. Thus
L.Q; f / L.P; f / D m01 .yxj /Cm02 .xj C1 y/mj .f /.xj C1 xj /

mj .f /.yxj /Cmj .f /.xj C1 y/mj .f /.xj C1 xj /
D 0:
The proof of the second inequality is similar. t

u
Corollary 7.1.2. Let P and Q be any two partitions of a; b. Then L.P; f /
U.Q; f /.
Proof. Since P [ Q is a refinement of both P and Q, Theorem 7.1.1 implies that
L.P; f / L.P [ Q; f / U.P [ Q; f / U.Q; f /:
t
u
Exercises 7.1.
1. Prove that, for any partitions P and Q of an interval a; b,
m.b a/ L.P; f / U.Q; f / M.b a/;
where M D supff .x/ j x 2 a; bg and m D infff .x/ j x 2 a; bg.

2. Consider the function f defined by f .x/ D x for all x 2 0; 1. For each positive
integer n, the partition

1 2
Pn D 0; ; ; : : : ; 1
n n
divides the interval 0; 1 into n subintervals of length 1=n. Find
lim U.Pn ; f /
n!1
and
lim L.Pn ; f /:
n!1
7.2 Upper and Lower Integrals
The collections of upper and lower sums of f over a; b are nonempty and bounded
below and above, respectively. Hence we can make the following definition.
R
Definition 7.2.1. The lower integral f of a bounded function f over a; b is
the least upper bound of the set of lower sums of f relative to partitions of a; b.
R
Similarly, the upper integral f of f over a; b is the greatest lower bound of the
set of upper sums of f relative to partitions of a; b.
From Corollary 7.1.2 it is clear that
Z Z
L.P; f / f f U.P; f / (7.3)
for every partition P of a; b. Moreover, if f and g are bounded functions such that
R R R R
f .x/ g.x/ for all x 2 a; b, then f g and f g. It is also clear from
inequalities (7.2) that
Z Z
f S.P; f / f
for every Riemann sum S.P; f / of f over a; b relative to P .

Example 7.2.1. Let f W 0; 1 ! R be defined by

f .x/ D
1 if x is rational.
7.3 The Riemann Integral 341
For each partition P D .x0 ; x1 ; : : : ; xn / of 0; 1 we have Mj .f / D 1 and

mj .f /D0 for all j < n. Hence
X
n1
U.P; f / D .xj C1 xj /
j D0
D xn x 0
D1
and
L.P; f / D 0:
R R
Thus f D 0 and f D 1. 4
7.3 The Riemann Integral
In this section we define the (Riemann) integral of a function and show that when
the integral exists, it coincides with both the upper and lower integrals.
Definition 7.3.1. Let f be a bounded function from a; b into R. We say that the
Riemann sums of f over a; b converge to a number I if for every " > 0 there is a
> 0 such that if P is a partition of a; b with norm .P / < , then
jS.P; f / I j < "
for every Riemann sum S.P; f / of f over a; b relative to P .
Remark 1. As in Proposition 2.2.3, convergence will follow if this inequality can

be established with " replaced by c" for some c > 0.
R f / < IC
Remark 2. If the Riemann sums of f over a; b converge to I , then S.P;
" for every Riemann sum S.P; f / of f over a; b relative to P . Thus f < I C ",
R R
and since " is arbitrary it follows that f I . Similarly, I f .
Example 7.3.1. If f .x/ D k for each x 2 a; b, then the Riemann sums of f over
a; b converge to k.b a/, the value of each such Riemann sum (cf. Example 7.1.2).
4
The proof of the following theorem is analogous to that of Theorem 2.2.2 and
therefore omitted.
Theorem 7.3.1. Let f be a bounded function on a; b whose Riemann sums
converge to I1 and I2 . Then I1 D I2 .
Definition 7.3.2. Let f be a function that is bounded on a; b. If the Riemann sums
converge to some number I , then we say that f is (Riemann) integrable over a; b.
We call I the (Riemann) integral of f over a; b, and write
Z b
f .x/ dx D I;
a
Rb R
or simply a f D I or f D I . The function f is called the integrand of I .
For instance, it follows from Example 7.3.1 that
Z b
k dx D k.b a/:
a
In particular,
Z b
0 dx D 0:
a
Rbf .x/
For typographical convenience, an integral of the form a g.x/ dx is sometimes
R b .x/ dx
written as a f g.x/ .
Theorem 7.3.2. Let f W a; b ! R be integrable. Then every sequence fS.Pn ; f /g
Rof Riemann sums for f over a; b satisfying limn!1 .Pn / D 0 must converge to
f.
R
Proof. Let I D f . For each " > 0, there exists > 0 such that
jS.P; f / I j < "
for every partition P of a; b with .P / < . For every sequence fS.Pn ; f /g of

Riemann sums for f over a; b satisfying limn!1 .Pn / D 0, there is an N such
that .Pn / < whenever n N . Therefore jS.Pn ; f / I j < " for each n N ,
and so
lim S.Pn ; f / D I: t
u
n!1
The converse of this theorem is given as an exercise.

Definition 7.3.3. We say that U.P; f / L.P; f / converges to 0 over a closed
interval a; b if for every " > 0 there is a > 0 such that if P is any partition of
a; b satisfying .P / < , then
U.P; f / L.P; f / < ":

Remark 1. Again, to establish convergence it suffices to prove the preceding

inequality with " replaced by c" for some c > 0.
Remark 2. Suppose that U.P; f / L.P; f / converges to 0 over a; b. By
inequalities (7.3) we see that
Z Z
U.P; f / L.P; f / f f
R R
for every partition P of a; b. However, if f > f , then
Z Z
f f > 0;
and so we may find > 0 such that

Z Z
U.P; f / L.P; f / < f f
for every partition P of a; b with norm less than . This contradiction shows that
R R
if U.P; f / L.P; f / converges to 0, then f D f .
Theorem 7.3.3. Let f be a function that is bounded on a; b. Then the following
two statements are equivalent:
1. U.P; f / L.P; f / converges to 0 over a; b;
2. f is integrable over a; b.
Proof. Suppose (1) holds. Given " > 0, there exists > 0 such that
U.P; f / L.P; f / < "
for every partition P of a; b with .P / < . Now

Z ! Z Z ! Z !
U.P; f /L.P; f / D U.P; f / f C f f C f L.P; f / :
Since each term in parentheses is nonnegative,

Z Z
f f U.P; f / L.P; f / < "
R R R
whenever .P / < . As " is arbitrary, f D f . Let I D f . It follows that
L.P; f / I U.P; f /, and as we also have L.P; f / S.P; f / U.P; f / for
each Riemann sum S.P; f /, we deduce that
jS.P; f / I j U.P; f / L.P; f / < ";
as required.
R
Conversely, suppose that (2) holds, and let I D f . Then given any " > 0, there
is a > 0 such that
I " < S.P; f / < I C "
for each partition P of a; b with norm less than and each Riemann sum S.P; f /
of f over a; b relative to P . Fix such a P D .x0 ; x1 ; : : : ; xn / and choose cj 2
xj ; xj C1 for each j < n. Since Mj .f / " is less than the least upper bound of f
on xj ; xj C1 , we may choose each cj so that f .cj / > Mj .f / ". Therefore
X
n1
U.P; f / D Mj .f /.xj C1 xj /
j D0
X
n1 X
n1
< f .cj /.xj C1 xj / C " .xj C1 xj /
j D0 j D0
D S.P; f / C ".b a/
 I ".b a C 1/;
and so
U.P; f / L.P; f / < 2".b a C 1/
whenever .P / < . Therefore U.P; f /L.P; f / converges to 0, as required. t

u
From Remarks 2 after Definitions 7.3.1 and 7.3.3 we obtain the following
corollary.
R R R
Corollary 7.3.4. If f is integrable over a; b, then f D f D f .
The proof of the next result is adapted from [7].
Lemma 7.3.5. Let f W a; b ! R be a bounded function. Then for every " > 0
there is a > 0 such that
Z
U.P; f / < f C "
and
Z
L.P; f / > f "
for every partition P of a; b for which .P / < .

Proof. The lemma is certainly true if f .x/ D 0 for all x 2 a; b, for in that case we
R R
have U.P; f / D f D L.P; f / D f D 0. We therefore assume that f .x/ 0
for some x 2 a; b.
Choose " > 0. By the definition of the upper integral there exists a partition P1
such that
Z
"
U.P1 ; f / < f C :
2
Suppose that P1 contains k interior points. Theorem 7.1.1 shows that P1 may be
refined if necessary so that k > 0. Let be the minimum length of any subinterval
of P1 . In view of Theorem 7.1.1, we may also assume that
"
< ;
6Mk
where M D supx2a;b jf .x/j > 0.
Now let P be any partition with .P / < , and let Q D P [ P1 . Then
Theorem 7.1.1 shows that
Z
"
U.Q; f / U.P1 ; f / < f C : (7.4)
2
We wish to obtain an upper bound for U.P; f / U.Q; f /. By the choice of
, there is at most one point of P1 in each subinterval of P . Let S be the set of
subintervals J of P such that some interior point of J belongs to P1 . These are
precisely the subintervals of P that are not subintervals of Q. Rather, each such
subinterval of P is the union of just two subintervals of Q. The contributions to
U.P; f / and to U.Q; f / of each subinterval of P not in S are equal and therefore
cancel in the expression U.P; f / U.Q; f /. Now choose an interval J D r; t in
S and let s be its unique interior point in P1 . Thus J1 D r; s and J2 D s; t are
subintervals of Q. The contribution of J to jU.P; f /j is no greater than M.t r/ <
M . Similarly, the contribution of each of J1 and J2 to jU.Q; f /j is less than M .
Therefore the sum of these contributions to jU.P; f /j and jU.Q; f /j is less than
3M . Moreover since P1 has just k interior points, we see that jS j k. Using the
triangle inequality, we therefore conclude that
"
U.P; f / U.Q; f / jU.P; f / U.Q; f /j 3kM < : (7.5)
2
Combining inequalities (7.4) and (7.5), we obtain
Z
U.P; f / D U.Q; f / C U.P; f / U.Q; f / < f C ";
and the first of the required inequalities is proved. The proof of the second is similar.
t
u
R R
Theorem 7.3.6. A bounded function f is integrable if and only if f D f .
Proof. We have already observed the necessity of the equation. Suppose therefore
R R
that f D f D I , and choose " > 0. By Lemma 7.3.5 there exists > 0 such
that
I " < L.P; f / U.P; f / < I C "
for every partition P for which .P / < . Hence
U.P; f / L.P; f / < " C " D 2";
so that f is integrable by Theorem 7.3.3. t

u
Corollary 7.3.7. A bounded function defined on a; b is integrable if and only if
for every " > 0 there is a partition P of a; b such that
U.P; f / L.P; f / < ": (7.6)
Proof. The necessity is clear from Theorem 7.3.3.

Given a partition P of a; b satisfying Eq. (7.6), we have
Z Z
L.P; f / f f U.P; f /:
Hence
Z Z
f f U.P; f / L.P; f / < ":
R R R R
Since " is arbitrary, it follows that f f . Therefore f D f , and so the
result is a consequence of Theorem 7.3.6. u
t
The next corollary is obtained by translating the previous one into terms of
sequences and is often useful.
Corollary 7.3.8. Let f W a; b ! R be a bounded function. Then f is integrable if
and only if there exists a sequence fPn g of partitions of a; b such that
lim .U.Pn ; f / L.Pn ; f // D 0:

n!1
If f is integrable, then
Z
f D lim U.Pn ; f / D lim L.Pn ; f /:
n!1 n!1
Theorem 7.3.9. If a function f is integrable over closed intervals a; c and c; b,

then f is integrable over a; b.
Proof. We must show that U.P; f / L.P; f / converges to 0 over a; b. Choose
" > 0. Since f is integrable on a; c, there exists 1 > 0 such that
U.P1 ; f / L.P1 ; f / < "
for every partition P1 of a; c satisfying .P1 / < 1 . Similarly, there exists 2 > 0
such that
U.P2 ; f / L.P2 ; f / < "
for every partition P2 of c; b satisfying .P2 / < 2 . Let D minf1 ; 2 g, and

choose a partition P of a; b such that .P / < . We may refine it if necessary to
ensure that c 2 PO . Then PO \ a; c is the set of division points of a partition P1 of
a; c with norm less than 1 . Similarly, PO \ c; b is the set of division points of a
partition P2 of c; b with norm less than 2 . Moreover
U.P; f / D U.P1 ; f / C U.P2 ; f /
and
L.P; f / D L.P1 ; f / C L.P2 ; f /I
hence
U.P; f / L.P; f / D U.P1 ; f / L.P1 ; f / C U.P2 ; f / L.P2 ; f / < " C " D 2";
as required. t
u
By means of an easy inductive argument, we obtain the following corollary.
Corollary 7.3.10. Let .x0 ; x1 ; : : : ; xn / be a partition of a closed interval a; b. Let
f be a function defined on a; b, and suppose that f is integrable over xj ; xj C1
for each j < n. Then f is integrable over a; b.
We shall show that if f is continuous on a; b, then it is integrable over a; b.
In fact, we prove the following more general theorem.
Theorem 7.3.11. If f is continuous on .a; b/ and bounded on a; b, then f is
integrable over a; b.
Proof. We suppose first that f is continuous at a (but not necessarily at b). Choose
" such that 0 < " a. Then f is continuous on a; c,
and Theorem 5.5.2 implies the existence of > 0 such that
jf .x/ f .y/j < "
whenever jx yj < and x; y 2 a; c.

Take any partition P D .x0 ; x1 ; : : : ; xn / of a; b with norm .P / < , and refine

it if necessary so that c D xr for some r such that 0 < r < n. By Theorem 7.3.3 it
suffices to show that
U.P; f / L.P; f / < k"
for some k > 0. Let P1 and P2 be the partitions of a; c and c; b, respectively,
such that Pc1 and P c2 are the intersections of PO with a; c and c; b, respectively.
The function f is continuous on xj ; xj C1 for each j such that 0 j < r, so
that Mj .f / D f .cj / for some cj 2 xj ; xj C1 and mj .f / D f .dj / for some
dj 2 xj ; xj C1 . Thus
X
r1
U.P1 ; f / L.P1 ; f / D .Mj .f / mj .f //.xj C1 xj /
j D0
X
r1
D .f .cj / f .dj //.xj C1 xj /
j D0
X
r1
<" .xj C1 xj /
j D0
D ".c a/:
Since f is bounded on c; b, there exists M such that jf .x/j < M for all x 2
c; b. Note also that
Mj .f / mj .f / D jMj .f / mj .f /j jMj .f /j C jmj .f /j 2M
for each j such that r j < n, by the triangle inequality. It follows that
X
n1
U.P2 ; f / L.P2 ; f / D .Mj .f / mj .f //.xj C1 xj /
j Dr
2M.b c/
D 2M ":
We therefore deduce that
U.P; f / L.P; f / D U.P1 ; f / L.P1 ; f / C U.P2 ; f / L.P2 ; f /

< .c a/" C 2M "
D .c a C 2M /";
as required.
We have now proved the theorem in the case where f is continuous at a.

Similarly, the theorem holds if f is continuous at b. The remaining case is dealt with
by noting that f is continuous at c and applying these results and Theorem 7.3.9 to
the intervals a; c and c; b. t
u
Corollary 7.3.12. Let f be a function that is continuous over an interval a; b
except at a finite number of points. If f is bounded on a; b, then f is integrable
over a; b.
Proof. Let P be a partition .x0 ; x1 ; : : : ; xn / of a; b such that PO contains all
points in a; b where f is not continuous. By Theorem 7.3.11, f is integrable on
xj ; xj C1 for each j < n. Now apply Corollary 7.3.10. t
u
By applying Theorems 5.3.1 and 7.3.11, we also obtain the following result.
Corollary 7.3.13. If a function f is continuous over an interval a; b, then f is
Lebesgue showed that a function may be discontinuous at infinitely many points
yet be integrable.
Example 7.3.2. Consider the function f W 0; 1 ! R defined by
(
x x1 if x 0;
f .x/ D
0 if x D 0:
Thus for each

1 1
x2 ; ;
nC1 n
where n 2 N, we have b1=xc D n. In this case it follows that f .x/ D nx, so that
n
< f .x/ 1:
nC1
Thus f is bounded on .1=.n C 1/; 1=n for all n 2 N. Consequently it is bounded

on 1=n; 1 for all n 2 N. Moreover
n
lim
C
f .x/ D < 1;
x! 1 nC1
nC1
whereas

1 1
f D bn C 1c D 1:
nC1 nC1
Hence f is discontinuous at 1=.n C 1/ for each n 2 N.

Choose " > 0 and N 2 N so large that 1=N < ". By Corollary 7.3.12, f
is integrable on 1=N; 1. Let P D .x0 ; x1 ; : : : ; xn / be a partition of 0; 1. By
Theorem 7.3.3, we may refine P so that xk D 1=N , for some k < n, and
X
n1
.Mj .f / mj .f //.xj C1 xj / < ":
j Dk
Now for each x < 1=N we have x 2 .1=.M C 1/; 1=M for some M 2 N such that
M N , so that
M N
1 f .x/ >
M C1 N C1
since the sequence fn=.n C 1/g is increasing. [This observation is easily checked
for all n > 0 by dividing both sides of the inequality
n2 > n2 1 D .n 1/.n C 1/
by n.n C 1/.] As
N N 1 1
> D1 ;
N C1 N N
it follows that
X
k1 k1
X
1
.Mj .f / mj .f //.xj C1 xj / < 1 1 .xj C1 xj /
j D0 j D0
N
1
D .xk x0 /
N
1
D 2
N
< "2 :
We conclude that
U.P; f / L.P; f / < "2 C " D "." C 1/I
therefore f is integrable over 0; 1. 4

Two more important classes of integrable functions are the increasing and
decreasing bounded functions.
Theorem 7.3.14. A bounded function that is increasing on an interval a; b or

decreasing on a; b is also integrable over a; b.
Proof. Let f be a bounded function that is increasing on a; b. Given " > 0, let P
be a partition .x0 ; x1 ; : : : ; xn / of a; b with .P / < ". Then
X
n1
U.P; f / L.P; f / D .Mj .f / mj .f //.xj C1 xj /
j D0
X
n1
<" .f .xj C1 / f .xj //
j D0
D ".f .b/ f .a//:
Moreover f .b/ > f .a/ since f is increasing on a; b. Thus U.P; f / L.P; f /
converges to 0, and so f is integrable over a; b.
The argument is similar if f is decreasing on a; b. t
u
Exercises 7.2.
1. Let f W a; b ! R. Prove that the following statements are equivalent:
(a) f is integrable over a; b.
(b) For each " > 0 there exists > 0 such that if P and Q are partitions of a; b
with norm less than , then
jS.P; f / S.Q; f /j < "
for all Riemann sums S.P; f / and S.Q; f / of f over a; b relative to P

and Q, respectively.
This result is known as the Cauchy criterion for integrability.
2. Let f W a; b ! R. Suppose that for each " > 0 there exist integrable functions
g and h from a; b to R such that
Z
.g h/ < "
and
g.x/ f .x/ h.x/
for all x 2 a; b. Show that f is integrable.

This result is called the sandwich theorem for integrals. In particular it shows
that if there are integrable functions g andR h mapping
R a; b into R such that
g.x/R f .x/ h.x/ for all x 2 a; b and g D h D I , then f is integrable
and f D I .
3. Let
nxn1
fn .x/ D
1Cx
R1
for all n > 1 and x 1. Define An D 0 fn for all n > 1, and show that
1
lim An D :
n!1 2
[Hint: nxn1 =2 fn .x/ nxn2 =2 for all n > 2.]

4. Let f W 0; 1 ! R be defined by
(
1 if x D 1
for some n 2 N;
f .x/ D n
0 otherwise:
R
Show that f is integrable and f D 0. [Hint: Choose " > 0 and an integer
m > 1=". Let

1
0; ; x2 ; x3 ; : : : ; x2m1 ; 1
m
be a partition P of 0; 1 satisfying the inequalities
1
x2k > > x2k1
mkC1
and
"
x2k x2k1 <
m
for each integer k such that 1 < k < m. Compute U.P; f / L.P; f /.]
5. Let f W a; b ! R be a function. Suppose that any sequence fS.Pn ; f /g of
Riemann sums of f over a; b is convergent if limn!1 .Pn / D 0. Prove that
f is integrable.
This result is the converse of Theorem 7.3.2 and therefore completes a
sequential characterization of integrability.
7.4 Basic Properties of Integrals 353
7.4 Basic Properties of Integrals
The next theorem summarizes the fundamental properties of integrals. We prepare

for it with the following lemma.
Lemma 7.4.1. Let f be a real-valued function defined on an interval a; b, and let
M D sup Rf and m D inf Rf . Define
g.x; y/ D f .y/ f .x/
for all .x; y/ 2 Df2 . Then
sup Rg D M m:
Proof. For each x; y 2 a; b we have f .y/ M and f .x/ m. Hence
g.x; y/ M m;
so that M m is an upper bound of Rg .

Now choose " > 0. We must show that M m " is not an upper bound of Rg .
Since M D sup Rf , there exists d 2 a; b such that
"
f .d / > M :
2
Similarly there exists c 2 a; b such that

"
f .c/ < m C :
2
Hence
g.c; d / D f .d / f .c/ > M m ";
as required. t
u
Theorem 7.4.2. Let f; g be functions and a; b an interval.
1. If f is integrable over a; b, then for every constant k, the function kf is also
integrable over a; b and
Z b Z b
kf D k f:
a a
2. If f and g are integrable over a; b, then so is f C g and

Z b Z b Z b
.f C g/ D f C g:
a a a
3. If f is integrable over a; b and c 2 .a; b/, then f is integrable over a; c and

c; b and
Z b Z c Z b
f D f C f:
a a c
4. If f and g are integrable over a; b, then so is fg.

Proof. 1. The result is clear if k D 0. Suppose therefore that k 0.
Let .x0 ; x1 ; : : : ; xn / be a partition P of a; b. Choose cj 2 xj ; xj C1 for each
j < n, and let c D .c0 ; c1 ; : : : ; cn1 /. It is immediate from Eq. (7.1) that
S.P; kf ; c/ D kS.P; f; c/:
The desired result now follows from the observation that if
jS.P; f; c/ I j < ";
then
jS.P; kf ; c/ kIj D jkjjS.P; f; c/ I j < jkj":
Rb Rb
2. Let I1 D a f and I2 D a g. Choose " > 0. There is a 1 > 0 such that if P is
a partition of a; b with norm less than 1 , then
jS.P; f / I1 j < "
for every Riemann sum S.P; f / of f over a; b relative to P . Similarly, there

is a 2 > 0 such that
jS.P; g/ I2 j < "
for every partition P of a; b with .P / < 2 and every Riemann sum S.P; g/
of g over a; b relative to P . Let D minf1 ; 2 g, and choose a partition P of
a; b with norm less than . Let P D .x0 ; x1 ; : : : ; xn /, choose cj 2 xj ; xj C1
for each j < n, and let c D .c0 ; c1 ; : : : ; cn1 /. Since .P / < 1 and .P / < 2 ,
we have
jS.P; f C g; c/ .I1 C I2 /j D jS.P; f; c/ C S.P; g; c/ .I1 C I2 /j

jS.P; f; c/ I1 j C jS.P; g; c/ I2 j
< 2";
as required.
3. Choose " > 0. Then
X
n1
.Mj .f / mj .f //.yj C1 yj / < "
j D0
for every partition Q D .y0 ; y1 ; : : : ; yn / of a; b with small enough norm.

Choose such a partition Q with yl D c for some l. Then we have both
X
l1
.Mj .f / mj .f //.yj C1 yj / < "
j D0
and
X
n1
.Mj .f / mj .f //.yj C1 yj / < ";
j Dl
and we deduce that f is integrable over both a; c and c; b.

Rb Rc Rb
Let a f D I , a f D I1 , and c f D I2 . There exists 0 > 0 such that
jS.P; f / I j < "
for every Riemann sum S.P; f / of f over a; b relative to any partition P of

a; b satisfying .P / < 0 . Similarly, there exist 1 > 0 such that
jS.P; f / I1 j < "
for every Riemann sum S.P; f / of f over a; c relative to any partition P of

a; c satisfying .P / < 1 and 2 > 0 such that
jS.P; f / I2 j < "
for every Riemann sum S.P; f / of f over c; b relative to any partition P

of c; b satisfying .P / < 2 . Let D minf0 ; 1 ; 2 g, and let P1 and P2
be partitions of a; c and c; b, respectively, with norm less than . Let P
be the partition of a; b for which PO D P c1 [ P c2 ; then .P / < . Let
P1 D .x0 ; x1 ; : : : ; xk / and P2 D .xk ; xkC1 ; : : : ; xn /. Choose cj 2 xj ; xj C1
for each j < n, and let c D .c0 ; c1 ; : : : ; cn1 /, c 0 D .c0 ; c1 ; : : : ; ck1 / and
c 00 D .ck ; ckC1 ; : : : ; cn1 /. Then
S.P; f; c/ D S.P1 ; f; c 0 / C S.P2 ; f; c 00 /:

Hence
jI .I1 C I2 /j
D jI S.P; f; c/ C S.P1 ; f; c 0 / C S.P2 ; f; c 00 / .I1 C I2 /j
jI S.P; f; c/j C jS.P1 ; f; c 0 / I1 j C jS.P2 ; f; c 00 / I2 j
< 3":
As " is arbitrary, we conclude that I D I1 C I2 , as required.

4. First we prove the result for the special case where f D g. Choose " > 0. There
X
n1
U.P; f / L.P; f / D .Mj .f / mj .f //.xj C1 xj / < "
j D0
for all partitions P D .x0 ; x1 ; : : : ; xn / of a; b with norm less than . Choose

such a partition P and let
M D supfjf .x/j j x 2 a; bg:
We may assume that M > 0, the required result being clear if f .x/ D 0 for all
x 2 a; b. Note that
Mj .f 2 / mj .f 2 / D supff 2 .cj / f 2 .dj / j fcj ; dj g xj ; xj C1 g
for each j < n, by Lemma 7.4.1. Moreover
f 2 .cj / f 2 .dj / D .f .cj / C f .dj //.f .cj / f .dj //

.jf .cj /j C jf .dj /j/jf .cj / f .dj /j
2M jf .cj / f .dj /j;
and so
Mj .f 2 / mj .f 2 / 2M supfjf .cj / f .dj /j j fcj ; dj g xj ; xj C1 g

D 2M supff .cj / f .dj / j fcj ; dj g xj ; xj C1 g
D 2M.Mj .f / mj .f //;
again by Lemma 7.4.1. Hence
U.P; f 2 / L.P; f 2 / 2M.U.P; f / L.P; f //

< 2M "
whenever .P / < . Thus U.P; f 2 / L.P; f 2 / converges to 0 and so f 2 is

The general case follows from this special case, the equation
.f C g/2 .f g/2
fg D
4
and parts (1) and (2).
t
u
We now introduce the conventions that
Z a
f D0
a
for every a, and

Z b Z a
f D f
a b
whenever b < a. With these conventions we can extend Theorem 7.4.2(3) as

follows. In this theorem no assumption is made about the sizes of a, b, and c.
Theorem 7.4.3. If a; b; c are any real numbers and f is integrable over a closed
interval containing a; b; c, then
Z b Z c Z b
f D f C f:
a a c
Proof. The definitions above imply the theorem immediately if a D b, a D c, or

b D c. If c < a < b, then
Z b Z a Z b
f D f C f;
c c a
so that
Z b Z b Z a Z c Z b
f D f f D f C f:
a c c a c
The argument is similar if a < b < c. The case where a < c < b is covered by
Theorem 7.4.2(3). If b < a, then
Z b Z a Z c Z a Z c Z b
f D f D f f D f C f: t
u
a b b c a c
Theorem 7.4.4. If m f .x/ M for all x 2 a; b and f is integrable over

a; b, then
Z b
m.b a/ f M.b a/:
a
Proof. Let P be a partition .x0 ; x1 ; : : : ; xn / of a; b. Since
X
n1
U.P; f / D Mj .f /.xj C1 xj /
j D0
X
n1
M .xj C1 xj /
j D0
D M.b a/;
it follows that
Z Z
f f M.b a/:
The remaining inequality is proved similarly. t

u
The following corollary is immediate.
Corollary 7.4.5. If f .x/ 0 for all x 2 a; b and f is integrable over a; b, then
Z b
f 0:
a
Corollary 7.4.6. If f .x/ g.x/ for all x 2 a; b and f and g are integrable over
a; b, then
Z b Z b
f g:
a a
Proof. This result is immediate from the fact that

Z b Z b Z b
0 .f g/ D f g: t
u
a a a
Corollary 7.4.7. Suppose that f is continuous on a; b. If f .x/ 0 for all x 2

a; b and f .x/ > 0 for some x 2 a; b, then
Z b
f > 0:
a
Proof. The continuous function f is integrable on a; b. Choose c 2 a; b such

that f .c/ > 0, and without loss of generality suppose that c b. By Theorem 4.4.6
and the continuity of f there exists a number 2 .0; bc/ such that f .x/ f .c/=2
for all x 2 c; d , where d D c C . Hence
Z d
f .c/
f .d c/ > 0;
c 2
and so
Z b Z c Z d Z b
f D f C f C f > 0;
a a c d
Rc Rb
since a f 0 and d f 0. t
u
If f W a; b ! R is a function, then we define jf j to be the function given by
jf j.x/ D jf .x/j
for all x 2 a; b.

Theorem 7.4.8. If a function f is integrable over a; b, then so is jf j. Moreover
Z
b Z b

f jf j:
a a
Proof. Choose " > 0. There exists a partition P D .x0 ; x1 ; : : : ; xn / of a; b such

that
U.P; f / L.P; f / < ":
For all j < n and all t; t 0 2 xj ; xj C1 we have

jf .t /j jf .t 0 /j jf .t / f .t 0 /j
supff .x/ f .x 0 / j fx; x 0 g xj ; xj C1 g

D Mj .f / mj .f /;
by Lemma 7.4.1. Hence
Mj .jf j/ mj .jf j/ D supfjf .x/j jf .x 0 /j j fx; x 0 g xj ; xj C1 g

Mj .f / mj .f /;
so that
U.P; jf j/ L.P; jf j/ U.P; f / L.P; f / < ":
We conclude that jf j is integrable over a; b. The desired inequality follows

Rb Rb
from Corollary 7.4.6, since j a f j D a f . t
u
Theorem 7.4.9 (Mean-Value Theorem for Integrals). Suppose that f and g are
continuous on a; b and that g.x/ 0 for all x 2 a; b. Then there exists 2 a; b
such that
Z b Z b
fg D f ./ g:
a a
Proof. Since f is continuous on a; b, f .x/ attains a minimum m and a maximum

M on a; b. As g.x/ 0 for all x 2 a; b, it follows that
Z Z Z
m g fg M g:
R
Therefore the theorem holds for every 2 a; b if g D 0. In the remaining case
we have
R
fg
m R M:
g
By the intermediate-value theorem there exists 2 a; b such that

R
fg
f ./ D R ;
g

u
By taking g.x/ D 1 for all x, we obtain the following corollary.
Corollary 7.4.10. If f is continuous on a; b, then there exists 2 a; b such that
Z b
f D f ./.b a/:
a
Fig. 7.6 Illustration for f (x)

Corollary 7.4.10
x
a b
Remark. This corollary asserts that, for a function f that is continuous on a; b,
the area bounded by the curve y D f .x/ and the lines x D a, x D b, and y D 0 is
equal to the area of some rectangle with side a; b (see Fig. 7.6).
Exercises 7.3.
1. For all x in a closed interval a; b, define
f C .x/ D maxff .x/; 0g
and
f .x/ D minff .x/; 0g:
Show that f is integrable over a; b if both f C and f are.

2. Prove that if both f and g are integrable over a closed interval a; b, then so are
maxff; gg and minff; gg. [Hint: Note that
1
maxff; gg D .f C g C jf gj/:
2
Find a similar formula for minff; gg.]

3. Let f W a; b ! R be continuous. Show that
Z 1 1=p
lim jf .x/j p
D max jf .x/j
p!1 0 x20;1
and
Z 1 1=p
lim jf .x/j p
D min jf .x/j:
p!1 0 x20;1
4. Let f W a; b ! R beRa continuous function, and suppose that f .x/ 0 for all
x 2 a; b. Show that f D 0 if and only if f .x/ D 0 for all x 2 a; b.
5. Let f W a; b
p! R be an integrable function such that f .x/ 0 for all x 2 a; b.
Show that f is integrable on a; b.
7.5 The Fundamental Theorem of Calculus
This section develops some techniques of integration. First we establish a relation-

ship between the concepts of an integral and an antiderivative.
Theorem 7.5.1 (Fundamental Theorem of Calculus). Let f be continuous on
a; b, and for each x 2 a; b define
Z x
F .x/ D f:
a
Then F 0 .x/ D f .x/ for all x 2 a; b. Moreover, if G is any function on a; b such
that G 0 D f , then
Z b
f D G.b/ G.a/:
a
Proof. Since f is continuous on a; b, the function F indeed exists. We show that
F 0 .t / D f .t / for all t 2 a; b. By definition,
F .x/ F .t /
F 0 .t / D lim
x!t xt
Z x Z t
1
D lim f f
x!t x t a a
Z x
1
D lim f
x!t x t t
f ./.x t /
D lim
x!t xt
D lim f ./
x!t
for some number between x and t . (In the penultimate step of the above calculation
we used the mean-value theorem for integrals.)
In order to prove that F 0 .t / D f .t /, it therefore suffices to show that
lim f ./ D f .t /:
x!t
7.5 The Fundamental Theorem of Calculus 363
Choose " > 0. Since f is continuous at t , there exists > 0 such that jf .x/
f .t /j < " whenever jx t j < . Choose x such that 0 < jx t j < . Since
lies between x and t , we have j t j jx t j < , and so jf ./ f .t /j < ", as
required.
Now suppose G is another antiderivative of f . Then G 0 .x/ D F 0 .x/ D f .x/ for
all x 2 a; b, so that
.F G/0 .x/ D 0
for each such x. Hence F .x/ G.x/ is a constant. Thus

Z b Z a Z b
G.b/ G.a/ D F .b/ F .a/ D f f D f:
a a a t
u
We sometimes write F .b/ F .a/ as F .x/jba .
Example 7.5.1. Since F 0 .x/ D 1=x if F .x/ D log x for all x > 0, we have
Z x
dt
D F .x/ F .1/ D log x
1 t
for all x > 0. This integral is sometimes used as a definition of the logarithm
function. 4
Example 7.5.2. Let
x nC1
f .x/ D
nC1
for all x 2 a; b, where n 1. Then
1
f 0 .x/ D .n C 1/x n D x n
nC1
for all such n. The fundamental theorem therefore implies that

Z b
b
x nC1
x n dx D
a n C 1 a
for all n 1. Similar arguments show that

Z b
sin x dx D cos xjba ;
a
Z b
cos x dx D sin xjba ;
a
and
Z b
e x dx D e x jba :
a 4
The fundamental theorem of calculus asserts that if we differentiate the integral
of a continuous function, then we recover the original function. It is pertinent to ask
the following question: Is it always true that
Z b
f 0 D f .b/ f .a/
a
The answer is no, for f 0 might not be integrable.

Example 7.5.3. Let f be defined by
(
x 2 sin x12 if x 0;
f .x/ D
0 if x D 0:
Then
1 2 1
f 0 .x/ D 2x sin 2
cos 2
x x x
for all x 0. It is easy to see that f 0 is not integrable on any interval that contains
0 and has positive length as it is not bounded on any such interval. 4
However, we do have the following theorem.
Theorem 7.5.2. If f is differentiable on a; b and f 0 is integrable over a; b, then
Z b
f 0 D f .b/ f .a/:
a
Proof. Take any partition P D .x0 ; x1 ; : : : ; xn / of a; b. By the mean-value

theorem, there exist numbers 0 ; 1 ; : : : ; n1 such that j 2 .xj ; xj C1 / and
f .xj C1 / f .xj / D f 0 .j /.xj C1 xj /
for each j < n. Therefore
X
n1
L.P; f 0 / D mj .f 0 /.xj C1 xj /
j D0
X
n1
f 0 .j /.xj C1 xj /
j D0
X
n1
D .f .xj C1 / f .xj //
j D0
D f .b/ f .a/;
by the telescoping property. Similarly,
U.P; f 0 / f .b/ f .a/:
Hence
Z Z
f 0 f .b/ f .a/ f 0:
As f 0 is integrable over a; b, it follows that

Z b
f 0 D f .b/ f .a/:
a
t
u
Theorem 7.5.3. Suppose that
1
X
f .x/ D aj x j
j D0
for all x such that jxj < r, where r is the radius of convergence of the power series
and is nonzero. Then f is integrable over every closed subinterval of .r; r/, and
for each x 2 .r; r/ we have
Z x X1
aj
f D x j C1 : (7.7)
0 j D0
j C 1
Proof. Being differentiable at all x 2 .r; r/ by Theorem 6.7.2, f is continuous,

and therefore integrable, on every closed subinterval of .r; r/.
By Theorem 6.7.1, r is also the radius of convergence for the series on the right-
hand side of Eq. (7.7). We may therefore define
X1
aj
G.x/ D x j C1
j D0
j C 1
for each x 2 .r; r/. It follows from Theorem 6.7.2 that G is differentiable on
.r; r/ and G 0 .x/ D f .x/ for each x 2 .r; r/. We conclude that G 0 is integrable
over every closed subinterval of .r; r/. Therefore Theorem 7.5.2 shows that
Z x
f D G.x/ G.0/ D G.x/: t
u
0
The technique implicit in the following theorem is called integration by

substitution.
Theorem 7.5.4. Let g be a function that is differentiable on an interval c; d and
assume that g 0 is continuous on c; d . Let f be a function that is continuous on the
range of g. Suppose that g.c/ D a and g.d / D b. Then
Z b Z d
f D .f g/g 0 :
a c
Proof. Note that since f g and g 0 are continuous, .f g/g 0 is also continuous,
and therefore integrable, over every interval on which it is defined.
For each x 2 Rg define
Z x
H.x/ D f
a
and for each x 2 c; d let

Z x
G.x/ D .f g/g 0 :
c
Then, by the chain rule and the fundamental theorem of calculus,
.H g/0 .x/ D H 0 .g.x//g 0 .x/

D f .g.x//g 0 .x/
D G 0 .x/:
Hence the functions H g and G differ by a constant, and so
G.d / G.c/ D H.g.d // H.g.c// D H.b/ H.a/:
The required result now follows since H.a/ D G.c/ D 0. t

u
Example 7.5.4. In order to evaluate

Z d
dx
;
c x log x
where c > 1 and d > 1, set f .x/ D 1=x for all x 0 and g.x/ D log x for all
x > 0. Then g 0 .x/ D 1=x for all x > 0. For all such x it follows that
1 g 0 .x/
D D f .g.x//g 0 .x/:
x log x g.x/
Hence
Z d Z log d
dx dx
D
c x log x log c x
log d
D log xjlog c
4
D log log d log log c:
The use of the next theorem is a technique called integration by parts.

Theorem 7.5.5. If f and g are both differentiable over an interval a; b and f 0
and g 0 are integrable over a; b, then
Z b Z b
fg0 D f .b/g.b/ f .a/g.a/ f 0 g:
a a
Proof. It is clear that both the integrals concerned exist. Let
F .x/ D f .x/g.x/
for all x 2 a; b. Then
f .b/g.b/ f .a/g.a/ D F .b/ F .a/

Z
D F0
Z
D .fg0 C f 0 g/
Z Z
0
D fg C f 0 g;

u
Example 7.5.5. For each nonnegative integer n, define

Z =2
In D sinn x dx:
0
For instance, I0 D =2 and

Z =2
=2
I1 D sin x dx D cos xj0 D 1:
0
For the case where n > 1, put f .x/ D sinn1 x and g.x/ D cos x for all x. Thus
Z
=2 =2
In D sinn1 x cos x 0 C .n 1/ sinn2 x cos2 x dx
0
Z =2
D .n 1/ sinn2 .x/.1 sin2 x/ dx
0
D .n 1/.In2 In /:
Hence
nI n D .n 1/In2 ;
so that
n1
In D In2 :
n
Easy inductions now show that
Y 2m 1
n
I2n D
2 mD1 2m
and
Y
n
2m
I2nC1 D
mD1
2m C1
for each positive integer n. For example,

Z =2

sin2 x dx D
0 4
and
Z =2
2
sin3 x dx D :
0 3
Note also that

Z =2
cosn x dx D In
0
for every nonnegative integer n. This formula can be verified either similarly or by
replacing x with =2 x. 4
Integration by parts has a number of interesting applications. We illustrate this
point by using the preceding example to derive a famous formula due to Wallis:
r
22n .n/2
lim p D : (7.8)
n!1 .2n/ 2n 2
When 0 x =2 and n 2 N, we have
0 sin2nC1 x sin2n x sin2n1 x:
Using the notation of Example 7.5.5, we therefore deduce that
0 I2nC1 I2n I2n1 ;
and so
Y
n
2m Y 2m 1
n Y 2m
n1
2n C 1 Y 2m
n
D :
mD1
2m C 1 2 mD1 2m mD1
2m C 1 2n mD1 2m C 1
Applying the result
Y
n
.2m 1/.2m C 1/ D 1 3 3 5 : : : .2n 1/.2n C 1/
mD1
Y
n
D .2n C 1/ .2m 1/2 ;
mD1
we find that
Yn
2m 2m

2 mD1
.2m 1/.2m C 1/
1 Yn
.2m/2
D
2n C 1 mD1 .2m 1/2
Pn
D ;
2n C 1
where
!2
Y
n
2m
Pn D
mD1
2m 1
for all positive integers n. Similarly,
2n C 1 Pn Pn
D :
2 2n 2n C 1 2n
In other words,

an bn ;
2
where an D Pn =.2n C 1/ and bn D Pn =.2n/ for each positive integer n. Therefore
Pn Pn Pn an
bn an D D D ;
2n 2n C 1 2n.2n C 1/ 2n 4n
so that bn an ! 0 as n ! 1 by the sandwich theorem. As

0 an bn an ;
2
the sandwich theorem also shows that

lim an D lim bn D :
n!1 n!1 2
Since
Y
n
2m D 2 4 : : : .2n/ D 2n n
mD1
and
Y
n
.2n/ .2n/
.2m 1/ D 1 3 : : : .2n 1/ D D n ;
mD1
2 4 : : : .2n/ 2 n
we find that
!2
1 Y
n
2m
bn D
2n mD1
2m 1
2
1 22n .n/2
D :
2n .2n/
Therefore
r
p
D lim bn
2 n!1
22n .n/2
D lim p ;
n!1 2n.2n/
as required.
We can use Walliss formula to establish a formula, attributed to Stirling, for
approximating factorials:
p 2nC1
2n 2
n :
en
If we write
ne n
cn D 2nC1
n 2
for all positive integers n, then our goal is to show that

p
lim cn D 2: (7.9)
n!1
Note that
2nC3
cn ne n.n C 1/ 2
D 2nC1
cnC1 n 2 .n C 1/e nC1
2nC1
1 nC1 2
D :
e n
Now define dn D log cn for all n 2 N. Then
dn dnC1 D log cn log cnC1

cn
D log
cnC1
2n C 1 nC1
D log 1:
2 n
Next we introduce a change of variable. Let
1
tD :
2n C 1
Noting that 0 < t < 1, we have
1
2n C 1 D ;
t
so that

1 1 1t
nD 1 D
2 t 2t
and
1t 1Ct
nC1D C1D I
2t 2t
hence
nC1 1Ct
D :
n 1t
Using Eq. (6.18), we therefore find that
1 1Ct
dn dnC1 D log 1
2t 1t
X1
1 t 2j C1
D 2 1
2t j D0
2j C 1
1
X t 2j
D
j D1
2j C 1
> 0:
Therefore the sequence fdn g is decreasing. In order to show that it is bounded below,
we resume the calculation above:
1
X t 2j
dn dnC1 D
j D1
2j C 1
1
X
< t2 t 2j 2
j D1
1
X
D t2 t 2j
j D0
t2
D
1 t2
1
D
.2n C 1/2 1 1
.2nC1/2
1
D
.2n C 1/2 1
1
D
4n.n C 1/
1 1
D :
4n 4.n C 1/
Hence
1 1
dn dnC1 C <0
4n 4.n C 1/
and we conclude that the sequence fdn 1=.4n/g is increasing. Therefore
1 1
d n > dn d1
4n 4
for all positive integers n.
Decreasing but bounded below, the sequence fdn g converges. Let C be its limit.
Then
lim cn D lim e dn D e C ;
n!1 n!1
p
and it remains only to show that e C D 2. We have already proved that
2nC1
n n 2 e C n :
Using Walliss formula, we therefore find that

r
22n n2nC1 e 2C 2n eC
4nC1 p D :
2 .2n/ 2 e C 2n 2n 2
p
Consequently e C 2, and the required equation follows because C is constant.
The proof of the next theorem also uses integration by parts.
Theorem 7.5.6. Let I be an open interval and let a 2 I . Let f W I ! R be

a function. Let n be a nonnegative integer and suppose that f .j / exists and is
continuous for each nonnegative integer j n C 1. Then for each x 2 I we
have
f .x/ D Pn .x/ C Rn .x/;
where
X
n
f .j / .a/
Pn .x/ D .x a/j ;
j D0
j
Z x
1
Rn .x/ D .x t /n f .nC1/ .t / dt
n a
and 00 D 1.
Proof. For every j 2 f0; 1; : : : ; n C 1g, the function f .j / is continuous, and
therefore integrable, on the closed interval with ends a and x. Consequently Rn .x/
exists, by Theorem 7.4.2(4).
We proceed by induction on n. First, for all x 2 I we have P0 .x/ D f .a/ and
Z x
R0 .x/ D f 0 .t / dt D f .x/ f .a/;
a
by Theorem 7.5.2. Therefore the theorem holds for n D 0.

Suppose therefore that n > 0 and the theorem holds for n 1. Integration by
parts yields
Z x
1 x
Rn .x/ D .x t /n f .n/ .t /a C n .x t /n1 f .n/ .t / dt
n a
.n/ Z x
f .a/ 1
D .x a/n C .x t /n1 f .n/ .t / dt
n .n 1/ a
for all x 2 I . By the inductive hypothesis, it follows that
f .x/ D Pn1 .x/ C Rn1 .x/

X
n1 Z x
f .j / .a/ 1
D .x a/j C .x t /n1 f .n/ .t / dt
j D0
j .n 1/ a
X
n1
f .j / .a/ f .n/ .a/
D .x a/j C .x a/n C Rn .x/
j D0
j n
X
n Z x
f .j / .a/ 1
D .x a/j C .x t /n f .nC1/ .t / dt;
j D0
j n a
as required. t
u
In 1947, Niven proved that is irrational. In 1986, Parks proved a more general
result from which the irrationality of both and e follows. We present this result
now. Its proof uses integration by parts.
Theorem 7.5.7. Let c be a positive number and let f W 0; c ! R be a differen-
tiable function such that f .x/ > 0 for all x 2 .0; c/. Suppose there exist functions
f1 ; f2 ; : : : that are differentiable on 0; c and satisfy the conditions that f10 D f ,
fk0 D fk1 for all k > 1 and fk .0/ and fk .c/ are integers for all k. Then c is
irrational.
Proof. Let S be the set of all real polynomials p such that p .k/ .0/ and p .k/ .c/ are
integers for all nonnegative integers k. In other words, p and all its derivatives yield
integers when evaluated at 0 and at c. The set S is closed under multiplication: If
p and q are polynomials in S , then so is pq, for it is easily seen by induction that
.pq/.k/ is a sum of products of derivatives of p and q. (We consider p and q to be
zeroth derivatives of themselves.) Note also that p 2 S if p.0/ and p.c/ are integers
and p 0 2 S . Rc
We show next that if p 2 S , then R c 0 fp is an integer. To this end, we first claim
that it is an integer if and only if 0 f1 p 0 is. Indeed, using integration by parts, we
find that
Z c Z c
fp D f1 .c/p.c/ f1 .0/p.0/ f1 p 0 :
0 0
From the hypotheses and the assumption that p 2 S we see that the first two terms
on the right-hand
R c side are integers, and our claimR c follows. Proceeding inductively,
we find that 0 fp is an integer if and only if 0 f p . / is an integer, where is
the degree of p. Note that p . / is a constant function, and the constant is an integer
0
since p 2 S . Moreover the differentiable function f D f C1 is continuous, and
hence integrable, on 0; c. Therefore, by the fundamental theorem of calculus, we
have
Z c Z c
0
f p . / D p . / .c/ f C1
0 0
Dp . /
.c/.f C1 .c/ f C1 .0//;
Rc
an integer. We conclude that 0 fp is an integer.
Assume now that c is rational. Then there are positive integers m and n such that
c D m=n. Define
x k .m nx/k
Pk .x/ D
k
for all x 2 0; c and all nonnegative integers k, where 00 D 1.
We shall show that Pk 2 S for all k. First, it is clear that the constant polynomial
P0 is in S . Suppose therefore that k > 0 and that Pk1 2 S . Observe first that
Pk .0/ D 0, and that Pk .c/ D 0 since m D nc. For all x 2 0; c we have
kxk1 .m nx/k knxk .m nx/k1

Pk0 .x/ D
k
kxk1 .m nx/k1 .m 2nx/
D
k
D Pk1 .x/.m 2nx/:
The first factor is a polynomial in S by the inductive hypothesis. The second is

in S since m, m 2nc D m, and 2n are all integers. Since S is closed under
multiplication, we conclude that Pk0 2 S and therefore that Pk 2 S for all k as
Pk .0/ and Pk .c/ are integers.
R c that Pk .x/ > 0 for all x 2 .0; c/ and all k, a property shared by
Next observe
f .x/. Thus 0 fPk > 0. But this integral must be an integer, since Pk 2 S . Hence
Z c
f .x/Pk .x/ dx 1 (7.10)
0
for all nonnegative integers k. However, setting
M D maxfx.m nx/ j x 2 0; cg
and
L D maxff .x/ j x 2 0; cg;
we obtain
Z c Z c
Mk
f .x/Pk .x/ dx L dx
0 0 k
cLM k
D :
k
This result gives a contradiction, for
Mk
lim D0
k!1 k
by Example 2.5.6. t
u
Corollary 7.5.8. If 0 < jj and cos and sin are rational, then is
irrational.
Proof. Since cos and sin are rational, so are cos jj and sin jj. Hence we can
find a positive integer n such that n cos jj and n sin jj are integers. Now apply the
theorem with c D jj and f .x/ D n sin x for all x 2 0; jj to conclude that jj,
and hence , is irrational. t
u
Corollary 7.5.9. For every positive rational number a 1, log a is irrational.
Proof. Suppose first that a > 1, so that log a > 0. Write a D m=n for positive
integers m and n, and apply the theorem with c D log a and f .x/ D nex for all
x 2 0; log a, noting that nec D nelog a D na D m.
If 0 < a < 1, then 1=a > 1 and we conclude that log 1=a is irrational by the
previous case. That log a is irrational now follows from the equation
1 t
u
log D log a:
a
Remark 1. By taking D in Corollary 7.5.8, we find that is irrational.
Remark 2. Let a; b; c be positive rational numbers and suppose that a2 C b 2 D c 2 .
There is a number such that 0 < < =2, sin D a=c, and cos D b=c. Then
both sin and cos are rational, so that is irrational. In other words, arctan.a=b/
is irrational.
Remark 3. Corollary 7.5.9 confirms the irrationality of e, since log e D 1. In fact,
e a is irrational for every nonzero rational number a.
In Sect. 6.4 we introduced the number . We now give it a geometric interpre-
tation by showing that the circumference of a unit circle is 2. For this purpose we
need the notion of arc length.
Let f be a continuous real-valued function defined on a closed interval I with
partition .x0 ; x1 ; : : : ; xn /. For each j let Pj be the point .xj ; f .xj //. For each j < n
the distance between Pj C1 and Pj is jPj C1 Pj j. The sum of these distances gives
an approximation for the length of the graph of f . If
X
n1
lim jPj C1 Pj j
n!1
j D0
exists and is finite, then we define it to be the (arc) length of the graph of f over I .
It is evaluated in the following theorem.
Theorem 7.5.10. Let f be a differentiable function from an interval a; b into R.
Suppose that f 0 is continuous on a; b. Then the length of the graph of f over
a; b is
Z bp
1 C .f 0 /2 :
a
Proof. By applying the mean-value theorem to f over xj ; xj C1 , where j < n, we

discover a j 2 .xj ; xj C1 / for which
f .xj C1 / f .xj / D f 0 .j /.xj C1 xj /:
Therefore
X
n1 n1 q
X
jPj C1 Pj j D .xj C1 xj /2 C .f .xj C1 / f .xj //2
j D0 j D0
n1 q
X
D .xj C1 xj /2 C .f 0 /2 .j /.xj C1 xj /2
j D0
X
n1 q
D .xj C1 xj / 1 C .f 0 /2 .j /;
j D0
p
which is a Riemann sum for the function 1 C .f 0 /2 . This function is continuous,
and therefore integrable, on a; b. Its Riemann sums converge to the required arc
length, and the result follows. t
u
p
Example 7.5.6. Since arccos.1= 2/ D =4, the length of the graph of the function
p
f .x/ D 1 x 2 ;
p
where x 2 0; 1= 2, is an eighth of the circumference of the unit circle. In order
to evaluate it, we first use the chain rule to compute
1
f 0 .x/ D p .2x/
2 1 x2
x
D p :
1 x2
This function is continuous at all x 1. Therefore
Z 1=p2 p Z 1=p2 r
x2
1 C .f 0 /2 D 1C dx
0 0 1 x2
Z p
1= 2
dx
D p
0 1 x2
p
1= 2
D arcsin xj0

D :
4
Consequently the circumference of the unit circle is 8 =4 D 2. 4

Exercises 7.4.
1. Use the result of Example 7.5.1 to prove the following:
(a) If F .x/ D log x for all x > 0 and cx > 0 for some constant c, then
1
F 0 .cx/ D
x
for all x > 0.
(b) If x > 0 and y > 0, then
log xy D log x C log y:
(c) If x > 0 and p is a rational number, then
log x p D p log x:
(d)
lim log x D 1
x!1
and
lim log x D 1:

x!0C
Rx
2. Let f be a function that is integrable over a; b and let F .x/ D a f for all
x 2 a; b.
(a) Show that for all x; y 2 a; b satisfying x < y we have
jF .x/ F .y/j < M.y x/;
where M is the least upper bound of jf j on a; b.

(b) Hence show that F is continuous on a; b. (Note that f need not be
continuous, and therefore the fundamental theorem of calculus cannot be
used.)
3. By choosing a particular type of partition of the interval 1; 2 and suitable
Riemann sums for
Z 2
dx
;
1 x2
evaluate
X
n
1
lim n :
n!1
j D1
.n C j /2
(The evaluation of this limit is due to Darboux.)

R1
Riemann sums for 0 e x dx, evaluate
1 X j 1
n
lim e n :
n!1 n
j D1

R1
Riemann sums for 0 x m dx, evaluate
1 X
n
lim j m:
n!1 nmC1 j D1
6. Let f W a; b ! RR be integrable. Show that if f is continuous at a point

x
c 2 .a; b/, then a f is differentiable at c. (Note that this result implies the
fundamental theorem of calculus.)
7. Let f W a; b ! RR be a continuous function such that f .x/ > 0 for all x 2
x
a; b. Show that a f is increasing on a; b.
8. Suppose that f 00 is continuous on a; b. Show that
Z b
xf 00 .x/ dx D bf 0 .b/ f .b/ .af 0 .a/ f .a//:
a
9. Let f be a continuous real-valued function and let u and v be differentiable

functions. Define
Z v.x/
F .x/ D f .t / dt:
u.x/
Show that
F 0 .x/ D f .v.x//v 0 .x/ f .u.x//u0 .x/:
10. Let f and g be functions from a; b to R such that g is continuous, f

monotonic, and f 0 integrable. Show that there exists c 2 a; b for which
Z b Z c Z b
fg D f .a/ g C f .b/ g:
a a c
7.6 The CauchySchwarz Inequality 381
This result is known as the second mean-value theorem. [Hint: Let

Z x
G.x/ D g.t / dt
a
for each x 2 a; b. Apply integration by parts to

Z b Z b
fg D fG0
a a
and then use the mean-value theorem.]

11. For each m; n 2 N evaluate the following integrals:
R
(a) R sin mx sin nx dx;

(b) R cos mx cos nx dx;

(c) sin mx cos nx dx.
7.6 The CauchySchwarz Inequality
Another theorem about integrals is known as the CauchySchwarz inequality.

Theorem 7.6.1 (CauchySchwarz Inequality). If f and g are integrable on a
closed interval, then
Z 2 Z Z
fg f 2
g2 :
Proof. Note that the functions fg, f 2 , and g 2 are integrable by Theorem 7.4.2(4).
For all real x we have
Z Z Z Z
.xf C g/2 D x 2 f 2 C 2x fg C g 2 :
The polynomial on the right-hand side of this equation is therefore nonnegative for
all real x. Thus its discriminant cannot be positive, by the results of Example 6.3.4.
In other words,
Z 2 Z Z
4 fg 4 f 2 g 2 0:
The required inequality follows. t

u
Let f W a; b ! R be integrable. We define the norm kf k of f by the equation
Z 12
kf k D f2 :
With this notation we may rewrite the CauchySchwarz inequality as

Z
fg kf kkgk:
The norm of a function satisfies the triangle inequality.

Corollary 7.6.2. Let f; gW a; b ! R be integrable. Then
kf C gk kf k C kgk:
Proof. Note first that kf k 0 for every integrable function f . Now

Z
kf C gk D 2
.f C g/2
Z Z Z
D f2C2 fg C g2
kf k2 C 2kf kkgk C kgk2

D .kf k C kgk/2 ;
and the result follows by taking square roots of both sides. t

u
A similar argument to that used to prove Theorem 7.6.1, combined with
Theorem 3.12.5, yields the following corresponding result for series. The details
of the proof are left as an exercise.
P P1 2
Theorem 7.6.3. Let fxn g and fyn g be sequences such that 1 2
j D0 xj and j D0 yj
P1
converge. Then j D0 jxj yj j converges and
0 12
1
X 1
X 1
X
@ jxj yj jA xj2 yj2 :
j D0 j D0 j D0
Exercises 7.5.
1. If f and g are integrable functions, prove that
kf k kgk kf gk
and
kf k D jjkf k
for every number .

7.7 Numerical Integration 383
2. Let f and g be integrable over an interval a; b. Prove that

0 !2 1
Z bZ b Z b
f .x/ g.x/ 2
@ f .x/g.x/ dx A :
f .y/ g.y/ dy dx D 2 kf k kgk
2 2
a a a
(Note that the integrand on the left-hand side is the square of a determinant.)
Deduce the CauchySchwarz inequality from this equation.
3. Let f and g be integrable over an interval a; b.
(a) Prove that
Z b Z b
.f .y/ f .x//.g.y/ g.x// dy dx
a a
Z Z Z !
b b b
D 2 .b a/ f .x/g.x/ dx f .x/ dx g.x/ dx :
a a a
(b) Suppose that f and g are both increasing or both decreasing. Deduce that
Z b Z b Z b
f g .b a/ fg:
a a a
4. Prove Theorem 7.6.3.
7.7 Numerical Integration
The fundamental theorem of calculus is a very useful tool for evaluating integrals.
However, it is not always applicable, as there are integrable functions such as
e x and sin sin x whose antiderivatives cannot be expressed explicitly in terms
2
of such elementary functions as polynomials, the trigonometric functions, and the

exponential and logarithm functions. In this section we therefore investigate the
question of approximating a definite integral by means of a Riemann sum. We also
wish to estimate the accuracy of the resulting approximations.
We begin with a lemma.
Lemma 7.7.1. Let f be a function with continuous derivative on an interval a; b.
For every c 2 a; b let
Z b
"D f f .c/.b a/:
a
Then
.b a/2
j"j sup jf 0 .x/j:
2 x2a;b
Proof. Since f 0 is continuous on the closed interval a; b, we may define
M D sup jf 0 .x/j:
x2a;b
By the mean-value theorem (for differentiation), for every x 2 a; b fcg there

exists between x and c such that
jf .x/ f .c/j D jf 0 ./.x c/j

M jx cj:
Hence
Z
b

j"j D .f .x/ f .c// dx
a
Z b
jf .x/ f .c/j dx
a
Z b
M jx cj dx
a
Z Z !
c b
DM .x c/ dx C .x c/ dx
a c
c b !
.x c/2 .x c/2
DM C
2 a 2 c

.a c/2 .b c/2
DM C
2 2
M
.b a/2
2
since
.b a/2 D ..b c/ C .c a//2

D .b c/2 C 2.b c/.c a/ C .c a/2
.b c/2 C .c a/2 :
t
u
Theorem 7.7.2. Let f be a function with continuous derivative on an interval a; b
with partition P D .x0 ; x1 ; : : : ; xn /. Then every Riemann sum S.P; f / of f over
a; b relative to P satisfies the inequality
Z
b ba

f S.P; f / sup jf 0 .x/j.P /:
a 2 x2a;b
Proof. As in the proof of the lemma, let
M D sup jf 0 .x/j:
x2a;b
For any Riemann sum S.P; f; c/ of f over a; b relative to P and intermediate

points c0 ; c1 ; : : : ; cn1 , we have
Z n1 Z

b X xj C1 X
n1

f .x/ dx S.P; f; c/ D f .x/ dx f .cj /.xj C1 xj /
a
j D0 xj j D0

n1 Z xj C1

X

f .x/ dx f .cj /.xj C1 xj /
xj
j D0
X
n1
M
.xj C1 xj /2
j D0
2
M X
n1
.P /.xj C1 xj /
2 j D0
1
D M.P /.b a/: t
u
2
Although the upper bound this theorem provides for the error may seem crude,
the following example shows that it cannot be improved.
R1
Example 7.7.1. Let f .x/ D x for all x 2 0; 1. Then 0 f D 1=2. For each n 2 N
let Pn be the partition

1 2
0; ; ; : : : ; 1 ;
n n
so that .Pn / D 1=n. By choosing as intermediate points the leftmost end of each
subinterval of Pn or the rightmost end of each such subinterval, we obtain the lower
sum L.Pn ; f / or the upper sum U.Pn ; f /, respectively, of f over 0; 1 relative to
Pn . Now for each j the j th subinterval of Pn is .j 1/=n; j=n, so that
1Xj
n1
n.n 1/ 1 1
L.Pn ; f / D D 2
D
n j D0 n 2n 2 2n
and
1Xj
n
n.n C 1/ 1 1
U.Pn ; f / D D D C :
n j D1 n 2n2 2 2n
Hence
Z Z
1 1 1
f .x/ dx L.Pn ; f / D f .x/ dx U.Pn ; f / D ;
2n
0 0
and this number is the upper bound provided by the theorem for the error. 4
We now consider the case where a Riemann sum is obtained by using the
midpoints of each subinterval of a partition.
Lemma 7.7.3. Let f be a function with continuous second derivative on an interval
a; b. Let c D .a C b/=2 and define
Z b
"D f f .c/.b a/:
a
Then
.b a/3
j"j sup jf 00 .x/j:
24 x2a;b
Proof. Using Theorem 7.5.6 with n D 1, we obtain

Z x
f .x/ f .c/ D f 0 .c/.x c/ C .x t /f 00 .t / dt
c
for all x 2 .a; b/. Thus

Z b
"D .f .x/ f .c// dx
a
Z b Z b Z x
D f 0 .c/.x c/ dx C .x t /f 00 .t / dt dx:
a a c
But
Z Z
b
0 0
b
aCb
f .c/.x c/ dx D f .c/ x dx
a a 2
Z !
0
b
.a C b/.b a/
D f .c/ x dx
a 2

0b 2 a2 b 2 a2
D f .c/
2 2
D 0:
Setting
M D sup jf 00 .x/j;
x2a;b
we therefore conclude that

Z Z x
b
j"j .x t /f 00 .t / dt dx

a c
Z b Z x
.x t /M dt dx
a c
Z x
b
.t x/2
D M dx
a 2 c
Z b
M
D .x c/2 dx
2 a
M
D .b c/3 .a c/3
6
!
M aCb 3 aCb 3
D b a
6 2 2
M 1
D .b a/3 .a b/3
6 8
M
D .b a/3 : t
u
24
Theorem 7.7.4. Let f be a function such that f 00 exists and is continuous on an

interval a; b with partition .x0 ; x1 ; : : : ; xn /. Let c D .c0 ; c1 ; : : : ; cn1 /, where
xj C1 C xj
cj D
2
for all j < n. Then
Z
b ba

f S.P; f; c/ sup jf 00 .x/j..P //2 :
a 24 x2a;b
Proof. Letting
M D sup jf 00 .x/j
x2a;b
and using Lemma 7.7.3, we obtain

Z n1 Z
b X xj C1

f .x/ dx S.P; F; c/ f .x/ dx f .cj /.xj C1 xj /
a xj
j D0
X
n1
M
.xj C1 xj /3
j D0
24
M X
n1
..P //2 .xj C1 xj /
24 j D0
1
D M..P //2 .b a/:
24 t
u
Again, the bound is sharp.
R1
Example 7.7.2. Let f .x/ D x 2 for all x 2 0; 1, so that 0 f D 1=3. For each
n 2 N let Pn be the partition used in Example 7.7.1. Defining c0 ; c1 ; : : : ; cn1 ; c as
in the theorem, we find that

1 j j C1 jC 1
cj D C D 2
2 n n n
for each j < n. Therefore

1X 1
n1
1 2
S.Pn ; f; c/ D jC
n j D0 n2 2
n1
1 X 2 1
D j C j C
n3 j D0 4

1 n.n 1/.2n 1/ n.n 1/ n
D 3 C C
n 6 2 4
4n2 1
D
12n2
1 1
D :
3 12n2
7.8 Improper Integrals 389
Hence
Z
1 1
f .x/ dx S.Pn ; f; c/ D ;
12n2
0
the bound given by the theorem. 4

The reader is referred to [3] for more on numerical integration.
Exercises 7.6.
1. Let f W a; b ! R be monotonic. For each n 2 N let h D .b a/=n and
Pn D .a; a C h; a C 2h; : : : ; b/:
For each Riemann sum S.Pn ; f / of f over a; b relative to Pn , show that

Z
b jb aj

f .x/ dx S.Pn ; f / jf .b/ f .a/j:
a n
2. Let b > 0 and
f .0/ C f .b/
SD b
2
for some function f . If jf 00 .x/j M for all x 2 0; b, show that

Z
b
1
f .x/ dx S M.b a/3 :
0 12
Rb
[Hint: Apply integration by parts twice to evaluate 0 t .b t /f 00 .t / dt.]
3. Using four subintervals and choosing their midpoints as the intermediate points,
R2
approximate log 2 D 1 dx x
by means of a Riemann sum and estimate the error.
7.8 Improper Integrals
Although there are several types of improper integrals, here we remove only the
condition that the domain of the function is a finite interval. Let f be a function that
is integrable over a; n, for every integer n a. We define
Z 1 Z n
f D lim f;
a n!1 a
provided the limit exists. In this case we say that the integral converges; otherwise
it diverges. If a b n, then
Z 1 Z n
f D lim f
a n!1 a
Z Z !
b n
D lim f C f
n!1 a b
Z b Z 1
D f C f:
a b
Example 7.8.1. We show that the integral

Z 1
x p dx
1
converges if and only if p > 1.

If p 1, then
Z 1 Z n
x p dx D lim x p dx
1 n!1 1
n
x 1p
D lim
n!1 1 p
1
n1p 1
D lim
n!1 1 p
(
1
if p > 1
D p1
1 if p < 1:
If p D 1, then the integral becomes

Z 1 Z n
dx dx
D lim
1 x n!1 1 x
D lim log xjn1
n!1
D lim log n
n!1
D 1: 4
One of the basic tools for establishing convergence of an improper integral is the
comparison test. It is analogous to the comparison test for convergence of a series.
Theorem 7.8.1 (Comparison Test). Let a 2 R. Suppose that f and g are

integrable
R1 R 1all b > a and that jf .x/j g.x/ for all x a.
functions over a; b for
If a g converges, then so does a f . Moreover
Z Z
1 1
f g:

a a
Proof. First note that jf j is also integrable over a; b for all b > a, by
Theorem 7.4.8.
Suppose that f .x/ 0 for all x a. Then the sequence

Z n
f
a
R1
is nondecreasing and bounded above by a g, according to Corollary 7.4.6. In this
case the result follows from Theorem 2.7.1.
For the general case, let
jf j C f
f1 D
2
and
jf j f
f2 D :
2
Then f1 and f2 are integrable over a; b for all b > a. Furthermore, for each x a
and j 2 f1; 2g we have
0 fj .x/ jf .x/j g.x/:

R1 R1 R1
Hence a f1 and a f2 both converge. As f D f1 f2 , a f also converges.
Finally, by Theorem 7.4.8 and Corollary 7.4.6,
Z
b Z b Z b

f jf j g:
a a a
The proof is completed by taking the limit as b ! 1. t

u
Ra
A corresponding result holds for integrals of the form 1 f .
The theorem immediately implies the following corollary.
Corollary
R1 7.8.2. Suppose f is Ran integrable function over a; b for all b > a. If
1
a jf j converges, then so does a f .

Z 1
t x1 e t dt
1
converges for all x > 0. We first locate the maximum of the function f defined by
f .t / D t xC1 e t
for all t > 1. Note that
f 0 .t / D .x C 1/t x e t t xC1 e t D t x e t .x C 1 t /
for all such t . Hence f 0 .t / D 0 if and only if t D x C 1. Since f 0 .t / > 0 when

t < x C 1 and f 0 .t / < 0 when t > x C 1, the maximum of f occurs at x C 1. Let
M D f .x C 1/. Then
Z n Z n Z n
x1 t 2
t e dt D f .t /t dt M t 2 dt
1 1 1
R1
for all n 2 N. By Example 7.8.1, 1 t 2 dt converges. Hence
Z 1
t x1 e t dt
1
converges by the comparison test. 4

R1 R1
The improper integral a f is said to be absolutely convergent if a jf j is
convergent.
The following example, which is due to Dirichlet, shows that a convergent
improper integral need not be absolutely convergent.
Z 1
sin x
dx
1 x
is convergent but not absolutely.

Using integration by parts (Theorem 7.5.5), we obtain
Z Z n
n
sin x cos x n cos x
dx D dx
1 x x 1 1 x2
for all n 2 N. Now

cos x n cos n
D C cos 1 ! cos 1
x 1 n
as n ! 1, and since
cos x
1
2 2;
x x
the integral
Z 1
cos x
dx
1 x2
converges by the comparison test. Hence the integral in question is convergent.

We show next that the integral is not absolutely convergent. For all

3
x2 C j; C j ;
4 4
where j 2 N, we have
1
j sin xj sin Dp
4 2
and x < .j C 1/. Hence

Z 1 X1 Z .j C 3 /
j sin xj 4 j sin xj
dx > dx
1 x 1
j D1 .j C 4 /
x
1 Z
1 X .j C 4 / dx
3
p
2 j D1 .j C 14 / x
1
1 X 1
> p
2 j D1 2 .j C 1/
1
1 X1
D p :
2 2 j D2 j
P1
As j D2 1=j diverges, we deduce that
Z 1
j sin xj
dx
1 x
also diverges. 4
Exercises 7.7.
1. Test the following integrals for convergence:
R1
(a) 0 xedxx ;
R0
(b) 1 e ax dx;
R1
(c) 0 sin x 2 dx.
2. Evaluate
Z 1
log x
dx:
1 x2
3. (a) Suppose that f .x/ 0 and g.x/ 0 for all x a, and let
f .x/
lim D L:
x!1 g.x/
Prove that
R1 R1
i. If 0 < L < 1,Rthen a f and a R g both converge or both diverge.
1 1
ii. If L D 0, then aR f converges if Ra g converges.
1 1
iii. If L D 1, then a g converges if a f converges.
(b) The result of part (a) is called the limit comparison test for integrals. Use it
to prove that
Z 1
dx
1 a C x2
2
is convergent.
7.9 Integral Test for Convergence of a Series
Since an integral is defined in terms of Riemann sums, there are many similarities
between the theories of integrals and series. One of the most important connections
between the two theories is the integral test for the convergence of a series. It
involves improper integrals.
Theorem 7.9.1 (Euler, Maclaurin). Let f be a nonincreasing continuous function
on the interval 1; 1/, and suppose that f .x/ 0 for all x 1. Let
X n Z n
dn D f .j / f
j D1 1
for all n 2 N. Then the sequence fdn g converges.

7.9 Integral Test for Convergence of a Series 395
Proof. It suffices to show that fdn g is nonincreasing and bounded below. Since f is
nonincreasing, for every positive integer j we have
f .j C 1/ D minff .x/ j x 2 j; j C 1g
and
f .j / D maxff .x/ j x 2 j; j C 1g:
Hence
Z j C1
f .j C 1/ f f .j /;
j
so that
0 1 0 1
X
nC1 Z nC1 X
n Z n
dnC1 dn D @ f .j / fA@ f .j / fA
j D1 1 j D1 1
Z nC1
D f .n C 1/ f
n
f .n C 1/ f .n C 1/
D0
for each n 2 N.
We have now proved that fdn g is nonincreasing. It is also bounded below by 0,
since
X
n n1 Z
X j C1
dn D f .j / f
j D1 j D1 j
X
n X
n1
f .j / f .j /
j D1 j D1
D f .n/
0
for each n 2 N. t
u
Corollary 7.9.2. Under the hypotheses of Theorem 7.9.1 we have

Z nC1 X
n Z n
f f .j / f .1/ C f
1 j D1 1
for all n 2 N.
Proof. From the proof of the theorem we have
Z j C1
f .j C 1/ f f .j /
j
for all j 2 N. From the second inequality we obtain

Z nC1 n Z
X j C1 X
n
f D f f .j /:
1 j D1 j j D1
The first inequality gives
X
n1 n1 Z
X j C1 Z n
f .j C 1/ f D f:
j D1 j D1 j 1
Consequently
X
n X
n X
n1 Z n
f .j / D f .1/ C f .j / D f .1/ C f .j C 1/ f .1/ C f: t
u
j D1 j D2 j D1 1
Corollary 7.9.3 (Integral Test). Let f be a nonincreasing continuous function on

P1interval 1; 1, and suppose
the R 1 that f .x/ 0 for all x 1. Then the series
j D1 f .j / and the integral 1 f both converge or both diverge.
Remark. The EulerMaclaurin theorem and the integral test are clearly true if there
exists a fixed number N such that the function satisfies all the conditions for each
x N.
P
Example 7.9.1. Use the integral test to show that 1 p
j D1 1=j is convergent if and
only if p > 1.
Solution. Let
1
f .x/ D >0
xp
for all x > 0.
7.9 Integral Test for Convergence of a Series 397
If p 0, we have
1
lim 0
n!1 np
and the series is divergent by the nth term test.

Suppose p > 0. Then f is clearly nonincreasing and continuous at all x 1
and the integral test can be applied. Therefore, by the integral test and the result of
Example 7.8.1, the series is convergent if and only if p > 1. 4
Example 7.9.2. Use the integral test to confirm the divergence of Abels series
1
X 1
:
j D2
j log j
Solution. Let
1
f .x/ D >0
x log x
for all x > 1. Then as x log x is increasing at all x 2, it follows that f is

decreasing at all x 2. Moreover f is continuous at all x 2. Hence the integral
test can be applied.
Using the result of Example 7.5.4, we have
Z 1
dx
D lim .log log n log log 2/ D 1:
2 x log x n!1
Therefore the series is divergent by the integral test. 4

Series for which the kth term involves log k are often suitable for the use of the
integral test. One problem with the integral test is that the required integral may not
be easy to evaluate. A comparison test is often used to modify the series so that the
corresponding integral may be found easily.
Example 7.9.3. Test the convergence of
1
X 1
2 C 3j C 1/
:
j D1
j log.j
Solution. Let
1
an D
n log.n2 C 3n C 1/
and
1
bn D
n log n
for all integers n > 1. Then, using lHpitals rule,
an n log n
lim D lim
n!1 bn n!1 n log.n2 C 3n C 1/
log n
D lim
n!1 log.n2 C 3n C 1/
1
D lim n
2nC3
n!1
n2 C3nC1
n2 C 3n C 1
D lim
n!1 2n2 C 3n
1
D :
2
P P1
Since 1 j D2 bj is divergent by the previous example, j D1 aj is also divergent by
the limit comparison test. 4
Exercises 7.8.
P1 1 P1 1
(a) j D2 j log2 j . (c) j D2 j log j log log j .
P1 1
(b) j D2 .j C5/ log2 j .
2. Does the result
Z 1 X1
dx 1
D 1
1 x2 j D1
j 2
contradict the argument used in the proof of the integral test?

3. Apply the integral test to test the convergence of
1
X 1
:
j D2
j logp j
Chapter 8
Taylor Polynomials and Taylor Series
8.1 Taylors Theorem
Because polynomials are so easy to study, it would be very convenient if they

were the only functions with which we had to deal. Although this is not the case,
it turns out that many functions that are not polynomials can be approximated as
accurately as we please by polynomials. In this section we present a theorem that
gives conditions under which such an approximation is possible.
Given a function f that is not a polynomial, we are interested in finding a
polynomial, in a variable x, that gives a good approximation for f .x/ at values
of x near some number a. It is convenient to write the polynomial in powers of
x a rather than x. If we take the polynomial to be of degree n 0 and denote it
by Pn , then we can write
X
n
Pn .x/ D aj .x a/j (8.1)
j D0
for all x, where a0 ; a1 ; : : : ; an are constants and we take 00 D 1. We also assume

that the graph of f is smooth at a in the sense that f .a/; f 0 .a/; : : : ; f .n/ .a/ exist
and are known. This may seem to be rather a strong requirement, but in fact it is
satisfied for every n 0 by many important functions, such as e x and sin x at 0 and
log x at 1.
It is reasonable to expect that at a the approximation is exact and the derivatives
of Pn and f are equal. In fact, we shall require that

400 8 Taylor Polynomials and Taylor Series
Pn.k/ .a/ D f .k/ .a/
for all integers k such that 0 k n.

We begin by deriving a formula for the coefficients of Pn . By differentiating the
sides of Eq. (8.1) k times, where 0 k n, we obtain
X
n
Pn.k/ .x/ D aj j.j 1/ .j k C 1/.x a/j k
j D0
X
n
j
D aj .x a/j k
.j k/
j Dk
X
n
j
D ak k C aj .x a/j k
.j k/
j DkC1
for all x. Therefore
f .k/ .a/ D Pn.k/ .a/ D ak k;
and so
f .k/ .a/
ak D :
k
Substitution into Eq. (8.1) yields
X
n
f .j / .a/
Pn .x/ D .x a/j
j D0
j
for all x. This polynomial is called the Taylor polynomial of order n for f
about a.
Example 8.1.1. Find the Taylor polynomial of order 5 for cos x about 0.
Solution. Taking f .x/ D cos x for all x and a D 0, we have f .0/ D 1. Moreover
f 0 .x/ D sin x
for all x, and so f 0 .0/ D 0. Similarly, we find that f .2/ .0/ D 1, f .4/ .0/ D 1, and
f .3/ .0/ D f .5/ .0/ D 0. Therefore
x2 x4
P5 .x/ D 1 C :
2 24
Notice that in this case the Taylor polynomial of order 5 is actually of degree 4
because the term in x 5 has a coefficient of 0. 4
8.1 Taylors Theorem 401
We now prove Taylors theorem, which gives a means of estimating the error
incurred in using a Taylor polynomial as an approximation for the value of a function
at some number.
Theorem 8.1.1 (Taylor). Let a be a real number, n a nonnegative integer, and f a
function such that f .nC1/ .x/ exists for all x in some open interval I containing a.
Let Pn be the Taylor polynomial of order n for f about a. Then for each x 2 I fag
there exists a number between a and x such that
f .nC1/ ./
f .x/ Pn .x/ D .x a/nC1 :
.n C 1/
Proof. Fix x 2 I fag, and for all t 2 I define
X
n
f .j / .t /
F .t / D f .x/ .x t /j
j D0
j
X
n
f .j / .t /
D f .x/ f .t / .x t /j :
j D1
j
Since f .nC1/ .t / exists for all t 2 I , we infer that f .j / must be continuous on I for
all j such that 0 j n. Hence F is continuous on the closed interval with ends
a and x and differentiable on the corresponding open interval. Note that
F .x/ D f .x/ f .x/ D 0

and
X
n
f .j / .a/
F .a/ D f .x/ .x a/j D f .x/ Pn .x/:
j D0
j
Thus we need to show that

f .nC1/ ./
F .a/ D .x a/nC1
.n C 1/
for some between a and x.
Using the product rule, the chain rule, and the telescoping property, we obtain
n .j C1/
X
f .t / f .j / .t /
F 0 .t / D f 0 .t / .x t /j C j.x t /j 1 .1/
j D1
j j
n .j C1/
X
f .t / f .j / .t /
D f 0 .t / .x t /j .x t /j 1
j D1
j .j 1/

0 f .nC1/ .t /
D f .t / .x t /n f 0 .t /
n
.x t /n .nC1/
D f .t /:
n
Now let
G.t / D .x t /nC1
for all t 2 I . Then
G 0 .t / D .n C 1/.x t /n
for all such t . Note also that G.x/ D 0 and that G 0 .t / 0 for all t x. By
Cauchys mean-value formula there exists between a and x such that
F 0 ./ F .a/ F .x/ F .a/

D D :
G 0 ./ G.a/ G.x/ G.a/
Hence
F 0 ./
F .a/ D G.a/
G 0 ./
n
.x/ f .nC1/ ./.x a/nC1
D n
.n C 1/.x /n
f .nC1/ ./
D .x a/nC1 ;
.n C 1/
as required. t
u
As a consequence of this theorem we have
f .nC1/ ./
f .x/ D Pn .x/ C .x a/nC1
.n C 1/
X
n
f .j / .a/ f .nC1/ ./
D .x a/j C .x a/nC1
j D0
j .n C 1/
for all x 2 I fag. Note also that the value of depends on the choice of x. For
each x 2 I fag we define
f .nC1/ ./
Rn .x/ D .x a/nC1 : (8.2)
.n C 1/
Thus
Rn .x/ D f .x/ Pn .x/
for each such x. The function Rn is called the Taylor remainder of degree n.
Example 8.1.2. Use the Taylor polynomial of order 5 for cos x about 0 to obtain an
approximation for cos 0:1 and use Taylors theorem to show that the approximation
gives the first eight digits in the decimal expansion of cos 0:1 correctly.
Solution. From Example 8.1.1 we have
x2 x4
P5 .x/ D 1 C
2 24
for all x. The required approximation for cos 0:1 is therefore
P5 .0:1/ D 0:9950041666 : : : :
Since the sixth derivative of cos x is cos x, Taylors theorem shows that there
exists a number between 0 and 0.1 such that
.0:1/6
cos 0:1 P5 .0:1/ D . cos / D 0:0000000013888 : : : .cos /:
6
Since 0 < cos < 1, we have
0:0000000013888 : : : < cos 0:1 P5 .0:1/ < 0:
Hence
0:0000000014 < cos 0:1 P5 .0:1/ < 0;
and so
P5 .0:1/ 0:0000000014 < cos 0:1 < P5 .0:1/:
We conclude that
0:99500416526 : : : < cos 0:1 < 0:9950041666 : : : :
Thus P5 .0:1/ gives the first eight digits in the decimal expansion of cos 0:1 correctly
as 0.99500416. 4
A function f is said to be smooth over an interval I if f .k/ .x/ exists for all
x 2 I and all nonnegative integers k. We write C 1 .I / for the set of functions that
are smooth over I .
Example 8.1.3. Let f 2 C 1 .I /, where I is an open interval containing 1, and

suppose that f .1/ D 0, f 0 .1/ D 1, f 00 .1/ D 2, and
2
ex
f 000 .x/ D
1Cx
for all x 2 I f1g. Write down the Taylor polynomial of order 3 for f about 1.
Suppose that this polynomial is used to approximate f .0:5/. Find an upper bound
for the error and write down the approximate value of f .0:5/.
Solution. The Taylor polynomial is
.x 1/2 00 .x 1/3 000

P3 .x/ D f .1/ C .x 1/f 0 .1/ C f .1/ C f .1/
2 3
e.x 1/3
D .x 1/ C .x 1/2 C :
12
The error incurred by the approximation is given by jR3 .0:5/j. Now,

2
e 2 2 C 2 1
f .4/
./ D ;
.1 C /2
where 0:5 < < 1, and since 2 2 C 2 1 is increasing at all > 0:5, it follows
that
3e 4
jf .4/ ./j < D e < 4:
.1:5/2 3
We thus have
jf .4/ ./jj0:5 1j4 4

jR3 .0:5/j D < < 0:011:
4 424
The approximation given by P3 .0:5/ is
.0:125/e
P3 .0:5/ D 0:5 C 0:25 0:28
12
with error less than 0:011. 4
A common problem in the approximation of numbers or functions is to make
the approximation correct to within a specified error tolerance. If we use the Taylor
polynomial of order n about a point a to approximate a function f at a point x near
a, then the error depends on n, the distance jx aj, and the absolute values of the
derivatives of f between x and a. The next example illustrates the problem and a
solution strategy.
Example 8.1.4. Let f .x/ D log.x C 1/ for all x > 1. What degree Taylor
polynomial about 0 is required to approximate log 1:1 with a maximum error of
0.001? What degree Taylor polynomial about 0 is required to approximate log 1:5
with the same maximum error?
Solution. For all n 1 and all > 0 we have
.n 1/
jf .n/ ./j D
.1 C /n
< .n 1/:
To approximate log 1:1, we note that x D 0:1 and
jf .nC1/ ./j.0:1/nC1
jRn .0:1/j D
.n C 1/
n
<
.n C 1/10nC1
1
D :
.n C 1/10nC1
We thus seek a value of n such that
.n C 1/10nC1 1000:
Evidently the preceding inequality is satisfied for n D 2 but not for n D 1. We

conclude that a Taylor polynomial of degree 2 suffices for the approximation.
To approximate log 1:5, we have
1
jRn .0:5/j < :
.n C 1/2nC1
We seek a value of n such that
.n C 1/2nC1 1000:
For n D 6,
7.27 / D 896;
and for n D 7,
8.28 / D 2048:
A Taylor polynomial of degree 7 thus suffices to make this approximation. 4

A function f 2 C 1 .I / is said to have a zero of order k at a number a 2 I

if f .j / .a/ D 0 for j D 0; 1; : : : ; k 1 and f .k/ .a/ 0. If f has a zero of order
k > 0 at a, then Taylors theorem implies that
f .k/ .a/ f .kC1/ ./

f .x/ D .x a/k C .x a/kC1
k .k C 1/
D .x a/k F .x/ (8.3)
for all x 2 I fag, where is a number between a and x and
f .k/ .a/ f .kC1/ ./

F .x/ D C .x a/
k .k C 1/
for all x 2 I . Thus
f .k/ .a/
F .a/ D 0:
k
Conversely, if f .x/ is given by Eq. (8.3) for all x 2 I fag and F .a/ 0, then f
has a zero of order k at a.
Example 8.1.5. Let f .x/ D sin x for all x and let a D 0. Then f .0/ D sin 0 D 0
and f 0 .0/ D cos 0 D 1. The function f thus has a zero of order 1 at 0. From the
definition of sin x we know that
1
X .1/j x 2j C1
sin x D
j D0
.2j C 1/
X1
.1/j x 2j
Dx
j D0
.2j C 1/
D xF .x/;
where
X1
.1/j x 2j
F .x/ D
j D0
.2j C 1/
for all x. Note that F .0/ D 1 0. 4

Taylors theorem provides an alternative perspective on lHpitals rule. Suppose
that f and g are functions that are smooth over an interval I and have zeros of order
m > 0 and n > 0, respectively, at a number a 2 I . Let
f .x/
h.x/ D
g.x/
for all x 2 I fag for which g.x/ 0. Now, limx!a h.x/ is of the indeterminate
form 0/0, and lHpitals rule can be invoked to evaluate the limit, if it exists.
Equation (8.3), however, shows that the functions f and g can be expressed as
f .x/ D .x a/m F .x/
and
g.x/ D .x a/n G.x/;
where F and G are continuous functions such that
f .m/ .a/
F .a/ D 0
m
and
g .n/ .a/
G.a/ D 0:
n
Thus
F .x/
h.x/ D .x a/mn :
G.x/
Evidently
F .x/ F .a/ nf .m/ .a/

lim D D 0:
x!a G.x/ G.a/ mg .n/ .a/
We thus have
(
0 if m > n;
lim h.x/ D f .m/ .a/
x!a
g .m/ .a/
if m D n:
If n > m, it is clear that jh.x/j ! 1 as x ! a.

In practice, this approach to evaluating limits is not notably shorter than using
lHpitals rule. It does, however, indicate how many applications of lHpitals
rule are needed, and in the case where the Taylor polynomials are known, it can
prove a shorter calculation.
In determining the order of a zero it is useful to observe that if f and g are
smooth functions that have zeros of order m > 0 and n > 0, respectively, at a then
the function p D fg can be expressed as
p.x/ D .x a/mCn F .x/G.x/;

where
f .m/ .a/ g .n/ .a/
F .a/G.a/ D 0I
m n
consequently, p has a zero of order m C n at a.
x sin2 x
lim :
x!0 1 cos x
Solution. Let f .x/ D x sin2 x for all x and let g.x/ D 1 cos x for all x such
that cos x 1. We know from Example 8.1.5 that sin x has a zero of order 1 at 0;
consequently sin2 x has a zero of order 2 at 0 and therefore f has a zero of order 3
at 0. Now g.0/ D 1 cos 0 D 0, g 0 .0/ D sin 0 D 0, and g 00 .0/ D cos 0 D 1. The
function g thus has a zero of order 2 at 0. Since 3 > 2,
x sin2 x
lim D 0:
x!0 1 cos x
4
x.1 e x /
lim :
x!0 1 cos x
Solution. Let f .x/ D x.1 e x / for all x and let g.x/ D 1 cos x for all x for
which cos x 1. It can be readily verified that 1 e x has a zero of order 1 at 0,
and therefore f must have a zero of order 2 at 0. We know from Example 8.1.6
that g has a zero of order 2 at 0. In this case m D n D 2, so that the limit exists
and is nonzero. We also know that two applications of lHpitals rule are required
to evaluate this limit. The limit can also be found by using Taylor polynomials. In
detail,
0 1
X1 j
x
f .x/ D x @1 A
j D0
j
0 1
1
X x j 1
D x @1 1 x x A
j D2
j
D x 2 .1 C s.x//
and
1
X .1/j x 2j
g.x/ D 1
j D0
.2j /
X1
x2 .1/j x 2j 2
D 11C x2
2 j D2
.2j /

1
D x2 C t .x/ ;
2
where s and t are functions such that limx!0 s.x/ D limx!0 t .x/ D 0:
Consequently
x.1 e x / x 2 .1 C s.x//

lim D lim 2 1 D 2:
x!0 1 cos x x!0 x C t .x/
2
4
If f has a zero of order k > 0 at a, then Eq. (8.3) implies that, near a, the function
f behaves like .x a/k F .a/. For example, if k D 3 and a D 0, then the graph of
f near 0 would be almost the same as that of x 3 F .0/. Thus in a neighborhood of 0
the graph would look like the cubic curve y D Cx3 , where C is a nonzero constant.
The local shape of a curve near a critical point provides a key insight into the nature
of the critical point. Recall that if a smooth function f has a relative extremum at
a, then f 0 .a/ D 0. The nature of the critical point depends on the higher-order
derivatives at a. It is the first nonzero derivative at a that determines whether the
critical point corresponds to a relative extremum. Briefly, suppose g has a critical
point at a. Then a is a zero of f .x/ D g.x/ g.a/ of order k > 1. The graph of
f near a is nearly the same as that of C.x a/k : If k is even, the critical point will
yield a local extremum; if k is odd, then the critical point cannot correspond to a
relative extremum. This simple observation is formalized in the next result.
Theorem 8.1.2. Let I be an open interval containing some number a, and let
n 2 N. Let f be a function such that f .k/ is defined and continuous on I for each
positive integer k n. Suppose that f .k/ .a/ D 0 for all positive integers k < n but
f .n/ .a/ 0.
1. If n is even and f .n/ .a/ > 0, then f .a/ is a relative minimum.
2. If n is even and f .n/ .a/ < 0, then f .a/ is a relative maximum.
3. If n is odd, then f .a/ is not a relative extremum.
Proof. Since f .n/ .a/ 0 and f .n/ is continuous, we may choose I so that
f .n/ .x/f .n/ .a/ > 0

for all x 2 I . Taylors theorem shows that for each fixed x 2 I fag there exists
between x and a such that
f .x/ D Pn1 .x/ C Rn1 .x/

f .n/ ./
D f .a/ C .x a/n :
n
Thus
f .n/ ./f .n/ .a/ > 0:
1. In this case f .n/ ./ > 0. Hence Rn1 .x/ > 0, so that f .x/ > f .a/.
2. The proof is similar in this case.
3. Since n is odd, .x a/n has the same sign as x a. Thus the sign of Rn1 .x/
when x > a is different from its sign when x < a. Consequently f .a/ is not a
relative extremum.
t
u
Example 8.1.8. Let
g.x/ D 1 C x.1 cos x/
for all x. It is readily verified that g has a critical point at 0. Let
f .x/ D g.x/ g.0/ D x.1 cos x/:
Since 1 cos x has a zero of order 2 at 0, f has a zero of order 3 at 0. We conclude

that f , and therefore g, does not have a relative extremum at 0. 4
Example 8.1.9. Let
g.x/ D 1 C x sin x x 2 x 5
for all x. Then g has a critical point at 0. Here f .x/ D x sin x x 2 x 5 for all x,
and the definition of sin x gives
x4
f .x/ D x 5 C x 6 L.x/;
3
where L is some smooth function. This expression shows that f has a zero of order
4 at 0 and so the critical point yields a relative extremum. Since g .4/ .0/ D 4 < 0,
this extremum is a relative maximum. 4
Exercises 8.1.
1. Let f .x/ D e x =x e for all x > 0.
(a) Show that the only relative minimum of f is at e.
(b) Deduce that e > e .
8.2 Taylor Series 411
2. Use the Taylor polynomial of order 6 for sin x about 0 to obtain an approximate
value for sin 0:3 and then use Taylors theorem to prove that this approximation
gives the first seven digits in the decimal expansion of sin 0:3 correctly.
3. Use the Taylor polynomial of order 3 for log x about 1 to obtain an approximate
value for log 1:06 and then use Taylors theorem to prove that this approximation
gives the first four digits in the decimal expansion of log 1:06 correctly.
4. Suppose f is a function that satisfies f .0/ D 1, f 0 .0/ D 0, and
f 00 .x/ C xf .x/ D 0
for all x 2 R. Find the Taylor polynomial of order 9 for f about 0. (The function
f is called an Airy function.)
5. Let f be a function such that f .1/ D 0, f 0 .1/ D 1, f 00 .1/ D 2, and
2x log.x C 2/
f 000 .x/ D
xC1
for all x > 2 satisfying x 1.
(a) Write down the Taylor polynomial of order 3 for f about 1.
(b) Find upper bounds for the error if the Taylor polynomial of order 2 for f
about 1 is used to approximate f .1:2/ and f .0:5/.
8.2 Taylor Series
Let f be a function, in a real variable x, given by

1
X
f .x/ D aj .x a/j (8.4)
j D0
for all x in the interior of the interval of convergence. Thus
f .j / .a/
aj D
j
for all j , as in Sect. 8.1, so that
1
X f .j / .a/
f .x/ D .x a/j : (8.5)
j D0
j
The series on the right-hand side is called the Taylor series for f about a. When
a D 0, it is also called the Maclaurin series for f . Note that the Taylor series about
a is the only possible representation of f as a power series with center a.
One application of a Taylor series gives a proof of the binomial theorem. Recall
that
!
n n
D ;
j j .n j /
where j and n are integers such that 0 j n.

Theorem 8.2.1. For every nonnegative integer n and every x and y,
!
X n
n j nj
.x C y/ D
n
x y :
j D0
j
Proof. For y D 0 we have

! !
X n
n j nj X
n1
n j nj
x y D x y C xn
j D0
j j D0
j
D xn;
as required.
Suppose y D 1, and let
f .x/ D .x C 1/n
for all x. Then f is a polynomial of degree n, and so we may write

1
X
.x C 1/n D aj x j
j D0
for some integers a0 ; a1 ; : : :, where aj D 0 for all j > n. Therefore
f .j / .0/
aj D
j
for each j . But for all j n and all x we have
n
f .j / .x/ D .x C 1/nj ;
.n j /
so that
n
f .j / .0/ D :
.n j /
Hence
!
n n
aj D D ;
j .n j / j
and substitution into the summation yields

!
X n
n
.x C 1/n D xj ;
j D0
j
as required.
In the general case with y 0 we therefore have
n
x
.x C y/n D y n C1
y
!
X n
n x j
Dy n
j D0
j y
!
X n
n j nj
D x y :
j D0
j
t
u
Taking x D y D 1 yields the following corollary.
Corollary 8.2.2. For every nonnegative integer n,
!
X n
n
D 2n :
j D0
j
We now give an example of an infinitely differentiable function f whose Taylor

series converges but to a sum different from f .
Example 8.2.1. Let
f .x/ D e 1=x
2
for all x 0, and let f .0/ D 0. This function is known as Cauchys function. We
show that its Maclaurin series does not converge to f .x/ for any x 0.
First we prove that if n is a positive integer, then
e 1=x
2
lim D 0: (8.6)
x!0 x n
We start with the case where n is even. Suppose therefore that n D 2k, where k is a
positive integer. Then
1
e 1=x
2
2k
D x1=x 2
xn e
and
1 2
lim 2k
D lim e 1=x D 1:
x!0 x x!0
Therefore we may apply lHpitals rule.

At this point we apply induction. For k D 1 we have
1 2
x2 x3 1
lim D lim D lim D 0:
x!0 2
2
x!0 e 1=x
x3
e 1=x 2 x!0 e 1=x 2
Now assume that k > 1 and

1
x 2.k1/
lim 2
D 0:
x!0 e 1=x
Then
1 2k 1
x 2k x 2kC1 x 2.k1/
lim D lim D k lim D k 0 D 0;
x!0 2
2 2
x!0 e 1=x
x3
e 1=x 2 x!0 e 1=x
as required.
If n is odd, we may write n D 2k C 1, where k this time is a nonnegative integer.
We then have
!
e 1=x e 1=x
2 2
lim D lim x 2kC2 D 0 0 D 0

x!0 x 2kC1 x!0 x
since 2k C 2 is even and positive. The proof of Eq. (8.6) is now complete.
Next we show by induction that for every x 0 and every positive integer n we
have
f .n/ .x/ D e 1=x gn .x/;

2
(8.7)
where
X
m
Aj
gn .x/ D
j D1
xj
for some positive integer m and constants A1 ; A2 ; : : : ; Am . For every x 0 we have
2
f 0 .x/ D e 1=x
2
:
x3
Thus f 0 .x/ is of the required form, with m D 3, A1 D A2 D 0, and A3 D 2.

Suppose that the desired result holds for n. Note that
2 1=x 2
gn .x/ C e 1=x gn0 .x/
2
f .nC1/ .x/ D e
x3

2
D e 1=x 0
2
g n .x/ C g n .x/ :
x3
Since
X m
jAj
gn0 .x/ D ;
j D1
x j C1
it follows that
2
gn .x/ C gn0 .x/
x3
X m
2Aj X m
jAj
D C3

j D1
x j
j D1
x j C1
X
mC3
2Aj 3 X .j 1/Aj 1
mC1
D
j D4
xj j D2
xj
X mC1
A1 2A2 2Aj 3 .j 1/Aj 1 2Am1 2Am
D 2
3 C C C mC3 :
x x j D4
xj xj x mC2 x
Hence
X
r
Bj
f .nC1/ .x/ D e 1=x
2
;
j D1
xj
where r D m C 3, B1 D 0, B2 D A1 , B3 D 2A2 , BmC2 D 2Am1 , BmC3 D

2Am , and
Bj D 2Aj 3 .j 1/Aj 1
for all integers j such that 4 j m C 1. The proof that f .n/ .x/ is of the desired
form is now complete.
Now we prove by induction that f .n/ .0/ D 0 for all positive integers n. First
e 1=x
2
0 f .x/ f .0/ f .x/
f .0/ D lim D lim D lim D0
x!0 x0 x!0 x x!0 x
by Eq. (8.6). Suppose that f .r/ .0/ D 0 for some positive integer r. Then, using
Eq. (8.7), for some positive integer m and constants A1 ; A2 ; : : : ; Am , we have
f .r/ .x/
f .rC1/ .0/ D lim
x!0 x
X m
Aj
D lim e 1=x
2
x!0
j D1
x j C1
D 0;
by Eq. (8.6).
We have now completed the proof that f .n/ .0/ D 0 for all positive integers n. We
deduce that the Maclaurin series for f converges to 0 regardless of the value of x.
Since f .x/ 0 for all x 0, it follows that the Maclaurin series for f converges
to f .x/ only when x D 0.
It is shown in complex analysis that functions that can be represented by a Taylor
series about a given point are precisely those that are differentiable throughout some
neighborhood of that point. If we consider the function f of this example as a
function in the complex plane, then it is not differentiable at 0. Indeed it is not even
continuous at 0: If we take
1
zn D p
n i
for all n > 0, then fzn g converges to 0, but the sequence ff .zn /g does not converge
to f .0/ D 0 since je 1=z j D je in j D 1:
2 2
4
Exercises 8.2.
1. Let f be a function such that f .n/ .x/ exists for all nonnegative integers n and for
all x in an open interval I containing a number a. Show that if there is a number
M such that jf .n/ .x/j M for all n and for all x 2 I , then f is representable
by a Taylor series about a.
2. Find the Maclaurin series for .1 C x/ for each 2 R and give its radius of
convergence.
8.3 Some Shortcuts for Computing Taylor Series
If a function f has a Taylor series about a number a, then the coefficients of the
series can be determined by evaluating the derivatives of f at a. Finding these
derivatives, however, can prove tedious and awkward. In this section we look at
some shortcuts when the functions involved are closely related to known Taylor
series.
8.3 Some Shortcuts for Computing Taylor Series 417
If f has a Taylor series about a, then the series is unique. If we can establish by
any means that a power series of the correct form represents f in a neighborhood
of a, then the power series must be the Taylor series. Often simple substitutions,
algebraic identities, differentiation, or integration can be used to facilitate the
computation of a Taylor series. The examples that follow illustrate how this is done.
Example 8.3.1. Find the Maclaurin series for cos x 4 .
Solution. For each w 2 R the definition of cos w gives
w2 w4 w6
cos w D 1 C C :
2 4 6
Let w D x 4 . Then
.x 4 /2 .x 4 /4 .x 4 /6
cos x 4 D 1 C C
2 4 6
x8 x 16 x 24
D 1 C C
2 4 6
1
X x 8j
D .1/j : 4
j D0
.2j /
Example 8.3.2. Find the Taylor series for e x about 1.

Solution. For all w 2 R,
w2 w3
ew D 1 C w C C C :
2 3
Let w D x 1. Then
.x 1/2 .x 1/3
e x1 D 1 C .x 1/ C C C ;
2 3
and hence

.x 1/2 .x 1/3
e x D e 1 C .x 1/ C C C
2 3
1
X .x 1/j
De :
j D0
j
4
Example 8.3.3. Find the Taylor series for 1=x about 2.
Solution. For all x 0 we have
1 1 1
D D :
x 2Cx2 2.1 .x2/
2
/
Now for all w such that jwj < 1,
1
D 1 C w C w2 C w3 C I
1w
hence for all x such that
jx 2j
< 1;
2
we have

1 1 x2 .x 2/2 .x 2/3
D 1 C C
x 2 2 22 23
1
X .1/j
D .x 2/j :
j D0
2j C1
4
Example 8.3.4. Find the Maclaurin series for
x
:
.1 C x/2
Solution. For all w such that jwj < 1 we have
X 1
1
D wj ;
1w j D0
and since power series can be differentiated term by term,

1
X 1
jwj 1 D :
j D1
.1 w/2
Let w D x. Then
X 1
1
D .1/j 1 jxj 1 ;
.1 C x/2 j D1
so that
X 1
x
D .1/j 1 jxj :
.1 C x/ 2
j D1
4
Example 8.3.5. Find the Maclaurin series for
log.1 C x 2 /:
Solution. The Maclaurin series can be readily obtained from the Maclaurin series
for log.1 C w/. We derive the series from the geometric series to illustrate the use
of integration.
Integration by substitution yields
Z x
2t
dt D log.1 C x 2 /
0 1 C t2
and for each x such that jxj < 1, a simple substitution in the geometric series gives
X 1
1
D .1/j x 2j ;
1Cx 2
j D0
so that
X1
2x
D 2 .1/j x 2j C1 :
1 C x2 j D0
We know that power series can be integrated term by term within the interval of
convergence. We thus have
Z x
2t
log.1 C x 2 / D dt
0 1Ct
2
X1 Z x
D2 .1/j t 2j C1 dt
j D0 0
X1
.1/j 2j C2
D2 x
j D0
2j C 2
1
X .1/j
D x 2j C2 :
j C1 4
j D0
Example 8.3.6. Find the Maclaurin series for cos2 x.

Solution. We know the Maclaurin series for cos x, but it is clear that simply
squaring this series involves a multiplication of infinite series that in itself can prove
awkward. Instead, we exploit the trigonometric identity
1 C cos 2x
cos2 x D :
2
The Maclaurin series for cos 2x can be derived by a simple substitution; hence
1
X .1/j 22j
cos 2x D x 2j :
j D0
.2j /
We thus have
1 1
1 X .1/j 22j 1 2j X .1/j 22j 1 2j
cos x D C
2
x D1C x :
2 j D0 .2j / j D1
.2j /
4
Exercises 8.3.
1. Use the geometric series to determine the Taylor series for the following
functions about the indicated point a:
(a) 2Cx 1
, a D 1; (c) arctan x, a D 0;
(b) 1
x.xC1/
, a D 1; (d) 1
.2Cx/3
, a D 0.
2. Use the Maclaurin series for sin x and cos x along with trigonometric identities
to determine the Taylor series for the following functions about a:
(a) sin x sin 2x, a D 0; (c) cos x, a D 2 ;

(b) 1cos x
x
, a D 0; (d) cos x sin x, a D 0.
3. Show that x 1=3 is representable by a Taylor series about 1 in the interval .0; 2/.
4. Show that cos x is representable by a Taylor series about any real number.
R1
5. Let f .x/ D .sin x/=x for all x 0 and let f .0/ D 1. Estimate 0 f by using
five terms of the Maclaurin series for f , and find an upper bound
R x for the error.
6. For the following functions f , find the Maclaurin series for 0 f .t / dt and give
the intervals of convergence:
(a) sin t 2 ;
(b) e t .
2
7. Find the Maclaurin series for sinh x and cosh x and give the intervals of
convergence.
8. Find the Maclaurin series for sin4 x by expressing the function in terms of sin kx
and cos kx, where k 2 Z.
9. Let
Z x
F .x/ D t t dt:
a
Show that
F 00 .x/ D .1 C log x/F 0 .x/:
Use this result to obtain the Taylor expansion

1 1 1
.x 1/ C .x 1/2 C .x 1/3 C .x 1/4 C : : :
2 3 8
for F about 1.
Chapter 9
The Fixed-Point Problem
9.1 The Fixed-Point Problem
The theory of sequences finds many applications in the study of numerical

techniques for solving equations, since many numerical methods involve the
construction of a sequence of successively better approximations for the desired
solution. In this chapter we show how analysis can help in the study of a numerical
approximation technique for solving nonlinear equations. The work in this chapter
is to a large extent based on the lecture notes of Michael Carter.
Suppose we are given a nonlinear equation, such as
3x 3 3x C 1 D 0: (9.1)
Such an equation can always be written in the form
x D g.x/
for some function g. For example, Eq. (9.1) can be rewritten as
x D 3x 3 2x C 1 (9.2)
or
3x 3 C 1
xD ;
3
and there are many other possibilities.

424 9 The Fixed-Point Problem
A number x satisfying the equation x D g.x/ is called a fixed point of the

function g because an application of g to x leaves x unchanged. For instance, the
function given by x 2 for all x has the two fixed points 0 and 1. Evidently if we can
find a good technique for determining the fixed points of a function, then we will
have a good technique for solving equations, for if an equation is written in the form
x D g.x/, then its solutions are precisely the fixed points of g.
9.2 Existence of Fixed Points
The first question to be considered is whether a given function has any fixed points
at all. Once we have that information, we can consider how the fixed points may
be found. We can gain considerable insight by looking at the problem graphically,
because a fixed point of a function g is simply a value of the argument x at which
the graph of g intersects the line y D x (see Fig. 9.1).
Clearly, a function need not have a fixed point. The exponential function, whose
graph is drawn in Fig. 9.2, is an example of a function with no fixed point.
The next theorem establishes a useful set of conditions under which we can be
sure that a given function has a fixed point.
Theorem 9.2.1. Let a and b be real numbers with a < b. Let g be a function such
that g.a/ a and g.b/ b and g is continuous on a; b. Then g has a fixed point
in a; b.
y
y =x
g(x2) = x2
x1
x
x2
g(x1) = x1
y = g(x)
Fig. 9.1 Graph of a function g with two fixed points

9.2 Existence of Fixed Points 425
Fig. 9.2 The exponential y

function has no fixed points
y = ex
y= x
Proof. Define
h.x/ D x g.x/
for all x 2 a; b. If either h.a/ D 0 or h.b/ D 0, then a or b, respectively, is

a fixed point of g. We may suppose, therefore, that g.a/ > a and g.b/ < b and
consequently that h.a/ < 0 < h.b/. Then the intermediate-value theorem shows
that there is a number 2 .a; b/ such that h./ D 0. Thus is a fixed point of g. u
t
Under the conditions of Theorem 9.2.1 the function g may have many fixed
points, for its graph may touch the line y D x several times. It is often important
to know that g has only one fixed point in a particular interval. Geometric intuition
suggests that this will be the case if g.x/ does not vary too rapidly as x changes, for
then the graph of g will not oscillate rapidly enough to cross the line y D x more
than once. It therefore seems that one way of guaranteeing uniqueness of the fixed
point in a particular interval is to restrict the size of g 0 . The next theorem makes this
idea precise.
Theorem 9.2.2. If a function g satisfies the hypotheses of Theorem 9.2.1 and in
addition jg 0 .x/j < 1 for all x 2 .a; b/, then g has a unique fixed point in a; b.
Proof. The existence of a fixed point of g in a; b follows from Theorem 9.2.1.
Suppose that x1 and x2 are fixed points of g in a; b such that x1 < x2 . By the
mean-value theorem applied to the interval x1 ; x2 , there exists 2 .x1 ; x2 / such
that
g.x2 / g.x1 /
g 0 ./ D :
x2 x1
As x1 and x2 are fixed points, it follows that g 0 ./ D 1, despite the hypothesis
that jg 0 .x/j < 1 for all x 2 .a; b/. This contradiction shows that the fixed point in
question is indeed unique. t
u
9.3 Fixed-Point Iteration
We suppose now that we are given a function g and we wish to find, as accurately
as may be required, the fixed points of g. The method we shall use is called fixed-
point iteration. It involves making an initial guess, x0 say, and then constructing
a sequence fxn g of successive approximations for the desired fixed point, using the
formula
xnC1 D g.xn /
for all nonnegative integers n. The process is illustrated graphically in Fig. 9.3.
If the sequence fxn g so constructed converges to some number s and g is
continuous, then it is easily seen that s is a fixed point of g. Indeed, since xn ! s
as n ! 1 and g is continuous, it follows that g.xn / ! g.s/. Therefore, taking
limits on both sides of the equation xnC1 D g.xn / yields s D g.s/, as required.
However, the sequence fxn g need not converge. The following example illustrates
this and other difficulties.
Example 9.3.1. Suppose we wish to solve Eq. (9.1) given at the beginning of this
chapter. First we sketch the graph of the function on the left-hand side of the
equation. For instance, if we draw the graph of 3x 3 3x and move it one unit
up, then Fig. 9.4 makes it appear that there are solutions near 1:1, 0.4, and 0.7.
Now let us try to solve the equation by writing it in the form (9.2) and looking
for the fixed points of the corresponding function given by
g.x/ D 3x 3 2x C 1
for all x. We show graphically in Fig. 9.5 the fates of seven iterations that start with
initial guesses near one of the three values obtained above.
The following facts should be clear from the figure.
y y
y=x y=x
y = g(x)
g(x0) = x1 g(x0) = x1
g(x2) = x3 g(x0) = x2
g(x1)= x2
y = g(x)
x0 x2 s x3x1 x s x2 x1 x0 x
fixed point of g fixed point of g
Fig. 9.3 Illustration of fixed-point iteration

9.3 Fixed-Point Iteration 427
Fig. 9.4 Graph of y

3x 3 3x C 1 y = 3x3 3x+1
y = 3x3 3x
0.58
x
1 1
1.15
1. No iteration can converge to the fixed points located near 1:1 and 0.7.
2. Iterations such as 35, which start close enough to the remaining fixed point, will
converge to that fixed point. The value obtained for the fixed point by this means
is 0.3949 (correct to four decimal places). A closer study of the graph shows that
in order for an iteration to converge to this fixed point, the initial guess must lie
in the interval 0:14; 0:74, approximately. Even then, the convergence is so slow
that it takes many steps of an iteration to get a good approximation for the fixed
point.
3. Some iterations that start near 1:1, such as iteration 7, which might be expected
to diverge like iteration 2, actually converge to 0.3949. . . because the third term
in the iteration falls in the interval 0:14; 0:74 referred to in (2), instead of
overshooting or undershooting it. The figure shows how this happens. 4
It is clear from this example that trying to solve Eq. (9.1) by applying fixed-point
iteration to version (9.2) is an approach whose value is very limited. Only one of the
three solutions can be found this way, and that at the cost of considerable calculation
because of the slow convergence of the iterations. Furthermore, the interval within
which the initial guess must be placed in order for us to be sure the iterations will
converge is not large. It would be helpful to know what sorts of conditions the
function g must satisfy if the fixed-point iterations are to converge and to have a
better understanding of what governs the rate of convergence of the iterations. The
next theorem deals with these questions. As well as giving conditions under which
a fixed-point iteration will converge to a fixed point of g, it also gives a bound on
the error incurred by stopping the iterations after a given number of steps.
Theorem 9.3.1. Suppose the function g satisfies the following conditions:
1. g is continuous on the closed interval a; b;
2. a g.x/ b for all x 2 a; b; and
3. there is a number L < 1 such that jg 0 .x/j L for all x 2 .a; b/.
2 6
y = 3x3 2x + 1
7
7 2
1
x
6 6
7 7
3 4 5 6
Fig. 9.5 Iterations for fixed points of 3x 3 2x C 1
Then
1. for any initial value x0 2 a; b, the sequence fxn g defined by xnC1 D g.xn /
for each n 0 converges to a number s, which is the unique fixed point of g in
a; b; and
2. the error en D s xn satisfies
Ln
jen j jx1 x0 j
1L
for each n 2 N.
Proof.
1. Taking x D a and x D b in hypothesis (2) gives g.a/ a and g.b/ b, respec-
tively, so that g satisfies the hypotheses of Theorem 9.2.1. By hypothesis (3), g
also satisfies the conditions of Theorem 9.2.2, and so we know that g has exactly
one fixed point s in a; b. Choose any x0 2 a; b, and let the sequence fxn g be
defined by
xnC1 D g.xn /
for each n 0. Then, by hypothesis (2), xn 2 a; b for all n. We may also

assume that xn s for all n. The mean-value theorem shows that for every
positive integer n there is a number n between xn1 and s such that
g.xn1 / g.s/
g 0 .n / D :
xn1 s
Thus
xn s D g 0 .n /.xn1 s/;
so that
jxn sj Ljxn1 sj
by hypothesis (3). In particular,
jx1 sj Ljx0 sj;
and if
jxk sj Lk jx0 sj
for some k > 0, then
jxkC1 sj Ljxk sj LkC1 jx0 sj:
Consequently
jxn sj Ln jx0 sj
for all n 2 N, by induction. Since 0 L < 1, we have
lim Ln D 0;
n!1
and so
lim jxn sj D 0:
n!1
It follows that
lim xn D s;
n!1
as required.
2. We may assume that xn xn1 for all n 2 N, as xn D s otherwise. By the
mean-value theorem, for every positive integer n there is a number n between
xn1 and xn such that
g.xn / g.xn1 /
g 0 .n / D :
xn xn1
Arguing as in the proof of (1), we obtain
jxnC1 xn j D jg 0 .n /.xn xn1 /j Ljxn xn1 j:
We conclude by induction that
jxnC1 xn j Ln jx1 x0 j
for all n 2 N.
Now fix n and let m be an integer such that m > n. Using the telescoping
property, we have

m1
X
jxm xn j D .xj C1 xj /
j Dn
X
m1
jxj C1 xj j
j Dn
X
m1
jx1 x0 j Lj
j Dn
0 1
X
m1 X
n1
D jx1 x0 j @ Lj Lj A
j D0 j D0

1L 1 Ln
m
D jx1 x0 j
1L 1L
Ln Lm
D jx1 x0 j ;
1L
since 0 L < 1. Now limm!1 xm D s, as proved in part (1), and so
Ln
jen j D js xn j jx1 x0 j;
1L
as required. t
u
Example 9.3.2. Let us return to Eq. (9.1) discussed in the previous example, but
this time we shall write it in the form
1=3
3x 1 1 1=3
xD D x
3 3
and look for fixed points of the function g given by

1 1=3
g.x/ D x
3
for all x, using the information provided by Theorem 9.3.1. In order to apply the
theorem we need to know for what values of x we have jg 0 .x/j < 1, and so we first
sketch the graph of

1 1 2=3
g 0 .x/ D x :
3 3
This graph is given in Fig. 9.6.

We can see from the graph that jg 0 .x/j < 1 if x < 0:1409 : : : or x > 0:5258 : : :.
Now we must look at the graph of g to find suitable intervals a; b satisfying
the condition that a g.x/ b for all x 2 a; b as well as the condition that
jg 0 .x/j L for some L < 1 and all x 2 a; b. We can recognize intervals a; b
0.8046
3( )
y = 1 x 1 2/3
3
0.5540
0.2752
0.2403
x
1.3 1 0.1409 1 0.6 0.8
3
0.5258
Fig. 9.6 Graph of .1=3/.x 1=3/2=3

Fig. 9.7 An interval a; b y

where a g.x/ b y =x
b
y = g(x)
x
a b
y= x
y = (x 13)1/3
0.8
0.6
1.3 1
x
0.6 0.8
fixed point at x = 0.3949...

1 already known
1.3
Fig. 9.8 Graph of .x 1=3/1=3
where a g.x/ b for all x 2 a; b by noting that in such a case the graph of g
on the interval a; b must lie entirely within the square box shown in Fig. 9.7. The
graph of g is drawn in Fig. 9.8.
It is clear that the fixed point 0.3949 . . . obtained by the method of the previous
example lies in a region of the graph where g 0 .x/ > 1, so that Theorem 9.3.1 cannot
be applied to locate that fixed point in the present case. In fact, graphical analysis or
numerical experimentation indicates that in the present case no fixed-point iteration
will converge to the fixed point 0.3949 . . . , and so that fixed point cannot be located
by the present method. In fact, iterations started just above 0.3949 . . . will converge
to the fixed point near 0.7, while those started just below 0.3949 . . . will converge to
the fixed point near 1:1. However, both the fixed point near 1:1 and that near 0.7
can be enclosed in intervals on which Theorem 9.3.1 may be applied, for example,
the intervals 1:3; 1 and 0:6; 0:8, as shown. Analysis of the graph of g 0 shows
that jg 0 .x/j 0:276 for all x 2 1:3; 1 and jg 0 .x/j 0:805 for all x 2 0:6; 0:8,
so that we may apply Theorem 9.3.1 on these intervals with L D 0:276 and L D
0:805, respectively. This method therefore locates the two solutions of Eq. (9.1) not
found by the method of Example 9.3.1. By combining the two methods, we can say
that, correct to four decimal places, the solutions of Eq. (9.1) are 1:1372, 0.3949,
and 0.7422. 4
Considering the amount of work involved in obtaining the solutions in this
example, one cannot help feeling that there must be an easier way. One point that
will be noticed is the relatively rapid convergence of the iterations in the interval
1:3; 1. Note that jg 0 .x/j is smaller in this interval, so that L is smaller and
therefore Ln tends to 0 relatively fast as n ! 1. Thus the error bound given
by Theorem 9.3.1 will also tend to 0 relatively quickly. This observation suggests
more rapid convergence of the iterations than in the case of iterations in the interval
0:6; 0:8. Thus one way of speeding up the process might be to seek functions g for
which jg 0 .x/j is very small near a fixed point of g. This idea is the reason for the
success of the following method.
Suppose we are given the nonlinear equation f .x/ D 0 for all x, and we wish to
write it in the form x D g.x/ in order to apply fixed-point iteration. One approach
is to write f .x/ D 0 in the equivalent form
f .x/h.x/ D 0;
where h is any differentiable function with the property that the equation h.x/ D 0
has no real solutions, so that f .x/ D 0 if and only if f .x/h.x/ D 0. Then we can
rewrite f .x/ D 0 as x D x C f .x/h.x/, so that
g.x/ D x C f .x/h.x/
for all x.
Now let s be a fixed point of g, that is, a solution of f .x/ D 0. In view of our
previous discussion, we should like jg 0 .x/j to be small when x is near s. If, in fact,
g 0 .s/ D 0, then as long as g 0 is continuous, we can be sure that jg 0 .x/j will be small
if x is close enough to s. Now
g 0 .x/ D 1 C f .x/h0 .x/ C h.x/f 0 .x/
for all x, and since f .s/ D 0 this equation gives
g 0 .s/ D 1 C h.s/f 0 .s/:
Therefore g 0 .s/ D 0 if f 0 .s/ 0 and h.s/ D 1=f 0 .s/. Thus we define

1
h.x/ D 0 .x/
f
for all x such that f 0 .x/ 0, for then the value of s does not need to be known. In
other words, we write the equation f .x/ D 0 in the equivalent form
f .x/
xDx
f 0 .x/
and then use fixed-point iteration. This procedure is called Newtons method. If
the fixed-point iteration starts close enough to a solution of f .x/ D 0, then L
in Theorem 9.3.1 will be small and so the iteration will converge rapidly to the
desired solution. Difficulties arise in certain cases, most obviously if f 0 .s/ D 0,
but Newtons method is very useful in cases where f 0 .x/ is easily calculated and a
reasonably good initial guess at the solutions can be made.
Example 9.3.3. Newtons method applied to Eq. (9.1) requires the equation to be
rewritten in the form
3x 3 3x C 1
xDx
9x 2 3
for all x such that 9x 2 3. Applying fixed-point iteration with starting values
near the solutions guessed at the beginning of Example 9.3.1 leads to iterations that
converge relatively rapidly compared with those in the previous examples. 4
Exercises 9.1.
1. Let g.x/ D x 2 for all x. From a graph similar to that used to trace the fates
of fixed-point iterations in Example 9.3.1, determine for what values of x0 the
sequence fxn g, defined by xnC1 D g.xn / for all nonnegative integers n, will
converge to a fixed point of g and for what values of x0 it will diverge.
2. Suppose Eq. (9.1) discussed in Examples 9.3.19.3.3 is rewritten as x D .3x 3 C
1/=3. Putting g.x/ D .3x 3 C1/=3 for all x, sketch the graphs of g and g 0 and use
the approach illustrated in Example 9.3.2 to find suitable intervals within which
Theorem 9.3.1 can be applied to locate fixed points. Use fixed-point iterations,
starting at each end of the intervals you have found, to locate the corresponding
fixed points correct to four decimal places. If there are any fixed points that
cannot be located in this way, use the graph of g to describe what will happen to
iterations that start near these fixed points.
3. Do the same as for the previous exercise but for the function g given by
3x 1
g.x/ D
3x 2
for all x 0. Note that in investigating the graph of g 0 it is not necessary to find
exactly the values of x for which g 0 .x/ D 1; just locate these values roughly
by calculating a few values of g 0 .x/ on either side of the apparent location of
these values of x.
4. Show that Newtons method applied to the equation x 2 D c, where c > 0, leads
to the fixed-point problem
x2 C c
xD ;
2x
and use graphical
p pmethods to show that Theorem p 9.3.1 can be applied on the
interval 23 c; 43 c to locate the fixed point c. Find the smallest possible
p
value of L for this interval. In the case c D 2, use the fact that 1 < 2 < 32
p p p
to show that 1 lies in the interval 23 2; 43 2 and hence locate 2 correct
to six decimal places by fixed-point iteration. Calculate, for each step in the
iteration, the size of the actual error and the theoretical bound on the error given
by Theorem 9.3.1.
5. For the fixed-point problem x D cos x for all x, use graphical methods to show
that Theorem 9.3.1 can be applied on the interval 0; 1. If your calculator can
evaluate trigonometric functions, use the initial value x0 D 0:5 and locate the
fixed point correct to four decimal places. Also, solve the equation by Newtons
method with the same initial value and note the rapid convergence in this case.
6. Suppose
f .x/ D .x r1 /.x r2 / .x rm /
for all x, where m 2 and r1 r2 : : : rm . Then f is a polynomial

of degree m with the coefficient of x m equal to 1, and the equation f .x/ D 0
has only real solutions, none of which exceeds the solution rm . Let fxn g be a
sequence produced by Newtons method applied to the equation f .x/ D 0.
(a) Explain why f .x/ > 0 for all x > rm , and show by direct differentiation
that f 0 .x/ > 0 and f 00 .x/ > 0 for all x > rm .
(b) Suppose that xk > rm for some integer k 0. Show that xkC1 < xk . By
using the mean-value theorem applied to f on the interval rm ; xk , show
also that f .xk / < .xk rm /f 0 .xk / and deduce that xk xkC1 < xk rm
and hence that xkC1 > rm .
(c) From (b) it follows that if x0 > rm , then the sequence fxn g is a decreasing
sequence bounded below by rm . Hence the sequence fxn g converges to a
limit s rm . Prove that in fact s D rm . Illustrate this conclusion for the
function given by
f .x/ D .x C 1/2 .x 2/;
for all x, by taking x0 D 3, x0 D 10, and x0 D 30 and carrying out the

iterations correct to four decimal places.
(d) In some cases the convergence of the sequence fxn g may be exceedingly
slow. For example, if f .x/ D .x s/m for all x, show that

1 s
xnC1 D 1 xn C
m m
for all n 2 N, and deduce that

1
s xnC1 D 1 .s xn /:
m
[Thus if m is large, so that 1 1=m is close to 1, then the decrease in the

error at each step of the iteration is slight. For example, if m D 100, then
enC1 D 99en =100. Note that this is a case where f 0 .s/ D 0, since m 2,
and so we might expect that Newtons method could run into difficulties.]
Chapter 10
Sequences of Functions
10.1 Introduction
The representation of a function as the limit of a sequence or, equivalently, an

infinite series is central to many topics in advanced analysis. In this chapter we
concentrate on sequences of functions and certain properties of limits that can be
gleaned from properties of the sequence terms. We begin with an application that
not only motivates the study of sequences of functions, but also highlights some key
questions regarding limits.
A first-order ordinary differential equation for a function y is an equation of the
form
y 0 D f .x; y/; (10.1)
where f is a given function of a variable x and y. The equation is usually

supplemented with an initial condition
y.a/ D c; (10.2)
where a and c are given numbers. The initial-value problem consists of determin-
ing a function y that satisfies Eqs. (10.1) and (10.2) for all x in some open interval
that contains a.
We gloss over the fundamental questions of the existence and uniqueness of
solutions to initial-value problems. It turns out that for most choices of f the
problem cannot be solved explicitly. Nonetheless, it can be shown, for example,
that if f is differentiable in a neighborhood of .a; c/ with respect to x and y, then

438 10 Sequences of Functions
the initial-value problem has a unique solution. The proof of this result requires
the use of a sequence of functions, and the solution is the limit of this sequence.
Picard proved this result by the method of successive approximations. Specifically,
the problem can be recast as an integral equation
Z x
y.x/ D c C f .; y.// d ;
a
and this formulation motivates the sequence defined by
y0 .x/ D c
and
Z x
ynC1 .x/ D c C f .; yn .// d
a
for all nonnegative integers n. It is then shown that fyn .x/g converges to a limit
y.x/ that solves the initial-value problem. The result is local in character: To
ensure convergence of the sequence, x is usually restricted to a small open interval
containing a.
Convergence questions aside, the claim that the limit is the solution of the
differential equation brings to the fore certain questions concerning the properties of
the function that is the limit of the sequence. The problem is that more than one limit
process is involved, and the order in which these limits are taken must be changed.
To prove that the function defined by
y.x/ D lim yn .x/

n!1
is a solution to the integral equation, we need to justify the following calculation:

Z x
lim yn .x/ D c C lim f .; yn .// d
n!1 n!1 a
Z x
DcC lim f .; yn .// d
a n!1
Z x
DcC f .; lim yn .// d
a n!1
Z x
DcC f .; y.// d :
a
In this calculation, the sequence limit migrates from outside the integral to inside
the integrand. There are two limits involved in this manipulation: the limit defining
y and the limit defining the integral. The problem in the second line is that we
10.1 Introduction 439
must change the order in which the sequence limit and the integral limit are taken.
Intuitive as the calculation seems, it turns out that changing the order of these limits
is not always valid.
Indeed, there is a third limit process in the background that stems from the
continuity of f and the functions yn . Given that f is differentiable with respect to
x and y in some neighborhood N of .a; c/, a fortiori f is continuous with respect
to y (and x) for each .x; y/ 2 N . Certainly, if fwn g is a sequence of numbers such
that .x; wn / 2 N and limn!1 wn D y, then the definition of continuity implies that
lim f .x; wn / D f .x; lim wn / D f .x; y/:

n!1 n!1
The sequence fyn g consists of functions continuous near a, but is the limit function
y continuous near a? For that matter, if y is not continuous, is f .x; y.x// integrable
with respect to x in some neighborhood of a?
Suppose that the problems with the calculation above are resolved. The function
y thus represents the solution to the integral equation. The original problem,
however, involved a differential equation. The integral equation is well defined for
every continuous function y; the differential equation requires y to be differentiable
near a. It is clear from the definition of the sequence that each yn is differentiable,
but is the limit y differentiable?
The initial-value problem highlights the need to examine conditions under which
the order of certain limiting processes can be changed. On a more fundamental
level it also raises questions as to whether properties of the sequence terms, such as
continuity and differentiability, are preserved in the limit.
Let I R be an interval and suppose the sequence ffn .x/g converges to f .x/
for all x 2 I . We have, in summary, the following questions concerning limits of
sequences of functions.
1. If fn is continuous on I for all n, is f continuous on I ?
2. If fn is integrable on I for all n, is f integrable on I ? If so, is
Z b Z b
lim fn .x/ dx D lim fn .x/ dx;
a n!1 n!1 a
where I D a; b?
3. If fn is differentiable on I for all n, is f differentiable on I ? If so, is
0
lim fn D lim fn0
n!1 n!1
In this chapter we develop conditions under which the order of limit processes can
be changed. We end this section with a few simple examples to illustrate that, in
general, the order of these processes cannot be changed.
Example 10.1.1 (Continuity). Let ffn g be the sequence defined by
1
fn .x/ D
1 C nx2
for all n and all x 2 1; 1. For every n, fn is continuous on 1; 1. If x 0, then
it is plain that fn .x/ ! 0 as n ! 1; however, fn .0/ D 1 for all n and consequently
fn .0/ ! 1 as n ! 1. The limit of the sequence is thus given by

0 if x 2 1; 1 f0g
f .x/ D
1 if x D 0;
hence f is not continuous on 1; 1. Note that

lim lim fn .x/ D 0 lim lim fn .x/ D 1:
x!0 n!1 n!1 x!0
4
Example 10.1.2 (Differentiation). Let ffn g be the sequence defined by
x
fn .x/ D
1 C nx2
for all n and all x 2 0; 1. Then
f .x/ D lim fn .x/ D 0

n!1
for all x 2 0; 1, so that f 0 .x/ D 0 for all such x. However,
1 nx2
fn0 .x/ D ;
.1 C nx2 /2
so that

1 if x D 0;
lim fn0 .x/ D
n!1 0 otherwise.
Hence
0
lim fn0 .x/ lim fn .x/
n!1 n!1
when x D 0. 4
10.2 Uniform Convergence 441
Example 10.1.3 (Differentiation). Let ffn g be the sequence defined by
sin nx
fn .x/ D
n
for all n 2 N and x 2 R. Now, j sin nxj 1; consequently fn .x/ ! 0 as n ! 1

for all x 2 R.
For every n 2 N we see that fn is differentiable on R. The limit f D 0 is also a
differentiable function on R. The derivative of fn , however, is defined by
fn0 .x/ D cos nx;
so that the sequence ffn0 g converges only for special values of x such as 0 and 2
and diverges for most x 2 R. 4
Example 10.1.4 (Integration). Let ffn g be the sequence defined by
fn .x/ D 2nxenx
2
for all n and all x 2 0; 1. For each x 2 0; 1, fn .x/ ! 0 as n ! 1; hence
Z 1 Z 1
lim fn .x/ dx D 0 dx D 0:
0 n!1 0
On the other hand,

Z 1
fn .x/ dx D 1 e n ;
0
so that
Z 1 Z 1
lim fn .x/ dx D 1 lim fn .x/ dx D 0:
n!1 0 0 n!1
10.2 Uniform Convergence
The examples in the previous section show that the order in which limits are taken is
important. We thus seek sufficient conditions under which this order can be changed.
In this section we present the key concept of uniform convergence. In the next
section we show that the order of limits for sequences that converge uniformly can
be changed.
Let I be a set of real numbers and let ffn g be a sequence of functions defined
on I . Suppose that for each x 2 I the sequence ffn .x/g converges to f .x/. For all
x 2 I , the definition of convergence implies that for every " > 0 there is an integer
N such that
jfn .x/ f .x/j < "
whenever n N . The value of N , in general, depends not only on the choice of "
but also on x. Suppose that we consider two distinct values x1 and x2 in I . Since
ffn .x1 /g and ffn .x2 /g are convergent sequences, for every " > 0 there exist integers
N."; x1 / and N."; x2 / such that
jfn .x1 / f .x1 /j < " (10.3)
whenever n N."; x1 / and
jfn .x2 / f .x2 /j < " (10.4)
whenever n N."; x2 /. Integers N."; x1 / and N."; x2 / are not necessarily equal,
but we could use N D maxfN."; x1 /; N."; x2 /g to ensure that inequalities (10.3)
and (10.4) are satisfied for all n N . Evidently, for any finite number of points
x1 ; x2 ; : : : ; xj in I we can always choose N so that if n N , then
jfn .xk / f .xk /j < "
for each k 2 f1; 2; : : : ; j g. It is not clear, however, that we can find an N such that,
for all x 2 I ,
jfn .x/ f .x/j < "
whenever n N . Generically, it is not possible to find such an N , and this situation

leads to the concept of uniform convergence.
Let ffn g be a sequence of functions defined on a set I R. The sequence is
said to converge uniformly to the function f on I if for each " > 0 there exists
an integer N , which may depend on " but not on any particular x 2 I , such that if
n N , then
jfn .x/ f .x/j < " (10.5)
for all x 2 I . Clearly, " in inequality (10.5) may be replaced by c" for any constant
c > 0. In applications we usually take I to be a closed interval. We always assume
it to be nonempty.
Note that if there exists a function f such that fn .x/ D f .x/ for each x 2 I and
each n, then the sequence ffn g converges uniformly to f on I .
Suppose that fn .x/ D an for each x 2 I and each n, and that the sequence fan g
converges to L. Choose " > 0. There exists N such that
jfn .x/ Lj D jan Lj < "
for all n N . Therefore the sequence ffn g of functions converges uniformly on I

to the constant L.
The geometric interpretation of uniform convergence is straightforward. If the
graph of f is drawn for all x 2 I , then a ribbon of width 2" can be constructed by
curves offset a distance of " from the graph of f . If ffn g converges uniformly to f
on I , then for each " > 0 there is an N such that the graph of fn lies in the ribbon
for all n N . The quantity limiting N is the maximum difference between fn .x/
and f .x/ for x 2 I . This observation motivates the investigation of the sequence
fMn g of numbers defined by
Mn D sup jfn .x/ f .x/j:

x2I
Inequality (10.5) shows that Mn necessarily exists for all n if ffn g is uniformly
convergent on I .
Theorem 10.2.1. The sequence ffn g converges uniformly on I to f if and only if
Mn ! 0 as n ! 1.
Proof. Suppose that ffn g converges uniformly on I to f . Choose " > 0. Since ffn g
converges uniformly, there is an integer N that is independent of x such that
jfn .x/ f .x/j < "
for all n N ; hence Mn exists and
jMn j D Mn D sup jfn .x/ f .x/j "

x2I
for all n N . Thus Mn ! 0 as n ! 1 from the definition of convergence.

Suppose that Mn ! 0 as n ! 1. For each " > 0 there is an N such that Mn < "
whenever n N ; therefore, for all x 2 I ,
jfn .x/ f .x/j Mn < "
for each n N . The choice of N is independent of x and we thus conclude that

ffn g converges uniformly to f on I . t
u
Corollary 10.2.2. Suppose that ffn g converges uniformly on I to f . If M is a
number such that jfn .x/j < M for all n and all x 2 I , then jf .x/j M for all
x 2 I.
Proof. Suppose that jf .x/j > M for some x 2 I . Then jf .x/j D M C " for some
" > 0. Thus for all n we have
jfn .x/ f .x/j jf .x/j jfn .x/j > M C " M D ":
Therefore
sup jfn .x/ f .x/j > "

x2I
for all n, and so we have the contradiction that
lim sup jfn .x/ f .x/j " > 0:

n!1 x2I
t
u
Of course, a corresponding result holds if jfn .x/j > M for all n and all x 2 I .
Example 10.2.1. Let ffn g be the sequence defined by
1 x nC1
fn .x/ D
1x
for each x 2 1=2; 1=2 D I . For all x 2 I , x nC1 ! 0 as n ! 1, and therefore
1
f .x/ D lim fn .x/ D :
n!1 1x
Now
jxjnC1
jfn .x/ f .x/j D I
1x
consequently
Mn D sup jfn .x/ f .x/j

x2I
jxjnC1
D sup
x2I 1x
1
D :
2n
Since Mn ! 0 as n ! 1, ffn g converges uniformly to f on I by Theorem 10.2.1.

4
1
fn .x/ D
1 C nx2
for all x 2 1; 1 D I . Example 10.1.1 shows that fn ! f as n ! 1, where

0 if x 2 I f0g;
f .x/ D
1 if x D 0;
therefore
(
1
if x 2 I f0g;
jfn .x/ f .x/j D 1Cnx2
0 if x D 0.
Now
Mn D sup jfn .x/ f .x/j D 1;

x2I
and it is clear that limn!1 Mn 0 as n ! 1. Theorem 10.2.1 thus implies that

the sequence does not converge uniformly to f on I . 4
Remark. A sequence ffn g of functions fails to be uniformly convergent to a function
f on a set I if and only if there is an " > 0 such that for all N there exist an integer
k N and an xk 2 I for which
jfk .xk / f .xk /j ":
We illustrate this remark in the next example.

x
fn .x/ D
n
for all n 2 N and all x 2 R. Since
f .x/ D lim fn .x/ D 0

n!1
for all x 2 R, the sequence converges to a continuous function. However, it is not

uniformly convergent: For all n 2 N let xn D n, so that fn .xn / D 1 and hence
jfn .xn / f .xn /j D 1 > 0:

Note also that

1 0
lim fn0 .x/ D lim D 0 D lim fn .x/
n!1 n!1 n n!1
and
Z Z
b
b 2 a2 b
lim fn .x/ dx D lim D0D lim fn .x/ dx:
n!1 a n!1 2n a n!1
4
Our next example shows that a sequence of functions may converge uniformly
on every closed subinterval of an open interval yet fail to be uniformly convergent
on the open interval.
Example 10.2.4. Let
fn .x/ D x n
for each n and each x 2 .0; 1/. We will show that if 0 < a < b < 1, then ffn g
converges uniformly on a; b, but it does not do so on .0; 1/.
For each x 2 .0; 1/ we have
f .x/ D lim fn .x/ D 0:

n!1
Therefore
sup jfn .x/ f .x/j D sup jfn .x/j D sup x n D b n ! 0

x2a;b x2a;b x2a;b
as n ! 1. Hence ffn g converges uniformly on a; b.

For each n > 0 let
1
xn D 1 :
n
Then

1 n 1
jfn .xn / f .xn /j D 1 !
n e
as n ! 1. Hence
1
sup jfn .x/ f .x/j > 0:
x2I e
We conclude from Theorem 10.2.1 that ffn g does not converge uniformly on .0; 1/.
4
Theorem 10.2.1 is a useful characterization of uniform convergence, but one must

find the supremum of jfn .x/ f .x/j. The problem can sometimes be mitigated by
using the sandwich theorem (Sect. 2.5). If there is a sequence fKn g such that
jfn .x/ f .x/j < Kn
for each n 2 N and each x 2 I , then 0 Mn Kn for each n. If Kn ! 0

as n ! 1, then Mn ! 0 as n ! 1. To show that a sequence is not uniformly
convergent on I , it suffices to establish a nonzero lower bound for the sequence
fMn g, valid for all n sufficiently large.
A more serious problem with this characterization of uniform convergence is that
it requires a candidate for f . Often there is not an obvious candidate and hence we
cannot use Theorem 10.2.1 directly. We thus seek an alternative characterization
that does not require a limit candidate. This line of thought leads to a generalization
of the Cauchy principle for convergence (Theorem 2.6.9).
Theorem 10.2.3 (Cauchy Principle). Let ffn g be a sequence of functions defined
on the set I . The sequence ffn g converges uniformly on I if and only if for each
" > 0 there is an integer N , which may depend on " and I but not on any x 2 I ,
such that for all x 2 I we have
jfn .x/ fm .x/j < "
whenever n N and m N .
Proof. Necessity: Suppose ffn g converges uniformly on I to f . Then for each " > 0
there is an integer N such that for all x 2 I we have
jfn .x/ f .x/j < "
whenever n N . If n N and m N , then, for all x 2 I ,
jfn .x/ fm .x/j D jfn .x/ f .x/j C jf .x/ fm .x/j

< 2":
The result follows.

Sufficiency: Suppose that for each " > 0 there is an N such that for all x 2 I we
have
jfn .x/ fm .x/j < "
whenever n N and m N . For every x 2 I , ffn .x/g is a Cauchy sequence of

numbers and therefore convergent. Define
f .x/ D lim fn .x/

n!1
for all x 2 I .
Choose any " > 0 and x 2 I . There is an N , independent of x, such that
jfn .x/ f .x/j jfn .x/ fm .x/j C jfm .x/ f .x/j

< " C jfm .x/ f .x/j
whenever m N and n N . Since fm .x/ ! f .x/ as m ! 1, we can choose

m N (m may depend on " and x) so that
jfm .x/ f .x/j < ":
Therefore
jfn .x/ f .x/j < " C " D 2"
whenever n N . This inequality is valid for all x 2 I , and N is independent of x.

We thus conclude that ffn g converges uniformly to f on I . t
u
Note the similarity between the proof of necessity in Theorem 10.2.3 and the
proof of Theorem 2.6.1.
In order to prove the uniform convergence of a sequence ffn g of Theorem 10.2.3
on an interval I , it is of course enough to prove the existence of a positive constant
c such that
jfn .x/ fm .x/j < c"
whenever n N and m N .
The following result provides a sequential characterization of uniform conver-
gence of a sequence of functions.
Theorem 10.2.4. Let ffn g be a sequence of functions on a nonempty set I . Then
the sequence converges uniformly to f on I if and only if
lim .fn .xn / f .xn // D 0 (10.6)

n!1
for each sequence fxn g in I .

Proof. Suppose that ffn g converges uniformly to f on I . Then for each sequence
fxn g in I we have
0 jfn .xn / f .xn /j sup jfn .x/ f .x/j ! 0

x2I
as n ! 1. Equation (10.6) follows by the sandwich theorem.

Suppose on the other hand that ffn g does not converge uniformly to f . Then
there is an " > 0 such that for all N there exist an integer k N and an xk 2 I for
which
jfk .xk / f .xk /j ":
In particular, there exist k1 1 and xk1 2 I for which
jfk1 .xk1 / f .xk1 /j ":
Moreover, suppose that positive integers k1 ; k2 ; : : : ; kn have been defined for some
n 2 N, and that xkj 2 I and
jfkj .xkj / f .xkj /j "
for all j . Suppose also that kj < kj C1 for each j < n. Then there exist an integer
knC1 kn C 1 and an xknC1 2 I such that
jfknC1 .xknC1 / f .xknC1 /j ":
We have now constructed a subsequence ffkn g of ffn g by induction. Let fxn g be any
sequence in I having fxkn g as a subsequence. For instance, since I ; we may
choose a 2 I and set xn D a for each positive integer n fk1 ; k2 ; : : :g. Then the
subsequence fjfkn .xkn / f .xkn /jg of fjfn .xn / f .xn /jg does not converge to 0,
u
The contrapositive of this theorem is often easier to use.
Corollary 10.2.5. If there exists a sequence fxn g in I such that
lim .fn .xn / f .xn // 0;

n!1
then ffn g does not converge uniformly to f .

Exercises 10.1.
1. For each of the following sequences defined on 0; 1, show that the sequence
is convergent and determine whether the convergence is uniform:
(a) fx
n g.
n
o
(b) sin.nxCn/
p .
x n
(c) 1Cnx .
n 2 o
nx
(d) 1Cnx .

q
(e) x 2 C n1 .
fsin1=n xg
converges uniformly on every closed proper subinterval of 0; but not on

0; itself.
( 1=n )
sin x
x
converges on .0; / but not uniformly.


nx
1 C n2 x 2
is uniformly convergent on c; 1, where 0 < c < 1. Does the sequence
converge uniformly on .0; 1/?
5. Use differentiation to find the maximum value of the function
n2 x
fn .x/ D
1 C n3 x 2
for all n 2 N and x 2 R, and hence show that the sequence ffn g is not
uniformly convergent on 0; 1.
6. Let
n x
fn .x/ D
1 C n x 2
for all n 2 N and x 2 R, where > 0. Show that the sequence ffn g
converges uniformly on 0; 1 if and only if > 2.
7. Let
1 nx
fn .x/ D C nx2
x e
for all n 2 N and x 2 .0; 1. Show that

1
f .x/ D lim fn .x/ D :
n!1 x
Use differentiation to find the maximum value of jfn .x/ f .x/j and hence
show that the sequence ffn g is not uniformly convergent on .0; 1.
8. Let ffn g be a sequence of functions defined on an interval I . Show that if

ffn g converges uniformly to f , then fjfn jg converges uniformly to jf j. Is the
converse true?
9. Let fan g be a convergent sequence and for each n let fn W I ! R be a function.
Suppose there exists N such that
supfjfn .x/ fm .x/jg jan am j

x2I
P
for all m N and n N . Show that 1 j D0 fj .x/ converges uniformly on I .
10. Suppose f W R ! R is uniformly continuous. Let

1
fn .x/ D f xC
n
for all n 2 N and x 2 R. Show that ffn g converges uniformly to f .

11. Let f W a; b ! R be continuous.
(a) Explain why, for each n 2 N, there exists n > 0 such that
1
jf .x/ f .y/j <
n
whenever jx yj < n .
(b) Let .x0 ; x1 ; : : : ; xkn / be a partition of a; b such that xj C1 xj < n for
each j < kn . For each n and each x 2 a; b define

f .xj / if xj x < xj C1
fn .x/ D
f .b/ if x D b.
Show that ffn g converges to f uniformly.

n xo
cos
n
converges uniformly on .a; a/ for every a. Is the convergence uniform on R?


x=n if x is even;
f .x/ D
1=n if x is odd.
Show that ffn g converges on R but not uniformly.

10.3 Properties of Uniformly Convergent Sequences
The examples in Sect. 10.1 illustrate the need for conditions under which it is valid
to change the order in which limits are taken. In this section we show that uniformly
convergent sequences are well behaved in the sense that the order in which limits
are taken is not important. We begin with continuity.
Theorem 10.3.1. Let I R and let ffn g be a sequence of functions that
is uniformly convergent on I . Let c be a limit point of I , and suppose that
limx!c fn .x/ exists for all n 2 N. Then
lim lim fn .x/

n!1 x!c
exists if and only if
lim lim fn .x/

x!c n!1
exists, and in this case those limits are equal.

Proof. As ffn g is uniformly convergent on I , we may define
f .x/ D lim fn .x/

n!1
for each x 2 I .
Suppose first that
lim f .x/ D L;
x!c
and choose " > 0. There exists > 0 such that
jf .x/ Lj < "
whenever x 2 I and 0 < jx cj < . The uniform convergence of ffn g on I shows

the existence of an N for which
jfn .x/ f .x/j < "
whenever n N and x 2 I . Hence
jfn .x/ Lj jfn .x/ f .x/j C jf .x/ Lj < 2"
whenever n N , x 2 I and 0 < jx cj < . Therefore, since limx!c fn .x/ exists

for all n, we have

lim fn .x/ L 2" < 3"
x!c
10.3 Properties of Uniformly Convergent Sequences 453
for all n N . We conclude that
lim lim fn .x/ D L: (10.7)

n!1 x!c
On the other hand, suppose that Eq. (10.7) holds. For each n 2 N, write
gn .c/ D lim fn .x/:

x!c
Thus
lim gn .c/ D L:
n!1
Choose " > 0. There exists M1 such that
jgn .c/ Lj < "
for all n M1 . Moreover, for each such n there exists n > 0 such that
jfn .x/ gn .c/j < "
whenever x 2 I and 0 < jx cj < n , and the uniform convergence of ffn g shows
the existence of M2 such that
jfn .x/ f .x/j < "
whenever n M2 and x 2 I . Fix n maxfM1 ; M2 g. Then
jf .x/ Lj jf .x/ fn .x/j C jfn .x/ gn .c/j C jgn .c/ Lj < 3"
for each x 2 I such that 0 < jx cj < n . Thus
lim f .x/ D L;
x!c
as required. t
u
Remark. The hypothesis that c be a limit point of I is needed only to ensure that
the limits in question are defined.
Corollary 10.3.2. Let ffn g be a sequence of functions that is uniformly convergent
to a function f on a set I R. If fn is continuous on I for each n, then f is
continuous on I .
Proof. It is immediate from the hypotheses that for each c 2 I we have
lim f .x/ D lim lim fn .x/

x!c x!c n!1
D lim lim fn .x/

n!1 x!c
D lim fn .c/
n!1
D f .c/:
t
u
The result above gives a sufficient condition for a sequence of continuous
functions to converge to a continuous function. The next example shows that it is
not a necessary condition.
fn .x/ D nx.1 x/n
for all x 2 0; 1 D I . It is clear that for all x 2 I , we have fn .x/ ! 0 as n ! 1.

The function defined by f .x/ D 0 for all x 2 I is continuous on I and hence the
sequence converges pointwise to a continuous function.
This sequence, however, is not uniformly convergent to f on I . Consider the
sequence fxn g in I defined by
1
xn D
1Cn
for all n 2 N.1 Since

n
n 1 1 1

jfn .xn / f .xn /j D 1 D > >0
1Cn 1Cn 1 nC1
1C n 2e
for all n 2 N, Theorem 10.2.1 shows that the sequence indeed fails to be uniformly
convergent on I . 4
Example 10.1.4 shows that there are sequences ffn g of integrable functions such
that fn .x/ ! f .x/ for all x in an interval I D a; b but
Z b Z b
lim fn .x/ dx f .x/ dx:
n!1 a a
Indeed, it may be that the limit f is not even integrable over I . The next result
shows that if ffn g converges uniformly to f on I , then f is integrable and the order
in which the limits are taken can be changed.
1
The reader may wonder what prompts the choice of this sequence. In fact, it can be shown using
elementary calculus that fn has a global maximum in I at xn .
Theorem 10.3.3. Let ffn g be a sequence of functions that are integrable over the
interval I D a; b and suppose that ffn g converges uniformly to f over I . Then f
is integrable on I , and
Z b Z b
lim fn D f: (10.8)
n!1 a a
Proof. Choose " > 0. Since ffn g converges uniformly to f on I , there exists N
such that
fn .x/ " < f .x/ < fn .x/ C "
for all n N and x 2 I . Taking the lower integrals over I of both sides of the first
inequality, we obtain
Z Z
fn ".b a/ f;
since fn is integrable over I . Similarly,

Z Z Z
f f fn C ".b a/:
Consequently,
Z Z
0 f f 2".b a/;
and since these inequalities must hold for every " > 0, it follows that
Z Z
f D f;
so that f is integrable on I .
We now have
Z Z Z
fn ".b a/ f fn C ".b a/:
Hence
Z Z
".b a/ f fn ".b a/
for all " > 0 and n N , and so

Z Z
lim fn D f:
n!1
t
u
Theorems 7.3.11 and 7.3.14 provide sufficient conditions under which a function
is integrable. These results can be used with Theorem 10.3.3 to glean the following
corollaries.
Corollary 10.3.4. Let ffn g be a sequence of functions that are continuous on the
interval a; b, and suppose that ffn g converges uniformly to f on a; b. Then f is
integrable over a; b and Eq. (10.8) is satisfied.
Corollary 10.3.5. Let ffn g be a sequence of functions that are bounded and
monotonic on the interval a; b. If ffn g converges uniformly to f on a; b, then
f is integrable over a; b and Eq. (10.8) is satisfied.
Example 10.3.2. Consider the sequence ffn g defined by
8
< xn
2 if x 2 0; 1=n/
fn .x/ D x n if x 2 1=n; 2=n/
: n
0 if x 2 2=n; 1
for all n > 0, where > 0 is a fixed number. The functions fn are continuous on
the interval 0; 1 and hence integrable by Theorem 7.3.11.
We show first that fn .x/ ! 0 as n ! 1 for each x 2 0; 1. The result is
obvious if x D 0. Suppose x 2 .0; 1. Let N be any integer such that N > 2=x.
Then for all n N , we have x > 2=n; hence fn .x/ D 0 and thus fn .x/ ! 0 as
n ! 1.
Evidently,
Z 1
fn .x/ dx D n2 ; (10.9)
0
so that if 0 < < 2,

Z 1 Z 1
lim fn .x/ dx D lim fn .x/ dx D 0I (10.10)
n!1 0 0 n!1
however, if D 2,
Z 1 Z 1
lim fn .x/ dx D 1 lim fn .x/ dx;
n!1 0 0 n!1
and the sequence fn2 g diverges if > 2. Now,
sup jfn .x/j D n1 ;

x20;1
and Theorem 10.2.1 implies that ffn g is uniformly convergent to the constant
function 0 on 0; 1 only if < 1. Corollary 10.3.4 can be used to deduce
relation (10.10), whenever 0 < < 1, without calculating the integral of fn . None
of the results concerning integration of uniformly convergent sequences, however,
can be applied when 1 < 2. This example thus shows that uniform convergence
is not a necessary condition for changing the order of the limits. 4
If ffn g is a convergent sequence of functions that are differentiable on an interval
I , Example 10.1.3 shows that the sequence ffn0 .x/g may diverge even for all x 2 I .
It is of interest to examine the relationship between these sequences when uniform
convergence is imposed. Uniform convergence of ffn g on I does not guarantee the
convergence of ffn0 g on I , but the next theorem shows that uniform convergence of
ffn0 g on I guarantees uniform convergence of ffn g on I , provided there is a c 2 I
such that ffn .c/g converges.
Theorem 10.3.6. Let ffn g be a sequence of functions that are differentiable on
an interval I D a; b. Suppose that the sequence ffn0 g converges uniformly on I
and that there exists c 2 I such that limn!1 fn .c/ exists. Then ffn g converges
uniformly on I to a differentiable function f , and
f 0 .x/ D lim fn0 .x/ (10.11)

n!1
for all x 2 I .
Proof. Choose " > 0. As ffn .c/g converges, there exists N1 such that
jfn .c/ fm .c/j < "
whenever m N1 and n N1 , and the uniform convergence of the sequence ffn0 g

on I implies the existence of an N2 such that
"
jfn0 .t / fm0 .t /j <
ba
whenever m N2 , n N2 and t 2 I .
Let N D maxfN1 ; N2 g and choose m N , n N and x 2 I . For each
t 2 I fxg we may apply the mean-value theorem to the function fn fm to
establish the existence of a between x and t such that
jfn .t / fm .t / fn .x/ C fm .x/j D jt xjjfn0 ./ fm0 ./j

jt xj
< "
ba
": (10.12)
Note that this inequality holds even if t D x. Substituting c for x, we therefore

obtain
jfn .t / fm .t /j jfn .t / fm .t / fn .c/ C fm .c/j C jfn .c/ fm .c/j

< 2":
Thus ffn g satisfies the Cauchy criterion for uniform convergence. Let this sequence
converge to a function f .
Next, for each n 2 N, x 2 I , and t 2 I fxg define
fn .t / fn .x/
hn .t / D I
t x
also let
f .t / f .x/
h.t / D :
t x
Fix x 2 I . Since fn is differentiable, we have
lim hn .t / D fn0 .x/:

t!x
Inequality (10.12) shows that

jfn .t / fn .x/ fm .t / C fm .x/j
jhn .t / hm .t /j D
jt xj
jt xj 1
< "
ba jt xj
"
D
ba
for all m N , n N , and t 2 I fxg. Therefore the sequence fhn g converges
uniformly to h on I fxg. As x is a limit point for I fxg and limt!x hn .t / exists,
we may therefore apply Theorem 10.3.1 to fhn g:
lim fn0 .x/ D lim lim hn .t /

n!1 n!1 t!x
D lim lim hn .t /
t!x n!1
D lim h.t /:
t!x
0
We conclude that f .x/ exists and
f 0 .x/ D lim fn0 .x/:

n!1
t
u
In Example 10.1.2 we found a sequence ffn g of functions that converges to the

constant function 0 but
lim fn0 .0/ 0:

n!1
In view of Theorem 10.3.6, we conclude that ffn0 g cannot be uniformly convergent.

The following example shows that Eq. (10.11) may hold even if ffn0 g is not
uniformly convergent.
Example 10.3.3. For all n > 0 and x 2 0; 1 let
log.1 C n2 x 2 /
fn .x/ D :
2n
Using lHpitals rule, we find that ffn g converges to 0. Furthermore

nx
fn0 .x/ D
1 C n2 x 2
for all x 2 0; 1, so that ffn0 g also converges to 0. Hence
0
lim fn .x/ D lim fn0 .x/:
n!1 n!1
However, we can show that ffn0 g is not uniformly convergent on 0; 1. Since
.1 nx/2 0;
we have
1 C n2 x 2 2nx;
so that fn0 .x/ 1=2 for each relevant n and x. Moreover fn0 .x/ attains its maximum
value of 1=2 at 1=n. Hence
1
sup jfn0 .x/ 0j D sup jfn0 .x/j D
x20;1 x20;1 2
for all n. As this result is nonzero, Theorem 10.2.1 shows that ffn0 g is not uniformly
convergent. 4
Exercises 10.2.
1. Let
n2 x
fn .x/ D
1 C n2 x 2
for all n 2 N and x 2 R.

(a) For each x find
f .x/ D lim fn .x/:

n!1
(b) Is f continuous?
(c) Is the convergence uniform?
2. Show that fsinn xg converges on 0; . Is the convergence uniform?
3. Let fn .x/ D x n =n for all n 2 N and x 2 0; 1.
(a) Does the sequence ffn g converge uniformly?
(b) Is
Z 1 Z 1
lim fn D lim fn
n!1 0 0 n!1
(c) Is
0
lim fn0 D lim fn
n!1 n!1
4. In view of Corollary 10.3.2 we can say that uniform convergence preserves

continuity. Use the following example to show that it does not necessarily
preserve discontinuity: For all n 2 N, let

1
if x is rational;
fn .x/ D n
0 if x is irrational.
5. For each n 2 N, let

1 nx if 0 x < n1 ;
fn .x/ D
0 if x n1 .
(a) Does the sequence ffn g converge uniformly?

(b) Show that fn is continuous for each n but the limit function is not
continuous.
6. Let
2n2 x
fn .x/ D
e n2 x 2
for each n 2 N and x 2 0; 1.

(a) Show that the sequence ffn g is not uniformly convergent on 0; 1.
(b) Is
Z 1 Z 1
lim fn D lim fn
n!1 0 0 n!1
7. Let
2nx
fn .x/ D
1 C n2 x 2
for each n 2 N and x 2 0; 1.

(a) Use differentiation to find the maximum value of each function and hence
show that the sequence ffn g is not uniformly convergent on 0; 1.
(b) Is
Z 1 Z 1
lim fn D lim fn
n!1 0 0 n!1
8. Show that

xn
x
n
converges uniformly on 0; 1 but the sequence of derivatives does not.

9. Let ffn g be a sequence of uniformly continuous functions that is uniformly
convergent to a function f on an interval I . Show that f is uniformly
continuous on I .
10. Let fn and gn be continuous on an interval I , and suppose that ffn g and
fgn g converge uniformly to f and g, respectively. Show that ffn gn g converges
uniformly to fg and that fg is continuous.
11. Let ffn g be a sequence of functions on an interval I . Suppose that fn .c/ D 0
for all n and some c 2 I , and that ffn0 g converges uniformly on I . Show that
ffn g converges uniformly and
lim fn0 .x/ D f 0 .x/:

n!1
12. It is crucial that the interval of integration be finite in Theorem 10.3.3. For all
n 2 N let
(
1
if 0 x n;
fn .x/ D n
0 if x > n.
(a) Show that ffn g is uniformly convergent on 0; 1/ and that fn is integrable

on 0; 1/ for each n 2 N.
(b) Show that

Z 1
lim fn D 1
n!1 0
but
Z 1
lim fn D 0:
0 n!1
(c) Construct a similar example where

Z 1
lim fn D 0
0 n!1
but

Z 1
fn
0
diverges.
10.4 Infinite Series
The results of Sect. 10.3 can be readily adapted to infinite series of functions. Let
ffn g be a sequence of functions defined on a set I and let fSn g be the sequence of
partial sums defined by
X
n
Sn .x/ D fj .x/
j D0
P
for all x 2 I . The series 1 j D0 fj .x/ is said to converge uniformly on I if fSn g
converges uniformly on I . Properties of uniformly convergent sequences can be
used to derive analogous results for uniformly convergent series. Moreover the
comparison, ratio, and root tests are applicable for uniformly convergent series.
P
Theorem 10.4.1. Let 1 j D0 jfj .x/j be a series that converges uniformly on a set
I . For all " > 0 there is an integer N such that jfk .x/j < " whenever k N and
x 2 I.
P
Proof. Noting that the sequence f nj D0 jfj .x/jg converges uniformly on I , we
apply the Cauchy principle. Thus for each " > 0 there exists N1 such that
X
n X
n X
N1
jfj .x/j D jfj .x/j jfj .x/j < "
j DN1 C1 j D0 j D0
whenever x 2 I and n > N1 , and the result follows by taking N D N1 C 1. t

u
10.4 Infinite Series 463
P
Theorem 10.4.2. Let 1 j D0 jfj .x/j be a series that converges uniformly on a set I .
P
Then 1 f
j D0 j .x/ converges uniformly on I .
Proof. By the Cauchy principle, for each " > 0 there exists N such that
X
n X
n X
m
jfj .x/j D jfj .x/j jfj .x/j < "
j DmC1 j D0 j D0
whenever x 2 I and n > m N . For each such x; n; m we have

n n
X X m
X Xn
f .x/ f .x/ D f .x/ jfj .x/j < ";
j j j
j D0 j D0 j DmC1 j DmC1
and the result therefore follows from the Cauchy principle. t

u
The next result follows immediately from Theorem 10.3.1.
P
Theorem 10.4.3. Suppose ffn g is a sequence of functions such that 1 j D0 fj .x/
converges uniformly on some set I R. Let c be a limit point of I . Then
1
X 1
X
lim fj .x/ D lim fj .x/;
x!c x!c
j D0 j D0
provided either side of the equation exists.

P
Corollary 10.4.4. If 1 f .x/ converges uniformly on I and fj is continuous
Pj D0 j
on I for each j , then 1 j D0 fj .x/ is continuous on I .
Proof. The hypotheses and Theorem 10.4.3 show that

1
X 1
X 1
X
lim fj .x/ D lim fj .x/ D fj .c/
x!c x!c
j D0 j D0 j D0
for each c 2 I . t
u
The function Sn is integrable or differentiable on I if each fj is integrable
or differentiable, respectively, on I . These simple observations coupled with
Theorems 10.3.3 and 10.3.6 give the following results immediately.
Theorem 10.4.5 (Term-by-Term Integration). Let ffn g be a sequence of func-
P1 that are integrable on the interval I D a; b. Suppose that the series
tions
j D0 fj .x/ is uniformly convergent on I . Then
0 1
1 Z
X b Z b 1
X
fj .x/ dx D @ fj .x/A dx:
j D0 a a j D0
Theorem 10.4.6 (Term-by-Term Differentiation). Let ffn g be a sequence of

functions that are differentiable on the interval I . Suppose that
P
1. the series 1 0
j D0 fj .x/ is uniformly convergent on I and
P1
2. there exists c 2 I such that the series j D0 fj .c/ converges.
P1
Then the series j D0 fj .x/ is uniformly convergent on I to a differentiable
function, and
0 10
1
X X1
@ fj .x/A D fj0 .x/
j D0 j D0
for all x 2 I .
Although the results concerning the uniform convergence of sequences can be
readily exported to get analogous results for series, the tests for uniform convergence
given in Sect. 10.2 rely on the use of the sequence of partial sums. The sequence
fSn g of partial sums can prove to be an elusive quantity to obtain in a form conducive
to evaluating limits. Indeed, fSn g can be found in closed form only for a few types
of series such as the geometric series. If fSn g is difficult to procure in a useful form,
then identifying a candidate for the limit is yet another potentially formidable task.
Ideally, one desires a test that avoids these problems and relies directly on the terms
of the series. The next result is a comparison test for uniform convergence.
Theorem 10.4.7 (Comparison Test). Let ffn g and fgn g be sequences of nonneg-
ative functions defined on the set I , and suppose that for all x 2 I and k 2 N we
have
fk .x/ gk .x/:
P P1
If the series 1j D0 gj .x/ is uniformly convergent on I , then the series j D0 fj .x/
is uniformly convergent on I .
Proof. Choose " > 0, and let
X
n
Sn .x/ D fj .x/
j D0
and
X
n
Tn .x/ D gj .x/
j D0
for all n 0 and x 2 I . Since fTn g is uniformly convergent on I , the Cauchy

criterion implies that there is an integer N such that
jTn .x/ Tm .x/j < "

for all x 2 I whenever n N and m N . Thus whenever x 2 I and n > m N ,

we have

X
n
jSn .x/ Sm .x/j D fj .x/
j DmC1
X
n
D fj .x/
j DmC1
X
n
gj .x/
j DmC1
D Tn .x/ Tm .x/
< ":
The integer N does not depend on x 2 I , and therefore fSn g is uniformly convergent
on I by Theorem 10.2.3. t
u
The next result, although limited in applications as we shall explain later,
nonetheless proves to be one of the most useful and convenient tests for uniform
convergence of series. It is an immediate consequence of Theorem 10.4.7.
Corollary 10.4.8 (Weierstrass M-Test). Let ffn g be a sequence of functions
sequence fMn g of constants
defined on a set I . Suppose there exists aP P1such that
jfn .x/j Mn for all x 2 I and all n. If 1 j D0 Mj converges, then j D0 fj .x/
converges uniformly on I .
Example 10.4.1. Let ffn g be the sequence of functions defined by
sin nx
fn .x/ D ;
n2 C jxj
for all n 2 N and x 2 R. We have j sin nxj 1 for all x 2 R; consequently,
1
jfn .x/j
n2
P P1
for all n 2 N. Now, 1 2
j D1 1=j is convergent, and therefore the series j D1 fj .x/
is uniformly convergent on R. 4
We know that power series define functions that are differentiable within the
interval of convergence of the series and that the derivative can be obtained by
differentiating the series term by term. A power series can also be integrated term
by term within the interval of convergence. It should thus occasion little surprise
that a power series is uniformly convergent within the interval of convergence.
Theorem 10.4.9. Suppose that the power series

1
X
aj .x a/j
j D0
has a radius of convergence r > 0 and let and be any numbers such that
a r < < < a C r:

P1
Then j D0 aj .x a/j converges uniformly on ; .
Proof. Without loss of generality we can assume that a D 0 and r D 1 since the
general case can be obtained by a translation and scaling of the variable. Let
D maxfjj; jjg:
P
The hypotheses show that < 1, so that the series 1 j
j D0 aj is absolutely
convergent. Moreover, if x 2 I D ; , then jxj . Therefore
jan x n j jan j n
for all x 2 I and all n. The uniform convergence of the power series thus follows
from the Weierstrass M-test with Mn D jan j n . t
u
The Weierstrass M-test is a comparison test for uniform convergence. The key
feature is that the comparison series consists of terms that are constants. Other series
tests that spawn from the comparison test can also be adapted to test for uniform
convergence. We give two such results now.
Theorem 10.4.10 (DAlembert Ratio Test). Let ffn g be a sequence of functions
defined on a set I . Suppose there exist numbers r < 1 and N such that fN is
bounded and for all x 2 I and n N we have fn .x/ 0 and
jfnC1 .x/j
r:
jfn .x/j
P1
Then the series j D0 fj .x/ converges uniformly on I .
Proof. Arguing as in the proof of Theorem 3.7.1, we see by induction that
jfN Cj .x/j r j jfN .x/j
for all x 2 I and j 2 N. The result now follows from the comparison test since fN
is bounded and r < 1. t
u
The reader should have no trouble proving the next theorem.
Theorem 10.4.11 (Cauchy Root Test). Let ffn g be a sequence of functions

defined on a set I , and suppose that there exist numbers r < 1 and N 2 N such that
jfn .x/j1=n r
P1
for all x 2 I and n N . Then the series j D0 fj .x/ converges uniformly on I .
The Weierstrass M-test is perhaps the most frequently used test for the uniform
convergence of series. Indeed, Bromwich ([4] p. 124) notes that series satisfying
the Weierstrass M-test were called normally convergent by Baire and that this
terminology has the advantage of emphasizing the fact that the M-test can be
applied to nearly all series in ordinary everyday use. A major limitation of the
test, however, is that a series must be absolutely convergent at each point in the set
I in order to apply the test successfully. More delicate tests are needed to cope with
series that are uniformly convergent in a set but conditionally convergent at points
in the set. However, we note the following result due to Baire that simply says that
every uniformly convergent series of bounded functions on a set I can be made into
a normally convergent series by a judicious grouping of terms. The proof of this
theorem may be found in [4].
Theorem 10.4.12 (Baire). For every uniformly convergent series of functions
bounded on some set, there exists a regrouping of terms such that the resulting series
satisfies the Weierstrass M-test.
Exercises 10.3.
P1
1. Determine whether the following series j D2 fj .x/ converge uniformly on the
set I :
(a) fn .x/ D sin nx
2n
, I D ; .
(b) fn .x/ D x
n2 Cx 2
, I D R.
x2
(c) fn .x/ D .1Cx 2 /n
, I D R.
.nC1/x
(d) fn .x/ D 2 , I D 0; 1.
e nx =2
log.1Cnx/
(e) fn .x/ D nxn
, I D c; 1/ for some c > 1 [hint: log.1 C h/ < h when
h > 0].
(f) fn .x/ D log 1 C x
n log2 n
, I D c; c for some c > 0.
(g) fn .x/ D 1
2n1 x
, I D .0; 1.
(h) fn .x/ D sinpnx sin x
nCx
,I D
0; 2.
c c
(i) fn .x/ D .1/ nC1 1
nx
,I D c; 1/ for some c > 0 [hint: nx D nx 2 n 2 ].
2. Let
sin nx
fn .x/ D
n2
P
for all n > 0 and x 2 R. Show that 1 j D1 fj .x/ converges uniformly over R
P
whereas 1 f
j D1 j
0
.x/ diverges at 0.
3. For all n > 0 let
(
0 if x D 0;
fn .x/ D x n log x
n
if 0 < x 1.
the supremum of jfn .x/j on 0; 1 occurs at e 1=n .

(a) Show that P
(b) Show that 1 j D1 fj .x/ converges uniformly on 0; 1.
4. For all n > 0 and x 2 R let
n2 x 2
fn .x/ D :
e n2 x
(a) Show that
4
sup jfn .x/j D :
x2R n2 e 2
P1
(b) Show that j D1 fj .x/ converges uniformly on R.
5. For all n > 0 and x 2 R let
(
1
if 1
< x n1 ;
fn .x/ D n nC1
0 otherwise.
P1
Show that j D1 fj .x/ converges uniformly but
1
X
sup jfj .x/j (10.13)
j D1 x2R
P1
diverges. Note that in general if (10.13) converges, then j D1 fj .x/ converges
uniformly (why?).
P
6. Show that if 1 j D0 aj is absolutely convergent, then
1
X
aj sin jx
j D1
and
1
X
aj cos jx
j D0
converge uniformly
P on R.
7. Suppose that 1 j D0 fj .x/ converges uniformly on an interval I to a bounded
P
function f W I ! R. Show that 1 j D0 f .x/fj .x/ converges uniformly on I .
8. Let ffn g be a sequence of functions defined on an interval I , and suppose that
fn .x/ 0 for all n and all x 2 I . Suppose also that ffn .x/g is a decreasing
sequence for all x 2 I and that
lim sup fn .x/ D 0:

n!1 x2I
Show that
1
X
.1/j fj .x/
j D0
converges uniformly on I .
9. Let
fn .x/ D .1/n .1 x/x n

P1
for all n and all x 2 0; 1. Show that j D0 fj .x/ converges uniformly but
P1
j D0 jfj .x/j does not.
10. Show that
1
X 1
j 2 C x2
j D0
is differentiable on R.
Bibliography
1. Ali, S.A.: The mth ratio test: new convergence tests for series. Math. Assoc. Am. 115, 514524
(2008)
2. Anderson, G.D., Vamanamurthy, M.K., Vuorinen, M.: Monotonicity of Some Functions in
Calculus, Research Report Series, vol. 538, University of Auckland (2005)
3. Bartle, R.G., Sherbert, D.R.: Introduction to Real Analysis. Wiley, New York (1982)
4. Bromwich, T.J.IA.: An Introduction to the Theory of Infinite Series. Macmillan, London
(1926)
5. Carlson, B.C.: Algorithms involving arithmetic and geometric means. Am. Math. Mon. 78(5),
496505 (1971)
6. Dunlap, R.A.: The Golden Ratio and Fibonacci Numbers. World Scientific, Singapore (1997)
7. Fulks, W.: Advanced Calculus, 3rd edn. Wiley, New York (1978)
8. Huang, X.-C.: A discrete lHopitals rule. Coll. Math. J. 19(4), 321329 (1988)
9. Kemeny, J.G.: The exponential function. Am. Math. Mon. 64(3), 158160 (1957)
10. Little, C.H.C., Teo, K.L., van Brunt, B.: The Number Systems of Analysis. World Scientific,
Singapore (2003)
11. Marsden, J.E., Hoffman, M.J.: Basic Complex Analysis, 2nd edn. Freeman, New York (1987)
12. Moritz, R.E.: On the extended form of Cauchys condensation test for the convergence of
infinite series. Bull. Am. Math. Soc. 44(6), 441442 (1938)
13. Piaggio, H.T.H.: An Elementary Treatise on Differential Equations and Their Applications. G.
Bell & Sons, London (1920). (Reprinted 1971)
14. Shiu, P.: A generalization of some convergence tests. Math. Gazette 56, 227228 (1972)
15. Smith, D., Teo, K.L.: Linear Algebra. New Zealand Mathematical Society, Wellington (1990)
16. Stone, M.H.: Developments in Hermite polynomials. Ann. Math. 29, 113 (1927)
17. Tong, J.-C.: Cauchys mean value theorem involving n functions. Coll. Math. J. 35, 50,51
(2004)
18. Zhang, X., Wang, G., Chu, Y.: Extensions and sharpenings of Jordans and Kobers inequalities.
J. Inequal. Pure Appl. Math. 7(2), Article 63 (2006)

Texts in Mathematics, DOI 10.1007/978-1-4939-2651-0
Index
Symbols Cauchy principle, 73

, 267 uniform convergence, 447
nth term test, 118 Cauchys condensation test, 124
p-series, 126 Cauchys function, 413
Cauchys mean-value formula, 288
CauchySchwarz inequality, 381
A cell, 4
Abels partial summation identity, 152 chain rule, 251
absolute value, 12 closed set, 76
accumulation point, 75 commutativity, 16
addition, 15 comparison test, 121
addition of complex numbers, 15 uniform convergence, 464
Airy function, 411 completeness, 67
Airys equation, 331 complex number
arc length, 377 argument of, 280
area, 333 polar form of, 280
arithmetic mean, 51, 63 component
arithmeticgeometric mean, 84 first, 3
associativity, 16 second, 3
composition of functions, 5
B convergence
Baires theorem, 467 uniform, 442
bijection, 4
binomial coefficient, 29
binomial expansion, 30 D
binomial theorem, 29, 412 Darbouxs theorem, 263
Bolzanos theorem, 226 de Moivres theorem, 280
BolzanoWeierstrass theorem, 74 density, 67
bounded above, 67 derivative, 243
bounded below, 67 nth, 248
Dirichlets test, 154
discriminant, 17, 262
C disjoint sets, 2
Cartesian product, 3 distinct sets, 2
Cauchy condition, 198 distributive, 25

Texts in Mathematics, DOI 10.1007/978-1-4939-2651-0
474 Index
distributivity, 16 inverse tangent, 269

division point, 333 limit of, 192
Lipschitz continuous, 240
local maximum of, 262
E local minimum of, 262
equal sets, 2 logarithm, 231
equation monotonic, 13
quadratic, 17 nondecreasing, 13
equivalence class, 4 nonincreasing, 13
equivalent sequences, 67 one-to-one, 4
Eulers constant, 326 order of zero of, 406
EulerMaclaurin theorem, 394 periodic, 268
exponentiation, 8, 94, 231 popcorn, 219
range of, 4
real-valued, 4
F reciprocal of, 178
factorial, 12 secant, 178
first derivative test, 262 sine, 177
fixed point, 228 smooth, 403
fixed point iteration, 426 strictly monotonic, 13
fixed-point theorem, 228 surjective, 4
Fuchss theorem, 330 tangent, 178
function, 4 Thomaes, 219
Airy, 411 trigonometric, 178
bijective, 4 uniformly continuous, 237
Cauchys, 413 fundamental theorem of calculus, 362
complex-valued, 4
constant, 4
continuous, 216 G
continuous on the left, 223 Gausss constant, 84
continuous on the right, 223 Gausss test, 234
continuous over a set, 224 generalised ratio test, 158
cosecant, 178 generalised root test, 159
cosine, 177 geometric mean, 61, 63
cotangent, 178 golden ratio, 35
critical point of, 263 greatest lower bound, 67
decreasing, 13
differentiable, 243
domain of, 4 H
exponential, 97, 176 harmonic mean, 63
extremal point of, 262 Hermite polynomial, 329
extremal value of, 262 Hermites equation, 328
extremum of, 262
fixed point of, 424
floor, 204 I
graph of, 5 image, 4
arc length of, 377 indeterminate form, 290
Heaviside, 203 induction, 5
increasing, 13 inductive hypothesis, 6
injective, 4 inequality
integrable, 342 Bernoullis, 7
inverse cosine, 269 triangle, 12
inverse of, 4 infimum, 67
inverse sine, 269 initial-value problem, 437
Index 475
injection, 4 multiplication of complex numbers, 15

integral, 342 mutually disjoint sets, 2
improper
absolutely convergent, 392
comparison test, 391 N
lower, 340 neighborhood, 37
upper, 340 Newtons method, 434
integral test, 396 norm, 334
integrand, 342 number
integration by parts, 367 complex, 15
integration by substitution, 366 conjugate of, 18
interior point, 334 exponentiation of, 17
intermediate-value theorem, 227 imaginary part of, 15
intersection of sets, 2 modulus of, 19
interval, 13 real part of, 15
closed, 13 natural, 5
ends of, 13
half open, 13
open, 13 O
inverse function theorem, 253 order of magnitude, 127
ordered pair, 3
component of, 3
J ordered triple, 3
Jordans inequality, 274
P
K partition, 4, 333
KummerJensen test, 142 refinement of, 338
peak index, 100
polynomial, 9
L coefficient of, 9
lHpitals rule, 290 degree of, 9
least upper bound, 67 root of, 9
Legendres equation, 332 product, 8, 165
Leibnizs rule, 256 Cartesian, 12
Leibnizs test, 147 factors of, 8
limit product of sums, 165
infinite, 208
one-sided, 205
limit comparison test, 128 R
limit point, 75 Raabes test, 143
Lipschitz condition, 248 ratio test, 132
Lipschitzs theorem, 247 uniform convergence, 466
lower bound, 67 relation, 3
lower sum, 336 equality, 3
equivalence, 3
inclusion, 3
M reflexive, 3
maximum- and minimum-value theorem, 225 symmetric, 3
mean-value theorem, 258 transitive, 3
mean-value theorem for integrals, 360 Riemann sum, 337
mesh, 334 convergence of, 341
method of successive approximations, 438 Riemanns theorem, 160
multiplication, 15 Rolles theorem, 257
476 Index
root test, 136 multiplication of, 174

uniform convergence, 467 radius of convergence, 172
real, 111
rearrangement of, 164
S Taylor, 411
sandwich theorem, 56, 201 telescoping, 113
sequence, 33 term of, 111
approaches infinity, 78 uniform convergence of, 462
approaches minus infinity, 79 set, 1
arithmetic, 34 compact, 224
bounded, 43 complement of, 2
Cauchy, 65 element of, 1
complex, 33 empty, 1
complex harmonic, 34 member of, 1
convergent, 37 ordered, 12
decreasing, 77 power, 2
divergent, 37 subset of, 1
Fibonacci, 35 subinterval, 334
geometric, 34 subsequence, 53
harmonic, 34 subtraction, 16
imaginary part of, 49 subtraction of complex numbers, 16
increasing, 53 sum, 8
limit of, 37 terms of, 8
monotonic, 77 supremum, 67
nondecreasing, 77 supremum property, 67
nonincreasing, 77 surjection, 4
null, 38
partial sum of, 111
rational, 33 T
real, 33 Taylor polynomial, 400
real part of, 49 Taylor remainder, 403
rearrangement of, 164 Taylors theorem, 401
sign, 34 telescoping property, 26
term of, 33
series, 111
Abels, 234 U
absolutely convergent, 157 uniform convergence, 442
alternating, 147 series, 462
condensed, 123 union of sets, 2
conditionally convergent, 157 upper bound, 67
derived, 316 upper sum, 335
geometric, 112
harmonic, 118
Maclaurin, 411 W
power, 170 Weierstrass M-test, 465
addition of, 174
center of, 170
circle of convergence, 172 Z
interval of convergence, 172 Zenos paradox, 109

Real Analysis Via Sequences An Series. Charles Little, Teo Kee, Bruce Van Brunt

Uploaded by

Copyright:

Available Formats

Real Analysis Via Sequences An Series. Charles Little, Teo Kee, Bruce Van Brunt

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

Real Analysis Via Sequences An Series. Charles Little, Teo Kee, Bruce Van Brunt

Uploaded by

Copyright:

Available Formats

What is the purpose of this text?

What is the purpose of this text?

What topics are covered in the text?

What topics are covered in the text?

Undergraduate Texts in Mathematics

Colin Adams, Williams College

Undergraduate Texts in Mathematics are generally aimed at third- and fourth-

More information about this series at http://www.springer.com/series/666

Real Analysis via Sequences

Bruce van Brunt

ISSN 0172-6056 ISSN 2197-5604 (electronic)

Library of Congress Control Number: 2015935731

Mathematics Subject Classification (2010): 26-01

Springer New York Heidelberg Dordrecht London

Printed on acid-free paper

Palmerston North, New Zealand Charles H.C. Little

3.15 Introduction to Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

10 Sequences of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437

Springer Science+Business Media New York 2015 1

1.2 Ordered Pairs, Relations, and Functions

As we said before, sets can be used to construct a large number of mathematical

.x; y; z/ D ..x; y/; z/

Equivalence relations are closely linked to partitions. A partition of a set X is a

such x, we see that h is a function from X into Z. We call it the composition of g

.g f /.x/ D g.f .x//

for each x 2 X . For instance, if f is a bijection from X onto Y , it follows that

1.3 Induction and Inequalities

Z D f: : : ; 2; 1; 0; 1; 2; : : :g:

Theorem 1.3.1. Let Y Z and a 2 Y . Suppose also that y C 1 2 Y whenever

if m > n, and we regard this expression as a sum with no terms.

for all n > m. We often write

and regard this expression as a product with no factors.

We may also define a0 D 1, though this definition is normally made only if a 0.

One can easily prove by induction that

if b 0. If ab 0, then these two equations extend to the case where n is any

for all n 2 N and then that

.am /n D amn (1.7)

for all numbers x is called a polynomial. Its degree is n if an 0. The numbers

then c is called a root of the polynomial (1.8).

On the other hand, if a D b n D b 2r D .b 2 /r , wherepb  0, then b 2 D c by the

Moreover 01=n D 0 and 11=n D 1.

Next, let a  0, let n 2 N, and let m be a nonnegative integer. We then define

We proceed to the generalization of Eqs. (1.4)(1.7). Let a  0, let m be an

Thus if p and q are also integers and q > 0, then

Now suppose that b is also a nonnegative real number. As

it then follows that

a result that generalizes Eq. (1.4). Thus if b > 0, then

.x1 ; x2 ; : : : ; xn / D ..x1 ; x2 ; : : : ; xn1 /; xn /:

Let X1 ; X2 ; : : : ; Xn be sets, where n > 1. We can also define the Cartesian

This result is known as the triangle inequality.

.a; b/ D fx j a < x < bg;

.a; 1/ D fx j a < xg;

for all such that 0   1.

for all real numbers x and y. If y 0, prove also that

3. Use the triangle inequality to prove that

jjxj jyjj  jx yj;

where x and y are real numbers.

1.4 Complex Numbers

where a and b are also real, then

On the other hand, if a D b n D b 2r D .b 2 /r , wherepb 0, then b 2 D c by the

Next, let a 0, let n 2 N, and let m be a nonnegative integer. We then define

We proceed to the generalization of Eqs. (1.4)(1.7). Let a 0, let m be an

for all such that 0 1.

jjxj jyjj jx yj;

N zN D .u iv/.x iy/ D ux vy C i.uy vx/ D wz:

z C zN D 2Re .z/ 2jRe .z/j 2jzj:

jw zj D jz wj jzj jwj D .jwj jzj/;

jzj jxj C jiyj D jxj C jyj D jRe .z/j C jIm .z/j:

D 1 if b 0 and D 1 if b < 0. Hence, conclude that the square root of a

d.z; w/ d.z; v/ C d.v; w/;

d.z1 ; zn / d.z1 ; z2 / C d.z2 ; z3 / C : : : C d.zn1 ; zn /:

jd.z; v/ d.v; w/j d.z; w/

Let m and n be integers with n m. Recall that

Thus if m n and am ; amC1 ; : : : ; anC1 are numbers, then

.anC1 an / C .an an1 / C C .amC1 am /