GUIDE TO ANALYSIS

Consultant Editor: David A. Towers,

Senior Lecturer in Mathematics,
University of Lancaster

Titles Available

Abstract Algebra
Linear Algebra
Analysis

Further titles are in preparation.

Guide to Analysis

F. Mary Hart
Department of Pure Mathematics
University of Sheffield

M
MACMILLAN
EDUCATION
© F. M. Hart 1988

All rights reserved. No reproduction, copy or transmission

of this publication may be made without written permission.

No paragraph of this publication may be reproduced, copied

or transmitted save with written permission or in accordance
with the provisions of the Copyright Act 1956 (as amended),
or under the terms of any licence permitting limited copying
issued by the Copyright Licensing Agency, 7 Ridgmount Street,
London WCIE 7AE.

Any person who does any unauthorised act in relation to

this publication may be liable to criminal prosecution and
civil claims for damages.

First published 1988

Published by
MACMILLAN EDUCATION LTD
Houndmills, Basingstoke, Hampshire RG21 2XS
and London
Companies and representatives
throughout the world

British Library Cataloguing in Publication Data

Hart, F. Mary
Guide to analysis.-(Macmillan
mathematical guides).
1. Calculus
I. Title
515 QA303
ISBN 978-0-333-43788-9 ISBN 978-1-349-09390-8 (eBook)
DOI 10.1007/978-1-349-09390-8
CONTENTS

Editor's foreword vii

Preface viii

Glossary of symbols and notation x

1 NUMBERS AND NUMBER SYSTEMS 1

1.1 Natural numbers 1
1.2 Integers 2
1.3 Rational numbers 3
1.4 The real number system 7
Appendix: Mathematical induction 20

2 SEQUENCES 25
2.1 Introduction 25
2.2 Sequences 26
2.3 Standard limits 42
2.4 Some general results for sequences 47
2.5 Subsequences 51
Appendix: Triangle inequalities 56
Table of standard limits 58

3 INFINITE SERIES 62
3.1 Introduction 62
3.2 Series and notation 64
3.3 Tests for convergence 69
3.4 Absolute convergence, rearrangement of series,
multiplication of series 92
Appendix 100
Tests for convergence-summary 103
v
CONTENTS

4 FUNCTIONS Limits of functions, continuous functions 106

4.1 Limits offunctions 106
4.2 Continuity 139
4.3 Some properties of continuous functions on closed
intervals 145

5 DIFFERENTIABLE FUNCTIONS Dilferentiable functions,

Rolle's theorem, mean value theorem, Taylor's theorem,
I'Hopital's rule, maxima and minima, Taylor series 160
5.1 Differentiation 160
5.2 Rolle's theorem and the mean value theorem 173
5.3 Taylor's theorem 180
5.4 L'Hopital's rule 183
5.5 Maxima and minima 187

Index 201

vi
EDITOR'S FOREWORD

Wide concern has been expressed in tertiary education about the difficulties
experienced by students during their first year of an undergraduate course
containing a substantial component of mathematics. These difficulties have a
number of underlying causes, including the change of emphasis from an
algorithmic approach at school to a more rigorous and abstract approach in
undergraduate studies, the greater expectation of independent study, and the
increased pace at which material is presented. The books in this series are
intended to be sensitive to these problems.
Each book is a carefully selected, short, introductory text on a key area of
the first-year syllabus; the areas are complementary and largely self-
contained. Throughout, the pace of development is gentle, sympathetic and
carefully motivated. Clear and detailed explanations are provided, and
important concepts and results are stressed.
As mathematics is a practical subject which is best learned by doing it,
rather than watching or reading about someone else doing it, a particular
effort has been made to include a plentiful supply of worked examples,
together with appropriate exercises, ranging in difficulty from the straight-
forward to the challenging.
When one goes fellwalking, the most breathtaking views require some
expenditure of effort in order to gain access to them: nevertheless, the peak is
more likely to be reached if a gentle and interesting route is chosen. The
mathematical peaks attainable in these books are every bit as exhilarating, the
paths are as gentle as we could find, and the interest and expectation are
maintained throughout to prevent the spirits from flagging on the journey.

Lancaster, 1987 David A. Towers

Consultant Editor

Vll
PREFACE

The transition from sixth form to university or polytechnic can be traumatic.

Students have to adapt to different styles of teaching with lectures replacing
lessons and at the same time they are often presented with material which is
unlike anything they have encountered before. The provision of suitable
textbooks can go some way to alleviating the problem. As far as pure
mathematics is concerned, analysis, which is central to most undergraduate
courses in mathematics, is likely to provide the biggest stumbling-block. The
subject itself is intrinsically difficult. The final polished version we have today
owes its form to years of concentrated work by many celebrated mathe-
maticians, and students often find it very daunting to be faced with this
finished product. They seldom realise just how much effort went into its
making, who did the work or why they did it. This account therefore includes
some historical notes and anecdotes which help to put the subject into context
and, it is hoped, enliven the text for the reader.
This volume is written specifically for students meeting analysis for the first
time. It is designed so that it can be used as a textbook in conjunction with a
lecture course in analysis or it can be used by students wishing to teach
themselves the subject-independent of any lecture course. For this reason
numerous worked examples are included with very detailed solutions to
enable the reader to check that the concepts and definitions are understood.
When working the exercises the reader is given the opportunity to receive a
little help (or inspiration!) without being given the answer, as the answers and
the hints for solutions are recorded in separate sections. In practice, one
exercise successfully completed by the reader (even with the help of a few hints)
can teach more than several examples read from the text.
Analysis traditionally contains numerous theorems and proofs. These
theorems are the tools of the trade and we cannot just dispense with large
numbers of them in order either to cut down the size of the volume or to make
the learning process appear easier. Neither can we simply omit their proofs.
Before using a result we must know for certain that it is true. The need for proof
is something which has always been recognised to a greater or lesser extent.
Historical records show that the ancient Greek mathematicians living
viii
 PREFACE

centuries before Christ acknowledged this, and, indeed, some of our present-
day proofs, like the contradiction argument, stem from their work. However,
the use of the link between limits of functions and sequences allows us a few
economies and reduces the number of separate proofs necessary.
For many undergraduates the greatest hurdle is the problem posed by the
idea of rigour. They are unsure about how much detail needs to be included
and how much may reasonably be omitted. The reader of more advanced texts
can be assumed to have sufficient judgement to be able to cope when details are
not carefully spelled out. A more elementary text, however, needs different
treatment and it seems prudent to include all the small details. This may make
the definitions and theorems appear a little more cumbersome, but it leaves the
reader in no doubt as to what is being assumed. The benefits far outweigh any
superficial disadvantage occurring from the longer, more detailed statements.
Since repetitions can be valuable for reinforcing knowledge, some statements
and proofs which closely resemble earlier ones are nevertheless recorded in
detail rather than being dismissed with an assertion of the form 'the proof is
similar to that of ... '.
Finally, I would like to record my profound thanks for all the secretarial
help I have received during the preparation of this volume. My thanks go to
Anne Hall and Janet Williams and particularly to Alex Cain who has
undertaken the bulk of the work.

Sheffield, 1987 F. M. Hart

ix
GLOSSARY OF SYMBOLS
AND NOTATION

Set of all natural numbers. This set consists of the numbers

0,1,2,3, ...
Set of all integers
Set of all positive integers
Set of all rational numbers
Set of all real numbers
Set of all non-zero real numbers
Modulus of a real number x. Its value is x if x ~ 0 and its value
is -x if x <0. Thus Ixl = +#
> Greater than
< Less than
;j> Not greater than
<f:: Not less than
~ Greater than or equal to
~ Less than or equal to
{x: P(x)} The set of all x such that P(x) holds
{a,b,c, ... } The set consisting of a, b, c, ...
Tends to
Does not tend to
Belongs to
Does not belong to
Infinite sequence ai' a 2 , a 3 , •• • whose nth term is an' (This notation
is used when we wish to make clear which values are assumed
by the subscript n)
Infinite sequence with terms of the form an' (This notation is
used when it is clear from the context which values are assumed
by n)
Subsequence an" an" an" '" of the sequence (an)
Sum of

x
 GLOSSARY OF SYMBOLS AND NOTATION

This is used in two senses. It is used formally to stand for the

infinite series a 1 + a2 + a 3 + ... It is also used for the sum of the
infinite series in the case in which the series is convergent. The
context makes clear which meaning is appropriate

The sum of the n terms a 1 + a2 + " . + an

y Euler's constant
/: A-+~ Function / from A into ~ i.e. / has domain of definition A and
range a subset of ~
log x The natural logarithm of x. Throughout this volume all
logarithms are to the base e
(a, (0) Set of all real numbers x such that x > a
[a, (0) Set of all real numbers x such that x ~ a
(- oo,b) Set of all real numbers x such that x < b
(- oo,b] Set of all real numbers x such that x ::::; b
(a,b) The open interval (a, b). It consists of all the real numbers x
such that a < x <b
[a,b] The closed interval [a, b]. It consists of all the real numbers x
such that a ::::; x ::::; b
x-+a x approaches a (two-sided limit)
x-+a+ x approaches a through values greater than a (one-sided limit)
x-+a x approaches a through values less than a (one-sided limit)
go/ 9 composition f This is the function such that
(gof)(x) = g(f(x))
/-1 Inverse function of /
l' Derivative. of /
IVT Intermediate value theorem
MVT Mean value theorem

Xl
1 NUMBERS AND
NUMBER SYSTEMS

This is a chapter in which we have another look at some familiar number

systems and meet some rather unexpected ideas.

1.1 NATURAL NUMBERS

We begin with the familiar numbers 0,1,2,3,4, ... , which we have used for
counting since early childhood. Not surprisingly, this set of numbers is called
the set of natural numbers and it is normally denoted by N. Thus the set N of
natural numbers consists of the numbers 0, 1,2, 3, 4, ....
Over the centuries various ways have been devised for recording these
numbers. The actual symbols we now use for them-the so-called arabic
numerals-originated in India many centuries ago. In time, they were
incorporated into Arab culture and were then finally brought to Europe by the
Arabs. Ancient civilisations, however, commonly used marks of various
shapes to denote the number of objects, with each mark corresponding to one
object in much the same way as primitive tribesmen sometimes make notches
on the handles of their weapons to show how many of the enemy they have
slain.
It is, of course, hardly surprising that it soon became recognised that
counting alone was not sufficient and the laws for the operations of addition
and multiplication were developed together with rules governing inequalities.
Thus two numbers m,.ti could be compared and statements of the form m < n
or m ~ n have sensible meanings. It is the order properties of the natural
numbers which gives rise to the powerful method of proof called the principle
of mathematical induction. Proofs using this principle permeate most
branches of mathematics and an appendix at the end of the chapter discusses it.
Many devices like the abacus were developed to facilitate addition and
multiplication and their users became amazingly skilled and displayed great
ingenuity in handling them-though they were naturally somewhat slower
than our modem computers! But at least users of the abacus obtained the
answer 4 for 2 x 2 and not 3.999. On the whole, the abacus'was a commercial
GUIDE TO ANALYSIS

tool. It was used widely by merchants and traders. The highly educated,
however, tended to prefer written records and some of the mathematics of the
Babylonians, Egyptians, Greeks, etc. can be seen in museums today.
If m and n are two natural numbers, then their sum m + n and their product
mn are also natural numbers. However, given two natural numbers m, n it is
not always possible to find a natural number x such that
m+x=n. (1)
For instance, there is no natural number x such that 3 + x = 2. In order to be
able to solve equation (1) to find x in every case, the number system must be
extended to the set Z of all integers.

1.2 INTEGERS
The set Z of integers consists of the numbers 0, ± 1, ± 2, ± 3, ± 4, .... The set of
positive integers which forms part of this set is denoted by Z +. Moreover,
given any two integers m, n it is always possible to find an integer x such that
m+x=n
and such equations no longer cause any difficulty. As in the case of the natural
numbers an order relation is defined on Z, so that statements of the form n ~ p
or p < q have a sensible meaning.
Suppose we have a non-empty set S of integers with the property that there
is some integer p such that
p~m (1)
for all mESo Then we can easily show that the set S possesses a smallest
element. This result is known as the well-ordering principle for the integers.
Since S is not empty, it must contain some element q. Let the set T consist of
all the integers I such that p ~ I ~ q,
i.e. T= {p,p + 1,p + 2,p + 3, ... ,q -l,q}.
Then T has only a finite number of elements, and therefore Tn S has only a
finite number of elements. Moreover, Tn S is not empty, because qE T and
qES; and TnS contains all the elements of S which are either less than q or
equal to q. Since TnS has only a finite number of elements it must have a
smallest element k. This integer k belongs to S and there are no elements of S
which are less than k, i.e. k is the smallest element of S.
Since every positive integer satisfies the inequality m;;:: 1 (which is a special
case of(1) with p = 1), we see that every non-empty set of positive integers has
a least element. This is precisely the property we need to prove the principle of
mathematical induction.
2
 NUMBERS AND NUMBER SYSTEMS

We record these results for future reference in the form of a theorem and its
corollary.

THEOREM 1.2.1 Let S be a non-empty set of integers with the property that
there is some integer p such that p ~ m for all mESo Then S has a least member.

CO RO LLARY Every non-empty set of positive integers has a least member.

We now recall that the sum and product of two integers is always an integer.
However, given two integers p, q with p -# 0, it is not always possible to find an
integer x such that
px=q.
For example, there is no integer x such that 3x = 2. This hurdle can be
surmounted only by again extending the number system. In this case we obtain
the set Q of rational numbers.

1.3 RATIONAL NUMBERS

The set of rational numbers consists of all numbers of the form p/q, where p, q
are integers and q -# o. For example t, -1, ii, etc. are all rational numbers. The
use of the rational numbers permits the solution of the equations
r+x= s,
tx=u
for x, where r, s, t, u are all rational numbers and t -# O. Thus operations of
subtraction and division (with non-zero divisor) are always possible with
rational numbers.
Any rational number can be represented as the quotient oftwo integers or as
a decimal in the familiar way. But, does this decimal have any special features?
Yes, it does. The decimal representation of a rational number is either
terminating, i.e. it has only a finite number of digits after the decimal point, or it
is a repeating decimal, i.e. after a certain time the digits keep on repeating a
definite pattern. For example! = 0.125, ~ = 0.142 857142 8571 ... with the

n
pattern 142857 continually repeating. Normally we put dots over the
repeating pattern of digits and we would write ~ = 0.i4~ 857, = 3.54 (which
means that n = 3.5454545454 ... ) and i = 0.83 (which means that i
= 0.833 333 333 ... ). In fact we can also prove the converse, viz. that
terminating decimals and repeating decimals always represent rational
numbers. The reader should try calculating the decimal representation of some

3
GUIDE TO ANALYSIS

rational numbers; this will provide insight into the reason why these patterns
occur.
Nowadays, pocket calculators are frequently used to find the quotient of
two numbers. If, however, we recall the old method oflong division and use it
to divide p by q, where p, q are integers and q > 0, then the reason for the
repeating pattern becomes immediately obvious. First let us suppose that
0< p < q and let us use the method oflong division to express p/q as a decimal.
At each stage we divide by q to give a remainder r which is one of the numbers
0, 1,2,3, ... , (q - 1). Then we bring down the digit 0 and divide r x 10 by q, to
give a remainder and keep on repeating this process. Since there are only q
different possible values for the remainder, we either obtain zero remainder at
some stage and the decimal terminates, or, after at most q steps of long
division, we obtain a remainder which is the same as one of the previous
remainders and the pattern of digits starts to repeat. The reader is encouraged
to try some examples to check what happens.
If 0 < q < p then ~can be expressed as~ = m +~, where m, n are integers and
o~ n < q. Thus n/q is represented by a decimal which either terminates or has
a repeating pattern, hence it follows that the decimal representation of p/q
either terminates or has a repeating pattern. The extension ofthe argument to
cover the case in which the rational number p/q is negative is now obvious.
Conversely, it can be proved that a decimal with a repeating pattern
represents a rational number, and a terminating decimal obviously represents
a rational number. The proof is left as an exercise for the reader.·

EXERCISE 1.3.1

First consider a decimal of the form 0.a 1a2 ••• an with a repeating pattern of n
digits. Write x = 0.a 1 a2 ••• an. Express 10nx as a decimal. Then subtract x and
check that 10n x - x is an integer. Deduce that x is rational. Now extend the
method to prove that all decimals with a repeating pattern represent rational
numbers.

It is interesting to consider how many rational numbers there are. Obviously

there are infinitely many, but we can get some idea of the size of II) by noticing
that its elements can be listed. To prepare this list we first notice that the
positive rationals can be displayed in a doubly infinite array as follows:
1 1._1. 1 --!- 1.
~/2/3/4/5/6
1/2/3/4/5 6
1. 1. 1. 1. 1. 1.

1/2/3/4 5 6
~~J:~J:J:

1~~~~5~6
1/2 3 4
.s.1 •............•••..••••..

This can now be reduced to a single list by starting at the top left-hand corner
4
 NUMBERS AND NUMBER SYSTEMS

and proceeding along the diagonals as shown by the arrows. Each time we
come to a number which has already been included, we pass over it and
continue on our way following the arrows. This gives a list beginning
1,2, t, 1. 3, 4, t~,!, t, 5, .... Now that the positive rationals are arranged in a
certain order, we include all the remaining rationals by inserting 0 at the
beginning and then inserting the negative element - p/q directly after p/q in
the list. We have now included all the rationals. Clearly there are infinitely
many of them, but not so many that they cannot be incorporated in a single
list. We will see later that the real numbers are too numerous to be put in order
in a list. We say that the set Q of rationals is countably infinite. But are there
enough of them for the purpose of analysis?
At the beginning of the chapter we started with the familar set of natural
numbers and it rapidly became apparent that this number system had to be
extended with the result that we now have the rational number system Q. Is
this adequate or do we need further extension? Certainly, many numerical
calculations use only rational numbers. A modern computer can use numbers
with a very large number of digits but its capacity is finite and so any decimal it
uses must have only a finite number of digits and it is, therefore, a rational
number. If a calculation requires a number whose square is 2, a computer or
pocket calculator will use a rational number for it. Is this justified? Is there a
rational number whose square is 2?
Centuries before Christ, the Babylonians had tables of roots. A surviving
tablet shows that they gave
24 51 10
1 + 60 + (60)2 + (60)3
as the value of the square root of 2 (in our decimal system this is 1.41421296
which is in remarkable agreement with the value given in our four-figure
tables). How they calculated their roots is not known, but their results are
surprisingly accurate. Obviously, they expressed their values in the form
Pl P2 P3 P4
Po + 60 + (60)2 + (60)3 + (60)4 + ... ,
where Po, Pl, P2,'" are integers and they used more and more terms to obtain
closer approximations. We imagine they believed that the process would
terminate eventually giving a finite number of terms for the square root of 2
and so assumed that there is a rational number whose square is 2. Surviving
records show that the Greeks and Egyptians also knew how to calculate good
approximations to roots. The Greeks, of course, concentrated on the
geometric aspects of mathematics. Using a ruler and compass they could
construct a unit square and they knew that the area of the square on its
diagonal was 2. So what is the length of this diagonal?
A leading Greek mathematician Pythagoras (who lived in the sixth century
Be) gathered together a group of disciples and founded the Pythagorean
Brotherhood which lasted probably two hundred years. They had strict rules
5
GUIDE TO ANALYSIS

for the conduct of their life and made known most of their mathematical
discoveries as a corporate endeavour. The Pythagoreans were able to prove
that there is no unit of length which can be used for measuring for which the
sides of the square and the diagonal are all whole numbers of units, i.e. they
recognised that the ratio ofthe length of the diagonal to the length ofthe side of
the square could not be expressed as the ratio of two natural numbers. Thus
there are no natural numbers p, q such that (p/q)2 = 2. In Pythagorean
terminology this meant that these two lengths were incommensurable. Thus
the square root of 2 exists as a length in geometry but it is not a rational
number. A quantity like this which cannot be expressed as the ratio of two
natural numbers was not acceptable to the Pythagoreans as a number at all; it
was called a magnitude and a distinction was drawn between numbers and
magnitudes. Today of course we would refer to Pythagorean magnitudes as
irrational numbers.
The strict adherence to the principle that the only numbers are those which
can be expressed as the ratio of two natural numbers gave rise in time to many
paradoxes. Perhaps the most famous were Zeno's paradoxes and the well-
known paradox of Achilles and the tortoise. Two centuries later the situation
was eased by the introduction of definitions and theorems about ratios of
magnitudes.
For us today, perhaps the simplest way of proving that there is no rational
number whose square is 2 is to use a contradiction argument. This method of
proof was invented by the Greeks and has become accepted as a traditional
method of proof. It begins by assuming that the result is false and shows that
this leads logically to a contradiction. Let us suppose that there is a rational
number p/q whose square is 2, where p, q are positive integers with no common
factor. Then

and therefore (1)

Hence p2 has a factor 2 and so p is even. Put p = 2r, where r is an integer in (1),
to obtain
i.e.

Thus q2 has a factor 2 and so q is also even, which means that p, q both have a
factor 2. This contradicts the assumption that p, q have no common factors.
Thus the initial assumption that there is a rational number p/q whose square is
2 leads to a contradiction. This assumption is therefore false, i.e. there is no
rational number whose square is 2 and there is a glaring gap in the rational
numbers. Whether the Greeks used this method to prove their conjecture is
not known. It is possible they used an argument along the same lines, as they
stated that they used a method depending on even and odd natural numbers. It
is easy to show that there are many other such gaps in the rationals. This,

6
 NUMBERS AND NUMBER SYSTEMS

certainly, has rather far-reaching consequences. Perhaps one simple example

gives some idea of the problems encountered because of the gaps in the
rationals.
Consider a very simple function given by f(x) = x 2 (XEQ). Thenf(1) = 1
andf(2) = 4. If we draw the graph of this function then we normally take it
for granted thatf(x) assumes every value between 1 and 4 as x increases from 1
to 2. However, if we only use rational values of x between 1 and 2, thenf(x)
cannot take the value 2, because there is no rational whose square is 2. Thus
f(x) does not take all rational values between 1 and 4 as x takes all
rational values between 1 and 2. This illustrates how our intuitive ideas about
continuous functions fall apart when only rational numbers are used. We must
therefore extend our number system. What we need for the purposes of
analysis is the real number system.

1.4 THE REAL NUMBER SYSTEM

By the fourth century Be the Greeks had seen the need for definitions and
theorems for ratios of magnitudes. Because of their preoccupation with
geometry and in particular their interest in right-angled triangles, the
magnitudes which probably interested them most were square roots of natural
numbers like 2,5, etc. These square roots are, of course, only very special
examples of the numbers we call irrational numbers today. Admittedly they
used the letter 1t for the ratio of the circumference of a circle to its diameter. But
we are not certain whether they knew that it could not be expressed as the ratio
of two natural numbers.
The Greeks had a preoccupation with whole numbers (i.e. natural
numbers). If they wished to consider the ratio oftwo lengths, they tried to find
some sufficiently small unit of length so that the two given lengths could both
be expressed as a whole number of units. When they compared the diagonal of
a square with the sides, their search for a unit of measurement led them to keep
on continually dividing to find smaller and smaller units in the vain hope of
expressing each length as a whole number of units. Thus, when theorems were
devised for ratios of magnitudes, the Greeks had to introduce the idea of
infinite divisibility. Ideas of infinity were foreign to Greek mathematics and
they preferred to avoid them, and to use rational numbers. Many centuries
elapsed before a satisfactory theory of real numbers and of infinite sets began
to emerge.
Probably the most significant advances in these areas began in the second
half of the nineteenth century with the work of five mathematicians-four of
them Germans and one a Frenchman. Their work provided the foundation for
much of the analysis we know today and their names have lived on in many
theorems. They were Karl Weierstrass (1815-97), Eduard Heine (1821-81),
Georg Cantor (1845-1918), Richard Dedekind (1831-1916) and Charles
Meray (1835-97), and they set about the task of defining the phrase 'real
7
GUIDE TO ANALYSIS

number'. The algebraic and order properties of real numbers had long been
known and raised no great problems. For the nineteenth-century mathe-
matician the real task was how to express in mathematical symbols the idea
that the real numbers form a continuum-a property which immediately
distinguishes them from the rational numbers. This property is embodied in an
axiom which we call the completeness axiom for the real number system. It
ensures that there are no gaps in the real number system. For example, we can
use it to show that there is a positive real number whose square is 2.
The completeness axiom can be stated in many equivalent forms. We choose
the formulation in terms ofthe 'supremum' (i.e. the least upper bound) because
it seems the most convenient form to handle in the present context.
We do not set about constructing the real numbers from the rationals.
Instead, we take the viewpoint that the real numbers are the familiar objects
which we have used throughout our time at secondary school and we list a set
of axioms which they must satisfy. These axioms distinguish the real number
system from other number systems (i.e. any number system which satisfies
them is mathematically indistinguishable from the real number system). We
use the letter IR for the set of all real numbers. The axioms (or basic rules) which
must be satisfied fall broadly into three groups. The first group governs the
way in which the familiar operations of addition and multiplication are carried
out.

Algebraic axioms
At For all x,yelR, x + yelR and xyelR.
A2 For all x,yelR,
(x + y) + z = x + (y + z). (associative law)
A3 For all x,yelR,
x+ y=y+x. (commutative law)
A4 There is a number OelR such that
x+O=x=O+x for all xelR.
AS For each xelR, there exists a corresponding number (-x)elR such that
x +( -x)=O=( -x)+ x.
A6 For all x,y,zelR,
(xy)z = x(yz). (associative law)
A7 For all x,yelR,
xy=yx. (commutative law)
A8 There is a number Ie IR such that
x·l =x= l·x for all xelR.

8
 NUMBERS AND NUMBER SYSTEMS

A9 For each XE ~ such that x # 0, there is a corresponding number

(X-l)E~ such that
x(x- 1)= 1 =(x-1)x.
AI0 Forallx,y,zE~

x(y+z)=xy+xz. (distributive law)

The second group of axioms governs the way we use inequalities.

Axioms for order relations

01 Any pair x, y of real numbers satisfies precisely one of the following
relations:
(a) x < y; (b) x = Y; (c) y < x.
02 If x < y and y < z then x < z.
03 If x < y then x + z < y + z.
04 If x < y and z > 0 then xz < yz.

These two groups of axioms are given for reference. Their relevance is not, of
course, restricted to the real number system as the rational numbers also
satisfy them. It is the next axiom (which embodies the idea that the real
numbers form a continuum) which underpins much of our analysis. Since the
completeness axiom is stated in terms of upper bounds, some preliminary
definitions are needed before the axiom can be formulated at all.

DEFINITION 1.4.1 A set S ofreal numbers is said to be bounded above if

there is some number M such that x ~ M for all XES. The number M is called
an upper bound for S.

DEFINITION 1.4.2 A set S of real numbers is said to be bounded below if

there is some real number m such that x ~ m for all XES. The number m is
called a lower bound for S.

DEFINITION 1.4.3 A set S ofreal numbers is said to be bounded if there is

some real number ksuch that Ixl ~ k for all XES.

If the set S is bounded then Ixl ~ k for all XES and some real number k. It
follows that - k ~ x ~ k for all XES and so S is bounded above and also
bounded below, i.e. a bounded set is bounded above and bounded below.

Examples 1.4.1
1. Let S consist of all the real numbers x such that
this in the standard way as
°
~ x ~ 1. We shall write

S = {XE~:O ~ x ~ 1},

9
GUIDE TO ANALYSIS

i.e. S consists of all the x E IR such that 0 ~ x ~ 1. Then S is bounded above

and 1 is an upper bound. Any number M ;::: 1 is also an upper bound for
x ~ 1 ~ M for all XES. Thus S has an infinite number of upper bounds.
Clearly S is also bounded below and 0 is a lower bound. Any number m ~ 0
is also a lower bound as m ~ 0 ~ X for all XES. Moreover, Ixi ~ 1 for all XES
and S is bounded.
2. Let S = {~: nE Z + }, i.e. S consists of the numbers 1,1,1, t, .... Then S is
bounded above and 1 is an upper bound (in fact any number M ;::: 1 is also
an upper bound), S is bounded below and 0 is a lower bound (any number
m ~ 0 is also a lower bound). Moreover, since Ixi ~ 1 for all XES, S is also
bounded.
3. Let S = {XE IR: X > 1}. Then S is not bounded above but it is bounded below
and any number m ~ 1 is a lower bound.
4. Let S = {XE IR: 0 < X < I}, then S is bounded above and any number M ;::: 1
is an upper bound. It is also bounded below and any number m ~ 0 is a
lower bound.

The set S in Example 4 has no least member, despite the fact that it is bounded
below. However its set of upper bounds has a least member, and, furthermore,
the sets in Examples 1 and 2 have the same property. This least element of the
set of upper bounds is called the supremum, i.e. the supremum is the least upper
bound. We also see that the sets in Examples 1 to 4 are bounded below and the
set oflower bounds has a greatest member. This greatest member of the set of
lower bounds is called the infimum, i.e. the infimum is the greatest lower bound.

Examples 1.4.2

1. The set S = {XE IR: 0 ~ X ~ 1} has supremum 1 and infimum 0 and both the
supremum and infimum are elements of the set.
2. The set S = U: nEZ+} has supremum 1 and infimum 0 and the supremum
is an element of S. The infimum, however, does not belong to S.
3. The set S = {XE IR: X > 1} is not bounded above and does not have a
supremum. It is bounded below and its infimum is 1, which does not belong
to the set.
4. The set S = {XE IR: 0 < X < 1} has supremum 1 and infimum 0 and neither
the supremum nor the infimum belong to S.

Perhaps the reader will find it instructive to spend time having a closer look at

10
 NUMBERS AND NUMBER SYSTEMS

some or all of the previous examples. For instance, in Example 4, the reader
will notice that 1 is certainly an upper bound for S, since x < 1 whenever XES.
Moreover, any number less than 1 is not an upper bound. In fact, if s > 0 is any
positive real number (however small), then 1 - s is not an upper bound, since
there are elements of S between 1 - sand 1. Hence 1 is the least member ofthe
set of upper bounds and so it is the supremum of S. These ideas lead to a very
useful characterisation of the supremum. Suppose the set S has supremum IX;
then (a) IX is an upper bound of S, i.e. x ~ IX for all XES and (b) any number less
than IX is not an upper bound; i.e., given any s > 0 (however small), IX - s is not
an upper bound of the set. There is, therefore, some element x' ES such that
IX - s < X' ~ IX. These ideas prove so valuable in practice that they are worth
incorporating in a theorem.

THEOREM 1.4.1 Let S be a non-empty set of real numbers. Then the real
number IX is the supremum of S if and only if both the following conditions are
satisfied:
(a) X ~ IX for all XES;
(b) for every s > 0, there is some X/ES such that IX - s< X' ~ IX.

Proof We notice that condition (a) ensures that IX is an upper bound of Sand
condition (b) guarantees that any number IX - s (which is less than IX) is not an
upper bound. Hence IX is the least of the upper bounds. It follows that if IX
satisfies (a) and (b) then IX is the supremum of S. Conversely, if IX is the
supremum of S, then IX must satisfy (a) and (b).

A similar result holds for the infimum.

THEOREM 1.4.2 Let S be a non-empty set of real numbers. Then the real
number f3 is the infimum of S if and only if both the following conditions are
satisfied:
(a) f3 ~ X for all XES,
(b) for every s > 0, there is some x' ES such that f3 ~ x' < f3 + s.
The first condition, f3 ~ X for all XES, ensures that f3 is a lower bound of S
and the second condition guarantees that any number f3 + s (which is greater
than f3) is not a lower bound. Thus f3 is the greatest of the lower bounds.

These results for the supremum and infimum may at first sight seem trivially
obvious and not particularly worthy of such emphasis. They will, however,
prove to be of great value later.
The supremum (and to a lesser extent the infimum) plays a central role in the
development of the real number system. For this reason it is convenient to
have some abbreviation of the notation. Normally the supremum of a set S is
denoted by either sup S or SUPxeS x. The infimum of S is denoted by either inf S

11
GUIDE TO ANALYSIS

or infxEs x. In the cases in which either sup S or inf S exist they may be elements
of S, but they do not necessarily belong to S.

EXERCISES 1.4.1
1 In each of the following cases, decide whether the set S has (i) a supremum
or (ii) an infimum:
(a) S = {xEIR:a ~ x ~ b};
(b) S = {xEIR:a < x < b};

(c) S={l+(-l)n~ :nEZ+}; (Remember, we have used

Z + for the set of positive
(d) s={n+(-l)n~ :nEZ+}; integers.)

(e) S = {1 + (-l)"n: nEZ+};

(f) S = {XEIR: 3X2 -lOx + 3 < OJ.
In each case in which either the supremum or infimum exists, decide
whether they are elements of S.
2 Let A, B be two non-empty sets of real numbers with supremums (X and /3
respectively, and let the sets A + Band AB be defined by
A + B = {a + b: aEA, bEB},
AB= {ab:aEA,bEB}.
Show that (X + /3 is the supremum of A + B. Give an example to show that
AB need not have a supremum. Prove also that even if AB has a supremum,
this supremum need not be equal to (X/3.

Now that the reader has acquired a certain familiarity with the idea of the
supremum, we use it to formulate the completeness axiom for the real
numbers.

COMPLETENESS AXIOM Every non-empty set of real numbers which is

bounded above has a supremum.

Later, we see that the requirement that the real number system satisfies'the
completeness axiom does indeed guarantee that the real numbers form a
continuum and ensures that we have sufficient numbers for the purposes of
analysis.
In a way, the completeness axiom may at first appear a little lopsided. It is
formulated in terms of upper bounds and the supremum. Lower bounds do
not seem to even merit a mention. However, appearances can at times be

12
 NUMBERS AND NUMBER SYSTEMS

deceptive! The completeness axiom ensures that non-empty sets which are
bounded below have an infimum as the next theorem shows. The completeness
axiom itself is not a theorem which has to be proved. It is a basic property
which a number system is required to possess if it is to be called the real
number system. Our next theorem has a different standing. It is a deduction
made on the understanding that real numbers, by definition, satisfy the
completeness axiom.

THEOREM 1.4.3 A non-empty set of real numbers which is bounded below

has an infimum.

Proof Let S be a non-empty set which is bounded below and let S* consist of
all the numbers of the form -x, where xeS.
i.e. S* = {-x:xeS}.
Since S is bounded below, it has a lower bound m and
m~x for all xeS.
Hence -m ~ -x for all xeS.
Thus - m is an upper bound for S*. Since S* is non-empty and bounded above
it has a supremum IX by the completeness axiom. Hence, using Theorem 1.4.1,
we see that
(a) every element of S* is less than or equal to IX,

i.e. for all xeS,

and
(b) for any 8> 0, there is an element - x' of S* with IX - 8 < - x' ~ IX
i.e. -IX + 8 > x' ~ -IX for some x' e S.

The above statements show that -IX is the infimum of S.

The remainder of the chapter will be devoted to consequences of the

completeness axiom. There are many properties of real numbers which the
reader may regard as obvious and never really question which, surprisingly,
require the completeness axiom for their proof. As the first illustration ofthis
we consider the Archimedean property which states that if x is any real
number then there is an integer greater than x. Of course, the result is trivially
true for x negative. For if x < 0 then zero is an integer exceeding x. It is the
proof for x > 0 which requires the completeness axiom.

THEOREM 1.4.4 (Archimedean Property) Given any real number x, there

is an integer p such that p > x.

13
GUIDE TO ANALYSIS

Proof We employ the traditional contradiction argument which we met

earlier in the chapter. If the result is not true, then there is some real number x*
such that no integer exists greater than x*, i.e. p ~ x* for all integers p. Hence
the set 7L of all integers is non-empty and bounded above by x*. By the
completeness axiom, 7L has a supremum oc, and therefore p ~ oc for all integers p.
Now, if p is an integer then p + 1 is also an integer and so
p+l~oc (1)
(since all integers are less than or equal to oc). Hence, from (1), it follows that
p~oc-1.

Since this is true for all integers p, we see that oc - 1 is an upper bound for 7L,
which contradicts the fact that oc is the supremum of 7L. We therefore have the
necessary contradiction. This contradiction follows as a logical consequence
of the assumption that the result is false. Thus the proof is complete.

COROLLARY Given any real number x, there is an integer n such that

n<x.

Proof We first note that -x is a real number whenever x is real. By the

Archimedean property, there is an integer p > - x and therefore - p < x. Now
- p is an integer whenever p is an integer and the proof is complete.

These properties are certainly of interest to the analyst, but what about our
previous claim that the completeness axiom ensures that the real numbers
form a continuum and there are no gaps? We made a promise that we could
show, for example, that there is a real number whose square is 2. Now is
the time to carry it out. The details of the proof are rather tedious, but the
result itself is exciting.

THEOREM 1.4.5 There is a positive real number c such that c2 = 2.

Proof Let the set S consist of all the positive real numbers x such that x 2 < 2.
i.e. S={xe~:x>Oandx2<2}.

Then 1 eS and S is not empty. Moreover, ifx > 2, then x 2 > 4 and xEtS. Hence
x ~ 2 for all xeS and S is bounded above. By the completeness axiom S has a
supremum c. We will now show that c 2 = 2, by proving (a) c2 <I: 2 and (b)
c2 ::1>2.
(a) We prove c2 <I: 2 This is proved using a contradiction argument. For
suppose the result is false, i.e. c 2 < 2, and write

14
 NUMBERS AND NUMBER SYSTEMS

2 - c2 1 c2
h=--=---- (1)
4 2 4'

Then, clearly,

c
and therefore c<--h' (2)
1-
But from (1),

I.e.

and [cj(1 - h)]ES. Since, from (2), [cj(l- h)] > c = supremum of S,
this gives the required contradiction and we cannot possibly have
c 2 < 2.
(b) We prove c 2 ::t- 2 Again a traditional contradiction argument is used.
For suppose the result is false, i.e. suppose c 2 > 2 and write
k _ c2 - 2 _ 1 1
-2C2-2- c 2 ' (3)

Then, clearly, O<k<t,

and therefore 0< c(1 - k) < C. (4)
Moreover,

[c(l - k)]2 = c 2 (1 - 2k + k 2) > c 2 (1 - 2k) = c2( 1 - {I - c~ }) = 2.

Now if XES, then X2 < 2 < [c(l - k)]2, and therefore 0 < X < c(l- k)
whenever XES. Thus c(1- k) is an upper bound for S, and, from (4),
c(l- k) < c = supS, which gives the needed contradiction. Thus the
assumption c 2 > 2 cannot hold.
Since c 2 1:: 2 and c2 ::t- 2 and c2 exists we must have c2 = 2. This
proves that there is a positive real number whose square is 2. Normally
we use J2for this number. We have already proved that it is not a
rational number and so j2 is a real number which is not rational. Such
a number is called an irrational number.

DEFINITION 1.4.4 A real number which is not a rational number is called

an irrational number.

15
GUIDE TO ANALYSIS

The irrational numbers are the numbers belonging to ~\ Q, i.e. the numbers
which belong to ~ but not to Q. The number .j2 is an example of an irrational
number. Other well-known examples of irrational numbers are 1t and e.
Lambert gave the first proof of the irrationality of 1t in a paper to the Berlin
Academy in 1761. Essentially his argument was quite simple. He showed that if
x is a non-zero rational number, then tan x is irrational. Since tan( 1t/4) = 1
which is rational, the number n/4 cannot be rational, i.e. n is not rational.
Subsequently Legendre proved the irrationality of 1t independently in 1794.
The irrationality of e had already been established by Euler in 1737.
Suppose we now take S* to be the set of all positive rational numbers x such
that x 2 < 2, then S* is not empty and it is bounded above. Ifwe take A to be the
set of all rational numbers a which are upper bounds of S*, then we will show
later that A does not have a least member, which demonstrates effectively that
the set Q of rational numbers does not satisfy the completeness axiom. The
completeness axiom, therefore, plainly distinguishes between the real number
system and the rational number system.

EXERCISES 1.4.2

1 Prove that if x 2 is irrational then x must be irrational. Is the converse true?

2 Show that if x is irrational and y is rational then x + y is irrational.
3 If x is irrational and y is irrational, is it true that x + y is irrational? Give
reasons for your answer. Is xy irrational?

We have already seen that early Greek mathematicians, living centuries before
the birth of Christ, knew of the existence of the irrational number .j2 as a
length in geometry. Their ideas suggest an obvious geometric representation
for real numbers.
Draw a straight line from left to right across the page. Now select a point
on it (called the origin) and label it with the number O. Then choose a unit
of length and label points to the right of 0 which are 1,2,3, ... units of length
from 0 with the numbers 1,2,3, .... Mark points to the left of 0, which are
distances 1,2,3, ... from 0 and label them -1, -2, -3, ... respectively (see
Fig. 1.1). Any point P to the right of 0 on the line represents the real number
x, where x is the length of OP. Any point P' to the left of 0 on the line
represents the real number - x', where x' is the length of OP'. Thus every
point on the line represents a unique real number. Conversely, given any real

p' p
I I I I I I I I I I I
-5 -4 -3 -2 -1 0 1 2 3 4 5
Fig. 1.1

16
 NUMBERS AND NUMBER SYSTEMS

number there is one and only one corresponding point on the line. This, of
course, represents pictorially the idea that the real numbers form a continuum.
Now suppose we take our picture and on the line we initially mark only
the points corresponding to rational numbers. Then, certainly, there will be
some holes. For example, there will be holes at .j2, .j3, )5, ... ,e, 1[, e 2 , 1[2, etc.
But how much ofthe line will be missing and how many holes will there be?
If we mark only the points representing rational numbers, then there will
be 'more holes' than points marked. Without going into detail as to what
precisely is meant by 'more holes', we can justify this statement by proving
that there are so many real numbers that they cannot be arranged in a list.
In fact, there are so many real numbers between 0 and 1 that they cannot
be so arranged. However hard we try and however careful we are in compiling
the list, we will always find that there are so many numbers that some have
been omitted.
Again we use a contradiction argument. Just suppose we can list the real
numbers between 0 and 1 in their decimal representation as

O.a ll a12a13a14···

O.a 21 a22a23a24 .. .

O.a 31 a32a33a34 .. .

O.a 41 a42a43a44 .. .

and recall that a real number has a unique decimal expansion if we exclude
decimals with recurring nines. Consider the real number x = O.b 1 b2 b3 b4 ... ,
where b k = 4 if 0 ~ a kk ~ 2 and bk = 1 if au ~ 3. Then x cannot be equal to
the first number in the list because of the way in which b 1 and all differ.
Similarly, x differs from the second number in the list and so on. Hence x is
not in the list, despite the fact that 0 < x < 1. This gives the required
contradiction, which shows us that the real numbers cannot be arranged in
a list. We express this idea by saying that the set of real numbers is
uncountable.
The set of all real numbers contains so many elements that it is impossible
to arrange them in a list. This contrasts with the set I(Jl of all rational numbers
which can be listed. This means that as far as the number of elements is
concerned the set of irrational numbers swamps the set of rational numbers.
The 'size' of these sets of numbers has been described in a pictorial manner,
which is adequate to give a preliminary idea.
A modern computer uses a decimal with a finite number of digits (which
must represent a rational number) as an approximation to.j2 in calculations.
We do not question the validity of this as we take it for granted that we can
approximate to .j2 by a terminating decimal to any required degree of
accuracy by taking a sufficient number of places after the decimal point. Does
the theory vindicate such an assumption? The answer is yes. For we can

17
GUIDE TO ANALYSIS

prove that given any two distinct real numbers a, b with a < b, there is a
rational number p/q such at a < p/q < b and there is an irrational number x
such that a < x < b. In particular, for each neZ+ there is a rational number
p/q such that J2 - l/n < p/q < J2 and so there is a rational number within
a distance l/n of J2. Similarly, we can find rational numbers within a distance
l/n of n, e, etc. Such statements necessarily require proof.

THEOREM 1.4.6 Let a, b be any two real numbers such that a < b. Then
there is a rational number p/q and an irrational number x such that a < p/q < b
and a<x<b.
(A brief geometric description of the problem helps to outline the method
of proof and shows how the integers p, q are chosen. The numbers a, b can
be represented by points on a line a distance (b - a) apart; see Fig. 1.2.
a b
I
Fig. 1.2

In order to ensure that a point representing a number of the form p/q

falls between a and b we must choose q sufficiently large so that l/q < b - a;
i.e. we want q> l/(b - a). Then we need the smallest integer p such that
p/q > a. This will give a rational number p/q with a < p/q < b. A slight
modification of the argument then provides a proof of the existence of an
irrational number x with a < x < b. Now that the geometrical picture has
pin-pointed the salient ideas we can write out a precise proof.)

Proof Since a < b, we have b - a > 0 and l/(b - a) > O. By the Archimedean
property there is an integer q such that
1
q>-b-·
-a
This integer q is positive, since l/(b - a) > O. Moreover,

1
- < b '-- a. (1)
q

Again using the Archimedean property, there is an integer p' such that
p' > qa. The set of all such integers p' is non-empty and bounded below by
qa. Using Theorem 1.2.1 and the fact that there is an integer less than qa,
we see that this set of integers p' has a least element. Let p be this least
element. Then
p>qa (2)

18
 NUMBERS AND NUMBER SYSTEMS

and p-1 ~qa,

p 1
I.e. -~-+a«b-a)+a=b (3)
q q

using relation (1). Relations (2) and (3) together give

a<r<b
q
and there are integers p, q such that a < p/q < b. Thus there is a rational
number p/q between a and b.
The proof of the existence of an irrational number x such that a < x < b,
is simplified if we stipulate that x should be of some definite form. Since j2
has received considerable prominence in this chapter, we will seek an
irrational number x of the form p* j2/q*, where p* and q* are integers and
q* > O. Obviously the reader could easily adapt the proof to obtain an
irrational number of a different form if so desired.
Using the first part, we see that there is a rational number p*/q* such that
a p* b
j2<q* <j2'

p*j2
I.e. a<--*-<b.
q
Now that it has been proved that there is an irrational x such that a < x < b,
a further application proves that there are irrationals Xl' X2' such that
a < Xl < x and x < x 2 < b. By repeated applications there are an infinite
number of irrationals between a and b; similarly, there are an infinite number
of rationals between a and b.

COROLLARY Let a,b be two real numbers such that a < b. Then there
are an infinite number of rationals p/q with a < p/q < b and an infinite number
of irrationals x such that a < x < b.

The section closes with a result which demonstrates that the rationals do
not satisfy the completeness axiom. Let S be the set of all positive rational
numbers x such that x 2 < 2,
i.e. S= {XEQ:X > 0 and x 2 <2}.
Then S is not empty. Moreover, if x ~ 2, then x 2 ~ 4 and x¢S, the number
2 is, therefore, an upper bound for S. Let A be the set of all rational numbers
IX such that IX is an upper bound of S,

i.e. A= {IXEQ:IX is an upper bound of S}.

19
GUIDE TO ANALYSIS

Then it can be proved that A does not have a least member, i.e. the
completeness axiom is not satisfied for Q. We first note that if aeA then
IX> ..ji. For if a < .J2, then there is a positive rational number p/q such that
IX < p/q <.J2. Since p/q <.J2, it follows that (p/q)2 < 2 and p/qeS and so a
is not an upper bound of S. Thus if IX <.J2 then a¢:A. Hence if lXeA, then
IX <I: .J2. Since J2 is irrational and all the elements of A are rational numbers
we see that if aEA, then IX> J2.
Now suppose 1X1 is any element of A. Then 1X1 >.J2 and so there is a
rational number p'/q' such that 1X1 > p'/q' >.J2. Moreover if xeS, then
x 2 < 2 < (p'/q,)2 and so p'/q' is an upper bound of S. Hence p'/q'eA and it
follows that if a l is any element of A then there is an element p'/q' of A with
p'/q' < a l . Thus A does not have a least element.
This demonstrates that (!) does not satisfy the completeness axiom, and
therefore the completeness axiom certainly distinguishes between ~ and Q.

APPENDIX: MATHEMATICAL INDUCTION

Suppose that there is a statement T" (involving n) associated with each positive
integer n. For example, T" might be the statement
1 + 2 + 3 + ... + n = 1n(n + 1)
It is certainly easy to check that this result is true for the first few cases
n = 1,2,3,4,5. To deal with the statement for all neZ+, the principle of
mathematical induction is needed.

Principle of mathematical induction

Suppose that there is a statement T" associated with each positive integer n. If
(a) the result is true when n= 1 (i.e. T. is true), and
(b) the result is true for n = k whenever it is true for n = k - 1 (i.e. 1k - 1 true
implies 1k true),
then T" is true for all positive integer n.
It is not difficult to see how this works. The underlying idea is quite simple.
Suppose we check in turn the statements for n = 1, followed by n = 2, then
n = 3, etc., and we record for each either T for true or F for false. Assumption (a),
that the result is true for n = 1, means that we begin with T. The hypothesis that
the result is true for n = k whenever it is true for n = k - 1 means that we can
never go from Tto F, i.e. we cannot have a true statement succeeded by a false
one. Since the first statement is true, we will record T, T, T, T, T, T, ... along the
line, and the result is true for all natural numbers. This describes the main
features. A proper proof depends on properties of the integers proved in the
preceding chapter.

20
 NUMBERS AND NUMBER SYSTEMS

Proof of the principle of mathematical induction

Suppose the conditions (a) and (b) are satisfied. Then either (I) the result is true
for all nEZ+ or (II) the result is not true for all nEZ+. We will show that (II)
leads to a contradiction.
Case (II) If the result is not true for all nEZ+ then by the corollary to
Theorem 1.2.1 there is a smallest natural number m for which the theorem is
talse and m ;;?; 2 because of condition (a). Hence the result is true for n = m - 1.
Condition (b) now ensures that the result is true for n = (m - 1) + 1, i.e.
n = m and we have the required contradiction. Hence case (II) is impossible and
so we must have case (I), i.e. T" is true for all nEZ+.

Examples

1. Prove that
1 + 2 + 3 + ... + n = !n(n + 1)
Solution In this case use T" for the statement
1 + 2 + 3 + ... + n =tn(n + 1).
Then clearly Tl is true. Moreover, if 1k-1 is true for some k ~ 2, then
1 + 2 + ... + (k -1) = t(k -1)k.
Add k to both sides:
1 +2+ ... +(k -1)+k=t(k-l)k+k=tk(k+ 1).
Hence 1k-l true implies 1k is also true. Since TI is true it follows by mathematical
induction that T" is true for all nEZ +.
2. Prove that 5311 - 1 + 19.34 (11-1) is divisible by 44.

Solution Let T" stand for the statement

5311 - 1 + 19.34 (11-1) is divisible by 44.
When n = 1, 5311 - 1 + 19.34 (11-1) = 52 + 19.30 = 25 + 19 = 44 and the result is true
when n= 1.
Suppose now that the result is true for n = k - 1 where k is some integer with
k~2.
i.e. 53(1-1)-1 + 19.34 (k-I-l) = 44p,
where p is an integer. This gives
531 - 4 + 19.3 4 (1-2)=44p. (1)
Multiply equation (1) by 53 to obtain
53k - 1 + 19.3 4 (1-2), 125 = 125.44p,
i.e. 53k - 1 + 19.34 (k-2)·(81 +44)= 125.44p,
i.e. 53k - 1 + 19.34(1-1) + 19.34(1-2),44 = 125.44p,
which gives 531 - 1 + 19.34(1-1) = 44(125p - 19.34(1-2».

21
GUIDE TO ANALYSIS

Thus 53k - 1 + 19.3 4 (k-l) is divisible by 44 and it therefore follows that 7;.-1 true
implies 7;. is also true. Since the result is true when n = 1, it follows by the principle of
mathematical induction that T" is true for all nEZ+.
Note: It is not necessary to start with n = 1 as the first case. Thus condition (a) could
be replaced by: Tn. is true for some positive integer no' The conclusion would then be
altered to read: T" is true for all n ~ no.

MISCELLANEOUS EXERCISES 1
1 The set A consists of all the real numbers of the form 1/2m + 1/3" + 1/5 q and
the set B consists of all the real numbers of the form 1/2m - 1/3", where
m, n, q are positive integers. Decide which of the following exist:
(a) supA, (b) sup B, (c) inf A, (d) infB.
Find those which exist.
2 Let A be a set of positive real numbers with infimum 0( > O. Let the set A-I
be defined by

Show that A -1 has a supremum. What is its value?

3 Decide which of the following statements is true and which is false.
(a) A set of real numbers which has a maximal and minimal element is
finite.
(b) Every non-empty set of rational numbers which has a rational upper
bound always has a rational supremum.
(c) If A, B are two non-empty sets of real numbers with 0( = sup A and
/3 = sup B and if the set 2A + 3B is given by
2A + 3B = {2a + 3b:aEA,bEB},
then 2A + 3B is bounded above and
sup(2A + 3B) = 20( + 3/3.
In the case of a true statement give a proof. In the case of a false statement
give a counter-example.
4 Let D, E be two non-empty sets such that D s; E and E is bounded above.
Show that
supD~supE.

22
 NUMBERS AND NUMBER SYSTEMS

5 Determine the supremum of each of the following sets.

(a) {1+~:n=1'2,3, .. -}, (b) {1-~:n=1,2,3, .. }

(c) {xEIR:3x2-10x+4<1}.
6 Let X be a set of positive real numbers with supremum IX. Write
y = {x 2 : XEX}. Show that 1X2 is the supremum of Y.

HINTS FOR SOLUTION OF EXERCISES

Exercises 1.4.1
1 (f) The set S is equal to {XEIR: i- < x < 3}, since
3X2 - lOx + 3 = (3x - 1)( x - 3) < 0 if i- < x < 3.

Exercise 1.4.2
1 Show that the square of a rational number is rational.
2 Use the fact that the difference of two rational numbers is rational. It
follows that if (x + y) is rational and y is rational then (x + y) - y is also
rational.

Miscellaneous Exercises 1
5 (c) The set is the set of all real numbers x for which 3x 2 - lOx + 3 < 0,
which we met earlier.

ANSWERS TO EXERCISES
Exercises 1.4.1
1 (a) (i) supS=b, (ii) infS=a;
the supremum and the infimum both belong to S.
(b) (i) sup S = b; (ii) inf S = a;
neither the supremum nor the infimum belongs to S.
(c) (i) supS=~, (ii) infS=O;
the supremum and infimum both belong to S.
(d) (i) S is not bounded above and so there is no supremum;
(ii) inf S = 0 and inf S belongs to S.
(e) S is not bounded above and it is not bounded below. Hence S does not
possess a supremum and it does not possess an infimum.
(f) (i) supS=3, (ii) infS=i-;
neither the supremum nor the infimum belongs to S.

23
GUIDE TO ANALYSIS

2 The following are just two possible examples:

(a) A=B={XEIR:X<O}. In this case AB={XEIR:X>O} which is not
bounded above.
(b) A = { -2, -1,0}, B = { -1,0, 1}. In this case AB = { -2, -1,0, 1,2}
and sup A = 0, supB = 1, supAB = 2.

Exercises 1.4.2
1 The converse is not true. For example, ../2 is irrational but its square is
rational.
3 False. Counter-example: x = J2, y = 1 - ../2. In this case x + y is rational,
but x and yare both irrational.
Not necessarily. If x = J2, y =../2, then x and yare both irrational,
but xy = 2, which is rational.

Miscellaneous Exercises 1
1 (a) supA = ~~, (b) supB=t, (c) infA = 0, (d) inf B = -j-.
2 sup A -1 = 1/1l.
3 (a) False. Counter-example: the set S = {XE IR: 0 ~ x ~ 10} has a maximal
element 10 and a minimal element 0, but S has an infinite number of
elements.
(b) False. Counter-example: let S = {XEQ: x > oand X2 < 2}. Then 2is an
upper bound of S, for every XES satisfies the inequality x ~ 2. However
the supremum of S is J2 which is irrational.
(c) True.
5 (a) The supremum is 2. (b) The supremum is 1.
(c) The supremum is 3.

24
2 SEQUENCES

2.1 INTRODUCTION
Early records of classical Greek mathematics show a marked aversion for any
methods involving infinite processes. For centuries this remained the prevail-
ing attitude among mathematicians. Indeed, there was no significant change
until the time of Sir Isaac Newton. In 1669 he composed his famous treatise De
Analysi per aequationes numero terminorum infinitas (published in 1711) which
included an account of his work on infinite series as well as the beginnings of
calculus. Because ofthe work of Newton the use of infinite processes became
regarded as a legitimate mathematical tool. In time, it was recognised that
careful rules needed to be developed governing the use of such infinite
processes and the calculus in order to ensure the validity of the final results,
and thus real analysis came into being. The rigorous subject as we know it
today stems, in the main, from the work of eighteenth and nineteenth-century
mathematicians on the continent. Many years of work by some of the most
outstanding mathematicians of the last few centuries have produced the final
polished version.
A fundamental concept is that of a limit and it is probably simplest to
introduce it initially in the context of sequences. Certainly an understanding of
limits of sequences is needed before we can tackle the problem of manipulating
infinite series. Mathematicians of Newton's time were not so careful how they
handled infinite series, but then they were occasionally forced to dismiss
certain results because they were patently absurd! Had they been in possession
of the knowledge of analysis which we have today, they would never have
obtained their occasional absurd results, but without them mathematicians
might never have seen the need to investigate the problem of what rules are
needed to govern the use of infinite processes. Nowadays we have precise laws
governing the manipulating of infinite sequences, infinite series and function
theory and we can be certain that our results are right provided that we have
followed the dictates of analysis correctly. With a view to presenting a logical
progression we begin with this chapter on infinite sequences.

25
GUIDE TO ANALYSIS

2.2 SEQUENCES
A sequence of real numbers is a succession of real numbers in a definite order
so that we know which number is in the first place, which number is in the
second place and for any positive natural number n we know which number is
in the nth place. Normally, the subscript n is used for the number in the nth
position and this number is called the nth term ofthe sequence. For example, if
the letter a is used for a particular sequence then an is normally used for the
number in the nth position. The sequence then starts a 1,a2'a3'a4,a S,a6' .... This
sequence would be denoted by either (a n ):'= 1 or just (an)-the latter notation
being used when it is felt that there is no need to emphasise the values which n
may assume. The former is particularly useful if the context does not make it
clear what values are assumed by n or if we wish to omit a few terms of a
sequence. Suppose we take the sequence a 1 , a 2, a 3, a4 , as, ... and we decide to
omit the first four terms and take as our new sequence as, a6, a 7 , .... Then we
could write (a n ):'= s for this new sequence and the subscript n = 5 at the bottom
of the bracket tells us we are going to begin by using as as our first term. Of
course, many other letters like b, s, x, etc. are frequently used for sequences,
giving the sequences (b n), (sn), (x n), etc.

Examples 2.2.1

2. The sequence « -1)"~):'= 1 starts

3. Let xn=cosnn(ne:r). Then (Xn):'=l is the sequence

-1, + 1, -1, + 1, -1, + 1, ....
4. The sequence (t/n2):,= 1 starts 1, t,~, /6"" . If we decide to omit the first four
terms and start with ls, {6' i9"" then this would be denoted by (l/n 2):'= s·

Now that we have a general idea of what is meant by a sequence, we can look
back and see that in each case the sequence provides a list of real numbers in a
definite order. We know precisely which real number is in the nth position and
so, corresponding to each positive integer n, there is a definite real number,
viz. the real number in the nth position. Thus a sequence provides a mapping
from the positive integers Z + into the reals. In fact this is probably the best
way to formulate the definition of a sequence.

DEFINITION 2.2.1 A sequence of real numbers is a function/: Z+ -+ R

26
 SEQUENCES

With function notation the real number fen) corresponds to the positive
integer n. In fact, we hardly ever use function notation. The number fen) is
normally replaced by a number of the form an> using the common subscript
notation.
For sequences the important thing is the order in which the terms appear
and the general trend of behaviour as we move along the sequence. Let us start
with some familiar examples where the pattern of behaviour seems very
obvious.

Examples 2.2.2

1. Let an=~(nE;r). This sequence starts l,t,j-,t,!,i-,~,i, .... Without too

much thought the reader will inevitably say that this sequence tends to zero,
and may very well be happy to write an -+ 0 as n -+ 00.
2. Let an = n + ( -1)n~(nEZ+). This sequence starts 0,2t,2~,4t,4~,6i,64,8i, ...
and again the reader will probably conclude that this sequence tends to
infinity and might even write an -+ 00 as n -+ 00.
3. The first example could be modified by alternating the signs +, - to give
1, -t, j-, -t,!, -i,··· .
No doubt the reader would still claim (and rightly claim) that the sequence
tends to zero.
4. Let an = 1-~(nE;r). This sequence begins o,t,~JJ,i,~,~, ... and it
would seem reasonable to say that an tends to 1 as n -+ 00.
5. Let an = G] + ( -1)"t(nEZ+), where [x] is the integer part of x, i.e. it is the
greatest integer p ~ x. The sequence now starts -t, t, t, i, t, f, i,1, .... The
2000th term a2000 is 1000 + 1, and the 2000 OOOth term is 1 000000 + t.
Despite the fact that the size of the terms fluctuates, it still seems reasonable
to say that an tends to infinity as n tends to infinity. If we draw a graph which
shows an plotted against n then this claim seems almost self-evident (see Fig.
2.1).

Now let us use these examples to pick out the salient facts. First let us try to
decide what properties are necessary if a sequence tends to infinity. Obviously
it is not enough to ensure that the terms get larger and larger. For instance,
Example 4 provides a sequence whose terms get larger and larger as n
increases, but we have already decided that it does not tend to infinity. Indeed,
it was claimed that this sequence tends to 1. Furthermore, Example 5 shows a
sequence in which the terms do not consistently get larger and larger and yet it
does tend to infinity. What leads us to this conclusion? If we consider the graph
of the sequence in Example 5, we notice that if we draw a horizontal line across

27
GUIDE TO ANALYSIS

an

6.0
5.5

5.0
4.5 X
4.0
3.5 X X
3.0
2.5 X X

2.0
1.5 X X

1.0
0.5 X

0.0 n
-0.5 X

-1.0
Fig. 2.1

the paper at a height 10 above the n-axis then all but the first few terms are
above this line. Similarly, if we draw a line at a height 100 above the n-axis, then
all but the first 201 terms are above this line. In general, if we take any positive
real number A (however large) and draw a horizontal line across the paper at a
height A above the n-axis, then all but a finite number of terms at the beginning
of the sequence are above the line, i.e. we can find N such that all the terms
aN + 1, aN + 2, ••. are above the line (see Fig. 2.2). There is therefore some N such
that an > A for all n > N. Obviously, the value taken for N will depend on the
size of A. For example, if we choose A = 10 in Example 5, then we notice
an> 10 = A for all n > 21, i.e. we can use N = 21 in this case. For if n > 21,
then n ~ 22 and
an = [!] + ( -1)nt ~ [¥] +I( -1)nt = 11 + (-I)nt > to.
If we take A = 100 in Example 5, then
an> l00=A for all n > 201.
In general, given any A> 0, we can always find a corresponding N. If we
choose N to be the smallest integer such that N > 2A + 3, then
an = [1] +( -l)nt~ [A + 2] +( _1)n!> A
for all n > N, since n > N implies n ~ 1 + N > 2A + 4.

28
 SEQUENCES

x X

X X
A

X X
AI
X

o N n
Fig. 2.2

Clearly, arguments of the same type could be used for the sequence in
Example 2. In this case, given any A > 0, all we need to do is to choose N to be
the smallest integer greater than A. Then, for all n > N, n ~ N + 1 and

However, the sequence in Example 4 does not have these properties. For if we
choose any number A greater than 1, then all the terms an < A. We are now in a
position to formulate the first definition.

DEFINITION 2.2.2 A sequence (an) of real numbers tends to infinity if(and

only if), given any A> 0 (however large), there exists a corresponding positive
integer N = N(A) such that

foralln>N.

We write an-+oo as n-+oo.

(In the definition we have written N = N(A) to emphasise the fact that the
value chosen for N depends on the value given to A initially. In general, larger
values of A give rise to larger values of N.)

29
GUIDE TO ANALYSIS

Examples 2.2.3

1. Let a" = In, then given any A > 0, let N be the smallest integer such that
N~A2. For all n>N,

2. Let a" = log n, where here and throughout this volume 'log' means 'the
logarithm to the base e' (i.e. the natural logarithm). Then, given any
A> 0, let N be the smallest integer such that N ~ eA. For all n > N,
a" = logn > logN ~ log(e A ) = A
and a" ..... 00 as n ..... 00.
n2Jn+n2+ 1
3. Let a" = n2 _ 43 (n ~ 7).

Then, for n ~ 7,

n2Jn n2Jn
a,,> n2 - 43 >--2-=Jn·
n
(1)

Given any A > 0, choose N to be the smallest integer such that N ~ A 2 and
N~7, i.e. N~max{A2, 7}. Then, for all n>N,

a,,>Jn>JN~A
and a,,"'" 00 as n ..... oo.

In Examples 1 and 2 it was easy to find a simple formula for N in terms of A.

However, in Example 3 it proved expedient to first use an inequality to obtain
a" > Inand then proceed to obtain a value for N corresponding to any given
A > 0. For any particular value of A, this would not necessarily give the
smallest possible corresponding value of N, but it will give a value of N which
will do. To demonstrate that a value of N exists, we need only find a value
which will do; we do not have to find the smallest one possible. Thus in
Example 3 we just need to show that there is some N such that

n2Jn+n 2 + 1
n2 _ 43 >A for all n > N;

we do not need to find the smallest N satisfying this inequality.

If we write b" = In,
then Example 3 reads as follows: For all n ~ 7, a" > b"
and b" ..... 00 as n -+ 00. Hence a" ..... 00 as n ..... 00.
The reasoning behind this type of method is patently obvious if we use a

30
 SEQUENCES

graphical representation. Draw a graph showing the sequences (an) and (b n)

on the same diagram. Then for each n ~ 7, the cross marking an is above the
cross for the corresponding bn • Now, if any line is drawn across the paper
at a height A > 0 above the n-axis then there is some N such that bn is above
this line whenever n > N. Hence an must also be above this line for all n> N.
In practice, this type of method is extremely useful and it is, therefore,
incorporated in a theorem.

THEOREM 2.2.1 Let (an) and (b n) be two sequences of real numbers, such
that

for all n ~ No (where No is some given positive integer). If bn --+ 00 as n --+ 00,
then this implies that
as n --+ 00.

Proof Let A > o. Since bn --+ 00 as n --+ 00, there is a positive integer N' such
that
bn > A for all n > N'.
Choose N = max { No,N'}. Then for all n > N,

Hence an --+ 00 as n --+ 00.

This result is frequently used in conjunction with the triangle inequalities

(see appendix at the end of this chapter):
Ilxl-lyll ~ Ix+ yl ~ Ixl + Iyl;
Ilxl-lyll ~ Ix- yl ~ Ixl +IYI·

Examples 2.2.4

1. Let an = n2 + n cos me. Then

an = n2 + ( - 1tn ~ n2 - n ~ n2 - tn 2 = tn 2 (n ~ 2).
Write bn = tn 2• Then bn--+ 00 and hence an --+ 00 as n --+ 00.
n2+Jn
2. Let an = . Then, for n> 1,
n+cosn

31
GUIDE TO ANALYSIS

Write bn = tn. Since an> bin> 1) and bn-+ 00 as n -+ 00, we see that an -+ 00
asn-+oo.

EXERCISES 2.2.1
1 Let an = log(logn) (n ~ 2). Given A> 0, find a formula for N such that
an > A for all n > N.
2 Prove that each of the following sequences (an) has the property that an -+ 00
asn-+oo:

(a) an = n + cosn (nEZ+),

(C) an = 2 n +.7 (nEZ.

+)
+smn
3 Show that if an -+ 00 as n -+ 00 and c is a positive real number, then can -+ 00

The preceding concepts all have analogues corresponding to the case

an -+ - 00 as n -+ 00. In this case we need to draw lines across the graph an
arbitrary distance A below the n-axis, and we require the existence of a
corresponding N such that the points representing an are below this line for
n> N (i.e. an < - A for all n > N). This gives the following definition.

DEFINITION 2.2.3 Let (an) be a sequence of real numbers. Then an -+ - 00

as n-+ 00 if (and only if) given any A > 0, there exists a.corresponding positive
integer N = N(A) such that
for aU n>N.

A comparison of the definitions shows immediately that an -+ - 00 as n -+ 00

if and only if - a. -+ 00 as n -+ 00. It also suggests that Theorem 2.2.1 has an
obvious analogue.

THEOREM 2.2.2 Let (an), (b n) be sequences of real numbers such that

for all n ~ No, where No is some given positive integer. If bn -+ - 00 as n -+ 00,

then this implies that an -+ - 00 as n -+ 00.

Proof Let A > O. Since bn -+ - 00 as n -+ 00, there is an integer N' such that
for all n> N'.

32
 SEQUENCES

Let N = max{No,N'}. Then for all n> N

and so an --+ - 00 as n --+ 00.

At the beginning of this section we met several examples of sequences which

seemed to have a limit which is not infinite. For example, let an = 1 - ~(nEZ+).
This gives a sequence which begins with 0, t,~, i,~, ... and appears to tend to 1.
Its graph is shown as Fig. 2.3.
We notice that we can ensure that an is within a given (arbitrarily small)
distance of 1 for all sufficiently large n. For example, 1 - < an < 1 + for all t t
n > 2, I - 150 < an < I + 150 for all n> lOO. In fact, if we take a pair of lines
across the graph, one at an arbitrarily small distance B below I and the other B
above 1, then the points representing an are between these lines for all

°
sufficiently large n, i.e. I - B < an < 1 + B for all sufficiently large n. For give.n
any B> (however small) we notice that
I-B<an<I+B for all n > liB.
If we choose N to be the smallest integer such that N ~ liB, then
I-B<a n < I +B for all n > N,
I.e. lan-II<B for all n > N.
These ideas are just what we need to define a finite limit.

t---------------------------------------l+e
1.0 t-------------------
'L )( X X X1 _ e

X
x X
X
X

1..1--
0.0 ~J.~I.....-I.L....--I..I....- 1 ---L.1----L.1---L.1---L.1---L.1----L1----L1----L1----l1_
1..L.
o 2 3 4 5 6 7 8 9 10 11 12 13 14 15 n
Fig. 2.3

33
GUIDE TO ANALYSIS

DEFINITION 2.2.4 Let (an) be a sequence of real numbers. Then we say that
an tends to I as n -+ 00 if given any e > 0, there is a corresponding N = N(e) such
that
Ian -II < e foralln>N,
i.e. l-e<an<l+e for all n> N.
We write an -+ I as n -+ 00 or limn-+ooan = I.
Note: The Greek letter e has been in common usage in the definition of a limit
for over a century. Influenced by Weierstrass's lectures, Heine first used the
symbole in his definition of the limit ofa function in his Elemente in 1872. His
definition'is in much the same form as we use today.

Example 2.2.5

n2 + n + 1
Let an = 2n 2 + I (neZ+). Then it seems reasonable to guess that an -.! as
n -+ 00. To justify this assertion we notice that
1 n2 +n+ 1 1
an - 2= 2n 2 +1 - 2
2n+ 1
2(2n2 + 1)'
1 2n+n 2n+n 3
Hence lan -2"1 < 2'2n 2 < 2'2n 2 = 4n (for n> 1),

and therefore, given e > 0

1 3
Ian - 2"1 < 4n < e
for all n > max{1,1.}. Thus if N is chosen to be the smallest integer such that
N~max{1,1.}, then

for all n > N.

EXERCISE 2.2.2

Use the definition to check that 3n + 1 -+ 3 as n -+ 00, by showing that, for any
n+2

34
 SEQUENCES

8 > 0, there is a corresponding N such that

3n + 1 _ 31 <8
1n+2 for all n>N.

The preceding work seems to suggest that a sequence cannot have more than
one limit and this is indeed true, as we now prove.

THEOREM 2.2.3 A sequence of real numbers cannot have more than one
limit.

Proof Let (an) be a sequence of real numbers. Suppose first that an -.1 as
n -. 00 and an -. m as n -. 00. Then either I = m or I =F m. We will show that the
assumption I =F m leads to a contradiction.
Case I =F m If I =I- m, then I - m =F 0 and II - mI > o. Using Definition 2.2.3
in the special case 8 = !II- ml, we see that as an -.1, there is an integer No such
that
for all n > No. (1)
Similarly, since an -. m, there is an integer N 1 such that
lan-ml<!II-ml for all n>N 1 • (2)
Now, II-ml = II-a n+ an-ml ~ II-ani + lall-ml
=lall-II+lan-ml,
and so, for n>max{No,N 1 }, we have
II-ml ~ lan-II + lall-ml <!Il-ml +!Il-ml = II-ml
from (1) and (2). The inequality
II-ml<ll-ml
provides the contradiction we were seeking.
Thus the assumption I =F m leads to a contradiction and is, therefore,
untenable. We must therefore have 1= m. This means that a sequence cannot
have two different finite limits.
Moreover, if all-.I as n -. 00, then there is N such that

for all n > N. Hence an ~ 00 as n -. 00 and an ~ - 00 as n -. 00. Thus an cannot

have a finite limit and also an infinite limit.
Finally, an cannot tend to both 00 and - 00. The results together show that a
sequence cannot have two different limits.

35
GUIDE TO ANALYSIS

DEFINITION 2.2.5 A sequence (an) of real numbers is called a convergent

sequence if an tends to a finite limit as n -+ 00.

From Theorem 2.2.3 we see that a convergent sequence has a unique limit.

DEFINITION 2.2.6 Ifan-+oo as n-+oo then we say that (an) diverges to

infinity. If an-+ - 00 as n-+ 00, we say that (an) diverges to minus infinity.

There are, of course, sequences which do not have any limit at all. For example,
the sequences [( -1 }n]:,= I and [( -1)"n ]:'= I do not have any limit.
It would be very difficult if we had to use the definitions every time we
needed the limit of a sequence. Using sand N each time would involve a
tremendous amount of work. It would be rather like going back to first
principles in calculus every time a derivative is required. We must, therefore,
now develop some new tools for use with sequences.

DEFINITION 2.2.7 A sequence (an) of real numbers is called a bounded

sequence if lanl::::; M for all n and some positive real number M.

THEOREM 2.2.4 A convergent sequence of real numbers is bounded.

Proof Let (an) be a convergent sequence of real numbers. Then there is some
real number a such that an -+ a as n -+ 00. Since an -+ a as n -+ 00, there is a
positive integer N such that
Ian - al < 1 for all n > N.
Thus lanl < 1 + lal for n > N. Let
M = max{lall, la 2 1, la 3 1, ... , IaN-II, laNI, 1 + lal}.
Then lanl::::; M for all n and (an) is a bounded sequence.

THEOREM 2.2.5 (Algebra of Limits)

If an -+ a and bn-+ b as n -+ 00, then, as n -+ 00
(a) an+bn-+a+b,
(b) anbn-+ ab,
() an a prOVl·dedb =F o.
c b-+/J'
n

36
 SEQUENCES

Proof
(a) Let e > o. Since a. -+ a and b. -+ b, there is a positive integer N such
that
a - e/2 < a. < a + e/2 for all n> N,
b - e/2 < b. < b + e/2 foralln>N.
By addition,
a + b - e < a. + b. < a + b + e for all n> N
and a. + b.-+a + bas n-+ 00.
(b) Since b.-+b as n-+ 00, the sequence (b.) is convergent and so it is
bounded by Theorem 2.2.4. Thus
(1)
for all n and some suitable positive real number M. Let e > O.
Then
e
--->0.
M+lal
Since a. -+ a and b. -+ b, there is some N such that

for all n> N (2)

and Ib.-bl < M + lal for all n> N. (3)

For all n> N,

la.b. - abl = la.b. - abo + abo - abl
= I(a. - a)b. + a(b. - b)1
:::;; la. - allb.1 + lallb. - bl
e e
< M +lal M +la lM +Ial =e
using (1), (2) and (3). Thus a.b.-+ab as n-+ 00.
(c) In view of (b), it is sufficient to prove that ~ -+~ as n -+ 00. Since b =F 0,
n
there is some positive integer N' such that
for all n> N'.
Hence, for all n > N'
Ib.1 = lb. - b + bl ~ Ibl-Ib. - bl > Ibl-tlbl =tlbl,
37
GUIDE TO ANALYSIS

i.e. Ib"l>tlbl for all n> N'. (4)

Let e > O. Since b" -+ b as n -+ 00, there is some N (with N ~ N') such
that

Ib,,-bl <el~12 for all n> N. (5)

Hence, for n > N,

1 11 Ib-blll elbl 2 2 1
1bll -b = Ib"llbl <-2-· fbi· fbI=e

from relations (4) and (5) and it follows that ; -+~ as n-+ 00. It now
a 1 1 a "
follows from (b) that b" = a" . "b-+ a . b = bas n-+ 00.
" II

The proof ofthe algebra oflimits may have appeared long and tedious, but its
manifold usefulness amply repays us for all the work, as the next example
illustrates.

Example 2.2.6

Let

. 7 5 8
1+-+-+-
n n n 2 1+ 0 + 0 + 0
3 5 1
Then a =-------+ -
" 3 27 5 +0+0 5
5+-+-5
n n
as n -+ 00, by Theorem 2.2.5.

Its use can be greatly extended if it is used in conjunction with the so-called
'sandwich rule' and some standard useful limits. The idea of the 'sandwich
rule' or 'squeeze rule' is to sandwich the sequence we are considering between
two sequences which are known to have the same limit I. The middle sequence
must then have the same limit I.

THEOREM 2.2.6 (Sandwich Rule) Let (a,,), (b,,), (e,,) be three sequences of
real numbers such that all ~ b" ~ ell for all n > No, where No is some positive
integer. If a" -+ I as n -+ 00 and ell -+ I as n -+ 00, then b" -+ 1 as n -+ 00.

38
 SEQUENCES

Proof Let e > O. Since an -+ I and en -+ I as n -+ ro there is some N (with

N ~ No) such that
l-e<an <I+6 for all n > N
and 1- 6 < en < 1+6 for all n>N.
Thus, for all n> N,

and therefore bn -+ I as n -+ 00.

THEOREM 2.2.7 Let (en) be a sequence of real numbers such that en ~ 0 for
all n > No, where No is some given positive real number. Suppose also that
en -+ e as n -+ 00. Then
e~O.

Proof The proof employs a traditional contradiction argument. Let us

suppose that
e<O. (1)
Use the definition of a limit with the particular value 6 = tiel, i.e. 6 = -e/2
because e < O. Since en -+ e as n -+ ro, there is some N (with N ~ No) such
that
le.-el<6= -te (2)
for all n> N. Now en ~ 0 for n ~ Nand e < 0, and hence (2) yields
en -e=lcn- el< -te
for all n > N,
i.e.
(by relation (1)) for all n > N. This is impossible as en ~ 0 for all n > No and we
have the required contradiction. Thus the only possibility is that e ~ 0 and the
proof is complete.

COROLLARY Let (an),(b n) be two sequences of real numbers such that

a. ~ bn for all n > No, where No is some given positive integer. Suppose also
that an -+ a, bn -+ b as n -+ 00. Then
a~b.

Proof Use the theorem with en = bn - an, and the result b - a ~ 0 is an

immediate consequence.

39
GUIDE TO ANALYSIS

This result shows that weak inequalities are preserved by limits. Is the same
true for strict inequalities? Surprisingly the answer is no. For suppose an < bn
for all n > No. Then it is certainly true that a", bn must satisfy the inequality
an ~ bn for all n> N. Thus if an -+ a and bn-+ b as n -+ 00, Theorem 2.2.6
guarantees that a ~ b. It is, however, possible to have a = b and we must,
therefore, always use the weak inequality a ~ b for the limit.

Example 2.2.7

Let an = 1/n2, b" = l/n for n = 1,2,3, .... Then clearly

an<b n
for all n> 1. However, an -+ 0 and bn-+ 0 as n -+ 00. The two sequences,
therefore, have the same limits despite the fact that an < b" for all n> 1.

THEOREM 2.2.8 Let (an) be a sequence of real numbers. If an -+ 00 as n -+ 00,

then l/an-+O as n-+ 00.
Proof Let e > O. Then l/e > O. Since an -+ 00 as n -+ 00, there is a positive
integer N such that
foralln>N.
Hence, for all n > N,

and l/an -+O as n-+oo.

We now return to look at some examples.

Examples 2.2.8

. 1 cos n I l 1 .
1. Smce -- ~-- ~-, and--+O as n-+ 00, and ---+0 as n-+ 00, It follows
n n n n n
from the sandwich rule that
cosn
---+0 as n-+oo.
n
40
 SEQUENCES

cosn 6
1 + - - + -5
n 5 + n4 cos n + 6 n n 1+ 0 + 0
--+
2. 4n5 + n3 + cos n 1 cosn 4+0+0
4 +2- + -5-
n n
=± as n--+oo.
3. Let (an) be a sequence of non-negative real numbers and let a ~ O. Suppose
that a; --+ a2 as n --+ 00. Prove that an --+ a as n --+ 00.

Solution We consider two separate cases, as follows.

(a) Case a = 0 In this case a; -+ 0 as n -+ 00. If e > Ois any positive number, then e2
is also positive and so there is some N such that
la;1 < e2
for all n > N,
i.e. lanl < e
for all n > N and an -+ 0 as n -+ 00 as required.
(b) Case a> 0 Since an ~ 0 and a> 0,
la; - a2 1= I(a n - a)(an + all = Ian - al(an + a) ~ Ian - ala. (1)
Now given any e > 0, the number ea is also positive and hence there is some N
such that
la; _a 2 1<ea
for all n> N. From relation (1) we see that
Ian - ala ~ la; - a2 1< ea
for all n> N,
i.e. lan-al < e
for all n > N and an -+ a as n -+ 00.

The tools we have developed obviously can be used to produce slick and easy
answers, but they must not be abused. Before the conclusion of any theorem is
applied, the reader must check that all the required conditions are satisfied. If
they are not satisfied, then the theorem is not applicable. For example, if
b = 0 the reader must never use the rule an/bn--+a/b. The correct use of the
algebra oflimits will never produce quantities like~, 00/00,0'00,00 - 00. If
the reader is ever faced by such monstrosities as the answer to a problem
then it is as a result of incorrect use of the theorems and further work on
these lines is a waste of time. The only solution is to go back to the beginning,
start again and remember not to break the rules this time!
The scope of our methods is greatly increased if we add certain standard
limits to our array of available tools.

41
GUIDE TO ANALYSIS

2.3 STANDARD LIMITS

2.3.1 na

As n- 00,
if oc > 0,
if oc = 0,
if oc < o.
For if IX > 0 and A > 0, then nlZ > NIZ ~ A for all n > N, where N is the smallest
integer such that oc log N ~ log A;
I.e. N ~ e(logA)/".

Hence nlZ _ 00 as n- 00 for IX> o.

When oc = 0, nlZ = 1 and therefore nlZ _ 1 as n _ 00.
When IX < 0, write p = - oc so that p > O. Then
1
nlZ=n-fJ=p_O
n

2.3.2 Ii'

I
As n-oo
00 if a > 1,
an _ 1 ifa=l,
o if lal < 1.
and a" does not tend to any limit if a ::;:;; - 1.
If a> 1, we can write a = 1 + h where h > O.

Hence an = (1 + h)" = 1 + nh + (~)h2 + ... + h" ~ 1 + nh.

Since h > 0, 1 + nh - 00 as n - 00 and therefore a" - 00 as n - 00 when a > 1.

°
If a = 1, then an = 1 for all nand a"-l as n- 00.
If < a < 1, then we can write a = lib where b> 1. Thus
1
a"=--O as
n b
since bn _ 00 as n- 00.
If a = 0, then an = 0 for all nand an_o as n- 00.
If -1 <a<O, then
-lain::;:;; a"::;:;; lal"
and laln_O as n- 00 because 0 < lal < 1.

42
 SEQUENCES

By the sandwich rule an -+ 0 as n -+ 00.

If a ~ -1, then an = (_l)nlal n.
Since laln~ 1 for all n, an cannot tend to any limit as n-+oo.

2.3.3 all/s"
For a > 1 and ex > 0, it has already been proved that an -+ 00 and nIX -+ 00 as
n-+ 00, but what happens to the quotient? Does it tend to any limit at all?
If it does have a limit, is the limit finite or infinite? Suprisingly, the answer
is that an/nIX -+ 00 as n -+ 00 for a > 1 and ex > O. If we take a close to 1 and
ex very large, then this seems at first sight somewhat unexpected. For example,
we are claiming that
( 1.000 000 00 1)"
nlOOOOOOOOl -+ 00

as n -+ 00, which seems quite incredible and the use of a large modern computer
to find the first million terms would be of little help in making this look even
plausible. Yet our analysis shows that the claim is indeed true. The proof is not
unduly difficult. The first step is to introduce an integral value for the power of
n. Let a > 1, ex > 0 and choose an integer p such that p ~ ex. Then

(1)

and it is sufficient to show that an/n P -+ 00 as n -+ 00. The actual details of the
proof can often be obscured by the notation in the general case and so we
illustrate the method by looking initially at the special case p = 2. Since a > 1,
we can write a = 1 + h, where h > O. Thus for n ~ 4

(1 + h)" ~(1 h n(n -1)h2 n(n -l)(n - 2) h3 hn)

n2 n2 +n + 2! + 3! + ... +
> n(n - 1)(n - 2) h3 = (n - 1)(n - 2) h3
n2 3! 6n
tn· tnh3 nh 3
>~ =24'

Now nh 3/24 -+ 00 as n -+ 00 and, therefore an /n2 -+ 00 as n -+ 00. A similar

argument can now be used to show that for all positive integers p, an /n P -+ 00 as
n -+ 00. In view of (1) and Theorem 2.2.1 we now have an/nIX -+ 00 as n -+ 00 for
a> 1 and ex > O.
Frequently it is useful to take the reciprocal of an/nIX. Since an/nIX-+ 00 as
n -+ 00, it follows that nIX/an -+ 0 as n -+ 00 for a > 1 and ex > O. The substitution
b = 1/ a then gives the useful form of the limit, viz.
nIXbn-+O as n-+ 00 for 0 < b < 1 and ex> O.

43
GUIDE TO ANALYSIS

By using the sandwich rule this result can be extended to show that for a> 0
and Ihl < 1,

as n---. 00.

EXERCISE 1.3.1

Choose any positive real number a and use a pocket calculator to find the first
few terms ofthe sequence (a l / n ):= 1. How does a l / n behave as n increases? Try a
few more positive values for a and check what happens to al/n.

2.3.4 slln

No doubt the previous exercise will have led the reader to the conclusion that
for all a > 0, a l /n ---.1 as n -400. Is this really correct? The answer is yes, as we
now prove.
First assume a> 1 and write a l /n = 1 + hn. Since a l /n > 1 when a > 1, we see
that hn > O. Moreover,

_( h)n- h
a- 1+ n - 1+ n n + n(n-1)
2!
2
hn + ... + hn
n

and therefore
a
0< hn <-.
n
By the sandwich rule, hn ---.0 as n ---. 00 and a l/n = 1 + h n -4 1 + 0 = 1 as n -400.
-If a = 1, then a l /n = 1 for all nand a l /n -41 as n-4 00.
If 0 < a < 1, then we can write a = lib, where b> 1. Then

a l /n = (b1)1/n =b 1l /n-4I=1
1

as n -4 00.
Thus for all a> 0, a l /n ---.1 as n ---. 00.

2.3.5 ,rln

Using a method similar to the one above, it can be shown that nl/n -41 as
n---' 00. Write nl/n = 1 + k n. If n > 1, then nl/n > 1 and k n > 0 for n> l.
Now, for n ~ 2,

-(1 k )n_ n(n-l) 2 n(n-1) 2

+ 1 + nkn + kn + ... + k n >
n
n- n - 2' kn
. 2!

44
 SEQUENCES

and therefore

(n ~ 2).

By the sandwich rule k" -.0 as n -+ 00 and nl/" = 1+ k" -.1 + 0 = 1as n -. 00.
2.3.6 tI'/ n!
Let a be any real number. Choose N to be the smallest positive integer such
that N ~ 21 a I. Then for each integer p ~ N,
lal lal 1
-~-~- (1)
p '" N '" 2'
For all n > N,

o~ la"1 =~. _la_l_ lal .laIN~!.! !lal N

'" n! n (n-l)"'(N+l) N! "'2 2'''2 N!

=(!)"-N~
2 N!

and ("2l)"-N -.0 as n-. 00. Hence la"1

n! -+0 as n-+ 00 and so a"
n! -+0 as n-' 00.

2.3.7 n!/n D
For all n

0< n! = n(n-l)".2 ·1 = 1(1-!)(1-~).,,~. !~!.

n" n" n n n n n
By the sandwich rule n !/n" -.0 as n -+ 00.

By using the algebra oflimits, the sandwich rule and the standard limits we can
now deal fairly simply with a large variety of sequences.

Examples 2.3.1

n77n+n 5 5n n7(i)"+n 5 (i)" 0+0

1. = -.--=0
3n + 8" (i)" + 1 0+1
as n -. 00 using the algebra of limits and subsections 2.3.2 and 2.3.3.
8"
1+-
n! + 8" n! 1 +0
2. - - = - - - . - - = 1
7" + n! 7" 0+1
-+1
n!
as n -. 00 using the algebra of limits and subsection 2.3.6.

45
GUIDE TO ANALYSIS

3. Find the value oflim n _ oo (4 10 + r")l/l, where r is any positive real number.
Solution
If 0 < , ~ 1, then
(4 10 )1/11 ~ (4 10 + ,11)1/11 ~ (410 + 1)1/11.
Now as n ..... 00, (4 10 )1/11 ..... 1 and (4 10 + 1) 1/11 ..... 1.
By the sandwich rule,

lim (4 10 + r")l/" = 1

when 0 <,~ 1.
If, > 1, then r" ..... 00 as n ..... 00 and so there is an integer N such that r" > 4 10 for all
n > N. It follows that for n > N
r = (,")1/" < (4 10 + r")I/" < (r" + r")1/11 = 2 1/o ·r.
Since 2 1/ 0 ..... 1 as n ..... 00, the sandwich rule shows that

lim (4 10 + r")I/" = r

when r> 1.

EXERCISES 1.3.1

Determine whether the following sequences have a limit. If the limit exists then
find it.

1 (COS(;II) ). 2 11 n3 911)
2 (n 10 + 211
1+ 7 .

3 ( n (n
+ ( - 1)"
2 + 1)1/2
In) . 4 (sin[ :2++
5 :J)-
5
(In(Jn 2 + ~ P+i»).
-
6 C"+41)-
2" + 5"

7 e" + 5 ).
3"
11 8 C 00" + 211)-
n! + 99"
2
9 Cn2 +2n+ 1). 10 (3n + 2n + 1 ).
n2 + 1
3"]
n+l
11 ([2 10 + (1)"] 1/11). 12 ([2 10 + 1/").
3 2
13 ( n 3- n 2cos n + 2 )
4n + n -4sinn .
14
([3:: 4T)- 2

Give reasons for your answers.

46
 SEQUENCES

2.4 SOME GENERAL RESULTS FOR SEQUENCES

We begin by considering increasing sequences and decreasing sequences. Not
surprisingly, we start in the time-honoured way with some definitions. The
first book of Euclid's elements opens abruptly with twenty-three definitions-
it has no introduction at all. Initially we content ourselves with a mere five
definitions!

DEFINITION 2.4.1 A sequence (an) of real numbers is called an increasing

sequence if an ~ an + 1 for all n. It is called a strictly increasing sequence if
an < an+ 1 for all n.
In a way the terminology is very obvious. A sequence whose terms strictly
increase as n increases is called a strictly increasing sequence.

DEFINITION 2.4.2 A sequence (an) of real numbers is called a decreasing

sequence if an ~ an + 1 for all n. It is called strictly decreasing if an > an + 1 for all n.

DEFINITION 2.4.3 A sequence of real numbers is called a monotonic

sequence if it is either an increasing sequence or a decreasing sequence.

DEFINITION 2.4.4 A sequence (an) of real numbers is said to be bounded

above if the set {an: nEZ+} is bounded above, i.e. if there is some M such that
an ~ M for all n.

DEFINITION 2.4.5 A sequence (an) of real numbers is said to be bounded

below if the set {an: nEZ+} is bounded below, i.e. if there is some m such that
m ~ an for all n.

If we draw graphs illustrating increasing sequences, then it looks as ifthere are

only two possible types of behaviour. Either the sequence is bounded above
and tends to a finite limit or the sequence is not bounded above and tends to
infinity. Using the completeness axiom it can easily be shown that this is
indeed the case.

THEOREM 2.4.1 Let (an) be an increasing sequence of real numbers. Then

either
(a) (an) is bounded above and an tends to a finite limit as n-+ 00, or
(b) (an) is not bounded above and an-+ 00 as n-+ 00.

47
GUIDE TO ANALYSIS

Proof
(a) Let S = {an: nE£:+}. If (an) is bounded above, then the set S is non-
empty and bounded above. By the completeness axiom it has a
supremum. Let a = sup S. Then, given any e > 0, a - e is not an upper
bound of S. Hence there is some XNES such that

Now the sequence (an) is increasing, and therefore for all n > N,

Hence xn -+a as n-+ 00.

(b) If( an) is not bounded above, then, given A > 0, there must be some term
X N such that

Since (xn) is increasing,

for all n > N and so Xn -+ 00 as n -+ 00.

Since a convergent sequence is bounded (see Theorem 2.2.4), we have the

following corollary.

COROLLARY An increasing sequence of real numbers converges if and

only if it is bounded above.
Note: The reader is advised to look back to Definition 2.2.5 in which a
convergent sequence is defined to be one with a finite limit.
Similar results hold for decreasing sequences.

THEOREM 2.4.2 Let (an) be a decreasing sequence of real numbers. Then

either
(a) (an) is bounded below and (an) converges, or
(b) (an) is not bounded below and an -+ - 00 as n -+ 00.

COROLLARY A decreasing sequence of real numbers converges ifand only

if it is bounded below.

These results are readily proved by applying the previous theorem and its
corollary to the sequence ( - an).
The previous two theorems, taken together, give the following theorem.

THEOREM 2.4.3 A bounded monotonic sequence converges.

48
 SEQUENCES

We first apply Theorem 2.4.1 to the sequence [( 1 + ~),,]:'= 1 to obtain a very

useful standard limit. As n--+ 00, 1 +~--+ 1 and it is a very common error to
imagine that this implies that (1 + ~t also tends to the same limit. However,
this seems much less likely when one realises that for each nE71+

(1+-nl)n =1+n (1)

-n + n(n-l)(1)2
2!
-n + ...
(n -1)
=1+1+~+ ...

~2.

Thus each of the terms (1 + lint

~ 2, and, if there is a limit, this limit must be
at least 2. A quick calculation ofthe first few terms suggests that the sequence is
increasing, which is indeed true. For let

Then

a=l+n (-n1) + n(n-l)(1)2

n
2!
n!(I)n
-n + ... +--
n! n

+~!(I_~). .. (I_n:l)

an +1 = 1+ 1+ ;! (
1- n~ 1 ) + ...
+~(1- n~l)(I- n~I).··(I-:~~)+ ...
+ ~!(1- n~l).··(l- :~~)+(n~l)!(l- n~l).··(l- n:l}
an an an
The first two terms of and + 1 are the same; the rth term of + 1 is greater
than the rth term of an for 3 ~ r ~ n + 1. Finally, an + 1 has an extra positive
term

1 (1 -n+l
(n+l)! -1)(
-
... 1 -n+1 -n)
-

at the end. Hence an < a and (an) is a strictly increasing sequence.

n+ 1
Moreover,

an = 1 + 1 + ;! (1 - ~ ) + ;! (1 - ~ ) ( 1 _ ~ ) + ... + ~! (1 _ ~ ). .. ( 1 _ n: 1 )
49
GUIDE TO ANALYSIS

1 1 1
:::;1+1+2+2'2+"'+2n-1

1 1 1 1
since - = :::; = ----.=--t (r ~ 2).
r! r(r-1) ... 2·1 2·2·2 ... 2·1 2
1- (t)n 1
Hence an < 1 + - 1
1. < 1 + - 11. = 1 + 2 = 3,
-2 - 2

and the sequence (an) is also bounded above. By Theorem 2.4.1 (an) converges,
i.e. an tends to a finite limit as n --+ CXJ. We denote the value of this limit bye,
which gives

(1+~r--+e as n--+ 00.

Note,' This is a special case of the rule (1 + i-)" --+ eX as n --+ 00, which will be
proved later in the volume.

Sometimes sequences are defined by a recurrence relation rather than an

explicit formula. Let us look at a particular example.

Example 2.4.1

The sequence (an) is defined by

(n = 1,2,3, ... ).

(a) Using induction it can be proved that

2<an <3
for all n. Obviously the result is true when n = 1, since a 1 = ~ by
definition. Now, suppose that it is true for n = k - 1, where k is some
integer such that k ~ 2. Then
2 < ak - 1 < 3. (1)
Now, 5ak = a~-1 + 6 < 32 + 6 = 15
by relation (1), and also
5a k = a~-1 + 6 > 22 + 6 = 10.
Hence 2 < ak < 3 and the assumption that 2 < ak - 1 < 3 implies that
2 < ak < 3. Since the result is true when n = 1, it follows by induction
that

50
 SEQUENCES

(2)
for all n.
(b) We now show that (an) decreases. Using the definition, we see that
an+ 1 - an = !(a; + 6) - an
= !(a; - 5an + 6)
=!(an - 3)(an - 2) < 0
since 2 < an < 3 from (a). Hence (an) is decreasing. Since it is bounded
below it converges. Let a be its limit. Then an -+ a as n -+ 00 and so
an + 1 -+ a as n -+ 00. But
a n + l =!(a; + 6)-+t(a 2 + 6)
as n -+ 00. Since the limit is unique,
a =!(a 2 + 6),
i.e. a2 - 5a+ 6 =0
which gives a = 2 or a = 3. Since the sequence is decreasing, a ~ a l =!
and so the only possibility is a = 2.

EXERCISE 2.4.1
The sequence (an) is defined by a l = 1, an+ 1 = 4 + 2a!/3 (nEZ+). Use mathe-
mati<;:al induction to prove that 1 ~ an ~ 8 for all nEZ+. Show also that
(a n+ llan ) ~ 1 for all n. Deduce that (an) converges and find its limit.

2.5 SUBSEQUENCES
If we take a sequence (an) and delete either a finite or an infinite number of
terms then the resulting progression is called a subsequence of the original
sequence. For example, suppose we delete all except every third term. This
gives a 3 , a6 , a 9 , a 12 , .•• and this is a subsequence of the original sequence.
Normally the first term of the subsequence is given the subscript n l , the second
is given the subscript n 2 , and so on. Thus we would write an" an" an" an., .. . for a
subsequence of a l , a 2 , a 3 , ••• • Normally such a subsequence would be denoted
by (an)f'=l. We notice that we must keep to the order of the original sequence
and so

51
GUIDE TO ANALYSIS

DEFINITION 2.5.1 Let (nk}f'= 1 be a strictly increasing sequence of positive

integers, i.e. 1 ~ n1 < n2 < n3 < .... Then (an)f'= 1 is called a subsequence of
(a n }:'= 1·

Thus a subsequence is obtained by taking the original sequence and deleting

terms (without any change of order).

Example 2.5.1

Let an=~ (nEZ+). Then the sequence (an) begins 1,!,!,!,t,!, .... Hence
1,!,t,~, ... is a subsequence of (an). Another example of a subsequence is
!,!,!, k, .... We could of course produce a subsequence by deleting all the
terms ~ for which n is not a prime. This would leave the subsequence

As we might expect, every subsequence of a convergent sequence has the

same limit as the original sequence.

THEOREM 2.5.1 Let (an) be a convergent sequence of real numbers and

suppose an -+ a as n -+ 00. Then every subsequence also converges to a.

Proof Let (an)k"= 1 be any subsequence of (an). Then

n 1 < n2 < n3 < n4 < ...
and nk~k (I)
for all k. Since an -+ a as n -+ 00, given any 6 > 0 there is a corresponding N such
that
for all n>N. (2)
Now if k > N, then nk > N by (I) and therefore
lan, -al <6 for all k> N.
Hence an, -+ a as k -+ 00.

COROLLARY If (an) has two subsequences which converge to different

limits then (an) cannot converge.

Using a similar proof we can show that the following results hold for sequences
which diverge to infinity or to minus infinity.

52
 SEQUENCES

THEOREM 2.5.2 Let (an) be a sequence of real numbers such that an -+ 00 as

n-+oo. Then every subsequence of (an) also diverges to 00.

THEOREM 2.5.3 Let (an) be a sequence of real numbers such that an -+ - 00

as n -+ 00. Then every subsequence of (an) also diverges to - 00.

Examples 2.5.2

1. Let an = ( - l)n. Then a ln = + 1, al n+ I = -1. Thus the subsequence

al , a4 , a6 , ••• converges to 1 and the subsequence ai' a 3 , as, . .. converges to
- 1. Since there are two different subsequences with two different limits we
see that an does not tend to any limit as n -+ 00.
2. Let bn = an where a is some real number with a ~ - 1.
Then b l ,. = al ,. = lal 2,. and b 2,.-1 = a 2n - 1 = _laI 2n - 1 •
If a = -1, the subsequence (b 2 ,.)::'= I converges to + 1 and the subse-
quence (b 2,.-I):,"',.1 converges to -1. Hence a" cannot tend to any limit as
n-+ 00.
If a < -1, the subsequence (b 2n ):,"',. 1 diverges to 00 and the subsequence
(b 2n - I):,"',. 1 diverges to - 00. Hence, again, a" does not tend to any limit as
n-+ 00.

3. Let an = (1 + ~)"(nE.z+).
Then ([ 1 + f,;-] 2n):,= I is the subsequence (a211):,"',.1 of (all). Since the original
sequence converges to e, every subsequence converges to the same limit e.
1 )2n
Hence lim ( 1 +-2 =e. (1)
"-+00 n

Now [ ( 1 + 21 n)"T = ( 1 + 21nyn, and therefore, from (1),

lim
n-+ 00
(1 + 21 ),. =
n
Je.
4. Let (a n)::'= I be a sequence of real numbers. Suppose we know that the
subsequence (a 2n )::'= 1 converges to a, and the subsequence (a 2n +d::'= 1
converges to a, then we can show that the original sequence (an) also
converges to a.
Since (a2n) converges to a, we know that given any f: > 0, there is some N'
such that
la 2n -al < f:
for all n > N'. However, (a2n+ d also converges to a and hence there is some

53
GUIDE TO ANALYSIS

Nil such that

Ia2n + 1 - a I < e
for all n > Nil. Let N = max{2N' + 1,2N" + 2}. Then for all n > N, we
have

and so an -+a as n -+ 00.

EXERCISES 2.5.1

1 By using the fact that (1 + Ijnt-+e as n-+oo, show that the following
sequences are convergent and find their limits in terms of e:

2 Let (an) and (b n) be two convergent sequences of real numbers such that
an -+a as n -+ 00 and bn-+b as n -+ 00. Prove that, if (an + ( -l)Rbn) and
« -It+ 1 an + bn ) both converge, then a = b = O.
3 Let (a n ):'= 1 be a sequence of real numbers. Suppose that each of the
subsequences

converges to the limit a. Prove that the original sequence (an) converges
to a.
4 Let (a n}:'= 1 be a sequence of real numbers. Let p be any positive integer such
that p ~ 2. Suppose each of the p subsequences (a pn }:'= l' (a pn + 1):'=o,
(apn + 2}:'=0,"" (apn + p- 1 }:'=o converges to the limit a. Prove that the
original sequence (an) converges to a. (The reader should memorise this
result as it is used in the next chapter-see Theorem 3.3.6 on the
alternating series test and the Appendix on rearrangements.)

The definition of convergence for a sequence (an) involved an investigation of

Ian - ai, where a is the limit of the sequence. We therefore have a criterion for
deciding whether (an) converges to a, but we have no means, at present, of
deciding whether a sequence converges if we do not know or are unable to
guess a suitable value for the limit. What we need is a criterion which involves
only the actual terms of the sequence and makes no mention of the value of the
limit. The Cauchy criterion for convergence provides the answer. In the
process of leading up to this proof we will need to show that every infinite
sequence has a monotonic subsequence-a result which is of interest in its
own right.

54
 SEQUENCES

THEOREM 2.5.4 Every infinite sequence of real numbers has a monotonic

subsequence.

Proof Let (an) be an infinite sequence of real numbers. We call a p a terrace

point if ap ~ an for all n ~ p. The sequence either has an infinite number of
terrace points or it does not.
(a) If (an) has an infinite number of terrace points, these can be labelled
an"an"an,,·· . where n 1 < n 2 < n3 < .... Since an, is a terrace point,
an, ~ an for all n ~ n 1 and so in particular an, ~ an,· Similarly, an, ~ an"
an, ~ an. and so on. Thus (an.)r'= 1 is a decreasing subsequence of (an).
(b) If (an) does not have an infinite number of terrace points, then it only
has a finite number (which may be zero). We can therefore find nl so
that all the terrace points ap occur for values of p < n 1 • Since an, is not a
terrace point, there exists n 2 with n 2 > n 1 such that an, ~ an,. Now an, is
not a terrace point and so there is an integer n3 > n 2 such that an, > an,.
Continuing in this way gives a sequence an" an" an" . .. such that
an, < an, < an, < .... Thus (anJr'= 1 is a strictly increasing subsequence
of (an).
Cases (a) and (b) together show that (an) has a monotonic subse-
quence.

COROLLARY Every bounded infinite sequence has a convergent subse-

quence.

Proof A bounded infinite sequence has a bounded monotonic subsequence.

This subsequence is convergent by Theorem 2.4.3.

With the help of this result we can now deduce Cauchy's criterion for
convergence.

THEOREM 2.5.5 (Cauchy's Criterion for Convergence)

Let (an) be a sequence of real numbers with the property that for every e > 0
there is corresponding N such that
for all m,n > N. (I)
Then (an) is convergent.

Proof From (I) we see that there is an integer M such that Ian - am I < 1 for all
m,n > M. In particular, Ian - aM + 11 < 1 for all n> M. Hence
lanl = Ian - aM+ 1+ aM+ 11 ~ Ian - aM+ 11 + laM+ 11 < 1 + laM+ 11

55
GUIDE TO ANALYSIS

for all n> M. Let

K = max{la11, la 2 1, la 3 1, ... , laM-II, laMI, 1 + laM+11}.
Then lanl ~ K for all n and the sequence (an) is bounded. By the corollary to the
previous theorem (an) has a convergent subsequence. Let this subsequence be
(anX'= 1 and suppose an. -+ a as k -+ 00. Let e > 0 be any positive real number;
then there exist positive integers No, Ko such that
for all m, n > No (1)
and for all k> Ko. (2)
Let N=max{No,Ko}. Since nk~k, we see that for all k>N
Iak - a I = Iak - an, + an, - a I ~ Iak - an. I + Ian. - a I
<e/2 +e/2=e
using (1) and (2). Hence ak-+a as k-+ 00 and the sequence (an) converges.
The converse is, of course, relatively easy to prove. Suppose the sequence
(an) converges to a. Then given e > 0, there is a corresponding N such that
Ian - al < e/2 for all n>N.
Thus ifn,m>N,
Ian - am I = Ian - a + a - am I ~ Ian - a I + Ia - am I < e.

APPENDIX: TRIANGLE INEQUALmES

Expressions involving both moduli and inequalities are common in analysis.
For example, we have the requirement Ian - al < e in the definition of a limit.
The reader, therefore, cannot avoid inequalities and would be wise to learn
how to handle them.
In analysis, the triangle inequalities prove invaluable. They are

lixl-Iyll ~ Ix + yl ~ Ixl + Iyl, (1)

Ilxl-lyll ~ Ix - yl ~ Ixl + Iyl, (2)

for all real numbers x,y. In order to prove them we recall that Ixl = x if
x ~ 0 and Ix I = - x if x < O. Hence
Ix12=X2 (3)
for all real x, and therefore

Ixl=+P·
Now for all real numbers x,

-x~+P,

56
 SEQUENCES

i.e. x:::;;lxl; -x:::;;lxl· (4)

Thus if x, yare any two real numbers, we have
Ix + yl2 =(x + y)2 =x 2 +2xy+ y2 = Ixl2 +2xy+ lyl2
by (3) and therefore
Ix + Y12:::;; Ixl2 + 21xYI + lyl2 = (Ixl + lyl)2
from (4). Taking positive square roots gives
Ix + yl :::;; Ixl + Iyl· (5)
Similarly, using (4) we have
Ix + yl2 = Ixl2 + 2xy + lyl2 ~ Ixl2 - 21xyl + lyl2 =(lxl-lyI)2.
Again take the positive square root to obtain
Ix + yl ~ lixl-Iyil· (6)
The outer modulus on the right-hand side of (6) occurs as there is no
information about whether Iyl is larger than Ixl or not. Inequalities (5) and
(6) together give the triangle inequality (1). Now that one of the triangle
inequalities has been established there is no problem in proving the other.
We simply need to note that 1- yl = Iyl, and the result follows immediately.
How do we use the triangle inequalities? Each time we have to decide
which part of the inequalities to use. There is one golden rule governing the
choice - select the part of the inequality which gives the inequality sign the
required way for the current problem. Remember also that if 0 :::;; a :::;; band
0:::;; e :::;; d, then ae:::;; bd, but these inequalities give no information about the
relative sizes of ale and bid. Thus inequalities may be multiplied, but they
cannot be divided. Let us try some examples.

Examples

1. Let a,b,e,d be real numbers such that lei ~ Idl. Then

0< Ilel-Idil:::;; Ic+dl:::;; Icl + Idl,
and therefore
o< 111
;!( - - ;!( .,--------,- (*)
lei + Idl ~ Ie + dl ~ Ilcl-ldil"

Since 0:::;; Iial-ibil ~Ia+ bl:::;; lal+ Ibl

and inequalities can be multiplied, we have
lial-Ibil la+bl la+bl lal+lbl
Icl+ldl :::;;Ic+dl = c+d :::;;llcl-ldl!"

57
GUIDE TO ANALYSIS

The reader should pay particular attention to the denominators. The signs
have been determined by inequality (*). It is the choice of the correct
expression for the denominator which frequently causes confusion-so
beware!
2. If Ix I ~ 1, then Ix - 31 ~ 4
and Ix 2- 31 ~ II x 21- 1311 = 3- IX 12 ~ 3- 1 = 2.
1 1
Hence ~--~-
Ix 2 -31 ~2
and therefore

whenever Ixl ~ 1.

TABLE OF STANDARD LIMITS

(IX> 0),
I (2.3.1) A, n~ 00. n" ~ ~
( (IX = 0),
(IX < 0).
(a> 1),
2 (23.2) As n~oo. a"~! ~ (a = 1),
(-1 <a< 1).
If a ~ - 1, then an does not tend to any limit as n -+ 00.
an
3 (2.3.3) If IX > 0 and a> 1, --+ 00 as n-+ 00.
n'"
If IX > 0 and Ibl < 1 then n"'bn-+O as n-+ 00.
4 (2.3.4) For a> 0, a 1/n-+ 1 as n-+ 00.
5 (2.3.5) As n -+ 00, nl/n -+ 1.
an
6 (2.3.6) For all real numbers a, ,-+0 as n-+ 00.
n.
n!
7 (2.3.7) As n-+ 00, --+0.
nn

8 As n -+ 00, ( 1 + ~ ) -+ eX for every real number x. In particular,

58
 SEQUENCES

as n-+oo.

MISCELLANEOUS EXERCISES 2
1 In each of the following cases, decide whether a. tends to a limit as n -+ 00.
When the limit exists find it. Give reasons for your answers.
n2 10" + n 3 9" n2 + 7
(a) a"= 72 "+ 1 (b) a'=2+n 3 •

2'· n 2 + 3"
(c) a. = 3" + nlOO . (d) a" = (3n 2 + n)l/n.

2n 3 + 3n + 1 2n 3 + 3n + 1
(e) a = (0 a• =
" 3n 3 +4 3n 2 +4
6'+ n! (n + I)"
(g) a = n! + (7)2"· (h) a• =
" n"
3"+(-2)"
(i) a =
" 1 +n!

2 For x ;;J!: 0, let

a,,(x) = (1 + X")l/".
Prove that ifO~x~ 1, then a,,(x)-+1 as n-+oo.
Show also that if x > 1, then a,,(x)-+x as n-+ 00.
3 The sequence (a,,) is defined by

al = 1, a" + 1 = Jl + a~/2 (n = 1,2,3, ... ).

Show that (a) a~ - 2 < 0, (b) a~+l - a~ > o.
Deduce that (a,,) converges and find its limit.
4 For each of the following sequences (a,,), decide whether a" tends to a
limit as n -+ 00. When the limit exists, find it.
(a) a" = [4 10 + (t)"] 1/"; (b) a" = (4 10 + 2")1/";
3n 3 + 1 3n 3 + ncos 2 n
(c) a _ . (d) a .
" - n3 + n2 ' " = n3 + sin 2 n '

59
GUIDE TO ANALYSIS

(g) an = n(Jn 2 + 144 - p=1).

Give reasons for your answers.
S Give an example of sequences (a,,), (b,,) which do not converge for which
[a,,+(-I)"b.] and [(-I)"+l a,,+b,,] both converge. (See Exercises
2.5.1, Question 2.)

HINTS FOR SOLUTION OF EXERCISES

Exercises 2.3.2
The algebra of limits is used in many of these parts. In addition, the following
are also used:
1 Use 2.3.3. 2 Use 2.3.3. 4 Use 2.3.3
5 Use Theorem 2.2.1 and the identity a2 - b2 = (a - b)(a + b). This allows
1/(Jn2 + 2 - P+i) to be expressed as Jn 2 + 2 +.J1i2+1.
6 Use 2.3.2. 10 Use Theorem 2.2.1.
11 and 12 Use 2.3.4 and the sandwich rule.

Exercises 2.5.1
1 [(1 + 1/3n)3,,] is a subsequence of the original sequence and so has the
same limit;

(I-n" ~(.:I)" ~ (I +~)'

and
1
( 1 + n-l
),,-1 -+e
n-l

as n-+oo;

( 1_~ __1 ) = (1-!)(1 +~).

2n 2n 2 n 2n

Miscellaneous Exercises 2
4 (g) Use the identity a 2 - b2 = (a + b)(a - b) with a = Jn 2 + 144,
b=p=1.

ANSWERS TO EXERCISES

Exercises 2.2.1
1 N is the smallest integer such that N ~ eeA •

60
 SEQUENCES

Exercises 2.3.2
1 1, 2 0, 3 1, 4 0, 5 00, 6 0, 7 00, 8 0, 9 3, 10 00, 11 1,
II 3, 13 i, 14 H>5.

Exercises 2.S.1
1 e 1/ 3 , lie, Je, 1/Je.
Miscellaneous Exercises 2
1 In each case an tends to a limit as n -+ 00. The limits are given below, with
reasons in brackets.
(a) 0 (algebra of limits and 2.3.3).
(b) 0 (algebra of limits). (c) 1 (2.3.3).
(d) 1 (sandwich rule and 2.3.5). (e) i (algebra of limits).
(f) 00 (theorem 2.2.1). (g) 1 (2.3.6).
(h) e (see section following Theorem 2.4.3).
(i) 0 (2.3.6).
3 .j2.
4 In each case an tends to a limit as n -+ 00. The limits are given below with
reasons in brackets.
(a) 1 (sandwich rule and 2.3.4). (b) 2 (sandwich rule and 2.3.4).
(c) 3 (algebra of limits). (d) 3 (algebra of limits and sandwich rule).
(e) 00 (theorem 2.2.1). (f) 0 (2.3.3). (g) 11 5 (algebra of limits).
5 One example is an = 1 + ( - Wand bn = - 1 + ( - l)n for n = 1,2, 3,4, ....
In this example an + (-l)nbn = 2 and (-It+ 1an + bn = -2.

61
3 INFINITE SERIES

3.1 INTRODUCTION
In the previous chapter we investigated the behaviour of infinite sequences
of real numbers and we realised that we had to look at the terms an and
consider the overall trend as n gets larger and larger. For series, however,
the important consideration is not just the individual terms themselves but
rather the behaviour as we keep on adding them together.
As we remarked in the previous chapter, the idea of using infinite processes
was first introduced by Newton in his treatise De analysi per aequationes
numero terminorum infinitas. Newton is remembered as one of the outstanding
mathematicians of all time. His work opened up numerous fields in
mathematics, physics and astronomy and he developed new ideas which were
to revolutionise mathematics. Among these ideas was the use of infinite series
as a mathematical tool. It was left to succeeding generations of mathe-
maticians to develop the subject rigorously and investigate the laws needed
to govern the manipulation of infinite series. Newton and his contemporaries
did not worry unduly about whether their use of the normal rules of algebra
in dealing with infinite series would give the correct answer. They simply got
on with their problems and on the whole their somewhat cavalier attitude
seemed to work and produced many exciting results. Of course, they
sometimes ran up against problems. For example, they were quite happy to
use relations of the type
1 2 3
-1-=1+x+x +x
-x
+ .... (1)

Now if we substitute x = 2 in relation (1) then the l.h.s. has the value -1
and the r.h.s. reads 1 + 2 + 4 + 8 + 16 + .... It would seem reasonable to write
00 on the r.h.s. of(1). How can we square this with the fact that the l.h.s. has
value -1? The answer is that we can't, but neither could the seventeenth-
century mathematicians and so they just dismissed it as absurd. If, however,
we substitute x = t in relation (1), then the l.h.s. has value 2 and the r.h.s. is
t
1 + t + + k+ .... Is the equality in relation (1) true for x = t? It certainly

62
 INFINITE SERIES

doesn't look absurd, but in view of what happened when we substituted x = 2,

we remain a little cautious. We are in good company: the Swiss mathematician
Leonhard Euler (1707-83) would also have advised caution. He recognised
that it is unwise to use infinite series unless they are convergent, and he set
the stage for a rigorous development of the subject. Rather surprisingly, he
did not always exhibit caution in his own work involving infinite series. He
did, however, produce many staggering results and used infinite series to
effect in solving difficult and long-standing problems. His recognition of the
underlying problem spurred others to develop the necessary theory which is
part of present day analysis.
A prime mover in the development of this new theory was the German
mathematician Carl Friedrich Gauss (1777-1855). Stories abound about his
childhood and he was certainly an amazingly precocious infant. Perhaps the
best-known story concerns his first arithmetic lesson at the age of nine. By
all accounts his teacher, Buttner, was an unpleasant and sullen character. As
the story goes, Buttner set his class a problem consisting of adding 100
numbers together. Whether these numbers were the first 100 positive integers,
as some commentators think or the numbers 81297 + 81495 + 81693
+ 81891 + ... + 100 899 as others claim is immaterial. The important point
was that the 100 numbers formed an arithmetic progression which the teacher
could readily sum by using a formula whereas the pupils in their first
arithmetic lesson would normally only obtain the required answer after long
and laborious additions. The school rule was that each pupil put his slate
on the teacher's desk when he had completed the task. Nine-year old Gauss
put his slate in front of the teacher almost immediately the teacher stopped
speaking. The other pupils toiled on for about an hour. When the slates were
finally checked only one pupil had the correct answer-the nine-y~ar-old
Gauss. Without prior knowledge of the formula for arithmetic progressions
he had managed to work out how to do the computation, while the teacher
spoke-a phenomenal achievement for a nine-year-old. Throughout his long
life he remained capable of doing complicated arithmetic operations in his
head. Fortunately, the fact that Gauss had outstanding ability was acknow-
ledged and he was given the opportunity to develop it.
One of his early interests was the binomial expansion of (1 + x)ft in the
case in which n is not a positive integer. Some so-called proofs had been
devised, but young Gauss realised that they were unsatisfactory and set about
making good the omission. Relation (1) which occurred earlier in this chapter
is just a particular example of the binomial expansion for the case n = -1.
We have seen that using the value x = 2 leads to an absurdity of the form:
'minus one equals infinity'. Gauss was the first to see that a 'proof' which
can lead to such a ridiculous statement is no proof at all. He realised that
the statement is correct for certain values, but not for all values of x and n.
In his opinion a formula which gave consistent results for some values had
no place in mathematics until the precise conditions had been determined
for which consistency is guaranteed. This criterion prescribes the form in

63
GUIDE TO ANALYSIS

which analytical theorems are presented right up to the present day. Typical
theorems in analysis are of the form: 'if certain specified conditions are
satisfied, then a given conclusion follows'.
We have Gauss to thank for this, but the man who more than any other
put this part of analysis in a form which we recognise readily today is the
French mathematician Cauchy (1789-1857). Not suprisingly, his name is
now associated with many facets of analysis. For example, we have Cauchy's
criterion for convergence. He was a very gifted teacher and a prolific writer.
In 1821 he wrote for publication the course of lectures on analysis which he
had been giving at the Poly technique. Much of what he wrote on convergence
of infinite series in this course oflectures would not surprise a modem reader,
as it is so like the accounts in present-day textbooks. In these lectures he
introduced the idea of taking the sum Sn of the first n terms of the series and
considering what happens to Sn as n-+ 00, which is exactly the line we will
follow in the next section.
Cauchy was a prolific writer-producing in his lifetime over 800 papers,
memoires, treatises, etc. (one of them being over 300 pages long), and his
terrific activity had a rather amusing result. In 1835 the Academy of Sciences
began publishing its weekly bulletin (the 'Comptes Rendus') and Cauchy
deluged the new publication with articles. In view ofthe rapidly rising printing
bill, the Academy passed a rule (which is still in force today) prohibiting the
publication of any paper over four pages long. Cauchy had to look for other
places to publish his longer works.
As a lecturer Cauchy was apparently very effective. According to old stories
one well-known French mathematician. (cum astronomer), Laplace (1749-
1827), is said to have listened to his lectures on infinite series with attention
and mounting apprehension. Unlike many mathematicians over the ages,
Laplace directed most of his energies towards one major project-the
investigation of the stability of the solar system. His magnum opus was the
Mecanique celeste-published in five volumes over a period of 26 years from
1799 to 1825. In the course of his investigations Laplace had used infinite
series. After listening to Cauchy's lecture, he rushed home to check whether
the series he had used were convergent, fearing that all his work on celestial
mechanics might be destroyed if his series turned out to be divergent. To his
great relief, he found that his series were indeed convergent and his work
was saved!
We too can use infinite series and use them with confidence provided we
keep the laws governing their use. The purpose of this chapter is to develop
these rules.

3.2 SERIES AND NOTATION

We first introduce the sigma notation. Let a l ,a2,a 3 , ••• be real numbers.
When we need notation for the sum of the first n terms a l + a 2 + ... + an.

64
 INFINITE SERIES

n
we will write I ak = a1 + a2 + ... + an·
k=1

If we wish to keep on adding the terms indefinitely, giving al + a2 + a 3 + ...

(where the dots signify that we keep on adding more and more terms), then
we will use
00

I an = a 1 + a2 + a3 + ....
n=1

At the moment this is just a formal sum as we do not know whether it

makes any sense to keep on adding more and more terms. In order to
investigate further, it would seem sensible to find the sum of the first n terms
and then consider what happens to this sum as n increases. Using the standard
notation we would normally write

If Sn tends to a finite limit S as n ~ 00, then the sequence (sn) converges and
it is sensible to say that the series I:'= 1 an converges and call the limit S its sum.
To facilitate printing, the limits are placed alongside the summation signs
when they appear in a line of text. A neater appearance can be achieved by
using Ir an instead of I:'= 1 an when there is no danger of confusion.

DEFINITION 3.2.1 Let aI' a2' ... be real numbers and let
n
Sn = a 1 + a2 + ... an = I ak·
k=1

The infinite series Iran is said to converge if Sn tends to a finite limit as

n ~ 00. The value of this finite limit is called the sum of the series and we
write
00

I an=s,
n=1

where S = limn .... oosn. The quantity Sft is called a partial sum of the infinite
series.

DEFINITION 3.2.2 An infinite series

which does not converge is said to diverge.

The symbols

65
GUIDE TO ANALYSIS

have been used in two different senses. In the first case they represent the
formal sum a1 + a2 + a3 + a4 + .... In the definition, however, they are also
used for the value of the sum of a convergent series and the equality

has a well-defined meaning. Normally, it will be clear from the context how

is being used and so the two different meanings do not cause any
confusion.

Examples 3.2.1

1. Consider the geometric series 1 + x + x 2 + ... Write

Sn = 1+ x + x 2 + ... + x" - 1,
i.e. s" is the sum of the first n terms. Then
XS. = x + x 2 + ... + x" - 1 + x",
and, therefore (1 - x)s" = 1 - x".
.
Thus If x * 1, S,,=--.
I-x"
I-x
If x = 1, s" = 1+ 1+ ... + 1 = n.
Now if Ix I < 1, x" -+ 0 as n -+ 00 and hence
1
S -+--
" I-x
as n -+ 00. The series
00

~ x",
n=O

therefore, converges for Ixl < 1 arid its sum is 1/(1 - x).
If x = 1, Sft = nand s" -+ 00 as n -+ 00 .
1-(-1)"
If x = -1, s. = 2 and s. does not tend to any limit as n-+ 00.
If Ixl > 1, then

Is I = Ixn-ll Ixl'-1
~'-------'---
n Ix - 11 Ixl + 1
66
 INFINITE SERIES

and ISftl ~ 00 as n~ 00. Hence Sft cannot tend to a finite limit as n~ 00 if

Ix I > 1. Hence the geometric series

diverges if Ixl ~ 1 and it converges if Ixl < 1. Moreover, for Ixl < 1
1
Lx
00
ft
=-.
n=O I-x

1 +
2. Let an = n(n + 1) (ne"Z. ). Then

1 1
a =----
n n n+t"
Hence a1 =1-!
a2=!-1
a3 =1-1
1 1
an - 1 =----
n-1 n
1 1
a =----
n n n+l
and it follows that

Thus Sft ~ 1 as n ~ 00, the series

converges and

f
n=ln(n
1
+ 1)
= 1.

3. Let

and

be convergent series. Write

67
GUIDE TO ANALYSIS

and s" + ttl -+ s + t as n -+ 00, where

00 00

s= Lan
n=1
and t= Lb
n=1
n•

00
Hence L (an + btl)
,,=1
converges and

00 00 00

L (a" + bn) = n=1

n=1
L an + n=1
L bn·
There is therefore no difficulty about adding convergent series. We shall
see later than multiplication of series is not quite so straightforward.

EXERCISES 3.2.1

1 Let an = 1/[n(n + 1)(n + 2)J (neZ+) and let Sn = a 1 + a 2 + ... + an. Express
an in partial fractions and hence, or otherwise, show that

1 1 1
Sn = 4- 2(n + 1) + 2(n + 2)"
Deduce that
00
L l/[n(n + l)(n + 2)]
n=1

converges and find its sum.

2 Find an explicit expression for the partial sums Sn of

"~1 log ( 1 + ~ ).
Deduce that

diverges.

In two of the worked examples, it was possible to write down an explicit

expression for the partial sum s" and to then check whether Sn tends to a
finite limit as n -+ 00. It rarely happens that this can be done in the general
case and so we must build up a collection of tests which can be used to
decide whether a series converges. The first test is a very simple one.

68
 INFINITE SERIES

3.3 TESTS FOR CONVERGENCE

In a sense, the first test is probably better considered as a test for divergence.

THEOREM 3.3.1 If an~O as n~ 00, then

diverges.

Proof Write
If the given series converges then Sn tends to a finite limit S as n ~ 00. Hence
Sn _ 1 -+ S as n ~ 00, and therefore

as n -+ 00. It follows that if an ~ 0, then Sn cannot tend to a finite limit as

n -+ 00 and the series must diverge.

COROLLARY If Lan converges, then an~O as n~oo.

The reader is warned to be extremely careful about the use of this corollary.
It is impossible to have convergence unless an -+0 as n -+ 00, but an ~ 0 as
n -+ 00 is not sufficient to guarantee convergence. For there are divergent
series Lan for which an -+ 0 as n ~ 00.

Example 3.3.1

Let an=~ (nEZ+). Then an-+O as n-+oo. Let

Sn = a 1 + a 2 + ... + an = 1 + t + t + t + ... +~.
We will show that Sn -+ 00 as n -+ 00 and so Lan diverges despite the fact that
an -+ 0 as n -+ 00. To achieve this first let us bracket the terms in groups in
the following way:

1+ ~ + ( ~ + ~ ) + ( ~ + ~ + ~ + ~ ) + ... + (2" -~ + 1+ ... + ;" ) + ... ,

so that the final term in each bracket is the reciprocal of a power of two.
Consider the sum of the first 2" terms,

69
GUIDE TO ANALYSIS

~ 1 + ~ + (~+ ~ ) + G+ ~ + ~ + ~) + ... + (;n + ;n + ... + ;n)

1 1 1 1
= 1 +2+2+2+'" +2
1 n+2
= 1 +(n)2=-2-'

since each of the brackets in the second line has sum t. As each of the terms
is positive we see that (s,,) is an increasing sequence. Moreover,
n+1
s2'~-2-'

and so (s,,) is not bounded above. Hence s" -+ 00 as n -+ 00 by Theorem 2.4.1.

Thus the series
00 1
L-
,,=I n

diverges.
This result may seem a little surprising at first sight. It was not immediately
obvious that we could make Sn arbitrarily large by taking a sufficiently large
number of terms, but it is indeed true. In fact the method we have used
demonstrates that we can be certain sn> 106 for all n > 21999998, which is,
of course, rather a large number of terms.
This example should be remembered; it is a divergent series which is
frequently used for reference.

Exercises 3.3.1 provide other examples of divergent infinite series La" for
which an -+ 0 as n -+ 00.

Examples 3.3.2

1. Let an
=(~)4
5n + 1 .

Then as n-+ 00

and

diverges because an ~ 0 as n -+ 00.

70
 INFINITE SERIES

1
2. Let
an = In'
1 1
Then an = Jn~n for all n. (1)

Write

where b n =~. Then

for all n by (l) and so Sn ~ 00 as n -+ 00 because tn ~ 00 as n ~ 00 i.e.

diverges.

The reader will notice that the series

was shown to diverge by using an inequality which compared its terms with
the terms of the series
00

L lin
n=l

which is known to diverge. This suggests that we may be able to test for
convergence or divergence by comparing the individual terms with those of
a series whose behaviour is already known. Indeed, such a method is
embodied in the well-known comparison test.

THEOREM 3.3.2 (Comparison Test)

00
Let L an
n=l
and

be two series of non-negative real numbers, such that

(a) a,,::::; Kb" for all n and some positive real number K, and
Lb
00
(b) n converges.
n=l

L an also converges.
00
Then
n= 1

71
GUIDE TO ANALYSIS

Proof Let

tn = b 1 + b 2 + ... + bn·
Since an ~ 0 and bn ~ 0 for all n, the sequences (sn) and (tn) are both increasing.
Moreover, tn tends to a finite limit t as n -+ 00, because

is convergent. Hence tn ~ t for all n. Now

Sn = a1 + a2 + ... + an ~ K(b 1 + b2 + ... + bn) = Ktn ~ Kt
for all n and, therefore, (s.) is bounded above. Since (sn) is also increasing,
we see that Sn tends to a finite limit as n -+ 00, and

therefore converges.

COROLLARY 1

Let

be two series of non-negative real numbers such that

(a) an ~ kCn for all n and some positive real number k,
I
00
(b) Cn diverges.
n=l

Then

also diverges.

Proof Write S. = a 1 + a 2 + ... + a",

From condition (a) it follows that

Sn ~ krn (1)
for all n. Now (r.) is an increasing sequence which does not converge by (b).
Hence rn -+ 00 as n --+ OCJ and, therefore, krn -+ 00 as n -+ 00. It now follows
from (1) that s" --+ 00 as n --+ 00 and so I:'= 1 an diverges. This completes the
proof.

72
 INFINITE SERIES

It is not essential that the inequality (1) should be satisfied for the first few
terms. In fact it is sufficient if this inequality is satisfied for all n ~ Nowhere
No is a given positive integer. Only a slight modification of the proof is
needed to take into account this change. Thus we have the following
corollaries.

COROLLARY 2

L an
00
Let and
n= 1

be two series of non-negative real numbers such that

(a) an ~ Kb n for all n ~ No and some positive real number K, where No
is a given positive integer; and
L bn converges.
00
(b)
n=l
00
Then Lan
n=l

also converges.

COROLLARY 3

L an L en
00 00
Let and
n=l n=l
be two series of non-negative real numbers such that
(a) an ~ KC n for all n ~ No and some positive real number K, where No
is a given positive integer; and
LC
00

(b) n diverges.
n=l
00
Then Lan
n=l
also diverges.

Inevitably, mistakes will be made unless inequalities are handled with care.
In some instances, however, the use of inequalities can be avoided by the use
of an alternative form of the comparison test. This alternative is called the
limit form of the comparison test.

THEOREM 3.3.2(a) (Limit Form of Comparison Test)

L an
00
Let and
n=l

73
GUIDE TO ANALYSIS

be series of positive real numbers such that (aJb n ) tends to a finite non-zero
limit as n -+ 00. Then either
00 00

I a" and I b" (1)

,,=1 ,,=1
both converge or
00 00

I a" and I b" (2)

,,=1 ,,=1
both diverge.

Proof Let the finite non-zero limit be I. Since I#-O and (aJb,,) -+ I as n -+ 00,
it follows that I> 0 and there is some integer N such that

bn -I I <211
Ian for all n> N,
a
I.e. 2 < ---.!!.
.11 bn <;!I 2 for all n > N,

which gives tlb" < a" < !Ib n for all n > N. (1)

Now ifI;,"'= 1bn converges, then I;,"'= 1a" must also converge by the comparison
test since 0 < an < !Ib n for all n> N. Moreover, if I;,"'= 1an converges then
I;,"'= 1bn must also converge by the comparison test since 0 < bn < 2aJI for
all n> N. Thus if one of the series converges, then the other also converges,
i.e. either
00 00

I an and I bn (1)
n=1 n=1
both converge or
00 00

I an and I bn (2)
n=1 ,,=1
both diverge.

Examples 3.3.3

1. Let an = l/n2, and bn = 1/[n(n + 1)]. Then for all n ~ 1, an> 0, ~,,> 0

and an _n(n+l)_(
--
bn n2
- 1+- -+1
n
1) as n-+ 00.

Now the second example of section 3.2 shows that I;,"'= 1 1/[ n( n + 1)]
converges, and therefore, by the limit from the comparison test, I;,"'= 11/n 2
converges.

74
 INFINITE SERIES

2. Let an = 11na where (X~2 and bn= 11n 2 • For all nEZ+, an>O, bn>O and
0< an = 11n a ~ 11n 2 = bn. Since L:'=
1 bn converges it follows by the com-

parison test that L:'=

1 an converges.

Hence converges for all (X ~ 2.

3. Let an = lin, bn = 11na where ~ 1. For all nE/r, an> 0, bn > 0 and
11
(X

O<a n =-~-=b
n na n

since (X ~ 1. By Corollary 1 of the comparison test, L~= 1 bn diverges.

Thus diverges for (X ~ 1.

4. Let an = sin( lin), bn = lin. For all nEZ +,

0< lin ~ 1 < nl2
and, therefore, sin(lln) > 0 for nEZ+.
Thus an > 0, bn > 0 and

an = sin(lln) .... 1 as n .... 00.

bn lin
Now L~= 1 bn diverges, and therefore L~= 1 an also diverges by the limit
form of the comparison test.

Examples 2 and 3 above provide useful standard results which can be used
in conjunction with the comparison test. At the moment they leave a small
gap. With the tests at present at our disposal we cannot determine what
happens to L~= l11n a for 1 < (X < 2. Later we will be able to show, with the
help of the integral test, that the series converges for these values of (X
also.

EXERCISES 3.3.1

1 Show that the following series converge:

(b) n~l (~: 11)\

3n + 4n
00

(c) nf:13n + 5n ; (d) f Jn2+1-n p-=t.

n=1

75
GUIDE TO ANALYSIS

2 Show that the following series diverge:

(b) f
n=l
(P+i-p-=t);
CXl 3n+ sn

(c) nf:l~;

3 Determine whether the following series converge ot diverge:

n+ 1
(a) I
CXl

n=ln
-2-1;
+
(b) f ~ sin(~);
n=ln n

(c) f sin2(~); (d) f In+l-In=l;

n=l n n=l .In
00 31/n
(f) nf:lnl/3"
Reasons should be given for answers. (In solving these problems, the reader
must be prepared to use all the tests given so far in section 3.3. Not all the
questions rely on the comparison test!)

Using the comparison test, we examine a series for convergence by comparing

it with another series whose behaviour is already known. It is, however,
possible to test for convergence by studying only the behaviour of the ratio
of successive terms. This test is known either as the ratio test or as
d' Alembert' s test in honour of the celebrated French mathematician Jean Ie
Rond d' Alembert (1717 -83). D' Alembert, an illegitimate son of the Chevalier
Destouches, took his name from the Chapel of St Jean Ie Rond on whose
steps he was abandoned, as an infant, by his mother. His name is known to
many generations of students in connection with the following result.

THEOREM 3.3.3 (D' Alembert's Test-Ratio Test)

Let I:'= 1 an be a series of positive real numbers such that an + lla n ~ I as
n-+ 00.

(a) If I < 1, then I:'= 1 an converges.

(b) If I > 1, then I:'= 1 an diverges.
Proof
(a) Case 1<1 Choose a number k such that 1< k < 1. Since an+ dan ~ I
as n -+ 00, there is some N such that

76
 INFINITE SERIES

for all n > N. This gives

an + l < k
an
for all n > N.
i.e. for all n>N.
Thus for all n > N +1
N+l )
an < kan- l < k 2 an- 2 <···< kn-N-l aN+l = kn(akN+l ·

Now aN+dk N+1 is just a constant. Moreover, L:;;,,= 1 kn converges since

o< k < 1. By the comparison test, L:;;,,= 1 an converges also.
(b) Case I> 1 Since an+dan-+I> 1 as n-+oo, there is some M such
that

for all n> M. Thus for n> M +1

an > an - 1 > an - 2 > ... > aM + 1 •
Hence an + 0 as n -+ 00, and therefore 2:;;,,= 1 an diverges.
Quite clearly the argument in the preceding paragraph can be used if
an+lla n -+ 00 as n -+ 00, which gives us the following corollary.

COROLLARY

Let

be a series of positive terms such that an+lla n-+ 00 as n -+ 00.

Then

diverges.

In the statement of the theorem one case is obviously omitted, viz. the case
1= 1. There is good reason for this omission. For if 1= 1, either convergence
or divergence is possible. For instance, if an = lin, then L:;;,,= 1 an diverges, but
an + dan = nln + 1 -+ 1 as n -+ 00. However, if an = 1/n 2 then L:;;,,= I an converges,
but again an + 1 Ian -+ 1 as n -+ 00. In both these examples 0 < an + 1 < an and
yet one of the series converges and the other diverges. It is therefore
inadmissible to omit the limit from the ratio test-a mistake frequently made
by undergraduates. The condition an+lla n < 1 is not sufficient to guarantee
convergence, as we can readily see by looking at the divergent series L:;;,,= 1 lin.

77
GUIDE TO ANALYSIS

Examples 3.3.4

1. Let an = [(2n)!J2/(4n)! for all ne:r. Then an> 0 and

an +l [(2n + 2)!J2(4n)!
an (4n + 4)![(2n)!J2
(2n + 2)(2n + 1)(2n + 2)(2n + 1)
(4n + 4)(4n + 3)(4n + 2)(4n + 1)
(2n + 2)(2n + 1)
-'---------'---'---------'-- -+-
1
2(4n + 3)2(4n + 1) 16
as n-+oo. Since the limit is less than 1, I:'=lan converges.
2. Let an = n2 e -nCo + 1) for all ne"Z. +. Then an> 0 and
(n + 1)2 en(n+ 1) 1)2
(n+l)(n+2) -- 1 + - e - 2(n + 1) -+ 0
(
--'------;;-''----_-;----;--~~
n2
e n
as n -+ 00. Hence I:'= 1 an converges.

EXERCISES 3.3.2

Give reasons to justify your answers to the following questions. (In question
3, the reader must be prepared to use all the tests in the preceding part of
section 3.3.)
1 Prove that the following series converge:
00 (n!)2 00 (2n)!
(a) .f:l(2n)!; (b) .f:17"(n!)2;

~ coshn
(d) 1...
n= 1 3" .

2 Prove that the following series diverge:

~ coshn
(c) 1... - - ;
n= 1 n

3 Determine whether the following series converge:

78
 INFINITE SERIES

00 n lO
(f) JIJ(2n)T;

(g) f
n= I
• 2
sm2 n.
n

The comparison test and d'Alembert's test provide us with very powerful
tools for testing for convergence. There is one further test for convergence
of positive terms which we wish to include. This is the integral test. Strictly
speaking it ought to be delayed until after the sections on functions, but it
seems a pity to separate it from the other tests and so we include it at this
point with a few preliminary definitions and the introduction of some
notation.
Let a be any real number; then the set of all real numbers x such that
x ~ a is denoted by [a, (0) and it is called an interval. Thus the interval
[2, (0) is the set of all real numbers x such that x ~ 2.

DEFINITION 3.3.1 A function f is said to be decreasing on [a, (0) if

f(x l ) ~ f(x 2 ) for all Xl' x 2 such that a ~ Xl ~ X 2 •

The reader should try sketching the graphs of some functions with the above
property. It will then be transparently obvious why the term decreasing is
used to describe the function.
Now let us suppose thatf(x) is defined for allxE[1, (0) and thatfis positive,
decreasing and continuous on [1, (0). (Note: For an explanation of the term
continuous the reader is asked to consult Chapter 4.) Write

f
n

In = f(x) dx (n~1), (1)

I

L f(k).
n
Sn = f(1) + f(2) + ... + f(n) = (2)
k=l

Then it can be shown that (sn - In) tends to a finite limit as n -+ 00.
It may prove illuminating and instructive to begin by drawing some
diagrams and using geometric reasoning. Before drawing any such diagrams,
we must, of course, remember that f is assumed to have certain properties,
viz. the function f is continuous, decreasing and positive (see Fig. 3.1).
We draw vertical lines at X = 1, X = 2, X = 3, X = 4, etc., and also horizontal
lines to give the rectangles below the curve which are shown in Fig. 3.1. The
rectangle on the extreme left-hand side has height f(2) and width 1. Its area
is, therefore, f(2). The next rectangle has height f(3) and width 1. Its area is
therefore f(3). The last rectangle, which appears at the far right-hand side of
the paper, has height f(n) and width 1. Its area is thus f(n). The sum of the

79
GUIDE TO ANALYSIS

2 3 n-2 n-1 n x
Fig. 3.1

areas of all these rectangles is f(2) + f(3) + f(4) + ... + f(n). From (2), it
follows that the sum of areas of all the rectangles below the curve between
x = 1 and x = n is Sn - f(I). Now the area under the curve between x = 1
and x = n is Hf(x)dx, i.e. the area under the curve is In (see equation (1».
It follows that the difference In - [sn - f(l)] measures the sum of the areas
of the (n - 1) shaded regions between the curve y = f(x) and the rectangles.
As n increases, the number of shaded regions increases and the sum of their
areas increases, i.e. In - Sn + f(l) increases as n increases. Hence In - Sn
increases as n increases, since f( 1) is a constant.
Now we could move each of these shaded areas sideways to the left (see
Fig. 3.2) and they would fit neatly into the rectangle ABCD without any
overlap, since f(x) decreases as x increases. The total area of these regions
together cannot exceed the area of the rectangle ABCD. Now this rectangle
has height f(l) and width 1 and its area is therefore f(I).
Hence In - Sn + f(l) ~f(l)
i.e. In - Sn ~O.

and the sequence (In - sn) is bounded above. Since the sequence is also
increasing, it must converge, i.e. In - Sn tends to a finite limit as n ~ 00.
If we wish to give a proof which uses only symbols and avoids geometric
reasoning, then we simply need to show that (In - sn) is increasing and
bounded above. Let us first write
80
 INFINITE SERIES

A D
o 2 3 4 n-2 n-l n x

Fig. 3.2

(3)
(a) (In - sn) is increasing For all n ~ 1,
Cn +1 -Cn =(In+1- Sn+1)-(In- sn)
=(In+1-[n)-(Sn+1- Sn)
n+ 1 n

= ( f f(x)dx - ff(X)dX ) - f(n + 1)

1 1
n+1

= f f(x)dx-f(n+1)~O,
n

sincef(x) ~f(n + 1)forn:::;; x:::;; n + 1. Thuscn +1 - Cn ~ O,i.e.cn + 1 ~ Cn

and the sequence is increasing.
(b) (In - sn) is bounded above Since f is decreasing we know that for
i:::;; X:::;; i + 1 (i = 1,2,3,4, ... )f(x):::;; f(i), and therefore
i+1 i+1 i+1

f f(x) dx:::;; f f(i) dx = f(i) f dx = f(i).

i i i

81
GUIDE TO ANALYSIS

Hence ff(X)dX ~ f(1),

1
3

f f(X) dx ~ f(2),
2

f f(x)dx~f(n-1),
n-l

and it follows, by addition, that

2 3 n

ff(X)dX + ff(X)dX + ... + f f(x)dx ~ f(1) + f(2) + ... + f(n -1),

1 2 n-l
n

i.e. ff(X)dX ~ f(l) + f(2) + ... + f(n -1).

1

Since f(x) ~ 0 for all x ~ 1, we see that

n

ff(X)dX ~ f(1) + f(2) + ... + f(n -1)

I

~f(1)+ f(2) + ... + f(n-1)+ f(n)

=Sn;

and therefore In - Sn ~ 0 (n ~ 1).

Since (In - sn) is increasing and bounded above, In - Sn tends to a
finite limit as n --+ 00.

In order to keep a record, for future reference, we incorporate these results

in a theorem.

THEOREM 3.3.4 Let the functionf be continuous, decreasing and positive

on [1,00). Write
Sn = f(l) + f(2) + ... + f(n) (n ~ 1),

82
 INFINITE SERIES

In = ff(X)dX (n ~ 1).
1

Then Sn - In tends to a finite limit as n-+ 00.

The integral test is now a simple deduction from this theorem.

THEOREM 3.3.5 (Integral Test) Let the function f be continuous,

decreasing and positive on [1, 00). Then
(a) I~ d(n) converges, if Hf(x) dx tends to a finite limit as n -+ 00,
(b) I:=
d(n) diverges, if Hf(x) dx does not tend to a finite limit as
n-+ 00.

Proof We use the standard notation

Sn = f(1) + f(2) + ... + f(n) (1)
n
(2)
In = ff(X)dx
1

and we write Cn = Sn -I". (3)

Clearly Cn tends to a finite limit as n-+ 00, by Theorem 3.3.4.
Since f(x) > 0 for x ~ 1, we see that s" and I" both increase as n increases.
Thus either
(a) (In) is bounded above and I" tends to a finite limit as n -+ 00, or
(b) (In) is not bounded above and I" -+ 00 as n-+ 00.

(a) If I" tends to a finite limit as n-+ 00, then Sn = I" + c" must also tend
to a finite limit as n -+ 00 (because c" tends to a finite limit as n -+ 00).
Hence I~ d(n) converges, which proves (a).
(b) If 1,,-+00, then s,,=c,,+In-+oo as n-+oo and therefore L:=d(n)
diverges, which proves (b).
It is, of course, not vital to start at x = 1; we could start at x = Nowhere
No is any given positive integer, provided all the conditions are satisfied on
[No, 00). This gives

COROLLARY Let the function f be continuous, decreasing and positive

on [No, 00) where No is some given positive integer. Then
(a) L:=NJ(n) converges if J~J(x)dx tends to a finite limit as n -+ 00,
83
GUIDE TO ANALYSIS

(b) L::'=N'/(n) diverges if IN./(x)dx does not tend to a finite limit as

n--+ 00.

So far all our integrals have been of the form JNj(x)dx where No, n are
positive integers. With the given conditions on f it is possible to prove that
J~./(x)dx tends to a finite limit as X --+ 00 if and only if JNj(x)dx tends to
a finite limit as n -+ 00. We say that IN'j(x) dx converges if and only if
J~./(x)dx tends to a finite limit as X --+00. The integral test is, therefore,
sometimes recorded in the following form.

THEOREM 3.3.5(a) (Integral Test) Let the function f be continuous,

decreasing and positive on [No, 00), where No is some given positive integer.
Then 'L,::'=Nj(n) and IN'j(x)dx either both converge or both diverge.

On the whole, the introduction of IN'j(x)dx and the idea of convergence of

integrals seems over-elaborate at this stage, since it is quite sufficient to
consider only limn-+ 00 JNj(X) dx. We will therefore use Theorem 3.3.5 and its
corollary in examples. The reader must, however, be aware that many authors
will use the second form (Theorem 3.3.5(a» and must be prepared to meet
it in books on analysis.
Our first example, using the integral test; settles a problem which arose
earlier in the chapter, viz. does the series 'L,::'= 11/n" converge if 1 < 0( < 2?
Fortunately the answer is yes. This means that we now know that

L -;
00

n
1
converges if 0( > 1,
n=l

and f ~ diverges if 0( ~ 1.
n=ln
Let us prove the result which closes the gap left earlier in the chapter.

Examples 3.3.5

1. Let f(x) = ~
x·
(x ~ 1), where 0( > O. Then f is continuous, decreasing and
positive on [1,00). Using the standard notation we write

(n = 1,2,3,4, ... ).

=fn~dX=([ -(0(~1)X}-lI (0( ~ 1),

[In
Then I
n
1 X
(T.
logx 1 (0( = 1).

84
 INFINITE SERIES

Now if IX > 1, I =_1_(1 __1_)-+_1_

n IX - 1 na - 1 IX - 1

as n -+ 00, and so L:'= 11/na converges by the integral test for IX> l.
However, if IX = 1, In = Ii 1/xadx = logn and In-+ 00 as n-+ 00. Thus
L:'= 1 1/ na diverges if IX = l.
If 0 < IX < 1, then In -+ 00 as n -+ 00, and again L:'= 1 1/na diverges.

The next example illustrates the fact that it is sometimes necessary to start
at a point other than 1.
2. We use the integral test to decide whether L:'=21/[nlogn] converges. To
do this we use
1
f(x) = xlogx (x ~ 2).

Then f is continuous, decreasing and positive on [2, (0). Moreover,

f-I-1_
n n

In = ff(X)dX = dx = [log(logx)]i
X ogx
2 2

= log(logn) -log(log2).
Hence In-+oo as n-+oo, and so L:'=21/[nlogn] diverges.

If Theorem 3.3.4 is applied with the f(x) =~, we derive some rather
interesting results.
3. For if f(x) = ~ for x ~ 1, then f is decreasing, continuous and positive
on [1, (0). With the standard notation, we have
n n

In = ff(X)dX = f~dX = [logx]~ = logn (1)

1 1
n 1
and Sn= L -k'
k=1
= 1 +1+-!+*+ ... +~.
Write (2)
Then, from Theorem 3.3.4, 1n tends to a finite limit 1 as n -+ 00. This number
1 is called Euler's constant. Its value has been computed to over 200
decimal places. It is 0.57721566490153 correct to fourteen decimal
places.
It follows from (1) and (2) that

where 1n-+1 as n-+oo.

85
GUIDE TO ANALYSIS

i.e. 1 +t+t+ ... +~=logn+Yn· (3)

This relation can be used to sum certain series.
Note: Formula (3) shows that the sum 1 + t + t + t + ... + -:. behaves like
log n + Y for large n. We already know that Sn -> 00 as n -> 00, but we now
have a better idea of how rapidly Sn grows as n increases.
4. Consider the series

Let tn be the sum of the first n terms. Because the signs alternate plus,
minus, plus, minus, etc., it is simpler to first consider an even number of
terms-say the first 2n terms. The sum of these first 2n terms is
1 1 1 1 1
t2n = 1 - 2' + 3" - 4 + ... + 2n - 1 - 2n

= log 2n + Y2n - [log n + YnJ

from equation (3) above. Hence
t2n = log 2n -log n + Y2n - Yn
= log2 + Y2n - Yn
and t2n -> log 2 + Y - Y = log 2
as n-> 00.
lf we add an odd number of terms (say the first 2n + 1 terms) we have
1 1 1 1 1 1
tZn+ 1 = 1- 2' + 3" - 4 + ... + 2n - 1 - 2n + 2n + 1

1
= tZn + 2n -+- l'
and therefore t2n+l->log2+0=log2

86
 INFINITE SERIES

as n --+ 00. Since t 2 " --+ log 2 and t 2" + 1 --+ log 2 as n --+ 00, we see that t" --+ log 2
as n --+ 00. The given series, therefore, converges and its sum is log 2,

i.e.

5. Suppose we now take the above series and rearrange its terms as

i.e. we take four positive terms, then one negative, then four positive terms
etc. Will this rearranged series still have sum log2? Well, let us see. Let
r" be the sum of the first n terms. Obviously it will be easier if we first see
what happens if we take the terms in bundles of five (four positive and
one negative term in each bundle). Let us take n bundles of 5, i.e. 5n terms
in all. The sum r 5" of these first 5n terms is given by
111111111
r 5 ,,= 1 +3+5+7-2+"9+11+ 13 + 15 -4+ ...
1 1 1 1 1
+8n-- -7 +8n-- -5 +8n-- -3 +8n-- -1 -2n
-

1 1 1 1) (1 1 1 1 1)
= ( 1 +2+3+4+ ... + 8n - 2+4+6+8+ ... + 8n
1 1 1 1 1)
- ( 2+4+6+8+ ... + 2n .
(Note: Many students frequently find it difficult at first to write down
suitable expressions for the terms in the nth bracket. The following
reasoning may help. The initial terms in each of the first three brackets
are t, ! and l7 respectively. The denominators increase by 8, and therefore
the positive terms have denominators of the form 8n + something. Now
the adjustment is easy. Similarly, the denominators of the negative term
go up by 2, and they therefore contain 2n + something.)

The first two brackets taken together (cancelling out terms as appropriate)
give the positive terms and the third bracket contains the negative terms.
Hence, from (3),

1 1 1 1) 1( 1 1 1)
r 5" = ( 1 + 2 + 3 + 4+ ... + 8n - 2 1 + 2+ 3 + ... + 4n

-2
1( +2+3+4+
1 1 1 ... +;;1)
1

= log 8n + ')Is,,-Hlog4n + ')14,,] -t[logn + ')I,,]

87
GUIDE TO ANALYSIS

= log Sn - ! log 4n -! log n + 18n - 11,4.n - 11"

.1 (Sn.sn) 1 1
=210g 4n'n + 18n-2"14n-2"1n

=!log 16 + 18n-114n-11n
-+!log 16+ 1 -11 -11 =tlog 16 = log 4.
as n-+oo. Thus rsn-+log4 as n-+oo. Now, we know that the positive
integers are not all divisible by 5. Thus if we take an arbitrary (finite)
number of terms and go along the series collecting these terms in bundles
of five, then there may be some terms left on their own at the end. This
remainder could comprise 0,1,2,3 or 4 extra terms. We know rSn -+ log 4
as n -+ 00. This takes care of the case in which there are no remaining
terms. We must, therefore, also consider what happens to rSn+ 1, r Sn + 2 ,
r Sn + 3 , r Sn +4' Now as n-+oo,

1
r Sn + 1 = rSn + Sn + 1 -+log4 + 0 = log 4,

1 1
r SII + 2 = rSn + Sn + 1 + 8n + 3 -+ log 4,

1 1 1
rsn+3=rsn+Sn+l +Sn+3 +Sn+5-+ log4 ,

1 1 1 1
r Sn +4 = rSn + Sn + 1 + Sn + 3 + Sn + 5 + 8n + 7 -+log4.

Gathering all the results together, we see that r n-+ log 4 as n -+ 00. Thus
the given series converges and its sum is log 4.
Now the series in the above example is only a rearrangement of the
series 1 -! +! - t +! - i + ... , which has sum log 2. So we have, clearly,
altered the sum of the series by rearranging the order of its terms. Does
this always happen? For the answer to this question, the reader will have
to be patient and wait a little while!

The reader may also wonder how many different sums can be obtained
by rearranging such a series. The answer is that the series
l-!+!-t+!-i+ ...
can be rearranged in such a way as to have any required sum, i.e. if the
reader chooses a favourite real number x, then there is a rearrangement
whose sum is x. Except for particular values of x it is not possible to write
down a nice, neat formula specifying which term is in the nth position of

88
 INFINITE SERIES

any rearrangement with sum x. This does not matter. The important thing
is that such rearrangements exist. (Note: The reader curious to know a little
more about this should consult the appendix at the end of this chapter,
which contains the outline of one possible method for constructing such
rearrangements. )
In previous sections we have seen that the series
1-t+!-i+!-i+~--k+ ...
converges-its sum being log 2. This is a series in which the signs alternate,
plus, minus, plus, minus, "etc. It is of the form :2::'= d - 1)n +1 an> where an = !.
(n = 1,2,3, ... ). In this case
(a) an> 0 for all n,
(b) an + 1 ::;;; an, i.e. the sequence (an) is decreasing,
(c) an--+O as n--+oo.
Are these properties of an sufficient to guarantee that every such series of the
form L:'= 1 ( _1)n + 1 an converges? The answer is yes. The result, which is
stated below, is normally called either the alternating series test or the
alternating signs test. Strictly speaking it should, probably, be called the test
for series whose terms have alternating signs, but this would be rather a
mouthful!

THEOREM 3.3.6 (Alternating Series Test) Let (a n ):'= 1 be a decreasing

sequence of positive real numbers such that an --+ 0 as n --+ 00. Then the series
L:'= 1 ( _1)n + 1 an converges.
Proof Since all the previous tests (except Theorem 3.3.1) are for series with
positive terms, they are not applicable. We must therefore go back to the
definition and consider the behaviour of the partial sums Sn> where
Sn = a 1 - a2 + a 3 - a4 + ... + (_1)n + 1 an.
Initially, it is probably easiest to consider the sum of an even number of
terms. The sum of the first 2n terms is given by

Hence S2n+2 - S2n = a2n +1 - a2n+2 ~O,

since (an) is decreasing and, therefore, a2n+ 1 ~ a2n+ 2. Thus
S2n+2~S2n

and the sequence (S2n) is increasing. Moreover,

89
GUIDE TO ANALYSIS

because a 2 ~ a 3 ~ a4 ~ •.• ~ a 2n - 2 ~ a 2n - 1 and a 2n > O. Thus S2n:;:; a 1 for all

n and the sequence (S2n) is bounded above. Since it is also increasing, it must
converge to some number s, i.e. S2n --+ S as n --+ 00. Of course, we must also
investigate what happens if we take an odd number of terms. The sum of
the first 2n + 1 terms is S2n+l, where

As n--+oo, a 2n +1 -+O and S2n-+S. It follows that S2n+l--+S as n--+oo. Since

S2n-+S and S2n+1--+S as n-+oo, we see that Sn--+S as n--+oo and the series,
therefore, converges (and its sum is s).

If, however, an + 0 as n --+ 00, then ( _1)n + 1 an + 0 as n --+ 00 and the series
L:'= 1 ( _1)n+ la n diverges by Theorem 3.3.1.

Examples 3.3.6

1. Let an = sin(l/n) (n = 1,2,3, ... ). For n ~ 1,

1 1 1t
0<--<-:>::: 1 <-
n+1 n"" 2'

and, therefore, an = sin(~) >0,

an+1 = sin(n ~ 1) < sin(~) = am

i.e. (an) is a decreasing sequence of positive real numbers. Furthermore,
an = sin( l/n) -+ 0 as n --+ 00. By the alternating series test,
L (_1)n+ 1 sin(l/n)
00

n=l
converges.
But what happens to this series if we dispense with the factor ( _l)n +1?
Using the limit form of the comparison test (with an = sin(l/n), bn = l/n),
we can show that L:'= 1 sin(l/n) diverges. The factors ( _1)n+ 1 therefore
made all the difference in this case.
2. Let an=cos(l/n). As n-+oo, cos(l/n)--+1. Hence (-l)"+lcos(l/n)+O as
n--+ 00 and so L:'= 1 ( -1)"+ 1 cos(l/n) diverges by Theorem 3.3.1.
3. Let an = sin 2(1/n). As in Example 1, it is easy to show that (a n):'= 1 is a
decreasing sequence of positive real numbers, such that an--+O as n--+ 00.
By the alternating series test L:'= 1 ( _1)n+ 1 sin 2 (1/n) converges.
Again, let us see what happens if we drop the factors ( - l)n +1. Using
the limit form of the comparison test (with an = sin 2(1/n), bn = 1/n 2), we

90
 INFINITE SERIES

can show that L:'=l sin2(1/n) also converges. (This fact will later give rise
to an alternative way of dealing with the question of the convergence of
L:'=d -1)"+lsin2(1/n).) In this case, therefore, we still have convergence
even when we omit the factors ( - 1)" +1. This contrasts with the series in
Example 1.

In stating the result, we used ( -1)" +1 in preference to ( -1 )", as the former

gives the first term of the series as a1 (rather than -a 1 ). However, it really
makes no difference to the conclusion if ( - 1)" is used in place of ( -1)"+ 1.
For if we write

8" = al - a2 + ... + (-1)"+ l a" = L" (_1)k+ l ak ,

k=l
tIl = -al + a2 + ... + (-I)"a" = L" (-I) kako
k=l
then tIl = - 8" and tIl tends to a finite limit as n -+ 00 if and only if 8" tends
to a finite limit as n-+oo. Thus if (a,,) satisfies all the conditions of
Theorem 3.3.6, then L:'= 1( - 1)"a" is convergent.

EXERCISES 3.3.3
1 Prove the following series converge:

(a) f (_1)"_1_; f (-1)" + 1_1;

In
(b)
,,=2 logn ,,=1

(c) f
/1=1
(-1)"+1(Jn+l- In).
2 Prove the following series diverges:

f (-1)"+ 1COSh(!).
,,=1 n
3 Decide whether the following series converge or diverge. Give reasons to
justify your answers. (The reader must be prepared to use all the tests in
section 3.3.)

(a) "~1 sin«n2 : l)n} (b) f (#+3 - P+i);

/1=1
1
(c) L (_1)"+lCOS 2_;1
00
(d) L (logn)3;
00

,,=2n
"=1 n

(e)
00 n 72/1 + n 23"
(f)
f 3"(n!)3.
"~1 n9 + n 23" ; /1= 1 (3n)!

91
GUIDE TO ANALYSIS

3.4 ABSOLUTE CONVERGENCE, MULTIPLICATION

OF SERIES, REARRANGEMENT OF SERIES

In the previous section we saw that the series L:'= 1(_l)n+ 1 sin 1 (1/n) is
convergent and the related series L:'= 1sinl(1/n) is also convergent. That is,
L:'= 11( -1 )n+ 1 sin 1 (1/n)1 is convergent. Such a series is said to be absolutely
convergent.

DEFINITION 3.4.1 Let L:'=lan be a series of real numbers. If L:'=llanl

is convergent then L:'= 1an is said to be absolutely convergent.

The phrase 'absolute convergence' immediately suggests convergence plus

something more. We would, therefore, expect that an absolutely convergent
series is certainly convergent. Fortunately, this is indeed the case.

THEOREM 3.4.1 Every absolutely convergent series of real numbers is

also convergent.

Proof Let L:'= 1an be an absolutely convergent series of real numbers, i.e.
it is a series such that L:'= 11 an I converges. Now some of the terms of L:'= 1an
may be positive and some may be negative. We therefore separate out the
positive and negative terms in the following way.
if an ~ 0, if an ~ 0,
Let bn = {an
0 if an < 0, if an < O.
Then bn ~ 0, en ~ 0 for all n, and
(1)
Write (2)
tn = a 1 + az + ... + an,
rn=b 1 +b1 + ... +bn> (3)
(4)
Now we are given that L:'=llanl converges and so Sn tends to a finite limit
S as n ..... 00. Moreover, lanl ~ 0 for all n (by definition of the modulus) and
so the sequence (sn) is increasing. Thus

for all n. Now, from the definition of bn , we see that

rn=b 1 +b 2 + ... +bn::::;;latl+la11+ ... +Ianl=sn::::;;s
92
 INFINITE SERIES

for all n,
i.e.
for all n. But bn ~ 0 for all n and therefore the sequence (rn) is increasing.
Since it is bounded above by s, (rn) must converge to some limit r as n--+ 00.
Similarly, (un) converges to some limit u as n --+ 00. Now from (1), (2), (3)
and (4)

as n --+ 00. Since tn tends to a finite limit as n --+ 00, I:'= 1an converges as
required.

We have, therefore, proved that an absolutely convergent series of real

numbers is also convergent. The converse, however, is not true. There are
series which are convergent, but not absolutely convergent. For example, the
series I:'= 1( _1)n+ ll/n is convergent. This is just the familiar series
l-t+i--i+!-i+ ....
However, it is not absolutely convergent, since L:'= 1 l/n diverges. Such a
series is called a conditionally convergent series.

DEFINITION 3.4.2 A series of real numbers which is convergent, but not

absolutely convergent, is called a conditionally convergent series.

Example 3.4.1

In the previous section we showed that L:'= 1( _1)n+ 1 sin2(1/n) is convergent

by using the alternating series test. It would, however, probably be easier to
prove absolute convergence, which of course implies convergence. We notice
that
I( _1)n+1 sin2(I/n)1 = sin 2(I/n).
Now write an = sin2(1/n), bn = l/n 2 (n = 1,2,3, ... ).
Then an > 0, bn > 0 and

an = (sin(l/n»)(sin(l/n») --+ 11 = 1 --+

bn (l/n) (l/n) . as n 00.

Now I:'= ll/n2 converges, and therefore, by the limit form of the comparison,
test I:'=lsin2(1/n) is convergent, i.e. I:'=11(-l)n+1sin2(1/n)1 is conver-
gent. Thus I:'= 1( _l)n+ 1 sin2(1/n) is absolutely convergent and therefore
convergent.

93
 GUIDE TO ANALYSIS

We have introduced the notion of absolute convergence. But what have we

gained? What advantages does absolute convergence bestow on us? To obtain
some insight we go back to some ideas introduced in the previous section.
We found that the conditionally convergent series
1-t+t-!+!-i+ ...
could be rearranged to give any desired sum. However, an absolutely
convergent series does not have this property. In fact every rearrangement
of an absolutely convergent series has the same sum. So absolutely convergent
series can be rearranged (and multiplied) without changing the sum. The
proof of this is quite long and rather complicated. Some readers may,
therefore, wish to omit the detailed proofs (which comprise the remainder of
this chapter) and may be content to make a note of the following statements.
(a) Every rearrangement of an absolutely convergent series has the same
sum as the original series.
(b) Absolutely convergent series can be multiplied.
For the sake of the brave souls who refuse to be deterred by threats of
long complicated proofs, we begin by showing that any rearrangement of an
absolutely convergent series is also absolutely convergent. The question of
whether the sum of the series is altered by the change of order of the terms
is deferred for just a little while.

THEOREM 3.4.2 Let I:; I an be an absolutely convergent series. Then

every rearrangement of it is also absolutely convergent. Moreover, if L::; I vn
is any such rearrangement, then

Proof Let L::; I vn be a rearrangement of L::; I a". Write

s" = lail + la21 + ... + la"l,
tIl = IVII + IV21 + ... + Iv"l·

Since L::; I an is absolutely convergent (i.e. L::; rianl converges), s" tends to
10 a finite limit s as n -+ 00. Now all the terms la,,1 are non-negative and so (s,,)
is an increasing sequence which converges to s. Thus
(1 )
for all n. Now if n is any given positive integer, there is a corresponding
positive integer M such that all the terms lVII, IV21, Iv 31, ... , Iv,,1 (and probably
many more) belong to the set {laII, la 21, la31, ... ,laM I} (since L::;Ilv,,1 is a
rearrangement of L::; I Ian I. Hence

94
 INFINITE SERIES

by relation (1). Thus (2)

for all n. Now the terms Ivnl are all non-negative. Hence (t n ) is an increasing
sequence. Since (tn) is also bounded above (see relation (2», it follows that
tn tends to a finite limit t as n-+ 00, where
t~s (3)
by (2). This means that I~= 11 vnl converges, i.e. I~= 1vn is absolutely
convergent and I~= 11 Vn I = t, where t ~ s by relation (3).
The inequality t ~ s is a direct consequence of the assumption that I ~= 1 IVn I
is a rearrangement of I~= danl. However, we could equally well treat
I~= 11 an I as a rearrangement of the series I~= 11 vnl. The preceding argument
(with an and Vn interchanged) would then give
s~t (4)
Relations (3) and (4) together give
s= t,

i.e.

Now we can address the question of whether the sum of an absolutely

convergent series is altered by rearrangement. Let us begin with the special
case of a series whose terms are all non-negative real numbers.

THEOREM 3.4.3 Let I~= 1 an be an absolutely convergent series of non-

negative real numbers. Then every rearrangement of this series is absolutely
convergent and has the same sum.

Proof Let I~= 1 vn be a rearrangement of I~= 1 an' Since an ~ 0 for all n, it

follows that Vn ~ 0 for all nand
(1)
In this case, therefore, absolute convergence and convergence mean the same
thing. Using the previous theorem we see that I:..
1 vn is absolutely convergent
and

In view of relation (1) this gives

and the proof is complete.

95
GUIDE TO ANALYSIS

But what happens in the general case when the series contains both positive
and negative terms? In this case, we separate out the positive and negative
terms, in much the same way as in the proof of Theorem 3.4.1.

THEOREM 3.4.4 Let L:'=

1 an be an absolutely convergent series of real
numbers. Then every rearrangement ofthis series is also absolutely convergent
and has the same sum.

Proof Let L:'= 1 Vn be a rearrangement of L:'= 1 all' Then L:'= 1 VII is

absolutely convergent by Theorem 3.4.2.
The positive and negative terms are separated out in the following way.
Define bn,cn,xlI,Yn by the relations
if an ~O, if an ~ 0,
bn = {an
0 if an < 0,
C ={O
-an
n if all < 0,
if Vn ~O, if VII ~ 0,
XII = {Vn
0 if VII < 0, YII = {~Vn if VII < O.
Then L:'= 1 bll contains the non-negative terms of L:'=
1 all' and 1 XIIL:'=
contains the non-negative terms of L:'=
1 VII' Since L:'=
1 VII is a rearrangement
of L:'=1 all' it follows thatL:'= 1 xn is a rearrangement of L:'=
1 bll. Now bn ~ 0
for all n, and L~ 1 bll is absolutely convergent by the comparison test, since
Ibnl ~ lanl for all nand L:'=tlanl converges. By the previous theorem
(Theorem 3.4.3), we see that L:'=1 xn is absolutely convergent and

(1)

Similarly, L:'=
1 Yn is a rearrangement of L:'=
1 Cn' Using the same argument
as before we see that L:'= 1 cn and L:'=
1 Yn are both absolutely convergent and

00 00

LYII=
11=1
L CII'
11=1
(2)

By definition, an = bll - CII and VII = X II - YII' From equations (1) and (2),

The rearranged series L:'=

1 VII therefore has the same sum as the original
series. Thus all rearrangements of an absolutely convergent series have the
same sum as the original series.

Finally, we consider what happens if we multiply two absolutely convergent

96
 INFINITE SERIES

series. Using the normal rules of algebra to take out the brackets, the product
(a 1 + a2 + a3 + .. .)(b 1 + b2 + b3 + ... )
of the two series L:'= 1 an and L:'= 1 bn> is the sum of terms of the form a,b"
where the sum is taken over all possible values of rand s. These terms can
be displayed in the following infinite array:
alb 1 al b2 a 1 b3 a1 b4
a2 b l a2 b2 a2 b3 a2b4
a3 b l a3b2 a3b3 a3b4
a4b 1 a4b2 a4b3 a4 b4

.. .. .. ..
They can be arranged in many different ways to give a series. For example,
we could start in the top left-hand comer and travel along diagonals to give
the series.
(1)
In this case the first bracket contains terms a,b. for which r + s = 3, the second
terms for which r + s = 4, etc. In fact, if we write
en = a 1 bn- 1 + a2bn-2 + ... + an- 1 b 1
n-l
= L a,bn _,
,= 1
(n ~ 2), (2)

then en is the sum of all terms a,b. for which r + s = n. The series (1) could
then be written as L:'= 2 en, a form which is commonly called the Cauchy form
of the product.
Another way of ordering the terms a,b. is to start again at the top left-hand
comer and then to travel along sides of a square. This would give the series
a 1 b 1 + (a 1 b2 + a2b2 + a2b d + (a 1 b3 + a2b3 + a 3b3 + a3b2 + a3bl) + ... ,
which is a very convenient order in some circumstances. For if we take the
first term and the first bracket, these four terms are (a 1 + a2)(b 1 + b2), while
the first term and first two brackets contain the terms which appear in the
product (a 1 + a2 + a 3)(b 1 + b2 + b3), and so on. The reader should remember
these two special examples as these two ways of rearranging the terms a,b.
in a series will be used in the following theorem.
Initially, we arranged the terms a,b. to produce a series L:'=2Cn, where
en = a l bn-l +a2bn-2+ ... +an - 1 b 1 •
The terms a,b. are those which appear when we multiply together the two
series L:'= 1 an and L:'= 1 bn· Is the value of L:'= 2 en equal to the value of
product (L:'= 1 an)(L:'= 1 bn )? More generally, suppose we arrange the terms

97
GUIDE TO ANALYSIS

arb. in some order to produce a single series, which we will denote by

L~s= 1 arb•. Is L~.= 1 arb. convergent? If it is convergent, is its sum equal to
the value of the product (L:'= 1 an)(L:'= 1 bn)? The next theorem supplies an
answer.

THEOREM 3.4.5 Let L~ 1 an and L~ 1 bn be two absolutely convergent

series ofreal numbers. Then the series L~ s= 1 arb. is absolutely convergent and

Proof Let tn be the sum of the moduli of the first n terms of the series
L~. = 1 arbs· Then there is some corresponding positive integer M such that
all these moduli (and probably many more besides) are contained in the
square array below:
lalbll lalb2 1 la l b3 1
la2bll la 2 b2 1 la 2 b3 1

The sum of all the terms in this square array is (L~ Ilarl)(L~ lib. I)· Hence

tn ~ (Jllarl)(Jllb.l) = RMSM, (1)

where RM = lall + la 2 + ... + laMI

1

SM = Ibll + Ib 2 + ... + IbMI·

1

Now the sequences (R M) and (SM) are both increasing. Moreover, RM and
SM both tend to finite limits as M ~ 00, because L:'= llanl and L:'= llbnl both
converge. Let R=limM-+coRM and S=limM-+coSM. Then

for all M. From (1), it now follows that

tn~RM SM~RS

for all n, i.e. the sequence (tn) is bounded above. Since tn is the sum of the
moduli of the first n terms, (tn) is increasing. Hence tn tends to a finite limit
as n ~ 00 and, therefore, the series L~ s = I arbs is absolutely convergent. This
guarantees that the sum is unaltered when the order of the terms is changed.
In order to prove that

we arrange the series in the following way. Start at the top left-hand corner.

98
 INFINITE SERIES

Then follow the arrows as in the diagram below:

a4 b l
This gives a series beginning
alb l + a2b l + a2 b2 + al b2 + a3 b l + a 3 b2 + ....
The first n2 terms of this series can be arranged in the following square array:
alb l alb2 a l b3 alb.
a2b l a2b2 a2b3 a2bn

anb l anb2 anb3 anb n

which has sum 1'n2 given by

As n -+ 00, 1'n2 -+ AB, where

"a
00 00

A= L... k' B= Ib ,.
k=l 1= 1

Hence the absolutely convergent series L~ s = 1 a,bs has sum AB. Thus

'.~l a,bs = AB = C~l an )(Jl bn)

and the proof is complete.

Since the series I~ s = 1 a,bs is absolutely convergent with sum AB, we can
rearrange it and the sum remains AB. If we collect together all the terms arbs
for which r + s = n, we obtain
(n ~ 2).
Write c.=albn-l+a2bn-2+ ... +an_lbl (n~2).

Then we need to sum over all n ~ 2, to obtain all the terms of L~s=la,bs.
Thus I~s=larbs can be rearranged so as to produce I:=2Cn. This gives the
following.

99
GUIDE TO ANALYSIS

COROLLARY Let L:=

1 an and L:= 1 bn be two absolutely convergent
series of real numbers. For n ~ 2, write
en = a 1 bn - 1 + a2 bn-2 + ... + a n - 1b 1•
Then L:= 2 en is absolutely convergent and

The series L:=

2 en if often called the Cauchy form of the product in honour
of the celebrated French mathematician Cauchy, who made many contri-
butions to analysis, as the reader may already have noticed!

APPENDIX
In Chapter 3 we discovered that the series
1-!+!-i+!-!+~-i+ ...
can be rearranged to give convergent series with different sums. In fact, it
was confidently asserted that rearrangements exist with any chosen sum. This
section contains a brief sketch of a method of constructing such rearrange-
ments. Only the outline of the method is given; the details are left to the reader.
We first note that the sum Sn of the first n positive terms is given by
11111
sn = 1 + 3" + 5" +"1 + 9 + ... + 2n - 1. (1)

1 1 1 1 1 1
Now, sn>2+4+"6+8+ 10 + ... + 2n

=~( 1 +~+~+ ... +~)

= !(log n + Yn),
and therefore Sn -+ 00 as n -+ 00.
Similarly, if tn is the sum of the first n negative terms, then
11111
tn = -2-4-"6-8-'" - 2n (2)

= -!(logn + Yn)

and tn - - 00 as n - 00. Thus if we choose any non-negative number x then

the sum of the first n positive terms exceeds x, provided n is sufficiently large.
Admittedly, the number n may need to be enormous if it is to be 'sufficiently
large'. For example, if x = 2000, then the number n would need to be very

100
 INFINITE SERIES

large indeed to ensure that Sn > x; but this doesn't matter. What is really
important is that such a number n exists.
Similarly, if x < 0 then we can choose a sufficiently large number of
negative terms so that the sum is less than x.
Suppose now that we choose any non-negative real number x and we wish
to develop a method for constructing a series with sum x. We begin the series
by writing down the first n positive terms, where n is chosen so that the sum
Sn (see equation (1)) exceeds x, but Sn-1 ~ x. For example, ifthe desired sum x
is n/4, then n = 1 and the series starts with just the first positive term 1.
However, if x = 1 then we need the first two positive terms 1 + t, if the sum
is to exceed x.
These first n positive terms are then followed by the first m negative terms,
where m is chosen so that the total sum Sn + tm of all m + n terms is less than
x, but Sn + t m - 1 ~ x, i.e. we include just enough negative terms so that the
total sum is less than x. For example, if the required sum x is n/4, then m = 1
and the series starts 1 - t, whereas if x = 1 then n = 2, m = 1 and the series
commences 1 + t - 1. However, if the desired sum x = /0' then m = 3 and
the series begins with the terms
I-t-t-i·
Now that the sum is less than x, we continue moving along the series, taking
positive terms and writing them down. As each one joins the new series, we
check the total sum and stop as soon as the total sum again exceeds x. It is
then time to include the next negative terms. We continue in this manner,
including blocks of positive terms and blocks of negative terms. Each time
we stop adding positive terms as soon as the sum exceeds x and we stop
including negative terms just as soon as the total sum is less than x. This
gives a recipe for constructing a rearrangement with sum x. Naturally, if
x < 0, then we need to start with negative terms.

Examples

The examples below show the first few terms of the rearrangements produced
by the above method for given values of x.
1. x = n/4. The rearranged series begins

1- t + t - t + ! + t - i + ~ - i + 1\ - /0 + ...
2. x = O. The rearranged series begins
1- t - t - i - i + t - /0 - /2 - l4 - /6 + ! - ...
3. x = n/3. The rearranged series begins
1+ t - t + ! - t + t + ~ - i + /1 + l3 - i + /5 + ...
101
GUIDE TO ANALYSIS

4. x = --b.. The rearranged series begins

- ! + 1- i - i- - i- - lo + !- - l2 - l4 - l6 - ls -lo + ! - ...
5. x = 1. The rearranged series begins
1+!--!+!-i+t+l-i+ ...

It can be shown that this construction gives a convergent series whose sum
is x. The following exercises provide a guide for the reader who is brave
enough to wish to embark on a proof.

EXERCISES
In all of the following exercises it is assumed that the above construction is
being used to produce a rearrangement of the series
1-!+!--i+!-i+t-i-+ ...
with sum x.
1 When x ~ 0, the rearranged series begins with a block of n positive terms.
Verify that
1
o< s" - x < 2n _ 1'

where s" is the sum of these n positive terms.

2 Check that for all real values of x the difference between x and the sum
of all the terms in the first block does not exceed the value of the modulus
of the last term in the block.
Check also that the difference between x and the sum of all the terms
in the first N blocks does not exceed the value of the modulus of the last
term in the Nth block.
3 (Harder) Let tk be the sum of the first k terms of the rearranged series.
Suppose the number of terms in the first N blocks is less than or equal
to k (i.e. all the terms of the first N blocks, and possibly many more besides,
are included in the first k terms). Show that
1
Itk-xl< N -1 (N~ 2).

(Take care! If x < 0 then the series begins with a block of negative terms.
In this case, the term 1 must be in the second block etc.)
Deduce that the rearranged series converges and has sum x.

102
 INFINITE SERIES

TESTS FOR CONVERGENCE-SUMMARY

Comparison test (po 71)
Limit form of comparison test (po 73)
D' Alembert's test (ratio test) (po 76)
Integral test (po 83)
Alternating series test (po 89)

MISCELLANEOUS EXERCISES 3
Decide whether the following series converge or diverge:

1
f (3n)!r n
n= 1 n!(2n)! '
0

2 L00 ( --n
n=1 n+l
r '
0
3 f: sin no
n= 1 n
2 ,

4 f -tan
1
n=1Jn
C2n -4
1 )1t)
'
0
5
00

L 3n
n=1
n2 + 2
3
+ 4;n
00
6 I-o
1
"=2 n log n'
(5n)!
L (_I)"n1/n; f (-I)nSin(~): 9
00
00
7
"=1
8
"-1 n "~1 (30)"(3n)!(2n)!;
00(n 2 + 1)4 2 + (-1)" n2 + 1
10
00
12
00

"~dn3 + 1)3; 11 L
"= 1 n
3/2;
n~1 4n4 + n2 - n - 1;
00 (2n)! sin(n!) (-I)"n2
13
00

L (1)2;
00
14 L 2 1; 15 I 1;
"=1 no "=1n +n+
3
"=1 n +

16
f: 9n(3n)!n! 0

17 L 2+(-1)"
00
18 f: sin ( 1tn2 )cos(?:n ):
(4n)! ' In '
0

"= 1
"=1 "=1

19 L00 n2 + n + 1
n=1Jn 7 +n4+ l'
0
20 L sm
00

"=1
0 C-I)")
--;
n
21
00 1
J1 n(log n)3/2 0

HINTS FOR SOLUTION OF EXERCISES

Exercises 3.2.1

2 Use IOg( 1 +~)=IOg[(n+ 1)/n] =log(n+ l)-logno

103
GUIDE TO ANALYSIS

ANSWERS TO EXERCISES
Exercises 3.2.1
1 !. 2 sn=log(n+ 1).

Exercises 3.3.1
3 In each case the comparison test or the limit form of the comparison test
is used. The series I:'=
1 bn used for comparison is given in brackets.

(a) Diverges (b n = lin). (b) Converges (b n = l/n 2 ).

(c) Converges (b n = 1/n ). (d) Diverges (b n = lin).
2

(e) Converges (use l/(nn+l/n) < 1/n2 for all n ~ 2).

(f) Diverges (b n = 1/nl/3).

Exercises 3.3.2
3 In parts (a) and (f) d'Alembert's test is used. The comparison test or its
limit form are used for all the rest. The series I:'=
1 b n used for comparison
is given brackets.
(a) Converges (d'Alembert's test or comparison test with bn = (tn.
(b) Converges (b n = 1/n 2). (c) Diverges (b n = n5 ).
(d) Diverges (b n = l/Jn).
(e) Diverges (b n = lin). Remember that (1 + l/n)n--+e as n--+ 00.
(f) Converges.
(g) Converges (b n = 1/n2). Remember that Isinnl ~ 1 for all n.

Exercises 3.3.3
3 (a) Converges (alternating series test,
sin[(n + 1)n/n] = sin(nn + n/n) = ( _1)n sin(n/n».
(b) Conver~es (comparison test, bn = l/nJn).
(c) Diverges (an+O as n-Hx:J). (d) Converges (integral test).
(e) Diverges (an+O as n-+oo). (f) Converges (d'Alembert's test).

Miscellaneous Exercises 3
1 Diverges (d'Alembert's test).
2 Diverges (an +0 as n -+ (0).
3 Converges (comparison test with bn = 1/n2 proves series is absolutely
convergent).
4 Converges (alternating series test and tan[(2n - l)n/4] = ( -It+ 1).
5 Diverges (comparison test, b = lin).

104
 INFINITE SERIES

6 Diverges (integral test).

7 Diverges (an ~ 0 as n -+ (0).
8 Converges (alternating series test).
9 Converges (d'Alembert's test).
10 Diverges (comparison test, bn = lin).
11 Converges (comparison test, bn = l/nJn; use the fact that
(2 + ( -1)")ln 3 / 2 :::; 3In 3 / 2 ).
12 Converges (comparison test, bn = 1/n 2 ).
13 Di verges (d' Alem bert's test).
14 Converges (comparison test with bn = 1/n 2 proves that the series is
absolutely convergent).
15 Converges (alternating series test).
16 Converges (d'Alembert's test).
17 Diverges (comparison test, bn = 1/Jn. Use [2 + (-I)]IJn ~ 1/Jn).
18 Diverges (a 2n + 1 =( _1)n+1 cos(nln) and an~O as n-+oo).
19 Converges (comparison test, bn = 1/nJn).
20 Converges (alternating series test; use the fact that
sin [( - l)n In] = ( -1)" sin( 1In)).
21 Converges (integral test).

105
4 FUNCTIONS
Limits of functions, continuous
functions

4.1 LIMITS OF FUNCTIONS

The principal characters in this chapter are real-valued functions. For the
benefit of any reader whose knowledge of functions may be a little shaky,
we recall some basic ideas. As we are only concerned with certain types of
functions, the definitions are restricted to those of interest to us, viz. the real
valued functions. (Note: A more comprehensive treatment of functions is
contained in the companion volume in this series, Guide to Abstract Algebra.)
First suppose X and Yare both sets of real numbers. Suppose also that there
is some rule which assigns to each real number xeX a unique real number
ye Y. Then this rule defines a function from X into Y. If we use the letter f
for the function then we would write f: X -+ Y to signify that the function is
from X into Y. Sometimes it is convenient to think in terms of the diagram
shown as Fig. 4.1.
If the number ye Y corresponds to the number xeX, then we write y = f(x).
The set X is called the domain of definition of f (sometimes abbreviated to
the 'domain off') and the phrase 'f(x) is defined' means xeX, i.e. x is in the
domain of definition of f and so f(x) has a definite value. The set of all
numbers ye Y such that y = f(x) for some xeX is denoted by f(X) and is
called the range of f Thus the range of f is the set f(X), where
f(X) = {yeY:y=f(x) for some xeX}.

Fig. 4.1

106
 FUNCTIONS

For any XEX, the number f(x) is called the image of x. The set f(X) is
therefore the set of images of all numbers in X. This set f(X), of course, need
not be the whole of the set Y. All the definition stipulates is that every element
of the form f(x) (XEX) belongs to Y; it does not require that these are the
only members of Y. In practice, therefore, it is often convenient to use the
whole of IR as the set Y. The majority of functions we consider will be treated
as functions f: X -+ IR. Occasionally, in particular circumstances, we will need
to use a set Y which does not include all the real numbers, but such instances
will be rare in this volume. Below we give a few examples of functions.

Examples 4.1.1

1. The function f: IR -+ IR is given by

f(x) = x2
for all xelR.
2. The function f: IR -+ IR is given by
if x is rational,
f(x) = {~ if x is irrational.
3. Let X be the set of all positive real numbers. The function f: X -+ IR is
given by
if x is irrational and x > 0,
if x is rational and x = p/q,
where p, q are positive integers with no
common factor greater than 1.
(The condition that p, q have no common factors means that the fraction
p/q has been cancelled down to its lowest terms.)
4. The function f: IR\ {O} -+ IR is defined by
f(x) = l/x (XEIR, x ~ 0).

In Example 1, we notice that x 2 ~ 0 for all real numbers x and so the range
is the set of all non-negative real numbers. In Example 2, f(x) is either 0 or
1 and the range is the set {O, I} with just two elements. The function in
Example 3, is at first sight a little strange. It satisfies the rules for a function
since there is precisely one corresponding value f(x) assigned to each positive
real number x and so it is undoubtedly a function. Its range consists of all
the rational numbers l/q for which q is a positive integer, together with the
number o. Example 4 is left to the reader.

107
 y
y = e"
55

50

45

40

35

30

25

20

15

10

-4 -3 -2 -1 o 2 3 4 x

Fig. 4.2

y = l/x

-4 -3 -2 -1 0 2 3 4 x

-1

-2

-3

-4

Fig. 4.3

108
 FUNCTIONS

It will not have escaped the reader's attention that Examples 1 and 4 are
familiar functions whose graphs are easy to sketch, whereas Example 3
illustrates that there are functions whose graphs cannot be sketched (even
with a considerable amount of ingenuity). We must bear this in mind in the
rest of the chapter. Our concepts must be framed in such a way that they
can be used with functions whose graphs cannot be drawn. They must,
therefore, be couched in terms of symbols rather than pictures, although we
will use sketches to give us the initial ideas.
Throughout the remainder of the chapter we will be considering functions
f: X --+~, whose domain of definition X contains an interval (or intervals). It
will probably simplify the discussion if we now introduce the required notation
for intervals. In evolving the notation we use two letters a, b to denote
arbitrary real numbers.

Intervals: notation
1. The set of all real numbers x such that x> a is denoted by (a, (0).
2. The set of all real numbers x such that x ~ a is denoted by [a, (0).
3. The set of all real numbers x such that x < b is denoted by ( - 00, b).
4. The set of all real numbers x such that x ~ b is denoted by (- oo,b].
5. Let a, b be real numbers with a < b. The set of all real numbers x such
that a < x < b is denoted by (a, b) and is called an open interval.
6. Let a, b be real numbers with a < b. The set of all real numbers x such
that a ~ x ~ b is denoted by [a, b] and it is called a closed interval.

We now turn to the question of limits of a function.

To give us some ideas on how to tackle the problem we begin by sketching
the graphs of some fairly elementary functions. These are shown as Figs 4.2
to 4.6.
It seems intuitively obvious that (a) eX --+ 00 as x--+ 00, (b) 1/x--+O as x--+ 00,
(c) 1/(1 + x 2 )--+O as x--+ 00 and (d) (x 2 -l)/(x 2 + 1)--+ 1 as x--+ 00. The last
example may cause a little hesitation. The graph seems to suggest that
x + sin x --+ 00 as x --+ 00. A moments thought suffices to confirm this
conjecture, since x + sin x ~ x - 1 for all x and quite obviously x - 1 --+ 00 as
x --+ 00. But is not sufficient just to sketch graphs. The functions we have used
as illustrations had graphs which could be drawn without undue difficulty.
There are, however, functions whose graphs are very difficult to draw and
functions whose graphs are impossible to draw. Our definitions of concepts
like the limit of a function must be applicable irrespective of whether the
graph of the function can be sketched or not. The definitions we eventually
give must therefore be in terms of symbols. The form in which they are cast
has not changed over the last century. In fact, they are essentially the same
as those given by Heine in 1872.
Let us now return to our examples. It seemed obvious that eX --+ 00 as
x --+ 00 and x + sin x --+ 00 as x --+ 00. What properties of the functions led to
this conclusion? Clearly eX increases as x increases and there is no limit on

109
 Y
1.1

1.0

0.9

I
0.8
I
0.7

,.6I
0.5

-8 -6 o 6 8 x
Fig. 4.4

Y
-----------------1.0 ------------------
x 2 -1
y=--
x2 + 1

-8 -6 0 2 4 6 8 x
-0.2

\
-0.4
\
-0.,

-0.8

Fig. 4.5

110
 FUNCTIONS

y
y =x +sinx
25

20

15

10

-25 -20 -15 -10 5 10 15 20 25 x

Fig. 4.6

its size i.e. there is no upper bound on the values assumed by eX. In fact, if
we choose any positive real number A (however large) then eX> A for all
sufficiently large x, i.e. eX remains greater than A for all sufficiently large x.
Since eX is a fairly easy function to handle we can be a little more precise.
Let A be any positive real number. Then eX> A whenever x> 10gA. If we
write Xl = log A, then we have eX> A for all x> Xl. Could we make a
similar statement about the last example? The answer is yes. For x + sin x > A
whenever x> A + 1. Suppose that Xl = A + 1. Then if x> Xl we have
x+sinx~x-l >X 1 -1 =A.
Since the graph fluctuates a little locally, there will also be values of x with
x ~ A + 1 for which x + sin x> A. But this doesn't matter. The vital property
is that the graph remains above the line y = A for all sufficiently large values
of x (in this case x> A + 1 will do). This is just the property we need to
develop our definition. We will, of course, need to assume that the function
f has the property that f(x) has a value for all very large values of x.

DEFINITION 4.1.1 Letf be a real valued function such thatf(x) is defined

for all x > a (where a is some given real number). Then we say that f(x) -+ 00

111
GUIDE TO ANALYSIS

as x -+ 00 if given any A > 0 (however large), there is a corresponding Xl

(whose value depends on A) such that Xl> a and
f(x»A for all x > Xl.
Note: The requirement that f(x) is defined for all x > a means that the domain
of definition off contains the interval (a, (0), i.e. it contains all the numbers
x for which x > a. Normally the symbol X is used in the definition rather
than X 1. However, the letter X has already appeared for the domain of
definition of the function, and it seemed unwise to use the same symbol for
two different things. The reader may wish to revert to the normal notation
and omit the SUbscript 1 when there is no danger of confusion.

Examples 4.1.2

1. Let f(x) = x 2 for all real x. Given any A> 0, we see that f(x) > A for all
x> +JA. In this case, therefore, we can use Xl = +JA
and f(x)-+ 00
asx-+oo.
2. Let f(x) = log(log x) for x> 1. Given any A > 0, we see that f(x) > A for
all x> eeA and in this case we can use Xl = eeA.
3. Let f(x) = X s + X4 + xsinx + 1. Then it looks obvious that f(x)-+ 00 as
x -+ 00. But how do we prove this using the definition? The straightforward
method of solving equations, which proved useful in the first two examples,
doesn't look as promising here. In this case the problem is simplified by
using some inequalities.
We first notice that if x > 1 then
x3+sinx~x3-1 >0,

and therefore
X4 + xsinx = x(x 3 + sin x) > o.
Thus, if x> 1 then
f(x) = XS + X4 + xsinx + 1 > X S + 1 > xs.
Given any A > 0; Xs > A for all x > A 115. Let Xl be the larger of the two
numbers 1, A11 s, i.e. Xl =max{1,A1/S}. Then for all x> Xl>
f(x»x s > A.
By definition f( x) -+ 00 as x -+ 00.

We notice that the working was simplified, in the above example, by the use
of inequalities. The reason is not hard to understand. The graph of y = f(x)

112
 FUNCTIONS

is above the graph of y = x 5 for all x> 1, since J(x) > X S for all x> 1. We
know what happens to the lower graph, since x 5 -+ 00 as x -+ 00. It follows
that J(x)-+ 00 as x-+ 00.
Naturally, this idea can be generalised. Suppose we have two functions J, g
with the property that the graph of flies above the graph of g, i.e.J(x) ~ g(x).
If g(x)-+oo as x-+oo, then it is clear thatJ(x)-+oo as x-+oo. Of course, it
is not really necessary to be able to draw the graphs. The inequality
J(x) ~ g(x), together with the assumption that g(x) -+ 00 as x -+ 00, is sufficient
to guarantee that J(x)-+ 00 as x-+ 00. A result like this is of such practical
value that it merits incorporation in a theorem.

THEOREM 4.1.1 LetJ,g be real valued functions such thatJ(x) and g(x)
are defined for all x> a and satisfy the inequality
J(x) ~g(x) (x> a),
where a is some given real number. If g(x) -+ 00 as x -+ 00, then J(x) -+ 00 as
x-+oo.

Proof' Let A be any positive real number. Since g(x)-+oo as x-+oo, there
is some Xl (with Xl> a) such that
g(x»A for all x>X 1 •
Thus for x > Xl J(x) ~ g(x) > A,
and, therefore J(x)-+ 00 as x-+oo.

Example 4.1.3

Let J(x) = x 2 + X + sin x,

g(x) =x 2 •
For x> 1, J(x) = x 2 + X + sin x ~ x 2 + x -1> x 2 = g(x)
and therefore J(x)-+ 00 as x-+oo because g(x)-+oo as x-+oo.

The reader may notice that there is a great similarity between the ideas in
section 2.2 on sequences and this section on limits of functions. For example
there is a great resemblance between Definitions 2.2.2 and 4.1.1, and between
Theorems 2.2.1 and 4.1.1. This is not accidental. The basic ideas are to a
great extent the same and the reader may find it helpful to refer back to
Chapter 2 when studying this section.

113
GUIDE TO ANALYSIS

EXERCISES 4.1.1

1 Use Definition 4.1.1 to show that x 3 -+ 00 as x -+ 00 (i.e. show that for each
A > 0 there is a corresponding XI with the required properties).
2 Use Definition 4.1.1 to show that, x"-+ 00 as x-+ 00 for each real number
IX >0.

3 Show that f(x) -+ 00 as x -+ 00 for each ofthe following functions f:

(a) f(x) = x 3 + x 2 + xcosx;
(b) f(x) = xlogx (x> 0).

We know what kind of picture to expect if f(x)-+oo as x-+oo. If for each

A> 0 (however large), the graph of y = f(x) remains above the line y = A for
all sufficiently large x, then f(x) -+ 00 as x -+ 00. But what type of behaviour
do we need if f(x) -+ - oo? If A> 0, then the line y = - A is below the axis.
The necessary requirement is that the graph of y = f(x) should remain below
the line y = -A for all sufficiently large x, i.e. f(x) < -A for all sufficiently
large x. Now the formulation of a precise definition in symbols causes no
problems.

DEFINITION 4.1.2 Suppose f(x) is defined for all x> a, where a is some
given real number. Then f(x) -+ - 00 as x -+ 00, if given any A> 0 (however
large) there exists a corresponding XI (with XI> a) such that
f(x) < -A for all x > X l'

A quick glance at Definitions 4.1.1 and 4.1.2 will immediately convince the
reader that f(x)-+ - 00 as x-+ 00 if and only if - f(x)-+ 00 as x-+ 00.
Obviously Theorem 4.1.1 has an analogue for functions with limit - 00.
In this case we want to use a function g whose graph is above that of f and
such that g(x)-+ -00 as x-+oo. This gives the following result.

THEOREM 4.1.2 LetJ,g be real valued functions such thatf(x) and g(x)
are defined for all x > a and
f(x) ~g(x)

for all x> a, where a is some given real number. If g(x) -+ - 00 as x -+ 00,
then f(x)-+ - 00 as x -+ 00.
The proof is left as an exercise for the reader.

EXERCISE 4.1.2

Write out a prooffor Theorem 4.1.2 (Hint: Either start with Definition 4.1.2
and construct a proof on similar lines to that of Theorem 4.1.1 or apply the

114
 FUNCTIONS

conclusions of Theorem 4.1.1 to the functions - f and - g and use the fact
that f(x) -+ - 00 as x -+ 00 if and only if - f(x)-+ 00 as x -+ 00.)

We recall that some of the examples near the beginning of the chapter had
infinite limits whilst others had finite limits. For example, it seemed intuitively
obvious that ~ -+ 0 as x -+ 00, and (x 2 - 1)/(x 2 + 1) -+ 1 as x -+ 00. How can
we set about writing down a precise definition in terms of symbols for finite
limits? Perhaps a glance at Chapter 2 will give us some ideas. Let us begin
by looking at Fig. 2.3. This shows a sequence (an) with the property that
an -+ 1 as n -+ 00. In this case every band about y = 1 (however narrow)
contains all the elements an for which n is sufficiently large. Normally, the
band width is denoted by 2e and so we have all the elements an for which n
is sufficiently large between 1 - e and 1 + e. In general, if the limit is I then
all the elements an for which n is sufficiently large lie between / - e and I + e
(see Definition 2.2.4). The same ideas carryover to functions and are used
to define the meaning of f(x) -+ I as x -+ 00. We now require that for all e > 0
(however small) the graph of y = f(x) lies between the lines y = 1- e and
y = I + e for all sufficiently large x, i.e.
1- e < f(x) < I + e (1)
for all sufficiently large x. For example, if f(x) = ~ (x > 0), then, given e > 0,
-e<f(x)<e
for all x>;. This is of the same form as (1) with 1=0, which is not really
suprising since it appears that ~-+O as x -+ 00 (see Fig. 4.7). It is fairly clear
that however narrow this band is chosen to be (i.e., however small we chose e),
the graph of y = ~ lies inside the band for all sufficiently large values of x.
We are now in a position to give a formal definition. In the statement of the
definition the relation /- e < f(x) < / + e will be expressed as If(x) -II < 6.

DEFINITION 4.1.3 Suppose f(x) is defined for all x> a, where a is some
given real number. Then f(x) -+ / as x -+ 00 if given any 6> 0 (however small)
there is a corresponding X 1 (with Xl> a) such that
If(x)-II <6

In each of the Definitions 4.1.1 to 4.1.3 the concept of a limit is defined

subject to the condition that f(x) is defined for all x> a (where a is some
given real number). Throughout this volume phrases like f(x) -+ I will be used
only when such a condition is satisfied. Thus, if it is stated that f(x)-+l as
x -+ 00, then the reader knows that there is some real number a such that
f(x) is defined for all x> a.

115
GUIDE TO ANALYSIS

Y = 1/x

X= 1 /e

y=B
o ________ .. ______________________________________ . x
Y= -B

Fig. 4.7

Examples 4-1.4

1. We have seen that given any e > 0,

1
-e<-<e
x
for all x >~,
i.e.

for all x > ~. Hence ~ ---+ 0 as x ---+ 00.

2. Given any e > 0, we see that

Isinx x I:s:; _1_

Ixl
<e

for all x >~. Hence (sinx)/x ---+0 as x ---+ 00. In this example and the previous
one Xl = t. As we already know, the value of X 1 depends on the value
of e, and the value of X 1 appears to increase as e gets smaller, which is
exactly what we would expect.

116
 FUNCTIONS

3. For all x,
x 2 -1 2
--=1---
x2 + 1 x 2 + 1·
Thus, given any e > 0,

I:: ~ ~ -11 < x2 ~ 1< 8,

provided x 2 + 1 > i. Thus the inequality is certainly satisfied for all x > 0
such that x 2 + 1 > i, i.e. it is satisfied for all sufficiently large x. Hence
(x 2 -1)/(x 2 + 1)-+ 1 as x-+ 00.

EXERCISES 4.1.3

1 Using Definition 4.1.3, show that 1/(x 2 + 1)-+0 as x-+oo.

2 Show that if f(x)-+oo as x-+oo, then l/f(x)-+O as x-+oo. Deduce that
l/xa -+O as x-+ 00 for all real numbers IX> O.

Many of the theorems given in Chapter 2 for sequences have obvious

analogues for functions. For example, a function cannot have more than one
limit.

THEOREM 4.1.3 Let f(x) be defined for all x> a, where a is some given
real number. Thenf(x) cannot tend to two different limits as x-+oo.

Proof We consider several cases separately.

(a) f(x) cannot tend to afinite limit and an infinite limit as x -+ 00 Suppose
f(x)-+l as x-+oo. Then, using Definition 4.1.3 (with the special case
e = 1), we see that there is a number X 1 (with X 1 > a) such that
If(x) -II < 1
for all x > X 1>
i.e. 1- 1 < f(x) < 1+1
for all x> X 1. Sincef(x) < 1+ 1 (x > X 1),f(X)-+ 00 as x -+ 00. Further,
f(x)-+-oo as x-+oo becausef(x»I-1 for allx>X 1 . Thusf(x)
cannot have a limit I and an infinite limit.
(b) If f(x)-+oo as x-+oo then f(x)-+-oo as x-+oo If f(x)-+oo as
x -+ 00 then, using Definition 4.1.1 (with the special case A = 1), we
see that there is some X 1 (with X 1 > a) such that

117
GUIDE TO ANALYSIS

f(x) > 1
for all x> X l' Hence f(x)-fr - 00 as x-+ 00.
Hence f(x) cannot have both 00 and - 00 as limits.
(c) f(x) cannot tend to two different finite limits Suppose f(x)-+I as
x-+oo andf(x)-+m as x-+oo. Then either I=m or I#m.
Case I -# m If 1-# m, there is no loss of generality in assuming I> m.
We now use Definition 4.1.3 (with the special case e= t(l- m». Since
f(x)-+I as x-+oo and f(x)-+m as x-+oo, there is some Xl (with
Xl> a) such that
If(x) -II < t(l- m) (x> Xl)' (1)
If(x) - ml < t(l- m) (x> Xl)' (2)
From relation (1) we have
I-t(l-m) < f(x) < I +t(l- m) (x >X 1 ),
i.e. t(l + m) < f(x) < t(3/- m) (x>X 1 )· (3)
Similarly, relation (2) reduces to
t(3m -I) < f(x) < t(l + m) (x>X 1 )· (4)
Now (3) and (4) cannot both be satisfied for all x> Xl, since
f(x) > t(l + m) from (3) andf(x) < t(l + m) from (4). Thus the assump-
tion 1# m leads to a contradiction and must therefore be false. We
are then left only with the possibility 1= m, i.e. the two limits are in
fact the same.
This completes the proof.

Since a function cannot have more than one limit, the limit, if it exists, is
unique.
The familar 'sandwich rule' also has an analogue for functions.

THEOREM 4.1.4 Supposef,g,h are three functions such thatf(x), g(x),

h(x) are all defined for x> a and satisfy the inequality
g(x) ~ f(x) ~ h(x)
for all x> a. If g(x) -+ I as x -+ 00 and h(x) -+ 1 as x -+ 00, then
f(x)-+I as x-+ 00.

Proof Let e>O. Since g(x)-+I as x-+oo and h(x)-+I as x-+oo, there is
some Xl (with Xl> a) such that
l-e<g(x)<l+e (1)
l-e<h(x)<I+e (2)

118
 FUNCTIONS

for all x> XI. Hence, for all X> XI'

1- B < g(X) ~ f(x) ~ h(x) < 1+ B.
And therefore f(x) -+ I as x -+ 00.

By this stage, the perceptive reader may be beginning to suspect that there
are definite similarities between limits for sequences and limits for functions.
Can these similarities be exploited to cut down on the number of proofs?
Fortunately the answer is yes. The next result demonstrates the relation
between limits for sequences and limits for functions. The underlying ideas
are intuitively obvious, as recourse to a graph quickly illustrates. Suppose
we have a function f such that f(x) -+ I as x -+ 00. Figure 4.8 shows one such
function. Now choose numbers X 1 ,X 2 ,X3, etc., such that Xn-+OO as n-+oo
and consider the sequence (f(x n ))::"= 1 of real numbers. Then it would appear
reasonable to guess that f(xn) -+ I as n -+ 00. This establishes a link between
the limit of f(x) and the limit of the sequence (f(xn)). Our result is contained
in the next theorem.

THEOREM 4.1.5 Suppose thatf(x) is defined for all x > a, where a is some
given real number. Thenf(x)-+l as X-+OO if and only if f(xn)-+I as n-+oo
for all sequences (x n)::"= 1 such that Xn > a for all n ~ 1 and Xn -+ 00 as n -+ 00.
y

y =I

I
i (Xl)
o X

Fig. 4.8

119
GUIDE TO ANALYSIS

Proof Since the theorem states that f(x) -+ 1as x -+ 00 if and only if a certain
condition is satisfied, we have two implications to prove. We must show that
f(x) -+ 1 as x -+ ex) implies that all suitable sequences behave in the required
way. Conversely we must show that the condition on sequences is sufficient
to guarantee that f(x) -+ 1as x -+ 00. The proof is, therefore, broken into two
separate parts. One part deals with the 'only if' from 'if and only if'. The
other concerns the 'if' in the double implication.
(a) 'Only if' (In this part we assume thatf(x)-+l as x-+oo and prove
that this implies f(xn)-+I as n-+ 00.)
Let (X n);:"= 1 be any sequence of real numbers such that Xn > a for all
n ~ 1 and Xn -+ 00 as n -+ 00, and let E > O. Since f(x) -+ I as x -+ 00,
there is some X 1 (with X 1> a) such that
If(x)-/I <E (1)
for all x> X 1. Since Xn -+ 00 as n -+ 00, there is a positive integer N
such that
x n >X 1 (2)
for all n> N. Thus, for all n> N, Xn > X 1 and
If(x n) -II < E

by (1). Hence f(xn)-+I as n-+ 00.

This proves the first part. The converse is considerably harder to
prove.
(b) 'If' (In this part we assume that f(xn) -+ I as n -+ 00 for all sequences
(X n);:"=1 such that xn>a for all n~ 1 and Xn-+OO as n-+oo. We then
have to prove that this guarantees thatf(x) -+ I as x -+ 00.) Let us begin
by considering the consequences of assuming f(x)~l as x-+oo. A
close look at Definition 4.1.3 helps with the problem of expressing in
symbols the statementf(x)~ I as x -+ 00. Sincef(x) ~ I as x -+ 00, there
must be at least one positive value of E for which there is no
corresponding X 1. Let this positive value be Eo. As there is no
corresponding value X 1 for this value Eo, it means that if we take any
number X 1 (however large) there is still a number x> a such that
x> X 1 and If(x) -II ~ Eo. Since this happens for all values of X l' we
can use X 1 = n, where n is a positive integer. Thus for each positive
integer n there is a corresponding number Xn such that Xn > a, Xn > n
and
(3)
From relation (3) we see that f(x n) ~ I as n -+ 00. However, (x n);:"= 1 is
a sequence such that Xn > a and Xn -+ 00 as n -+ 00 (because Xn > n).
We have, therefore, proved that if f(x)~1 as x-+ 00, then there is
a sequence (X n );:"= 1 such that Xn > a, Xn -+ 00 as n -+ 00 and f(xn) ~ I as
n-+ 00. If no such sequence exists then it is impossible to havef(x)~l

120
 FUNCTIONS

as x -+ 00. This means that if f(x,,) -+ 1 as n -+ 00 for every sequence

(x,,):'= 1 such that x" > a (n ~ 1) and XII -+ 00 as n -+ 00, then we must
also have f(x)-+l as x-+oo. The sequential condition is therefore
sufficient to guarantee thatf(x)-+l as X--+CX).
This completes the proof.

Similar results hold for the casesf(x)-+ 00 as x-+ 00 andf(x)-+ - 00 as x-+ 00.
The theorems are stated without proof. The reader who wishes to check their
veracity can produce a proof using the above arguments with appropriate
modifications to suit the particular case.

THEOREMS 4.1.6 Suppose that f(x) is defined for all x> a, where a is
some given real number. Thenf(x)-+ 00 as x-+ 00 if and only if f(x,,)-+ 00 as
n -+ 00 for all sequences (x,,):'= 1 such that x" > a for all n ~ 1 and x" -+ 00 as
n-+ 00.

THEOREM 4.1.7 Suppose thatf(x) is defined for all x > a, where a is some
given real number. Thenf(x)-+ -00 as x-+oo if and only if f(x,,)-+ -00 as
n -+ 00 for all sequences (x,,):'= 1 such that XII> a for all n ~ 1 and XII -+ 00 as
n-+oo.

The connection between limits of sequences and limits of functions allows

us to adapt theorems for sequences to produce their analogues for functions
without the need for additional proof. For example, Theorem 2.2.5 concerning
the algebra of limits immediately gives the corresponding result for functions.

THEOREM 4.1.8 (Algebra of Limits) Letf(x), g(x) be defined for all X > a,
where a is some given real number. Suppose also that f(x) -+ 1as x -+ 00 and
g(x)-+m as x-+oo. Then
(a) f(x) + g(x)-+l + mas x-+ 00,
(b) f(x)g(x)-+lm as x-+ 00.
If, in addition, g(x) satisfies the condition g(x) =F- 0 for all x > a and if m =F- 0,
then
f(x) 1
(c) ---+-asx-+oo.
g(x) m
Proof The prooffollows immediately from Theorems 2.2.5 and 4.1.5. As an
example of how these two results are used we look briefly at (c).
First use Theorem 4.1.5. It guarantees that for every sequence (xll):'= 1 such
that x" > a (n ~ 1) and X,,-+ 00 as n-+ 00
121
GUIDE TO ANALYSIS

f(xn)-+I as n-+ 00,

g(xn ) -+ m as n -+ 00,
since f(x)-+I as x-+ 00 and g(x)-+m as x-+ 00. Moreover, we are given that
m =F 0 and g(x n ) =F O. By Theorem 2.2.5, we see that
f(x n )
---+-
g(xn ) m
asn-+oo.
Since this is true for all sequences such that x n > a, and Xn -+ 00, as n -+ 00
it follows from Theorem 4.1.5 that
f(x)
---+-
g(x) m
asn-+oo.
The proofs of (a) and (b) are also rather obvious consequences of the results
in Theorem 2.2.5 and Theorem 4.1.5.

Naturally, a proof of Theorem 4.1.8 is possible without recourse to sequences

at all. For example, the first proof of the theorem could be proved in the
following way. Let 8> O. Since f(x)-+l as x-+ 00 and g(x)-+m as x-+ 00,
there is some Xl (with Xl> a) such that
1-8/2 <f(x) < 1+ 8/2 for all x> Xl
and m- 8/2 <g(x)< m + 8/2 for all x > Xl.
By addition, 1+ m-8<f(x) + g(x) < 1+ m+ 8
for all x>X 1 and sof(x)+g(x)-+I+m as x-+oo. This form of proof is
modelled on the proof of Theorem 2.2.5 with the obvious modifications to
allow for the fact that we are now dealing with functions and not sequences.
The reader may feel that the direct proof of (a) (without recourse to sequences)
is relatively painless, which is true. However, the advantages of using
sequences becomes abundantly clear when the rule for the quotientf(x)/g(x)
is considered. In this case the direct proof contains many messy details.
Theorem 2.2.7 and its corollary also have obvious analogues for functions.

THEOREM 4.1.9 Suppose thatf(x) and g(x) are defined for all x> a and
satisfy the inequality
f(x) ~'g(x) (x > a),
where a is some given real number. If f(x)-+I as x-+ 00 and g(x)-+m as
x-+ 00, then
I~m.

122
 FUNCTIONS

The proof of this result is left as an exercise for the reader. It demonstrates
that weak inequalities are preserved when limits are taken. In the case of
sequences, strict inequalities did not obey this rule exactly and the wary
reader may suspect that the same is true for limits of functions. Such caution
is exemplary! Even if the condition f(x) ~ g(x) for all x> a is replaced by
the condition f(x) < g(x) (x> a), the conclusion remains in the form I ~ m
and that is the best we can do. For example, let f(x) = 1/x 2 and g(x) = 1/x
for x> 1. Then clearly f(x) < g(x) for all x> 1, but

lim f(x) = 0 = lim g(x).

X-+ 00 X-+OO

Thus the limits are equal despite the strictly inequality f(x) < g(x) for x > 1.
In any particular situation the reader now has two different ways of
attacking the solution of a problem concerning limits of functions. It is
possible to use I> and X, and tackle the problem directly. Alternatively, the
link between sequences and limits of functions can be exploited and the
problem resolved using sequences. Some problems are amenable to one
approach and some to the other. The reader should be prepared to use either
method, and will learn by experience which gives the neater solution to
certain types of problems. Some examples are given below.

Examples 4.1.5

1. Let X be the set of all positive real numbers, i.e. X = {XEIR: x > O}. Let
f: X --+ IR be defined in the following way:
if XEX and x is irrational,
if XEX and x is rational with x = p/q,
where p, q are positive integers with no
common factor greater than 1.
(The condition that p, q have no common factor exceeding 1 means that
the fraction p/q is in its lowest terms, i.e. common factors of numerator
and denominator have been cancelled out.)

This example looks formidably complicated, but there is no need to be intimidated.

It is really quite easy if we use sequences.
From Theorems 4.1.5 to 4.1.7 we notice that if f(x) tends to any limit (finite or
infinite) as x -> 00, then f(x.) must also tend to this limit as n -> 00 for all suitable
sequences. Thus if we find two sequences (x.)~ 1 and (y.)~ 1 such that x. -+ 00 as
n -> 00 and y. -> 00 as n -> 00 and f(x.) and f(y.) tend to two different limits as
n -> 00, then f( x) cannot tend to any limit as x -> 00. In this case we choose such
sequences, one consisting entirely of suitable rational numbers and the other
consisting of irrational numbers. Let us use
2n-l
x·=-2- (n= 1,2,3,4, ... ),

123
GUIDE TO ANALYSIS

Yn=nJ2 (n= 1,2,3,4, ... ).

Clearly Xn -+ 00 as n -+ 00, Yn -+ 00 as n -+ 00 and Xn > 0, Yn > O. Now (x n):;' 1 is the
sequence t,t,!,!, ... , and f(xn)=!. By definition f(y.)=O. Hence f(xn)-+! as
n-+oo, andf(Yn)-+O as n-+oo. The fact that

n .... oo n-+oo

guarantees that f(x) does not tend to any limit as x -+ 00.

The above example illustrates one type of problem for which the sequential
criterion is particularly powerful. It can be used to great effect to prove that
a functionf does not have a limit. For it is then sufficient to find two sequences
(x.) and (y.) (satisfying appropriate conditions) such that f(x.} and f(y.} do
not have the same limit.
On the other hand if we want to prove that a function does possess a limit,
then the e, X criterion is frequently more useful, since the sequential criterion
requires a certain condition to be proved for aU sequences.
Of course, it would be possible to formulate a solution for Example 1 using
e and X, but the reader will find it difficult to express the explanation clearly
and convincingly using this approach.
x 2 -1
2. Show that ---+1 as x-+ 00.
x2 + 1

Solution For x > 0

X2 -1 1-1/x2
(1)
x 2 + 1 = 1 + 1/x2 •
As x -+ 00, 1/x2 -+ 0 and so by the algebra of limits
1 - 1/x2 -+ 1 - 0 = 1
1 + 1/x2 -+ 1 + 0 = 1
as x-+ 00. Hence, by equation (1),

3. For all x> 0

l-l/x x2 - x 2 + x sin x x2 +x 1 + llx
.
X
--~ ~--=
1 + l/x 2 x2 1+ x2 + 1 x2 1 + 1 + l1x 2
As x-+oo,
l-l/x 1-0
--,---=-+-- = 1
1 + l/x 2 1+0 '

124
 FUNCTIONS

1 + l/x -+ 1 + 0 = 1
1 + l/x2 1 + 0
by the algebra of limits. Using the sandwich rule we see that
x 2 + x sin x -+ 1
x2 + 1
asx-+oo.
4. For all x> 1,
x 3 +xsinx x 3 -x
- - = - - - >- - - - x - -
(X2-1)
x 2 + 1 ,.... x 2 + 1 - x2+ 1 .

x 2 -1 2 2 3
Now --=1--->1--=-
x +1
2 x +1
2 5 5
for all x > 2. Thus for x > 2
x 3 +xsinx 3
x2+1 >5 x .
Now 3x/5 -+ 00 asx-+ 00 and, by Theorem 4.1.1, (x 3 + x sin x)/(x 2 + 1)-+ 00
as x-+ 00.

EXERCISES 4.1.4
1 For each of the following functions J, decide whether f( x) tends to a limit
as x -+ 00. When the limit exists find it.

(b) f(x)
IX
= 1- VA;
I+Jx
3 .
(d) f(x)=x 2smlx;
x +

(e) f(x)
x3 sin 2
= x2 + 1 ;
x
(f) f(x) = x
x
:m+ t;
2 . 2

xsinx
(g) f(x) = x2 + 1 ; (h) f(x) = sinh x;

(i) f(x) = P+T - JXCl.

In each case justify your answer.
2 Supposef(x) and g(x) are defined for all x > a, where a is some given real
number. Suppose also thatf(x)-+ 00 as x-+ 00 and Jg(x)J ~ k for all x > a,

125
GUIDE TO ANALYSIS

wherekis some positive real number. Showthatf(x) + g(x)-+ 00 asx-+ 00.

Give examples to show that (a) f(x)g(x) need not tend to any limit as
x -+ 00, (b) if f(x)g(x) does tend to a limit as x -+ 00, then this limit can be
infinite or it can be finite.
3 Show that if f(x)-+l and g(x)-+oo as x-+oo, then f(x) + g(x)-+ 00 as
x-+ 00.
4 Give examples of functions f and g to illustrate each of the following
possibilities.
(a) f(x)-+ 00 and g(x)-+ - 00 as x -+ 00, but f(x) + g(x) does not tend to
any limit as x -+ 00.
(b) f(x)-+ 00 and g(x)-+ - 00 as x -+ 00, butf(x) + g(x)-+ - 00 as x-+ 00.
(c) f(x)-+oo and g(x)-+ -00 as x-+ -00, but f(x) + g(x)-+l as x-+oo,
where I is some given real number.

The beginning of this section was devoted to limits as x -+ 00 and the material
was presented in a very leisurely manner. Naturally, a corresponding theory
exists for limits as x -+ - 00. By now, the reader should have sufficient
knowledge and experience to be able to envisage how to frame the definitions
and to decide which results will be useful in actually finding limits. The
following definitions and theorems will, therefore, be given without further
explanations and proofs. They are all, in fact, rather obvious analogues of
the ones we have already met.

DEFINITION 4.1.4 Suppose that f(x) is defined for all x < a, where a is
some given real number. (Note: The phrasef(x) is defined for all x < a means
that the domain of definition of f contains the interval ( - 00, a).) We say that
f(x) -+ 00 as x -+ - 00 if given any A> 0 (however large) there exists a
corresponding Xl (with Xl < a) such that f(x) > A for all x < Xl'

DEFINITION 4.1.5 Suppose that f(x) is defined for all x < a, where a is
some given real number. We say that f(x) -+ I as x -+ - 00 if given any 8> 0
(however small) there exists a corresponding Xl (with Xl < a) such that
If(x)-ll < 8

DEFINITION 4.1.6 Suppose that f(x) is defined for all x < a, where a is
some given real number. We say that f(x) -+ - 00 as x -+ - 00 if given A > 0

126
 FUNCTIONS

(however 1arge) there exists a corresponding Xl (with Xl < a) such that

f(x) < -A for all x<X 1 •

As in the case oflimits as x --+ 00, we find that the limit (if it exists) is unique.

THEOREM 4.1.10 Suppose that f(x) is defined for all x < a, where a is
some given real number. Then f(x) cannot tend to more than one limit as
x--+-oo.

Theorems 4.1.2 to 4.1.4, which contain practical tools for establishing the
value of limits, have their obvious analogues.

THEOREM 4.1.11 Suppose thatf(x) and g(x) are defined for all x < a and
satisfy the inequality f(x) ~ g(x) for all x < a, where a is some given real
number. If g(x)--+oo as x--+ -00, then f(x)--+ 00 as x--+ -00.

THEOREM 4.1.12 (Sandwich Rule) Suppose that f(x), g(x), h(x) are all
defined for all x < a and satisfy the inequality
g(x) ~ f(x) ~ h(x)
for all x < a, where a is some given real number. If g( x) --+ I as x --+ - 00 and
h(x)--+l as x--+ -00, thenf(x)--+l as x-+ -00.

THEOREM 4.1.13 Suppose that f(x), g(x) are defined for all x < a and
satisfy the inequality
f(x) ~ g(x)
for all x < a, where a is some given real number. If g(x) --+ - 00 as x --+ - 00
thenf(x)--+ -00 as x--+ -00.

In the case of limits as x --+ - 00, it is again possible to write down results
demonstrating the link between sequences and limits of functions. As before,
there are rules for limits of sums, products etc.

THEOREM 4.1.14 Suppose that f(x), g(x) are defined for all x < a, where
a is some given real number. If f(x)--+l, g(x)--+m as x--+ -00, then
(a) f(x) + g(x) --+ I + m as x --+ - 00;
(b) f(x)g(x)--+lm as x--+ -00.

127
 GUIDE TO ANALYSIS

Moreover, if f satisfies the additional conditions that g(x) =1= 0 for x < a and
m =1= 0, then
f(x) I
(c) - - - + - as x ..... -00.
g(x) m

THEOREM 4.1.15 Suppose that f(x), g(x) are defined for all x < a, and
satisfy the inequality
f(x) ~g(x) (x < a),
where a is some given real number. If f(x)-+l and g(x)-+m as x-+ - 00, then
l~m.

Let us now return to one of the functions sketched at the beginning of the
section. In Fig. 4.3 we have the graph of y = l/x. It is easy to see that l/x -+ 0
as x ..... 00 and also that l/x -+ 0 as x -+ - 00. But what happens if x -+ O?
The graph of y = l/x is in two separate pieces and if we start with a positive
value for x and let x approach zero on the right-hand portion (i.e. on the
piece in the first quadrant), then it looks as if y -+ 00, i.e. if we let x approach
zero in such a way that it only takes positive values, then it seems that y -+ 00.
On the other hand if we let x approach zero on the left-hand piece of the
graph (i.e. the piece in the third quadrant for which x < 0), then it looks as
if y -+ - 00. So in this case it looks as if we have a reasonably sensible notation
of a limit provided x approaches zero from just one side. If we want to use
symbols to express the fact that x is approaching zero from the right (i.e.
only through values of x greater than 0) then we write either limx-+o+ or we
use the phrase l/x -+ 00 as x ..... 0 +. This limit is called a one-sided limit
because we only approach zero from the right-hand side. In general, if we
are interested in a one-sided limit of f(x) as x approaches a point a from the
right, then we use limx .... a+f(x) or we discuss what happens tof(x) asx-+a+.
The phrase x -+ a + expresses the fact that we are only considering what
happens to f(x) as x approaches a through values which exceed a.
Naturally, we also have one-sided limits from the left. Suppose we wish to
, investigate what happens to f(x) as x approaches a point a through values
less than a. For such limits we use either lim x-+ a _ or we discuss what happens
to f(x) as x-+a-, the symbols a- denoting the fact that we are only
considering what happens to f(x) as we approach a through values less than
a (i.e. we approach a from the left).
One-sided limits can be useful when there is a nasty break in the graph of
a function or when the function is defined only at one side of the point in
question. For example l/x has a nasty break at 0, but we can write l/x -+ 00
as x -+ 0 + and l/x -+ - 00 as x -+ 0 -. The natural logarithm log x, however,
is defined only for x > 0 and the reader can check by drawing the graph that
it looks as if log x -+ - 00 as x -+ 0 + .

128
 FUNCTIONS

In both the examples above, we notice that the function was not defined
at the point itself, for neither l/x nor log x have a value when x = O. All that
we required in taking the one-sided limits, limx_o+ and limx-o_, is that the
function is defined at all points sufficiently close to 0, either on the right of
o or on the left of 0, whichever is appropriate. In taking limits we totally
ignore what happens at the point; we only make statements about what
happens as x approaches the point. Thus, for example, when we define
limx_a+f(x) we will need to stipulate that f(x) is defined for a < x < a + R,
where R is some positive real number, i.e. we will need to know that the
domain of definition of f contains some interval (a, a + R) for some R > O.
This guarantees that f(x) has a value when x is sufficiently close to a, but
on the right of a. Once the one-sided limits have been defined, they can be
used to define the two-sided limit. The same theorems are needed for the
one-sided limits and the two-sided limits. We will therefore write down all
the definitions and then simply state results like the sandwich rule and the
algebra of limits once. The reader can make appropriate minor adjustments
to adapt them for use with one-sided or two-sided limits. The definitions
follow a similar pattern to that of those we have already met, with one
significant change.
To pinpoint the exact place at which the alteration must be made, let us
return to the example f(x) = l/x (x ¥- 0). We have already suggested that
l/x -+ 00 as x -+ 0 + . In fact, given any positive real number A (however large)
we notice that f(x) > A for all x such that 0 < x < l1A, i.e. f(x) > A for all
positive values of x sufficiently close to zero. If we write f> = 11A, then f> > 0
and in this example f(x) > A for all x such that 0 < x < f> (see Fig. 4.9). Thus
the conditions we need are similar to those in Definition 4.1.1 except that
we now have 0 < x < f> instead of x> X l ' For we are now interested in values
of x close to 0, whereas Definition 4.1.1 was concerned with all sufficiently
large values of x.
In general, if we are considering what happens to f(x) as x -+ a +, then we
will be interested in determining whether certain conditions are satisfied for
all x sufficiently close to a, but greater than a, i.e. we will need to know
whether the criteria are satisfied for all x such that a < x < a + f> where f> is
some positive real number. We are now in a position to formulate the
definitions.

DEFINITION 4.1.7 Suppose that f(x) is defined for all x such that
a < x < a + R, where R is some given positive real number. Then we say that
f(x~-+ 00 as x -+ a +, if given any A> 0 (however large) there is a corre-
spondingf> > o(with f> ~ R)such thatf(x) > AforaUxsuch that a < x < a + f>.

DEFINITION 4.1.8 Supposef(x) is defined for all x such that a - R < x < a,
where R is some given positive real number. Then we say that f(x) -+ 00 as

129
GUIDE TO ANALYSIS

y=A

o~____======~~============
6
___ x
Fig. 4.9

x -+ a -, if given any A > 0 (however large) there is a corresponding f> > 0

(with f> :::;; R) such that j(x) > A for all x such that a - f> < x < a.

DEFINITION 4.1.9 Supposej(x) is defined for all x such that a < x < a + R,
where R is some given positive real number. Then we say that j(x)-+I as
x -+ a + if given any 6 > 0 (however small) there is a corresponding f> > 0
(with f> ~ R) such that
Ij(x) -II < 6

for all x such that a < x < a + f>.

DEFINITION 4.1.10 Suppose j(x) is defined for all x such that a - R < x < a,
where R is some given positive real number. Then we say that f(x)-+I as
x -+ a - if given any 6 > 0 (however small) there is a corresponding f> > 0
(with f> :::;; R) such that
If(x) -II < 6

for all x such that a - f> < x < a.

130
 FUNCTIONS

DEFINITION 4.1.11 Supposef(x) is defined for all x such that a < x < a + R,
where R is some given positive real number. Then we say that f(x) -+ - 00
as x -+ a + if, given any A > 0 (however large), there is a corresponding ~ > 0
(with ~ ~ R) such that f(x) < -A for aU x such that a < x < a +~.

DEFINITION 4.1.12 Suppose f(x) is defined for all x such that a - R < x < a,
where R is some given positive real number. Then we say that f(x) -+ - 00
as x -+ a - if, given any A > 0 (however large), there is a corresponding ~ > 0
(with ~ ~ R) such that f(x) < -A for aU x such that a - ~ < x < a.

The two-sided limit, limx--+af(x), exists iflimx--+a+f(x) and limx-+a_f(x) both

exist and are the same. This common limit is called limx--+af(x). By combining
the previous definitions, it is easy to formulate definitions for the two-sided
limits. One case, in particular, merits mention as it turns out to be extremely
important in practice. This special case is the one in which limx-+a+f(x) and
limx -+ a _ f(x) are finite and equal. The relevant definition is given below. Since
limx--+a+f(x) and limx--+a_f(x) both exist,J(x) must be defined for all x such
that a < x < a + R and also for all x such that a - R < x < a. The first
condition gives 0 < x - a < R and the second condition is - R < x - a < O.
These two can be combined to give the inequality 0 < Ix - al < R.

DEFINITION 4.1.13 Suppose f(x) is defined for all x such that

0< Ix - al < R, where R is some given positive real number. Then we say
that f(x)-+l as x-+a, if, given any 8>0 (however small), there is a
corresponding ~ > 0 (with ~ ~ R) such that
If(x)-ll <8
for all x such that 0 < Ix - al <~.

(Note: This definition is couched in the same terms as Heine's definition

which dates back to over a century ago. The one superficial difference is that
Heine used the Greek letter '1 in place of the letter ~ we use today.)

We notice from the inequality 0 < Ix - al < ~ that the point x = a is

specifically excluded. Thus the statementf(x)-+l as x-+a tells us how f(x)
behaves as x approaches a. However, it gives us absolutely no information
about what happens at a itself. As before, it can be proved that a function
cannot have more than one limit.
Two results are of great practical importance as they are frequently used
in determining limits. They are analogues of results which we have met
already.

THEOREM 4.1.16 (Sandwich Rule) Suppose f(x), g(x), h(x) are defined
and satisfy the inequality

131
GUIDE TO ANALYSIS

g(x) ~ f(x) ~ h(x)

for all x such that 0 < Ix - al < R, where R is some given positive real number.
If g(x) -+ 1 as x -+ a and h(x) -+ 1 as x -+ a, then f(x) -+ I as x -+ a.

THEOREM 4.1.17 (Algebra of Limits) Suppose f(x), g(x) are defined for
all x such that 0 < Ix - al < R, where R is some given positive real number.
If f(x)-+l as x-+a and g(x)-+m as x-+a, then
(a) f(x)+g(x)-+l+m as x-+a;
(b) f(x)g(x)-+lm as x-+a.

If in addition g(x) satisfies the condition g(x) =F 0 for all x such that
o< Ix - a I < R and if m =F 0, then
f(x) I
(c) ---+- as x-+a.
g(x) m

THEOREM 4.1.18 Suppose f(x), g(x) are defined for all x such that
0< Ix - al < R and satisfy the inequality
f(x) ~ g(x) (O<lx-al<R),
where R is some given positive real number. If f(x)-+l and g(x)-+m as x-+a,
then
l~m.
Proofs of these results can easily be constructed along the same lines as
those given for earlier theorems of the same form. These results are not
restricted to the case oflimits as x -+ a; they can be used for limits as x -+ a +
and limits as x -+ a - , provided that the wording is very slightly modified (in
the obvious way) to make it appropriate for the case being considered.
As the reader may expect, Theorems 4.1.1 and 4.1.2 have obvious analogues
for limits as x -+ a, etc. However, these analogues for infinite limits are less
frequently used than the results for finite limits.
We have already noticed that the link between sequences and limits of
functions can be of practical use in some circumstances. In the case in which
the limit as x-+a is being considered we have the following statement.

THEOREM 4.1.19 Supposef(x) is defined for all x such that 0 < Ix - al < R,
where R is some given positive real number. Then f(x)-+l as x-+a if and
only if f(xn)-+l as n-+ 00 for all sequences (X n);:O=l such that 0 < IX n - al < R
(for all n ~ 1) and xn-+a as n-+ 00.

Let us now see if we can put these theorems to good use by working some
examples.

132
 FUNCTIONS

Examples 4.1.6

1. The function f: IR ..... IR is defined by

x (x < 0),
(x = 0),
f(x) = 1 +x (0<x<1),
3 (x = 1),
2X2 (x> 1).
Decide whether the following limits exist and find their value when they
do exist:

(a) lim f(x); (b) lim f(x); (c) limf(x); (d) limf(x}.
x-+o- x-+O+ x-->O x-->1
Solution

(a) Ifx<O,f(x)=x. As x-+O-,f(x)-+O. Hence

lim f(x)=O.
x-+o-
(b) If 0 < x < l,f(x) = 1 + x. We see that 1 + x -+ 1 as x -+0+ (using the algebra
of limits). Hence

lim f(x) = 1.
x-->O+
(c) Since
lim f(x) #- lim f(x),
x-->O- x-->O+
we see that limx-->of(x) does not exist.
(d) IfO<x< l,f(x) = 1 +x. We see thatf(x)-+2 as x-+l-.

i.e. lim f(x) = 2.

x-+l-

If 1 < x, f(x) = 2x 2 • We see that f(x)-+2 as x-+ 1 +.

Hence lim f(x) = 2 = lim f(x).

x--+l+ x-+l-

Thus limx-->d(x) = 2.
The fact that f(l) = 3 has no bearing at all on whether limx--> d(x) exists or not.
In deciding whether the limit exists or not we are interested in what happens as
x approaches 1. We ignore what happens at x = 1.

2. Let f(x} = xsinx

x2 + 1
for all xeR For all x,
Ixl
If(x)1 ~ x2 + 1'

133
GUIDE TO ANALYSIS

-Ixl Ixl
i.e. X2 + 1 ::::;f(x)::::; x 2 + 1"
Using the algebra of limits, we see that
-Ixl Ixl
-2---+0, -2---+0 as x-+O.
x +1 x +1
By the sandwich rule, therefore,
f(x)-+O as x-+O.
In this example a slightly modified version of the sandwich rule can be
used. Suppose that
If(x)1 ::::; Ig(x)1 (1)
for all x such that 0 < Ix - al < R, where R is some positive real number.
Suppose also that g(x)-+O as x-+a. Then we can deduce from inequality
(1) thatf(x)-+O as x-+a. For, by definition, given any 8>0, there is a
corresponding b > 0 (with b::::; R) such that Ig(x)1 < 8 for all x such that
0< Ix - al < b because g(x)-+O as x-+a. Hence, for 0 < Ix - al < b,
If(x)1 ::::; Ig(x)1 < 8
and, therefore, f(x) -+ 0 as x -+ a.
3. Let f be defined on IR\ {O} by
f(x) = x sin(±) (x #0).
For x # 0 we have
If(x)1 = Ix sin(±)1 ::::; lxi,
and thereforef(x)-+O as x-+O.
The graph of Y = x sin(l/x) is shown as Fig. 4.10.
4. Let f be defined on IR\ {O} by
f(x) = sin{1/x) (x # 0).

Then, using sequences, we can show that f(x) does not tend to any limit
as x-+O. We prove this by choosing two sequences (X n):'=l and (Yn):'=l so
that xn-+O, Yn-+O as n-+oo, but

n~oo n-+oo

To do this choose
xn = l/nn (n = 1,2,3,4, ... )
2
(n= 1,2,3,4, ... ).

134
 FUNCTIONS

,, y
1.4 '" '"
',y=x
,, ,,/y=x

, 1.2
y= 1 ",'" "" ""
-------~..... ;:_------ ------1.0 ----------------/--------
,, ""
~ 0.8 /.:: '" '"

-1.4 -1.2 -1.0 -0.8 -0.6 -0.4 \~..2 0.6 0.8 1.0 1.2 1.4 x

", , ,
T-O.2
","" '" "" -0.4
'"
,,
",,""'" -0.6 ,,
// '" -0.8 ""
"" ,,
",,"""" -1.0 ,,
'" ,,
",'" -12 ,,
"" ,,
"" '" -1.4 ,,
""
Fig. 4.10

Clearly Xn -+ 0 and Yn -+ 0 as n -+ 00.

Moreover, f( xn) = sin( l/xn) = sin nn = 0
f(Yn) = sin(1/Yn) = sin[(4n + 1)n/2] = 1.
Hence f(xn) -+ 0 as n -+ 00 and f(Yn) -+ 1 as n -+ ro. Since the two limits are
different, f(x) does not tend to any limit as x -+0. Since xn > 0 and Yn > 0
for all n, we see that f(x) does not tend to any limit as x -+ 0 +. The graph
is shown as Fig. 4.11.
Our final example in this section concerns a limit which has rather important
implications, as we will see later.
S. Let C be a circle centre 0 and radius 1. Suppose the arc AB of the circle
C subtends an angle x radians at 0, where 0 < x < n/2. Suppose also that
the tangent to the circle C at the point A cuts OB extended at Q, and that
P is the point on 0 B such that AP is perpendicular to 0 B. Using elementary
trigonometry we see that the length of AP is sin x and the length of AQ
is tanx. Now the area of a triangle is t base x height. Since OB is oflength
1, the triangle OBA has area tsinx. The sector OAB of the circle has
area tx (since x is measured in radians) and the triangle OAQ has area
ttanx. Using Fig. 4.12 we can see that

135
GUIDE TO ANALYSIS

1.2

-1.4 -1.2 -1.0 -0.8 -0.6 1.2 1.4 x

-1.2

Fig. 4.11

Fig. 4.12

area of ~OBA < area of sector DBA < area ~OQA,

i.e. tsinx<tx<ttanx.
Thus, for 0 < x < n12, sin x < x < tanx. (1)

136
 FUNCTIONS

Now if 0 < x < n12, then 0 < xl2 < nl4 < nl2 and, therefore,
0< sin(xI2) < xl2
by (1). Hence for 0 < x < n12,
cos x = 1- 2 sin 2 (xI2) > 1 - 2(x12f = 1- !x 2 , (2)
and therefore, from (1),
1 2 sin x
1-2x <cosx <--< 1 (3)
x
for 0 < x < n12. Since cos( -x) = cosx and sin( -x) = -sinx we see that
(3) is also true if -n12 < x < O. Thus if 0 < Ixl < n12, then
1 2 sin x
1- 2 x <cosx <--< 1.
x
Using the sandwich rule we see that
sin x
---+1 as x-+O
x
and also cos x -+ 1 as x -+ O. The limit for (sin x)lx will be used later when
we come to the question of differentiation of trigonometric functions.
Incidently, this limit also helps to clear up a point which arose in
connection with Example 3. The graph appears to indicate that x(sin 1/x) -+ 1
as x -+ 00, which we can easily verify. For if x > 0, then,

x
As x-+oo, l/x-+O and therefore
sin(1/x) -+ 1
(1/x) .

i.e. xsin(l/x)-+ 1 as x-+ 00.

Similarly, xsin(1/x)-+ 1 as x-+ - 00.

EXERCISES 4.1.5

1 For each of the following functions f and points a, decide whether f(x)
tends to a limit as x -+ a. When the limit exists find it.
2X2 +1
(a) f(x) = 3x 2 + 3x + l' a = 0;

137
GUIDE TO ANALYSIS

2x 2 + 1
(b) f(x) = 3x2 + 3x + l' a= 1;

(c) f(x) = (Xl + l)sinx (x #0), a=O;

X

(d) f(x) = xsin(l/x) (x #0), a = 0;

x2 + 1

(e) f(x) = sin[ X + (~) ] (x # 0), a =0;

1
(f) f(x) = r:;2;11 ~1 (x#I),a=O;
yX +~ -yX -~
SInQ(X
(g) f ( x ) = - (x # 0), a = 0, where Q( is any real number;
x
(h) f(x) = sin x (x#O), a=O;
X2

(i) f {x (x rational) a=O;

(x) = x2 (x irrational),
(j) f )= {O (x rational) a = 1.
(x 1 (irrational) ,
In each case give reasons to justify your answer.

I
2 Let f: IR -+ IR be defined by

sinx (x < 0),

(x = 0),
f(x) = ~2 (O<x<I),
2x-l (1 ~ x).

Decide whether limx-+of(x) and limx-+d(x) exist. Find any limit which
exists.

3 Give examples of functions f and g to illustrate each of the following

possibilities:
(a) f(x)-+O as x-+ 1, g(x)-+O as x-+ 1, butf(x)/g(x) does not tend to any
limit as x -+ 1.
(b) f(x)-+O as x-+l, g(x)-+O as x-+l and f(x)/g(x) tends to 00 as
x-+ 1.
(c) f(x)-+O as x-+ 1, g(x)-+O as x-+ 1 and f(x)/g(x) tends to 3 as
x-+1.

4 Show that for all real numbers x

Isinxl ~ IxI-

138
 FUNCTIONS

By using the formula

. (x+a)
cosx-cosa=2sm . (a-x)
-2- sm -2- ,

prove that Icosx-cosal ~Ix-al.

Deduce that cos x -+ cos a as x -+ a.

Use the same method to show that sinx-+sina as x-+a.
(Warning: Remember that x is measured in radians).

4.2 CONTINUITY
In the sixth form, students frequently think of a continuous function as one
whose graph can be sketched without taking the pencil oft'the paper. Their
idea of a continuous function, therefore, is one which has no breaks. Suppose
we examine this notion in a little more detail. A function f does not have a
break at a point a if f(x)-+ f(a) as x-+a. This is precisely the condition we
need for continuity. Since limits are defined in terms of symbols, the
requirementf(x)-+ f(a) as x-+a does not depend on existence ofa graphical
representation of f

DEFINITION 4.2.1 Suppose that f(x) is defined for all x such that
a - R < x < a + R, where R is some given positive real number. Then we say
thatf is continuous at a if f(x) -+ f(a) as x -+ a, i.e. if given any Il > 0, there is
a corresponding l> > 0 (with l> ~ R) such that
If(x) - f(a) I < Il
for all x such that Ix - al < l>.

We notice that our definiton of continuity of f at a point a is subject to the

condition thatf(x) is defined for all x such that a - R < x < a + R. This means
that the domain of definition of the function f contains some interval of the
form (a - R, a + R) (R > 0). It will be tactily assumed that such a condition
is satisfied whenever the phrase 'f is continuous at a' is used. (Note: This
convention is not used universally. Frequently the restriction on the type of
point is lifted and the 8, l> condition in the definition is replaced by the
requirement that If(x) - f(a)1 < 8 for all x in the domain of definition of f
for which Ix - al < l>.) Extensions of these ideas to continuity at other types
of points are possible, but we will not use them.
In analysis the concept of continuity on an interval is frequently needed.

DEFINITION 4.2.2 Let a, b be two real numbers with a < b and suppose
that f(x) is defined for all x such that a < x < b. We say that f is continuous
on the open interval (a, b) iff is continuous at all points c for which a < c < b.

139
GUIDE TO ANALYSIS

DEFINITION 4.2.3 Let a, b be two real numbers with a < b and suppose
that f(x) is defined for all x such that a ~ x ~ b. We say that fis continuous
on the closed interval [a, b] iff is continuous at all points c such that a < c < b
and limx-+a+f(x) = f(a), limx--+b-f(x) = f(b).
The use of the one-sided limits at the end points a, b ensures that we do
not need to consider what happens outside the interval [a, b]. Instead of
using the phrase 'f is continuous on the closed interval [a, b]', we sometimes
just write f: [a, b] -+ IR is continuous.

As the reader might expect, there are results concerning sums, products,
quotients, etc., of continuous functions. In preparation for the proof of the
result for quotients we establish the following result.

THEOREM 4.2.1 Suppose that f(x) is defined for all x such that
a - R < x < a + R, for some positive real number R. If f is continuous at a
and f(a) =F 0, then there is some 15 > 0, (with 15 ~ R) such that f(x) =F 0 for
all x such that a -15 < x < a + 15.

Proof Since f(a) =F 0, it follows that

If(a)1 >0.
Now f is continuous at a and so using the definition of continuity with the
special case B = If(a)l, we see that there is a 15 > 0 (with 15 ~ R) such that
If(x) - f(a) I < If(a)1 (1)
for all x such that a -15 < x < a + 15. From (1) we see that f(x) =F 0 for
a -15 < x < a + o.

We can now deal with sums, products and quotients of continuous functions.

THEOREM 4.2.2 Suppose that f(x), g(x) are defined for all x such that
a- R < x < a + R, for some positive real number R. If J, g are continuous at
a, then
(a) f + g is continuous at a;
(b) fg is continuous at a;
(c) fig is continuous at a, provided g(a) =F O.

Proof SinceJ,g are continuous at a,f(x)-+ f(a) andg(x)-+g(a) as x-+a. By

Theorem4.1.17,J(x) + g(x)-+ f(a) + g(a) as x-+a andf(x)g(x)-+ f(a)g(a) as
x-+a. This proves (a) and (b).
Since g(a) =F 0, Theorem 4.2.1 ensures that there is some 15 > 0 (with 15 ~ R)

140
 FUNCTIONS

such that g(x) of. 0 for all x such that a - J < x < a + J. From Theorem 4.1.17
it now follows that
f(x) f(a)
---+--
g(x) g(a)

as x -+ a. Hence fig is continuous at a.

Similar results follow for functions continuous on an interval. For example

we have the following:

THEOREM 4.2.3 Let f: [a,b] -+ IR be continuous and let g: [a,b] -+ IR be

continuous. Then
(a) f + 9 is continuous on [a,b],
(b) fg is continuous on [a,b],
(c) fig is continuous at all points of [a, b] at which 9 is not zero.

Theorems about continuous functions are hardly very productive without a

stock of continuous functions. Do we have such a collection available? From
the definition (with J = e) it is clear that the identity function f given by
f(x) = x for all xEIR is continuous at all points of R Using the algebra of
limits, it follows that all positive integral powers of x are continuous at all
points of IR, and therefore every polynomial is continuous at all points of R
Again, an appeal to the algebra of limits assures us that the quotient of two
polynomials (i.e. a rational function) is continuous at all points of IR at which
the denominator is not zero. The familiar trigonometric functions sine and
cosine, the hyperbolic functions sinh and cosh, and the exponential function
provide us with a further stock of functions which are continuous at every
point of IR. Since these functions are usually formally defined in terms of
power series, the reader is asked to accept on trust the statement that they
are continuous. A rigorous detailed proof requires properties of power series
and is beyond the scope of this volume. We can further enlarge our stock of
continuous functions by using the theorem on inverse functions which appears
later in this section. With the help of this result we can conclude that log x
is continuous at all points x for which x > 0 and we can investigate where
nth roots are continuous.
In examples we may need more than just sums, products and quotients of
continuous functions. For example these rules, on their own, are not really
adequate when we come to handle objects like sin(e'') which is often referred
to as a function of a function or a composite function-to use mathematical
terminology. Let us recall how composite functions are defined. Suppose we
have two functions g:X -+ Y andf: Y-+Z (see Fig. 4.13). If we take XEX,
then g(X)E Y and then, using f, we see that f(g(X»EZ. The two together,
therefore, associate with each XEX a unique element ZEZ, and so they define

141
GUIDE TO ANALYSIS

g f

fog

Fig. 4.13

a function from X into Z. This function is denoted by Jog. It has the property
that, for each xeX, the image (f og)(x) is the element J(g(x»eZ.

Examples 4.2.1

1. Let g: IR -+ IR and J: IR -+ IR be defined by

g(x) = eX (all xelR),
J(x) = sin x (all xelR).
Then (f 0 g): IR -+ IR and (g 0 f): IR -+ IR are both defined and
(f og)(x) = J(g(x» = sin(eX)
(gof)(x) = g(f(x» = esinx
for all xeR
2. Let Y be the set of all positive real numbers and suppose that g: IR -+ Y
and J: Y -+ IR are defined by
g(x) = x 2 + 1 (all xelR),
J(y) = log y (all ye Y, i.e. all y > 0)
Then (Jog): IR -+ IR is defined and
(J og)(x) = J(g(x» = log(x 2 + 1)
for all xeR The composite function (gof): Y -+ Y is also defined and
(go f)(y) = g(f(y» = (log y)2 + 1
for all ye Y, i.e. for all y > o.

142
 FUNCTIONS

In the present context the question which exercises us is whether the continuity
of a composite function can be deduced from the continuity of the separate
functions. Let us take two functions g: X --+ Y, andf: Y --+Z where X, Y,Z are
all sets of real numbers. Suppose also that all the following conditions are
satisfied:
(a) X contains an interval of the form (a - R, a + R) for some positive
real number R;
(b) g(a) = b;
(c) g is continuous at a;
(d) Y contains an interval of the form (b - s, b + s) for some positive real
number s;
(e) f is continuous at b.
Now let 6> O. Since f is continuous at b, there is some 15 1 > 0 (with 15 1 ~ s)
such that
If(y) - f(b)1 < 6 (1)
for all y such that b-t5 1 <y<b+t5 1 •
Since g is continuous at a, there is some 15 > 0 (with 15 ~ R) such that
Ig(x) - g(a)1 < 15 1 (2)
for all x such that a - 15 < x < a + 15. Relation (2) can be written as
b - 15 1 < g(x) < b + 15 1 (3)
for all x such that a - 15 < x < a + 15, because g( a) = b. Hence, if
a - 15 < x < a + 15, then

If(g(x)) - f(b)1 < 6,

i.e. If(g(x)) - f(g(a)) I < 6.

This means that given any 6 > 0, there exists a corresponding 15 > 0 such that
If(g(x)) - f(g(a)) I < 6
for all x such that a - 15 < x < a + 15. Hence fog is continuous at a.
We now record this result in the following theorem.

THEOREM 4.2.4 Let f, g be two real valued functions such that

(a) g is continuous at a;
(b) g(a) = b,
(c) f is continuous at b.
Then the composite function (f 0 g) given by
(f og)(x) = f(g(x))
is continuous at a.

143
GUIDE TO ANALYSIS

Examples 4.2.2

1. Let f: ~ -+ ~, g: ~ -+ ~ be given by
g(x) = e% (all real x)
f(x) = sin x (all real x).
Thenf is continuous at all points of ~ and 9 is continuous at all points of~.
Thus the composite functions fog and 9 of are both continuous at all
points of ~, i.e. sin(e%) and esiu are both continuous at all points of ~.
2. Let h: ~ -+ ~ be defined by
h(x) = {sin(1/X) x =F 0,
o x=O.
Write f(x) = sin x for all real x and let g(x) = l/x (x =F 0). Then 9 is
continuous at all points other than the origin, i.e. 9 is continuous at all
points of ~\{O}. Moreover, f is continuous at all points of ~. The
composite function fog is therefore continuous at all points of ~\{O},
i.e. h is continuous at all points of ~\ {O}. Now h(x) does not tend to any
limit as x -+0 (see Examples 4.1.6). Hence h(x) ~ h(O) as x -+0 and h is not
continuous at the origin.
3. Let h: ~ -+ ~ be defined by

h(x) = {~sin(1/X) (x =F 0),

(x = 0).
We see from the previous example that sin(1/x) is continuous at all points
of ~\{O}. Using the result for the product of two continuous functions,
we see that xsin(l/x) is continuous at all points of ~\{O}, i.e. h is
continuous at all points of ~\{O}. But what happens at the origin? Is h
continuous there? To decide what happens at the origin we consider
Ih(x) - h(O)I. If x =F 0 then
Ih(x) - h(O)1 = Ixsin(1/x) - 01
= Ix sin(l/x) I ~ Ixl.
Hence, Ih(x)-h(O)I-+O as x-+O, i.e. h(x)-+h(O) as x-+O and h is also
continuous at the origin. Thus the function h is continuous at all points
ofR

EXERCISES 4.2.1
1 Use the results of questions in Exercises 4.1.5 to show that the sine and
cosine are continuous at all points of ~.

144
 FUNCTIONS

2 Decide whether each of the following functions f is continuous at the

point a indicated.
(a) f(x) = [x], a = 0
where [x] is the integer part of x, i.e. [x] is the largest integer which
does not exceed x.
X2 (x < 0)
(b) f(x) =
{
sinx (x ~ 0) (a = 0).

(c) f(x) = {~sin(1/X) (x =f. 0)

(a = 0).
(x =0)
3 The function f: IR -+ IR is defined by

f(x) = {:2 when x is rational,

when x is irrational.
Show that f is continuous at 0 and continuous at 1, but discontinuous
at all other points.
4 The function f: IR -+ IR is defined by
when x is rational,
f(x) = {~ when x is irrational.
Show that f is discontinuous everywhere.
5 (a) The function f: IR -+ IR is defined by
when x is rational,
f(X)={~ when x is irrational.
Show that f is continuous at the origin and discontinuous at all other
points.
(b) Construct a function f: IR -+ IR which is continuous at andJ2
discontinuous at all other points.

4.3 SOME PROPERTIES OF CONTINUOUS

FUNCTIONS ON CLOSED INTERVALS
In section 4.2 we saw how certain combinations of continuous functions are
themselves continuous. For example, the product of two continuous functions
is continuous. So we have had plenty of practice in deciding whether a
function is continuous or not. But what interesting properties do these
continuous functions possess anyway?
Suppose we take a continuous function whose graph is fairly simple to
draw and consider its behaviour on a closed interval. It is hard to imagine
that a function continuous on a closed interval could be unbounded on this
interval. Perhaps that is hardly surprising as we can prove that a function

145
GUIDE TO ANALYSIS

continuous on a closed is bounded on that interval. Again, our result needs

to be proved using symbols as it must cover the case of a continuous function
whose graph cannot be drawn.

THEOREM 4.3.1 Suppose f(x) is defined for all x such that a ~ x ~ band
f is continuous on the closed interval [a,b]. Thenf is bounded on [a,b], i.e.
there is some positive real number k such that If(x)1 ~ k for all xE[a, b].

Proof Let us see what would happen if f is not bounded on [a, b]. In this
case, if we choose any positive integer n, then there is some point Xn with
a ~ x. ~ b such that
(1)
The points of the sequence (x n ):'= 1 all satisfy the inequality a ~ x. ~ b, i.e. it
is bounded. It must, therefore, have a convergent subsequence (x n);;"= 1 by
the corollary to Theorem 2.5.4. Let

c = lim x n.'
k-+oo

Then a ~ c ~ b since a ~ xn ~ b for all n. Since x n• --+ c as k --+ 00 and f is

continuous on [a, b] it follows that
f(xn)--+ f(c) (2)
as k --+ 00. However, condition (1) shows that
If(xnJ I > nk ~ k
and so If(xnJI--+ 00 (3)
as k --+ 00. This contradicts relation (2). Hence the assumption that f is not
bounded on [a, b] leads to a contradiction and this assumption must be
false. It therefore follows that f is bounded on [a, b] and the proof is complete.

In the statement of the theorem we deliberately included the phrase 'the

closed interval [a, b]' because the result is not true for open intervals, which
may suprise the reader considerably. For example, if f(x) = 1/x (x> 0) then
f is continuous on the open interval (0, 1), but f is not bounded on this interval.
In fact, given any A> 0 (however large) we see that f(x) > A for all x such
that 0 < x < 1/A.
Now suppose we consider the set S of all values assumed by f(x) when
a ~ x ~ b. Iff is continuous on [a, b], then S is bounded. Since it is bounded
above it has a supremum and similarly it has an infimum. Are there values
of x for which f(x) has the same value as this supremum? The next result
answers this question.

146
 FUNCTIONS

THEOREM 4.3.2 Suppose f(x) is defined for all x such that a ~ x ~ band
suppose also that f is continuous on the closed interval [a,b]. Let
M = sup f(x), m= inf f(x).

Then there are points c,d with a ~ c, d ~ b such thatf(c) = M,f(d) = m. (This
means that M is the largest value off(x) on [a, b] and m is the smallest value.)

Proof We begin by considering what happens if the result is not true. Since

M= sup f(x),

it follows that f(x) ~ M for all xe[a, b]. If there is no point ce[a, b] for
which f(c) = M, then f(x) < M for all xe[a, b]. Hence M - f(x) =F 0 for all
xe[a,b]. Using the continuity off and the algebra of limits we see that the
function 9 given by
1
g(x) = M - f(x)

is continuous on [a,b]. By Theorem 4.3.1, 9 must therefore be bounded on

[a, b] and so there is some positive real number k such that Ig(x)1 ~ k for
all xe[a, b].

i.e. M _1f(X) = 1M _If(X) I~k (a~x ~b),

1
which gives f(x)~M -Tc
for all xe[a,b]. Hence M -1/k is an upper bound for f on [a,b] and it is
less than the supremum M. Thus the assumption that there is no c for which
f( c) = M leads to a contradiction. This assumption is therefore false and there
is some point ce[a, b] for whichf(c) = M. Similarly, there is a point de[a, b]
for which f(d) = m. This completes the proof.

The fact that there are points c, de [a, b] at which f( c) = M and f( d) = m

means that the values M,m are actually attained by f(x) on [a,b]. For this
reason Theorems 4.3.1 and 4.3.2 are often taken together and stated succinctly
in the following way. A function which is continuous on a closed interval is
bounded and attains its bounds.
What other properties do we expect a continuous function to possess? Let
us look at sin x. We know that sin(n/2) = 1 and sin(3n/2) = -1, and we
always take it for granted that sin x takes every value between 1 and -1 for
n/2 ~ x ~ 3n/2, because sin x is continuous. In general, if the function f is
continuous on [a, b], then we draw the graph y = f(x) so that y takes every
value between f(a) and f(b) for a ~ x ~ b. This property is captured in the
intermediate value theorem.

147
GUIDE TO ANALYSIS

THEOREM 4.3.3 (Intermediate Value Theorem) Suppose f(x) is defined

for all x such that a ~ x ~ b. Suppose also that f is continuous on the closed
interval [a, b], thatf(a) oF f(b) and that K is any number strictly betweenf(a)
and f(b). Then there is some c such that a < c < band f(c) = K. (Le. f(x)
assumes every value between f( a) and f( b) at least once on [a, b].)

Proof Since f(a) oF f(b), we begin by assuming f(a) < f(b). Let K be any
number such that
f(a) < K < f(b).
(The proof that the value K is not omitted hinges on the fact that the real
numbers obey the completeness axiom and so there are no gaps in the real
numbers. Since the completeness axiom is vital, it is not suprising that the
proof contains essentially two sets of real numbers belonging to [a, b ]-the
set for which f(x) < K and the set for which f(x) ~ K. As x crosses from one
set to the other, f(x) assumes the value K.)
Let S be the set of all real numbers x such that a ~ x ~ band f(x) < K.
Then the set S is bounded above and it is not empty as aES. By the
completeness axiom S has a supremum. Let c = supS. Then
(1)
since S only contains real numbers x for which a ~ x ~ b. We now need to
prove c > a and c < b.
(a) c > a Since f is continuous on [a, b],f(x) --. f(a) as x --.a+. Using
the definition of a limit (with the special value e = K - f(a», we see
that there is some (j > 0 (with (j ~ b - a) such that
If(x) - f(a)1 < K - f(a)
for all x such that a ~ x < a + (j. This inequality gives
f(x)<K
for all x such that a ~ x < a + (j. In particular f(a + (j12) < K and hence
a + (j12ES, which implies that c ~ a + (j12, i.e. c > a.
(b) c < b Since f is continuous on [a, b], f(x) --. f(b) > K as x --. b -.
From this we can deduce that there is some (jl > 0 (with (jl ~ b - a)
such that f(x) ~ K for all x such that b - (jl < X ~ b and these values
of x do not belong to S. Thus all numbers x in S satisfy the inequality
a ~ x ~ b - (jl < b and therefore c = sup S ~ b - (jl < b.
Our results together imply that
a<c<b. (2)
To complete the proof, all that now remains is to demonstrate that c has the
required property, viz. that f(c) = K. To do this we show that f(c) satisfies
both the inequalities f(c) ~ K and f(c) ~ K.

148
 FUNCTIONS

(c) f(c) ~ K By definition c = supS and so for every positive integer n

there is a point x"E[a, b] such that c - lin < x" ~ c and X"ES. Thus
f(x,,)<K
because X"ES. Now f is continuous at c and X,,-+C as n-+oo. Hence:

f(c) = lim f(x lI ) ~ K, (3)

11--+00

using the sequential criterion.

(d) f(c) ~ K If b ~ x > c = supS, then x¢S and f(x) ~ K. Since f is
continuous at c,

f(c) = lim f(x) ~ K. (4)

x-+c+

From relations (3) and (4) we see that f(c) =K and the proof is
complete.

Th~ reader can readily see how the proof needs to be modified for a function
for which f(a) > f(b).

Theorems 4.3.3 and 4.3.2 together show that if f is continuous on [a, b], then
f(x) assumes every value between M and m at least once on [a,b], where
M = sUPa.;;u,bf(x) and m = infa.;;x.;;bf(x). If we happen to be considering an
example in whichf(x) takes both positive and negative values on [a,b], then
m < 0 and M > O. The intermediate value theorem then allows us to deduce
that there is a point CE [a, b] at which f( c) = O. This idea lies behind many
of the examples on continuous functions given to undergraduates.

Examples 4.3.1

1. Show that the equation xesinx = cos x has a solution in the interval (0, nI2).

Solution Whenever trigonometric functions are used in analysis it is always

assumed (unless the contrary is specifically stated) that the unit of measure is a
radian. In this particular example, therefore, the quantity x is measured in radians.
The problem is solved by introducing a function f given by
f(x) = xesinx - cos x.
Using the rules for composite functions, and products and differences of continuous
functions, we see that f is continuous at all points of IR. In particular, it is continuous
on the closed interval [0, n/2].
Moreover, f(O) = -1,
f(n/2) = ne/2 > 0,
and, therefore, by the intermediate value theorem (with K = 0) there is a point Xo
with 0 < Xo < n/2 such thatf(xo) = 0, i.e. x = Xo is a solution ofthe given equation.

149
GUIDE TO ANALYSIS

2. Suppose f(x} is defined for all x such that a ~ x ~ b and suppose also that
a ~ f(x} ~ b whenever a ~ x ~ b. Show that if f is continuous on [a, b],
then the equation
f(x}=x
has a root in [a,b].

Solution Define a function 9 by

g(x) = f(x) - x (a,,;;x";;b).
Then 9 is continuous on [a,b]. Moreover,
g(a) = f(a) - a ~ 0
g(b) = f(b) - b,.;; O.
If f(a) = a, then x = a is a root of the given equation. If f(b) = b, then x = b is a
root of the given equation. If f(a) oF a and f(b) oF b, then
g(a) = f(a) - a> 0,
g(b)=f(b)-b<O,
and there is a number c with a < c < b such that g(c) = 0 by the intermediate value
theorem,
i.e. f(c) = c.
Thus in all cases the equation f( x) = x has a root in [a, b].

When the function f satisfies the conditions given in Example 2 above,

it may be possible to locate a root of f(x} = x by using successive
approximations. As a first approximation choose a number Xl with a ~ Xl ~ b.
Define the sequence (xn}:'= 1 by the recurrence relation
(n= 1,2,3, ... ).
Then a ~ Xn ~ b for all n, since a ~ f(x} ~ b whenever a ~ x ~ b. If the
sequence (Xn}~l converges to a point d, then a ~ d ~ band f(x n)-+ f(d} as
n-+ 00, becausefiscontinuous on [a,b]. However,f(xn } = xn + 1 andxn + l -+d
as n-+ 00, and therefore f(xn}-+d as n-+ 00. Since a sequence cannot have
more than one limit we must have f(d} = d. Thus if (x n ) converges, its limit
d satisfies the equationf(d} = d. So x = d is a root off(x} = x, if(xn } converges.
Of course, there is still an element of doubt, as we do not know whether
the sequence converges. We overcome this difficulty by using the intermediate
value theorem.

Let us look at a particular example. Let f(x} = cos x, a = 0 and b = 1. Then

f satisfies all the required conditions since f is continuous on [0,1] and
o~ cos x ~ 1 when 0 ~ x ~ 1. We therefore have a root of the equation
cos x = x in the interval [0,1].
Now let us see that happens if we use successive approximations. We will
use Xl = 0.5 and

150
 FUNCTIONS

Xn+l = COS Xn (n= 1,2,3, ... ).

Using a pocket calculator we have
X 2 = 0.878, X3 = 0.639, X4 = 0.803, Xs = 0.695, ... , XlO = 0.745, X11 = 0.735,
X 12 = 0.742, X13 = 0.737, Xl4 = 0.740, .... (The reader is reminded that we
are using radians not degrees for measuring x.) It looks as if the sequence
(xn) is converging to 0.74 (correct to two decimal places) and it is reasonable
to guess that x = 0.74 may be a root of cosx = x correct to two decimal
places. Let us now check to see whether we have indeed located a root.
We again use g(X) = f(x) - x = cos X - x,
and evaluate g(0.74 ± 0.005). Using a pocket calculator, we see that
g(0.735) > 0.006,
g(0.745) < -0.009,
and so the equation g(x) = 0 has a root c such that 0.735 < c < 0.745, by the
intermediate value theorem (since g is continuous on [0.735, 0.745]). Thus
c=0.74 correct to 2 decimal places, i.e. the equation cosx=x has a root
x = 0.74 correct to two decimal places.
If the reader requires a root correct to six decimal places then it will be
necessary to find the first forty terms of the sequence (xn). This will give a
value of c which can be checked by taking a very small interval about c and
applying the intermediate value theorem to g. For accuracy to six decimal
places the intermediate value theorem has to be applied to the interval
[0.7390845, 0.7390855]. Later, we will be able to prove (using the results
of Chapter 5) that the equation cos x = x has just one real root.

EXERCISES 4.3.1

1 Show that the equation 2sinx =x 2 -1

has a solution between 1 and 2.
2 Show that the equation 2 tan x = 1 + cos x
has a solution in (0, n/4).
3 Show that the equation xeX = 1
has a solution between 0 and 1. Write the equation as

Put Xl = 1, X2 = e- x " ••• , x n+1= e- x , (n = 1,2, ... ). Use a calculator to

find the first few terms of the sequence (Xn):'l. Hence, find a root of the
equation correct to three decimal places. (Remember to use the intermediate
value theorem to check that you really have found a root.) Use the same
method to find a root correct to six decimal places.

151
GUIDE TO ANALYSIS

(-1

Fig. 4.14
Finally, we conclude with a section on the inverse function. Suppose f: X ..... Y
is a function such that Y = f(X) and f(XI):F f(x 2) whenever Xl :F X2 and
Xl' X2EX, i.e. f: X ..... Y is a bijection. (For a more detailed explanation of the
term bijection, the reader should consult the companion volume, Guide to
Abstract Algebra.) Then for each yE Y there is one and only one XEX such
that y = f(x). The inverse function f- l from Y onto X is defined by
f-l(y) = X, where X is the (unique) element of X such that y = f(x) (see
Fig. 4.14). Thus for each XEX,
f-l(f(x»=x
and for each yE Y

We are interested in the question of whether continuity of f implies

continuity of f-l. Since the existence of an inverse function requires that
f(xl):F f(X2) if Xl :F X2 we will need to stipulate more than just continuity
for f. It proves convenient to use strictly monotonic functions.

DEFINITION 4.3.1 Let X, Y be two sets of real numbers. The function

f: X ..... Y is said to be strictly increasing on X iff(x l ) < f(x 2) for all Xl' X2EX
such that Xl < X2.

DEFINITION 4.3.2 Let X, Ybesetsofrealnumbers. Thefunctionf:X ..... Y

is said to be strictly decreasing on X if f(XI) > f(X2) for all Xt>X 2EX such
that Xl <X2.

A strictly monotonic function is one which is either strictly increasing or

strictly decreasing.

THEOREM 4.3.4 (Inverse Function Theorem) Suppose thatf(x) is defined

for all X such that a ~ X ~ b. Suppose also that f is strictly increasing on

152
 FUNCTIONS

[a,b], that f is continuous on [a,b], and that f(a) = c,j(b) = d. Then the
inverse function f - 1: [c, d] - [a, b] exists, is strictly increasing and is
continuous on [c, d].

Proof Sincef is strictly increasing, it follows thatf(x l ) < f(X2) for all Xl ,X2
such that a:;;;; Xl < X2 :;;;; b and so f(xtl =!- f(x 2). Now f(a) = c,J(b) = d andf is
continuous on [a, b]. By the intermediate value theorem every value between
c and d is assumed on [a,b], i.e., given any y with c:;;;; y:;;;; d, there is some
xE[a,b] such that
f(x)=y.
As f is strictly increasing there is only one such value of x. The function
f-l:[c,d]-[a,b] is defined in the following way. For each y such that
c:;;;;y:;;;;d,
(1)
where X is the number such that a:;;;; x:;;;; band f(x) = y. Now we must prove
that f - 1 is strictly increasing and continuous on [c, d].
(I) f -1 is strictly increasing If we draw a graph the result looks patently
obvious (Fig. 4.15).

y:= !(x)
v
d ------------ - --------- ------------

V2

VI ------------------

c ----------

o a XI

Fig. 4.15

153
GUIDE TO ANALYSIS

Let Y1'Y2 be any two numbers such that c ~ Y1 < Y2 ~ d and let
Xl = f- 1(yd, X2 = f- 1(Y2). Then, by definition, we have
Y1 = f(x 1), Y2 = f(X2),
and we wish to prove that Xl < X2. There are only two possibilities;
either (a) a ~X2 ~ Xl ~ bor(b)a ~ Xl < X2 ~ b. If condition (a) is satisfied,
then f(X2) ~ f(xd because f is strictly increasing, i.e. Y2 ~ Y1' which we
know is untrue. Hence condition (a) is not satisfied and so Xl' X2 must
satisfy (b), i.e. a ~ Xl < X2 ~ b and therefore f- 1(Y1) < f- 1(Y2) whenever
c ~ Y1 < Y2 ~ d. Hence f- 1 is strictly increasing.
(II) f - 1 is continuous on [c, dJ First choose any number Y 1 such that
c < Y1 < d. To prove continuity at Y1 we have to show that given any
e > 0, If- 1(y) - f- 1(Y1)1 < e for all Y sufficiently close to Y1. If we write
x=f- 1(y), x 1 =f- 1(Y1)' then we need to prove that Ix-x 11<e
whenever Y is sufficiently close to Y1. Figure 4.16 might prove helpful in
disentangling the variables. What we have to prove is that X is within e
of X 1 if the corresponding Y is sufficiently close to Y1. The diagram suggests
that sufficiently close to Y1 means within [) of Y1' where [) is the smaller
of the numbers h, k. The reader may find it useful to keep referring to
Fig. 4.16 during the execution of the proof in the next section.

y,+k - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

y, -------------------------

y, -h

o a X, -E x, X, +E x

Fig. 4.16

154
 FUNCTIONS

To prove the result using symbols, let Xl = f-l(yd and let e > 0 be
any positive real number such that a:::; Xl - e < Xl + e:::; b.

Let /(Xl-e)=Yl-h,

/(Xl +e)=Yl +k.

Then h > 0 and k > 0 since / is strictly increasing and we also have
C:::;Yl -h<Yl +k:::;d.
Let (j be the smaller of the two numbers h, k. Then (j > 0 and

Now f- l is strictly increasing, and therefore if Y is any number such that

f-l(Yl-h):::;/-l(Yl-(j)</-l(y)</-l(Yl + (j):::;/-l(Yl +k)

i.e. Xl -e:::;/-l(Yl-(j)</-l(y)</-l(Yl + (j):::;X l +e.

Thus I/-l(y) - xIi < e for all Y such that Yl - (j < Y < Yl + (j, i.e.
If-l(y) - f-l(ydl < e for all Y such that Iy - Yll < (j. This means that
if we are given any e > 0 then there is a corresponding (j > 0 such that
I/-l(y) - /-l(Yl)1 < e whenever Iy - Yti < (j. Hence /-1 is continuous
at Yl> which is any point of (c, d). A similar argument shows that
/-l(y)_/-l(C) as y-c+ and /-1(y)_/-1(d) as y-d-. This
completes the proof that /-1 is continuous.

Naturally, this result remains true if the phrase 'strictly increasing' in the
statement of the theorem is replaced by the phrase 'strictly decreasing'
throughout.

* * * * *
This completes a long and somewhat difficult section on limits and continuity.
The reader who has found the going tough may be pardoned for asking if
there is an easier way to learn analysis. Such a question is not original.
Ptolemy, who became ruler of Egypt after the death of Alexander the Great,
found Euclid's elements tough reading and so he asked Euclid whether there
was an easier way to a knowledge of geometry. Euclid made a very simple
reply: 'There is no royal road to geometry.' Today we could make a similar
pronouncement about analysis. There is no royal road to analysis. An
understanding of analysis can be gained only with hard work, but it will
prove worth the effort in the end. The final chapter of this volume furnishes
~ome respite for the weary reader. It introduces the notion of differentiability
and then illustrates some of its uses.

155
GUIDE TO ANALYSIS

MISCELLANEOUS EXERCISES 4
1 The function f: IR --+ IR is given by

{~:t
(x < 0),
(O~x<I),
f(x) =
(x= 1),
x 2 -1 (x> 1).
Determine whether the following limits exist:
(a) limx~o+f(x), (c) limx~l_f(x),
(d) limx~ J(x).
When the limits exist find them. Decide where f is continuous.
2 The function f: IR --+ IR is given by
if x is rational,
f(x) = {~ if x is irrational.
Show that f is continuous at the origin and discontinuous at all other
points.
3 The function f: IR --+ IR is given by f(x) = x[x] for all real x, where [x] is
the greatest integer which is either less than x or equal to x. Sketch a
graph of f Determine where f is continuous.
4 Letthefunctionsf, gboth be continuous on [a, b]. Suppose thatf(a) < g(a)
and f(b) > g(b). Show that there is a number c with a < c < b such that
f(c) = g(c).
5 Suppose the function f has the property that a ~ f(x) ~ b whenever
a~x~b, i.e. f:[a,b]--+[a,b]. By considering the function g given by
g(x) = f(x) - x, show that there is a number c such that a ~ c ~ b, and
f(c) = c.
Deduce that the equation
has a solution in [ -1, + 1].
6 The function f: IR --+ IR satisfies the relation f(x + y) = f(x)f(y) for all
x,YER Prove that
(a) f(l) ~ 0;
(b) if f(l) = 0, then f(x) = 0 for all x;
(c) if f( 1) = a > 0, then f(p) = a P for all positive integers p, and
f(p/q) = aP/ q for all rationals p/q.
Show further that if f(l) = a > 0, and f is continuous at each point of
IR, then f(x) = aX for all XER

156
 FUNCTIONS

7 Show that the equation

sin 19 x + COS ll1 X = X

has a solution in the interval [ - 2, 2].

HINTS FOR SOLUTION OF EXERCISES

Exercises 4.1.1
3 Use Theorem 4.1.1.

Exercises 4.1.4
1 (a) and (b). Use algebra of limits. (c) Use Theorem 4.1.1.
(d) Use sequences. The following are suitable:
xn=mt(ne;z+), Yn=2mt+n/2 (neZ+).
In this case f(x n ) = 0 and f(Yn) = y~/(y; + 1).
(e) and (f). Use same sequences as in (d).
(g) Use the inequality
x
If(x)1 ~ x 2 + 1 (x> 0).

Exercises 4.1.5
1 Many of these parts use the algebra of limits. In addition the following
points may be helpful:
(b) Remember, this is a limit as x -+ 1.
(c) Remember that sinx/x-+l as x-+O.
(e) Use sequences. Suitable ones are (x n ) and (Yn), where
Xn = l/nn and Yn = 1/(2nn + n/2) (neZ+).
In this casef(xn) = (-I)nsin(l/nn) andf(Yn) = cos[I/(2nn + n/2)].
(f) Use the identity a2 - b 2 = (a - b)(a + b) to deal with the difference of
two square roots-see previous sets of exercises (Chapter 2).
(g) Deal with the cases a = 0 and a =1= 0 separately. Note that if a =1= 0 then
sin ax/ax -+ 1 as
(i) Since sinx/x-+ 1 as x-+O, there is some b >0 such that!< sin x/x <j
for 0 < Ixl < b. Hence if 0 <x < b then
sin x 1
-- >-
x
2 2x
and f(x)-+ 00 as x-+O+. Similady,J(x)-+ -00 as x-+O-.

157
GUIDE TO ANALYSIS

2 Find limits as x -+ 0 + and as x -+ 0 - from the definition of f

Exercises 4.2.1
2 (a) Use the fact th~(x) = -1 for -1 x < O.
x'"
~
(c) Use If(x)1 ~ .Jlxl for O.
3 and 4 Use sequences.
5 (a) Use sequences.

Miscellaneous Exercises 4
2 Use sequences.
3 If nelL and n ~ x < n + 1 then [x] = nand f(x) = nx for n ~ x < n + 1.
6 (a) Usef(1)=f(t+t)=[f(t)]2.
(b) Use f(x) = f(x -1 + 1) = f(x - 1)f(1).
Finally, to prove that f(x) = aX for all real x, use a sequence of rational
numbers which converges to x.

ANSWERS TO EXERCISES

Exercises 4.1.4
1 (a) f(x)-+l (b) f(x)-+-1. (c) f(x)-+oo.
(d) No limit. (e) No limit. (f) No limit.
(g) f(x)-+O. (h) f(x)-+ 00. (i) f(x)-+O.
2 The following are just some possible examples:
(a) f(x) = x and g(x) = sinx. Use sequences to show that f(x)g(x) does
not tend to any limit as x -+ 00. The sequences (x n) and (Yn) would be
suitable, where Xn = mr and Yn = 2mr + n/2.
(b) If f(x) =x 2 and g(x) = 1/(1 + x 2 ), thenf(x)g(x)-+ 1 as x-+ 00.
If f(x) = x 3 and g(x) = 1/(1 + x 2 ), then f(x)g(x)-+ 00 as x -+ 00.
4 The following are some possible examples:
(a) f(x)=xandg(x)= -x+sinx. (b) f(x) = x and g(x) = -x - x 2 •
(c) f(x) = x + 1 and g(x) = -x.

Exercises 4.1.5
1 (a) f(x) -+ 1. (b) f(x)-+t. (c) f(x) -+ 1.
(d) f(x)-+O. (e) No limit. (f) f(x) -+ 1.
(g) f(x) -+ IX. (h) No limit. (i) f(x)-+O.
U) No limit.
2 As x-+O,J(x)-+O. As x-+ 1,J(x)-+ 1.

158
 FUNCTIONS

3 The following are some suitable examples:

(a) f(x) = x -1 and g(x) = (x -1)sin(x -1).
(b) f(x) = x -1 and g(x) = (x _1)3.
(c) f(x) = 3(x - 1) and g(x) = (x -1).

Exercises 4.2.1
2 (a) f is not continuous at the origin. (b) f is continuous at the origin.
(c) f is continuous at the origin.
5 (b) One possible example is the function f given by

f(X)={f when x is rational,

when x is irrational.

Miscellaneous Exercises 4
1 (a) f(x)~O as x~O+. (b) No limit.
(c) f(x)~Oasx~1-. (d) f(x)~Oasx~1.

The function f is discontinuous at the origin and continuous at all other

points of R
3 f is continuous at all points of R\Z and also continuous at the origin. It
is discontinuous at all other points.
4 Apply the intermediate value theorem to f - 9 on the interval [a, b].

159
5 DIFFERENTIABLE
FUNCTIONS
Differentiable functions, Rolle's
theorem, mean value theorem,
Taylor's theorem, power series
expansions of functions, I' Hopital 's
rule, maxima and minima

5.1 DIFFERENTIATION
Calculus originated with the work of two celebrated mathematicians, both
born in the seventeenth century. One of these was Sir Isaac Newton
(1642-1727), the other was Baron Gottfried Wilhelm von Leibniz (1646-
1716).
Newton's extraordinary ability was noticed early in his life by a maternal
uncle who encouraged his mother to enter him for Cambridge. As a result,
Newton entered Trinity College, Cambridge, in 1661. Soon after he gained
his degree the college was temporarily closed because of the plague and
Newton returned home to continue his work. This period of enforced absence
from Cambridge in 1665-6 saw the birth of many new ideas which laid the
foundation for Newton's subsequent work. In particular, the period was
marked by the beginnings of calculus and the emergence of infinite series as
a tool in mathematics. During the next few years Newton wrote several
accounts of his work. The first of these was the treatise already mentioned
in earlier chapters, De analysi per aequationes numero terminorum injinitas,
which he wrote in 1669. Initially the manuscript was used mainly for
circulation among his friends and its publication was delayed until 1711. In
fact, Newton's first published account of calculus did not appear until 1687.
It was his celebrated Philosophiae naturalis principia mathematica-frequently
referred to in literature as 'Newton's Principia'. Newton's actual terminology
might seem a little strange to us today and his notation, though not totally
unfamiliar, is not the one in most common use. However, this work does
include some of the familiar rules for differentiating products, quotients, etc.
which we will derive later in the chapter, though we will not use Newton's
notation.
Leibniz was born in Leipzig and entered the local university at the early
age of fifteen, to study theology, law, philosophy and mathematics. After

160
 DIFFERENTIABLE FUNCTIONS

obtaining his doctorate at the University of Altdorf, he was offered a

professorship in law, which he declined in favour of a career in the diplomatic
service. During his travels abroad he maintained an interest in mathematics
and became a member of the Royal Society when he visited London in 1673.
At that time, manuscripts of Newton's De Analysi were in existence and it
would have been possible for him to have obtained one. It is, however, very
doubtful whether he could have gained much from reading such a manuscript
as he lacked much of the necessary background knowledge of mathematics at
that stage.
The year 1684 was marked by the appearance ofthe first published accounts
of differential calculus written by Leibniz. It is known that he developed his
theory of differential calculus between his visits to London in 1673 and 1676,
and it is generally accepted nowadays that Newton and Leibniz worked
completely independently. In the first edition of his Principia (1687) Newton
acknowledged Leibniz's contribution to calculus. The following years saw
heated arguments between adherents of Newton and Leibniz about whether
Leibniz had been guilty of plagiarism and the Royal Society was asked to
~djudicate. When the third edition of Newton's Principia appeared (1726)
he had deleted all reference to Leibniz. The most serious consequence of
these disagreements was the rift it produced between mathematicians in
England and those on the Continent. At a time when new disciplines like
analysis were beginning to emerge on the Continent, English mathematicians
were virtually isolated-a situation which was not reversed until the early
nineteenth century.
Leibniz was a firm believer in the vital importance of clear, comprehensible
notation. He was the one to introduce the 'd' notation and the calculus of
today owes its form primarily to the influence of Leibniz. Like Newton he
derived formulae for derivatives of products and quotients. He is particularly
remembered by generations of students for his formula for the nth-order
derivative of the product of two functions-a formula which is called
Leibniz's rule to this day.
In the sixth form differentiation is normally introduced in the following
way. Write y = f(x) and y + (iy = f(x + (ix), where (iX:F O. Then subtract to
obtain
(iy = f(x + (ix) - f(x),

(iy =f(x + (ix) - f(x)

which gives
(ix (ix
If we can draw the graph y = f(x), then the ratio (iy/(ix measures the gradient
of the chord PQ (see Fig. 5.1), where P is the point with coordinates (x,y)
and Q has coordinates (x + (ix,y + (iy). Thus «(iy/(ix) = tan IX, where IX is the
angle which the chord PQ makes with the positive direction of the x-axis.
The final step is to consider what happens to «(iy/(ix) as (ix-+O, i.e. we need
to consider whether [f(x + (ix) - f(x)]/(ix tends to a limit as (ix -+0. This

161
GUIDE TO ANALYSIS

p
- - - - - - - -- - - - - - - - - -=..--:;,.;-;;.;;-=-,;;;,;--=-...- --f"=---;::----1
/)x

o x
Fig. 5.1

should not present an insurmountable problem as we know how to deal with

limits of functions.
The whole process we have outlined can be carried out using only symbols
without any recourse to a diagram at all. Of course, if it is to make sense,
then f(x) must be defined and f(x + <5x) must be defined for all sufficiently
small <5x. This means we will have to stipulate that the domain of definition
off has certain properties, but this will not cause undue difficulty. Normally,
in this context, the symbol h is used in place of <5x, and we therefore consider
what happens to
f(x + h) - f(x)
h
as h --+ O. Finally, in order to ensure that our definition is precise, we restrict
our attention to the behaviour of f near a particular point Xl.

DEFINITION 5.1.1 Suppose f(x) is defined for all X such that Xl - R < X
< Xl + R, where R is some given positive real number, i.e. f is defined for all
points sufficiently close to Xl. If the quotient
f(x l + h) - f(xd
h

162
 DIFFERENTIABLE FUNCTIONS

tends to a finite limit as h -+ 0, then f is said to be differentiable at x 1. The

value of this finite limit is then called the derivative of f at x 1 and it is denoted
by f'(X1)'
i.e. f '( Xl ) -I· f(X1 + h) - f(X1)
- 1m h .
h-+O

DEFINITION 5.1.2 Suppose f(x) is defined for all x such that a < x < b.
Then f is said to be differentiable on tbe open interval (a, b), if f is differentiable
at all points of the interval.

The reader will notice that Definition 5.1.1 includes an explicit statement
concerning the domain of definition of f It is assumed that this domain
contains some interval (Xl - R, Xl + R) (R > 0). Whenever differentiability
is used it is taken for granted that some such condition is satisfied unless it
is explicitly stated otherwise.
Sometimes it is convenient to use x in place of Xl + h. In this case h is
replaced by x - Xl' and the condition h-+O is replaced by the condition
X-+X 1. With this notation,

f '( Xl ) -- I·1m f(x) - f(x 1).

x-+x, X - Xl
Indeed, the definition of differentiability is frequently stated using just this
notation. The statement would then read as follows.
f(x)- f(xd
If the quotient

tends to a finite limit as X-+X1' thenf is said to be differentiable at Xl.

Now suppose that f is differentiable at Xl. Then

and hence
lim [f(x) - f(x l )] = lim [f(X) - f(X I ) ] lim (x - Xl) =0
X-+Xl X-+Xl X - Xl X-+Xl

using the algebra of limits.

i.e. lim f(x) = f(xd.

Thus f is continuous at Xl and we have shown that differentiability implies

continuity. This result is now stated in the form of a theorem.

THEOREM 5.1.1 Suppose that f(x) is defined for all X such that

163
GUIDE TO ANALYSIS

Xl - R < X < Xl + R, where R is some positive real number. If f is

differentiable at X I ' then f is continuous at X I .

COROLLARY Suppose f(x) is defined for all X such that a < X < b. Iff is
differentiable on (a, b), then f is continuous on (a, b).

There are, however, continuous functions which are not differentiable and
the converse ofTheorem 5.1.1 is not true. For example, letf: ~ -+ ~ be defined
by f(x) = Ixl. Thenf is clearly continuous at the origin since f(x)-+O = f(O)
asx-+O. Nowifx>O,
f(x) - f(O) = X - 0= 1
x-o x-o
and lim f(x) - f(O) = 1.
(1)
x .... o+ x-o
H owever, if X < o, f(x)-f(O)_lxl-IOI_
0 - 0 -
-x-O_ 1
0 --
x- x- x-

and lim f(x) - f(O) = -1.

(2)
o- x-o
x ....

From equations (1) and (2) we see that

f(x) -f(O)
x-o
does not tend to any limit as x -+ O. Thus f is not differentiable at the origin.
The graph of this particular function is easily sketched (see Fig. 5.2). It has

Fig. 5.1

164
 DIFFERENTIABLE FUNCTIONS

a V-shape at the origin. The slope is + 1 on the right of the origin and - 1
on the left, which are the right and left limits we obtained in equations (1)
and (2).
This is just one example used to illustrate the fact that continuity does not
imply differentiability. It is a particularly simple example, in which the
derivative does not exist at the origin, but it exists at all other points. There
are far more complicated situations than this one. There are, for example,
functions f: IR -+ IR which are continuous at all points of IR and not differentiable
anywhere at all. Such functions are continuous everywhere, but their graphs
cannot be sketched as they do not have a tangent at any point at all. It is
therefore fortunate that our definitions of concepts like continuity are not
dependent on geometric representations.
In their first accounts of the theory of differential calculus, Newton and
Leibniz both included rules for differentiating sums, products, etc. These
familiar rules can now be proved without any difficulty.

THEOREM 5.1.2 Suppose that f(x) and g(x) are both defined for
x I - R < x < X I + R where R is some given positive real number. If f, 9 are
differentiable at Xl' then
(a) f + 9 is differentiable at Xl' with derivative !'(x I ) + g'(xd;
(b) fg is differentiable at Xl' with derivative f(xl)g'(xd + !'(xdg(x l );
(c) fig is differentiable at Xl with derivative
f'(XI)g(X I ) - f(xl)g'(X I )
[g(xl)F
provided g(xd =I- O.
Proof Since f,g are differentiable at Xl'
f(x) - f(xd f'( ) (1)
--------'--+ Xl'
X-Xl
g(x) - g(x l ) '()
- - - - - - + g Xl (2)
X-Xl
as X -+ X l ' Moreover, f, 9 are both continuous at XI' since differentiability
implies continuity,
I.e. f(x)-+ f(xd (3)
g(x)-+g(x l ) (4)

(a) LetO<lx-xll<R.Then
[f(x) + g(x)] - [f(x l ) + g(x l )]
=------'----------=- = f(x) - f(x l ) g(x) - g(x l )
+ -'---'-------=-.:--"'-
X-Xl X-Xl X-Xl

165
GUIDE TO ANALYSIS

and this tends to !'(x l ) + g'(xd as X--+Xl' using the algebra of limits
and relations (1) and (2). Hence f + g is differentiable at Xl and its
derivative at Xl is !'(Xl) + g'(Xl).
(b) LetO<!x-xl!<R. Then
f(x)g(x) - f(xl)g(xd =f(x)g(x) - f(xl)g(x) + f(Xl)g(X) - f(xdg(xd

and this tends to g(xl)f'(xd + f(xdg'(Xl) as X--+Xl using the algebra

of limits and (1), (2) and (4). Thus the product fg is differentiable at
Xl and its derivative at Xl is
!'(Xl)g(X l ) + f(xl)g'(xd·
(c) Since g is continuous at Xl and g(xd # 0, there is some J > 0 (with
J ~ R) such that g(x) # 0 for all X such that Xl - J < X < Xl + J, by
Theorem 4.2.1. Let 0 <!x - Xl! < J. Then
1 1
-----
g(x) g(xd

1 1 g(x) - g(xd
----
g(x) g(xd X - Xl
and this tends to -g'(xd/[g(xl)]2 as X--+Xl' using the algebra of
limits and relations (2) and (4). Thus 1/g is differentiable at Xl' and
its derivative at this point is
-g'(Xl)
[g(x l )] 2·
Using the rule for differentiating a product, we see that fig is
differentiable at Xl; its derivative is
!'(Xl)g(X l ) - f(xl)g'(xd
[g(Xl)]2
This completes the proof.

Up to this point we have taken great care to state explicitly in each theorem
that the functions under consideration are defined on certain specified
intervals. Since continuity and differentiability are defined on the under-
standing that the domain of definition of the functions satisfies certain
conditions, such care is not really necessary. In future we will omit such

166
 DIFFERENTIABLE FUNCTIONS

details, and will take it for granted that the concepts of continuity and
differentiability are now sufficiently familiar to the reader that such reminders
are no longer needed.
We can now differentiate sums, products and quotients of differentiable
functions. Can we manage to differentiate more complicated functions like
cos (eX)? If we are to handle such functions then we must have a rule for
differentiating composite functions.

THEOREM 5.1.3 Suppose that the function g is differentiable at Xl and

g(Xl) = Yl' Suppose also that the function f is differentiable at Yl' Then the
composite function fog is differentiable at Xl and its derivative at Xl is
f'(Ydg'(xd·
(If we write Y = g(x), U = f(y) so that u = f(g(x», then the above rule can be
written as
du du dy
dx = dy 'dx

since :; = f'(y) and :~ = g'(x), which is the familiar rule for differentiating
a function of a function.)

Proof Since g is differentiable at Xl,g(X) is defined for all X sufficiently close

to Xl and g is continuous at Xl' Using the fact that f is differentiable at Yl'
we can conclude thatf(g(x» is defined for all X such that Xl - R < X < Xl + R
(where R is some given positive real number)-see the section preceding
Theorem 4.2.4.
Since f is differentiable at Yl'

as k~O. We define the function e by e(k) = 0 when k = 0 and

e(k) =f(Yl + k) - f(yd f'(Yt) (k ~O). (1)

k
Then f(Yl + k) - f(Yd = kf'(Yl) + ke(k), (2)
where e(k)~O as k~O.
Now let 0 < Ihl < R, and write
g(x l + h) = Yl +k
so that k = g(Xl + h) - g(x 1 ), (3)

167
GUIDE TO ANALYSIS

because Yl = g(Xl). Then we have

f(g(x 1 + h)) - f(g(x 1 )) = f(Yl + k) - f(Ytl
= kj'(Yl) + h(k)
f(g(x l + h)) - f(g(x l ))
by (2). Hence (4)
h
k g(x l + h) - g(x l )
Now, by definition,
h h

and therefore ~--+g'(xtl (5)

h
as Moreover g is continuous at Xl and so k = g(x l + h) - g(xd~O as
h~O.
h~O. Thus as h--+O, we must have k~O which in turn implies that e(k)~O.
From (4) it now follows that as h ~ 0
f(g(x l + h)) - f(g(x l )) '( )f'( )
h ~g Xl Yl

and so fog is differentiable at Xl with derivative f'(Yl)g'(X l ). This completes

the proof.

The above proof was rather long and involved. The reader who is wide awake
might be tempted to ask whether we could use a more straightforward method.
Using the same notation it is rather tempting to use the following argument.
Write
f(g(Xl +h))- f(g(x l )) f(Yl + k) - f(Yl) . ~
(*)
h k h·
Now k/h~g'(Xl) as h--+O and it all looks plain sailing. Unfortunately there
is one serious snag. We do not know whether there are arbitrarily small
non-zero Ihl for which the corresponding value of k is zero. The relation (*)
does not make sense if k = O. To allow for the possibility of such an occurrence,
we have been forced to use the longer argument.
Finally we need a rule for differentiating inverse functions.

THEOREM 5.1.4 Suppose the function f is continuous and strictly

increasing on the closed interval [Xl - R,Xl + R] where R is some given
positive real number. Suppose also thatf is differentiable at Xl andf'(xl) =F O.
Let Yl = f(x l )· Then the inverse function is differentiable at Yl and its
derivative at that point is l/f'(x l ).

Proof Let c = f(x l - R), d = f(x l + R), thenfhas an inverse function which
is continuous and strictly increasing on [c,d] by Theorem 4.3.4. We will use

168
 DIFFERENTIABLE FUNCTIONS

g for this inverse function rather than f - t to prevent confusion with

superscripts.
Let k be any non-zero number such that Yt + k lies in [c,d], and let
g(Yt +k)=x t +h. (1)
Then h # 0 since g is strictly increasing and
Yt +k=f(x l +h). (2)
Moreover g(Yt + k) - g(Yt) (Xt + h) - Xl h
k k k
h
f(Xt + h) - f(xd·
Now k-+O implies h-+O as g is continuous and the condition h-.O in tum
implies that

Thus as k-+O
g(Yt + k) - g(Yd 1
=-=---=-------:--=-=-= -+--
k !'(x t )
and the inverse function g is differentiable at Yt, with derivative 1/!'(x t ).
This completes the proof.

We have shown that g'(Yl) = l/!'(x t ). Now, using the notation Y = f(x) and
X = g(y), we see that

'()
g Y = dx
dy"

The relation g'(Yt) = 1/!'(x l ), therefore, gives the familiar rule

dx =
dy
1/dY.
dx
The result in Theorem 5.1.4 remains true if the phrase 'strictly increasing'
is replaced by the phrase 'strictly decreasing' throughout the statement of
the theorem.

Example 5.1.1

The functions f: IR -+ IR and g: IR -+ IR are defined by

f(x) = {xo sin(1/x) (x # 0), g(x) = {X02 sin(1/x) (x #0),

(x =0); (x =0).

169
GUIDE TO ANALYSIS

Decide where J,g are (a) continuous, (b) differentiable. Find the derivatives
where they exist.

Solution
(a) In Example 3 at the end of section 4.2 we showed that 1 is continuous at every
point of JR. Moreover, g(x) = xl(x) for all xeJR and hence g is also continuous
at all points of JR, since the product of two continuous functions is continuous.
(b) Since l/x is differentiable at all points of JR\{O} and the sine is differentiable
at all points of JR, it follows from the law for differentiating composite functions
that the derivative of sin(l/x) is

(x i' 0).

Using the rule for products we see that

f'(x) = sin(l/x) -~cos(~) (x i' 0)

g'(x) = 2xsin(l/x) - cos(1/x) (x i'0).

Thus J, g are both differentiable at all points of JR\ {O}.
The question of behaviour at the origin still remains. If x i' 0
I(x) - 1(0) = x sin(1/x) = sin{1/x)
x-O x
and we know that sin{1/x) does not tend to any limit as x --+ O. Hence 1 is not
differentiable at the origin.
g(x) - g(O) x 2 sin{1/x) . (1/ )
I f x i' 0, = x sm x.
x-O x
and xsin(l/x)--+O as x--+O. Thus g is also differentiable at the origin with
g'(O) = O.

EXERCISES 5.1.1
1 Use the identity
. x+a . a-x
cos x - cos a = 2 slD-2- slD-2-

to show that cos x is differentiable at every point xelR and its derivative
is -sinx. (You may use the fact that (sint)/t~l as t~O.)
2 The function f: IR -+ IR is defined by
e% (x < 0),
f(x)= { x+l (x ~ 0).

Show that f is differentiable at all points of IR.

170
 DIFFERENTIABLE FUNCTIONS

3 Find the derivative of each of the following functions f In each case state
explicitly the values of x for which the formula for f' is valid.
(a) f(x) = x 3 sinh x sin(x2); (b) f(x) = exsinx;
(c) f(x)=cos[1og(1+x2)]; (d) f(x)=sinh- 1(1+x 2);
(e) f(x) = x 10gx (x> 0).

Higher-order derivatives
Suppose the function f is differentiable at all points of the open interval
(a - R, a + R), where R is some positive real number. Then f'(x) is defined
for all x such that a - R < x < a + R and we could ask the question: Is f'
differentiable at a? If [f'(x)- f'(a)J/(x-a) tends to a finite limit as x~a,
thenf' is itself differentiable at a. Its derivative will be denoted by f"(a), which
is normally called the second-order derivative of f at a. Higher-order
derivatives can be defined in the same way. There are no new principles
involved.

DEFINITION 5.1.3 Suppose the functionf is differentiable at all points of

an open interval (a - R, a + R), where R is some positive real number. If the
derivative f' is itself differentiable at a, thenf is said to possess a second-order
derivative at a, which is denoted by f"(a).
Thus f"(a) = limf' (x) - f'(a).
X"" a x- a
This can be generalised to give the definition of an nth-order derivative.

DEFINITION 5.1.4 Suppose the function f possesses an (n -1)th-order

derivative at all points of an open interval (a - R, a + R), where R is some
positive real number and n is positive integer with n ~ 2. If the (n -1)th-order
derivative j<n - 1) is differentiable at a, then its derivative is called the nth-order
derivative of f at a and denoted by f(n)(a). Thus
j<n-1)(x) _ j<n-1)(a)
j<n)(a) = lim .
X"" a x-a

DEFINITION 5.1.5 If f possesses an nth-order derivative at all points of

an open interval (a, b), we say that f is n times differentiable on (a, b).

As one might expect, slight modifications are made when dealing with a
closed interval. In the case of continuity of a function f on a closed interval
[a, bJ we used only one-sided limits at the end-points a, b in order to avoid

171
GUIDE TO ANALYSIS

considering what happens to f(x) when x is outside [a,b]. The same

convention holds for differentiability on a closed interval.

DEFINITION 5.1.6 Let f(x) be defined for all x such that a ~ x ~ b.1f f is
differentiable at all points of the open interval (a, b) and the one-sided limits
. f(x) - f(a)
I1m . f(x) - f(b)
I1m
, ------::--
x-+a+ X - a x-+b- x-b
both exist and are finite, then f is said to be differentiable on the closed
inte"al [a, b]. In these circumstances, it is normal to write f' (a) for
. f(x) - f(a)
II m----
x-+a+ X - a
and f'(b) for
. f(x) - f(b)
11m •
x-+b- x-a

Naturally, if the domain of definition of f extends outside [a,b] and if the

normal derivatives exist at a and b, they have the same value as the respective
one-sided limits and the end-points need no special treatment. For example,
we have seen that if f(x) = cos x, then f is differentiable at all points of~,
and therefore it is differentiable at all points of any closed interval [a, b].
Hence f is differentiable on [a, b]. In this case there was no special problem
at the end-points.
The same convention is adopted for higher-order derivatives on a closed
interval. If n is a positive integer with n ~ 2, then f is said to possess an
nth-order derivative on a closed interval [a, b] if the (n - 1)th-order derivative
pn-l) exists on [a,b] and is differentiable on [a,b].

We are now the proud possessors of a rigorous theory of differentiation, but

have we any differentiable functions on which to try it out? Perhaps the most
obvious place in which to commence our search is the stock of continuous
functions from Chapter 4. Clearly the function f: ~ -+ ~ given by f(x) = x is
differentiable at all points of ~. For if XoE~ and x =I xo, then
f(x) - f(xo) = x - Xo = 1
x-xo X-Xo
f(x) - f(xo) -+ 1
and
X-X o

as x -+ xo. Hence f is differentiable at all points of ~ and f' (xo) = 1 for all
XoE~. Similarly, we can easily show that a constant function is differentiable
with derivative zero. Since sums and products of differentiable functions are

172
 DIFFERENTIABLE FUNCTIONS

differentiable, it follows that positive integral powers are differentiable

everywhere, and the derivative of x" is nx"-l. Thus polynomials are
differentiable at all points of R Hence a rational function (i.e. the quotient
of two polynomials) is differentiable at all points at which the denominator
is not zero. Using the fact that sin tit -+ 1 as t -+ 0, we can prove that the sine
and cosine are differentiable everywhere and have the familiar derivatives.

5.2 ROLLE'S THEOREM AND THE MEAN VALUE

THEOREM

In the previous section we showed how the technical rules for differentiation
can be derived using as a basis the theory of limits. This section contains
some rather pretty results for differentiable functions.
The first result is named after the French mathematician Michel Rolle
(1652-1719). It is ironic that Rolle should above all be remembered in
connection with a theorem concerning differentiation, as he was never
impressed with calculus. In fact, he is supposed to have dismissed the subject
as 'an ingenious collection of fallacies'. His result was originally published
in 1691 in an obscure book on geometry and algebra. It is really very simple,
but its consequences are far-reaching.

THEOREM 5.2.1 (Rolle'S Theorem) Suppose that

(i) the function f is continuous on the closed interval [a,b],
(ii) f is differentiable on the open interval (a,b),
(iii)f(a) = f(b).
Then there is some c with a < c < b such that
f'(c) = o.

Proof Let us begin by sketching a diagram (Fig. 5.3). Since f(a) = f(b), the
graph of y = f(x) is the same height above the x-axis at the two ends x = a
and x = b. It seems patently obvious that there is some point c between x = a
and x = b at which the tangent is parallel to the x-axis, i.e. there is some
point c such thatf'(c) = O. Now let us see if we can prove this using symbols
rather than geometric reasoning.
Sincef is continuous on the closed interval [a,b],! is bounded on [a,b]
by Theorem 4.3.1. Let

M= sup f(x), m= inf f(x).

and let f(a) = f(b) = k. Then we must have m ~ k ~ M.

173
GUIDE TO ANALYSIS

I
I
I
I
I
I
I
I
-------1-------
I
I
I
I
1
I
I
I
I
I
I
I
I
I
I
I
o a c b x
Fig. 5.3

Ifm = k = M, then f(x) = k for all x such that a:S;; x:S;; b. Hence f'(c) = 0
for all c such that a < c < b.
If m #- M, then m < k or k < M. Suppose k < M. Then there is some point
c with a < c < b at which f(c) = M by Theorem 4.3.2. Moreover, f'(c) exists
because a < c < b by condition (ii). Now, for all xE[a, b ],f(x):s;; M and hence
if a :s;; x < c, then
f(x) - f(c) =f(x) - M ~ O.
x-c x-c

As x--+c-, f(x) - f(c) --+ f'(c)

x-c
and therefore f'(c) ~O. (1)
However, if c < x :s;; b,
f(x) --f(c) =f(x) - M :s;; 0
x-c x-c

and as x --+ c + , f(x) - f(c) --+ f'(c).

x-c
Hence f'(c):S;; o. (2)

174
 DIFFERENTIABLE FUNCTIONS

Since f'(c) satisfies conditions (1) and (2) we must have f'(c) = o.
Similarly, if m < k, it can be proved thatf'(c 1 ) = 0 at the point Cl at which
f(c 1 ) = m. This completes the proof.

Suppose we apply this theorem in the special case in which f(a) = f(b) = O.
Then we have a point C with a < C < b such that !,(c) = 0, i.e. between any
pair of zeros of f there is a zero of f'.

Example 5.2.1

In the previous chapter we used the intermediate value theorem to show that
the equation
xesinx = cos x
has a root in the interval (0, n/2). Now we can use Rolle's theorem to prove
that it does not have more than one. Write
f(x) = xesinx - cosx.
Then f is continuous at all points of ~ and differentiable at all points of R
If there are two distinct roots x = a, x = b of the equation in [0, n/2] , then
0= f(a) = f(b) and by Rolle's theorem!' is zero at some point between a and
b. However,
f'(x) = esinx + xcosxe sinx + sin x
and for 0 < x < n/2, f'(x) > eO = 1. Since f' does not have a zero on (0, n/2)
the equation cannot have more than one root in (0, n/2). Thus the equation
has precisely one root between 0 and n/2.

Rolle's theorem is an interesting result in its own right. It also has a part to
play in proving one of the most useful results in elementary analysis-the
mean value theorem.

THEOREM 5.2.2 (Mean Value Theorem) Suppose that

(i) the function f is continuous on the closed interval [a,b],
(ii) f is differentiable on the open interval (a,b).
Then there is a number c with a < c < b such that
f(b) - f(a)
f'(c).
b-a
Proof Again it is instructive to sketch a diagram (see Fig. 5.4). The quantity
[f(b) - f(a)]/(b - a) measures the gradient of the chord AB. The theorem

175
GUIDE TO ANALYSIS

B
v

f(b) -f(a)
I
I
I
I
I
I
I
I I
I I

--------T
I I

-----1------------ - - - --
I b-a I
I I
I I
I I
I I
I I
o a c b x

Fig. 5.4
asserts that there is a point C between A and B with the property that the
tangent at C is parallel to the chord AB, which certainly looks a reasonable
result. Indeed it can be proved rigorously.
Define a new function g: [a, b] -+ IR by the relation
g(x) = f(x) - kx, (1)
where the number k is chosen so that g(a) = g(b)
i.e. f(a) - ka = f(b) - kb

k =f(b) - f(a).
which gives (2)
b-a
Then g satisfies all the necessary conditions for the validity of Rolle's theorem.
For g is continuous on [a,b] and g is differentiable on (a, b). Further, the
relation g(a) = g(b) is satisfied when k is given by (2). Hence there is a number
c such that a < c < band g'(c) = 0 by Rolle's theorem. Now
g'(x) = f'(x) - k,
and therefore 0= g'(c) = f'(c) - k

i.e. f'(c) = k =f(b) -f(a)

b-a
by equation (2). This completes the proof.

176
 DIFFERENTIABLE FUNCTIONS

Some of the applications of this result are fairly obvious. For example,
whenever some estimate is required comparing the sizes of [f(b) - f(a)] and
(b - a), the mean-value theorem is an obvious candidate. It is also used to
justify the addition of an arbitrary constant when finding indefinite integrals
as we will now see.

Examples 5.2.2

1. Suppose the function f is continuous on the closed interval [a, b] and

differentiable on the open interval (a, b). Suppose also that f'(x) = 0 for
all x such that a < x < b. Then we can use the mean value theorem to
prove that f is a constant on [a, b].
Suppose Xl is any number such that a < Xl ~ b. Then f is continuous
on [a, Xl] and differentiable on (a, x d. By the mean value theorem, there
is some c with a < c < Xl such that
f(Xl) - f(a) = (Xl - a)f'(c).
Now f'(x) = 0 for all X such that a < x < b, and thereforef'(c) = O. Hence
f(xd - f(a) = 0,
i.e. f(x l ) = f(a). Since this is true for all Xl with a < Xl ~ b, we see that f
is constant on [a, b].
2. Suppose the functions f, 9 are continuous on [a, b] and differentiable on
(a,b). Suppose also that
f'(X) = g'(x) (1)
for all X such that a < X < b. Then the function f - 9 is continuous on
[a, b] and differentiable on (a, b). Moreover,J'(x) - g'(x) = 0 for all x such
that a < x < b. Using the previous example we see that f - 9 is constant
on [a, b] i.e. for all xE[a, b ],J(x) = g(x) + constant, which explains why
we add an arbitrary constant when we are finding indefinite integrals.
3. Let the function f be continuous on [a,b] and differentiable on (a,b).
Suppose that f'(x) > 0, for all x such that a < x < b. Then we can prove
that f is strictly increasing on [a, b].
Suppose Xl' x 2 are any two numbers such that a ~ Xl < X2 ~ b. Then f
is continuous on [X l ,X 2 ] and differentiable on (Xl,X2). By the mean value
theorem, there is a number c with Xl < c < x 2 , such that
(1)
Now f'(c) > 0 and, therefore, equation (1) gives f(X2) - f(x l ) > 0, i.e.
f(x 2) > f(xd. Since this is true for all Xl' x 2 such that a ~ Xl < x 2 ~ b,f is
strictly increasing on [a, b].

177
GUIDE TO ANALYSIS

Finally we work an example where the solution depends on a combi-

nation of Rolle's theorem and the mean value theorem. In future we will
use MVT as an abbreviation for the mean value theorem.
4. The function f satisfies all the following conditions:
(i) f is continuous on the closed interval [a,b];
(ii) f is twice differentiable on the open interval (a,b);
(iii)f"(x) > 0 for all x such that a < x < b;
(iv)f(a) = f(b) = o.
Show that f(x) ~ 0 for all xE[a, b].

Solution This is a fairly complicated problem and the solution is not immediately
obvious. The reader may find it helpful to begin by trying to sketch a possible
graph for J, incorporating properties (i) to (iv). Usually, very few students are lucky
enough to produce a suitable picture at the first attempt. Several modifications
may be needed before the sketch shows a function with all the required properties.
We notice that f"(x) > 0 for a < x < b and so f' is strictly increasing (see the
previous example), i.e. the gradient of the tangent increases. The graph, therefore,
must be drawn so that f(a) = f(b) = 0 and the gradient of the tangent increases
as x increases. After a few trials the reader should come up with a picture similar
to Fig. 5.5. Pictures cannot be used to prove results in analysis, but they can be
used to show us the way to construct a proof. In this case the diagram seems to
indicate that there is a point between a and b at which the tangent is parallel to
the x-axis. Rolle's theorem shows us that this is indeed true.
Since f"(x) exists for a < x < b, the first-order derivative f' is differentiable on
(a,b) and so f is differentiable on (a, b). However, f also satisfies (i) and (iv). By
Rolle's theorem, therefore, there is a number c with a < c < b such that
f'(c) = o. (1)
Now let a < Xl < X 2 < b. Then!, is continuous on [XI,X2] andf' is differentiable
on(xl,x2). By the MVT(mean value theorem) there is somed,such that Xl < d < X2
and

o~ __________ ~a ____________~C____________Tb____________~
x

Fig. 5.5

178
 DIFFERENTIABLE FUNCTIONS

by (iii). Thus f'(X2) > f'(Xl) and, therefore, f' is strictly increasing on (a, b). In
particular,
f'(x) < 0 for a < x < C, (2)
f'(x»O for c<x<b, (3)
by ( 1). (The geometric interpretation of relation (2) is that the tangent has a negative
gradient between a and C and so the graph goes down, which the picture led us
to expect. Relation (3) tells us that the graph goes up between c and b, and this
verifies that our sketch is not unreasonable. A further' glance at the diagram also
suggests that an appeal to the MVT may complete the solution.)
Again, using the MVT we can show thatf(x) ~ 0 for a ~ x ~ C and for C ~ x ~ b.
First let x be any number such that a < x ~ c. Then f is continuous on [a, x]. By
the MVT there is some number C1 with a < C1 < X ~ C such that
f(x) - f(a) = (x - a)f'(c 1 ) < 0
by (2). Hence f(x) < f(a) = 0 for a < x ~ c. Similarly, if C ~ x < b, we can show that
f(x) < f(b) = 0 by applying the MVT to f on the interval [x,b] and using (3). The
results together give f(x) ~ 0 for a ~ x ~ b.

This long and difficult last example may have taught the reader not to despise
diagrams. They certainly do not constitute a proof. They may, however, give
guidance on how to set about constructing a proof and they may inspire the
reader with sufficient courage to dare to tackle seemingly intractable
problems.

EXERCISES 5.2.1
1 Prove that the equation x 3 - 4x 2 + cos X = 0 has one and only one solution
between 0 and 1.
2 Prove that if 0 < a < b < n/2, then
(b - a)cosb < sinb - sin a < (b- a)cosa.
3 Prove that if 0 < a < b then

(~ - ~
1 ) > log > ( 1 - ~ ).
4 By applying the MVT to sin -1 x on the interval [0.5, 0.6] show that

n .j3 . -1 1l 1
6"+15<sm (0.6)<6"+g

(Use the fact that sin n/6 = t i.e. 1l/6 = sin -1 t.)
5 The function f satisfies all the following conditions:
(i) f is continuous on the closed interval [a,b],
(ii) f is twice differentiable on the open interval (a, b);

179
GUIDE TO ANALYSIS

(iii)f"(x) < 0 for all x such that a < x < b;

(iv)f(a) = f(b) = O.
Show that f(x) ~ 0 for all xE[a, b].
6 Show that if 0 < a < b < 1 then
b-a . -lb . -1 b-a
~<sm -sm a< ~.
V l-a 2 V I-b 2
7 The function f: IR -+ IR is differentiable at all points of IR, and f(O) > 0 and
f'(x) < t for all x such that x ~ O. By considering the function given by
g(x) = f(x) - x, show there is one and only one positive real number c
such that f(c) = c.

5.3 TAYLOR'S THEOREM

The conclusions of the MVT can be presented as
f(b) = f(a) + (b - a)f'(c). (1)

This relation suggests that it might be possible to extend the result to include
higher-order derivatives. Under suitable conditions, the corresponding formula
would appear to be
f"(a)
f(b) = f(a) + f'(a)(b - a) + 2!(b - a)2 + ...

+
f (n-1) (b_a)"-l+~_c
pn)( )
(b-a)",
(n - I)! n!

where a < c < b. This result is sometimes known as the nth mean value
theorem or it is called Taylor's theorem. Many proofs exist in the literature
and some of them are amazingly slick and somewhat bewildering at first
sight. We will give one of the longer proofs which has the merit of being
relatively easy to understand.

THEOREM 5.3.1 (Taylor's Theorem; nth MVT)

Suppose that f possesses derivatives of all orders up to and including the
(n-l)th-order derivative on the closed interval [a,b]. If

(i) f and its first (n -1) derivatives are continuous on the closed interval
[a,b];
(ii) f has an nth-order derivative on the open interval (a,b);

then there is some number c such that a < c < band

180
 DIFFERENTIABLE FUNCTIONS

f(b) = f(a) + f'(a)(b - a) + f~\a) (b - a)2 + ...

+
f (n -1)( a) (b_a)n-1 +--C-(b-a)R.
pn)( )
(n -1)! n!
Proof We proceed in a similar fashion to the proof of the MVT. First
construct a suitable function g and then apply Rolle's theorem. In this case
Rolle's theorem has to be used repeatedly.
Define the function g by

g(x) = f(x) - f(a) - f'(a)(x - a) - f~(~) (x - a)2 - ...

pn-1)(a)
- (n-1)! (x-a)n-1-k(x-a)n,

n-1 p')(a)
= f(x) - f(a) - L --(x -
,=1 r!
a)' - k(x - a)n, (1)

where the number k is chosen so that g(b) = g(a). The value of k is therefore
n-1 p')(a)
f(b)- f(a)- L --(b-a)'
k= ,=1 r! (2)
(b - a)n
Now the function g is continuous on [a,b] and differentiable on (a,b) and
g(a) = g(b). By Rolle's theorem there is some number C l with a < Cl < b such
that
g'(cd=O. (3)
Now differentiating (1) we have
pn-1)(a)
g'(x) = f'(x) - f'(a) - f"(a)(x - a) - ... - (n _ 2)! (x - a)n-2 - kn(x - a)n-l,

and therefore g'(a) = o. Assuming that n ~ 2, we see that the function g' is
continuous on [a,c1] and it is differentiable on (a,c l ), and g'(a) = g'(c 1) = O.
By Rolle's theorem, there is a number C2 such that a < C2 < C1 < band
g"(C2)=O. (4)
We can check, by differentiation, that
g(k)(a)=O (1 ~ k~n-1).

In particular, if n ~ 3, then g"(a) = 0 = g"(C2) and applying Rolle's theorem

to g" on [a, C2] we obtain a number C3 such that a < C3 < C2 < C1 < band
g"'(C3) = o. Continue in this way, and repeat the process a further n - 3 times,
to obtain a number Cn such that
a < Cn < Cn -l < ... < C3 < C2 < C1 < b

181
GUIDE TO ANALYSIS

and g(n)(Cn) = o. (5)

From (1), g(n)(x) = pn)(x) - kn! (a < x < b),
and therefore equation (5) gives (writing C in place of cn)
pn)(c) = n!k.
When this is combined with equation (2), we obtain the required result, and
the proof is complete.

There are dozens of practical applications of Taylor's theorem as we will see

a little later. Some of them will use Taylor's theorem in precisely the form
in which it is stated in Theorem 5.3.1. Others use very slight variants of it.
For example, Theorem 5.3.1 is stated for the case a < b. With a very slight
modification of the proof and an appropriate change to the intervals, it
remains true when b < a. In this case the result is stated in the following way.

THEOREM 5.3.1(a) Let a be two real numbers with b < a. Suppose I

possesses derivatives of all orders up to and including the (n -1)th order
derivative on the closed interval [b,a]. If
(a) I and its first (n - 1) derivatives are continuous on the closed interval
[b,a], and
(b) I has an nth order derivative on the open interval (b, a),
then there is some number c such that b < c < a and

f(b) = I(a) + f'(a)(b - a) + I~\a) (b - a)2 + ...

I (n -1)( ) I(n)( )
+ a (b_a)n-1+ _ _ C (b-a)n.
(n - 1)! n!
The proof uses exactly the same function g as before. In this case the only
slight modification is that we apply Rolle's theorem to intervals [b, a], [c 1> a],
[c2,a], etc. instead of the intervals [a,b], [a,c 1], [a,c2l

Our first application, commonly known as I'Hopital's rule (or I'Hospital's

rule) is named after a French mathematician, the Marquis G. F. A. de
I'Hopital (1661-1704). The spelling is somewhat doubtful-sometimes the
letter's' is included, at other times the's' is omitted and a circumflex placed
over the letter '0'. There is, however, little doubt as to the identity of the
discoverer ofthe rule, and it wasn't the Marquis! For many years it has been
known that the result was originally discovered by the Swiss mathematician
Jean Bemouilli (1667-1748), a member ofthe famous Bemouilli family which
produced a remarkable number of celebrated mathematians over the years.
The story begins in 1692, when the two men met in Paris. Bemouilli had

182
 DIFFERENTIABLE FUNCTIONS

already spent some time studying Leibniz's work on calculus and he taught
it to I'H6pital. But what was more significant is that they signed an agreement.
According to this contract, the Marquis guaranteed to pay Bernouilli a regular
salary and Bernouilli undertook to send all his mathematical discoveries to
I'H6pital. The Marquis was then free to use them in any way he pleased.
In 1696, the Marquis de l'H6pital published a text book on calculus. It
included material which had been communicated by Bernouilli and among
this material was the law which we now know as l'H6pital's rule. The book
was well written and its author proved to be a good teacher. The preface
includes an acknowledgement of the importance of the work of Leibniz and
Bernouilli. To this day, however, the Marquis still gets the credit, in name
at least, for the following law.

5.4 L'HOPITAL'S RULE

In Chapter 4 we saw that if f(x)-+ 1and g(x)-+m as x -+a, thenf(x)/g(x)-l/m
as x - a provided m "# 0 and g(x) "# 0 sufficiently close to a. If 1= 0 and m = 0,
then the algebra of limits is not applicable. This is just the case in which
I'H6pital's rule may be applicable.

THEOREM 5.4.1 (L'H6pital's Rule) Suppose there is some positive real

number R such that the functions f and g and their first n derivatives are
continuous on the open interval (a - R, a + R), where n ~ 1. Suppose also that
(i) f(a) = O,f(k)(a) = 0 for k~ n-l,
(ii) g(a) = 0, g(k)(a) = 0 for n-l,

-----
k~
(iii) g(n)(a)"# O.
f(x) f(n)(a)
Then
g(x) g(n)(a)
as x-a.

Proof Since gIn) is continuous on (a - R, a + R) and g(n)(a)"# 0, it follows

from Theorem 4.2.1 that there is some r > 0 (with r ~ R) such that g(n)(t)"# 0
for all t such that a - r < t < a + r.
Now let x be any real number such that 0 < Ix - al < r, thenf,g and their
first n derivatives are continuous on the closed interval with end-points a, x.
By Taylor's theorem,
j"-l(a) pn)(c)
f(x) = f(a) + f'(a)(x - a) + ... + ( _ 1)' (x - a)"-l + -,-(x - a)n
n. n.
f(n)(c) n
= - - ( x - a) (1)
n!

183
GUIDE TO ANALYSIS

by condition (i), where c lies between a and x. Similarly,

, g(n-l)(a) n-l g(n)(d) n
g(x)=g(a)+g(a)(x-a)+ ... + ( -1)1 (x-a) +-,-(x-a)
n. n.
g(n)(d)
=- - ( x - a)n (2)
n!
by condition (ii), where d lies between a and x. Hence 0 < la - dl < r, because
O<la-xl<r. Since g(n)(t)-#O for all t such that a-r<t<a+r, we see
g(n)(d) -# 0 and so g(x) -# 0 provided 0 < Ix - al < r. Thus, if 0 < Ix - al < r,
then
f(x) pn)(c)
(3)
g(x) g(n)(d)
from (1) and (2). As x ..... a, we must have c ..... a and d ..... a, because c,d are
between a and x. Moreover, pn)(c) ..... pn)(a) as c--.a and g(n)(d)--.g(n)(a) as
d ..... a, because pn), g(n) are continuous. Hence
j<n)(c) j<n)(a)
- ---.--
g(n)(d) g(n)(a)
as x --. a, and therefore, from (3),
f(x) pn)(a)
----.--
g(x) g(n)(a)
as x --. a. This completes the proof.

Taylor's theorem states that a number c exists with given properties when
f satisfies certain conditions. The number c is determined by the function f
and the interval [a, b]; it is not just an arbitrary number between a and b.
It was therefore necessary in the above proof to use two different symbols
c, d for the intermediate points-one for the function f and the other for g.
Now let us test how this rule works in practice by trying an example.

Examples 5.4.1

Evaluate the following limits:

e x2 -1 · cos x -1
2. I1m · (x-l)3
3. I1m 4. lI ·msin x
1. lim -.-2-. 2 . --.
x-+o sIn x x-+o X x.... 1 logx x-+ocoshx

Solutions
1. We notice that if
f(x) = ex2 -1, g(x) = sinx 2

184
 DIFFERENTIABLE FUNCTIONS

for all real x, then f(O) = 0, g(O) = O. So we have an obvious candidate for a
trial of I'Hopital's rule. Obviously, we must check that all the conditions
are satisfied before applying it. We must, therefore, decide what value to use for
n and we must check that there is some suitable R. How do we find n? Theorem
5.4.1 says that n is the smallest positive integer for which the nth-order derivative
of 9 is non-zero. We must, therefore, find some derivatives of g. Now g'(x) = 2x cos x 2
and g'(O) = 0,
g"(x) = -4x 2 sinx 2 + 2 cos x 2 and g"(O) = 2 #0.
The first non-zero derivative at the origin is the second and we therefore use n = 2.
Naturally, we still have to check the conditions on f We notice that
f'(x) = 2xe x2 and 1'(0) = 0,
f"(x) = 4x 2 ex2 + 2ex2•
Moreover, J,g and all their derivatives are continuous at all points of IR, and we
may therefore choose any positive value for R. Suppose we choose R = 1. Clearly
J,g,f',g',f" and g" are continuous on (-1, +1). Further,f(O) =1'(0) =0, and
g(O) = g'(O) = 0, g"(O) # O. By I'Hopital's rule

lim f(x) =/,,(0) =~= 1

x-+Og(x) g"(O) 2 '

ex2 -1
i.e. lim - ' - - 2 = 1.
x-+O SlDX

For most undergraduates, one ofthe main problems is what value to use for n.
Students expect somehow to look at the question and immediately know which
value to choose, but in practice this is impossible. The functions f and 9 are usually
fairly obvious. After that it is a question of finding the derivatives and evaluating
them at the appropriate point. The integer n is the order of the lowest-order
non-zero derivative of g. The reader should experience no difficulty whatsoever
in finding n, when this method is followed. Finally, a word ofwaming. Don't use
I'Hopital's rule if it is not appropriate. If f(a) # 0, g(a) # 0 then I'Hopital's rule is
not appropriate and should not be used.
2. We follow the same method as before.
Let f(x) = cos x - 1, g(x) = x 2
for all x. Then g'(x)=2x and g'(O) = 0,
g"(x) = 2 and g"(O) = 2#0
and we again have a problem for which n = 2. Moreover,
f'(x) = -sinx and 1'(0)=0,
f"(x) = -cosx.
In this case J, 9 and all their derivatives are continuous at every point of IR and
we can choose any positive value for R. For example, we can use R = 1, since
J,g,/',g',f" and g" are all continuous on (-1,1). Hence, using I'Hopital's rule,
. cos x -1
I1m 2
,,-0 X

3. By now, the reader is sufficiently familiar with the method to be able to write
down the solution quite quickly. There are, however, one or two details which

185
GUIDE TO ANALYSIS

require just a little care. The choice of the functions f, g is fairly obvious. Let
f(x) = (x _1)3 (all xelR)
g(x) = logx (x> 0)
Then f(l) = 0, g(l) = 0
g'(x) = l/x (x> 0),
and g'(I) = 1.
This problem seems an obvious candidate for the use ofl'Hopital's rule with n = 1
for this particular example. Since we require a limit as x tends to 1, the continuity
conditions must be checked on an interval about 1. In this case the interval (0,2)
will do. Clearly,f,g,/"g' are continuous on (0,2). By I'Hopital's rule,
lim (x-l)3 =/,(1) =~=O.
x-+! logx g'(l) 1
In this example n = 1 and /,(1) = O. This causes no problem, as the conditions for
I'Hopital's rule require yCn)(a) #0; they do not stipulate thatpn)(a) is non-zero.
One final word of warning about this example-the question asks for the limit
as x tends to 1 and so f, g and their derivatives are evaluated at 1. When the
question asks for a limit as x tends to a non-zero value, then answer the question
you were asked-don't evaluate the functions at zero and then wonder why your
answer is wrong.
4. Use f(x) = sin x,
g(x) = cosh x
for all real x. Then f(O) = 0, g(O) = 1 # 0, and this is a case in which l'Hopital's
rule is Dot appropriate. In fact, we can see immediately that f(x) ..... O as x ..... O,
g(x) ..... 1 as x ..... O and so
f(x) 0
- - ..... -
g(x) 1
as x ..... 0 by the algebra of limits.

EXERCISES 5.4.1
Decide which of the following limits exist. When the limit exists find its value.

· Iogx 2 •
2 I1m (x 2 _1)2
1 1Im--, . ,
x-+! 1- x x.... ! l - sm(nx/2)

3
rtm-.-,
cosx r cosx
4 Im--
x-+J sIn x x.... !1-x'
r (1_X)2 rtm (X-1)3 ,
5 1m . 2 ' 6
x-+ J sm (nx) x-+! logx

7
r 1-log(e+x2)
8
rIm-
log x
- -,
1m (eX - 1) Slnx
. , 2 1
x-+O x-+!x -

186
 DIFFERENTIABLE FUNCTIONS

x-2 1- cosh x
9 lim~, 10 lim Xl
,
x-+2X - x-+o
. h x1
rlm-·-
sm r sin 1tX
11 -, 12 1m--
x-+l logx '
l
x-+o SIn x
r log(cosx) X3 _X 2 _X-2
13 1m 2 ' 14 lim 3 2 '
x-+o X x-+2X -3x +3x-2
enx_e- nx r1m-.-.
cosh x
15 lim , 16
x-+olog(1 + 1tx) x-+n/2 smx

(Don't use I'Hopital's rule when it is not appropriate. You have been
warned!)

5.5 MAXIMA AND MINIMA

Most sixth-formers who study mathematics spend at least some time working
on problems on maxima and minima. Given a reasonably well-behaved
function f such that f'(a) = 0, f"(a) > 0, the conclusion is that f has a
minimum at a. If f'(a) = 0 and f"(a) < 0, then there is a maximum at a. But
what happens if f'(a) = 0 andf"(a) = O? Among sixth-formers brave enough
to hazard any guess at all many jump to the conclusion that f has a point
of inflection at a. Is this really true? Perhaps we were a little hasty, for if
f(x) = (x - a)4, thenf'(a) = O,f"(a) = o. However,f(x) = (x - a)4 > 0 = f(a)
for all x such that x # a and so it would appear that f has a minimum at a
despite the fact that f"(a) = o. This suggests that it might be profitable to
look into the subject in a little more detail. Perhaps we should begin by
spelling out carefully the meaning of the terms we intend to use. First of all
we introduce the terms local maximum and local minimum.

DEFINITION 5.5.1 Suppose there is some R > 0 such that f(x) is defined
and satisfies the inequality f(x) :::;; f(a) for all x such that a - R < x < a + R.
Then f is said to have a local maximum at a.

DEFINITION 5.5.2 Suppose there is some R > 0 such that f(x) is defined
and satisfies the inequality f(x) ~ f(a) for all x such that a - R < x < a + R.
Then f is said to have a local minimum at a.

If f has a local maximum at a, thenf(a) is the greatest value assumed by f(x)

for a - R < x < a + R for some R > 0, i.e. f(a) is the greatest value assumed
by f(x) for x close to a. For this reason we call it a local maximum. There
may, of course, be points outside the interval (a - R, a + R) at which f(x)

187
GUIDE TO ANALYSIS

assumes values greater than f(a). In contrast, we have the notion of a global
maximum. Let f: [a, b] -+ IR have the property that there is some ce[a, b]
such that f(x) ~ f(c) for all xe[a, b]. Then f(c) is the largest value assumed
by f(x) on [a, b]. Thus f(c) is a global maximum rather than a mere local
phenomenon. Similarly, if de [a, b] is such that f(x) ~ f(d) for all xe[a, b],
then f(d) is the smallest value assumed by f(x) on [a, b] and it is, therefore,
a global minimum. Now let us begin the search for the conditions needed
for a maximum or minimum.

THEOREM 5.5.l Suppose f(x) is defined for all x such that a - R < x < a + R,
where R is a given positive real number. If f has a local maximum
(or minimum) at a and if f'(a) exists, then
f'(a) = O.

Proof Since f has a local maximum at a, there is some r > 0 (with r ~ R)

such thatf(x)~f(a) for all x such that a-r<x<a+r.
Thus if a - r < x < a then
f(x) - f(a) ~ O.
(1)
x-a
Now, f'(a) exists and so

f'(a) = lim f(x) - f(a) ~ 0 (2)

x-+a- X - a
by relation (1). Similarly, if a < x < a + r, then
f(x) - f(a) 0
'----'---'--~
x-a

and f'(a) = lim f(x) - f(a) ~ O. (3)

x-+a+ X - a

From (2) and (3) we see that f'(a) = O.

A similar proof shows that if f has a local minimum at a, then f'(a) = O.
This completes the proof.

DEFINITION 5.5.3 Suppose that f(x) is defined for all x such that
a - R < x < a + R, for some positive real number R, and suppose also that
f'(a) exists. If f'(a) = 0 then a is called a stationary point or critical point.

Now, how do we set about distinguishing the characteristics of our stationary

points? Probably our wisest course of action is to use Taylor's theorem.

188
 DIFFERENTIABLE FUNCTIONS

Suppose f and all its first n derivatives (n ~ 2) are continuous on some

interval (a - R, a + R), where R is some given positive real number. Suppose
also that f'(a) = 0 so that a is a stationary point of f If pk)(a) = 0 for
1::::; k::::; n - 1 and p")(a) #- 0, then Taylor's theorem gives us
p"-l)(a) P")(c)
f(x) = f(a) + f'(a)(x - a) + ... + (n _1)! (x - a)"-l + ---;;!(x - a)"

= f(a) +P")~C)(x - a)"

n.
for 0 < Ix - al < R, where c lies between a and x. Now let us look at some
separate cases.
(i) Suppose n is even (n ~ 2) and p")(a) > o. Since p") is continuous on
(a - R, a + R) there is some r > 0 (with r::::; R) such that P")(t) > 0 for
all t such that a - r < t < a + r. Thus for all x with 0 < Ix - al < r, we
have
P")(c)
f(x) - f(a) = -,-(x - a)" > O.
n.
Since p")(c) > 0 because c lies between a and x. Hence
f(x) > f(a)
for all x such that 0 <Ix - al < rand f has a local minimum at a.
(ii) Suppose n is even (n ~ 2) and p")(a) < o. As before, the continuity of
P") means that there is some r > 0 (with r::::; R) such thatp")(t) < 0 for
all t such that a-r<t<a+r. Thus, for all x with O<lx-al<r,
we have
P")(c)
f(x) - f(a) = -,-(x - a)" < 0,
n.
i.e. f(x) <f(a)
and f has a local maximum at a.
(iii) Suppose n is odd (n ~ 2) and p")(a) > O. We use the same r as in (i).
Then for a<x<a+r,
P")(c)
f(x) - f(a) = -,-(x - a)";> 0,
n.
and for a - r < x < a,
P")(c)
f(x) - f(a) = -,-(x - a)" < 0,
n.
since P")(c) > 0 because c lies between a and x. Thus
f(x) > f(a) (a < x < a + r)
f(x) < f(a) (a - r < x < a)

189
GUIDE TO ANALYSIS

and I has neither a local maximum nor a local minimum at a. Such

a point is called a point of inflection.
(iv) Suppose n is odd (n ~ 2) and pn)(a) < O. We use the same r > 0 as in
(ii). If we follow the method used in (iii) then we see that
I(x) < I(a) (a<x <a +r),
I(x) > I(a) (a-r<x < a).
Again there is neither a local maximum nor a local minimum at a,
and I has a point of inflection at a.

Let us gather together these results and incorporate them in a theorem.

THEOREM 5.5.2 Suppose I and its first n derivatives (n ~ 2) are continuous

on the interval (a - R, a + R), where R is some given positive real number.
Suppose also that p")(a) = 0 (1 ~ k ~ n - 1) and pn)(a) =F O. If
(i) pn)(a) > 0 and n is an even integer, then I has a local minimum at a;
(ii) pn)(a) < 0 and n is an even integer, then I has a local maximum at a;
(iii) n is an odd integer, then I has a point of inflection at a.
The special case n = 2 is the one normally encountered in the sixth form.
From our theorem we see that if f'(a) = 0, f"(a) > 0, then we have a local
minimum. However, if f'(a) = 0 and f" (a) < 0 then we have a local maximum.
These results are, of course, very familiar already. However, with a well-
behaved function the case f'(a) = 0, I"(a) = 0 causes no problem, now that
our general theorem is available. All we have to do in such cases is to keep
on differentiating until we reach a derivative which is non-zero at a. Let us
look at some particular examples.

Examples 5.5.1

1. Let f(x) = (x - a)4. Then I and all its derivatives are continuous at every
point of IR and so any positive number can be used for R. Moreover,
f'(a) = 0, f"(a) = 0, f"'(a) = 0, j<4)(a) = 4! This function satisfies all the
conditions ofthe theorem with n = 4 and pn)(a) = 4! = 24 > O. Thus I has
a local minimum at a (see Theorem S.S.2(i», which is precisely what we
conjectured earlier.
2. Let I( x) = sin x-x. Then I and all its derivatives are continuous at every
point of IR. Now,
f'(x) = cos x - 1,
f"(x) = -sin x,
1111 (x) = -cosx,
190
 DIFFERENTIABLE FUNCTIONS

and, for every integer k,

f'(2kn) = 0, f"(2kn) = 0, f"'(2kn) = -1.
We therefore need to use Theorem 5.5.2(iii). Since the appropriate value
for n is n = 3, this shows us that f has a point of inflection at each of the
points 2kn (k = 0, ± 1, ± 2, ± 3, ... ). In this example, all the stationary
points are points of inflection.
Suppose now that we have a function f which is continuous on a closed
interval [a, b]. Then Theorem 4.3.1 and 4.3.2 tell us that f( x) has a largest
value M on [a, b]. How do we set about finding this largest value? There
are two possibilities: either f(a) = M or f(b) = M or there is a point c
with a < c < b such that f(c) = M. If f(c) = M, where a < c < b, then
f(x) ~ f(c) = M for all x such that c - R < x < c + R where R is the smaller
of the two numbers c - a, b - c. In this case, therefore, f has a local
maximum at c. We already know that, if f'(c) exists and f has a local
maximum at c, then f'(c) = o. A similar result holds for the minimum
value of f(x) on [a,b].
3. The function f: [0,5] --+ IR is given by
f(x) = x3 - 5x 2 + 3x + 23.
Find the largest and smallest values assumed by f(x) on [0,5].
Solution This function f is differentiable on (0,5). Hence the largest value of f(x)
will occur either at one of the end-points or at a point in between at which there
is a local maximum, i.e. it will either occur at an end-point or at an intermediate
point at which the derivative vanishes. A similar result holds for the smallest value.
We therefore need only find the points e with 0 < e < 5 at which f'(e) = 0 and
evaluate f(x) at such points e and at the end-points 0 and 5.
Now, f'(x) = 3x 2 -lOx + 3 = (3x -l)(x - 3) = 0,
if x = t 3. It is easy to check that frO) = 23, fm = 23 nJ(3) = 14, f(5) = 48. We
can now simply read off the result. The largest value of f(x) on the interval is 48,
the smallest value is 14.

Finally we use Taylor's theorem to develop power series expansions.

5.5.1 Taylor series

Let a, x be two distinct real numbers and let J be the closed interval with
end-points a,x. Suppose that the functionf possesses continuous derivatives
of all orders on J. Then by Taylor's theorem we have, for all n ~ 1,
pn-1)(a) f(n)(c)
f(x) = f(a) + f'(a)(x - a) + ... + (n _ 1)! (x - a)n-1 + -----;;!(x - a)"

n-1 pn)(c)
= I ak(x -
k=O
a)k + - - ( x - a)",
n!
where a o =f(a), ak =pk)(a)/k! (k= 1, ... ,n -1) and c lies between a and x.

191
GUIDE TO ANALYSIS

ft-1 Pft}(c)
Thus f(x) - L aA:(x - a)A: = -,-(x - a)n;
k=O n.
and, if we can show (for a particular value of x) that
j<ft}(c) n
--(x-a)
n!
tends to zero as n-+ 00, then (for this value of x) the infinite series
Lk=Oat(x - a)k converges and its sum is f(x). We can then write
<Xl
f(x) = L ak(x - a)k
t=o
where ao = f(a), aA: = j<1c}(a)/k! (k ~ 1). This series is called the Taylor series
of f(x) about x = a. Let us try a few examples.

Examples 5.5.2

1. Let f(x) = eX for all x. Then f and all its derivatives are continuous at all
points of R Let x be any non-zero real number and apply Taylor's theorem
on the closed interval J with end-points 0 and x to obtain
f"(0) 2 j<n-1}(0) n-1 j<ft}(C)
= f(O) + f (O)X + ~X + ... + (n _ I)! x
I

f(x) + -----n!x"' (1)

where c lies between 0 and x. Now we can easily check that

f(0) = 1,
for all positive integers k. Hence (1) reduces to
x2 x3 x n- 1 eC
.+ ... + (n
f(x) = 1 +x+-' +-3' -1)' (2)
2. . +,x".
n.
Now c lies between 0 and x and therefore
ecxnl elxllxlft
l -~---
n! n!
For any given fixed real number x,
e1x1lxlft
----+0
n!
as n -+ 00. Thus for all real numbers x
x2 x3 <Xl x"
2 ' +-3' + ...
f(x)=e x = 1 +x+-
..
= n=on.
L ,. (3)

This expansion can be used to obtain a rather neat limit.

192
 DIFFERENTIABLE FUNCTIONS

Let a. be any real number and choose a positive integer q such that
q> a.. Then, for x> 1, we have
(4)
Now, as x ---+ 00, x q ---+ 00 and eX -+ 00. What happens to the quotient x qleX?
Relation (3) can be used to give an answer. For, if x > 1, then
eX > x q + 1I( q + 1)!
from (3) and combining this with (4) yields the inequality
xlt x q xq(q + 1)! (q + 1)!
0<-<-<
eX eX xq + 1
=---
X

i.e. (x> 1).

As x---+oo, (q+ l)!/x---+O, and therefore, using the sandwich rule,

(5)

It is therefore clear that as far as this limit is concerned, the exponential

dominates the power. The reader will do well to make an effort to memorise
the result. The corresponding rule for logarithms is that for all p > 0

logy
-P- ---+ 0 as y ---+ 00.
Y

This can be quite easily deduced from (5). For using a. = 1 we see that
xe-X---+O as x---+ 00. Now write x = pt (P > 0). As x-+ 00, t-+ 00 and
pte - PI ---+ 0
and therefore
as t ---+ 00. Now write t = log y, so that e' = y and the relation becomes

log y ---+ 0
yP
as y---+oo, for P>O.
2. Let f(x) = 10g(1 + x) (x> -1), and use a = 0 in Taylor's theorem. First
let x be any real number such that 0 < Ix I < 1. Then f and its derivatives
of all orders are continuous on the closed interval with end-points 0 and
x. Using Taylor's theorem,

f(x) = f(O) + f'(O)x + f"(0) x2 + + pn-l)(O) xn-1 + pn)(c) xn

2! ... (n - 1)! n!'

193
GUIDE TO ANALYSIS

where c lies between 0 and x. By differentiating, we find that

(k=I,2,3, ... )

and therefore

Thus (1)

where c lies between 0 and x. Now, if 0 < x < 1 then

f(")(C) I I I
I-----n!x" I)!
(n -
= (1 + c)"n!x" = (1 + c)"n'
x"

x"
~­
n
because 0 < c < x < 1 and x"In -+ 0 as n -+ 00. Hence if 0 < x < 1 then
x2 x3 x"
f(x) = 10g(1 + x) = x --2 +-3 ... = L (_1)"-1_,
00
(2)
"=1 n
which is the familiar expansion of log( 1 + x). The same reasoning could
be applied when x = 1. In this case we observe that
x" 1 1
---= <-
(I + c)"n (1 + c)"n n
where 0 < c < x = 1. Thus the remainder tends to zero as n-+ 00, and the
relation (2) is true for x = 1. This is hardly surprising in view of the
conclusion of Example 4 in the section following Theorem 3.3.5(a).
If, however, - 1 < x < 0, then the remainder term is
f--x"=(_1)"-1
(")(c) x"
,
n! (1 + c)"n
and we do not know what happens to this as n -+ 00 because the relation
-1 < x < C < 0 does not give us sufficient information about [xl(1 + c)]n.
It can actually be proved that relation (2) is also true when - 1 < x < 0,
but the method we used above will not provide that proof. Moreover, if
Ix I > 1, then xnIn -f+ 0 as n -+ 00 and the series on the right-hand side of (2)
does not converge for any value of x for which Ix I > 1.
Taylor's theorem can be used extensively to develop power series expan-
sions for suitable functions. The reader is warned, however, to treat series
expansions with care. Don't just take it for granted that the series
f"(a)
f(a) + f'(a)(x - a) + 2"!(x - a)2 + ...

will have the same value as f(x). There are functions J, for which this series

194
 DIFFERENTIABLE FUNCTIONS

converges, but its sum is notf(x)-so be warned! The only safe way is to use
Taylor's theorem correctly and check what happens to the remainder term.
As a warning to the reader of what can happen we now include a further
example.
3. Let f: ~ -+ ~ be defined by
-1/x2
(x # 0),
f(x)=
{
~ (x =0).

Thenf(x)-+ f(O) as x-+o andf is continuous at the origin. Using the rule
for composite functions, we see that f is continuous at all other points of
R Now, if x #0 then
f_(_x)_-_f,-(O_) =.!.e- 1/ X2

x-O x

and 1 -1/x2 -+ 0
-e
x
as x -+ O. Hence f is differentiable at the origin and I' (0) = O. Moreover, if
x # 0, thenf'(x) = (2/x3)e- 1/x2 using the rule for differentiating composite
functions. Therefore if x # 0 then
f'(x) - 1'(0) _ 2 -1/;x2
x-o -x4e ,
and this tends to zero as x -+ O. Hence f" (0) exists and f" (0) = O.
Continuing in this way we can show that j<k)(O) = 0 (k = 3,4, ... ). Thus the
series
, f"(0) 2
f(O) +f (O)x + 2 ! x + ...
consists entirely of zeros and its sum is zero for all real numbers x.
Hence the series
, f"(0) 2
f(O) +f (O)x +2"!x + ...
converges for all real numbers x and has sum zero, which is not equal to
f(x) if x #0.

This final chapter has introduced the idea of differentiability and demon-
strated that differential calculus has a rigorous foundation. The basic idea
in the development has been the concept of a limit, a theme which has recurred
repeatedly through the volume. First it appeared in the context of sequences
and then later it returned in connection with functions. This notion of a limit,
together with the concept of the real number system, underpins the analysis

195
GUIDE TO ANALYSIS

of this volume. The outcome of our work has been the development of
numerous theorems-perhaps a somewhat daunting achievement for the
beginner. These theorems, however, are the tools of the trade to the budding
analyst. Without theorems, life would be much harder for the mathematician
and problems would take much longer to solve. Alexander the Great, in his
day, found that mathematics contained very many theorems, propositions
and definitions and a great deal of time and effort was needed to learn them.
Like his general Ptolemy, Alexander asked if there was an easier way to
master mathematics, and was assured that there are no shortcuts. This is still
true today. It requires time and effort to learn analysis and to learn how to
apply it.

MISCELLANEOUS EXERCISES 5
1 Find the local maxima, local minima and points of inflection of x 2 (x - 1)3.
x 2 -x-2
2 Let h(x) = (x _1)2(X + 1)2 (x~ ±1).

Decide whether limx .... ah(x) exists in each of the following cases:
a = - 1,0, 1,2. In any case in which a limit exists find it.
3 The function g: ~ -+ ~ is given by
(x rational),
g(x) = {~2 (x irrational).
Show that 9 is differentiable at the origin. Show also thatg is
discontinuous at all points of ~\ {O}.
4 The function g: ~ -+ ~ is given by

g(X)={:: (x rational),
(x irrational).
Show that 9 is continuous at 0, 1 and discontinuous at all other points.
Where is 9 differentiable?
5 Decide whether the following limits exist:
. e
11m
x' - 1

x-+ocosx -1'
. lim (cot x
x-+o
_.!..);
x
lim (eX + X)l/X.
x-+o
In any case in which the limit exists find it.
6 Show that the equation
tanh 2x + cos(x/2) = x
has a solution in the interval [ - 2, 2].

196
 DIFFERENTIABLE FUNCTIONS

7 Letf,g,h:[ -1, +1]-+1R be given by

f(x)=lxI 3 (-1~x~1),

g(x)=x+lxl (-1~x~1),

h(x)=x 3 + [x] (-1~x~1),

where [x] is the greatest integer which does not exceed x.

State whether each of the functions f, g, h satisfy the conditions of the
MVT. Do the conclusions of the MVT hold for any of these functions?
8 The function f is continuous on [a, b] and differentiable on (a, b) and
f'(x) > 0 for all x such that a < x < b. Show that f is strictly increasing
on [a,b].
Hence show that for all x > 0,
X -tx 2 < log(1 + x) < X -tx 2 + i X 3.
9 The function g is continuous on [a, b] and differentiable on (a, b) and
o~ g'(x) ~ 1 for all x such that a < x < b. Moreover, g(a) = a. Prove that
if a ~ x ~ b, then a ~ g(x) ~ b. Show also that g(b) = b if and only if
g(x) = x for all xe[a, b].
10 Use Taylor's theorem to develop series expansion for sin x, cos x, sinh x,
coshx. Check carefully that the remainder term tends to zero as n-+ 00.
11 Show that for all real numbers ex,
log(l + exx)
--=--'----'- -+ ex
x
as x -+ o. Deduce that
nlog(l + ex/n)-+ex
as n -+ 00. Hence, show that

as n-+ 00.

HINTS FOR SOLUTION OF EXERCISES

Exercises 5.1.1
2 For x > O,f(x) = eX and so f'(x) = eX(x > 0). Similarly,f'(x) = 1 for x < O.
To prove differentiability at the origin, consider what happens to the
quotient [f(x) - f(O)]/(x - 0) as x-+O.
3 (e) Write f(x) = e(logx)(logx) (x> 0).

197
GUIDE TO ANALYSIS

Exercises 5.2.1
1 Use the intermediate value theorem to show there is a solution in [0,1].
Use Rolle's theorem to show that there cannot be more than one solution
in [0,1].
2 Apply MVT to sin x on the interval [a, b ].
3 Apply MVT to logx on the interval [a,b].
5 See example immediately preceding the exercises.
6 Apply MVT to sin- 1 x on [a,b].
7 Let ex = /(0), then ex> 0 and g(O) = ex. Use the MVT to show that
g(3ex) < -!ex. Then the intermediate value theorem will give the existence
of c for which g (c) = 0.)

Miscellaneous Exercises 5
2 Note that x 2 - x - 2 = (x - 2)(x + 1) and therefore
x-2
h(x) = (x _ 1)2(X + 1) (x#±1).

3 To prove differentiability at the origin, consider what happens to

[g(x) - g (O)]/(x - 0) as x -+ O. Notice that the modulus of this quotient
never exceeds Ixl when x # O.
Use sequences to prove the discontinuity.
4 Remember that differentiability implies continuity, and so g is not
differentiable at any point where it is discontinuous.
5 Write cotx -1/x as (xcosx - sinx)/x sin x and use I'Hopital's rule.
Show that

as x-+O

and deduce that (eX + x)1/x-+e 2 as x-+O.

6 Use the intermediate value theorem.
7 Show that Ix 3 is differentiable everywhere. Show that g is differentiable
1

at all points of [ - 1, 1] except the origin. Show that h is not differentiable

at the origin, but is differentiable at all other points of ( - 1, 1). Since /
satisfies the conditions of MVT, the conclusions of MVT must hold for f.
By definition g(l) - g( -1) = 2 and so [g(l) - g( -1)]/2 = 1, but g'(x) = 2
for x > 0 and g'(x) = 0 for x < 0 and the conclusion ofthe MVT does not
hold. By definition [h(l) - h( -1)]/2 =! = 2 and h'(x) = 3x 2 for 0 < x < 1.
Hence h'(y1) = 2.

198
 DIFFERENTIABLE FUNCTIONS

ANSWERS TO EXERCISES
Exercises 5.1.1
3 (a) f'(x) = 3x 2 sinh x sin(x2) + x 3 cosh x sin(x2) + 2X4 sinh xcos(x 2 )
(XE~).
(b) f'(x) = (sin x + xcosx)eXSinX (XE~).
(c) f'(x) = - 2x sin [log( 1 + x 2)]/(l + x 2) (XE~).
(d) f'(x) = 2x/J2 + 2X2 + X4 (XE~).
(e) f'(x) = 2xlogx-llogx (x> 0).

Exercises 5.4.1
1 - 2. (Remember, you want the limit as x --+ 1.)
2 32/n 2 • 3 cot 1 (I'Hopital's rule not appropriate!)
4 sin 1. 5 1/n 2 • 6 O. 7 -l/e. 8 t. 9 .!
4' 10 1
-"2'

11 1. 12 -n. 13 -"2'
1
14 7
3' 15 2.
16 cosh(n/2) (don't use l'Hopital).

Miscellaneous Exercises 5
1 Local maximum at x = 0; local minimum at x =~; point of inflection at
X= 1.

2 There is no limit as x--+ -1; h(x)--+ -2 as x--+O; h(x)--+ -00 as x--+1;

h(x)--+O as x--+2.
4 9 is differentiable at the origin.
5 All the limits exist. Their values are - 2, 0, e 2.
7 f satisfies the conditions of the MVT on [ -1,1] but g, h do not. The
conclusions of the MVT hold for f and h, but not for g.

199
INDEX

absolute convergence 92 decimal expansion of rationals 3

algebra of limits decreasing function 79
for sequences 36 decreasing sequence 47
for functions 121 derivative 163
alternating series test 89 differentiability 163
Archimedean property 13 divergence
of sequences 36
bijection 152 of series 65
bounded above 9 domain of definition of a function 106
bounded below 9
bounded set 9 Euler's constant 85
bounded sequence 36
bounded function 146 function 106
bounds
lower bound 9 geometric series 66
upper bound 9
higher-order derivatives 171
Cauchy criterion for convergence 55
Cauchy form of product 97
closed interval 109 image 107
comparison test 71 increasing sequences 47
comparison test (limit form) 73 induction 20
completeness axiom 12 inequalities 31, 56
composition infimum 10
of continuous functions 141-3 integer 2
of differentiable functions 167 integral test 83
conditional convergence 93 intermediate value theorem 148
continuity 139 intervals
convergence closed interval 109
of sequences 36 open interval 109
of series 65 inverse function 152
countability 5 irrational number 15
critical point 188 irrationality of J2 6, 14

D' Alembert's test 76 I'Hopital's rule 183

201
INDEX

limit standard limits

of a sequence 29, 32, 34 for n« 42
ofa function 111,114,115,126,131 for aft 42
one-sided 129, 130 for aftln« 43
limit form of comparison test 73 for a 1/ ft 44
local maximum 187 for n 1/ ft 44
local minimum 187 for aftln! 45
lower bound 9 for n!lnft 45
for (1 + ~ )ft 49
mathematical induction 20 strictly decreasing
maximum (local) 187 sequence 47
mean value theorem (MVT) 175 function 152
minimum (local) 187 strictly increasing
modulus of a real number 56 sequence 47
monotonic sequence 47 function 152
multiplication of series 97-100 strictly monotonic function 152
subsequence 51
supremum 10
natural number
Taylor series 191
open interval 109 Taylor's theorem 180
tests for convergence of series
partial sum 65 alternating series test 89
comparison test 71
D' Alembert's (ratio) test 76
range of a function 106 integral test 83
ratio test 76 limit form of comparison test 73
rational number 3 triangle inequalities 31, 56
real number 7
rearrangement of series 94-6, 100 uniqueness of limit
Rolle's theorem 173 of sequence 35
of function 117
sandwich rule 38, 118, 127, 131 upper bound 9
sequence 26
series 64 well-ordering principle 2

202