Sei sulla pagina 1di 283

Further calculus

A. Ostaszewski, J.M. Ward


MT2176, 2790176
2012

Undergraduate study in
Economics, Management,
Finance and the Social Sciences

This subject guide is for a 200 course offered as part of the University of London
International Programmes in Economics, Management, Finance and the Social Sciences.
This is equivalent to Level 5 within the Framework for Higher Education Qualifications in
England, Wales and Northern Ireland (FHEQ).
For more information about the University of London International Programmes
undergraduate study in Economics, Management, Finance and the Social Sciences, see:
www.londoninternational.ac.uk
This guide was prepared for the University of London International Programmes by:
A. Ostaszewski, Department of Mathematics, The London School of Economics and
Political Science.
J.M. Ward, Department of Mathematics, The London School of Economics and Political
Science.
This is one of a series of subject guides published by the University. We regret that due to
pressure of work the authors are unable to enter into any correspondence relating to, or aris-
ing from, the guide. If you have any comments on this subject guide, favourable or unfavour-
able, please use the form at the back of this guide.

University of London International Programmes


Publications Office
Stewart House
32 Russell Square
London WC1B 5DN
United Kingdom
www.londoninternational.ac.uk

Published by: University of London


© University of London 2012
The University of London asserts copyright over all material in this subject guide except where
otherwise indicated. All rights reserved. No part of this work may be reproduced in any form,
or by any means, without permission in writing from the publisher.
We make every effort to contact copyright holders. If you think we have inadvertently used
your copyright material, please let us know.
Contents

Contents

1 Introduction 1
1.1 This subject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Online study resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.1 The VLE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.2 Making use of the Online Library . . . . . . . . . . . . . . . . . . 5
1.4 Using this subject guide . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Examination advice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 The use of calculators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Limits 7
2.1 Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Limits at infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.2 Limits at a point . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Some useful results that involve limits . . . . . . . . . . . . . . . . . . . . 28
2.2.1 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.2 Differentiability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.3 Taylor series and Taylor’s theorem . . . . . . . . . . . . . . . . . 31
2.2.4 L’Hôpital’s rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Learning outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Solutions to activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

3 The Riemann integral 53


3.1 The Riemann integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.1.1 Lower and upper estimates of an area . . . . . . . . . . . . . . . . 54
3.1.2 Getting better lower and upper estimates . . . . . . . . . . . . . . 60
3.1.3 The definition of the Riemann integral . . . . . . . . . . . . . . . 63
3.1.4 What happens if the integrand isn’t continuous? . . . . . . . . . . 66
3.1.5 Some properties of the Riemann integral . . . . . . . . . . . . . . 71

i
Contents

3.2 The Fundamental Theorem of Calculus . . . . . . . . . . . . . . . . . . . 72


3.2.1 Motivating the FTC . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.2.2 Notation: Dummy variables . . . . . . . . . . . . . . . . . . . . . 75
3.2.3 The relationship between integration and differentiation . . . . . . 76
3.2.4 Some applications of the FTC . . . . . . . . . . . . . . . . . . . . 77
3.2.5 An extension of the FTC . . . . . . . . . . . . . . . . . . . . . . . 79
Learning outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Solutions to activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

4 Improper integrals 99
4.1 Improper integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.1.1 Improper integrals of two kinds, and a third kind . . . . . . . . . 100
4.1.2 Some further thoughts on improper integrals . . . . . . . . . . . . 102
4.2 Tests for convergence and divergence . . . . . . . . . . . . . . . . . . . . 106
4.2.1 The Direct Comparison Test . . . . . . . . . . . . . . . . . . . . . 107
4.2.2 The Limit Comparison Test . . . . . . . . . . . . . . . . . . . . . 111
4.2.3 Variable sign integrands . . . . . . . . . . . . . . . . . . . . . . . 121
Learning outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Solutions to activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

5 Double integrals 135


5.1 Double integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.1.1 Volumes over rectangular bases . . . . . . . . . . . . . . . . . . . 136
5.1.2 Defining double integrals in terms of volumes . . . . . . . . . . . 137
5.1.3 Motivating Fubini’s theorem . . . . . . . . . . . . . . . . . . . . . 139
5.1.4 Fubini’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.1.5 Volumes over other bases . . . . . . . . . . . . . . . . . . . . . . . 143
5.2 Change of variable techniques . . . . . . . . . . . . . . . . . . . . . . . . 152
5.3 Improper double integrals . . . . . . . . . . . . . . . . . . . . . . . . . . 162
Learning outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
Solutions to activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

ii
Contents

Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180

6 Manipulation of integrals 187


6.1 The manipulation of proper integrals . . . . . . . . . . . . . . . . . . . . 187
6.1.1 Joint continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
6.1.2 The manipulation rules for proper integrals . . . . . . . . . . . . . 191
6.1.3 Applications of the rules for manipulating proper integrals . . . . 194
6.2 The manipulation of improper integrals . . . . . . . . . . . . . . . . . . . 197
6.2.1 Dominated convergence . . . . . . . . . . . . . . . . . . . . . . . 198
6.2.2 The manipulation rules for improper integrals . . . . . . . . . . . 205
6.2.3 Using the rules for manipulating improper integrals . . . . . . . . 206
Learning outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
Solutions to activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217

7 Laplace transforms 221


7.1 What is a Laplace transform? . . . . . . . . . . . . . . . . . . . . . . . . 221
7.1.1 Some properties of the Laplace transform . . . . . . . . . . . . . . 223
7.1.2 Extending our view of Laplace transforms . . . . . . . . . . . . . 230
7.2 Using Laplace transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
7.2.1 Solving ODEs with constant coefficients . . . . . . . . . . . . . . 233
7.2.2 Convolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
Learning outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
Solutions to activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252

A Sample examination paper 257

B Solutions to the sample examination paper 261

iii
Contents

iv
1

Chapter 1
Introduction

In this very brief introduction, we aim to give you an idea of the nature of this subject
and to advise you on how best to approach it. We give general information about the
contents and use of this subject guide, and on recommended reading and how to use the
textbooks.

1.1 This subject


Calculus, as studied in this 200 course, is primarily the study of integrals of functions of
one and two variables.
Our approach here is not just to help you acquire proficiency in techniques and
methods, but also to help you understand some of the theoretical ideas behind these.
For example, after completing this course, you will hopefully understand how certain
kinds of definite integral are defined and how to deal with integrals where the integrand
is a function of two variables.

Aims of the course

The broad aims of this course are:

to enable students to acquire skills in the methods of calculus, as required for their
use in further mathematics subjects and economics-based subjects;

to prepare students for further courses in mathematics and/or related disciplines.


However, as emphasised above, we do also want you to understand why certain methods
work: this is one of the ‘skills’ that you should acquire. Indeed, the examination will not
simply test your ability to perform routine calculations, it will also probe your
knowledge and understanding of the principles that underlie the material.

Learning outcomes

We now state the broad learning outcomes of this course, as a whole. At the end of this
course and having completed the essential reading and activities, you should be able to:

1
1. Introduction
1
demonstrate knowledge of the subject matter, terminology, techniques and
conventions covered in the subject;

demonstrate an understanding of the underlying principles of the subject;

demonstrate the ability to solve problems involving an understanding of the


concepts.
There are a couple of things that we should stress at this point. Firstly, note the
intention that you will be able to solve unseen problems. This means simply that you
will be expected to be able to use your knowledge and understanding of the material to
solve problems that are not completely standard. This is not something you should
worry unduly about: all courses in mathematics expect this, and you will never be
expected to do anything that cannot be done using the material of this course.
Secondly, we expect you to be able to ‘demonstrate knowledge and understanding’ and
you might well wonder how you would demonstrate this in the examination. Well, it is
precisely by being able to grapple successfully with unseen, non-routine, questions that
you will indicate that you have a proper understanding of the topic.

Topics covered

Descriptions of the topics to be covered appear in the relevant chapters. However, it is


useful to give a brief overview at this stage.
We start by introducing the limit of a function of one variable and, in particular, how
this can be used to define what it means to say that a function is continuous. We then
introduce the Riemann integral and explain its relationship to differentiation via the
Fundamental Theorem of Calculus. This leads on to a discussion of improper integrals
and, in particular, some tests that we can use to determine whether such integrals are
convergent or divergent. We then turn our attention to functions of two variables, in
particular, how we can integrate such functions over certain regions and how we can
manipulate such integrals. We then discuss Laplace transforms and some of their
important applications.
Throughout this subject guide, the emphasis will be on the theory as much as on the
methods. That is to say, our aim in this subject is not only to provide you with some
useful techniques and methods from calculus, but to also enable you to understand why
these techniques work.

1.2 Reading
There are many books that would be useful for this subject. We recommend two in
particular, and a couple of others for additional, further reading. (You should note,
however, that there are very many books suitable for this course. Indeed, almost any
text on first-year university calculus will cover the majority of the material.)
Textbook reading is essential as textbooks will provide you with more in-depth
explanations than you will find in this subject guide, and they will also provide many
more examples to study and exercises to work through. The books listed are the ones
we have referred to in this subject guide.

2
1.3. Online study resources
1
Essential reading

Detailed reading references in this subject guide refer to the editions of the set
textbooks listed below. New editions of one or more of these textbooks may have been
published by the time you study this course. You can use a more recent edition of any
of the books; use the detailed chapter and section headings and the index to identify
relevant readings. Also check the virtual learning environment (VLE) regularly for
updated guidance on readings.

+ Binmore, K. and J. Davies Calculus: concepts and methods. (Cambridge:


Cambridge University Press, 2002) second revised edition [ISBN 9780521775410].

+ Ostaszewski, A. Advanced mathematical methods. (Cambridge: Cambridge


University Press, 1991) [ISBN 9780521289641].

Both of these texts, when used wisely, will provide you with a large number of examples
for you to study and exercises for you to attempt. It is recommended that you purchase
both of these.

Further reading

Once you have covered the essential reading you are then free to read around the
subject area in any text, paper or online resource. You will need to support your
learning by reading as widely as possible and by thinking about how these principles
apply in the real world. To help you read extensively, you have free access to the VLE
and University of London Online Library (see Section 1.3.2). However, two useful
textbooks that we have referred to in this subject guide are the following.

+ Adams, R.A. and C. Essex Calculus: A complete course. (Toronto: Pearson, 2009)
seventh edition [ISBN 9780321549280].

+ Wrede, R. C. and M. Spiegel Schaum’s outline of advanced calculus. (London:


McGraw-Hill, 2010) third edition [ISBN 9780071623667].

Adams and Essex (which is merely an example from a large range of very similar
calculus textbooks) is a detailed calculus textbook which contains much material which
is beyond the scope of this course. Wrede and Spiegel contains a brief summary of some
of the course material but is useful as it contains a large number of worked examples
and exercises. Both of these texts are suitable as sources of additional explanation,
examples and exercises, but they are probably not worth purchasing.

1.3 Online study resources


In addition to the subject guide and the essential reading, it is crucial that you take
advantage of the study resources that are available online for this course, including the
VLE and the Online Library.

3
1. Introduction
1
You can access the VLE, the Online Library and your University of London email
account via the Student Portal at

http://my.londoninternational.ac.uk

You should have received your login details for the Student Portal with your official
offer, which was emailed to the address that you gave on your application form. You
have probably already logged in to the Student Portal in order to register! As soon as
you registered, you will automatically have been granted access to the VLE, Online
Library and your fully functional University of London email account.
If you forget your login details at any point, please email uolia.support@london.ac.uk
quoting your student number.

1.3.1 The VLE


The VLE, which complements this subject guide, has been designed to enhance your
learning experience, providing additional support and a sense of community. It forms an
important part of your study experience with the University of London and you should
access it regularly.
The VLE provides a range of resources for EMFSS courses:

Self-testing activities: Doing these allows you to test your own understanding of
subject material.

Electronic study materials: The printed materials that you receive from the
University of London are available to download, including updated reading lists
and references.

Past examination papers and Examiners’ commentaries: These provide advice on


how each examination question might best be answered.

A student discussion forum: This is an open space for you to discuss interests and
experiences, seek support from your peers, work collaboratively to solve problems
and discuss subject material.

Videos: There are recorded academic introductions to the subject, interviews and
debates and, for some courses, audio-visual tutorials and conclusions.

Recorded lectures: For some courses, where appropriate, the sessions from previous
years’ Study Weekends have been recorded and made available.

Study skills: Expert advice on preparing for examinations and developing your
digital literacy skills.

Feedback forms.
Some of these resources are available for certain courses only, but we are expanding our
provision all the time and you should check the VLE regularly for updates.

4
1.4. Using this subject guide
1
1.3.2 Making use of the Online Library
The Online Library contains a huge array of journal articles and other resources to help
you read widely and extensively.
To access the majority of resources via the Online Library at

http://tinyurl.com/ollathens

you will either need to use your University of London Student Portal login details, or
you will be required to register and use an Athens login.
The easiest way to locate relevant content and journal articles in the Online Library is
to use the Summon search engine.
If you are having trouble finding an article listed in a reading list, try removing any
punctuation from the title, such as single quotation marks, question marks and colons.
For further advice, please see the online help pages at

www.external.shl.lon.ac.uk/summon/about.php

1.4 Using this subject guide


We have already mentioned that this subject guide is not a textbook. It is important
that you read textbooks in conjunction with the subject guide and that you try
problems from the textbooks. The exercises at the end of the main chapters of this
subject guide are a very useful resource and you should try them once you think you
have mastered the material from the chapter. You should really try these exercises
before consulting the solutions, as simply reading the solutions provided will not help
you at all. Sometimes, the solutions we provide will just be an overview of what is
required, i.e. an indication of how you should answer the questions, but in the
examination, you must always show all of your calculations. It is vital that you develop
and enhance your problem-solving skills and the only way to do this is to try lots of
exercises.

1.5 Examination advice


Important: the information and advice given here are based on the examination
structure used at the time this subject guide was written. Please note that subject
guides may be used for several years. Because of this we strongly advise you to always
check both the current Regulations for relevant information about the examination, and
the VLE where you should be advised of any forthcoming changes. You should also
carefully check the rubric/instructions on the paper you actually sit and follow those
instructions.

Remember, it is important to check the VLE for:

Up-to-date information on examination and assessment arrangements for this


course.

5
1. Introduction
1
Where available, past examination papers and Examiners’ commentaries for the
course which give advice on how each question might best be answered.
This course is assessed by a two hour unseen written examination. There are no
optional topics in this subject: you should study them all and this is reflected in the
structure of the examination paper. There are five questions (each worth 20 marks) and
all questions are compulsory. A sample examination paper may be found in an appendix
to this subject guide.
Please do not think that the questions in your real examination will necessarily be very
similar to the exercises in this subject guide or those in the sample examination paper.
The examination is designed to test you. You will get examination questions unlike the
questions in this subject guide. The whole point of examining is to see whether you can
apply your knowledge in familiar and unfamiliar settings. The Examiners (nice people
though they are) have an obligation to surprise you! For this reason, it is important
that you try as many examples as possible from the subject guide and from the
textbooks. This is not so that you can cover any possible type of question the
Examiners can think of! It is so that you get used to confronting unfamiliar questions,
grappling with them, and finally coming up with the solution.
Do not panic if you cannot completely solve an examination question. There are many
marks to be awarded for using the correct approach or method.

1.6 The use of calculators


You will not be permitted to use calculators of any type in the examination. This is not
something that you should worry about: the Examiners are interested in assessing that
you understand the key concepts, ideas, methods and techniques, and will set questions
which do not require the use of a calculator.

6
Chapter 2 2
Limits

Essential reading

(For full publication details, see Chapter 1.)

+ Ostaszewski (1991) Sections 17.4–17.6 and Section 18.8.

Further reading

+ Adams and Essex (2010) Sections 1.2–1.4, parts of Section 2.2, Sections 4.3 and
4.9–4.10.

+ Wrede and Spiegel (2010) parts of Chapters 3, 4 and 11.

Aims and objectives

The objectives of this chapter are:

to see what a limit is and how they can be found in a variety of different situations;

to examine the relationship between limits, continuity and differentiability.


Specific learning outcomes can be found near the end of this chapter.

2.1 Limits
We encountered the general idea behind limits in 174 Calculus and, although we used
the idea there, we never gave a thorough account of what was involved. In this section,
we will make the idea behind a limit more precise, but our account will still be fairly
informal.1 And, once we have done this, we will look at some useful results that involve
limits such as the use of limits to define what it means to say that a function is
continuous or differentiable. We will also see how our understanding of Taylor series can
be extended by using Taylor’s theorem and we will end this chapter by considering
L’Hôpital’s rule which will allow us to calculate some of the ‘trickier’ limits that we will
encounter.
1
That is, we will say enough to make the idea of a limit precise and see how to calculate limits, but
we will not give a rigorous mathematical treatment of limits like the one you will see in 116 Abstract
Mathematics.

7
2. Limits

2.1.1 Limits at infinity


Limits ‘at infinity’ are concerned with the behaviour of a function, f (x), as x tends to
2 infinity, a situation we denoted by ‘x → ∞’ in 174 Calculus. Indeed, as we saw in that
course, this kind of information is useful when we were sketching the graph of a function
because it told us what was happening as x gets ‘very large’. In this section, we consider
exactly what this kind of limit means and see how we can find such a limit (if it exists)
in some straightforward cases.

Finite limits

Suppose that l is a real number and that f (x) is a function, we start by asking what it
means to say that f (x) → l as x → ∞, i.e. what it means to say that

lim f (x) = l.
x→∞

Intuitively, based on what we saw in 174 Calculus, we would want to say that this
means that the curve y = f (x) has a horizontal asymptote given by y = l, i.e. a
horizontal line that the curve gets arbitrarily close to. But, more specifically, we mean
that however close we would like f (x) to be to l, there is some suitably large value of x,
say X, for which f (x) is as close to l as we wanted if x ≥ X.2 For instance, in Figure 2.1
we have the graphs of the two functions where f (x) → l as x → ∞.

y y

y = f (x) y = f (x)

l l

O x O x
(a) (b)

Figure 2.1: Two functions for which f (x) → l as x → ∞ where l > 0 is some real number.
2
Technically, as you will see in 116 Abstract Mathematics, we say that f (x) → l as x → ∞ for some
real number, l, if

for any ε > 0, there is an X such that, for all x ≥ X, |f (x) − l| < ε.

Here, the value of ε > 0 tells us how close we want to be and, if we can show that there is an X such that
|f (x) − l| < ε for all x ≥ X, then we have established that f (x) is as close to l as we wanted for these
values of x. Indeed, if we can do this for any value of ε, we can guarantee that f (x) is getting ‘arbitrarily
close’ to l as x → ∞. However, we will not make use of this formal definition here.

8
2.1. Limits

Notice, however, that we must take some care when we describe the fact that f (x) → l
as x → ∞. In particular, we don’t want to say that f (x) → l as x → ∞ because f (x)
gets ‘closer and closer’ to l when we take larger and larger values of x. This is because,
if we consider the graph in
2
Figure 2.1(a), we see that f (x) also gets ‘closer and closer’ to, say, zero even
though that is clearly not its limit as x → ∞.
Figure 2.1(b), we see that at some points f (x) is ‘heading towards’ l and sometimes
it is ‘heading away’ from l and so f (x) is not always getting ‘closer and closer’ to l
even though l clearly is its limit as x → ∞.
Let’s now consider how we can actually find such limits.

Finding finite limits

Usually, we can find finite limits by considering some basic functions that have finite
limits and then, by using some appropriate rules about how limits work, we can find the
limits of certain combinations of these basic functions. So, we start by stating some
finite limits that arise from basic functions, i.e.

(a) If f (x) = a where a is a constant, then f (x) → a as x → ∞.


(b) If f (x) = 1/xb where b > 0 is a constant, then f (x) → 0 as x → ∞.
(c) If f (x) = 1/cx where c > 1 is a constant, then f (x) → 0 as x → ∞.
Then, we have the rules which tell us how to find the limits of certain combinations of
these basic functions which are stated in the following theorem.
Theorem 2.1 If f (x) → l and g(x) → m as x → ∞ where l, m ∈ R, then

(a) if c is a constant, cf (x) → cl as x → ∞;


(b) f (x) + g(x) → l + m as x → ∞;
(c) f (x)g(x) → lm as x → ∞;
f (x) l
(d) if m 6= 0, → as x → ∞;
g(x) m
(e) if b is a constant and l > 0, [f (x)]b → lb .

We will not prove this theorem here even though some of the rules may be fairly
obvious, but it is important that you treat these rules with some care. In particular, as
we require that m 6= 0 in Theorem 2.1(d), this tells us nothing about
f (x)
lim ,
x→∞ g(x)

if g(x) → 0 as x → ∞,3 but we will say more about this later. Also observe that in
Theorem 2.1(e), we require that l > 0 because, if we had b = 1/2 (say), this would make
3
As we should expect since l/m makes no sense if m = 0 because we can never divide by zero!

9
2. Limits

no sense in cases where l < 0 or, indeed, in some cases where l = 0 as we will see in
Activity 2.7. However, we now consider some examples of how we use the results above.
2 
1 1

Example 2.1 Find lim 3+ − x .
x→∞ x 2

Using (a), (b) and (c) respectively, it should be obvious that


1 1
lim 3 = 3, lim = 0 and lim = 0,
x→∞ x→∞ x x→∞ 2x

so, using Theorem 2.1(a) and (b) respectively, we have


   
1 1
lim − x = −0 = 0 and lim 3 + = 3 + 0 = 3.
x→∞ 2 x→∞ x

Then, using Theorem 2.1(b) again, this gives us


     
1 1 1 1
lim 3 + − x = lim 3+ + − x = 3 + 0 = 3,
x→∞ x 2 x→∞ x 2

as the answer.

Note: As this is fairly obvious, once you have understood the results above, you
would normally just write
 
1 1
lim 3 + − x = 3 + 0 − 0 = 3,
x→∞ x 2

since it is easy to find the limit of each of the three terms and hence the limit of this
combination of them.
p
4 − 1/x
Example 2.2 Find lim .
x→∞ 1 − 4/x3

Using (b) and Theorem 2.1(a), it should be obvious that


   
1 4
lim − = −0 = 0 and lim − 3 = −4(0) = 0,
x→∞ x x→∞ x
so, using (a) and Theorem 2.1(b), we have
    
1 1
lim 4 − = lim 4 + − = 4 + 0 = 4,
x→∞ x x→∞ x
and     
4 4
lim 1 − 3 = lim 1 + − 3 1 + 0 = 1.
x→∞ x x→∞ x
Now, the first of these limits is positive and so, using Theorem 2.1(e), we have
r  1/2
1 1
lim 4 − = lim 4 − = 41/2 = 2,
x→∞ x x→∞ x

10
2.1. Limits

and the second of these limits is non-zero and so, using Theorem 2.1(d), we have
p
4 − 1/x 2
lim
x→∞ 1 − 4/x 3
= = 2, 2
1

as the answer.

Note: As this is also fairly obvious, once you have understood the results above, you
would normally just write
p √
4 − 1/x 4−0 2
lim 3
= = = 2,
x→∞ 1 − 4/x 1−0 1

having taken care to observe that we are taking the square root of a positive number
and that we are not dividing by zero.

Of course, given what we have seen in these two examples, it should be obvious that we
could extend Theorem 2.1 by including the two results in the next activity.

Activity 2.1 Use Theorem 2.1 to show that: If f (x) → l and g(x) → m as x → ∞
where l, m ∈ R, then
f (x) − g(x) → l − m as x → ∞,
and
cf (x) + dg(x) → cl + dm as x → ∞,
where c and d are constants.

It is also useful to note that, sometimes, it is necessary to rewrite the function we are
considering before we attempt to find the limit.

x3 − 3x2 + 2
Example 2.3 Find lim .
x→∞ 4x3 + 6x
We start by noting that, in this case, we can not simply work out the limit of this
quotient by considering the limits of the numerator and the denominator as x → ∞
because neither of these have a limit which is a real number.4

In cases such as this, we employ the useful ‘trick’ of dividing the numerator and the
denominator by the highest power of x that occurs in the quotient.5 Indeed, here,
this highest power of x is x3 and so, dividing the numerator and denominator by
this, we get
x3 − 3x2 + 2 1 − (3/x) + (2/x3 )
lim = lim ,
x→∞ 4x3 + 6x x→∞ 4 + (6/x2 )
and, we can deal with this by considering the limits of the numerator and the
denominator as x → ∞ because now, the limit of the numerator is 1 and the limit of
the denominator is 4. As such, we can see that the limit we are asked to find is 1/4.

Note: As this is fairly obvious once you understand that, in such cases, we need to
divide by the highest power of x in the quotient in order to get finite limits in its

11
2. Limits

numerator and denominator as x → ∞, you would normally just write

x3 − 3x2 + 2 1 − (3/x) + (2/x3 ) 1−0+0 1


2 lim
x→∞ 3
4x + 6x
= lim
x→∞ 2
4 + (6/x )
=
4+0
= ,
4

as it is easy to find the limit once we have rewritten the quotient in this way.

Lastly, it will sometimes be useful to appeal to the so-called ‘Sandwich theorem’ when
we are asked to find the limits of certain functions that can not be usefully analysed
using the methods above.
Theorem 2.2 (The Sandwich theorem) If, for some X ∈ R, the functions f , g and h
are related by the inequality
f (x) ≤ g(x) ≤ h(x),

for all x ≥ X and there is some l ∈ R such that

lim f (x) = l and lim h(x) = l,


x→∞ x→∞

then we can conclude that lim g(x) = l too.


x→∞

Of course, we will not prove this here, but we will motivate it by considering an
example where it can be usefully applied.

sin x
Example 2.4 Show that lim = 0.
x→∞ x

As, for x > 0, we have


1 sin x 1
−1 ≤ sin x ≤ 1 =⇒ − ≤ ≤ ,
x x x
and we also have
 
1 1
lim = 0 which means that lim − = 0 as well,
x→∞ x x→∞ x

we can use the Sandwich theorem to conclude that


sin x
lim = 0,
x→∞ x

as required. Of course, this is obvious if we look at the graph of this function which is
illustrated in Figure 2.2 along with the graphs of the functions ±1/x that bound it.

4
In fact, both the numerator and the denominator of this quotient ‘tend to infinity’ as x → ∞, a
situation we shall consider in more detail in a moment.
5
We could call this highest power of x the ‘dominant term’ as its behaviour will ‘determine’ the
behaviour of the function in the limit.

12
2.1. Limits

Figure 2.2: The dashed curve in the positive quadrant is the graph of the function 1/x
and the dashed curve in the quadrant below this is the graph of the function −1/x. The
function sinx x , whose graph is the solid line, always lies between these two curves for x > 0
and, because of this, we see that it must tend to zero as x → ∞.

Activity 2.2 Find the following limits.

x2 + x + 1 (x + sin x − 2)1/2
(a) lim , (b) lim √ .
x→∞ x + x2 + x3 x→∞ x + sin x − 2

Infinite limits

If f (x) is a function, we now want to ask what it means to say that this function tends
to infinity as x → ∞, i.e. what it means to say that f (x) → ∞ as x → ∞ or, slightly
abusing our notation,6 that
lim f (x) = ∞.
x→∞

Intuitively, based on what we saw in 174 Calculus, we want to say that this means that
the function, f (x), can take arbitrarily large values as we let x get larger and larger.
But, more specifically, we mean that however large we want f (x) to be, let’s say we
want it to be larger than some real number, M , we can find a value of x, say X, for
which f (x) is larger than M for all x ≥ X.7
6
This is an abuse of our notation since, technically, what we denote by ‘∞’ is not a real number and
so nothing can be equal to it. However, what we write here will be a very useful ‘notational convenience’.
7
Technically, as you will see in 116 Abstract Mathematics, we say that f (x) → ∞ as x → ∞ if,

for any M > 0, there is an X such that, for all x ≥ X, f (x) ≥ M .

13
2. Limits

We also want to consider functions, f (x), which tend to minus infinity as x → ∞, i.e.
where f (x) → −∞ as x → ∞ or, again slightly abusing our notation, where
2 lim f (x) = −∞.
x→∞

Of course, we saw this kind of behaviour in 174 Calculus too and we want to say that
this means that the function, f (x), can take negative values which are arbitrarily large
in magnitude as we let x get larger and larger. But, it is perhaps easier to define this in
terms of what we have just seen, i.e. we say that
 
lim f (x) = −∞ if lim − f (x) = ∞,
x→∞ x→∞

i.e. if −f (x) → ∞ as x → ∞, then we must have f (x) → −∞ as x → ∞. Let’s now


consider how we can actually find such limits.

Finding infinite limits

Again, as we saw above with finite limits, we can find infinite limits by considering some
basic functions that have infinite limits and then, by using some appropriate rules about
how limits work, we can then find the limits of certain combinations of these basic
functions. So, we start by stating some infinite limits that arise from basic functions, i.e.

(a) If f (x) = xb where b > 0 is a constant, then f (x) → ∞ as x → ∞.

(b) If f (x) = cx where c > 1 is a constant, then f (x) → ∞ as x → ∞.

(c) If f (x) = logd x where d > 1 is a constant, then f (x) → ∞ as x → ∞.


Then we have the rules which tell us how to find the limits of certain combinations of
these basic functions which are stated in the following theorem.
Theorem 2.3 If f (x) → ∞ as x → ∞, then

(a) if c > 0 is a constant, then cf (x) → ∞ as x → ∞.

And, if we also have g(x) → ∞ as x → ∞, then

(b) f (x) + g(x) → ∞ as x → ∞.

(c) f (x)g(x) → ∞ as x → ∞.

Whereas, if we have g(x) → m as x → ∞ where m > 0 is a real number, then

(d) f (x) + g(x) → ∞ as x → ∞.

(e) f (x)g(x) → ∞ as x → ∞.
Here, the value of M > 0 tells us how large we want f (x) to be and, if we can show that there is an
X such that f (x) ≥ M for all x ≥ X, then we have established that f (x) is always larger than M for
these values of x. Indeed, if we can do this for any value of M , we can guarantee that f (x) is getting
‘arbitrarily large’ as x → ∞. However, we will not make use of this formal definition here.

14
2.1. Limits

f (x)
(f) → ∞ as x → ∞.
g(x)

We will not prove this theorem here even though some of the rules may be fairly 2
obvious, but it is important that you treat these rules with some care. In particular,
although you can extend what we have seen in Theorem 2.3 fairly simply by doing
Activity 2.3, there are some things that we won’t be able to do at the moment as you’ll
see in Activity 2.4.

Activity 2.3 What is the analogue of Theorem 2.3(a) when c < 0 and what are the
analogues of Theorem 2.3(d)-(f) when m < 0?

Hence show that, if 0 < d < 1 is a constant, then logd x → −∞ as x → ∞.

Activity 2.4 What, if anything, can you say about the analogue of

(i) Theorem 2.3(a) when c = 0?

(ii) Theorem 2.3(b) and Theorem 2.3(c) when g(x) → −∞ as x → ∞?

(iii) Theorem 2.3(d)-(f) when m = 0?

Let’s now see how these results work by considering some examples.

Example 2.5 Find lim (x3 + x + 2).


x→∞

Using Theorem 2.3(a) and what we saw before, we have

lim x3 = ∞, lim x = ∞ and lim 2 = 2,


x→∞ x→∞ x→∞

so, using Theorem 2.3(d), we have

lim (x + 2) = ∞,
x→∞

so that, using Theorem 2.3(b), we get

lim (x3 + x + 2) = ∞,
x→∞

as the answer.

x3 + 2x + 2
Example 2.6 Find lim .
x→∞ x2 + 1
We saw in Example 2.5 that the numerator of the function

x3 + 2x + 2
,
x2 + 1
tends to infinity as x → ∞ and, using similar reasoning, the denominator tends to
infinity as x → ∞ too. In particular, this means that we can’t use any of the results

15
2. Limits

in Theorem 2.3 on this function as it stands. However, if we divide the numerator


and the denominator of this function by the highest power of x in the denominator,8
2 i.e. x2 , we get
x3 + 2x + 2 x + (2/x) + (2/x2 )
= ,
x2 + 1 1 + (1/x2 )
and in this form the numerator still tends to infinity as x → ∞, but the denominator
now tends to one which is a positive real number. Consequently, we can use
Theorem 2.3(f) to see that

lim x + (2/x) + (2/x2 )


 
x3 + 2x + 2 x + (2/x) + (2/x2 )
lim = lim = x→∞ = ∞,
x→∞ x2 + 1 x→∞ 1 + (1/x2 ) 1

is the answer.

x+1
Example 2.7 Find lim √ .
x→∞ 4x − 1
It should be clear that the numerator and the denominator of the function
x+1
√ ,
4x − 1
both tend to infinity as x → ∞ and so we are in a similar situation to the one in
Example 2.6. So, as we did in that example, we divide the numerator and the √
denominator of this function by the highest power of x in the denominator, i.e. x,
to get √ √
x+1 x + (1/ x)
√ = p ,
4x − 1 4 − (1/x)
and in this form the numerator still tends to infinity as x → ∞, but the denominator
now tends to two — as we saw in Example 2.2 — which is a positive real number.
Consequently, we can use Theorem 2.3(f) to see that
√ √
√ √ 
x+1 x + (1/ x) lim x + (1/ x)
lim √ = lim p = x→∞ = ∞,
x→∞ 4x − 1 x→∞ 4 − (1/x) 2

is the answer.

x2 − sin x
Activity 2.5 Find the limits (a) lim (x2 − x3 ) and (b) lim .
x→∞ x→∞ x + sin x

8
Of course, the highest power of x in the quotient is x3 (i.e. this is the ‘dominant term’), but if we
divide the numerator and denominator by this, we get

x3 + 2x + 2 1 + (2/x2 ) + (2/x)
2
= .
x +1 (1/x) + (1/x3 )

And, this means that, as x → ∞, the numerator tends to one and the denominator tends to zero, a case
that we can not deal with using Theorem 2.1(d). However, we will see in Example 2.8 that we can make
sense of this once we have Theorem 2.4.

16
2.1. Limits

The relationship between infinite and finite limits

We have seen that, as x → ∞, some functions have finite limits and others have infinite 2
limits but now we want to briefly discuss how these two types of limit are related. The
key result here is the following theorem.
Theorem 2.4 (a) If f (x) → ∞ as x → ∞, then

1
lim = 0.
x→∞ f (x)

(b) If f (x) → 0 as x → ∞ and there is an M ∈ R such that

1
f (x) > 0 for all x > M , then lim = ∞.
x→∞ f (x)

1
f (x) < 0 for all x > M , then lim = −∞.
x→∞ f (x)

Of course, the results in this theorem should be fairly obvious and we can see why if we
consider an example.

Example 2.8 Following on from Example 2.6, use Theorem 2.4 to verify that

x3 + 2x + 2
lim = ∞,
x→∞ x2 + 1
as we found there.9

The highest power of x in the quotient is x3 ,10 and if we divide the numerator and
denominator by this, we get

x3 + 2x + 2 1 + (2/x2 ) + (2/x3 )
 
2 2 1
2
= 3
= 1+ 2 + 3 .
x +1 (1/x) + (1/x ) x x (1/x) + (1/x3 )

Now, as x → ∞, the first term in this product tends to one whereas, by


Theorem 2.4, the second term tends to infinity as
1 1
+ 3 > 0,
x x
for x > 0. Consequently, by Theorem 2.3, we see that

x3 + 2x + 2
lim = ∞,
x→∞ x2 + 1
as expected.

9
This example follows on from the discussion in footnote 8.
10
That is, x3 is the ‘dominant term’ here.

17
2. Limits

Limits that don’t exist

2 If we have a function f (x) and we find that

lim f (x) = c,
x→∞

where c is a real number or we find that f (x) → ∞ (or −∞) as x → ∞, we say that the
limit of f (x) as x → ∞ exists. However, not every function has a limit as x → ∞ and,
in such cases, we say that this limit does not exist.

Example 2.9 Explain why lim sin x does not exist.


x→∞

As we know from 174 Calculus, the function sin x, which is illustrated in Figure 2.3,
oscillates between the values of 1 and −1 with a period of 2π. As such, this function
has no limit as x → ∞ since it never stays arbitrarily close to any value.

Figure 2.3: The graph of the function sin x for x ≥ 0.

However, although in this case the oscillations mean that a limit doesn’t exist, we saw
in Example 2.4 that the function
sin x
,
x
tends to zero as x → ∞ even though it is oscillating.11 But, generally, some care must
be taken when deciding whether an oscillating function has a limit as the next two
activities illustrate.

11
Of course, in this case it is the fact that the ‘amplitude’ of the oscillations decreases to zero as x → ∞
that guarantees that this limit is zero!

18
2.1. Limits

Activity 2.6 Consider the limits

(i) lim x(1 + sin x) and (ii)


x→∞
lim x(2 + sin x).
x→∞
2
Do either of these limits exist? If the limit exists, what is it? (Hint: A sketch may
help!)

Activity 2.7 Consider the limits


r s
sin x sin x
(i) lim and (ii) lim
.
x→∞ x x→∞ x

Do either of these limits exist? If the limit exists, what is it? (Hint: A sketch may
help!)

2.1.2 Limits at a point


Limits ‘at a point’ are concerned with the behaviour of a function, f (x), as x tends to
some finite value, say a, a situation we denoted by ‘x → a’ in 174 Calculus. Indeed, as
we saw in that course, this kind of information is useful when we were sketching the
graph of a function because it told us what was happening as x gets ‘very close to a’,
especially if the behaviour of the function as x tends to a from above (denoted by
x → a+ ) was different to its behaviour as x tends to a from below (denoted by x → a− ).
In this section, we consider exactly what this kind of limit means and see how we can
find such a limit (if it exists) in some straightforward cases.

Finite limits

Suppose that l is a real number and that f (x) is a function that is defined for all values
of x < a.12 We start by asking what it means to say that f (x) → l as x → a− , or ‘as x
tends to a from below’, i.e. what it means to say that

lim f (x) = l.
x→a−

Intuitively, we would want to say that this means that we can ensure that f (x) is as
close to l as we want by taking values of x that are less than a but close enough to a.
But, more specifically, we mean that however close we would like f (x) to be to l, there
is some value of x, say X, for which f (x) is as close to l as we wanted if X < x < a.13
12
That is the function needs to be defined for all values of x less than a, but it need not be defined at
x = a.
13
Technically, as you will see in 116 Abstract Mathematics, we say that f (x) → l as x → a− for some
real number, l, if

for any ε > 0, there is a δ > 0 such that, for all x ∈ (a − δ, a), |f (x) − l| < ε.

Here, the value of ε > 0 tells us how close we want to be and, if we can show that there is a δ > 0 such
that |f (x) − l| < ε for all x in the interval (a − δ, a), then we have established that f (x) is as close to
l as we wanted for these values of x below x = a. Indeed, if we can do this for any value of ε, we can

19
2. Limits

Of course, we can also ask what it means to say that f (x) → l as x → a+ , or ‘as x tends
to a from above’, i.e. what it means to say that
2 lim f (x) = l,
x→a+

and, intuitively, we would want to say that this means that we can ensure that f (x) is as
close to l as we want by taking values of x that are greater than a but close enough to a.
But, more specifically, we mean that however close we would like f (x) to be to l, there
is some value of x, say X, for which f (x) is as close to l as we wanted if a < x < X.14
Indeed, if both of the limits

lim f (x) and lim f (x),


x→a− x→a+

exist and, furthermore, they are the same, then we say that the limit

lim f (x),
x→a

exists and is equal to this common value. However, having said that, unless we need to
worry about the limits of f (x) as x → a− and as x → a+ individually,15 we will often be
able to find the limit of f (x) as x → a straightaway. Let’s now consider how we can
actually do this.

Finding finite limits

Again, as we saw with limits at infinity in Section 2.1.1, we can find finite limits by
considering some basic functions that have finite limits and then, by using some
appropriate rules about how limits work, we can then find the limits of certain
combinations of these basic functions. So, we start by stating some finite limits that
arise from basic functions, i.e.

(a) If f (x) = xn where n ∈ N, then f (x) → an as x → a.

(b) If f (x) = cx where c > 0, then f (x) → ca as x → a.


Then we have the rules which tell us how to find the limits of certain combinations of
these functions which are stated in the following theorem.

guarantee that f (x) is getting ‘arbitrarily close’ to l as x → a− . However, we will not make use of this
formal definition here.
14
Technically, as you will see in 116 Abstract Mathematics, we say that f (x) → l as x → a+ for some
real number, l, if

for any ε > 0, there is a δ > 0 such that, for all x ∈ (a, a − δ), |f (x) − l| < ε.

Here, the value of ε > 0 tells us how close we want to be and, if we can show that there is a δ > 0 such
that |f (x) − l| < ε for all x in the interval (a, a − δ), then we have established that f (x) is as close to
l as we wanted for these values of x above x = a. Indeed, if we can do this for any value of ε, we can
guarantee that f (x) is getting ‘arbitrarily close’ to l as x → a+ . However, we will not make use of this
formal definition here.
15
For instance, if it is possible that one of them doesn’t exist or, if both of them exist, it is possible
that they aren’t equal.

20
2.1. Limits

Theorem 2.5 Theorem 2.1 holds if x → ∞ is replaced by x → a− , x → a+ or x → a.

Of course, the same caveats apply as the ones we saw after the statement of 2
Theorem 2.1.

x2 − 9
Example 2.10 Find lim (x + 3) and lim .
x→3 x→3 x − 3

As we should expect, for the first limit, we have

lim (x + 3) = 3 + 3 = 6,
x→3

as illustrated in Figure 2.4(a) whereas for the second limit we note that, as long as
x 6= 3, we have
x2 − 9 (x − 3)(x + 3)
= = x + 3,
x−3 x−3
even though this function is not actually defined at x = 3. So, as the limit as x → 3
only considers what the function is doing around x = 3,16 we see that

x2 − 9
lim = lim (x + 3) = 3 + 3 = 6,
x→3 x − 3 x→3

as illustrated in Figure 2.4(b). In particular, we observe that this limit is 6 even


though this function is undefined at x = 3 and, as such, the point (3, 6) is not part
of the graph of this function.

y y

x2 −9
6 y =x+3 6 y= x+3

3 3

O x O x
−3 3 −3 3
(a) (b)

Figure 2.4: Graphs of the two functions from Example 2.10. (Note that a ‘•’ means that
this point is actually part of the graph of the function whereas a ‘◦’ means that this point
is not actually part of the graph of the function.)

16
That is, whether it is actually defined at x = 3 is irrelevant when we consider the limit as x → 3.

21
2. Limits

x2
Example 2.11 Find lim .
x→2 x − 3
2
As we anticipate no problems as x → 2, we use Theorem 2.2 to get

lim x2 = 22 = 4 and lim (x − 3) = 2 − 3 = −1,


x→2 x→2

and, as the second of these limits is non-zero, Theorem 2.2(d) then gives us

x2 4
lim = = −4,
x→2 x − 3 −1
as the answer.

Note: As this is fairly obvious, once you have understood the results above, you
would normally just write

x2 22 4
lim = = = −4,
x→2 x − 3 2−3 −1
having taken care to observe that we are not dividing by zero.


1+x−1
Example 2.12 Find lim .
x→0 x
We start by noting that, in this case, we can not simply work out the limit of this
quotient by considering the limits of the numerator and the denominator as x → 0
because both of these limits are zero.

In cases such as this, we employ the useful ‘trick’, called


√ rationalisation, of
multiplying the numerator and the denominator by 1 + x + 1 as this gives us
√ √ √
1+x−1 ( 1 + x − 1)( 1 + x + 1) (1 + x) − 1 1
= √ = √ =√ .
x x( 1 + x + 1) x( 1 + x + 1) 1+x+1

Having done this, we now see that the limits of the numerator and the denominator
as x → 0 are both non-zero, and this gives us

1+x−1 1 1 1 1
lim = lim √ =√ = = ,
x→0 x x→0 1+x+1 1+0+1 1+1 2

as the answer.

x2 − 3x + 2
Example 2.13 Find lim .
x→1 1 − x2
We start by noting that, in this case, we can not simply work out the limit of this
quotient by considering the limits of the numerator and the denominator as x → 1
because both of these limits are zero.

In cases such as this, where the numerator and the denominator are polynomials, the

22
2.1. Limits

fact that both of them are zero at x = 1 guarantees that they both have x − 1 as a
factor. So, if we employ the useful ‘trick’ of factorising the numerator and
denominator, we see that we have 2
2
x − 3x + 2 (x − 1)(x − 2) x−2
= = − ,
1 − x2 (1 − x)(1 + x) 1+x

as long as x 6= 1.17 So, as the limit x → 1 only considers what the function is doing
around x = 1, we have

x2 − 3x + 2 x−2 1−2 1
lim 2
= − lim =− = ,
x→1 1−x x→1 1 + x 1+1 2
as the answer.

Infinite limits

If f (x) is a function, we now want to ask what it means to say that this function tends
to infinity as x → a− or as x → a+ , i.e. what it means to say that f (x) → ∞ as x → a−
or as x → a+ which, again abusing our notation, we would write as
lim f (x) = ∞ or lim f (x) = ∞.
x→a− x→a+

Of course, intuitively, as we saw in 174 Calculus, we would want to say this means that
the curve y = f (x) has a vertical asymptote given by x = a, i.e. a vertical line that the
curve gets arbitrarily close to. But, more specifically, when we say that

f (x) → ∞ as x → a− we mean that however large we want f (x) to be, let’s say we
want it to be larger than some real number, M , we can find a value of x, say X, for
which f (x) is larger than M if X < x < a.18
f (x) → ∞ as x → a+ we mean that however large we want f (x) to be, let’s say we
want it to be larger than some real number, M , we can find a value of x, say X, for
which f (x) is larger than M if a < x < X.19
17
Of course, this function is not actually defined at x = 1.
18
Technically, as you will see in 116 Abstract Mathematics, we say that f (x) → ∞ as x → a− , if

for any M > 0, there is a δ > 0 such that, for all x ∈ (a − δ, a), f (x) > M .

Here, the value of M > 0 tells us how large we want f (x) to be and, if we can show that there is a δ > 0
such that f (x) > M for all x in the interval (a − δ, a), then we have established that f (x) is always larger
than M for these values of x below x = a. Indeed, if we can do this for any value of M , we can guarantee
that f (x) is getting ‘arbitrarily large’ as x → a− . However, we will not make use of this formal definition
here.
19
Technically, as you will see in 116 Abstract Mathematics, we say that f (x) → ∞ as x → a+ , if

for any M > 0, there is a δ > 0 such that, for all x ∈ (a, a + δ), f (x) > M .

Here, the value of M > 0 tells us how large we want f (x) to be and, if we can show that there is a δ > 0
such that f (x) > M for all x in the interval (a, a + δ), then we have established that f (x) is always larger
than M for these values of x above x = a. Indeed, if we can do this for any value of M , we can guarantee
that f (x) is getting ‘arbitrarily large’ as x → a+ . However, we will not make use of this formal definition
here.

23
2. Limits

And, if we find that


   
lim − f (x) = ∞ or lim − f (x) = ∞,
2 x→a− x→a+

we say that f (x) tends to minus infinity as x → a− or as x → a+ respectively. That is,


this is what it means to say that f (x) → −∞ as x → a− or as x → a+ which, again
abusing our notation, we could write as
lim f (x) = −∞ or lim f (x) = −∞,
x→a− x→a+

respectively.

Finding infinite limits

Let’s start with a simple example.

1 1
Example 2.14 Find lim− and lim+ .
x→0 x x→0 x

For the first limit, we see that for values of x < 0, the function 1/x is negative and
so −1/x is positive. Indeed, as these x < 0 get closer to zero, we find that −1/x
tends to infinity, i.e.
 
1 1
lim− − = ∞ =⇒ lim− = −∞.
x→0 x x→0 x

For the second limit, we see that for values of x > 0, the function 1/x is positive
and, as these x > 0 get closer to zero, we find that 1/x tends to infinity, i.e.
1
lim+ = ∞.
x→0 x
Of course, as these two limits are not the same, this means that the limit
1
lim ,
x→0 x

does not exist and, indeed, we can see that this function has a vertical asymptote at
x = 0.

But, generally, as we saw above with finite limits, we can find infinite limits by
considering some basic functions that have infinite limits and then, by using some
appropriate rules about how limits work, we can find the limits of certain combinations
of these basic functions. So, we start by stating some infinite limits that arise from basic
functions, i.e.

(a) If f (x) = xb where b < 0 is a constant, then f (x) → ∞ as x → 0+ .


(b) If f (x) = logd x where d > 1 is a constant, then f (x) → −∞ as x → 0+ .
However, as you will see in Activity 2.8, some care must be taken with the analogue of
(a) when we are considering limits as x → 0− .

24
2.1. Limits

Activity 2.8 Suppose that f (x) = xb where b < 0 is a constant. What can you say
about the limit of this function as x → 0− when (i) b = −1, (ii) b = −2 and (iii)
b = −1/2? 2
Then, we have the rules which tell us how to find the limits of certain combinations of
these basic functions which are stated in the following theorem.
Theorem 2.6 Theorem 2.3 holds if x → ∞ is replaced by x → a− , x → a+ or x → a.

Of course, the same caveats apply as the ones we saw after the statement of
Theorem 2.3. As an example of why these rules work, we can think of something similar
to Example 2.13, which will no longer give us a finite limit.

x2 − 3x + 3 x2 − 3x + 3
Example 2.15 Find lim− and lim .
x→1 1 − x2 x→1+ 1 − x2
We start by noting that, in this case, we can not simply work out the limit of this
quotient by considering the limits of the numerator and the denominator as x → 1−
or as x → 1+ because, in both cases, the limit of the denominator is zero.

In cases such as this, where the denominator is a polynomial, the fact that it is zero
at x = 1 guarantees that it has x − 1 as a factor. So, if we employ the useful ‘trick’
of factorising the denominator, we see that we have

x2 − 3x + 3 x2 − 3x + 3
= ,
1 − x2 (1 − x)(1 + x)

and, of course, this function is not actually defined at x = 1. So, as the limit x → 1−
only considers what the function is doing below x = 1, we have

x2 − 3x + 3 x2 − 3x + 3
lim− = lim .
x→1 1 − x2 x→1− (1 − x)(1 + x)

Now, we can see that, without the 1 − x term in the denominator, we have

x2 − 3x + 3 12 − 3 + 3 1
lim = = ,
x→1− 1+x 2 2
and this is positive whereas 1 − x itself tends to zero through positive values as
x → 1− since this gives values of x < 1. Consequently, as we are dividing a positive
number, i.e. 1/2, by smaller and smaller positive numbers, we see that

x2 − 3x + 3 x2 − 3x + 3
lim− = lim = ∞.
x→1 1 − x2 x→1− (1 − x)(1 + x)

Then, applying similar reasoning, we see that as the limit x → 1+ only considers
what the function is doing above x = 1, we have

x2 − 3x + 3 x2 − 3x + 3
lim+ = lim .
x→1 1 − x2 x→1+ (1 − x)(1 + x)

25
2. Limits

So, we still have


x2 − 3x + 3 12 − 3 + 3 1
lim = = ,
2 x→1+ 1+x 2 2
and this is positive whereas 1 − x itself now tends to zero through negative values as
x → 1+ since this gives values of x > 1. Consequently, as we are dividing a positive
number, i.e. 1/2, by smaller and smaller [in magnitude] negative numbers, we see
that
x2 − 3x + 3 x2 − 3x + 3
lim+ = lim = −∞.
x→1 1 − x2 x→1+ (1 − x)(1 + x)

Of course, as these two limits are not the same, this means that the limit

x2 − 3x + 3
lim ,
x→1 1 − x2
does not exist and, indeed, we can see that this function has a vertical asymptote at
x = 1.

The relationship between infinite and finite limits

We have seen that, as x → a− or as x → a+ , some functions have finite limits and


others have infinite limits but now we want to briefly discuss how these two types of
limit are related. The key result here is the following theorem.
Theorem 2.7 (a) If f (x) → ∞ as x → a− , then
1
lim− = 0,
x→a f (x)
and this result also holds if a− is replaced by a+ .

(b) If f (x) → 0 as x → a− and there is a δ ∈ R such that

1
f (x) > 0 for all a − δ < x < a, then lim = ∞.
x→∞ f (x)

1
f (x) < 0 for all a − δ < x < a, then lim = −∞.
x→∞ f (x)

The analogue of this result holds if a− is replaced by a+ .

Of course, the results in this theorem should be fairly obvious and we have used them
implicitly in Example 2.15.

Limits that don’t exist

If we have a function f (x) and we find that

lim f (x) = c,
x→a−

where c is a real number or we find that f (x) → ∞ (or −∞) as x → a− , we say that the
limit of f (x) as x → a− exists. And, of course, we can say similar things in the limit as

26
2.1. Limits

x → a+ or x → a. However, not every function has a limit as x → a− or x → a+ or


x → a and, in such cases, we say that this limit does not exist.
2
Example 2.16 Consider the function
(
x+3 if x ≤ 3
f (x) = .
12 − 3x if x > 3

What can we say about the limit of this function as x → 3?

Here we can see that we may have to worry about what is happening with this
function at x = 3 and so it makes sense to find its limits as x → 3− and as x → 3+ .
In particular, we see that as x → 3− , we are concerned with values of x < 3 that are
getting closer to x = 3 and so we have

lim f (x) = lim− (x + 3) = 3 + 3 = 6,


x→3− x→3

whereas as x → 3+ , we are concerned with values of x > 3 that are getting closer to
x = 3 and so we have

lim f (x) = lim+ (12 − 3x) = 12 − 3(3) = 3.


x→3+ x→3

So, although both of these limits exist, they are not equal and so the limit of f (x) as
x → 3 does not exist.

Of course, the limits that we have found here are obvious if you look at a sketch of
the graph of this function such as the one in Figure 2.5.

6 y = f (x)

O x
−3 3 4

Figure 2.5: The graph of the function f (x) from Example 2.16. (Note that a ‘•’ means
that this point is actually part of the graph of the function whereas a ‘◦’ means that this
point is not actually part of the graph of the function.)

27
2. Limits

2.2 Some useful results that involve limits

2 2.2.1 Continuity
Intuitively, we say that a function is continuous at a point if it has no ‘breaks’ or
‘jumps’ at that point and we can use limits to make this idea more precise. In
particular, if a function, f (x), is such that both
lim f (x) and f (c),
x→c

are defined and, furthermore, we have


lim f (x) = f (c),
x→c

then we say that f (x) is continuous at x = c. Indeed, if the function is continuous at


every point in the interval (a, b), we say that it is continuous on that interval and if the
function is continuous at every point in R, we simply say that it is continuous.

Some continuous functions

Most of the basic functions that we considered in 174 Calculus are continuous, in
particular, it should be clear that the following functions are all continuous over their
domains.

Power functions f : R → R where f (x) = xn for n ∈ N.


Exponential functions f : R → (0, ∞) where f (x) = ax for positive a 6= 1.
Trigonometric functions f : R → [−1, 1] where f (x) = sin x or f (x) = cos x.
Logarithmic functions f : (0, ∞) → R where f (x) = loga (x) for positive a 6= 1.
If you are in any doubt about the continuity of any of these functions, you should
consider what we know about their graphs from 174 Calculus.

Some useful results

Having seen which of our basic functions are continuous, we can then see how continuity
is preserved when we take combinations of these functions by using the following
theorem.
Theorem 2.8 If the functions f (x) and g(x) are continuous at x = c, then so are the
functions

kf (x) where k ∈ R;
f (x) + g(x);
f (x)g(x);
f (x)
as long as g(c) 6= 0.
g(x)

28
2.2. Some useful results that involve limits

Moreover, if f (x) is continuous at x = c and g(x) is continuous at d = f (c),20 then the


composition (g ◦ f )(x) = g(f (x)) is also continuous at x = c.

Lastly, suppose that f −1 (x) is the inverse of a function f (x) which is strictly increasing 2
(or decreasing) over some interval (a, b). If f (x) is continuous at x = c for some
c ∈ (a, b), then f −1 (x) is also continuous at x = f (c).

Let’s take a moment to see how this all works by considering an example.

sin x
Example 2.17 Explain why the function is continuous for all x 6= 0.
x
We know from above that the functions sin x and x are continuous for all x ∈ R and
so, by Theorem 2.6, we see that the function
sin x
,
x
is also continuous as long as x 6= 0. Indeed, it should be clear that this function is
not continuous at x = 0 since it is not even defined there.

Notice, however, that we can sometimes ‘repair’ failures in continuity by taking a little
more care with the definition of a function as the next example shows.

Example 2.18 Following on from Example 2.17, consider the function f : R → R


given by f (0) = 1 and
sin x
f (x) = ,
x
when x 6= 0. Show that this function is continuous for all x ∈ R.
sin x
[Hint: You may assume that → 1 as x → 0. (See Example 2.22.)]
x
We saw in Example 2.17 that f (x) is continuous for all x 6= 0 and, using the hint, we
see that
sin x
lim f (x) = lim = 1 = f (0),
x→0 x→0 x

which means that f (x) is continuous at x = 0 too.

2.2.2 Differentiability
We saw in 174 Calculus if f (x) is a function, then its derivative, f 0 (x), is the function
defined by
f (x + h) − f (x)
f 0 (x) = lim ,
h→0 h
and, if f 0 (c) exists, we say that f (x) is differentiable at x = c. Indeed, if the function is
differentiable at every point in the interval (a, b), we say that it is differentiable on that
interval and if the function is differentiable at every point in R, we simply say that it is
20
So that, in particular, d is in the range of f which is, in turn, taken to be in the domain of g.

29
2. Limits

differentiable. Of course, we explored derivatives and what they tell us about functions
in some detail in 174 Calculus and so we will settle for a simple example of how this
works.
2

Example 2.19 Find the derivative of the function f (x) = x for x > 0.

Using the definition of the derivative, we have


√ √
0 f (x + h) − f (x) x+h− x
f (x) = lim = lim ,
h→0 h h→0 h

where we require that x > 0 so that x is defined. Of course, we can evaluate this
using the rationalisation ‘trick’ that we saw in Example 2.12, i.e. we multiply the
numerator and the denominator of
√ √
x+h− x
,
h
√ √
by x + h + x to get
√ √ √ √  √ √ 
x+h− x x+h− x x+h+ x (x + h) − x 1
= √ √ = √ √ =√ √ ,
h h x+h+ x h( x + h + x) x+h+ x

so that we have
√ √
x+h− x 1 1 1
=√ √ →√ √ = √ ,
h x+h+ x x+ x 2 x

as h → 0. This then means that we have


√ √
0 f (x + h) − f (x) x+h− x 1
f (x) = lim = lim = √ ,
h→0 h h→0 h 2 x

which is the answer we should have been expecting.

Activity 2.9 Using only the definition, find the derivative of the function
f (x) = x−1 for x 6= 0.

One useful thing to note is that ‘differentiability implies continuity’ in the following
sense.
Theorem 2.9 If the function, f (x), is differentiable at x = c, then it is continuous at
x = c.

To see why this works, consider that if the function, f (x), is differentiable at x = c,
then we know that f 0 (c) exists and is given by
f (c + h) − f (c)
f 0 (c) = lim ,
h→0 h
which means that if we consider the limit as h → 0 of the function
f (c + h) − f (c),

30
2.2. Some useful results that involve limits

we can multiply the numerator and denominator by h to get


f (c + h) − f (c)
   
lim f (c + h) − f (c) = lim h = (0)f 0 (c) = 0.
h→0 h→0 h 2
But, this means that
lim f (c + h) = f (c),
h→0
which establishes that f (x) is continuous at x = c.
However, the converse of this theorem is not true as the next activity shows.

Activity 2.10 Show that the function f (x) = |x| is continuous at x = 0, but that it
is not differentiable at that point.

2.2.3 Taylor series and Taylor’s theorem


In 174 Calculus, we saw that the Taylor series for a function, f (x), around some point
x = c was given by
(x − c)2 00 (x − c)n (n)
f (x) = f (c) + (x − c)f 0 (c) + f (c) + · · · + f (c) + · · · ,
2! n!
and, in the case where c = 0, we called this the Maclaurin series for f (x). Indeed, in
that course, we used such series to find approximations to the value of the function f (x)
for values of x close to x = c. In particular, we defined the nth-order approximation to
f (x) around the point x = c to be the polynomial, let’s call it Pn (x), given by
(x − c)2 00 (x − c)n (n)
Pn (x) = f (c) + (x − c)f 0 (c) + f (c) + · · · + f (c).
2! n!
But, we now want to use a result, called Taylor’s theorem, which will allow us to
investigate how accurate such approximations are.
Theorem 2.10 (Taylor’s theorem) If f (x) is a function defined on the interval (a, b)
and all of its derivatives up to f (n+1) (x) exist on (a, b), then for any c, x ∈ (a, b) and any
n ∈ N, we have
(x − c)n+1 (n+1)
f (x) = Pn (x) + f (d),
(n + 1)!
for some d lying [strictly] between c and x. In particular, we call
(x − c)n+1 (n+1)
f (d),
(n + 1)!
the remainder term associated with Pn (x).

Notice that, in Taylor’s theorem, the remainder term tells you about the size of the
difference between f (x) and Pn (x), i.e. the size of the difference between what we are
trying to approximate and the approximation that we have found. As such, if we have
reason to believe that the remainder term is small, then we can be assured that Pn (x) is
a good approximation to f (x). Indeed, using Taylor’s theorem, we can actually find
bounds that determine just how accurate our approximation is as the next example
shows.

31
2. Limits

Example 2.20 Find the Taylor series for e−x about x = 0 and use it to find a
third-order approximation to e−1 . Then use Taylor’s theorem to find an upper bound
2 on the difference between the true value of e−1 and this approximation to it.

If we take f (x) = e−x we have

f 0 (x) = − e−x , f 00 (x) = e−x , f 000 (x) = − e−x ,

and, spotting the pattern, we see that f (n) (x) = (−1)n e−x for n ≥ 1. Thus, we have
f (0) = 1 and f (n) (0) = (−1)n for n ≥ 1, which means that

−x (x − 0)2 (x − 0)3 (x − 0)n


e = 1 + (x − 0)(−1) + (1) + (−1) + · · · + (−1)n + · · ·
2! 3! n!
x2 x3 x n
=1−x+ − + · · · + (−1)n + · · · ,
2! 3! n!
is the Taylor series for e−x about x = 0. Indeed, using this, we can see that a
third-order approximation to e−1 is given by
1 1 1 1 1
1−1+ − = − = ,
2! 3! 2 6 3
i.e. our third-order approximation to e−1 is 0.3333 to 4dp.

Referring to Taylor’s theorem, we can then see that the remainder term in this case
is given by
(1 − 0)4 −d e−d
e = ,
4! 24
for some d lying between c = 0 and x = 1. That is, using Taylor’s theorem, we have
found that
1 e−d
e−1 = + ,
3 24
for some d ∈ (0, 1). This means that the difference between the true value of e−1 and
our third-order approximation to it is given by

1 e−d 1
e−1 − = < ,
3 24 24
as d > 0 means that e−d < 1. Thus, the difference between the true value of e−1 and
our approximation is at most 1/24 (or 0.0417 to 4dp) and so this is the required
upper bound.

Incidentally, the true value of e−1 is 0.3679 to 4dp and so the difference between the
true value of e−1 and our approximation is
1
e−1 − = 0.3679 − 0.3333 = 0.0346,
3
and, as expected, this difference is less than the upper bound that we found for this
quantity.

32
2.2. Some useful results that involve limits

If we extend this idea, we can see that Taylor’s theorem can also be used to find useful
bounds on functions as the next example illustrates.
2
Example 2.21 Use Taylor’s theorem to show that

x2
1 − x < e−x < 1 − x + ,
2
when x > 0.

As we’ll soon see, it makes sense to use the first-order Taylor series for e−x around
x = 0 and the associated remainder term, i.e. using Taylor’s theorem and what we
saw at the beginning of Example 2.20, we have

x2 −d
e−x = 1 − x + e ,
2
for some d ∈ (0, x) where x > 0. Now, as d > 0, we have e−d < 1 and so we can write

x2 −d x2
e−x = 1 − x + e <1−x+ ,
2 2
but, we also know that e−d > 0 for all d ∈ R, and so we can also write

x2 −d
e−x = 1 − x + e > 1 − x.
2
So, putting these two inequalities together, we get

x2
1 − x < e−x < 1 − x + ,
2
when x > 0, as required.

And, although we won’t dwell too much on it here, we can see that the values of x for
which the Taylor series ‘works’ — an idea we encountered briefly in 174 Calculus — are
those values of x for which the remainder term tends to zero as n → ∞. This is because,
if we have a value of x for which the remainder term tends to zero as n → ∞, then it
should be clear that Pn (x) tends to f (x) as n → ∞. That is, the value of f (x) is the
same as the value of
n
X (x − c)i (i)
lim Pn (x) = lim f (c),
n→∞ n→∞
i=0
i!

which is just what we get when we keep all of the terms in the Taylor series.

Using Taylor series to find limits

One particularly important application of Taylor series in this course is that they allow
us to find certain limits. In particular, if we have the Taylor series of some function,
f (x), about the point x = c, we can often use it to deduce the limit of expressions that
involve f (x) as x → c simply because, as x gets closer to c, we would expect f (x) to

33
2. Limits

take values that get closer to the values that arise from its Taylor series. Let’s consider
an example of how this works.
2
sin x
Example 2.22 Use the Taylor series for sin x about x = 0 to find lim .
x→0 x

We know that the Taylor series for sin x about x = 0 gives us

x3
sin x = x − + ··· ,
3!
and so, we have
x3
sin x x− + ··· x2
= 3! =1− + ··· .
x x 3!
Now, as x → 0, we see that the x2 term (and all the higher-order terms that we have
omitted) should tend to zero,21 leaving us with

x2
 
sin x
lim = lim 1 − + · · · = 1,
x→0 x x→0 3!

as the answer.

In particular, we observe that this method is especially useful when we are looking at
limits as x → 0 because we know a lot about the Taylor series of our basic functions
about x = 0 from Section 3.4 of 174 Calculus.

Activity 2.11 Use the appropriate Taylor series to find the following limits.

1 − cos x ln(1 + x)
(a) lim , (b) lim .
x→0 x2 x→0 x

We now consider another important method for working out certain limits and, as we
shall see, this also implicitly relies on the use of Taylor series.

2.2.4 L’Hôpital’s rule


Following on from our earlier discussion of limits, we know that there are still certain
cases which we can not deal with. For instance, suppose that we have to find the limit

f (x)
lim+ ,
x→a g(x)

in the case where f (x) and g(x) both tend to zero as x → a+ . In such cases, we can’t
use Theorem 2.5 because g(x) → 0 as x → a+ but, as long as the stated conditions hold,
we can use L’Hôpital’s rule which runs as follows.
21
If necessary, we can make this idea precise by using the appropriate remainder term, but we will
generally be content with the more ‘intuitive’ calculation that we have presented here.

34
2.2. Some useful results that involve limits

Theorem 2.11 (L’Hôpital’s rule: first form) Suppose that the functions f and g are
differentiable on the interval (a, b) and that g 0 (x) 6= 0 for x ∈ (a, b). If

both lim+ f (x) = 0 and lim+ g(x) = 0, and 2


x→a x→a

f 0 (x)
lim+ = L where L is a real number or ∞ or −∞,22
x→a g 0 (x)
then
f (x)
lim+ = L,
x→a g(x)
too. Indeed, analogues of this rule hold when lim+ is replaced by
x→a

lim or lim,
x→b− x→c

for some c ∈ (a, b) as well as when we have a is −∞ or b is ∞.

We won’t prove this rule here, but we can see why it works by using Taylor’s theorem.
For instance, looking at f (x) about x = a, Taylor’s theorem dictates that

f (x) = f (a) + (x − a)f 0 (d1 )

for some d1 ∈ (a, x) as we are interested in x ∈ (a, b), whereas looking at g(x) about
x = a, Taylor’s theorem dictates that

g(x) = g(a) + (x − a)g 0 (d2 ),

for some d2 ∈ (a, x) as, again, we are interested in x ∈ (a, b). This means that, looking
at our quotient, we have
f (x) f (a) + (x − a)f 0 (d1 )
= .
g(x) g(a) + (x − a)g 0 (d2 )
But, we are given that

lim f (x) = 0 and lim g(x) = 0,


x→a+ x→a+

and so we have f (a) = 0 and g(a) = 0,23 which means that our quotient becomes
f (x) (x − a)f 0 (d1 ) f 0 (d1 )
= = ,
g(x) (x − a)g 0 (d2 ) g 0 (d2 )
as we can assume that x 6= a if we are taking the limit as x → a+ . Thus, we have
f (x) f 0 (d1 ) f 0 (x)
lim+ = lim+ 0 = lim+ 0 ,
x→a g(x) x→a g (d2 ) x→a g (x)
provided that, as assumed in Theorem 2.11, this last limit exists.24

22
That is, this limit exists.
23
Observe that, using Theorem 2.9, the fact that f and g are differentiable on (a, b) implies that they
must be continuous on (a, b) as well.
24
Observe that the last equality arises because, if d1 and d2 are in the interval (a, x), we can see that
they must both tend to a from above when x → a+ .

35
2. Limits

Activity 2.12 Given that f (x) and g(x) are differentiable functions which both
tend to zero as x → ∞, use the substitution x = 1/t and Theorem 2.11 to show that
2
f (x) f 0 (x)
lim = lim 0 ,
x→∞ g(x) x→∞ g (x)

provided that the limit on the right-hand side exists.

Let’s now look at some examples of how L’Hôpital’s rule can be applied.

1 − cos x ln(1 + x)
Example 2.23 Use L’Hôpital’s rule to find lim and lim .
x→0 x x→0 x
For the first limit, we note that the numerator and the denominator both tend to
zero as x → 0 and we also have
f (x) 1 − cos x f 0 (x) sin x
= =⇒ 0
= ,
g(x) x g (x) 1

where, as x → 0, the limit of this second quotient is zero. So, using L’Hôpital’s rule,
we have
1 − cos x sin x
lim = lim = 0,
x→0 x x→0 1

as the answer.

For the second limit, we again note that the numerator and the denominator both
tend to zero as x → 0 and we also have
f (x) ln(1 + x) f 0 (x) 1/(1 + x)
= =⇒ 0
= ,
g(x) x g (x) 1

where, as x → 0, the limit of this second quotient is one. So, using L’Hôpital’s rule,
we have
ln(1 + x) 1/(1 + x)
lim = lim = 1,
x→0 x x→0 1
as the answer in agreement with what we saw in Activity 2.11(b)

Activity 2.13 Following on from Example 2.22, use L’Hôpital’s rule to verify that
sin x
lim = 1.
x→0 x

Another case where we run into problems involves the limit


f (x)
lim+ ,
x→a g(x)
in the case where f (x) and g(x) both tend to infinity as x → a+ . In such cases, we can’t
use Theorem 2.7 because this case isn’t covered but, as long as the stated conditions
hold, we can use another form of L’Hôpital’s rule which runs as follows.

36
2.2. Some useful results that involve limits

Theorem 2.12 (L’Hôpital’s rule: second form) Suppose that the functions f and g
are differentiable on the interval (a, b) and that g 0 (x) 6= 0 for x ∈ (a, b). If







2
both lim+ f (x) = ∞ and lim+ g(x) = ∞, and
x→a x→a

f 0 (x)
lim+ = L where L is a real number or ∞ or −∞,25
x→a g 0 (x)
then
f (x)
lim+
= L,
x→a g(x)

too. Indeed, analogues of this rule hold when lim+ is replaced by


x→a

lim or lim,
x→b− x→c

where c ∈ (a, b) as well as when we have a is −∞ or b is ∞.

To get a sense of why this works observe that, if the limits of f (x) and g(x) are both
infinity as x → a+ , we can see that 1/f (x) and 1/g(x) both tend to zero as x → a+ and
so, using Theorem 2.11, we have
2 
g 0 (x)/[g(x)]2 g 0 (x)
 
f (x) 1/g(x) f (x)
lim = lim+ = lim+ 0 = lim+ lim ,
x→a+ g(x) x→a 1/f (x) x→a f (x)/[f (x)]2 x→a g(x) x→a+ f 0 (x)

assuming that these limits exist. Then, if we also have (for the sake of argument) some
L 6= 0 such that
f 0 (x)
lim+ 0 = L,
x→a g (x)

we see that this gives us


f (x)
lim+ = L,
x→a g(x)

too. Of course, this isn’t a proof of Theorem 2.12, but it does indicate how it is related
to Theorem 2.11 if we aren’t being that careful. Let’s look at an example of how it can
be applied.

ln(1 + x)
Example 2.24 Use L’Hôpital’s rule to find lim .
x→∞ x
We note that the numerator and the denominator both tend to infinity as x → ∞
and we also have
f (x) ln(1 + x) f 0 (x) 1/(1 + x)
= =⇒ 0
= ,
g(x) x g (x) 1

where, as x → ∞, the limit of this second quotient is zero. So, using L’Hôpital’s rule,
we have
ln(1 + x) 1/(1 + x)
lim = lim = 0,
x→∞ x x→∞ 1
as the answer.
25
That is, this limit exists.

37
2. Limits

π 
Activity 2.14 Use L’Hôpital’s rule to find lim x − tan−1 x .
x→∞ 2
2
Lastly, as you will see in the next activity, some care must be taken when finding limits
with L’Hôpital’s rule as it is not always applicable.

Activity 2.15 Find the limits


ln x x + sin x
(a) lim+ and (b) lim .
x→0 x x→∞ x
Why is L’Hôpital’s rule not applicable for either of these limits?

Learning outcomes
At the end of this chapter and having completed the relevant reading and activities, you
should be able to:

find limits or explain why they don’t exist;


assess whether a function is continuous;
find derivatives using the definition of the derivative;
use Taylor series to find limits and Taylor’s theorem to find bounds;
use L’Hôpital’s rule to find limits.

Solutions to activities
Solution to activity 2.1
We are given that f (x) → l and g(x) → m as x → ∞ where l, m ∈ R, and so using
Theorem 2.1(a) with c = −1 and (b), we see that
f (x) − g(x) = f (x) + [−g(x)] → l + [−m] = l − m as x → ∞,
and, for any constants c and d, we also have
cf (x) + dg(x) = [cf (x)] + [dg(x)] → cl + dm as x → ∞,
again using Theorem 2.1(a) and (b).

Solution to activity 2.2


For (a), we follow the method in Example 2.3 and divide the numerator and
denominator by the highest power of x that occurs in the given quotient. In this case,
the highest power of x in the quotient is x3 and so we get
x2 + x + 1 (1/x) + (1/x2 ) + (1/x3 ) 0+0+0 0
lim = lim = = = 0.
x→∞ x + x2 + x3 x→∞ 2
(1/x ) + (1/x) + 1 0+0+1 1

38
2.2. Solutions to activities

For (b), we note that the highest power of x in the quotient is x1/2 and so, using the
same method as in (a), we get

(x + sin x − 2)1/2 1 + sinx x − x2


1/2 2
lim √ = lim .
x→∞ x + sin x − 2 √ x − √2
x→∞ 1 + sin
x x

Now, from Example 2.4, we know that


sin x
lim = 0,
x→∞ x

and so, as x → ∞, the numerator has a limit given by (1 + 0 + 0)1/2 = 1 if we use


Theorem 2.1(e). Further, in the denominator, we see that
sin x
lim √ = 0,
x→∞ x

if, following Example 2.4, we use the Sandwich theorem with the inequality
1 sin x 1
−√ ≤ √ ≤ √ ,
x x x

which clearly holds for x > 0. So, overall, we have


1/2
(x + sin x − 2)1/2 1 + sinx x − x2 1
lim √ = lim sin x 2
= = 1,
x→∞ x + sin x − 2 x→∞ 1 + √ − √
x x
1+0+0

as our final answer.

Solution to activity 2.3


Given that f (x) → ∞ as x → ∞, we see that the analogue of Theorem 2.3(a) when
c < 0 is
cf (x) → −∞ as x → ∞,
whereas, given that g(x) → m as x → ∞ where m < 0 is a real number, we see that the
analogues of Theorem 2.3(d)-(f) are

f (x)
f (x) + g(x) → ∞, f (x)g(x) → −∞ and → −∞,
g(x)
as x → ∞.

We also see that, if 0 < d < 1, we have (say) ln d < 0 and ln x → ∞ as x → ∞ so, using
the ‘change of base formula’ for logarithms, we have
ln x
logd x = → −∞,
ln d
as x → ∞ using our analogue of Theorem 2.3(a) from above.

39
2. Limits

Solution to activity 2.4


For (i), we consider the analogue of Theorem 2.3(a) and note that regardless of the fact
2 that f (x) → ∞ as x → ∞, if we take c = 0, we have

cf (x) = 0 → 0 as x → ∞.

For (ii), we see that if f (x) → ∞ and g(x) → −∞ as x → ∞, then the analogue of
Theorem 2.3(b) tells us nothing since we can make no sense of ‘∞ − ∞’. But, the
analogue of Theorem 2.3(c) is

f (x)g(x) → −∞ as x → ∞,

as the result of the product will be negative and arbitrarily large in magnitude.

For (iii), we see that if f (x) → ∞ and g(x) → 0 as x → ∞, then the analogue of
Theorem 2.3(d) is
f (x) + g(x) → ∞ as x → ∞,
but the analogue of Theorem 2.3(e) tells us nothing since we can make no sense of
‘∞ · 0’. The analogue of Theorem 2.3(f) also tells us nothing in this case as we can make
no sense of ‘∞/0’.26

Solution to activity 2.5


For (a), notice that, as it stands, we can make no sense of x2 − x3 as x → ∞ because we
can make no sense of ‘∞ − ∞’. However, if we consider that the highest power of x here
is x3 (i.e. this is the ‘dominant term’), we see that using the analogue of Theorem 2.3(e)
that we saw in Activity 2.3, we have
 
2 3 3 1
lim (x − x ) = lim x − 1 = −∞,
x→∞ x→∞ x

since, as x → ∞, the first term in the product tends to ∞ whereas the second term
term tends to −1.

For (b), following on from what we saw in Example 2.6, we see that dividing the
numerator and denominator by the highest power of x in the denominator, i.e. x, we get
 
sin x
lim x −
x2 − sin x x − sinx x x→∞ x
lim = lim = = ∞,
x→∞ x + sin x x→∞ 1 + sin x 1
x

if we use Theorem 2.3(f) and the result from Example 2.4.

26
However, as we will see in Theorem 2.4, we can make some sense of this if we have some information
about the sign of g(x).

40
2.2. Solutions to activities

Solution to activity 2.6


For (i), let’s consider the function

f (x) = x(1 + sin x),


2
for x ≥ 0 so that, since −1 ≤ sin x ≤ 1, we have

0 ≤ 1 + sin x ≤ 2 =⇒ 0 ≤ x(1 + sin x) ≤ 2x =⇒ 0 ≤ f (x) ≤ 2x,

and so the graph of the curve y = f (x) is always between the straight lines y = 0 (i.e.
the x-axis) and y = 2x. In particular, given the sin x term in f (x), we would expect this
curve to ‘oscillate’ between these two straight lines. Consequently, we should anticipate
that a rough sketch of this curve would look something like the one in Figure 2.6(a).
Indeed, with this sketch, it should be clear that the limit

lim x(1 + sin x),


x→∞

does not exist since, as x → ∞, f (x) neither stays arbitrarily close to any finite value
(i.e. there is clearly no finite limit) nor is it always getting arbitrarily large (i.e. it
clearly doesn’t tend to infinity).

For (ii), let’s consider the function

g(x) = x(2 + sin x),

for x ≥ 0 so that, since −1 ≤ sin x ≤ 1, we have

1 ≤ 2 + sin x ≤ 3 =⇒ x ≤ x(2 + sin x) ≤ 3x =⇒ x ≤ g(x) ≤ 3x,

and so the graph of the curve y = g(x) is always between the straight lines y = x and
y = 3x. In particular, given the sin x term in g(x), we would expect this curve to
‘oscillate’ between these two straight lines. Consequently, we should anticipate that a
rough sketch of this curve would look something like the one in Figure 2.6(b). Indeed,
with this sketch, it should be clear that we have

lim x(2 + sin x) = ∞,


x→∞

as, for x ≥ 0, g(x) ≥ x and so g(x) must tend to infinity as x → ∞.

Solution to activity 2.7


For (i), we see that the limit r
sin x
lim ,
x→∞ x
does not exist because, although
sin x
lim = 0,
x→∞ x

as we saw in Example 2.4, we cannot take the limit of


r
sin x
,
x

41
2. Limits

(a) (b)
Figure 2.6: The sketches for Activity 2.6. (a) The dashed line is y = 2x and the solid
curve that ‘oscillates’ between this line and the x-axis (i.e. the line y = 0) is y = f (x).
(b) The dotted line is y = x, the dashed line is y = 3x and the solid curve that ‘oscillates’
between these two lines is y = g(x).

as x → ∞ because this is not a well-defined function from R to R. In particular, we see


that if we want to consider its values (and whether they are staying arbitrarily close to
some limiting number — like, say, zero) for suitably large values of x, we find that for
some of those values of x, sin x will be negative and so the square root will not return a
value (i.e. a real number) for comparison.27 See, for instance, the rough sketch in
Figure 2.7(a).

For (ii), we see that the function


s
sin x
f (x) = ,
x

is well-defined for all x > 0 and so, unlike with (i), we can begin to make sense of the
limit of f (x) as x → ∞. Indeed, since 0 ≤ | sin x| ≤ 1 for all x > 0, we can see that

1
0 ≤ f (x) ≤ √ ,
x

and so, as 1/ x → 0 as x → ∞, we can conclude
s
sin x
lim = 0,
x→∞ x
27
Notice, in particular, that this is a case where Theorem 2.1(e) fails because with b = 1/2 and

sin x
l = lim f (x) = lim = 0,
x→∞ x→∞ x
we can not make sense of
lim [f (x)]b ,
x→∞

even though lb = 01/2 = 0. This is why we insist that l > 0 in the statement of the theorem. (Because
l > 0 guarantees that f (x) > 0 for suitably large values of x and so we can meaningfully find the
corresponding values of, say, [f (x)]1/2 .)

42
2.2. Solutions to activities

using Theorem 2.2.28 Of course, this can be seen very easily if we consider the rough
sketch in Figure 2.7(b).
2

(a) (b)

Figure 2.7: The sketches for Activity 2.7. (a) The dashed curve is y = 1/ x and the
solid
p curve that ‘oscillates’ between this curve and the x-axis (i.e. the line y = 0) is y =
(sin x)/x where it exists! In particular, for (2k −1)π < x < 2kπ with k ∈ N, although
it may look like this curve is giving us y = 0 in the sketch, this is misleading because
√ it
does not actually exist for these values of x! (b) The dashed curve is y = 1/ x and
the solid curve that ‘oscillates’ between this curve and the x-axis (i.e. the line y = 0) is
y = f (x).

Solution to activity 2.8


For (i), we have f (x) = x−1 , and as we saw in Example 2.14, we have f (x) → −∞ as
x → 0− .

For (ii), we have f (x) = x−2 , and thinking about this in light of Example 2.14, we see
that when x < 0, the function 1/x2 is positive. Indeed, as we take values of x < 0 that
get closer to zero, we find that 1/x2 takes larger and larger positive values. That is,
f (x) → ∞ as x → 0− in this case.

For (iii), we have f (x) = x−1/2 and, if x < 0, this is not a well-defined function from R
to R as we have to take the square root of a negative number. Consequently, the limit of
f (x) as x → 0− can not exist either.

Solution to activity 2.9


Given the function f (x) = x−1 for x 6= 0, we use the definition of the derivative to see
that
f (x + h) − f (x) (x + h)−1 − x−1
f 0 (x) = lim = lim ,
h→0 h h→0 h
where we have x 6= 0 so that x−1 is defined. Of course, we can then see that
1 1
(x + h)−1 − x−1 − −h −1
= x+h x = = ,
h h x(x + h)h x(x + h)
28
In particular, with reference to the previous footnote, we still can’t use Theorem 2.1(e) here because
‘l’ is still zero!

43
2. Limits

as we have h 6= 0 when we are considering the limit as h → 0. This then means that we
have
−1 1
2 f 0 (x) = lim = − 2,
h→0 x(x + h) x
which is the answer we should have been expecting.

Solution to activity 2.10


We see that the function f (x) = |x| is continuous at x = 0 because

lim f (x) = lim |x| = |0| = 0 = f (0).


x→0 x→0

However, if we consider the definition of the derivative at x = 0, we need to look at

f (0 + h) − f (0) |h| − 0 |h|


= = ,
h h h
so that, when h < 0, we have |h| = −h and

f (0 + h) − f (0) |h| −h
lim− = lim− = lim− = lim− (−1) = −1,
h→0 h h→0 h h→0 h h→0

whereas, when h > 0, we have |h| = h and

f (0 + h) − f (0) |h| h
lim+ = lim+ = lim+ = lim+ 1 = 1.
h→0 h h→0 h h→0 h h→0

Consequently, we see that f (x) is not differentiable at x = 0 because

f (0 + h) − f (0)
f 0 (0) = lim ,
h→0 h
does not exist since the limits as h → 0− and as h → 0+ are not equal.

Solution to activity 2.11


For (a), we know that the Taylor series for cos x about x = 0 gives us

x2 x4
cos x = 1 − + − ··· ,
2! 4!
and so, we have

x2 x 4
 
1− 1− + − ···
1 − cos x 2! 4! 1 x2
= = − + ··· .
x2 x2 2! 4!
Now, as x → 0, we see that the x2 term (and all the higher-order terms that we have
omitted) should tend to zero, leaving us with

1 − cos x x2
 
1 1
lim 2
= lim − + ··· = ,
x→0 x x→0 2! 4! 2

as the answer.

44
2.2. Solutions to activities

For (b), we know that the Taylor series for ln(1 + x) about x = 0 gives us

x2
ln(1 + x) = x −
2
+ ··· , 2
and so, we have
x2
ln(1 + x) x−+ ··· x
= 2 = 1 − + ··· .
x x 2
Now, as x → 0, we see that the x term (and all the higher-order terms that we have
omitted) should tend to zero, leaving us with

ln(1 + x)  x 
lim = lim 1 − + · · · = 1,
x→0 x x→0 2
as the answer.

Solution to activity 2.12


Essentially, the substitution x = 1/t gives us t → 0+ as x → ∞ and

f (x) f (1/t)
lim = lim+ ,
x→∞ g(x) t→0 g(1/t)

where, as f (x) and g(x) both tend to zero as x → ∞, we see that the functions f (1/t)
and g(1/t) must also both tend to zero as t → 0+ . So, using the chain rule,
Theorem 2.11 gives us

f (1/t) f 0 (1/t)(−1/t2 ) f 0 (1/t) f 0 (x)


lim+ = lim+ 0 = lim = lim ,
t→0 g(1/t) t→0 g (1/t)(−1/t2 ) t→0+ g 0 (1/t) x→∞ g 0 (x)

if we use the substitution x = 1/t again. So, putting this all together, we have shown
that
f (x) f 0 (x)
lim = lim 0 ,
x→∞ g(x) x→∞ g (x)

provided that the limit on the right-hand side exists.

Solution to activity 2.13


To find the given limit, we note that the numerator and the denominator both tend to
zero as x → 0 and we also have
f (x) sin x f 0 (x) cos x
= =⇒ = ,
g(x) x g 0 (x) 1

where, as x → 0, the limit of this second quotient is one. So, using L’Hôpital’s rule, we
have
sin x cos x
lim = lim = 1,
x→0 x x→0 1
as the answer in agreement with what we saw in Example 2.22.

45
2. Limits

Solution to activity 2.14


In order to use L’Hôpital’s rule to find the given limit, we first need to write the given
2 product as a quotient. Of course, the sensible way to do this is to write it as
π
π  − tan−1 x
lim x − tan−1 x = lim 2
,
x→∞ 2 x→∞ 1/x
so that we now have a quotient where the numerator and the denominator both tend to
zero as x → ∞. We also have
π
f (x) 2
− tan−1 x f 0 (x) −1/(1 + x2 ) x2 1
= =⇒ = = = 1 ,
g(x) 1/x g 0 (x) −1/x2 1 + x2 x2
+1

where, as x → ∞, the limit of this second quotient is one. So, using L’Hôpital’s rule, we
have
π

−1

2
− tan−1 x −1/(1 + x2 ) x2
lim x − tan x = lim = lim = lim = 1,
x→∞ 2 x→∞ 1/x x→∞ −1/x2 x→∞ 1 + x2

as the answer.

Solution to activity 2.15


For (a), L’Hôpital’s rule is not applicable because, even though we have a quotient, the
numerator tends to minus infinity as x → 0+ whereas the denominator tends to zero as
x → 0+ . However, we can see that
 
ln x 1
lim+ = lim+ ln x = −∞,
x→0 x x→0 x

using the analogue of Theorem 2.3(c) that we saw in Activity 2.4(ii).

For (b), we see that L’Hôpital’s rule is not applicable because, even though we have a
quotient where the numerator and denominator both tend to infinity as x → ∞, we find
that
f (x) x + sin x f 0 (x) 1 + cos x
= =⇒ 0
= ,
g(x) x g (x) 1
and, as x → ∞, the limit of this second quotient does not exist. However, we can see
that  
x + sin x sin x
lim = lim 1 + = 1 + 0 = 1,
x→∞ x x→∞ x
if we use the result from Example 2.4.

Exercises
Exercise 2.1
Find functions f and g which are such that, as x → ∞, f (x) → ∞ and g(x) → 0 but

(a) f (x)g(x) → ∞,

46
2.2. Solutions to exercises

(b) f (x)g(x) → 0,

(c) f (x)g(x) → −2,


2
(d) the limit of f (x)g(x) does not exist.

Exercise 2.2
Evaluate the following limits.
   
2 1/2 3 2 1/3
(a) lim (x + x) − x , (b) lim (x + x ) −x .
x→∞ x→∞

Exercise 2.3
Use L’Hôpital’s rule to determine the following limits.

ln(1 − cos x)
(a) lim , (b) lim (1 − cos x)tan x .
x→π/2 cos x x→π/2

Exercise 2.4
Using only the definition, find the derivative of the function f (x) = ln x for x > 0.

Exercise 2.5
For x 6= 0, consider the function

1 − cos(2x)
f (x) = .
2x2
Find the limit of f (x) as x → 0 and hence define a function g(x) that is continuous for
all x ∈ R which is the same as f (x) for all x 6= 0. Hence use the definition of the
derivative to find g 0 (0).

Solutions to exercises
Solution to exercise 2.1
We can easily see that if we pick the functions f and g as follows they will satisfy the
requirement that, as x → ∞, f (x) → ∞ and g(x) → 0, as well as the additional
requirement stated in the question.

1
(a) f (x) = x2 and g(x) = gives us f (x)g(x) = x → ∞ as x → ∞.
x
1 1
(b) f (x) = x and g(x) = gives us f (x)g(x) = → 0 as x → ∞.
x2 x
2
(c) f (x) = x and g(x) = − gives us f (x)g(x) = −2 → −2 as x → ∞.
x

47
2. Limits

sin x
(d) f (x) = x and g(x) = gives us f (x)g(x) = sin x.
x
Of course, in (d), the limit of f (x)g(x) does not exist as x → ∞.
2
The lesson here is that, as we observed in Activity 2.4(iii), if we only know that
f (x) → ∞ and g(x) → 0 as x → ∞, we can infer nothing about the limit of f (x)g(x) as
x → ∞.

Solution to exercise 2.2


For (a), we can use the rationalisation ‘trick’ from Example 2.12 to see that

[(x2 + x)1/2 − x][(x2 + x)1/2 + x] (x2 + x) − x2 x


(x2 + x)1/2 − x = 2 1/2
= 2 1/2
= 2 .
(x + x) + x (x + x) + x (x + x)1/2 + x

We then observe that, as in Example 2.3, we can divide the numerator and the
denominator by the highest power of x that occurs in the quotient, i.e. x itself, to get

x 1 1 1
= 1/2
→ = ,
(x2 1/2
+ x) + x 1 + x1 1+1 2

+1

as x → ∞. That is, we have shown that


 
2 1/2 1
lim (x + x) − x = .
x→∞ 2

For (b), we can also rationalise29 to get

[(x3 + x2 )1/3 − x][(x3 + x2 )2/3 + x(x3 + x2 )1/3 + x2 ]


(x3 + x2 )1/3 − x =
(x3 + x2 )2/3 + x(x3 + x2 )1/3 + x2
(x3 + x2 ) − x3
=
(x3 + x2 )2/3 + x(x3 + x2 )1/3 + x2
x2
= .
(x3 + x2 )2/3 + x(x3 + x2 )1/3 + x2

We can then divide the numerator and the denominator by the highest power of x that
occurs in the quotient, i.e. x2 , to get

x2 1 1 1
3 2 2/3 3 2 1/3 2
= 2/3 1/3
→ = ,
(x + x ) + x(x + x ) + x 1 + x1 + 1 + x1 1+1+1 3
 
+1
29
The rationalisation ‘trick’ in (a) takes ‘a − b’ and uses the well-known difference of two squares, i.e.

(a − b)(a + b) = a2 − b2 ,

to ‘remove’ the square root. In (b), the rationalisation ‘trick’ also takes ‘a − b’ but we need to use the
less well-known difference of two cubes, i.e.

(a − b)(a2 + ab + b2 ) = a3 − b3 ,

to ‘remove’ the cube root.

48
2.2. Solutions to exercises

as x → ∞. That is, we have shown that


 
3 2 1/3 1
lim (x + x ) − x = .
x→∞ 3 2
Incidentally, notice that, if we take n ∈ N we can consider the function
" 1/n #
1
(xn + xn−1 )1/n − x = x 1 + −1 ,
x

which replicates what we had in (a) and (b) if we take n = 2 and n = 3 respectively.
Now, for suitably large x, 1/x is close to zero and so we can use the binomial theorem30
to see that 1/n 1 1
( − 1)

1 1 n n
1+ =1+ + + ··· ,
x nx 2! x2
which means that
1 1 1 1
( − 1) ( − 1)
  
n n−1 1/n 1 n n 1 n n
(x + x ) −x=x 1+ + 2
+ · · · − 1 = + + ··· .
nx 2! x n 2! x

Consequently, as x → ∞, we find that the term in 1/x (and all of the neglected terms)
tend to zero, leaving us with
 
2 1/n 1
lim (x + x) − x = ,
x→∞ n
in agreement with what we saw above.

Solution to exercise 2.3


For (a), we see that the numerator and the denominator both tend to zero as x → π/2
and we also have
f (x) ln(1 − cos x) f 0 (x) (sin x)/(1 − cos x) −1
= =⇒ = = ,
g(x) cos x g 0 (x) − sin x 1 − cos x

where, as x → π/2, the limit of this second quotient is −1. So, using L’Hôpital’s rule,
we have
ln(1 − cos x) (sin x)/(1 − cos x) −1
lim = lim = lim = −1,
x→π/2 cos x x→π/2 − sin x x→π/2 1 − cos x

as the answer.

For (b), we note that we can write

(1 − cos x)tan x = e(tan x) ln(1−cos x) ,

and, as x → π/2, we have

ln(1 − cos x)
(tan x) ln(1 − cos x) = (sin x) → (1)(−1) = −1,
cos x
30
See Example 3.25 in Section 3.4.1 of 174 Calculus.

49
2. Limits

if we use the answer to (a). Consequently, we find that

lim (1 − cos x)tan x = lim e(tan x) ln(1−cos x) = e−1 ,


2 x→π/2 x→π/2

is the answer.

Solution to exercise 2.4


We are given f (x) = ln x and, for x > 0, the definition of the derivative gives us

f (x + h) − f (x) ln(x + h) − ln x
f 0 (x) = lim = lim ,
h→0 h h→0 h
and, as the numerator and the denominator of this quotient both tend to zero as h → 0,
we can use L’Hôpital’s rule to see that
1
1 1
f 0 (x) = lim x + h = lim = ,
h→0 1 h→0 x+h x
which is the answer we should expect.

Solution to exercise 2.5


We are given the function
1 − cos(2x)
f (x) = ,
2x2
for x 6= 0 and so we can see that, as the numerator and the denominator of this quotient
tend to zero as x → 0, we can use L’Hôpital’s rule to see that

1 − cos(2x) 2 sin(2x) sin u


lim f (x) = lim 2
= lim = lim = 1,
x→0 x→0 2x x→0 4x u→0 u
if we let u = 2x and use the result from Example 2.22. This means that if we define the
function, g : R → R, to be such that g(0) = 1 and g(x) = f (x) for x 6= 0, we will have a
function that is continuous for all x ∈ R. In particular, g(x) is continuous for all x 6= 0
because f (x) is by Theorem 2.8 and, as

lim g(x) = lim f (x) = 1 = g(0),


x→0 x→0

g(x) is continuous at x = 0 too.

We can now use the definition of the derivative to see that


1 − cos(2h)
g(0 + h) − g(0) −1 1 − cos(2h) − 2h2
g 0 (0) = lim = lim 2h2 = lim ,
h→0 h h→0 h h→0 2h3
and, as the numerator and the denominator of this quotient tend to zero as h → 0, we
can use L’Hôpital’s rule to see that

0 + 2 sin(2h) − 4h sin(2h) − 2h
g 0 (0) = lim 2
= lim ,
h→0 6h h→0 3h2

50
2.2. Solutions to exercises

and, as the numerator and the denominator of this quotient also tend to zero as h → 0,
we can use L’Hôpital’s rule again to see that

2 cos(2h) − 2 cos(2h) − 1 2
g 0 (0) = lim = lim ,
h→0 6h h→0 3h
and, as the numerator and the denominator of this quotient also tend to zero as h → 0,
we can use L’Hôpital’s rule yet again to see that

−2 sin(2h)
g 0 (0) = lim = 0,
h→0 3
is the final answer. Of course, this is obvious if we use the Taylor series for cos(2h)
about h = 0 to observe that
(2h)2 (2h)4
 
2 2
1 − cos(2h) − 2h = 1 − 1 − + + · · · − 2h2 = − h4 + · · · ,
2! 4! 3

so that we get
4
0 1 − cos(2h) − 2h2 − 2h3 + · · · 1
g (0) = lim 3
= lim 3
= lim − h = 0,
h→0 2h h→0 2h h→0 3
as the final answer.

51
2. Limits

52
Chapter 3
The Riemann integral
3
Essential reading

(For full publication details, see Chapter 1.)

+ Binmore and Davies (2002) Sections 10.2–10.4.

+ Ostaszewski (1991) Sections 17.1–17.3.

Further reading

+ Adams and Essex (2010) pp.90–95 and 102–104.

+ Wrede and Spiegel (2010) Sections 5.2–5.5.

Aims and objectives

The objectives of this chapter are:

to introduce the Riemann integral and see how it allows us to define a definite
integral in terms of the area it represents;

to see how the Riemann integral and differentiation are related via the
Fundamental Theorem of Calculus.
Specific learning outcomes can be found near the end of this chapter.

3.1 The Riemann integral


In Section 5.3 of 174 Calculus we interpreted the definite integral
Z b
f (x) dx,
a

where the integrand, f (x), is continuous and non-negative over the interval [a, b] as the
area of the region between the curve y = f (x), the x-axis and the vertical lines x = a
and x = b. Indeed, we saw how to find such integrals, and hence the corresponding
areas, by using the idea that integrals can be seen as antiderivatives. However, some
definite integrals, such as Z 1
2
e−x dx,
0

53
3. The Riemann integral

2
can not be found in this way as the integrand, which is e−x in this case, has no
antiderivative1 and so, in such cases, we would need another way of interpreting what
this integral is. One way of doing this would be to consider the area of the region
2
between the curve y = e−x , the x-axis and the vertical lines x = 0 and x = 1 directly,
and having done this, we could interpret this area as the value of this integral.
Indeed, in this chapter, we will see how to define the definite integral
3 Z b
f (x) dx,
a

in terms of the area of the region between the curve y = f (x), the x-axis and the
vertical lines x = a and x = b and we will explore some properties of this integral. In
particular, we will call this kind of integral a Riemann integral.

3.1.1 Lower and upper estimates of an area


Suppose that we have a non-negative continuous function, f (x), over some finite interval
[a, b] and we want to estimate the area, A, of the region between the curve y = f (x), the
x-axis and the vertical lines x = a and x = b as illustrated in Figure 3.1(a). One way to
do this is to take a partition, P, of the interval [a, b], i.e. a finite set of points

P = {x0 , x1 , x2 , . . . , xn−1 , xn } where a = x0 < x1 < x2 < · · · < xn = b,

and use this to divide the interval [a, b] into n sub-intervals of the form [xi−1 , xi ] for
i = 1, . . . , n as illustrated in Figure 3.1(b).

y y

y = f (x) y = f (x)

A
...

O a x O x x1 x2 ... xn−1 xn x
b 0

(a) (b)

Figure 3.1: (a) The area, A, of the region between the curve y = f (x), the x-axis and the
vertical lines x = a and x = b. (b) The partition, P, of the interval [a, b] where x0 = a
and xn = b. Note that, for clarity, only the first three and last two points of the partition
are shown.
1
By this we mean that there is no combination of our basic functions which can be differentiated to
2
give the function e−x .

54
3.1. The Riemann integral

Each of these sub-intervals can then be taken to be the base of a rectangle whose height
in some way depends on the values of f (x) in that sub-interval. In particular, we can
take the height of the rectangle in each sub-interval to be the

minimum value of f (x) in that sub-interval so that the sum of the areas of these
rectangles gives us a lower estimate of the area, or the

maximum value of f (x) in that sub-interval so that the sum of the areas of these 3
rectangles gives us an upper estimate of the area.
Clearly, these rectangles will allow us to estimate the area of the region involved.
Specifically, we can see that the

lower estimate, L(P), will be the sum of the areas of the rectangles illustrated in
Figure 3.2(a), i.e.
Xn
L(P) = Li ,
i=1

and this will clearly be less than the area, A, that we are after as the rectangles
give us ‘too little’ area if we compare their area with the area in Figure 3.1(a).

upper estimate, U (P), will be the sum of the areas of the rectangles illustrated in
Figure 3.2(b), i.e.
Xn
U (P) = Ui ,
i=1

and this will clearly be greater than the area, A, that we are after as the rectangles
give us ‘too much’ area if we compare their area with the area in Figure 3.1(a).

y y

y = f (x) y = f (x)

L1 L2 ... Ln U1 U2 ... Un

O x x1 x2 ... xn−1 xn x O x x1 x2 ... xn−1 xn x


0 0

(a) (b)

Figure 3.2: In (a) and (b) we have, respectively, the rectangles that contribute towards
the lower estimate, L(P), and the upper estimate, U (P), of A based on the partition P.
Note that, for clarity, only the first, second and last rectangles are shown.

55
3. The Riemann integral

That is, we can see that using these estimates we have


L(P) < A < U (P),
and so these estimates provide us with bounds on the true value of A. Indeed, since it is
easy to work out the area of a rectangle using ‘base times height’, we can see that the
base of each of the rectangles is given by the length of the relevant sub-interval, i.e. for
i = 1, 2, . . . , n we have a
3
sub-interval [xi−1 , xi ] with a length of xi − xi−1 ,
which means that, thinking about the heights as well, we have a

lower estimate given by


n
X
L(P) = (xi − xi−1 )mi ,
i=1

where mi is the minimum value of f (x) for values of x in the sub-interval [xi−1 , xi ],
i.e.
mi = min{f (x)|xi−1 ≤ x ≤ xi },
is the height of this rectangle.
upper estimate given by
n
X
U (P) = (xi − xi−1 )Mi ,
i=1

where Mi is the maximum value of f (x) for values of x in the sub-interval [xi−1 , xi ],
i.e.
Mi = max{f (x)|xi−1 ≤ x ≤ xi },
is the height of this rectangle.
Let’s have a look at a simple example to see how these estimates can be found in
practice.

Example 3.1 Suppose that we have the function f (x) = 1 + x2 defined over the
interval [−2, 2] and the partition P = {−2, 0, 1, 2} of this interval. Find the lower
and upper estimates, L(P) and U (P) respectively, of the area of the region bounded
by the curve y = f (x), the x-axis and the vertical lines x = −2 and x = 2.

If we sketch the curve y = 1 + x2 over the interval [−2, 2] and indicate the points in
the partition P = {−2, 0, 1, 2} we see that we will be looking at three rectangles
whose bases are given by the sub-intervals [−2, 0], [0, 1] and [1, 2]. Indeed, for the

lower estimate, we need to sum the areas of the three rectangles illustrated in
Figure 3.3(a) where the height of each rectangle is given by the minimum value
of f (x) in each of the sub-intervals. This gives us,

L(P) = (2)(1) + (1)(1) + (1)(2) = 5,

where we have taken each of the sub-intervals in turn and used ‘base times
height’ to find the area of each rectangle.

56
3.1. The Riemann integral

upper estimate, we need to sum the areas of the three rectangles illustrated in
Figure 3.3(b) where the height of each rectangle is given by the maximum value
of f (x) in each of the sub-intervals. This gives us,

U (P) = (2)(5) + (1)(2) + (1)(5) = 17,

where we have taken each of the sub-intervals in turn and used ‘base times
height’ to find the area of each rectangle. 3
In particular, observe that if A is the area of the region bounded by the curve
y = 1 + x2 , the x-axis and the vertical lines x = −2 and x = 2, then we can see from
the illustrations in Figure 3.3 that we have

L(P) < A < U (P),

as we should expect from the lower and upper estimates respectively.

y y

y = 1 + x2 5 y = 1 + x2

2 2

1 1

O x O x
−2 1 2 −2 1 2
(a) (b)

Figure 3.3: In (a) and (b) we have, respectively, the rectangles that contribute towards
the lower estimate, L(P), and the upper estimate, U (P), of A based on the partition P.

Activity 3.1 Use integration to find the area, A, of the region bounded by the
curve y = 1 + x2 , the x-axis and the vertical lines x = −2 and x = 2. Hence verify
that the lower and upper estimates found in Example 3.1 do indeed satisfy the
inequality L(P) < A < U (P).

Of course, although we can find lower and upper estimates using simple partitions like
the one in Example 3.1, it will be necessary for us to use more sophisticated partitions
if we want to make any real progress towards more accurate estimates and, ultimately, a
definition of the Riemann integral.

57
3. The Riemann integral

Example 3.2 Suppose that we have the function f (x) = 1 + x defined over the
interval [0, 1] and the partitions
Pn = 0, n1 , n2 , . . . , n−1

n
,1 ,
of this interval for some n ∈ N. Find the lower and upper estimates, L(Pn ) and
U (Pn ) respectively, of the area of the region bounded by the curve y = f (x), the
3 x-axis and the vertical lines x = 0 and x = 1.

If we sketch the curve y = 1 + x over the interval [0, 1] and indicate the points in the
partition Pn , we see
 that  we will be looking at n rectangles whose bases are given by
the sub-intervals k−1n
, k
n
for k = 1, 2, . . . , n. Indeed, we see that, for k = 1, 2, . . . , n,
each of these sub-intervals gives us a base whose length is
k k−1 1
− = ,
n n n
and so, for the

lower estimate, we need to sum the areas of the n rectangles illustrated in


Figure 3.4(a) where the height of each rectangle is given by the minimum value
of f (x) in each of the sub-intervals. This gives us,
n−1
       
1 0 1 1 1 2 1
L(Pn ) = 1+ + 1+ + 1+ + ··· + 1+ ,
n n n n n n n n
where we have taken each of the sub-intervals in turn and used ‘base times
height’ to find the area of each rectangle. Then, summing this series we find that
 
1 1
L(Pn ) = 3− ,
2 n
as you can verify in Activity 3.2.
upper estimate, we need to sum the areas of the n rectangles illustrated in
Figure 3.4(b) where the height of each rectangle is given by the maximum value
of f (x) in each of the sub-intervals. This gives us,
       
1 1 1 2 1 3 1 n
U (Pn ) = 1+ + 1+ + 1+ + ··· + 1+ ,
n n n n n n n n
where we have taken each of the sub-intervals in turn and used ‘base times
height’ to find the area of each rectangle. Then, summing this series we find that
 
1 1
U (Pn ) = 3+ ,
2 n
as you can verify in Activity 3.2.
In particular, observe that if A is the area of the region bounded by the curve
y = f (x), the x-axis and the vertical lines x = 0 and x = 1, then we can see from the
illustrations in Figure 3.4 that we have
L(Pn ) < A < U (Pn ),
as we should expect from the lower and upper estimates respectively.

58
3.1. The Riemann integral

y y
y =1+x y =1+x
2 2

3
1 1

... ...

O x O x
1 2 3 ... n−1 1 1 2 3 ... n−1 1
n n n n n n n n

(a) (b)

Figure 3.4: In (a) and (b) we have, respectively, the rectangles that contribute towards
the lower estimates, L(Pn ), and the upper estimates, U (Pn ), of A based on the partition
Pn . Note that, for clarity, some of the y-intercepts have been omitted and only four of
the rectangles are shown.

Activity 3.2 Given that the sum of the first k natural numbers is given by the
formula
k
1 + 2 + ··· + k =
(k + 1),
2
show that the lower and upper estimates in Example 3.2 can be written as
   
1 1 1 1
L(Pn ) = 3− and U (Pn ) = 3+ ,
2 n 2 n

when the series are summed.

Activity 3.3 Find the area, A, of the region bounded by the curve y = 1 + x, the
x-axis and the vertical lines x = 0 and x = 1. Hence verify that the lower and upper
estimates found in Example 3.2 do indeed satisfy the inequality L(Pn ) < A < U (Pn ).

Another thing to notice from Example 3.2 is that, as the function is increasing, there is
a straightforward relationship between the height of each rectangle and the value of the
function at the end-points of each sub-interval. In particular, we see that when we have
an increasing function over an interval, we find the

lower estimate takes the height of each rectangle to be the value of the function
at the left-hand end-point of the relevant sub-interval.
upper estimate takes the height of each rectangle to be the value of the function
at the right-hand end-point of the relevant sub-interval.
This idea is explored in Activity 3.4.

59
3. The Riemann integral

Activity 3.4 Suppose that f (x) is a non-negative continuous function defined over
some interval [a, b] and that P is the partition

{x0 , x1 , x2 , . . . , xn−1 , xn } where a = x0 < x1 < x2 < · · · < xn = b.

If f (x) is an increasing function on the interval [a, b], show that the lower and upper
3 estimates, L(P) and U (P) respectively, of the area of the region bounded by the
curve y = f (x), the x-axis and the vertical lines x = a and x = b are given by
n
X n
X
L(P) = (xi − xi−1 )f (xi−1 ) and U (P) = (xi − xi−1 )f (xi ).
i=1 i=1

What are L(P) and U (P) if f (x) is a decreasing function on the interval [a, b]?

However, even though we can come up with such useful results about the lower and
upper estimates of an area for increasing or decreasing functions over an interval,
generally speaking, the best way to proceed is to sketch the relevant rectangles and then
use these to infer their areas as we did in Examples 3.1 and 3.2.

3.1.2 Getting better lower and upper estimates


Having seen how to find lower and upper estimates of an area, we now want to see how
we can get better lower and upper estimates with a view to finding the best lower and
upper estimates which should, naturally, be the value of the area itself. In particular, as
the next example makes clear, we should be able to get better lower and upper
estimates by adding points to our partition so that we get ‘thinner’ rectangles and hence
better estimates of the area involved.

Example 3.3 Use the partition P 0 = {−2, −1, 0, 1, 2} to find the lower and upper
estimates, L(P 0 ) and U (P 0 ) respectively, of the area discussed in Example 3.1.
Compare these values to what we found in that example.

If we sketch the curve y = 1 + x2 over the interval [−2, 2] and indicate the points in
the partition P 0 = {−2, −1, 0, 1, 2} we see that we will now be looking at four
rectangles whose bases are given by the sub-intervals [−2, −1], [−1, 0], [0, 1] and
[1, 2]. Indeed, for the

lower estimate, we need to sum the areas of the four rectangles illustrated in
Figure 3.5(a) where the height of each rectangle is given by the minimum value
of f (x) in each of the sub-intervals. This gives us,

L(P 0 ) = (1)(2) + (1)(1) + (1)(1) + (1)(2) = 6,

where we have taken each of the sub-intervals in turn and used ‘base times
height’ to find the area of each rectangle.

upper estimate, we need to sum the areas of the four rectangles illustrated in
Figure 3.5(b) where the height of each rectangle is given by the maximum value

60
3.1. The Riemann integral

of f (x) in each of the sub-intervals. This gives us,

U (P 0 ) = (1)(5) + (1)(2) + (1)(2) + (1)(5) = 14,

where we have taken each of the sub-intervals in turn and used ‘base times
height’ to find the area of each rectangle.
Comparing this with our answers from Example 3.1, we can see that the 3
lower estimates give us

L(P) = 5 < 6 = L(P 0 ) =⇒ L(P) < L(P 0 ),

i.e. L(P 0 ) is a larger, and hence better, lower estimate of the area than L(P).

upper estimates give us

U (P 0 ) = 14 < 17 = U (P) =⇒ U (P 0 ) < U (P),

i.e. U (P 0 ) is a smaller, and hence better, upper estimate of the area than U (P).
Indeed, it should be clear that this improvement in the lower and upper estimates
has come about because P 0 has one more point than P and this has enabled us to
replace the single rectangle for the sub-interval [−2, 0] from P (see Figure 3.3) with
two ‘thinner’ rectangles for the sub-intervals [−2, −1] and [−1, 0] from P 0 (see
Figure 3.5).
In particular, observe that if A is the area of the region bounded by the curve
y = 1 + x2 , the x-axis and the vertical lines x = −2 and x = 2, then we can see from
the results above and the illustrations in Figure 3.5 that we have

L(P) < L(P 0 ) < A < U (P 0 ) < U (P),

i.e. P 0 gives us better lower and upper estimates of A than P.

Activity 3.5 Use your answer to Activity 3.1 to verify that the lower and upper
estimates found in Example 3.2 satisfy the inequality

L(P) < L(P 0 ) < A < U (P 0 ) < U (P),

as one would expect.

In fact, this is what we should expect more generally, i.e.

lower (and upper) estimates should always be less (or greater) than the area they
are estimates of, and

increasing the number of the points in the partition should always give us better
lower (or upper) estimates.
To make this clear, let’s consider another example.

61
3. The Riemann integral

y y

y = 1 + x2 5 y = 1 + x2

3
2 2

1 1

O x O x
−2 −1 1 2 −2 −1 1 2
(a) (b)

Figure 3.5: In (a) and (b) we have, respectively, the rectangles that contribute towards
the lower estimate, L(P 0 ), and the upper estimate, U (P 0 ), of A based on the partition P.

Example 3.4 Use the results from Example 3.2 to verify the two points above.

We start by noting that in this case, as we saw in Activity 3.3, we can meaningfully
talk about the value of the area, A, of the region bounded by the curve y = 1 + x,
the x-axis and the vertical lines x = 0 and x = 1 because it is simply the area of a
trapezium. Indeed, for the first point, we have already shown in Activity 3.3 that
L(Pn ) < A < U (Pn ),
i.e. if we are estimating the area A, then whatever value of n ∈ N we take, the lower
estimates, L(Pn ), are always less than A whereas the upper estimates, U (Pn ) are
always greater than A.
For the second point, consider that the partition, Pn has n + 1 points and, if we take
some other natural number m > n, the partition Pm will have m + 1 points, i.e. Pm
has m − n more points than Pn . We then have

lower estimates given by


   
1 1 1 1
L(Pn ) = 3− and L(Pm ) = 3− .
2 n 2 m
But, m > n > 0 means that
   
1 1 1 1 1 1 1 1
> =⇒ 3− <3− =⇒ 3− < 3− ,
n m n m 2 n 2 m
which means that we have L(Pn ) < L(Pm ). Indeed, using the first point, both of
these lower estimates must be less than A, and so we have
L(Pn ) < L(Pm ) < A,
i.e. increasing the number of points in the partition has given us a better lower
estimate.

62
3.1. The Riemann integral

upper estimates given by


   
1 1 1 1
U (Pn ) = 3+ and U (Pm ) = 3+ .
2 n 2 m

But, m > n > 0 means that

1
>
1
=⇒
1
3+ >3+
1
=⇒
1

3+
1

>
1

3+
1

,
3
n m n m 2 n 2 m

which means that we have U (Pn ) > U (Pm ). Indeed, using the first point, both
of these upper estimates must be greater than A, and so we have

U (Pn ) > U (Pm ) > A,

i.e. increasing the number of points in the partition has given us a better upper
estimate.

Notice, however, that although we have been using various graphical and geometric
‘intuitions’ about the area under a curve in our examples, we have not actually defined
what the area under a curve is. In particular, this is why all reference to using definite
integrals to find areas have been relegated to Activities 3.1 and 3.3 where they can serve
to illustrate the points being made without interfering with the flow of our argument.

3.1.3 The definition of the Riemann integral


We now turn to the definition of the Riemann integral which will allow us to assign a
value to the definite integral
Z b
f (x) dx,
a

where f (x) is a non-negative continuous function over the interval [a, b] by taking it to
be the value of the area of the region between the curve y = f (x), the x-axis and the
vertical lines x = a and x = b. So, we have to ask ourselves, what is the value of this
area?
It should be clear, from what we have seen above, that in the case we are considering:

Every partition of [a, b] gives us a lower and upper estimate of the area.

Every lower estimate is less than every upper estimate, i.e. we have L(P) ≤ U (P).

This means that there is at least one number A such that L(P) ≤ A ≤ U (P).
Now, if there is only one such number, then A is the value of the area and hence the
value of the Riemann integral
Z b
f (x) dx,
a

whereas if there is more than one such number, then we say that the area is undefined
and so this Riemann integral is undefined too.

63
3. The Riemann integral

In practice, we can use this to find the value of the Riemann integral,
Z b
I= f (x) dx,
a

where f (x) is a non-negative continuous function as follows. We find the lower and
upper estimates in terms of some appropriately general partition, P, and then find the
3
smallest number, L∗ , that is greater than all the lower estimates, L(P), and the

largest number, U ∗ , that is smaller than all the upper estimates, U (P).
Then, according to what we have seen above, we should find that

L∗ ≤ I ≤ U ∗ ,

so that, if there is only one number, I, that satisfies this inequality we can take this to
be the value of the Riemann integral.2

Example 3.5 Continuing on from Examples 3.2 and 3.4, find the value of
Z 1
(1 + x) dx,
0

using the definition of the Riemann integral.

We saw in Example 3.2 that we have

lower estimates given by


 
1 1 3
L(Pn ) = 3− ≤ .
2 n 2

for all n ∈ N. That is, following on from Example 3.4, we have found that, for
m, n ∈ N with m > n,
L(Pn ) ≤ L(Pm ) ≤ L∗ ,
where L∗ = 3/2 is the smallest number that is greater than all the lower
bounds.3

upper estimates given by


 
1 1 3
U (Pn ) = 3+ ≥ ,
2 n 2
2
Technically, we are looking at every possible partition, P, of the interval [a, b] and, having found the
corresponding lower and upper estimates, we take L∗ to be the least upper bound (or supremum) of the
lower estimates and U ∗ to be the greatest lower bound (or infimum) of the upper estimates, i.e.

L∗ = sup{L(P)|P is a partition of [a, b]} and U ∗ = inf{U (P)|P is a partition of [a, b]}.

This is because, generally, as we have taken a partition to be a finite set of points, we shall find that
there is no partition P that makes L∗ = L(P) or U ∗ = U (P), even though L∗ and U ∗ act as the relevant
bounds on L(P) and U (P) respectively.

64
3.1. The Riemann integral

for all n ∈ N. That is, following on from Example 3.4, we have found that, for
m, n ∈ N with m > n,
U ∗ ≤ U (Pm ) ≤ U (Pn ),
where U ∗ = 3/2 is the largest number that is smaller than all the upper bounds.
This means that there is only one number, I, that satisfies the inequality

L∗ ≤ I ≤ U ∗ , 3
in this case, i.e. I = 3/2, and so we take this to be the value of the given Riemann
integral. That is, Z 1
3
(1 + x) dx = ,
0 2
as we should expect from Activity 3.3.

Activity 3.6 Use the definition of the Riemann integral to find the value of
Z 2
(1 + x2 ) dx,
0

by considering, for m ∈ N, the partition

Pm = {0, m1 , m2 , . . . , 2m−1
m
, 2},

of the interval [0, 2].

Note: You will need to use the formula


k
12 + 22 + · · · + k 2 = (k + 1)(2k + 1),
6
for the sum of the first k square numbers.

Of course, more generally, we should expect the Riemann integral


Z b
f (x) dx,
a

to be well-defined if f (x) is a non-negative continuous function over the interval [a, b].
This is because it is just the area between the curve y = f (x), the x-axis and the
vertical lines x = a and x = b, which is clearly well-defined if f (x) is continuous. Indeed,
although we do not dwell on this here, we can drop the requirement that f (x) is
non-negative and extend what we have seen to functions where f (x) is negative for
some values of x in the interval [a, b] by utilising the methods discussed in Section 5.3 of
174 Calculus. All of this leads to the following fact.
3
That is, we have
L∗ = sup{L(Pn )|n ∈ N},
as, for all n ∈ N, L(Pn ) ≤ L∗ and so L∗ is an upper bound on the lower estimates and L∗ is the least
upper bound as there is no other upper bound, l, such that l < L∗ . Also observe that, for any n ∈ N,
L(Pn ) < L∗ and so there is no partition with a finite number of points that makes L(Pn ) = L∗ .

65
3. The Riemann integral

If f (x) is a continuous function over the interval [a, b], then the Riemann
integral Z b
f (x) dx,
a
exists.
Although, we shall not prove this here.
3
3.1.4 What happens if the integrand isn’t continuous?
We now begin to consider the Riemann integral
Z b
f (x) dx,
a

when f (x) is not continuous over the interval [a, b]. In particular, we consider how the
definition needs to be slightly modified when the integrand has finite discontinuities at
certain points and encounter a case where the Riemann integral is undefined.
In particular, one problem we will encounter is that a function which has finite
discontinuities may not have a minimum or maximum value over every interval. To
motivate the discussion that follows, let’s look at an example.

Example 3.6 Consider the integral


(
1
2x if 0 ≤ x < 1
Z
f (x) dx where f (x) =
0 1 if x = 1

and, for n ∈ N, the partition

Pn = {0, n1 , n2 , . . . , 1},

of the interval [0, 1]. Sketch f (x) over the interval [0, 1] and explain why the upper
estimates (as we defined them in Section 3.1.1) are problematic in this case. Does
this seem reasonable?

If we sketch f (x) over the interval [0, 1], as illustrated in Figure 3.6(a), we see that
the function has a finite discontinuity at x = 1 because f (x) → 2 as x → 1− but
f (1) = 1 6= 2. In particular, even though f (x) → 2 as x → 1− , the function never
attains the value two and so, over the interval [0, 1], this function has no maximum
value.4

This poses a problem for the upper estimates since, if we consider the sub-interval
[ n−1
n
, 1], as illustrated in Figure 3.6(b), it means that we can not ascribe a value to

Mn = max{f (x)| n−1


n
≤ x ≤ 1},

and so, bearing in mind that the base of each rectangle that arises from the partition
is 1/n, the upper estimates
n
X 1
U (Pn ) = Mi ,
i=1
n

66
3.1. The Riemann integral

for n ∈ N are not defined since we cannot ascribe a value to its last term.

However, given that we want to interpret the given Riemann integral as the area of
the region under the curve y = f (x), the x-axis and the vertical lines x = 0 and
x = 1, we would surely expect this area to exist. Indeed, we would expect this to be
the area of a triangular region which, using ‘half base times height’, is one! So, in
this case, it seems unreasonable that the Riemann integral does not exist because we
can not define its upper estimates.
3

y y

y = f (x) y = f (x)
2 2

1 1

O x O n−1
x
1 n
1
(a) (b)

Figure 3.6: In (a) we have a sketch of the function y = f (x) from Example 3.6 and in (b)
we examine the effect of the problematic sub-interval [ n−1
n
, 1].

To deal with cases like this, we should use a slightly more general definition of the
Riemann integral Z b
f (x) dx,
a
in terms of a partition, P, of the interval [a, b] given by
P = {x0 , x1 , x2 , . . . , xn−1 , xn } where a = x0 < x1 < x2 < · · · < xn = b,
which runs as before except that, now, we have

lower estimates, L(P), given by


n
X
L(P) = (xi − xi−1 )mi ,
i=1

where mi is the greatest lower bound, or infimum, of f (x) for values of x in the
sub-interval [xi−1 , xi ], i.e.
mi = inf{f (x)|xi−1 ≤ x ≤ xi },
4
Of course, the function g(x) = 2x defined over the interval [0, 1] gives us g(1) = 2 and this is its
maximum value, but the function f (x) that we are considering here has f (1) = 1!

67
3. The Riemann integral

with the understanding that, in cases where this value is actually attained by the
function in the sub-interval, this will just be the minimum value of the function as
before.
upper estimates, U (P), given by
n
X
U (P) = (xi − xi−1 )Mi ,
3 i=1

where Mi is the supremum of f (x) for values of x in the sub-interval [xi−1 , xi ], i.e.
Mi = sup{f (x)|xi−1 ≤ x ≤ xi },
with the understanding that, in cases where this value is actually attained by the
function in the sub-interval, this will just be the maximum value of the function as
before.
Of course, with this in mind, we can now make sense of what we saw in Example 3.6 as
the next example shows.

Example 3.7 Following on from Example 3.6, use this more general definition of
the Riemann integral to find the value of
Z 1 (
2x if 0 ≤ x < 1
f (x) dx where f (x) =
0 1 if x = 1
by considering, for n ∈ N with n ≥ 2, the partitions
Pn = {0, n1 , n2 , . . . , 1},
of the interval [0, 1].

If we sketch the curve y = f (x) over the interval [0, 1] and indicate the points in the
partition Pn we see
 that we will be looking at n rectangles whose bases are given by
the sub-intervals k−1
n
, k
n
for k = 1, 2, . . . , n. Indeed, we see that, for k = 1, 2, . . . , n,
each of these sub-intervals gives us a base whose length is
k k−1 1
− = ,
n n n
and so, for the

lower estimates, we need to sum the areas of the n rectangles illustrated in


Figure 3.7(a) where the height of each rectangle is given by the minimum value
of f (x) in each of the sub-intervals. This gives us,
n−2
         
1 0 1 1 1 2 1 1
L(Pn ) = 2· + 2· + 2· + ··· + 2· + 1 ,
n n n n n n n n n
where we have taken each of the sub-intervals in turn and used ‘base times
height’ to find the area of each rectangle. Then, summing this series we find that
n−1
L(Pn ) = 1 − 2 · ≤ 1,
n2
for n ≥ 2, as you can verify in Activity 3.7. That is, L∗ = 1 is the smallest
number that is greater than all the lower bounds.

68
3.1. The Riemann integral

upper estimates, we need to sum the areas of the n rectangles illustrated in


Figure 3.7(b) where the height of each rectangle is given by the maximum value
of f (x) in the first n − 1 sub-intervals and the supremum of f (x) in the last
sub-interval. This gives us,

n−1
         
1 1 1 2 1 3 1 1 n
U (Pn ) = 2· + 2· + 2· +···+ 2· + 2· ,
n n n n n n n n n n
3
where we have taken each of the sub-intervals in turn and used ‘base times
height’ to find the area of each rectangle. Then, summing this series we find that
1
U (Pn ) = 1 + ≥ 1,
n
for n ≥ 2, as you can verify in Activity 3.7. That is, U ∗ = 1 is the largest
number that is less than all the upper bounds.
This means that there is only one number, I, that satisfies the inequality

L∗ ≤ I ≤ U ∗ ,

in this case, i.e. I = 1, and so we take this to be the value of the given Riemann
integral. That is,
Z 1 (
2x if 0 ≤ x < 1,
f (x) dx = 1 when f (x) =
0 1 if x = 1,

as we should expect from our discussion in Example 3.6.

Activity 3.7 Using the formula for the sum of the first k natural numbers given in
Activity 3.2, show that the lower and upper estimates in Example 3.7 can be written
as
n−1 1
L(Pn ) = 1 − 2 · 2
and U (Pn ) = 1 + ,
n n
when the series are summed and verify that, for n ≥ 2, we have L(Pn ) ≤ 1 and
U (Pn ) ≥ 1.

However, if there are too many finite discontinuities, the Riemann integral will not exist
as the next example shows.

Example 3.8 Consider the integral


(
1
1 if x is rational,
Z
f (x) dx where f (x) =
0 0 if x is irrational,

and the partition

P = {x0 , x1 , x2 , . . . , xn−1 , xn } where 0 = x0 < x1 < x2 < · · · < xn−1 < xn = 1,

69
3. The Riemann integral

y y
y = f (x) y = f (x)
2 2

3
1 1

... ...

O x O x
1 2 3 ... n−2 n−1 1 1 2 3 ... n−2 n−1 1
n n n n n n n n n n

(a) (b)

Figure 3.7: In (a) and (b) we have, respectively, the rectangles that contribute towards
the lower estimates, L(Pn ), and the upper estimates, U (Pn ), of the Riemann integral in
Example 3.7 based on the partition Pn for n ≥ 2. Note that, for clarity, some of the
y-intercepts have been omitted and only five of the rectangles are shown. (Of course, in
(a), the first rectangle has a height of zero and so we can not see it!)

of the interval [0, 1]. Find the lower and upper estimates, L(P) and U (P)
respectively, of this integral. What do these tell you about the existence of this
Riemann integral?

Note: A real number is said to be rational if it is of the form p/q where p, q ∈ Z and
q 6= 0 and, if it is not rational, a number is said to be irrational.5 Indeed, we will
need the facts that

(1) every interval [a, b] with a < b contains at least one rational number, and
(2) every interval [a, b] with a < b contains at least one irrational number,
to answer this question.

In this case, we can not sketch the function as it is discontinuous at every point, but
we can easily find the

lower estimate by noting that in every sub-interval [xi−1 , xi ] of length


xi − xi−1 the minimum value of f (x) is zero as, by fact (1), there is at least one
rational number in that sub-interval. This gives us
L(P) = (x1 − x0 )(0) + (x2 − x1 )(0) + · · · + (xn−1 − xn−2 )(0) + (xn − xn−1 )(0) = 0,
i.e. the lower estimate is zero for any partition, P.
upper estimate by noting that in every sub-interval [xi−1 , xi ] of length
xi − xi−1 the maximum value of f (x) is one as, by fact (2), there is at least one
irrational number in that sub-interval. This gives us
U (P) = (x1 −x0 )(1)+(x2 −x1 )(1)+· · ·+(xn−1 −xn−2 )(1)+(xn −xn−1 )(1) = xn −x0 ,

70
3.1. The Riemann integral

if we cancel the corresponding intermediate terms and, noting that x0 = 0 and


xn = 1, we then see that the upper estimate is one for any partition, P.
Consequently, we see that as L(P) = 0 and U (P) = 1 for any partition P, there are
many numbers I that satisfy the inequality

L(P) ≤ I ≤ U (P),
3
and so, as there is not only one such number, we can not assign a single value to this
Riemann integral, i.e. it does not exist.6

In the next chapter, we will see how to deal with Riemann integrals in cases where the
interval we are concerned with is not finite and where the integrand fails to be
continuous in other, less straightforward, ways (such as the presence of an infinite
discontinuity).

3.1.5 Some properties of the Riemann integral


So far, we have been concerned with Riemann integrals of the form
Z b
f (x) dx,
a

where a < b and we now want to extend our understanding of such integrals by seeing
what happens when a = b or a > b. To do this, we introduce some basic properties of
the Riemann integral in Theorem 3.1 and, in particular, these will be useful in the next
section.
Theorem 3.1 If the Riemann integral of f exists over some interval containing the
points a, b and c with a ≤ c ≤ b, then

(a) The integral over an interval of zero length is zero, i.e.


Z a
f (x) dx = 0.
a

(b) The integral is additive over the interval of integration, i.e.


Z b Z c Z b
f (x) dx = f (x) dx + f (x) dx.
a a c

(c) Interchanging the limits of integration changes the sign of the integral, i.e.
Z b Z a
f (x) dx = − f (x) dx.
a b
5

For example, you may have encountered the fact that the real number, 2, is irrational because it
can not be written in the form p/q where p, q ∈ Z and q 6= 0.
6
Actually, there is a well-defined sense in which ‘almost all’ real numbers are irrational and so the
value of this integral ‘ought’ to be zero as its value shouldn’t be affected by the value of the integrand
at a ‘few’ points where its argument is rational. Indeed, although it is beyond the scope of this course,
you might care to note that there is a more general notion of integration, called the Lebesgue integral,
whose value would be zero in this case.

71
3. The Riemann integral

(d) It follows that


Z b Z b Z a
f (x) dx = f (x) dx − f (x) dx.
a c c

We will not prove this theorem here but, in Activity 3.8, you will be able to motivate
these results by appealing to our interpretation of the Riemann integral as an area.
3
Activity 3.8 Suppose that f is a non-negative function whose Riemann integral
exists over some interval containing the points a, b and c. By considering the areas
involved, explain why parts (a) and (b) of Theorem 3.1 hold. Hence deduce parts (c)
and (d) of this theorem.

3.2 The Fundamental Theorem of Calculus


Returning to the case where the function, f (x), is continuous and non-negative over an
interval [a, b], we have defined the Riemann integral,
Z b
f (x) dx,
a

in terms of the area between the curve y = f (x), the x-axis and the vertical lines x = a
and x = b. What we want to do now is make sure that this view of integration agrees
with what we saw in Chapter 5 of 174 Calculus. In particular, we want to be sure that
the relationship between integrals and derivatives that we used there still works if
integration is defined in this way. Indeed, the key result here is the Fundamental
Theorem of Calculus (which we will abbreviate to FTC) and this runs as follows.
Theorem 3.2 (FTC) If f is a continuous function defined on the interval [a, b], then
Z t 
d
f (x) dx = f (t),
dt c

for any numbers c and t in the interval (a, b).

We will spend the rest of this chapter seeing why this works and what it tells us about
the relationship between integrals and derivatives.

3.2.1 Motivating the FTC


To see how the FTC arises, consider the case where f (x) is a decreasing non-negative
continuous function over the interval [a, b]. In particular, we will take c, t ∈ (a, b) where
x > c and consider the integral
Z t
A(t) = f (x) dx,
c

so that A(t) represents the area between the curve y = f (x), the x-axis and the vertical
lines x = c and x = t as illustrated in Figure 3.8(a). Indeed, this means that if h > 0, we

72
3.2. The Fundamental Theorem of Calculus

also have the integral Z t+h


A(t + h) = f (x) dx,
c

so that A(t) represents the area between the curve y = f (x), the x-axis and the vertical
lines x = c and x = t + h as illustrated in Figure 3.8(b).

y y 3

y = f (x) y = f (x)
A(t) A(t + h)

O x O x
a c t b a c t t+h b
(a) (b)

Figure 3.8: The shaded regions in (a) and (b) have areas A(t) and A(t + h) respectively.

We now consider the difference between these two areas, i.e. the quantity

A(t + h) − A(t),

which is the area of the vertically hatched region indicated in Figure 3.9.

y y

1111
010000 1111
0000
101111 0000
1111
y = f (x) y = f (x)
0000
101111
0000 0000
1111
101111
0000
101111 0000
1111
0000
1111
O
a c t t+h b
0000
10 x O
a c 0000
1111
t t+h b
x

(a) (b)
Figure 3.9: The areas of the shaded regions in (a) and (b) are the lower and upper
estimates, respectively, of the area A(t + h) − A(t) of the vertically hatched region.

73
3. The Riemann integral

In particular, we see that since f (x) is a decreasing function over the interval [a, b], we
have a

lower estimate of this area given by the area of the shaded rectangle in
Figure 3.9(a) which, using ‘base times height’, is

hf (t + h),
3
and so we have hf (t + h) ≤ A(t + h) − A(t).

upper estimate of this area given by the area of the shaded rectangle in
Figure 3.9(b) which, using ‘base times height’, is

hf (t),

and so we have A(t + h) − A(t) ≤ hf (t).


Consequently, putting these two inequalities together, we have
A(t + h) − A(t)
hf (t + h) ≤ A(t + h) − A(t) ≤ hf (t) =⇒ f (t + h) ≤ ≤ f (t),
h
as h > 0. Then, we take the limit as h → 0+ on both sides of this inequality to get
A(t + h) − A(t) A(t + h) − A(t)
lim+ f (t+h) ≤ lim+ ≤ lim+ f (t) =⇒ f (t) ≤ lim+ ≤ f (t),
h→0 h→0 h h→0 h→0 h
if we use the continuity of f (x) over [a, b].7 This means that
A(t + h) − A(t)
lim+ = f (t),
h→0 h
and, repeating the argument with h < 0, we also get
A(t + h) − A(t)
lim− = f (t),
h→0 h
as you can verify in Activity 3.9. Then, putting these two results together, we have
A(t + h) − A(t)
lim = f (t) =⇒ A0 (t) = f (t),
h→0 h
if we use the definition of the derivative, A0 (t), of A(t). Thus, we have shown that when
f (x) is a continuous decreasing function over [a, b] with c, t ∈ (a, b) and c < t, we have
Z t 
0 d
A (t) = f (x) dx = f (t),
dt c

as expected from the FTC.

7
Notice that the continuity of f (x) over [a, b] is essential here in order to justify the assertion that
the limit
lim f (t + h),
h→0+

exists and that it is equal to f (t). Also, notice that we can only take the limit as h → 0+ here as we
have assumed that h > 0.

74
3.2. The Fundamental Theorem of Calculus

Activity 3.9 Show that

A(t + h) − A(t)
lim− = f (t),
h→0 h
using an argument similar to the one above with h < 0.
3
3.2.2 Notation: Dummy variables
Before we proceed, we should make sure that we understand the role that dummy
variables are playing here. Generally, given a function, f , defined over some interval
[a, b] it makes no difference what we call the independent variable. That is, we could use
x and look at values of
f (x) where x ∈ [a, b],
or we could use y and look at values of
f (y) where y ∈ [a, b],
or we could use any other letter to represent the independent variable.8 Indeed, when
we look at an integral like Z b
f (x) dx,
a
where a and b are constants, we know that the answer will not depend on x and, in
particular, it makes no difference if we write this integral as
Z b
f (y) dy.
a

In such cases, the variable that is used inside the integral is a dummy variable as it does
not matter what we call it since it does not occur in the final answer. However, when we
have an integral like Z t
f (x) dx,
c
where only c is a constant, we now have two variables. One of these variables, x, is a
dummy variable as it will not figure in the final answer but the other variable, t, is not
as the answer will be a function of t, i.e. we may write
Z t
F (t) = f (x) dx.
c

Of course, this means that


Z t Z t
f (x) dx and f (y) dy,
c c

say, represent the same integral whereas


Z t Z s
F (t) = f (x) dx and F (s) = f (x) dx,
c c
8
Although, by convention, we usually use letters from the end of the English alphabet to denote
variables.

75
3. The Riemann integral

say, represent the same function, namely F . In particular, this means that some care
must be taken when choosing names for our variables as we would not want to write
Z x
F (x) = f (x) dx,
c

because this uses x as both the independent variable of F and the dummy variable of
3 the integral. To see why this is a problem, ask yourself what is, say, F (1) using this
definition of F ? Is it Z 1 Z 1
f (x) dx or f (1) d1 ?
c c
Obviously, we want to be able to distinguish between these two options and that is why
we use dummy variables!

3.2.3 The relationship between integration and differentiation


Having motivated the FTC, we are now in a position to clarify the relationship between
integration and differentiation. That is, we want to find that differentiation ‘undoes’
integration and that, up to an arbitrary constant, integration ‘undoes’ differentiation.

Differentiation ‘undoes’ integration

Recall that, in Section 5.1 of 174 Calculus, we defined an antiderivative of the function,
f (t), to be a function F (t) such that
dF
= f (t).
dt
Indeed, using the FTC, we can now see that the integral
Z t
F (t) = f (x) dx,
c

where c is a constant is an antiderivative of f (t) since


Z t 
0 dF d
F (t) = = f (x) dx = f (t).
dt dt c

That is, differentiation ‘undoes’ (or ‘reverses’) the process of integration as, given the
integral of f , differentiation allows us to find f .

Integration ‘undoes’ differentiation

Conversely, if F (t) is an antiderivative of f (t), say we have


Z t
F (t) = f (x) dx,
c

where c is a constant, as before, then for some constant d, we also have


Z t Z t Z d
f (x) dx = f (x) dx − f (x) dx = F (t) − F (d),
d c c

76
3.2. The Fundamental Theorem of Calculus

using Theorem 3.1(d) and, of course, F (d) is a constant. That is, integration ‘undoes’
(or ‘reverses’) the process of differentiation as, given the derivative of F , i.e. f ,
integration allows us to find F up to a constant. Indeed, we find that if F1 (t) and F2 (t)
are any two antiderivatives of f , we have

F10 (t) = f (t) and F20 (t) = f (t),

which means that 3


 
d
F1 (t) − F2 (t) = F10 (t) − F20 (t) = f (t) − f (t) = 0,
dt
and so the function F1 (t) − F2 (t) must be a constant as its derivative is zero.
Consequently, any two antiderivatives of f can only differ by a constant and so, as we
should expect, integration only ‘undoes’ differentiation up to an arbitrary constant.

In summary

Having established that the expected relationships between integration and


differentiation hold, we can now evaluate Riemann integrals in the manner described in
Chapter 5 of 174 Calculus, i.e. we can use our standard integrals and rules of
integration.

3.2.4 Some applications of the FTC


We are now in a position to consider some other applications of the FTC.

Using the FTC and Taylor series to approximate the value of an integral

At the beginning of this chapter, we observed that the integral


Z 1
2
e−x dx,
0
2
could not be found in the standard way as the integrand, e−x , has no antiderivative. Of
course, we could use our interpretation of the integral as an area to gain some
approximation to the value of this integral, but the FTC allows us to approximate its
value in another way as the next example shows.

Example 3.9 Given that Z t


2
F (t) = e−x dx,
0

find the Taylor series for F (t) about t = 0 in terms up to t3 . Hence find an
approximate value for F (1).

To find the Taylor series for F (t) about t = 0, we note that


Z t Z 0
−x2 2
F (t) = e dx =⇒ F (0) = e−x dx = 0,
0 0

77
3. The Riemann integral

using Theorem 3.1(a) and that


Z t 
0 d −x2 2
F (t) = e dx = e−t =⇒ F 0 (0) = 1,
dt 0

using the FTC. We then see that,


3 F 00 (t) = −2t e−t
2
=⇒
2 2
F 000 (t) = (−2)(e−t ) + (−2t)(−2t e−t ) = 2(2t2 − 1) e−t ,
2

which means that F 00 (0) = 0 and F 000 (0) = −2. Thus, we have

t2 00 t3 t3
F (t) = F (0) + tF 0 (0) + F (0) + F 000 (0) + · · · = t − + · · · ,
2! 3! 3
as the Taylor series for F (t) about t = 0 in terms up to t3 . Consequently, we can see
that
1 2
F (1) ' 1 − = ,
3 3
is the required approximate value for F (1). Indeed, as
Z 1
2
F (1) = e−x dx,
0

this gives us an approximate value of 0.67 (2dp) for this integral and this is fairly
close to the true value which turns out to be 0.75 (2dp).

Using the FTC to define functions in terms of definite integrals

We can also use the FTC to define certain functions in terms of definite integrals as the
next activity shows.
2
Activity 3.10 The function, F (x), satisfies the conditions F 0 (x) = e−x and
F (0) = 0. Write F (x) as a definite integral of the form
Z x
F (x) = f (t) dt,
A

for a suitable choice of f and A.

Using the FTC to solve integral equations

We can also use the FTC to solve certain integral equations, i.e. equations that involve
definite integrals, as the next example shows.
Z t
Example 3.10 Use the FTC to solve the integral equation f (t) = 1 − f (x) dx.
0

If we differentiate both sides of this integral equation with respect to t, we get


f 0 (t) = −f (t),

78
3.2. The Fundamental Theorem of Calculus

where we have used the FTC to differentiate the integral. Solving this simple
separable differential equation in the standard way9 we get

f (t) = A e−t ,

for some constant A ∈ R. Indeed, setting t = 0 in the integral equation, we see that
Z 0
3
f (0) = 1 − f (x) dx =⇒ f (0) = 1 − 0 = 1,
0

using Theorem 3.1(a), and so A = 1. Consequently, a function f that satisfies this


integral equation is
f (t) = e−t ,
and so this is the required solution.

In fact, we will see another way of solving integral equations like this in Chapter 7 when
we look at Laplace transforms.

3.2.5 An extension of the FTC


We can also use the FTC to differentiate more complicated integrals if we use the chain
rule. For instance, if we have the integral
Z q(t)
F (t) = f (x) dx,
p(t)

we can use Theorem 3.1(d) to write it as


Z q(t) Z p(t)
F (t) = f (x) dx − f (x) dx,
c c

where c is some constant. Then, differentiating both sides of this expression with
respect to t, we get
Z q(t) ! Z p(t) !
d d
F 0 (t) = f (x) dx − f (x) dx ,
dt c dt c

so that, using the chain rule, we have


Z q  Z p 
0 dq d dp d
F (t) = f (x) dx − f (x) dx .
dt dq c dt dp c

Then, applying the FTC, this gives us


d q(t)
Z
0
F (t) = f (x) dx = q 0 (t)f (q(t)) − p0 (t)f (p(t)),
dt p(t)
where, of course, this result assumes that p and q are both differentiable functions of t
and that f is continuous over the two intervals of integration. Let’s look at an example
to see how this works.

9
See, for example, Section 8.2.1 of 174 Calculus.

79
3. The Riemann integral

Example 3.11 Given that


Z t3
F (t) = x2 dx,
4t
0
find F (t) using the result above.

Here we have p(t) = 4t and q(t) = t3 , which means that


3
F 0 (t) = (3t2 )(t3 )2 − (4)(4t)2 = 3t8 − 64t2 ,

if we use the result above.

Activity 3.11 Following on from Example 3.11, use integration to find F (t) and
hence verify your answer for F 0 (t).

Indeed, we will see in Section 6.1.3 that this result can be extended even further when
we meet the Leibniz rule for differentiating integrals.

Using the FTC to derive identities

This extension of the FTC also has an interesting application because it can be used to
derive certain identities like the ones we saw in Section 2.1.4 of 174 Calculus.

Example 3.12 Use integration to verify that


Z t
1
ln t = dx,
1 x

where t > 0. Hence show that if a > 0 is a constant and F is given by

F (t) = ln(at) − ln t,

where t > 0, then F 0 (t) is zero. Deduce that F (t) is a constant and, by finding this
constant, deduce that
ln(ab) = ln a + ln b,
where b > 0 is a constant too.

Note: Of course, if we use the fact that

ln(ab) = ln a + ln b =⇒ ln(ab) − ln b = ln a,

for a, b > 0, then it is obvious that

F (t) = ln(at) − ln t = ln a,

for a, t > 0. Indeed, as a is a constant, this means that F (t) = ln a is a constant too
and so, obviously, F 0 (t) = 0! However, here we are using an integral that represents
ln x and the function F (t) to derive this fact and so we can not use it in this way
here.

80
3.2. The Fundamental Theorem of Calculus

Firstly, it is easy to verify that, for t > 0, we have


Z t  t
1
dx = ln x = ln t − ln 1 = ln t,
1 x 1

as ln 1 = 0. Then, for a, t > 0, we can use this to write


Z at
1
Z t
1 3
F (t) = ln(at) − ln t = dx − dx,
1 x 1 x

so that, using the FTC and its extension as appropriate, we have


 
0 1 1 1 1
F (t) = (a) − = − = 0,
at t t t

as required. But, of course, if F 0 (t) = 0, this means that F (t) = c where c is a


constant which we can find by taking any value of t > 0. So, if we let t = 1, say, we
have
c = F (1) = ln a − ln 1 = ln a,
as the value of the constant. Consequently, we have

F (t) = ln a = ln(at) − ln t,

which means that, if we rearrange this and let t = b where b > 0, we have

ln(ab) = ln a + ln b,

as required.

You can establish a slightly trickier identity for yourself in Activity 3.12.

Activity 3.12 Use integration to verify that


Z t
−1 1
tan t = 2
dx.
0 1+x

Hence show that if a is a constant and F is given by


 
−1 a+t
F (t) = tan − tan−1 t,
1 − at

then F 0 (t) is zero. Deduce that F (t) is a constant and, by finding this constant,
deduce that  
−1 −1 −1 a+b
tan a + tan b = tan ,
1 − ab
where b is a constant too.

81
3. The Riemann integral

Learning outcomes
At the end of this chapter and having completed the relevant reading and activities, you
should be able to:

use the definition of the Riemann integral to find the value of a definite integral;
3
use the Fundamental Theorem of Calculus to differentiate definite integrals;

use definite integrals and their derivatives in various applications.

Solutions to activities
Solution to activity 3.1
Given what we saw in Section 5.3 of 174 Calculus, we would expect the area, A, of the
region bounded by the curve y = 1 + x2 , the x-axis and the vertical lines x = −2 and
x = 2 to be given by the integral
Z 2
A= (1 + x2 ) dx,
−2

as the integrand, 1 + x2 , is a non-negative function over the interval [−2, 2]. Evaluating
this integral, we then find that
2
x3 23 (−2)3
        
8 8 1
A= x+ = 2+ − (−2) + = 2+ − −2 − =9 .
3 −2 3 3 3 3 3

We can then use our results from Example 3.1, i.e. that L(P) = 5 and U (P) = 17, to
see that as
1
5 < 9 < 17,
3
they do indeed satisfy the inequality L(P) < A < U (P).

Solution to activity 3.2


Looking at the lower estimate, we have

n−1
       
1 0 1 1 1 2 1
L(Pn ) = 1+ + 1+ + 1+ + ··· + 1+
n n n n n n n n
!
1 0 + 1 + 2 + · · · + (n − 1)
= 1| + 1 + 1{z+ · · · + 1} +
n n
n times
1 + 2 + · · · + (n − 1)
 
1
= n+
n n
1 n−1
  
1
= n+ (n − 1 + 1) ,
n n 2

82
3.2. Solutions to activities

where we have used the formula for the sum of the first n − 1 natural numbers to get
the term in the square brackets. If we now simplify this expression, we get

n−1 1 3n − 1
     
1 1 1
L(Pn ) = n+ = = 3− ,
n 2 n 2 2 n

as required. Similarly, looking at the upper estimate, we have


      3
1 1 1 2 1 3 1 n
U (Pn ) = 1+ + 1+ + 1+ + ··· + 1+
n n n n n n n n
!
1 1 + 2 + 3 + ··· + n
= 1 + 1 + 1{z+ · · · + 1} +
n | n
n times
 
1 1 hn i
= n+ (n + 1) ,
n n 2

where we have used the formula for the sum of the first n natural numbers to get the
term in the square brackets. If we now simplify this expression, we get
     
1 n+1 1 3n + 1 1 1
U (Pn ) = n+ = = 3+ ,
n 2 n 2 2 n

as required.

Solution to activity 3.3


Looking at the illustrations in Figure 3.4, it should be clear that the area, A, of the
region bounded by the curve y = 1 + x, the x-axis and the vertical lines x = 0 and
x = 1 is 3/2 as it is the area of a trapezium (or the area of a right-angled triangle plus
the area of a square).10 We can then use our results from Example 3.2, i.e. that
   
1 1 1 1
L(Pn ) = 3− and U (Pn ) = 3+ ,
2 n 2 n

to see that, as n > 0, we have


   
1 1 1 1 3 1 1
3− <3<3+ =⇒ 3− < < 3+ ,
n n 2 n 2 2 n

and so they do indeed satisfy the inequality L(Pn ) < A < U (Pn ).

10
Alternatively, we can see that it is given by the integral
Z 1
A= (1 + x) dx,
0

as the integrand, 1 + x, is a non-negative function over the interval [0, 1]. Evaluating this integral, we
then find that 1 
x2 12 02
   
1 3
A= x+ = 1+ − 0+ =1+ = ,
2 0 2 2 2 2
as we would expect from above.

83
3. The Riemann integral

Solution to activity 3.4


Suppose that f (x) is a non-negative continuous function defined over some interval [a, b]
and that P is the partition
{x0 , x1 , x2 , . . . , xn−1 , xn } where a = x0 < x1 < x2 < · · · < xn = b.
Now, if f (x) is an increasing function on the interval [a, b] and A is the area of the
3 region bounded by the curve y = f (x), the x-axis and the vertical lines x = a and x = b,
we can see that the

lower estimate is given by


n
X n
X
L(P) = (xi − xi−1 )mi = (xi − xi−1 )f (xi−1 ),
i=1 i=1

where mi , the minimum value of f (x) for values of x in the sub-interval [xi−1 , xi ], is
simply f (xi−1 ) as the function f (x) is increasing over [a, b] and so it is increasing
over all of the sub-intervals as illustrated in Figure 3.10(a).
upper estimate is given by
n
X n
X
U (P) = (xi − xi−1 )Mi = (xi − xi−1 )f (xi ),
i=1 i=1

where Mi , the maximum value of f (x) for values of x in the sub-interval [xi−1 , xi ],
is simply f (xi ) as the function f (x) is increasing over [a, b] and so it is increasing
over all of the sub-intervals as illustrated in Figure 3.10(b).

y y
y = f (x) y = f (x)

... ...

O x x1 x2 ... xn−1 xn x O x x1 x2 ... xn−1 xn x


0 0

(a) (b)

Figure 3.10: When f (x) is an increasing function. In (a) and (b) we have, respectively,
the rectangles that contribute towards the lower estimate, L(P), and the upper estimate,
U (P), of A based on the partition P. Note that, for clarity, only the first, second and last
rectangles are shown.

Of course, if f (x) is a decreasing function on the interval [a, b] and A is the area of the
region bounded by the curve y = f (x), the x-axis and the vertical lines x = a and x = b,
we can see that the

84
3.2. Solutions to activities

lower estimate is given by


n
X n
X
L(P) = (xi − xi−1 )mi = (xi − xi−1 )f (xi ),
i=1 i=1

where mi , the minimum value of f (x) for values of x in the sub-interval [xi−1 , xi ], is
simply f (xi ) as the function f (x) is decreasing over [a, b] and so it is decreasing
over all of the sub-intervals.
3
upper estimate is given by
n
X n
X
U (P) = (xi − xi−1 )Mi = (xi − xi−1 )f (xi−1 ),
i=1 i=1

where Mi , the maximum value of f (x) for values of x in the sub-interval [xi−1 , xi ],
is simply f (xi−1 ) as the function f (x) is decreasing over [a, b] and so it is decreasing
over all of the sub-intervals.
If you are having any trouble understanding this, try sketching the analogue of the
illustrations in Figure 3.10 with a function, f (x), which is now decreasing over the
interval [a, b].

Solution to activity 3.5


Comparing this with our answers from Example 3.1 and Activity 3.1, we now have the
values
1
5 < 6 < 9 < 14 < 17,
3
which means that we have

L(P) < L(P 0 ) < A < U (P 0 ) < U (P).

That is, both of our lower estimates are less than the value of A but L(P 0 ) is a better
lower estimate of A than L(P) whereas both of our upper estimates are greater than
the value of A but U (P 0 ) is a better upper estimate of A than U (P).
Indeed, it should be clear that this improvement in the lower and upper estimates has
come about because P 0 has one more point than P and this has enabled us to replace
the single rectangle for the sub-interval [−2, 0] from P (see Figure 3.3) with two
‘thinner’ rectangles for the sub-intervals [−2, −1] and [−1, 0] from P 0 (see Figure 3.5).

Solution to activity 3.6


If we sketch the curve y = 1 + x2 over the interval [0, 2] and indicate the points in the
partition Pm we see
 k−1that we will be looking at 2m rectangles whose bases are given by
k
the sub-intervals m , m for k = 1, 2, . . . , 2m. Indeed, we see that, for k = 1, 2, . . . , 2m,
each of these sub-intervals gives us a base whose length is

k k−1 1
− = ,
m m m
and so, for the

85
3. The Riemann integral

lower estimate, we need to sum the areas of the 2m rectangles illustrated in


Figure 3.11(a) where the height of each rectangle is given by the minimum value of
1 + x2 in each of the sub-intervals to get
2 ! 2 ! 2 !
2m − 1
  
1 0 1 1 1
L(Pm ) = 1+ + 1+ + ··· + 1+ ,
m m m m m m
3
where we have taken each of the sub-intervals in turn and used ‘base times height’
to find the area of each rectangle. This gives us
!
1 02 + 12 + 22 + · · · + (2m − 1)2
L(Pm ) = 1| + 1 + 1{z+ · · · + 1} +
m m2
2m times
12 + 22 + · · · + (2m − 1)2
 
1
= 2m +
m m2
1 2m − 1
  
1
= 2m + 2 ([2m − 1] + 1)(2[2m − 1] + 1) ,
m m 6

where we have used the formula for the sum of the first 2m − 1 square numbers to
get the term in the big square brackets. If we now simplify this expression, we get

(2m − 1)(2m)(4m − 1)
 
1
L(Pm ) = 2m +
m 6m2
 
2m 2
= 6m + (2m − 1)(4m − 1)
6m3
 
1 2 2
= 6m + (8m − 6m + 1)
3m2
 
1 2
= 14m − 6m + 1
3m2
6m − 1
 
1
∴ L(Pm ) = 14 − .
3 m2

Now we can see that, for m ≥ 1, we have

6m − 1
 
1 14
L(Pm ) = 14 − ≤ ,
3 m2 3

and so L∗ = 14/3 is the smallest number that is larger than all the lower bounds.

upper estimate, we need to sum the areas of the 2m rectangles illustrated in


Figure 3.11(b) where the height of each rectangle is given by the maximum value of
1 + x2 in each of the sub-intervals to get
 2 !  2 !  2 !
1 1 1 2 1 2m
U (Pm ) = 1+ + 1+ + ··· + 1+ ,
m m m m m m

where we have taken each of the sub-intervals in turn and used ‘base times height’

86
3.2. Solutions to activities

to find the area of each rectangle. This gives us


!
2 2 2
1 1 + 2 + · · · + (2m)
U (Pm ) =
m |1 + 1 + 1{z+ · · · + 1} + m2
2m times
12 + 22 + · · · + (2m)2
 
1
= 2m +
m m2
1
 
1 2m
 3
= 2m + 2 (2m + 1)(2[2m] + 1) ,
m m 6

where we have used the formula for the sum of the first 2m square numbers to get
the term in the big square brackets. If we now simplify this expression, we get
 
1 (2m)(2m + 1)(4m + 1)
U (Pm ) = 2m +
m 6m2
 
2m 2
= 6m + (2m + 1)(4m + 1)
6m3
 
1 2 2
= 6m + (8m + 6m + 1)
3m2
 
1 2
= 14m + 6m + 1
3m2
 
1 6m + 1
∴ U (Pm ) = 14 + .
3 m2

Now we can see that, for m ≥ 1, we have


 
1 6m + 1 14
U (Pm ) = 14 + 2
≥ ,
3 m 3

and so U ∗ = 14/3 is the largest number that is smaller than all the upper bounds.
This means that there is only one number, I, that satisfies the inequality

L∗ ≤ I ≤ U ∗ ,

in this case, i.e. I = 14/3, and so we take this to be the value of the given Riemann
integral. That is, Z 2
14
(1 + x2 ) dx = ,
0 3
and, indeed, this is what we should expect since
2 2 
x3 23 03
   
8 14
Z
2
(1 + x ) dx = x + = 2+ − 0+ =2+ = ,
0 3 0 3 3 3 3

if we evaluate the integral in the usual way.

87
3. The Riemann integral

y y

y = 1 + x2 y = 1 + x2
5 5

3
... ...
1 1

O x O x
1 2 ... 2m−1
2 1 2 ... 2m−1
2
m m m m m m

(a) (b)

Figure 3.11: In (a) and (b) we have, respectively, the rectangles that contribute towards
the lower estimates, L(Pm ), and the upper estimates, U (Pm ), based on the partition Pm .
Note that, for clarity, some of the y-intercepts have been omitted and only three of the
rectangles are shown.

Solution to activity 3.7


Looking at the lower estimates, we have

n−2
         
1 0 1 1 1 2 1 1
L(Pn ) = 2· + 2· + 2· + ··· + 2· + 1
n n n n n n n n n
2 0 + 1 + 2 + · · · + (n − 2) 1
 
= +
n n 2
2 1 + 2 + · · · + (n − 2) 1
 
= +
n n 2
2 1 n−2
   
1
= (n − 2 + 1) + ,
n n 2 2

where we have used the formula for the sum of the first n − 2 natural numbers to get
the term in the square brackets. If we now simplify this expression, we get

n−1
   
1 1 2
L(Pn ) = 2 (n − 2)(n − 1) + n = 2 n − 2n + 2 = 1 − 2 · ,
n n n2

as required. Indeed, since n ≥ 2, we have n − 1 > 0 and n2 > 0, which means that

n−1
L(Pn ) = 1 − 2 · ≤ 1,
n2

as we were asked to verify.

88
3.2. Solutions to activities

Similarly, for the upper estimates, we have


n−1
       
1 1 1 2 1 3 1 1  n
U (Pn ) = 2· + 2· + 2· + ··· + 2· + 2·
n n n n n n n n n n
 
2
= 2 1 + 2 + 3 + · · · + (n − 1) + n
n
3
 
2 n
= 2 (n + 1) ,
n 2
where we have used the formula for the sum of the first n natural numbers to get the
term in the square brackets. If we now simplify this expression, we get
1 1
U (Pn ) = (n + 1) = 1 + ,
n n
as required. Indeed, since n ≥ 2, we have
1
U (Pn ) = 1 + ≥ 1,
n
as we were asked to verify.

Solution to activity 3.8


As we are assuming that f is a non-negative function whose Riemann integral exists
over some interval containing the points a, b and c we know that all of the Riemann
integrals will represent the corresponding areas. So, by considering these areas, we can
see that Theorem 3.1(a) holds, i.e. that
Z a
f (x) dx = 0,
a

since the area bounded by the curve y = f (x), the x-axis and the vertical lines x = a
and x = a must be zero as the ‘base’ of this region is zero. We can also see that
Theorem 3.1(b) holds, i.e. that
Z b Z c Z b
f (x) dx = f (x) dx + f (x) dx,
a a c

since, for c ∈ [a, b], the area bounded by the curve y = f (x), the x-axis and the vertical
lines x = a and x = b is the same as the sum of the area bounded by the curve
y = f (x), the x-axis and the vertical lines x = a and x = c and the area bounded by the
curve y = f (x), the x-axis and the vertical lines x = c and x = b.

We can now deduce that Theorem 3.1(c) holds, i.e. that


Z b Z a
f (x) dx = − f (x) dx,
a b

since, thinking about Theorem 3.1(a) and (b), we have


Z a Z b Z a
0= f (x) dx = f (x) dx + f (x) dx.
a a b

89
3. The Riemann integral

Of course, using this in Theorem 3.1(b), Theorem 3.1(d) follows immediately.

Solution to activity 3.9


Following on from what we saw in Section 3.2.1, when we have h < 0 so that t + h < t,
we are now interested in the difference between the two areas A(t + h) and A(t) given by

3 A(t) − A(t + h),

which is the area of the vertically hatched region of base |h| indicated in Figure 3.12. In
particular, we see that since f (x) is a decreasing function over the interval [a, b], we
have a

lower estimate of this area given by the area of the shaded rectangle in
Figure 3.9(a) which, using ‘base times height’, is

|h|f (t),

and so we have |h|f (t) ≤ A(t) − A(t + h).

upper estimate of this area given by the area of the shaded rectangle in
Figure 3.9(b) which, using ‘base times height’, is

|h|f (t + h),

and so we have A(t) − A(t + h) ≤ |h|f (t + h).


Consequently, putting these two inequalities together, we have
A(t) − A(t + h)
|h|f (t) ≤ A(t) − A(t + h) ≤ |h|f (t + h) =⇒ f (t) ≤ ≤ f (t + h),
|h|

as |h| > 0. Indeed, as h < 0, we have |h| = −h and so this gives us

A(t) − A(t + h) A(t + h) − A(t)


f (t) ≤ ≤ f (t + h) =⇒ f (t) ≤ ≤ f (t + h),
−h h
which means that, taking the limit as h → 0− on both sides of this inequality, we get
A(t + h) − A(t) A(t + h) − A(t)
lim− f (t) ≤ lim− ≤ lim− f (t+h) =⇒ f (t) ≤ lim− ≤ f (t),
h→0 h→0 h h→0 h→0 h
if we use the continuity of f (x) over [a, b].11 Of course, this means that

A(t + h) − A(t)
lim− = f (t),
h→0 h
as required.

11
Notice that the continuity of f (x) over [a, b] is essential here in order to justify the assertion that
the limit
lim− f (t + h),
h→0

exists and that it is equal to f (t). Also, notice that we can only take the limit as h → 0− here as we
have assumed that h < 0.

90
3.2. Solutions to activities

y y

0000
011111 1111
0000 3
0000
101111 0000
1111
101111 0000
1111
y = f (x) y = f (x)
0000
101111
0000 0000
1111
101111
0000
101111 0000
1111
0000
1111
O
a c t+h t
0000
10 b
x O
a c 0000
1111
t+h t b
x

(a) (b)

Figure 3.12: The areas of the shaded regions in (a) and (b) are the lower and upper
estimates, respectively, of the area A(t) − A(t + h) of the vertically hatched region. Notice
that here we have h < 0 and so the ‘base’ of the rectangles is |h|.

Solution to activity 3.10


If, as instructed, we write F (x) as a definite integral of the form
Z x
F (x) = f (t) dt,
A

then, assuming that A is a constant,12 the FTC tells us that


Z x 
0 d
F (x) = f (t) dt = f (x),
dx A
2
and, as we are told that F 0 (x) = e−x , this gives us
Z x
−x2 2
f (x) = e =⇒ F (x) = e−t dt,
A

for a suitable choice of the [assumed] constant, A. Now, we are also told that F (0) = 0,
which means that Z 0
2
F (0) = e−t dt = 0,
A
2
and, since this is the area between the positive function e−t , the t-axis and the vertical
lines t = A and t = 0, this means that we must take A = 0 [which is a constant] for this
to hold. Consequently, we see that
Z x
2
F (x) = e−t dt,
0

is the required definite integral.


12
If this assumption turns out to give us nonsense, then we would have to try something more
complicated.

91
3. The Riemann integral

Solution to activity 3.11


Here, we have
t3 t3
x3 (t3 )3 (4t)3 t9 64
Z 
2
F (t) = x dx = = − = − t3 ,
4t 3 4t 3 3 3 3

3 and so
F 0 (t) = 3t8 − 64t2 ,
in agreement with what we found in Example 3.11.

Solution to activity 3.12


Firstly, it is easy to verify that we have
Z t  t
1
2
−1
dx = tan x = tan−1 t − tan−1 0 = tan−1 t,
0 1+x 0

using the result from Example 5.17 of 174 Calculus and noting that tan−1 0 = 0 because
tan 0 = 0. Then, if a is a constant, we can use this to write
a+t
  t
a+t 1 1
Z Z
1−at
−1 −1
F (t) = tan − tan t= dx − dx,
1 − at 0 1 + x2 0 1 + x2
so that, using the FTC and its extension as appropriate, we have
!
− −
 
(1)(1 at) (a + t)(−a) 1 1
F 0 (t) = 2 − ,
(1 − at)2 1 + a+t 1 + t2

1−at

where we have used the quotient rule to differentiate the upper limit of the first
integral. Now, considering the first term, we have
!
(1)(1 − at) − (a + t)(−a) 1 − at + a2 + at
 
1
2 =
(1 − at)2 1 + a+t (1 − at)2 + (a + t)2

1−at
1 + a2
=
(1 − 2at + a2 t2 ) + (a2 + 2at + t2 )
1 + a2
=
1 + a2 t2 + a2 + t2
1 + a2
=
(1 + a2 )(1 + t2 )
1
= ,
1 + t2
which means that
1 1
F 0 (t) =
2
− = 0,
1+t 1 + t2
as required. But, of course, if F 0 (t) = 0, this means that F (t) = c where c is a constant
which we can find by taking any value of t. So, if we let t = 0, say, we have

c = F (0) = tan−1 a − tan−1 0 = tan−1 a,

92
3.2. Exercises

as the value of the constant. Consequently, we have


 
−1 −1 a+t
F (t) = tan a = tan − tan−1 t,
1 − at
which means that, if we rearrange this and let t = b, we have
 
−1 −1 −1 a+b 3
tan a + tan b = tan ,
1 − ab
as required.

Exercises
Exercise 3.1
Consider, for n ∈ N, the partition P = {1, 2, . . . , n} of the interval [1, n]. Find the lower
and upper estimates, L(P) and U (P) respectively, of the integral
Z n
1
dx,
1 x(x + 1)

simplifying your answers as far as possible.

Exercise 3.2
Consider, for n ∈ N, the partition

Pn = 0, n1 , n2 , . . . , 1 ,


of the interval [0, 1]. Find the lower and upper estimates, L(Pn ) and U (Pn ) respectively,
of the integral Z 1
ex dx,
0
simplifying your answers as far as possible.

Find lim L(Pn ) and lim U (Pn ). What do these tell us about the value of the integral?
n→∞ n→∞

Exercise 3.3
Z sin t
0
Find f (t) when f (t) = x3 dx.
cos t

Exercise 3.4
Find the following limits.
t
1
Z
t
(a) lim+ t ln t, (b) lim+ t , (c) lim+ xx dx.
t→0 t→0 t→0 t 0
Z t
Hence find lim+ (ln t) xx dx.
t→0 0

93
3. The Riemann integral

Exercise 3.5
Let f be a continuous function that takes positive real values and suppose that
Z t2
G(t) = f (x) dx.
t
0
Find G (t) and hence use a Taylor series to find a first-order approximation to G(t) for
3 values of t close to 1.

Solutions to exercises
Solution to exercise 3.1
If we sketch the curve y = 1/x(x + 1) over the interval [1, n] and indicate the points in
the partition P we see that we will be looking at n − 1 rectangles whose bases are given
by the sub-intervals [k − 1, k] for k = 2, 3, . . . , n. Indeed, we see that, for k = 2, 3, . . . , n,
each of these sub-intervals gives us a base whose length is
k − (k − 1) = 1,
and so, for the

lower estimate, we need to sum the areas of the n − 1 rectangles illustrated in


Figure 3.13(a) where the height of each rectangle is given by the minimum value of
1/x(x + 1) in each of the sub-intervals to get
       
1 1 1 1
L(P) = 1 +1 +1 + ··· + 1
2(2 + 1) 3(3 + 1) 4(4 + 1) n(n + 1)
1 1 1 1
= + + + ··· + ,
2·3 3·4 4·5 n(n + 1)
where we have taken each of the sub-intervals in turn and used ‘base times height’
to find the area of each rectangle. Then, using partial fractions, we see that
1 1 1
= − ,
k(k + 1) k k+1
and so we can write
       
1 1 1 1 1 1 1 1 1 1
L(P) = − + − + − + ··· + − = − ,
2 3 3 4 4 5 n n+1 2 n+1
as the intermediate terms cancel.
upper estimate, we need to sum the areas of the n − 1 rectangles illustrated in
Figure 3.13(b) where the height of each rectangle is given by the maximum value of
1/x(x + 1) in each of the sub-intervals to get
       
1 1 1 1
U (P) = 1 +1 +1 + ··· + 1
1(1 + 1) 2(2 + 1) 3(3 + 1) (n − 1)(n − 1 + 1)
1 1 1 1
= + + + ··· + ,
1·2 2·3 3·4 (n − 1)n

94
3.2. Solutions to exercises

where we have taken each of the sub-intervals in turn and used ‘base times height’
to find the area of each rectangle. Then, using partial fractions again, we can write
       
1 1 1 1 1 1 1 1 1
U (P) = − + − + − + ··· + − =1− ,
1 2 2 3 3 4 n−1 n n
as the intermediate terms cancel.
Thus, we have found that the lower and upper estimates of the given integral are 3
1 1 1
L(P) = − and U (P) = 1 − ,
2 n+1 n
respectively.

y y

1 1
y= y=
x(x + 1) x(x + 1)

... ...
O ... x O ... x
1 2 3 4 n−1 n 1 2 3 4 n−1 n
(a) (b)

Figure 3.13: The sketches for Exercise 3.1. In (a) and (b) we have, respectively, the
rectangles that contribute towards the lower estimates, L(P), and the upper estimates,
U (P), based on the partition P. Note that, for clarity, the y-intercepts have been omitted
and only four of the rectangles are shown.

Solution to exercise 3.2


If we sketch the curve y = ex over the interval [0, 1] and indicate the points in the
partition Pn we
 seekthat we will be looking at n rectangles whose bases are given by the
sub-intervals k−1

n
, n
for k = 1, 2, . . . , n. Indeed, we see that, for k = 1, 2, . . . , n, each of
these sub-intervals give us a base whose length is
k k−1 1
− = ,
n n n
and so, for the

lower estimate, we need to sum the areas of the n rectangles illustrated in


Figure 3.14(a) where the height of each rectangle is given by the minimum value of
ex in each of the sub-intervals to get
1 0 1 1/n 1
L(Pn ) = e + e + · · · + e(n−1)/n ,
n n n

95
3. The Riemann integral

where we have taken each of the sub-intervals in turn and used ‘base times height’
to find the area of each rectangle. This gives us
1 0
e + e1/n + · · · + e(n−1)/n

L(Pn ) =
n
1
1 + e1/n + · · · + e(n−1)/n

=
n
3 1 1 − (e1/n )n

=
n 1 − e1/n
1−e
 
1
= ,
n 1 − e1/n

where we have used the formula for the sum of a geometric series to get the final
answer. We now see that writing

1−e
 
1 1/n
lim L(Pn ) = lim 1/n
= (e −1) lim 1/n ,
n→∞ n→∞ n 1−e n→∞ e −1

we are dealing with the limit of a quotient where the numerator and the
denominator both tend to zero as n → ∞. As such, we can use L’Hôpital’s rule to
see that
−1/n2 1
lim L(Pn ) = (e −1) lim 2 1/n
= (e −1) lim 1/n = e −1,
n→∞ n→∞ (−1/n ) e n→∞ e

is the first of the sought after limits.

upper estimate, we need to sum the areas of the n rectangles illustrated in


Figure 3.14(b) where the height of each rectangle is given by the maximum value of
ex in each of the sub-intervals to get
1 1/n 1 2/n 1
U (Pn ) = e + e + · · · + e1 ,
n n n
where we have taken each of the sub-intervals in turn and used ‘base times height’
to find the area of each rectangle. This gives us
1 1/n
e + e2/n + · · · + e1

U (Pn ) =
n
e1/n
1 + e1/n · · · + e(n−1)/n

=
n 
1/n
1 − (e1/n )n

e
=
n 1 − e1/n
e1/n 1−e
 
= ,
n 1 − e1/n

where we have used the formula for the sum of a geometric series to get the final
answer. We now see that writing

e1/n 1−e
 
1/n
lim U (Pn ) = lim = (e −1) lim ,
n→∞ n→∞ n 1−e 1/n n→∞ 1 − e−1/n

96
3.2. Solutions to exercises

we are dealing with the limit of a quotient where the numerator and the
denominator both tend to zero as n → ∞. As such, we can use L’Hôpital’s rule to
see that
−1/n2 1
lim U (Pn ) = (e −1) lim = (e −1) lim = e −1,
n→∞ n→∞ −(1/n2 ) e−1/n n→∞ e−1/n

is the second of the sought after limits. 3


What do these limits tell us about the value of the integral? Well, as n increases, the
estimates we get from the partition Pn should get better and better so that, in
particular, as we take the limit as n → ∞, we get the best estimates that Pn can provide.
Indeed, as these best lower and upper estimates are the same, this should lead us to
conclude that this common value is the value of the integral, i.e. we have found that
Z 1
ex dx = e −1,
0

which is, of course, exactly what we would expect to get if we found the integral in the
usual way!

y y

y = ex y = ex

... ...
1 1

O x O x
1 2 ... n−1
1 1 2 ... n−1
1
n n n n n n
(a) (b)

Figure 3.14: The sketches for Exercise 3.2. In (a) and (b) we have, respectively, the
rectangles that contribute towards the lower estimates, L(Pn ), and the upper estimates,
U (Pn ), based on the partition Pn . Note that, for clarity, the y-intercepts have been
omitted and only three of the rectangles are shown.

Solution to exercise 3.3


Given that Z sin t
f (t) = x3 dx,
cos t

we can apply our extension to the FTC from Section 3.2.5 to see that

f 0 (t) = (cos t)(sin t)3 − (− sin t)(cos t)3 = sin t cos t(sin2 t + cos2 t) = sin t cos t,

97
3. The Riemann integral

if we use the trigonometric identity sin2 t + cos2 t = 1.

Solution to exercise 3.4


For (a), we write
ln t
lim+ t ln t = lim+
,
t→0 t→0 1/t

3 so that we have a quotient where the numerator and the denominator both tend to
infinity [in magnitude] as t → 0+ . As such, we can use L’Hôpital’s rule to see that
ln t 1/t
lim+ t ln t = lim+ = lim+ = lim −t = 0.
t→0 t→0 1/t t→0 −1/t2 t→0+
For (b), we use the fact that tt = et ln t so that
lim tt = lim+ et ln t = e0 = 1,
t→0+ t→0

if we use our answer from (a).

For (c), we write Rt x


1 t x x dx
Z
lim+ x dx = lim+ 0 ,
t→0 t 0 t→0 t
so that we have a quotient where the numerator and the denominator both tend to zero
as t → 0+ . As such, we can use L’Hôpital’s rule and the FTC to see that
Rt x
1 t x x dx tt
Z
lim+ x dx = lim+ 0 = lim+ = 1,
t→0 t 0 t→0 t t→0 1

if we use our answer from (b).

Then, lastly, we can see that if we use our results from (a) and (c), we have
Z t   Z t 
x 1 x
lim (ln t) x dx = lim+ t ln t x dx = (0)(1) = 0,
t→0+ 0 t→0 t 0
as the final answer.

Solution to exercise 3.5


If f is a continuous function that takes positive real values and we are given that
Z t2
G(t) = f (x) dx,
t
then, using our extension of the FTC from Section 3.2.5, we find that
G0 (t) = (2t)f (t2 ) − (1)f (t) = 2tf (t2 ) − f (t).
As such, noting that
Z 1
G(1) = f (x) dx = 0 and G0 (1) = 2(1)f (1) − f (1) = f (1),
1
we see that a first-order Taylor series about t = 1 gives us
G(t) = G(1) + (t − 1)G0 (1) + · · · = 0 + (t − 1)f (1) + · · · = (t − 1)f (1) + · · · ,
and so (t − 1)f (1) is the required first-order approximation to G(t) for values of t close
to 1.

98
Chapter 4
Improper integrals

Essential reading

(For full publication details, see Chapter 1.)


4
+ Binmore and Davies (2002) Section 10.11.

+ Ostaszewski (1991) Sections 18.6, 18.7, 18.9 and 18.10.

Further reading

+ Adams and Essex (2010) Section 6.5.

+ Wrede and Spiegel (2010) parts of Chapter 12.

Aims and objectives

The objectives of this chapter are:

to introduce various kinds of improper integral and see what it means to say that
they are convergent or divergent;

to establish various tests that allow us to see whether an improper integral is


convergent or divergent.
Specific learning outcomes can be found near the end of this chapter.

4.1 Improper integrals


The definition of the definite integral in Chapter 3 assumed that the range of integration
is a finite closed interval (i.e. a set of real numbers t restricted so that a ≤ t ≤ b for some
a, b ∈ R) and that the integrand was continuous on that closed interval. These are the
proper integrals. This chapter is concerned with continuous integrands and extends the
definition of integral in two ways thus creating improper integrals of two kinds, namely:

The first kind permits a semi-infinite range of integration, i.e. the range takes the
form a ≤ t or t ≤ b.

The second kind permits a finite range of the form a ≤ t < b or a < t ≤ b and so
permits the integrand to be undefined at an end-point. This, in particular, admits

99
4. Improper integrals

unbounded continuous functions that may, for instance, tend to plus infinity as t
tends to b or a respectively.
More generally, we may encounter improper integrals that are the sum of two improper
integrals of the first or second kind. We will call these improper integrals of the third
kind. For instance, the infinite range of all real numbers may be regarded as a sum of
two half-infinite ranges.
All three kinds of improper integral can be given meaning as a limit of proper integrals
and so the only issue at stake is whether the limit exists. As such, the main thrust of this
chapter is to develop tools which allow us to verify that the relevant limit exists. The
most basic of these tools will, for example, take a given integrand, f (t), and compare its
4 behaviour against a simpler integrand, g(t). If this latter integrand has an integral
which is known to be convergent or divergent, then we should (under appropriate
circumstances) be able to deduce a similar result about the integral of the former. Thus,
the task of this chapter is to identify the circumstances under which information about
the integral of g(t) will allow us to make deductions about the integral of f (t).

4.1.1 Improper integrals of two kinds, and a third kind


Problem 1. If f (t) is defined and continuous for all t ≥ a, we know the meaning of the
proper integral Z x
F (x) = f (t) dt,
a

for x ≥ a. We now want to define an improper integral of the first kind to be an integral
of the form Z ∞
f (t) dt,
a

and we will interpret this integral as the limit

lim F (x),
x→∞

if it exists.
In some simple cases it may be easy to compute the integral and to have it available for
inspection as a well known function F (x). But in general the problem might be that all
we know about F (x) is numerical computer output for certain values of x and a graph
on some necessarily finite range. Such circumstantial finite evidence is of limited use.
We need tools for verifying the behaviour of F (x) for large x from knowledge of f (t). In
a word: we are up against calculation of some specific limits! And, in Problem 1, the
problem location is behaviour at infinity.
We know from Chapter 2 that, speaking quite generally, a function F (x) might not have
a limit as x tends to infinity. But, given special properties of f (t), perhaps we can verify
that the limit of F (x) as x → ∞ is finite. Our objective is to find such verification tools.
Problem 2. If f (t) is defined and continuous for all t with a ≤ t < b, but is not even
defined at t = b, can we make sense of
Z b
f (t) dt = lim− F (x)?
a x→b

100
4.1. Improper integrals

A similar problem arises in the other direction, when f (t) is defined and continuous for
all t with a < t ≤ b, but is not even defined at t = a, can we make sense of
Z b
f (t) dt = lim+ F (x)?
a x→a

These are called improper integrals of the second kind. For now, we’ll consider the
former case, but similar remarks will apply to the latter.
Comment on removable discontinuities: If we find that f (t) is undefined at t = b
but the limit of f (t) as t → b− is some finite number, say l, then we can ‘repair’ it by
extending its definition, i.e. we can just set f (b) = l. With this ‘repair job’ done, f (t) is
now continuous at t = b, and the integral 4
Z b
f (t) dt,
a

can now be interpreted as a ‘proper’ integral in the sense that we discussed earlier.1

Example 4.1 The integral


1
sin t
Z
dt,
0 t
is an improper integral of the second kind because the integrand is not defined at
t = 0. But, as we know from Example 2.22,
sin t
lim+ = 1,
t→0 t
and so we can ‘repair’ the integrand by taking f (0) = 1 and
sin t
f (t) = ,
t
if t 6= 0. With this done, we see that the integrand is now continuous on the interval
[0, 1] and so we can interpret the given integral as a ‘proper’ integral.

Consequently, we see that Problem 2 is really an issue when the limit of f (b) as t → b−
is ±∞ or does not exist at all. However, even when this limit is ±∞, we can still find
cases where the limit of F (x) as x → b− is finite and, as such, the improper integral will
be convergent. And, in Problem 2, the problem location is behaviour at t = b.
Problem 3 Of course we can be confronted by a combination of the two previous
problems. For instance, can we make sense of the integral
Z ∞ −t
e
dt,
0 t
which has a semi-infinite range of integration and an integrand which is undefined at
t = 0? The easy answer is to divide the given problem into two problems and see what
1
Strictly speaking, the function obtained by setting f (b) = l is a different function (because it has a
larger domain of definition) and so one ought to use a different name, e.g. f ∗ , to denote the new function.
However, it is normal practice to avoid this extra formalism.

101
4. Improper integrals

we can do with them separately. That is, we consider separately (and independently)
whether the limits Z 1 −t Z b −t
e e
lim+ dt and lim dt,
a→0 a t b→∞ 1 t
exist. Integrals such as these (which mix one or two kinds of improperness) are called
improper integrals of the third kind.
Objective: When all is said and done the material of the present chapter is devoted to
answering the following specific question and showing examples.
Knowing the dominant behaviour of f (t) at the problem location, what can we
4 say about the behaviour of F (x)?
What we will do is build up a tool-kit comprising: (1) test functions which are
important examples of convergent and divergent behaviour, representing simple
standard modes of decay (‘rates of convergence’); and (2) tests with which to draw
conclusions from test functions. No tool-kit is complete without Taylor series (for
creating approximations) and L’Hôpital’s rule (for carrying out tests).

4.1.2 Some further thoughts on improper integrals


Recall that an improper integral of the first kind has a semi-infinite range, say [a, ∞),
and an integrand, say f (t), which is continuous over the range. Its definition is
Z ∞ Z b
f (t) dt = lim f (t) dt,
a b→∞ a
assuming the limit on the right-hand side exists. Thus the terminology corresponds to a
problem location at infinity.

Positive integrands

Note, in particular, that if the integrand, f (t), is positive on the range of integration,
then the integral Z b
F (b) = f (t) dt,
a
is an increasing function of b. That is, more range gives more area. In Figure 4.1 we see
that as b changes from b1 to b2 the fainter area represents the increase in F (b). This
observation makes the story particularly straightforward because it can only give rise to
two possible scenarios:
(i) If the values of F (b) increase to a finite limit, i.e. we have
Z b
lim f (t) dt = L
b→∞ a
for some L ∈ R, then we say that the integral is convergent.
(ii) Otherwise, if the values of F (b) do not increase to a finite limit, i.e. we have
Z b
lim f (t) dt = ∞,
b→∞ a
then we say that the integral is divergent.

102
4.1. Improper integrals

y = f (t)

4
O t
a b1 b2

Figure 4.1: If f (t) is a positive function for a ≤ t ≤ b, we see that as b increases from b1
to b2 , the area under the curve y = f (t), i.e. the value of F (b), increases.

A comment on the behaviour of ‘tails’

As a consequence of the definition of improper integrals of the first kind, an equivalent


way to test Z ∞
f (t) dt
a
for convergence is to verify that the area under the ‘tail’ of the integrand tends to zero.
That is, this integral will converge if
Z ∞
f (t) dt → 0
T

as T → ∞.2

Tool-kit 1: Inverse power decay at infinity

At infinity, inverse powers above one behave rather like an inverse square. We have
Z b  −1 b
dx x 1
2
= = 1 − → 1 as b → ∞.
1 x −1 1 b
Now note the distinction between this last result and the following two examples of
inverse powers with powers at or below one:
Z b  b
dx
= ln x = ln b → ∞ as b → ∞,
1 x 1

and  1/2 b
Z b
dx x √
= = 2( b − 1) → ∞ as b → ∞.
1 x1/2 1/2 1
2
But notice that this ‘area under the tail’ is itself an improper integral and so we can not use this as
a definition of what it means to say that an improper integral converges.

103
4. Improper integrals

The inverse square case is covered by the general observation that for m > 0
Z b  −m b
dx x 1 −m 1
1+m
= = (1 − b ) → as b → ∞.
1 x −m 1 m m

Note that as m → 0 this area increases unboundedly; this is consistent with the m = 0
result we saw above.
On the other hand for m > 0 we have
Z b  m b
dx x 1
1−m
= = (bm − 1) → ∞ as b → ∞.
1 x m 1 m
4
If you are surprised by the special case of the integrand x−1 (corresponding to m = 0 in
the last two calculations) take a look at Figure 4.2. The diagram is meant to suggest
that, for x ≥ 1, as m decreases from 1 to 1/2 and so on towards zero, the corresponding
graphs of 1/xm+1 ‘rise’ as m decreases leading to an area which increases unboundedly
with m as m decreases. However, for strictly positive m, the area between the graph
and the x-axis for the infinite range x ≥ 1 is finite.

Figure 4.2: The graph of the function 1/xm+1 for m = 1 (dotted line), m = 1/2 (dashed
line) and m = 0 (solid line). Notice that as m decreases to zero, the graphs ‘rise’ leading
to areas that increase unboundedly with m as m decreases.

Tool-kit 2: Inverse power growth at the origin

The results here need to be contrasted with those in Tool-kit 1.


Z 1  1/2 1
dx x √
1/2
= = 2(1 − a) → 2 as a → 0+ .
a x 1/2 a

104
4.1. Improper integrals

On the other hand


1  1
dx
Z
= ln x = − ln a → ∞ as a → 0+ ,
a x a

and  −1 1
1
dx x 1
Z
2
= = − 1 → ∞ as a → 0+ .
a x −1 a a
Here, in contrast with what we saw in Tool-kit 1, we have convergence for inverse
powers when the power is below one; and we have divergence for inverse powers above
or at one.
For instance, with m > 0, we have
4
Z 1  −m 1
dx x 1
1+m
= = (a−m − 1) → ∞ as a → 0+ .
a x −m a m
To see the reason for the difference, draw the graph of
1
y= ,
xk
for x > 0 and then turn round the diagram to make the y-axis horizontal. This way
behaviour at infinity is interchanged with the behaviour at the origin. The graph above
the y-axis (now in horizontal position) is that of
1
x= ,
y 1/k
but we have k > 1 if and only if 1/k < 1.

Tool-kit 3: Exponential decay at infinity

For any r > 0 we have the familiar result


Z b  −rt b
−rt e 1 1
e dt = = (1 − e−rb ) → as b → ∞.
0 −r 0 r r
We consider whether a more general version of this result might hold, and we ask for
what functions p(t) does the following improper integral
Z ∞
e−rt p(t)dt
0

exist? The answer will need to depend on how fast p(t) grows as t tends to infinity
relative to e−rt . We stop to motivate the significance of this question.
Z ∞
Example 4.2 For r > 0, consider the improper integral e−rt p(t) dt.
0
When does this improper integral exist? A good question to ask is: what happens
when p(t) itself is an exponential? Consider, as a special case of p(t), the function

q(t) = eγt ,

105
4. Improper integrals

where γ > 0. We then find that


T T T T
e(γ−r)t e(γ−r)T −1
Z Z Z 
−rt −rt γt (γ−r)t
e q(t) dt = e e dt = e dt = =
0 0 0 γ−r 0 γ−r

and this converges as T → ∞ if and only if r > γ. The limiting value, when it exists,
is
1
.
r−γ
Thus the answer depends on whether or not r is larger than γ.
4
Looking for similar behaviour

What happens to the functions

t 1
f1 (t) = and f2 (t) = .
1 + t2 1 + t2

as t → ∞? On identifying the dominant behaviour in the denominators we have


     
t 1 1 1 1 1
f1 (t) = 2 1 = 1 and f2 (t) = 2 1 ,
t 1+ t2
t 1+ t2
t 1+ t2

and so we might well presume (correctly) that f2 (t) has a convergent integral, being like
1/t2 , and f1 has a divergent integral, being like 1/t. For these examples it is easy to
compute directly what is happening.

b  b
t 1 1
Z
2
dt = ln(1 + t ) = ln(1 + b2 ) → ∞ as b → ∞,
2
0 1+t 2 0 2

whereas
b  b
1 π
Z
2
dt = tan (t) = tan−1 (b) → as b → ∞.
−1
0 1+t 0 2

The choice of example here was dictated by our ability to do the integration. We now
justify the intuitive view, based on dominant behaviour, when we are unable to do the
integration. There will be two methods: the Direct Comparison Test and the, generally
faster, Limit Comparison Test. Indeed, for the latter test, the approach below follows
Ostaszewski (1991) in wanting to stress what is a valid rule, and what is not.

4.2 Tests for convergence and divergence


We now introduce the tests that will allow us to determine whether an improper
integral with a positive integrand is convergent or divergent. We will consider how to
deal with integrands of variable sign in the next section.

106
4.2. Tests for convergence and divergence

4.2.1 The Direct Comparison Test


There are evidently two comparison procedures that one can use for a function, f (t),
that is positive, namely:

over-estimation: that is consideration of a test function g(t) such that


f (t) ≤ g(t), for all t in some tail t ≥ T.

under-estimation: that is consideration of a test function g(t) such that


g(t) ≤ f (t), for all t in some tail t ≥ T.
For current purposes (checking for convergence of an improper integral), this technique 4
may seem to present too much awkwardness either in finding the T beyond which the
comparison of f and g is valid, or in identifying the right test function (even though the
dominant behaviour may be obvious). The easier Limit Comparison Test (coming soon!)
fixes all that, but we should not therefore avoid direct comparison as it has other
important applications later – both in this chapter when we consider integrands with
variable sign and when we study dominated convergence.
The first form of the Direct Comparison Test uses over-estimation as follows.

(DCT1) If 0 ≤ f (t) ≤ g(t) for all t > T , then the convergence of the integral
Z ∞
g(t) dt,
T

implies the convergence of the integral


Z ∞
f (t) dt.
T

The reason why this works is that, as 0 ≤ f (t) ≤ g(t) for t ≥ T , the test function g(t)
over-estimates the integrand f (t) and so we have
Z ∞ Z ∞
0≤ f (t) dt ≤ g(t) dt,
T T

i.e. f (t) has less area in its tail than g(t). In Figure 4.3, the over-estimate of the area
from the test function g(t) is indicated by the shading.


dt
Z
Example 4.3 Test the improper integral for convergence.
1 t2
+t
The dominant term in the denominator as t → ∞ is t2 . Fortunately, we can omit the
‘smaller’ term, t, because it is positive. Indeed, for t ≥ 0 we have

t2 + t ≥ t2 ,

so that
1 1
f (t) = ≤ 2 = g(t).
t2 +t t

107
4. Improper integrals

1111111111111
0000000000000
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
4 0000000000000
1111111111111
0000000000000
1111111111111 t
O
T
Figure 4.3: If, for t ≥ T , we have 0 ≤ f (t) ≤ g(t), then the graph of f (t) (solid line) gives
an area (hatched region) which is less than the area under the graph of g(t) (dashed line).
The corresponding over-estimate to this area from g(t) is indicated by the shading.

Of course, g(t) has a convergent tail, i.e.


Z ∞
g(t) dt,
T

is finite. Hence the area under the tail of f (t), i.e.


Z ∞
f (t) dt,
T

is at most finite and so the given integral converges.

In light of what we have just done, it is also useful to note that if we had used the
fact that
t2 + t ≥ t,
for t ≥ 0 so that we have
1 1
f (t) = ≤ = g(t),
t2
+t t
then we would not be able to draw any useful conclusion. This is because the area
under the tail of g(t) is now infinite and so the area under the tail of f (t) is at most
infinite which doesn’t tell us anything useful!

However, generally speaking, the DCT can be quite difficult to apply in cases where it is
difficult to find an appropriate over-estimate as the next example shows.


dt
Z
Example 4.4 Test the improper integral for convergence.
1 t2 − t + 1
Like the improper integral in Example 4.3, this improper integral does converge for
ultimately the same reason: its similarity to 1/t2 . But finding the right over-estimate
is a bit awkward. We can omit the positive term +1, as in the previous example, so

108
4.2. Tests for convergence and divergence

making the integrand larger. But we cannot omit the troublesome negative term −t
as that would make the integrand smaller rather than larger. So instead we make the
negative term more negative but not too negative by replacing −t by −t2 /2. Indeed,
we have
t2 t2
−t ≥ − ⇐⇒ t≤ ⇐⇒ 2 ≤ t.
2 2
and so, for t ≥ 2, this gives us

t2 t2
t2 − t + 1 ≥ t2 − t ≥ t2 − = .
2 2
Thus 4
1 2
f (t) = ≤ 2 = g(t).
t2 −t+1 t
Again, g(t) has a convergent tail, i.e.
Z ∞ ∞
2
Z
g(t) dt = dt,
2 2 t2

is finite. Hence the area under the tail of f (t), i.e.


Z ∞
f (t) dt,
2

is at most finite. Hence the given integral converges. Note that to make good the
comparisons, we had to look at the range t ≥ 2 rather than t ≥ 1. But, this is fine
here as the integral Z 2
f (t) dt,
1

is finite because f (t) is continuous.

The second form of the Direct Comparison Test uses under-estimation as follows.

(DCT2) If 0 ≤ g(t) ≤ f (t) for all t > T , then the divergence of the integral
Z ∞
g(t) dt,
T

implies the divergence of the integral


Z ∞
f (t) dt.
T

The reason why this works is that, as 0 ≤ g(t) ≤ f (t) for t ≥ T , the test function g(t)
under-estimates the integrand f (t) and so we have
Z ∞ Z ∞
0≤ g(t) dt ≤ f (t) dt,
T T

i.e. f (t) has more area in its tail than g(t). In Figure 4.4, the under-estimate of the area
from the test function g(t) is indicated by the shading.

109
4. Improper integrals

1111111111111
0000000000000
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
4 0000000000000
1111111111111
0000000000000
1111111111111 t
O
T
Figure 4.4: If, for t ≥ T , we have 0 ≤ f (t) ≤ g(t), then the graph of f (t) (solid line)
gives an area (hatched region) which is greater than the area under the graph of g(t)
(dashed line). The corresponding under-estimate to this area from g(t) is indicated by
the shading.


dt
Z
Example 4.5 Test the improper integral for convergence.
1 t1/2
+t
For t ≥ 1, we have
t1/2 + t ≤ t + t = 2t,
and so
1 1
f (t) = ≥ = g(t).
t1/2 + t 2t
Of course, g(t) has a divergent tail, i.e.
Z ∞
g(t) dt,
1

is infinite. Hence, the area under the tail of f (t), i.e.


Z ∞
f (t) dt,
1

being larger, is also infinite. Hence the given integral diverges.


2 + sin t
Z
Example 4.6 Test the improper integral dt for convergence.
1 t
For all t ∈ R, we have
2 + sin t ≥ 1,
and so, for t ≥ 1, we have
2 + sin t 1
f (t) = ≥ = g(t).
t t

110
4.2. Tests for convergence and divergence

Of course, g(t) has a divergent tail, i.e.


Z ∞
g(t) dt,
1

is infinite. Hence, the area under the tail of f (t), i.e.


Z ∞
f (t) dt,
1

being larger, is also infinite. Hence the given integral diverges.


4
Activity 4.1 Decide whether each of the following statements is true or false and
give a reason for your answers.
R∞
(a) If f and g are non-negative continuous functions, such
R∞ that 0
f (t) dt
converges, while g(t) ≤ M for some fixed M , then 0 f (t)g(t) dt converges.
R∞
(b) If f is a continuous function, and f (t) ≥ c for some c > 0, then 0 f (t) dt
diverges.

Activity 4.2
R∞
(a) Let f be a non-negative
R ∞continuous function, and suppose that 1
f (x) dx
2
converges. Show that 1 f (y ) dy converges.

(b) For t ≥ 1, give


R ∞an example of a non-negative
R ∞ continuous function, f (t), which
2
is such that 1 f (x ) dx converges but 1 f (y) dy diverges.

4.2.2 The Limit Comparison Test


The first form of the Limit Comparison Test runs as follows.

(LCT1) If f (t) and g(t) are both positive for all t ≥ a and
f (t)
lim = 1,
t→∞ g(t)
then
Z ∞ Z ∞
f (t) dt is convergent if and only if g(t) dt is convergent.
a a

Interpretation: This test is saying that if the test function g(t) is a good
approximation to f (t) for arbitrarily large values of t (i.e. the limit of the ratio of these
two functions is one), then what you know about the integral of g(t) being finite or
infinite is also true of the integral of f (t).
Comment: It is best to learn this rule as stated above where the limit of the ratio of
f (t) and g(t) as t → ∞ is one, rather than the more general statement where we can

111
4. Improper integrals

replace the limit in (LCT1) with

f (t)
lim = c where 0 < c < ∞.
t→∞ g(t)

Of course, it makes no real difference because, if we find that this latter limit holds, we
actually have
f (t)
lim = 1 where 0 < c < ∞,
t→∞ cg(t)

and all this says is that the re-scaled function cg(t), rather than g(t) itself, is the ‘right’
4 approximation to f (t) for (LCT1).
Proof of (LCT1): Given that f (t) and g(t) are both positive for all t ≥ a, if we find
that
f (t)
lim = 1,
t→∞ g(t)

then we can guarantee that, for suitably large T , we have (say)

1 f (t) 3
< < ,
2 g(t) 2

for all t > T . That is, we have the over and under-estimations given by

1 3
g(t) < f (t) < g(t),
2 2
which means that
∞ ∞ ∞
1 3
Z Z Z
g(t) dt ≤ f (t) dt ≤ g(t) dt.
2 T T 2 T

Thus, if Z ∞
g(t) dt = G,
T

where G is some finite number, then we find that


Z ∞
3
f (t) dt ≤ G,
T 2

and the tail area of f (t) is finite. But, on the other hand, if
Z ∞ Z ∞
g(t) dt = ∞, then we know that g(t) dt ≥ M,
T T

for any finite M , and so as this implies that


Z ∞
1
M≤ f (t) dt,
2 T

for any finite M , the tail area of f (t) is also infinite.

112
4.2. Tests for convergence and divergence

Example 4.7 Following on from Example 4.3, test the improper integral
Z ∞
dt
2
,
1 t +t
for convergence.

We apply the Limit Comparison Test using the following procedure:


Step 0: Classification. This is an improper integral of the first kind, i.e. the problem
location is at infinity.
Step 1: Approximation. Investigate the integrand, f (t), to discover the dominant 4
behaviour as t → ∞. As
1 1
f (t) = = 2 ,
t+t 2 t (1 + 1t )
we see that, for large t, the dominant behaviour is given by
1
g(t) = .
t2

Step 2: Compute the ratio. We see that

f (t) t2 t2 1
= = 1 = → 1,
g(t) t + t2 t2 (1 + t ) (1 + 1t )

as t → ∞.
Z ∞ Z ∞
Step 3: Draw conclusions. As g(t) dt is finite, so is f (t) dt by the LCT.
1 1

Activity 4.3 Use the Limit Comparison Test to determine whether the following
integrals converge.

x1/3 + 5
Z
(a) dx.
1 x2 − 6x + 10
Z ∞
π 
(b) − tan−1 x dx.
1 2
[Hint: For (b), the answer to Activity 2.14 will help.]


dx
Z
Activity 4.4 For which real numbers r does the integral converge?
2 x(ln x)r


x
Z
Activity 4.5 Determine whether the integral dx converges.
−∞ 1 + x2


dx
Z
Activity 4.6 Determine whether the integral √ converges.
1 x4 + x3 − x2

113
4. Improper integrals

The one-way convergence test

What happens when the ratio in the Limit Comparison Test has a limit of zero? The
answer is that we get the second form of the Limit Comparison Test which runs as
follows.

(LCT2) If f (t) and g(t) are both positive for all t ≥ a and
f (t)
lim = 0,
t→∞ g(t)

4 then
Z ∞ Z ∞
g(t) dt is convergent implies that f (t) dt is convergent.
a a

Interpretation: This test is saying that if the test function g(t) over-estimates f (t) for
arbitrarily large values of t (i.e. the limit of the ratio of these two functions is zero) and
the integral of g(t) is finite, then the integral of f (t) is finite too (hence a possible
mnemonic here is ‘the zero-finite test’).
Comment: Notice that this is a one-way test because it tells us nothing about the
integral of f (t) in cases where the integral of g(t) is divergent.
Proof of (LCT2): Given that f (t) and g(t) are both positive for all t ≥ a, if we find
that
f (t)
lim = 0,
t→∞ g(t)

then we can guarantee that, for all suitably large T , we have (say)
f (t)
< 1,
g(t)
for all t > T . That is, we have an over-estimation given by
f (t) < g(t),
which means that Z ∞ Z ∞
f (t) dt < g(t) dt.
T T
Thus, if Z ∞
g(t) dt,
a
is finite, then the tail area of f (t) is also finite.
We’ll illustrate this test by looking at functions, f (t), that we can think of as the
product of simpler factors. But, first, let’s describe our strategy in words before we firm
the ideas up with examples. We begin by following intuition in order to isolate a single
‘significant’ factor that one suspects might be responsible for the convergence of the
integral of f (t). One wants the other factors to remain bounded. But, it may happen
that the other potentially non-significant factors are actually unbounded. In such a
situation, we can try borrowing some of the ‘convergence power’ of the significant factor
so as to ‘drown out the unboundedness’. This borrowing needs to be sufficiently ‘small’
so that the altered significant term still gives us a convergent integral.

114
4.2. Tests for convergence and divergence


1 −t
Z
Example 4.8 Show that the improper integral e dt converges.
1 t
Here, the integrand, f (t) = e−t /t, is composed of two factors. The factor e−t will
give us convergence and the factor 1/t is bounded (by one as 1/t ≤ 1 if t ≥ 1). As
such, we can try to apply the Limit Comparison Test with the test function
g(t) = e−t . Consequently, as
f (t) 1
= → 0,
g(t) t
as t → ∞ and we have
∞ ∞ ∞ 4
e−t
Z Z 
g(t) dt = e−t dt = = 1,
1 1 −1 1

which is finite, we can conclude that the given integral is also convergent by the
LCT.


| sin t| −t
Z
Example 4.9 Show that the improper integral e dt converges.
1 t
As the integrand, f (t) = | sin t| e−t /t is positive for t ≥ 1, we can try to apply the
LCT and we again take g(t) = e−t . The other factor of f (t) remains bounded, indeed

f (t) | sin t|
= → 0,
g(t) t

as t → ∞. Then, as we again have


Z ∞ ∞ ∞
e−t
Z 
−t
g(t) dt = e dt = = 1,
1 1 −1 1

which is finite, we can conclude that the given integral is also convergent by the
LCT.


1 −t
Z
Example 4.10 Show that the improper integral e dt converges.
1 t2
We are spoilt for choice here as either of the factors e−t and 1/t2 will generate
convergence. You can check this for yourself.


ln(1 + t2 )
Z
Example 4.11 Show that the improper integral dt converges.
1 t2
As f (t) = ln(1 + t2 )/t2 is positive for t ≥ 1, we can try to apply the LCT. Here
ln(1 + t2 ) is unbounded and so we will not choose 1/t2 for the test function, but
something a little less quick in its convergence to zero, namely 1/t3/2 . This leaves
something over to get a ratio whose limit is zero. Indeed, we have
f (t) ln(1 + t2 ) 3/2 ln(1 + t2 )
lim = lim t = lim ,
t→∞ g(t) t→∞ t2 t→∞ t1/2

115
4. Improper integrals

and so, noting that the numerator and the denominator both tend to ∞ as t → ∞,
we can use L’Hôpital’s rule to see that

ln(1 + t2 ) 2t/(1 + t2 ) 4t3/2


lim = lim = lim = 0.
t→∞ t1/2 t→∞ 1/(2t1/2 ) t→∞ 1 + t2

Then, as we have  −1/2 ∞


∞ ∞
dt t
Z Z
g(t) dt = 3/2
= = 2,
1 1 t −1/2 1
which is finite, we can conclude that the given integral is also convergent by the
4 LCT.

Of course, as it is a one-way test, it doesn’t always work as the next example shows.

Example 4.12 Consider, for t ≥ 1, the functions


1 1
f (t) = and g(t) = ,
t2 t
which are such that
f (t) t 1
= 2 = → 0,
g(t) t t
as t → ∞. In this case, we have a ratio whose limit is zero, but the integral
Z ∞ Z ∞
dt
g(t) dt = ,
1 1 t
is divergent and so our one-way rule tells us nothing in this case. In fact here we see
that the integral Z ∞
f (t) dt,
1
is actually convergent.

Z ∞ √
Activity 4.7 Show that the improper integral t8 e− t
dt is convergent.
0

The one-way divergence test

What happens when the ratio in the Limit Comparison Test has a limit of infinity? The
answer is that we get the third form of the Limit Comparison Test which runs as follows.

(LCT3) If f (t) and g(t) are both positive for all t ≥ a and
f (t)
lim = ∞,
t→∞ g(t)
then
Z ∞ Z ∞
g(t) dt is divergent implies that f (t) dt is divergent.
a a

116
4.2. Tests for convergence and divergence

Interpretation: This test is saying that if the test function g(t) under-estimates f (t)
for arbitrarily large values of t (i.e. the limit of the ratio of these two functions is
infinity) and the integral of g(t) is infinite, then the integral of f (t) is infinite too (hence
a possible mnemonic here is ‘the infinite-infinite test’).
Comment: Notice that this is a one-way test because it tells us nothing about the
integral of f (t) in cases where the integral of g(t) is convergent.
Proof of (LCT3): Given that f (t) and g(t) are both positive for all t ≥ a, if we find
that
f (t)
lim = ∞,
t→∞ g(t)
4
then we can guarantee that, for all suitably large T , we have (say)

f (t)
> 1,
g(t)

for all t > T . That is, we have an over-estimation given by

f (t) > g(t),

which means that Z ∞ Z ∞


f (t) dt > g(t) dt.
T T

Thus, if Z ∞
g(t) dt,
a

is infinite, then the tail area of f (t) is also infinite.


We apply this test in much the same way as we applied the previous one. But, of
course, as it is a one-way test, it doesn’t always work as the next example shows.

Example 4.13 Consider, for t ≥ 1, the functions


1 1
f (t) = and g(t) = ,
t t2
which are such that
f (t) t2
= = t → ∞,
g(t) t
as t → ∞. In this case, we have a ratio whose limit is infinity, but the integral
Z ∞ Z ∞
dt
g(t) dt = ,
1 1 t2
is convergent and so our one-way rule tells us nothing in this case. In fact here we
see that the integral Z ∞
f (t) dt,
1
is actually divergent.

117
4. Improper integrals

Improper integrals of the second kind

So far, we have looked at the Limit Comparison Test for improper integrals of the first
kind. But, all three forms of this test also apply to improper integrals of the second
kind. For instance, consider a function f (t) that is defined on the interval [a, b) so that
the integral Z bZ x
f (t) dt = lim− f (t) dt,
a x→b a
has a problem at t = b. If we appropriately replace references to
Z b Z x
4 lim f (t) dt with references to lim− f (t) dt,
b→∞ a x→b a

with similar replacements in the limits and the integrals involving g(t), then all three
forms of the Limit Comparison Test now hold for such an improper integral of the
second kind.

1
ln(1 + t)
Z
Example 4.14 Show that the improper integral dt converges.
0 t
There’s nothing obvious to do here as the integral
Z 1
dt
,
0 t

is divergent. However, we can try a lower power, say g(t) = 1/ t, as the integral
Z 1
dt
√ ,
0 t
is convergent. Indeed, this is a good idea, because

f (t) ln(1 + t) √ ln(1 + t)


lim+ = lim+ t = lim+ √ ,
t→0 g(t) t→0 t t→0 t

and so, noting that the numerator and the denominator both tend to zero as t → 0+ ,
we can use L’Hôpital’s rule to see that

ln(1 + t) 1/(1 + t) 2t1/2


lim+ √ = lim+ = lim = 0.
t→0 t t→0 1/(2t1/2 ) t→0+ 1 + t

So, by the zero-finite form of the LCT, the given integral converges.

Activity 4.8 Determine whether each of these integrals converges.


1
dt
Z
(a) .
0 sin t
Z 1
(b) (− ln t) dt.
0

118
4.2. Tests for convergence and divergence

1
dt
Z
(c) .
0 t1/3 − t4/3

1 − e−u
Z
(d) du.
0 u
[Hint: It’s useful to start by carefully thinking about the integrands so that you can
see what the issues are. (For example, does the integrand have an asymptote?) If
necessary you can ‘divide and conquer’ by breaking the range of integration into
pieces to isolate the various potential problems.]

Some further examples to improve technique 4


The examples we consider here may seem remote from applications, but their purpose is
to improve your technique when using the Limit Comparison Test.

dt
Z
Example 4.15 Test the integral √  for convergence.
1 t t2 +t−t

Step 0: Classification. This is an improper integral of the first kind, i.e. the problem
location is at infinity.
Step 1: Approximation. Investigate the integrand, f (t), to discover the dominant
behaviour as t → ∞. As the denominator can be written as
!
√  r
1
t t2 + t − t = t2 1+ −1 ,
t
we can approximate the square-root by using its Taylor series around the origin.
(This is justifiable since, if t is large, 1/t can be assumed to be small.) That is, since
r  1/2
1 1 1 1
1+ = 1+ = 1 + − 2 + ··· ,
t t 2t 8t
we have
r !   
1 1 1 t 1
t2 1+ −1 =t 2
1 + − 2 + ··· − 1 = − + ··· .
t 2t 8t 2 8
As such, we find that
1 1
f (t) = √ = t 1 ,
t t2 +t−t 2
− + ···
8

and so, for large t, the dominant behaviour is given by the test function
2
g(t) = .
t
Step 2: Compute the ratio. We see that
t
f (t) 2
− 18 + ... 1
= = 1 − + ... → 1,
g(t) t/2 4t
as t → ∞.

119
4. Improper integrals

Z ∞ Z ∞
Step 3: Draw conclusions. As g(t) dt is divergent, so is f (t) dt by the LCT.
1 1

Z 1
Example 4.16 Test the integral t−4/3 tan−1 t dt for convergence.
0

Step 0: Classification. This is an improper integral of the second kind, i.e. the
problem location is at t = 0.
Step 1: Approximation. Investigate the integrand, f (t), to discover the dominant
behaviour as t → 0+ . The Taylor series for tan−1 t near the origin gives us
4
tan−1 (t) = 0 + t + ... = t + · · · ,

as tan−1 (0) = 0 and its first derivative is 1/(1 + t2 ) which is one when t = 0. So, we
can see that
f (t) = t−4/3 tan−1 t = t−4/3 (t + · · · ) = t−1/3 + · · · ,
and so, for small t, the dominant behaviour is given by the test function

g(t) = t−1/3 .

Step 2: Compute the ratio. We see that

f (t) t−4/3 tan−1 (t) tan−1 (t)


lim+ = lim+ = lim ,
t→0 g(t) t→0 t−1/3 t→0+ t

and so, noting that the numerator and the denominator both tend to zero as t → 0+ ,
we can use L’Hôpital’s rule to see that

tan−1 (t) 1/(1 + t2 ) 1


lim+ = lim+ = lim+ = 1.
t→0 t t→0 1 t→0 1 + t2

Z 1 Z 1
Step 3: Draw conclusions. Since g(t) dt is convergent, so too is f (t) dt by the
0 0
LCT.
Z ∞
Example 4.17 Test the integral t−4/3 tan−1 t dt for convergence.
1

Step 0: Classification. This is an improper integral of the first kind, i.e. the problem
location is at infinity.
Step 1: Approximation: Investigate the integrand, f (t), to discover the dominant
behaviour as t → 0+ . In this case, we know that as t gets very large, tan−1 t gets
closer to π/2 and so we take our test function to be
π
g(t) = t−4/3 .
2
Step 2: Compute the ratio. We see that
f (t) t−4/3 tan−1 (t) 2 tan−1 (t)
= π −4/3 = → 1,
g(t) 2
t π

120
4.2. Tests for convergence and divergence

as t → ∞.
Z ∞ Z ∞
Step 3: Draw conclusions. Since g(t) dt is convergent, so too is f (t) dt by
1 1
the LCT.

Z ∞
Example 4.18 Test the integral t−4/3 tan−1 t dt for convergence.
0

This is an improper integral of the third kind and, as it is the sum of the improper
integrals from the last two examples, both of which were convergent, it is convergent
too. 4
4.2.3 Variable sign integrands
So far in this chapter, we have been concerned with improper integrals which have
positive integrands. We now consider what happens when the integrand has variable
sign. In particular, we will be making use of the following result.

(VSI) If the function, f (t), is continuous for t ≥ a and the integral


Z ∞
|f (t)| dt,
a

is convergent, Z ∞
f (t) dt,
a
is also convergent.
We will not even consider a full justification of this result, which is based on considering
separately (i) the area under the curve where the function f (t) is positive and (ii) the
area under the curve corresponding to where the function f (t) is negative. Indeed,
under the assumptions of the result both of these areas are finite. Instead, in what
follows, we will use a direct comparison technique to get around the variable sign.
Indeed, sometimes, this will need a preparatory step that uses integration by parts as
this can be used to alter the integrand so that it contains a factor which converges to
zero more rapidly.


cos t
Z
Example 4.19 Show that the improper integral dt is convergent.
1 t2
Here we can use a direct comparison by referring to a crude inequality. That is, as
| cos t| ≤ 1 for all t ∈ R, we have

cos t | cos t| 1
|f (t)| = 2 = 2
≤ 2,
t t t
Z ∞ Z ∞
dt
for t ≥ 1. So, as the integral is convergent, so is the integral |f (t)| dt.
1 t2 1
Consequently, the given integral is convergent by (VSI).

121
4. Improper integrals


sin t
Z
Example 4.20 Show that the improper integral dt is convergent.
1 t
Clearly the method of the last example won’t help since, even though

sin t
t ≤ 1,

for t ∈ R, the integral



dt
Z
,
t
4 1
is divergent. However we can ‘speed up’ the integrand if we use integration by parts
to get
b b Z b Z b
− cos t

sin t 1 cos b cos t
Z
dt = (− cos t) − 2
dt = − + cos 1 − dt.
1 t t 1 1 −t b 1 t2

Now, taking the limit as b → ∞, we note that


Z b Z ∞
cos b cos t cos t
→ 0 and dt → dt,
b 1 t2 1 t2
if this limit exists. However, as we saw in Example 4.19, this last integral is
convergent which means that we have
Z ∞ Z ∞
sin t cos t
dt = cos 1 − dt,
1 t 1 t2
and as this is finite, the given integral is convergent too.

Activity 4.9 Use Taylor’s theorem to show that, for 0 < t < π,
sin t
0< < 1,
t
and hence deduce that
sin t
0 < < 1,
t
for all t > 0. Use this to show that, for x > 0,
Z ∞
sin t −xt
1
e dt ≤ ,

0 t x

and hence verify that the equality


Z ∞ Z ∞  
sin t −xt sin t −xt
lim e dt = lim e dt,
x→∞ 0 t 0 x→∞ t

holds.

122
4.2. Learning outcomes

Learning outcomes
At the end of this chapter and having completed the relevant reading and activities, you
should be able to:

identify different kinds of improper integral;

test whether a given improper integral is convergent or divergent.

Solutions to activities 4
Solution to activity 4.1
For (a), we can use the DCT to see that the statement is true. Indeed, as g(t) ≤ M , we
have f (t)g(t) ≤ M f (t) and so, as the integral
Z ∞
f (t) dt,
0

is convergent, so is the integral


Z ∞ Z ∞
M f (t) dt = M f (t) dt,
0 0

which means that the integral Z ∞


f (t)g(t) dt,
0
is also convergent by the DCT.

For (b), we can also use the DCT to see that the statement is true. We know that
f (t) ≥ c > 0 and we can easily see that, for c > 0, the integral
Z ∞
c dt,
0

is divergent and so the integral Z ∞


f (t) dt,
0
is also divergent by the DCT.

Solution to activity 4.2


For (a), we are given that the integral
Z ∞
f (x) dx,
1

is convergent and so, making the substitution x = y 2 so that dx = 2ydy, we see that the
integral Z ∞ Z ∞
f (x) dx = 2yf (y 2 ) dy,
1 1

123
4. Improper integrals

is convergent too. Indeed, as we have

2yf (y 2 ) ≥ f (y 2 ),

for y ≥ 1, this allows us to infer that the integral


Z ∞
f (y 2 ) dy,
1

is also convergent by the DCT.

4 For (b), we can take, for t ≥ 1, the non-negative continuous function f (t) = 1/t so that
we have Z ∞ Z ∞
2 dx
f (x ) dx = ,
1 1 x2
which is convergent, but
∞ ∞
dy
Z Z
f (y) dy = ,
1 1 y
is divergent.

Solution to activity 4.3


(a) This is an improper integral of the first kind, i.e. the problem location is at infinity.
The integrand is

x1/3 + 5 x1/3 1 + 5/x1/3 1 + 5/x1/3


   
1
f (x) = 2 = 2 = 5/3 ,
x − 6x + 10 x 1 − (6/x) + (10/x2 ) x 1 − (6/x) + (10/x2 )

and so, for large x, its dominant behaviour is given by


1
g(x) = .
x5/3
This means that we have
f (x) x1/3 + 5 1 + 5/x1/3
= 2 x5/3 = → 1,
g(x) x − 6x + 10 1 − (6/x) + (10/x2 )

as x → ∞ and, as the integral


∞ ∞
dx
Z Z
g(x) dx = ,
1 1 x5/3

is convergent, we see that the given integral is also convergent by the LCT.

(b) This is an improper integral of the first kind, i.e. the problem location is at infinity.
The integrand is
π
f (x) = − tan−1 x,
2
and we know from Activity 2.14 that
π 
lim x − tan−1 x = 1.
x→∞ 2

124
4.2. Solutions to activities

This means that if we take g(x) = 1/x, we have


π
f (x) − tan−1 x π 
lim = lim 2
= lim x − tan−1 x = 1,
x→∞ g(x) x→∞ 1/x x→∞ 2

so that the divergence of the integral


Z ∞ ∞
dx
Z
g(x) dx = ,
1 1 x

implies that the given integral is also divergent by the LCT.


4
Solution to activity 4.4
If we substitute y = ln x so that dy = x−1 dx, we have
Z ∞ Z ∞
dx dy
r
= r
,
2 x(ln x) ln 2 y

and we know that this is convergent if r > 1 and divergent if r ≤ 1.

Solution to activity 4.5


The given integral is an improper integral of the third kind as it can be written as
∞ 0 ∞
x x x
Z Z Z
dx = dx + dx,
−∞ 1 + x2 −∞ 1 + x2 0 1 + x2

which is the sum of two improper integrals of the first kind. But, looking at the second
of the integrals on the right-hand side, we can see that
b b
ln(1 + b2 ) − ln 2

x 1
Z
2
dx = ln(1 + x ) = → ∞,
0 1 + x2 2 1 2

as b → ∞ and so the improper integral


Z ∞
x
dx,
0 1 + x2

is divergent. As such, the given integral is also divergent.3

Solution to activity 4.6


Here the integrand is
1
f (x) = √ ,
x4 + x3 − x2
3
In particular, observe that even though
b b
ln(1 + b2 ) − ln(1 + b2 )

x 1
Z
2
dx = ln(1 + x ) = = 0,
−b 1 + x2 2 −b 2

which is zero as b → ∞, this tells us nothing about the convergence of the given integral as this is not
how we treat improper integrals of the third kind!

125
4. Improper integrals

and to see how this behaves, we note that


√ ! √ p
1 x4 + x3 + x2 x4 + x3 + x2 1 + 1/x + 1
f (x) = √ √ = 3
= ,
x 4 + x3 − x2 x4 + x3 + x2 x x

so that, for large x, the dominant behaviour is given by


1
g(x) = .
x
This means that we have
4
p
f (x) 1 + 1/x + 1 p
= x = 1 + 1/x + 1 → 2,
g(x) x
as x → ∞ and, as the integral
∞ ∞
dx
Z Z
g(x) dx = ,
1 1 x
is divergent, we see that the given integral is also divergent by the LCT.

Solution to activity 4.7



If we substitute u = t so that t = u2 and dt = 2udu, we have
Z ∞ √
Z ∞
8 − t
t e dt = 2u17 e−u du,
0 0

and, if we take g(u) = e−u/2 , this gives us


f (u) 2u17 e−u
= −u/2 = 2u17 e−u/2 → 0,
g(u) e
as u → ∞. Thus, as the integral
Z ∞ Z ∞
g(u) du = e−u/2 du,
0 0

is convergent, the given integral is also convergent by LCT.

Solution to activity 4.8


(a) This is an improper integral of the second kind and the problem here is that the
integrand
1
f (t) = ,
sin t
has an asymptote at t = 0. But, we know from Example 2.22 that
sin t
lim = 1,
t→0 t

and so, if we take g(t) = 1/t, we have


 −1
f (t) t sin t
lim+ = lim+ = lim+ = 1−1 = 1.
t→0 g(t) t→0 sin t t→0 t

126
4.2. Solutions to activities

As such, the divergence of the integral


Z 1 1
dt
Z
g(t) dt = ,
0 0 t
implies that the given integral is also divergent by LCT.

(b) This is an improper integral of the second kind and the problem here is that the
integrand
f (t) = − ln t,
has an asymptote at t = 0. Here, it is easiest to use integration by parts to see that
Z 1  1
4
(− ln t) dt = − (t ln t − t) = −(ln 1 − 1) + (a ln a − a) = a ln a − a + 1,
a a

so that we have Z 1  
(− ln t) dt = lim+ a ln a − a + 1 .
0 a→0

Of course, we can now see that


ln a
lim+ a ln a = lim+ ,
a→0 a→0 1/a

gives us a quotient where the numerator and the denominator both tend to infinity [in
magnitude] and so, using L’Hôpital’s rule, we have

ln a 1/a
lim+ = lim+ = lim −a = 0,
a→0 1/a a→0 −1/a2 a→0+

and so, in conclusion, we have


Z 1  
(− ln t) dt = lim+ a ln a − a + 1 = 0 − 0 + 1 = 1,
0 a→0

which is finite, i.e. the given integral is convergent.

(c) If we write the integrand as


1 1
f (t) = = 1/3 ,
t1/3 −t4/3 t (1 − t)

then we see that it has asymptotes at t = 0 and t = 1. This means that we have an
improper integral of the third kind which we can deal with by writing it as
Z 1 Z 1/2 Z 1
f (t) dt = f (t) dt + f (t) dt,
0 0 1/2

which is the sum of two improper integrals of the second kind. But, looking at the
second of the integrals on the right-hand side, we can see that the dominant behaviour
of
1
f (t) = 1/3 ,
t (1 − t)

127
4. Improper integrals

for t close to one, is given by


1
g(t) = .
1−t
This means that we have
f (t) 1 1
= 1/3 (1 − t) = 1/3 → 1,
g(t) t (1 − t) t

as t → 1− and, as the integral


1 1
dt
Z Z
g(t) dt = ,
4 1/2 1/2 1−t

is divergent, we can see that the improper integral


Z 1
f (t) dt,
1/2

is also divergent by LCT. As such, the given integral is also divergent.

(d) If we look at the integrand


1 − e−u
f (u) =
,
u
we see that, due to a potential problem at u = 0 and the infinite upper limit of
integration, we have an improper integral of the third kind. We can deal with this by
writing it as Z ∞ Z Z 1 ∞
f (u) du = f (u) du + f (u) du,
0 0 1
which is the sum of an improper integral of the first kind and an improper integral of
the second kind. But, looking at the second of these, we can see that the dominant
behaviour of f (u) is given by
1
g(u) = ,
u
for large u. This means that we have

f (u) 1 − e−u
= u = 1 − e−u = 1,
g(u) u

as u → ∞ and, as the integral


∞ ∞
du
Z Z
g(u) du = ,
1 1 u
is divergent, we can see that the improper integral
Z ∞
f (u) du,
1

is also divergent by LCT. As such, the given integral is also divergent.

128
4.2. Exercises

Solution to activity 4.9


With f (t) = sin t, we have f 0 (t) = cos t and f 00 (t) = − sin t so that Taylor’s theorem
gives us, for some 0 < c < t,
t2 00 t2
f (t) = f (0) + tf 0 (0) + f (c) =⇒ sin t = t − sin c.
2! 2
So, given that 0 < t < π, one has 0 < c < π and so sin c > 0 which means that sin t < t,
i.e.
sin t
0< < 1,
t
for 0 < t < π. Of course, for t ≥ π we have t > 1 and so we also have 4
| sin t|
0< < 1,
t
for t > π. Combining these two results then gives us
| sin t|
0< < 1,
t
for all t > 0.

For x > 0, we can now see that, if we use the preceding result, we have
Z ∞ Z ∞ Z ∞  −xt ∞
sin t −xt
sin t −xt −xt e 1
e dt ≤ e dt ≤ e dt = = ,

0 t 0
t 0 −x t=0 x
as required. In particular, if we let x → ∞, 1/x → 0 and the Sandwich theorem then
assures us that Z ∞
sin t −xt
lim e dt = 0.
x→∞ 0 t
On the other hand, Z ∞   Z ∞
sin t −xt
lim e dt = 0 dt = 0,
0 x→∞ t 0
and so, as both sides of the equality are zero, we have verified it.

Exercises
Exercise 4.1
Use L’Hôpital’s rule to determine the following limits.
(ln x)2
 
x ln x 1 1
(i) lim , (ii) lim √ , (iii) lim − .
x→∞ ex x→∞ x x→0+ ln(1 + x) x
[Hint: You will need to use a common denominator in (iii).]

Hence determine whether the following integrals converge.


Z ∞ Z ∞ Z 1
(ln x)2

−2x 1 1
(a) x(ln x) e dx, (b) dx, (c) − dx.
1 1 x2 0 ln(1 + x) x

129
4. Improper integrals

Exercise 4.2
Use L’Hôpital’s rule to show that
(1 + h)2/3 − 1 2
lim+ = ,
h→0 h 3
and hence deduce whether the integral
Z ∞
(1 + x6 )2/3 − x4 dx,
 
1

converges. [Hint: Let h = x−6 .]


4
Exercise 4.3
Consider the integral

1
Z
√ dx,
1 x2a + xb − xa
where a and b are real numbers.

(i) Show that, if 2a > b, then the integral converges if and only if b > a + 1.
(ii) If 2a ≤ b, find the values of b for which the integral converges.
(iii) Sketch the region in the ab-plane containing the values of a and b for which the
integral converges.

Solutions to exercises
Solution to exercise 4.1
As instructed, we use L’Hôpital’s rule to determine the three limits provided, of course,
that its use is justified.

For (i), we notice that the numerator and denominator both tend to infinity as x → ∞
and so we can use L’Hôpital’s rule to see that
x ln x (1)(ln x) + (x)(1/x) 1 + ln x
lim = lim = lim ,
x→∞ ex x→∞ ex x→∞ ex
and then, as the numerator and denominator still both tend to infinity as x → ∞, we
can use L’Hôpital’s rule again to see that
1 + ln x 1/x 1
lim x
= lim x = lim = 0,
x→∞ e x→∞ e x→∞ x ex

is the answer.

For (ii), we notice that the numerator and denominator both tend to infinity as x → ∞
and so we can use L’Hôpital’s rule to see that
(ln x)2 2(ln x)/x 4 ln x
lim √ = lim √ = lim √ ,
x→∞ x x→∞ 1/(2 x) x→∞ x

130
4.2. Solutions to exercises

and then, as the numerator and denominator still both tend to infinity as x → ∞, we
can use L’Hôpital’s rule again to see that
4 ln x 4/x 8
lim √ = lim √ = lim √ = 0,
x→∞ x x→∞ 1/(2 x) x→∞ x
is the answer.

For (iii), we follow the hint and write

x − ln(1 + x)
 
1 1
lim+ − = lim+ ,
x→0 ln(1 + x) x x→0 x ln(1 + x) 4
and we notice that the numerator and denominator both tend to zero as x → 0+ and so
we can use L’Hôpital’s rule to see that
1
x − ln(1 + x) 1 − 1+x x
lim = lim+ x = lim+ ,
x→0+ x ln(1 + x) x→0 ln(1 + x) + x→0 (1 + x) ln(1 + x) + x
1+x

and then, as the numerator and denominator still both tend to zero as x → 0+ , we can
use L’Hôpital’s rule again to see that
x 1 1 1
lim+ = lim+ 1+x = lim+ = ,
x→0 (1 + x) ln(1 + x) + x x→0 ln(1 + x) + 1+x
+1 x→0 ln(1 + x) + 2 2

is the answer.

We can then determine whether any of the given integrals are convergent.

For (a), we have an improper integral of the first kind where, for large x, the behaviour
of the integrand
f (x) = x(ln x) e−2x ,
can be captured by g(x) = e−x . This means that we have

f (x) x(ln x) e−2x x ln x


lim = lim −x
= lim = 0,
x→∞ g(x) x→∞ e x→∞ ex

using (i) and, as the integral


Z ∞ Z ∞
g(x) dx = e−x dx,
1 1

is convergent, this means that the given integral is convergent too by the LCT.

For (b), we have an improper integral of the first kind where, for large x, the behaviour
of the integrand
(ln x)2
f (x) = ,
x2
can be captured by g(x) = 1/x3/2 since, using (ii), we see that this gives us

f (x) (ln x)2 /x2 (ln x)2


lim = lim = lim √ = 0.
x→∞ g(x) x→∞ 1/x3/2 x→∞ x

131
4. Improper integrals

Thus, as the integral


∞ ∞
dx
Z Z
g(x) dx = ,
1 1 x3/2
is convergent, we see that the given integral is convergent too by the LCT.

For (c), we have an improper integral of the second kind where the integrand
1 1
f (x) = − ,
ln(1 + x) x

is undefined at x = 0. However, using (iii), we have


4  
1 1 1
lim+ − = ,
x→0 ln(1 + x) x 2

and so, using this to ‘repair’ the integrand, we see that this integral is convergent.

Solution to exercise 4.2


In the given limit, we note that as the numerator and denominator both tend to zero as
h → 0+ , we can use L’Hôpital’s rule to see that
2
(1 + h)2/3 − 1 (1 + h)−1/3 2/3 2
lim+ = lim+ 3 = lim+ = ,
h→0 h h→0 1 h→0 (1 + h)1/3 3

as required. We then follow the hint and use the substitution h = x−6 so that
dh = −6x−7 dx to see that the given integral can be written as
Z ∞ Z 0 " 2/3 # 
 6 2/3 4
 1 1 dh
(1 + x ) − x dx = 1+ − 2/3 − 7/6
1 1 h h 6h
1
(h + 1)2/3 − 1
Z  
= dh.
0 6h11/6

This gives us an improper integral of the second kind where the integrand

(h + 1)2/3 − 1
f (h) = ,
6h11/6
is undefined at h = 0. However, using the limit that we have just found, we see that if
we take g(h) = h−5/6 , we have

(h + 1)2/3 − 1 (1 + h)2/3 − 1
  
f (h) 1 2 1
lim+ = lim+ 11/6 −5/6
= lim+ = = ,
h→0 g(h) h→0 6h h h→0 6h 6 3 9

and we see that the integral


Z 1 Z 1
g(h) dh = h−5/6 dh,
0 0

is convergent. That is, by the LCT, the given integral must be convergent too.

132
4.2. Solutions to exercises

Solution to exercise 4.3


As in Activity 4.6, the integrand can be rewritten as
√ ! √
1 1 x2a + xb + xa x2a + xb + xa
f (x) = √ =√ √ = .
x2a + xb − xa x2a + xb − xa x2a + xb + xa xb

This means that we have:

(i) Given that 2a > b, we can write the integrand as

xa p
 
f (x) = b
x
1 + 1/x 2a−b +1 , 4
so that, for large x, the dominant behaviour is given by

g(x) = 2xa−b .

This means that we have


 
a−b
p
x 1 + 1/x 2a−b +1 p
f (x) 1 + 1/x2a−b + 1
= = → 1,
g(x) 2xa−b 2
as x → ∞ and, as the integral
Z ∞ Z ∞
g(x) dx = 2xa−b dx,
1 1

is convergent if and only if b > a + 1, then so too is the given integral by the
LCT.

(ii) Given that 2a ≤ b, we can write the integrand as

xb/2 p
 
b−2a (b/2)−a
f (x) = b 1/x + 1 + 1/x ,
x
so that, for large x, the dominant behaviour is given by
1
g(x) = .
xb/2
This means that we have
 
−b/2
p (b/2)−a
x 1/xb−2a + 1 + 1/x
f (x) p
= = 1/xb−2a + 1 + 1/x(b/2)−a → 1,
g(x) x−b/2
as x → ∞ and, as the integral
Z ∞ ∞
dx
Z
g(x) dx = ,
1 1 xb/2
is convergent if and only if b/2 > 1, then so too is the given integral by the
LCT. Thus, when 2a ≤ b, the given integral converges if and only if b > 2.

133
4. Improper integrals

(iii) The required region is shown in Figure 4.5. Make sure that, if your picture
doesn’t match this one, you understand how you went wrong.

b
b = 2a

b=a+1
2
b=2
4 1

O a
−1 1
Figure 4.5: The region of convergence for part (iii) of Exercise 4.3. Here we get the region
to the right of the line b = 2a from part (i) and the region which is both on and to the
left of the line b = 2a from part (ii). [Note that ‘dashed lines’ indicate points that are not
in the region.]

134
Chapter 5
Double integrals

Essential reading

(For full publication details, see Chapter 1.)

+ Binmore and Davies (2002) Sections 11.1–11.4.


+ Ostaszewski (1991) Chapter 19. 5
Further reading

+ Adams and Essex (2010) Sections 14.1–14.4.


+ Wrede and Spiegel (2010) parts of Chapter 9.

Aims and objectives

The objectives of this chapter are:

to introduce double integrals and see how they represent certain volumes;
to develop techniques that will allow us to evaluate double integrals.
Specific learning outcomes can be found near the end of this chapter.

5.1 Double integrals


In Chapter 3 we defined two inter-related notions: the area under a curve y = f (x) and
the definite integral of f . We now adapt the story to look at volumes under the surface
z = f (x, y). Just as the area or integral is a limiting sum over chosen x values, so too
this volume and the corresponding notion of a double integral is a limiting double
summation over chosen x and y values.
In the case of a function f (x, y) of two variables, domains can be rather more
complicated. In the simplest situation the domain is a rectangle. But other regions arise
quite naturally, as we will see later in this chapter.
If the domain is a rectangle, a double integral can be reduced to two integrations, i.e.
one with respect to each of the two independent variables. This is an example of what is
known as repeated integration.
If the domain can be bounded by the graphs of two functions of x, say, then the double
integral can still be reduced to a repeated integral although now the y integration will

135
5. Double integrals

be between limits determined by x. Sometimes a region can be regarded as bounded in


two alternate ways, the second being by two graphs of functions of y. In this case, the
evaluation of a repeated integral may be simplified by what is called a ‘change in the
order’ of integration.
In other circumstances a change of variable may transform the region of integration into
one bounded by two more convenient graphs and, once again, the double integral is
reduced to a repeated integral.

5.1.1 Volumes over rectangular bases

In the upcoming definition, we will want to refer to the following example.

5 Example 5.1 The surface z = ex−y is shown in Figure 5.1(a) for 0 ≤ x ≤ 2 and
0 ≤ y ≤ 2. It is the graph of a function which is increasing with x and decreasing
with y. We want to ask: What is the volume, illustrated in Figure 5.1(b), that lies
under the surface in (a) and above the rectangle R = [0, 1] × [0, 1] in the z = 0
plane?

(a) (b)
Figure 5.1: (a) The surface z = ex−y for 0 ≤ x ≤ 2 and 0 ≤ y ≤ 2. (b) The volume that
lies under the surface in (a) and above the rectangle R = [0, 1] × [0, 1] in the z = 0 plane.

Our definition below will answer the question posed in Example 5.1 by generalising,
from area to volume, the approach we used when dealing with the Riemann integral in
Chapter 3. This will allow us to define what we will call a double integral. A subsequent
result, Fubini’s theorem, will then show us how to evaluate such double integrals by
using our standard single variable integration.

136
5.1. Double integrals

5.1.2 Defining double integrals in terms of volumes


We follow the approach that we used for the Riemann integral in Chapter 3 and
partition the intervals on the x and y-axes independently. For instance, suppose that P
is the partition:
0 = x0 < · · · < xi < xi+1 < · · · < xn = 1,

on the x-axis and that Q is the partition

0 = y0 < · · · < yj < yj+1 < · · · < ym = 1,

on the y-axis. We want to estimate the volume standing on the rectangle


R = [0, 1] × [0, 1] in the z = 0 plane (as far as Example 5.1 is concerned, this will
actually be an under-estimate) by forming a sum of terms like

Vij = hij (xi − xi−1 )(yj − yj−1 ), 5


where hij = f (xi , yj ). That is, this estimate of the volume is the sum of the volumes of
the rectangular prisms of height
hij = f (xi , yj ),

and rectangular bases Rij of area

(xi − xi−1 )(yj − yj−1 ).

Each rectangular base, Rij , is located in the z = 0 plane and the height of each prism is
given by the value of f at the point (xi , yj ) as illustrated in Figure 5.2(a). Indeed, with
reference to Example 5.1, we can see in Figure 5.2(b) that this prism lies under the
surface z = ex−y that we saw in Figure 5.1(a).
We can now define an intermediate estimate of the volume under the surface z = f (x, y)
over the rectangle R = [0, 1] × [0, 1] in the z = 0 plane to be
X
E(P, Q) = f (xi , yj )(xi − xi−1 )(yj − yj−1 ),
1≤i≤n,
1≤j≤m

in terms of the partitions P and Q. In particular, as with the definition of the Riemann
integral, we can define lower estimates and upper estimates of this volume by reference
to the numbers

mij = min{f (x, y) : xi−1 ≤ x ≤ xi and yj−1 ≤ y ≤ yj },

for the former, and

Mij = max{f (x, y) : xi−1 ≤ x ≤ xi and yj−1 ≤ y ≤ yj },

for the latter, so that we get lower estimates given by


X
L(P, Q) = mij (xi − xi−1 )(yj − yj−1 ),
1≤i≤n,
1≤j≤m

137
5. Double integrals

5
(a) (b)

Figure 5.2: (a) A rectangular prism. (b) With reference to Example 5.1, notice that this
rectangular prism lies under the surface z = ex−y that we saw in Figure 5.1(a).

and upper estimates given by


X
U (P, Q) = Mij (xi − xi−1 )(yj − yj−1 ),
1≤i≤n,
1≤j≤m

in terms of the partitions P and Q. In particular, if we find that

sup L(P, Q) = inf U (P, Q), (5.1)


P,Q P,Q

we say that f (x, y) is Riemann integrable and we take this common value to be the
value of the double integral
Z Z
f (x, y) dxdy.
R

In Chapter 6 we will discuss what it means for a function f (x, y) to be jointly


continuous in the two variables x and y in a way that extends our one-variable approach
from Chapter 2.1 And, for reasons similar to those discussed in Chapter 3 for the
one-variable case, if we do have joint continuity, then we will also have the equality in
(5.1), with the result that jointly continuous functions are Riemann integrable in the
sense defined above, just as we saw in the one-variable case.
But, for now, all of the integrands that we will see in this chapter will be assumed to be
Riemann integrable and we will briefly review this issue once we have more tools
available to us in Chapter 6.

1
We will see there that the definition ensures that f (x, y) is a good approximation to f (x0 , y 0 ) for all
nearby points (x0 , y 0 ), just as we have in the one-variable case.

138
5.1. Double integrals

5.1.3 Motivating Fubini’s theorem


To motivate the key result of this chapter, Fubini’s theorem, let’s imagine the preceding
calculation being done on a spreadsheet where cell (i, j), i.e. the one in row i and
column j, corresponds to the grid point (xi , yj ) arising from our partitions P and Q of
the x and y-axes and contains the value f (xi , yj ) which represents the height of the
surface at that grid point. Such a spreadsheet would look like the following where we
have added up the entries in each column to get the column sums in the last row and
added up the entries in each row to get the row sums in the last column.

i\j 1 2 ··· m Row sums


P
1 f (x1 , y1 ) f (x1 , y2 ) ... f (x1 , ym ) f (x1 , yj )
Pj
2 f (x2 , y1 ) f (x2 , y2 ) ... f (x2 , ym ) f (x2 , yj )
Pj
3 f (x3 , y1 ) f (x3 , y2 ) ... f (x3 , ym ) j f (x3 , yj )
.. .. .. .. .. ..
. . . . . P . 5
n f (x , y ) f (x , y ) ··· f (x , y ) j f (xn , yj )
P n 1 P n 2 P n m
Column sums i f (xi , y1 ) i f (xi , y2 ) ... i f (xi , ym )

Now, if the grid is such that the differences xi − xi−1 for 1 ≤ i ≤ n and yj − yj−1 for
1 ≤ j ≤ m are both given by some δ > 0, i.e. we are dealing with a grid which consists
of squares of side δ, then it should be clear that if we add up the column sums we get
m X
X n
2
E(P, Q) = δ f (xi , yj ),
j=1 i=1

whereas if we add up the row sums we get


n X
X m
E(P, Q) = δ 2 f (xi , yj ),
i=1 j=1

and these two expressions give us the same intermediate estimate for the volume based
on the partitions P and Q.
Indeed, more generally, in the case where the partitions P and Q do not give us a grid
which consists of squares of side δ, i.e. the grid now consists of rectangles which have an
area given by
δij = (xi − xi−1 )(yj − yj−1 ),
for 1 ≤ i ≤ n and 1 ≤ j ≤ m, we see that a similar spreadsheet argument should
convince us that if we were to add up the row sums, we would get
n
" m # n
" m #
X X X X
E(P, Q) = f (xi , yj )δij = f (xi , yj )(xi − xi−1 )(yj − yj−1 ) ,
i=1 j=1 i=1 j=1

for the intermediate estimate of the volume based on the partitions P and Q. Indeed,
rewriting this as
n
" m #
X X
E(P, Q) = (xi − xi−1 ) f (xi , yj )(yj − yj−1 ) ,
i=1 j=1

139
5. Double integrals

we see that the sum on the right-hand side begins to look like
X n Z 1
(xi − xi−1 ) f (xi , y) dy,
i=1 y=0

and so, if we wanted to write


Z 1 Z 1
F (xi ) = f (xi , y) dy we would want to let F (x) = f (x, y) dy,
y=0 y=0

where we can think of the latter as the ‘opposite’ of partial differentiation with respect
to y. That is, as defined, F (x) is what we get when we hold x fixed and integrate with
respect to y. Indeed, now it becomes plausible to say that
Xn Z 1 Z 1 Z 1 
E(P, Q) ' (xi − xi−1 )F (xi ) ' F (x) dx = f (x, y) dy dx,
5 i=1 x=0 x=0 y=0

provided that the grid created by the partitions P and Q gives us areas δij that are
suitably small.
Of course, adding up the column sums and running the same argument, we could also
get
m
" n # m
" n #
X X X X
E(P, Q) = f (xi , yj )δij = f (xi , yj )(xi − xi−1 )(yj − yj−1 ) ,
j=1 i=1 j=1 i=1

for the intermediate estimate of the volume based on the partitions P and Q. Indeed,
rewriting this as
m
" m #
X X
E(P, Q) = (yj − yj−1 ) f (xi , yj )(xi − xi−1 ) ,
j=1 j=1

we see that the sum on the right-hand side begins to look like
Xm Z 1
(yj − yj−1 ) f (x, yj ) dx,
j=1 x=0

and so, if we wanted to write


Z 1 Z 1
G(yj ) = f (x, yj ) dx we would want to let G(y) = f (x, y) dx,
x=0 x=0

where we can think of the latter as the ‘opposite’ of partial differentiation with respect
to x. That is, as defined, G(y) is what we get when we hold y fixed and integrate with
respect to x. Indeed, now it becomes plausible to say that
Xm Z 1 Z 1 Z 1 
E(P, Q) ' (yj − yj−1 )F (yj ) ' G(y) dy = f (x, y) dx dy,
j=1 y=0 y=0 x=0

provided that the grid created by the partitions P and Q gives us areas δij that are
suitably small.

140
5.1. Double integrals

In summary, this intuitive use of approximation is supposed to indicate our belief that
as the partitions P and Q get ‘finer’, the value of E(P, Q) should get closer to either
one of the integrals
Z 1 Z 1  Z 1 Z 1 
f (x, y) dx dy, or f (x, y) dy dx,
y=0 x=0 x=0 y=0

and as we shall now see, this is what Fubini’s theorem says. That is, under certain
circumstances, the volume represented by a double integral can be found by evaluating
either of the so-called repeated integrals that we have just found.

5.1.4 Fubini’s theorem


The argument just given suggests (but does not fully prove) the following very
important result.
Theorem 5.1 (Fubini’s theorem) If f (x, y) is an integrable (in particular, jointly
5
continuous)2 function defined over a rectangle R = [a, b] × [c, d], then
Z Z Z d Z b  Z b Z d 
f (x, y) dxdy = f (x, y) dx dy = f (x, y) dy dx.
R y=c x=a x=a y=c

That is, we may evaluate a double integral by using repeated integration, i.e. we may
integrate with respect to one variable (whilst keeping the other fixed) and then we can
integrate with respect to the other remaining variable. When we rewrite double
integrals in this way, we say that we have a repeated integral. Indeed, whilst we may
think of this theorem as reducing a double integral to two separate integrations, we will
also benefit from reversing these ideas for, as we will see later, a repeated integration
can sometimes be simplified by first viewing it as a double integral.

Example 5.2 Following on from Example 5.1, we can now see that the volume
that lies under the surface z = ex−y and above the rectangle R = [0, 1] × [0, 1] in the
z = 0 plane is given by the double integral
Z Z
ex−y dxdy,
R

and, using Fubini’s theorem, we can evaluate this by viewing it as a repeated


integral. Indeed, we find that:

Writing the given double integral as the repeated integral


Z Z Z 1 Z 1 
x−y x−y
e dxdy = e dx dy,
R y=0 x=0

we do the x-integral first (with y fixed) to find that


Z 1  1
e x−y
dx = e x−y
= e1−y − e−y ,
x=0 x=0
2
We will discuss what it means to say that a function f (x, y) is jointly continuous in Chapter 6.

141
5. Double integrals

and then, integrating this with respect to y, we have


1 1
e1−y e−y e0 e−1 e1 e0
Z     
1−y −y
(e − e ) dy = − = − − − ,
y=0 −1 −1 y=0 −1 −1 −1 −1

so that we find that Z Z


ex−y dxdy = e−1 −2 + e,
R
is the value of the given double integral.

Writing the given double integral as the repeated integral


Z Z Z 1 Z 1 
x−y x−y
e dxdy = e dy dx,
R x=0 y=0

5 we do the y-integral first (with x fixed) to find that


1 1
ex−y
Z 
x−y
e dy = = − ex−1 + ex ,
y=0 −1 y=0

and then, integrating this with respect to x, we have


Z 1  1    
x−1 x x−1 x 0 1 −1 0
(− e + e ) dx = − e +e = − e +e − − e +e ,
x=0 x=0

so that we again find that


Z Z
ex−y dxdy = e −2 + e−1 ,
R

is the value of the given double integral.


Indeed, it is interesting to note that, if we write ex−y = ex e−y , we can in fact
simplify this calculation by noting that
Z Z Z 1 Z 1  Z 1  Z 1 
x−y x −y x −y
e dxdy = e e dx dy = e dx e dy ,
R y=0 x=0 x=0 y=0

as the whole integral with respect to x is a constant as far as the integral with
respect to y is concerned. This then gives us
Z 1  Z 1   1  −y 1   
x −y x e 1 −1
e dx e dy = e = e −1 − e +1 ,
x=0 y=0 x=0 −1 y=0

which, if you multiply out the brackets, gives us the same answer as before.

In order to avoid excessive use of brackets, but still be able to identify the two dummy
variables on their respective integral signs, we will often write the repeated integral
Z d Z b  Z d Z b
f (x, y) dx dy as dy f (x, y) dx,
y=c x=a y=c x=a

142
5.1. Double integrals

and we will also write


Z b Z d  Z b Z d
f (x, y) dy dx as dx f (x, y) dy.
x=a y=c x=a y=c

Activity 5.1 Evaluate the following repeated integrals.


Z 2 Z 4 √
(i) dy (x2 y − x ey +1) dx.
y=1 x=0

1 1
2xy 3
Z Z
(ii) dy dx.
y=0 x=0 x2 y 2 + 1
1 1
x
Z Z
(iii) dx dy.
x=0 y=0 1 + x2 y 2
5
5.1.5 Volumes over other bases
So far, we have been considering double integrals which represent volumes over a
rectangular base. We now consider the case where the base is some non-rectangular
bounded region such as the one in the following example.

Example 5.3 What is the volume that lies under the surface z = ex−y and above
the region D which is given by the unit disk centred on the point (1, 1) in the z = 0
plane?

We saw this surface in Figure 5.1(a) and the required base, which can be represented
by the inequality
(x − 1)2 + (y − 1)2 ≤ 1,
and the volume we seek is illustrated in Figures 5.3(a) and (b) respectively.

It should come as no surprise that we will represent the volume in Example 5.3 by the
double integral Z Z
f (x, y) dxdy,
D
x−y
with f (x, y) = e but, in order to evaluate this, we first need to give it a proper
definition. We do this in two steps.
Definition 5.1 For any subset S of a rectangle R = [a, b] × [c, d] the function

1 if (x, y) ∈ S,
1S (x, y) =
0 otherwise,
is called the indicator function of S.

This notation reminds us that the indicator function of S takes the value 1 at the point
(x, y) if and only if that point lies in S and is zero otherwise. So, for the base in
Example 5.3, we have
1D (x, y) = 1 if and only if (x − 1)2 + (y − 1)2 ≤ 1.

143
5. Double integrals

1 D

O x
5 1 2
(a) (b)

Figure 5.3: (a) The region D discussed in Example 5.3 is the interior of this disk. (b) The
volume that lies under the surface z = ex−y and above the disk, D, in the z = 0 plane.

We can then use this to define a double integral over a non-rectangular region in terms
of a double integral which is over a rectangular region as follows.
Definition 5.2 For any subset S of a rectangle R = [a, b] × [c, d], we define the double
integral over S by reference to the double integral over the containing rectangle R by
setting
Z Z Z Z
f (x, y) dxdy = f (x, y)1S (x, y) dxdy,
S R

provided that the function f (x, y)1S (x, y) is Riemann integrable.

In the case we are considering in Example 5.3, we see that 1D is continuous at all points
(x, y) such that (x − 1)2 + (y − 1)2 < 1 and also at points (x, y) such that
(x − 1)2 + (y − 1)2 > 1. However, 1D is not continuous at points on its boundary, i.e. the
circle (x − 1)2 + (y − 1)2 = 1, but nevertheless, the function f (x, y)1D (x, y) is still
Riemann integrable.
Comment: The volume integral defined above may be interpreted as formalising a
limiting sum idea as follows. Find the upper and lower estimates as discussed in
Section 5.1.1, but now treat them as the sums we get when we omit any terms which
arise from a rectangular base [xi−1 , xi ] × [yj−1 , yj ] which lies wholly outside of the
region, say D, that we are considering.
Definition 5.2 now reduces volume integrals over general bounded regions to volume
integrals over rectangular regions and this immediately suggests that we can still apply
Fubini’s theorem to turn a double integral into a repeated integral. It will be instructive
to illustrate this in a simpler case where the region involved is simpler than the one that
we have been considering.

144
5.1. Double integrals

Example 5.4 Consider the rectangle R = [0, 1] × [0, 1] and let T denote the
triangular region bounded by the lines y = 0, x = 1 and x = y as illustrated in
Figure 5.4. One mathematical description of T involves regarding x as an
independent variable ranging between 0 and 1 and then taking y to be a dependent
variable whose values are determined in terms of x so that T can be described as the
set of points
T = {(x, y) : 0 ≤ x ≤ 1 and 0 ≤ y ≤ x}.
This then allows us to apply Definition 5.2 to obtain
Z Z Z Z
x−y
e dxdy = ex−y 1T (x, y) dxdy,
T R

so that Fubini’s theorem gives us


Z 1
dx
Z 1
e x−y
1T (x, y) dy =
Z 1
dx
Z x
ex−y dy.
5
x=0 y=0 x=0 y=0

This last integral is explained by the fact that, if we use our definition of T above,
we can see that for any fixed x in the interval [0, 1] we have (x, y) ∈ T if and only if
0 ≤ y ≤ x.3
Having done this, we see that evaluating the y-integral (holding x fixed) we get
Z x  x−y x
x−y e e0 ex
e dy = = − = ex −1,
y=0 −1 y=0 −1 −1

so that we have
Z 1 Z x Z 1    1
x−y x x
dx e dy = e −1 dx = e −x = (e1 −1) − (e0 −0) = e −2,
x=0 y=0 x=0 x=0

as the final answer for this double integral.

Activity 5.2 How does the calculation in Example 5.4 work out if you treat y as
the independent variable?

Comment: You may find it helpful to imagine the spreadsheet being used here with
f (x, y) replaced by f (x, y)1T (x, y). In this case cell entries in the lower triangular area
with cells i, j where j ≤ i would be unchanged because 1T (x, y) = 1 whereas those in the
upper triangular area where j > i would all contain zero entries because 1T (x, y) = 0.
What we saw in Example 5.4 has a natural generalisation to other subsets, S, of the
rectangle R = [a, b] × [c, d] which are defined by referring to a pair of functions such as
g(x) and h(x), let’s call them bounding functions, that allow us to describe S as the set
of points
S = {(x, y) : a ≤ x ≤ b and g(x) ≤ y ≤ h(x)}.
3
Of course, (x, y) ∈ T if and only if 1T (x, y) = 1, and this is why we can omit any points (x, y) in the
‘inner’ integral where 1T (x, y) = 0.

145
5. Double integrals

x
=

x=1
y
T

O x
1
Figure 5.4: The triangular region, T , bounded by the lines y = 0, x = 1 and x = y from
Example 5.4.
5
That is, in cases like this, the boundary of S is given by the curves y = g(x) and
y = h(x) which means that we would have
Z Z Z Z Z x=b Z y=h(x)
f (x, y) dxdy = f (x, y)1S (x, y) dxdy = dx f (x, y) dy,
S R x=a y=g(x)

if we extend the argument given for the region T in Example 5.4 where we had g(x) = 0
and h(x) = x. Let’s see how this works if we return to the situation we described in
Example 5.3.

Example 5.5 Following on from Example 5.3, we now see that one way to
approach the evaluation of that volume is to describe the circular region illustrated
in Figure 5.3(a) by two bounding functions. To do this, we need to identify the two
functions whose graphs describe the boundary of the region.
Evidently, the disk is enclosed in the rectangle [0, 2] × [0, 2] and so, if we choose
0 ≤ x ≤ 2 to be the independent variable, we need to find the functions g(x) and
h(x) that bound the dependent variable, y. To do this, we solve the equation that
relates x and y on the boundary, i.e.
(x − 1)2 + (y − 1)2 = 1,
and see that, writing y in terms of x, we have
p
y = 1 ± 1 − (x − 1)2 .
This gives us the two bounding functions for, as illustrated in Figure 5.5(a), we have
p
g(x) = 1 − 1 − (x − 1)2 ,
which describes the lower semicircle so that, within the disk, y ≥ g(x) for 0 ≤ x ≤ 2,
and we have p
h(x) = 1 + 1 − (x − 1)2 ,
which describes the upper semicircle so that, within the disk, y ≤ h(x) for
0 ≤ x ≤ 2. In particular, this means that we now have
Z Z Z 2 Z h(x)
x−y
e dxdy = dx ex−y dy,
D x=0 y=g(x)

146
5.1. Double integrals

if we want to evaluate the double integral that represents the volume in Example 5.3.
Of course, we could also choose 0 ≤ y ≤ 2 to be the independent variable here and
then we would need to find the functions m(y) and n(y) that bound the dependent
variable, x. To do this, we solve the equation that relates x and y on the boundary,
i.e.
(x − 1)2 + (y − 1)2 = 1,
and see that, writing x in terms of y, we have
p
x = 1 ± 1 − (y − 1)2 .

This gives us the two bounding functions for, as illustrated in Figure 5.5(b), we have
p
m(y) = 1 − 1 − (y − 1)2 ,

which describes the left semicircle so that, within the disk, x ≥ m(y) for 0 ≤ y ≤ 2, 5
and we have p
n(y) = 1 + 1 − (y − 1)2 ,
which describes the right semicircle so that, within the disk, x ≤ n(y) for 0 ≤ y ≤ 2.
In particular, this means that we now have
Z Z Z 2 Z n(y)
x−y
e dxdy = dy ex−y dx,
D y=0 x=m(y)

if we want to evaluate the double integral that represents the volume in


Example 5.3.

y y

2 2

1 D 1 D

O x O x
1 2 1 2
(a) (b)

Figure 5.5: (a) For fixed 0 ≤ x ≤ 2, the region D discussed in Example 5.3 is bounded
by the curves y = g(x) and y = h(x) which correspond to the lower (solid) and upper
(dashed) semicircles respectively. (b) For fixed 0 ≤ y ≤ 2, the region D discussed in
Example 5.3 is bounded by the curves y = m(y) and y = n(y) which correspond to the
left (solid) and right (dashed) semicircles respectively.

147
5. Double integrals

Let’s look at some more examples of this.

Example 5.6 Evaluate the double integral


Z Z
xy dxdy,
Q

where Q is the quarter disk bounded by the curves x = 0, y = 0 and x2 + y 2 = 1 in


the positive quadrant.

The region Q is illustrated in Figure 5.6(a) and, if we take x to be the independent


variable, we find that

Z Z Z 1 Z 1−x2
xy dxdy = dx xy dy
Q 0 0
5 Z 1 Z √
1−x2
= x dx y dy
0 0

1  √1−x2
1
Z
= x y2 dx
0 2 0
1
x
Z
= (1 − x2 ) dx
0 2
1
1
Z
= (x − x3 ) dx
2 0
1
1 x2 x4

= −
2 2 4 0
 
1 1 1 1
= − = ,
2 2 4 8

is the answer.

Example 5.7 Evaluate the double integral


Z Z
x dxdy,
R

where the region R is the finite region bounded by the curves y = 2x and y = 3 − x2 .

The two curves are illustrated in Figure 5.6(b) and we find that the two curves
intersect when

3 − x2 = 2x =⇒ x2 + 2x − 3 = 0 =⇒ (x + 3)(x − 1) = 0 =⇒ x = −3, 1.

With this information, we should be able to determine the region of integration, R,


and taking x to be the independent variable, the problem at hand requires us to
identify the following.

148
5.1. Double integrals

y
y
1

y = 2x
−3 R x
O
1
y = 3 − x2
Q

O x
1
(a) (b) 5
Figure 5.6: (a) The region Q for Example 5.6. (b) The curves which determine the region
R in Example 5.7.

(i) What is the x range? It is from −3 ≤ x ≤ 1.

(ii) What is the lower bounding curve? It is y = 2x.

(iii) What is the upper bounding curve? It is y = 3 − x2 .


Once we have this information, we find that
Z Z Z 1 Z 3−x2
x dxdy = dx x dy
R −3 2x
Z 1  3−x2
= x y dx
−3 2x
Z 1
= x(3 − x2 − 2x) dx
−3
1
3x2 x4 2x3

= − −
2 4 3 −3

32
=− ,
3
is the answer.

Comment: In Example 5.7, the answer is negative. This is because volumes standing
above z = 0 are positive, whereas volumes standing below z = 0 are deemed negative.
Double integrals, in general, thus compute not the absolute value of the volume, but a
signed volume and, in Example 5.7, on the sub-range −3 ≤ x ≤ 0 the integrand is
negative. As such, the double integral here is not adding two absolute values of volume,
but two signed contributions, and the net result is negative since there is more negative
volume than positive volume in this case.

149
5. Double integrals

π/2 π/2
sin y
Z Z
Example 5.8 Evaluate the repeated integral dx dy.
0 x y
The given repeated integral requires integration with respect to y first (the ‘inner
integral’) and this looks hard. So we regard this repeated integration as an
evaluation of the double integral
sin y
Z Z
dxdy,
R y

over some region R, and we aim to see if repeated integration with respect to x first
is any easier. It will be!

Firstly, we need to identify the region of integration, R. To do this, we sketch, for


0 ≤ x ≤ π/2, the two curves y = x and y = π/2 as illustrated in Figure 5.7(a). Then,
5 we can identify the region R by noting that, for any chosen 0 ≤ x ≤ π/2, y runs
from y = x (the lower bounding curve) to y = π/2 (the upper bounding curve). That
is, the region R must be the upper triangle in Figure 5.7(a).

Secondly, we need to change the order of integration. To do this, we now treat y as


the independent variable and note that, for any chosen 0 ≤ y ≤ π/2, x runs from
x = 0 to x = y. Hence, we can rewrite the given repeated integral (which is the same
as the double integral above) as
π/2 y π/2 Z y
sin y sin y
Z Z Z
dy dx = dy dx
0 0 y 0 y 0
π/2  y
sin y
Z
= x dy
0 y 0
π/2
sin y
Z
= (y) dy
0 y
Z π/2
= sin y dy
0
 π/2
= − cos(y)
0

= 1,

and this is the answer.


1 1− 1−x2
x ey
Z Z
Example 5.9 Evaluate the repeated integral dx dy.
0 0 (y − 1)2
The given repeated integral requires integration with respect to y first (the ‘inner
integral’) and this looks hard. So, as in Example 5.8, we want to change the order of
integration and see if repeated integration with respect to x first is any easier.

150
5.1. Double integrals

y y

π y = π/2
2 1

R
x
=

x=1
y

O π x O x
2 1
(a) (b) 5
Figure 5.7: (a) The curves which determine the region R in Example 5.8. (b) The curves
which determine the region in Example 5.9.

For 0 ≤ x ≤ 1, we can sketch the line y = 0 and, since



y = 1 − 1 − x2 =⇒ x2 + (y − 1)2 = 1,

we see that the other curve is the lower right-hand quarter of the circle of radius 1
that is centred on the point (0, 1). This means that, for any chosen 0 ≤ x ≤ 1, y runs
from the x-axis to the quarter circle and so the region we are integrating over is the
‘curvilinear triangle’ in Figure 5.7(b).
So, to change the order of integration, we now treat y as the independent variable
and note that, for any chosen 0 ≤ y ≤ 1, x runs from
p
x2 + (y − 1)2 = 1 =⇒ x = 1 − (y − 1)2 (as x > 0),

to x = 1. Hence, we can rewrite the given repeated integral to find that


Z 1 Z 1 Z 1 Z 1
x ey ey
dy √ 2
dx = 2
dy √ x dx
y=0 x= 1−(y−1)2 (y − 1) y=0 (y − 1) x= 1−(y−1)2

1  2 1
ey x
Z
= dy
y=0 (y − 1)2 2 x=√2y−y2
1
ey 1 − (2y − y 2 )
Z  
= dy
0 (y − 1)2 2
1
1 ey
Z
= (y − 1)2 dy
2 0 (y − 1)2
1
1 1
Z
= ey dy = (e −1),
2 0 2
and this is the answer.

151
5. Double integrals

Activity 5.3 Evaluate the double integral


Z Z
x ey dxdy,
D

where D is the region bounded by the curves y = 1 + x2 and x = 1 in the positive


quadrant.

Activity 5.4 Consider the repeated integral


Z 1 Z 2−x
x
I= dx dy.
x=0 y=x y

Sketch the region of integration, and by changing the order of integration, evaluate I.
5 [Hint: You will need to split the region of integration into two parts.]

Activity 5.5 For x ≥ y ≥ 0, the function p(x, y) satisfies the conditions

∂p 2
= e−x ,
∂x
and p(y, y) = 0 for all y.
Z x
(i) Show that p(x, y) can be written as f (t) dt for suitable A and f .
t=A
Z x
(ii) For each fixed x, find p(x, y) dy.
y=0

5.2 Change of variable techniques


We now want to consider another way of evaluating double integrals that is, in essence,
the double integral equivalent of integration by substitution. We start our discussion by
focusing on a specific example.

Example 5.10 Consider the double integral


dxdy
Z Z
,
D xy

where D is the finite region bounded by the four lines


y = x, y = 2x, y = 1 − x and y = 2 − x,
as illustrated in Figure 5.8.
The region D in the xy-plane can be described more easily as a rectangle in the
uv-plane under the change of variable where
y
u=x+y and v= ,
x

152
5.2. Change of variable techniques

for x 6= 0. This is because a point (x, y) ∈ D is between the lines y = 1 − x and


y = 2 − x if

1 ≤ x + y ≤ 2 which is the same as requiring that 1 ≤ u ≤ 2,

whereas, for x 6= 0 (as is the case if the point is in D), it will be between the lines
y = x and y = 2x if
y
1≤ ≤ 2 which is the same as requiring that 1 ≤ v ≤ 2,
x
and this means that every point in D is in the rectangle in the uv-plane given by the
inequalities 1 ≤ u ≤ 2 and 1 ≤ v ≤ 2. Indeed, we can see that the lines where u is a
constant, i.e. lines of the form y = u − x, are parallel to two of the boundaries
whereas the other two boundaries are lines where v is a constant, i.e. they are lines
of the form y = vx. Notice that these latter lines have a gradient given by the value
of v and, in particular, to be in the region D we need to consider gradients (i.e.
5
values of v) which range from the smallest gradient when v = 1 to the largest
gradient when v = 2.
Thus, the region D in the xy-plane can be described by a rectangle, let’s call it ∆, in
the uv-plane. But, if we perform this change of variable, what happens to the
integral? There are, in fact, two questions we now have to ask:

(1) What happens to the integrand when we change from (x, y) to (u, v)?

(2) What happens to dxdy when we change to dudv?


The following discussion should enable us to answer these questions.

y
2x

2
y=

x
=
y

1
y

D
=
2

x
y
=
1

x

O x
1 2
Figure 5.8: The region D in Example 5.10.

Given a change of variable where u = u(x, y) and v = v(x, y), the chain rule tells us that
small changes in x and y will give rise to changes in u and v that are reasonably well

153
5. Double integrals

approximated by

∂u ∂u ∂v ∂v
du = dx + dy and dv = dx + dy.
∂x ∂y ∂x ∂y

Thus, in matrix form we have


∂u ∂u
    !
du d(u, v) dx d(u, v) ∂x ∂y
= where = ,
dv d(x, y) dy d(x, y) ∂v ∂v
∂x ∂y

is the coefficient matrix. This matrix equation describes the transformation from (x, y)
to (u, v) with reasonable accuracy, at least locally, where the changes we are considering
are small. It should not be surprising, therefore, that the determinant
 
∂(u, v) d(u, v)
5 ∂(x, y)
= det
d(x, y)
,

which is called the Jacobian of the transformation from (x, y) to (u, v), will play an
important role in our discussion.
We also need to assume that the transformation from (x, y) to (u, v) can be inverted.
That is, we need to be able to write x and y in terms of u and v to get the inverse
transformation x = x(u, v) and y = y(u, v). In fact, the Jacobian for this inverse
transformation, i.e.
∂x ∂x
  !
∂(x, y) d(x, y) d(x, y) ∂u ∂v
= det where = ∂y ∂y ,
∂(u, v) d(u, v) d(u, v)
∂u ∂v

is related to the Jacobian for the original transformation by the formula


 −1
∂(x, y) ∂(u, v)
= ,
∂(u, v) ∂(x, y)
provided, of course, that both of these Jacobians are non-zero.
If this all holds, then subject to some further provisos that we will outline after
Activity 5.6, we have the following result.
Theorem 5.2 (Change of variable) If a transformation from (x, y) to (u, v) maps the
region D to its image ∆, then

∂(x, y)
Z Z Z Z
f (x, y) dxdy = f (x(u, v), y(u, v))
dudv,
D ∆ ∂(u, v)

subject to the provisos discussed after Activity 5.6.

The reason why we need to use the absolute value of the Jacobian in this formula will
become clear when we look at the provisos. But, it will be instructive if we take a
moment to consider why the change from dxdy to dudv requires us to multiply the
latter by the [magnitude of the] Jacobian. Let’s see how this works in the case of a very
simple transformation.

154
5.2. Change of variable techniques

Example 5.11 Consider the transformation from (x, y) to (u, v) where

u = λx and v = µy,

for some non-zero constants λ and µ. In this case, we clearly have

du = λdx and dv = µdy,

which means that a rectangle of area dxdy in the xy-plane transforms, under this
change of variable, into a rectangle of area (λµ)−1 dudv in the uv-plane, i.e. the area
is scaled by a ‘transformation factor’ of (λµ)−1 .
But, on the other hand, we observe that the Jacobian of this transformation is
 
∂(u, v) λ 0
= det = λµ,
∂(x, y) 0 µ 5
and, indeed, we can easily find the inverse of this transformation by writing
u v
x= and y= ,
λ µ
to see that we also have the Jacobian
 
∂(x, y) 1/λ 0 1
= det = .
∂(u, v) 0 1/µ λµ
This verifies that  −1
∂(x, y) ∂(u, v)
= ,
∂(u, v) ∂(x, y)
as we noted above and, given our discussion of the area of a rectangle, it also
motivates replacing dxdy with something that involves a ‘transformation factor’
since we have
dudv ∂(x, y)
= dudv,
λµ ∂(u, v)
as we do in the change of variable formula given in Theorem 5.2.

Let’s now see how this helps us with the double integral that we were considering in
Example 5.10.

Example 5.12 Following on from Example 5.10, we can check that the given
change of variable is invertible by noticing that the equations
y
u=x+y and v= ,
x
allow us to write y = vx so that
u
u = x + vx = x(1 + v) =⇒ x = ,
1+v
as long as v 6= −1. Then, using y = vx again, we also have
uv
y= .
1+v

155
5. Double integrals

From this, we can see that the change of variable where u = u(x, y) and v = v(x, y)
that we started with can be inverted to give us x = x(u, v) and y = y(u, v) where
u uv
x(u, v) = and y(u, v) = ,
1+v 1+v
as long as v 6= −1.
This means that, to answer our first question in Example 5.10, we can use the
change of variable formula in Theorem 5.2 to change our current integrand, i.e.
1
f (x, y) = ,
xy
to the new integrand
(1 + v)2
 
u uv
5 f (x(u, v), y(u, v)) = f ,
1+v 1+v
=
u2 v
.

To answer the second question in Example 5.10, we need to find the Jacobian of the
transformation and it is easiest to do this using the original transformation as this
gives us
∂u ∂u
!  
d(u, v) ∂x ∂y 1 1
= ∂v ∂v = ,
d(x, y) −y/x2 1/x
∂x ∂y

so that the Jacobian is


   
∂(u, v) d(u, v) 1 1 1 y x+y
= det = det 2 = + 2 = .
∂(x, y) d(x, y) −y/x 1/x x x x2
We now use the formula  −1
∂(x, y) ∂(u, v)
= ,
∂(u, v) ∂(x, y)
to see that −1
x2

∂(x, y) x+y
= = ,
∂(u, v) x2 x+y
and this works as long as both of the Jacobians are non-zero, i.e. as long as
x + y 6= 0 and x 6= 0. And, using the original transformation to see that u = x + y
and the inverse of this transformation to see that
u
x= ,
1+v
we now have, for u 6= 0 and 1 + v 6= 0,
∂(x, y) u2 u
= 2
= ,
∂(u, v) u(1 + v) (1 + v)2

as we will want this Jacobian in terms of u and v only. (In Activity 5.6 you can
verify that this agrees with what we would have found by direct calculation.)
We can now use the change of variable formula from Theorem 5.2 to evaluate the
double integral that we were considering in Example 5.10. We know, from that
example, that the image of the region D is the rectangular region ∆ given by the

156
5.2. Change of variable techniques

inequalities 1 ≤ u ≤ 2 and 1 ≤ v ≤ 2 whereas, from this example, we know that the


integrand becomes
(1 + v)2
,
u2 v
and the magnitude of the Jacobian will be given by

∂(x, y) u u
∂(u, v) (1 + v)2 = (1 + v)2 ,
=

as this is certainly positive at all points (u, v) ∈ ∆. Thus, we have

(1 + v)2
Z Z   
dxdy u dudv
Z Z Z Z
= 2 2
dudv = .
D xy ∆ uv (1 + v) ∆ uv

Notice, in particular, that this change of variable has made the region of integration
much easier to deal with even though the integrand will require the same amount of 5
work. Consequently, we have
Z 2 Z 2  Z 2  Z 2 
dxdy du dv du
Z Z
= dv = ,
D xy v=1 u=1 uv v=1 v u=1 u

if we use the observation that we saw at the end of Example 5.2, and then we have
 2  2
dxdy
Z Z
= ln v ln u = (ln 2 − ln 1)(ln 2 − ln 1) = (ln 2)2 ,
D xy v=1 u=1

as the final answer.

Activity 5.6 Following on from Example 5.12, use the inverse transformation to
find the Jacobian
∂(x, y)
,
∂(u, v)
and verify that it agrees with what we found in the example when u 6= 0 and
1 + v 6= 0.
How are these latter conditions related to the conditions x + y 6= 0 and x 6= 0 that
we saw in the example?

We now take a moment to consider the provisos that apply to the use of our change of
variable formula for double integrals.

The provisos that apply to the use of the formula in Theorem 5.2

These provisos are discussed in detail in Sections 19.5–19.9 of Ostaszewski (1991), but
the highlights are as follows.

(1) We use the magnitude of the Jacobian in the formula to ensure that the region ∆
is always described by uv-coordinates where u and v are increasing regardless of
whether the change of variable itself implies that the order is reversed. (For

157
5. Double integrals

instance, if increasing x and y leads to decreasing u and v under the


transformation.) This means that if ∆ is, say, the rectangle 1 ≤ u ≤ 2 and
1 ≤ v ≤ 2, then the double integral
Z Z
· · · dudv,

is always to be computed as
Z 2 Z 2  Z 2 Z 2 
· · · dv du or · · · du dv,
u=1 v=1 v=1 u=1

depending on which of the repeated integrals we want to evaluate.


(2) The partial derivatives that are required for the calculation of the Jacobian must
be jointly continuous throughout the region of integration being considered.4
5 (3) The transformation given by u = u(x, y) and v = v(x, y) which takes us from xy to
uv-coordinates must be one-to-one over the region of integration and, in particular,
we need to know that it has an inverse over that region, i.e. that there are functions

x = x(u, v) and y = y(u, v),

that will take us back from uv to xy-coordinates. In particular, even though we


may not need to use this inverse transformation (or even find it), we do need to
know that it is there.
(4) The Jacobian must be of one sign only (i.e. either positive or negative) over the
region of integration. This assumption, in fact, also assures us that the
transformation will have a local inverse over the region of integration as required
by the previous proviso.
In particular, if provisos (3) or (4) are violated, then we can usually rectify the situation
by splitting the region of integration into sub-regions where they are fulfilled.

Activity 5.7 Use the transformation u = x + y and v = y/x, discussed in


Examples 5.10 and 5.12, to evaluate the double integral
y
Z Z
dxdy,
D x

where, as in Example 5.10, D is the region bounded by the lines y = x, y = 2x,


y = 1 − x and y = 2 − x.

Let’s now look at some more examples of how this all works.

3 x−2
dy
Z Z
Example 5.13 Evaluate the repeated integral dx .
x=2 y=0 (x + y)2 (x − y)

The repeated integral takes x as the independent variable and, for any chosen
2 ≤ x ≤ 3, we see that 0 ≤ y ≤ x − 2. This gives us the region of integration, D,
4
We will discuss joint continuity in Chapter 6.

158
5.2. Change of variable techniques

illustrated in Figure 5.9(a). We will evaluate the repeated integral by using the
change of variables given by the transformation
u=x+y and v = x − y.
Indeed, we see that adding these two equations together we get
u+v
u + v = 2x =⇒ x= ,
2
whereas subtracting these two equations gives us
u−v
u − v = 2y =⇒ y= ,
2
and so the inverse transformation is given by
u+v u−v
x= and y = .
2 2 5
In particular, the inverse transformation allows us to easily see that the inequality
u+v
2≤x≤3 =⇒ 2≤ ≤3 =⇒ 4 − u ≤ v ≤ 6 − v,
2
whereas the inequality
u−v u+v
0≤y ≤x−2 =⇒ 0≤ ≤ − 2 =⇒ v ≤ u and v ≥ 2,
2 2
and these inequalities determine the image, ∆, of the region of integration in the
uv-plane which is illustrated in Figure 5.9(b).
Of course, in order to perform our change of variables we also need the Jacobian of
the transformation which is given by
∂u ∂u
!  
∂(u, v) ∂x ∂y 1 1
= det ∂v ∂v = det = −2,
∂(x, y) 1 −1
∂x ∂y

which is always non-zero and gives us


 −1
∂(x, y) ∂(u, v) 1
= =− .
∂(u, v) ∂(x, y) 2
Then, taking the magnitude of this and applying the change of variable formula
from Theorem 5.10, we get
Z 3 Z x−2
dy dxdy dudv
Z Z Z Z
dx 2
= 2
= 2
,
x=2 y=0 (x + y) (x − y) D (x + y) (x − y) ∆ 2u v

if we use the transformation to see that the (x − y)(x + y)2 in the integrand is simply
uv 2 . It is now best to take v as the independent variable. To evaluate this new double
integral it is easiest to take v to be the independent variable so that, for any chosen
2 ≤ v ≤ 3, we see that v ≤ u ≤ 6 − v. Hence, we can rewrite the double integral as
Z 3 Z 6−v
dudv du
Z Z
2
= 2
dv,
∆ 2u v v=2 u=v 2u v

and this repeated integral is easy to evaluate as you will see in Activity 5.8.

159
5. Double integrals

y v

u
1 6 v=
v=
6−
4 v= u

2
x−
4−

x=3
3
u
y=
∆ v=2
2
D

O x O u
2 3 2 3 4 6
5 (a) (b)

Figure 5.9: (a) The region of integration, D, for the repeated integral in Example 5.13
and (b) its image, ∆, in the uv-plane under the transformation discussed in that example.

Activity 5.8 Complete the evaluation of the repeated integral in Example 5.13.
Why did we choose v to be the independent variable in this calculation?

Activity 5.9 Consider the transformation mapping a point (x, y) ∈ R2 with x > 0
to the point (u, v) given by
y
u = x2 + y 2 and v= .
x
Observe, in particular, that u > 0.

(i) For u > 0, find the inverse of this transformation.


∂(u, v)
(ii) Find the Jacobian of the given transformation.
∂(x, y)
(iii) Sketch the region D = {(x, y) : 0 < y ≤ x ≤ 1} and find its image ∆ under
the given transformation.

(iv) Use the given transformation to evaluate the double integral

y2
 
y x2 +y2
Z Z
e 1 + 2 dxdy.
D x x
[Hint: The answer to Activity 5.3 may be useful.]

Polar coordinates

If we take a point (x, y), then its polar coordinates are given by (r, θ) where r > 0 is the
distance of that point from the origin and 0 ≤ θ ≤ 2π is the angle between the x-axis

160
5.2. Change of variable techniques

and the line segment joining the point to the origin as illustrated in Figure 5.10. Indeed,
simple trigonometry dictates that the transformation that takes us from (r, θ) to (x, y)
is given by
x = r cos θ and y = r sin θ,
and the inverse transformation is given by
p y
r = x2 + y 2 and θ = tan−1 ,
x
for 0 ≤ θ < 2π. The Jacobian for this transformation is also very simple because we have
∂x ∂x
!  
∂(x, y) ∂r ∂θ cos θ −r sin θ
= det ∂y ∂y = det = r cos2 θ + r sin2 θ = r,
∂(r, θ) sin θ r cos θ
∂r ∂θ

if we use the trigonometric identity sin2 θ + cos2 θ = 1. As we shall see, this particular
transformation is very useful.

y 5

(x, y)
r

θ
O x

Figure 5.10: The polar coordinates, (r, θ), of the point (x, y).

Example 5.14 Use polar coordinates to evaluate the double integral


Z Z
1 2 2
e− 2 (x +y ) dxdy,
Q(R)

where Q(R), as illustrated in Figure 5.11(a), is the quarter disk of radius R centred
on the origin that lies in the positive quadrant of the xy-plane.

Every point (x, y) in Q(R), has polar coordinates (r, θ) where 0 ≤ r ≤ R and
0 ≤ θ ≤ π/2 and so this tells us how to rewrite the limits of integration under this
transformation. We also see that the integrand can be easily rewritten in polar
coordinates since
x2 + y 2 = r2 cos2 θ + r2 sin2 θ = r2 (cos2 θ + sin2 θ) = r2 ,
if we use the trigonometric identity sin2 θ + cos2 θ = 1. Consequently, using the
Jacobian that we found above, we have
Z π/2 Z R " 1 2 #R
π e− 2 r π
Z Z 
1 2 2 1 2 1 2
e− 2 (x +y ) dxdy = dθ e− 2 r r dr = = 1 − e− 2 R ,
Q(R) 0 0 2 −1 2
0

161
5. Double integrals

as the value of the given double integral.

y y

R R

S(R)
Q(R)

5 O x O x
R R
(a) (b)

Figure 5.11: (a) For Examples 5.14 and 5.15, Q(R) is the quarter disk of radius R centred
on the origin that lies in the positive quadrant of the xy-plane. (b) For Example 5.15,
S(R) is the square of side R that has one corner at the origin and also lies in the positive
xy-plane.

Activity 5.10 Use polar coordinates, as discussed above, to sketch the region of
integration for each of the following integrals in the xy-plane.
Z π/2 Z 1/ cos θ
(i) dθ f (r, θ) dr.
θ=0 r=0
Z 2√2 Z sin−1 (2/r)
(ii) dr f (r, θ)dθ.
r=2 θ=π/4

5.3 Improper double integrals


We now briefly consider an example where we have to evaluate an improper double
integral. In particular, we want to show that
Z ∞ r
− 12 x2 π
e dx = ,
0 2
by showing that 2 Z ∞
π − 12 x2
e dx = .
x=0 2
To do this, we could introduce a new dummy variable, y, so that we can write
Z ∞ 2 Z ∞  Z ∞ 
− 12 x2 − 12 y 2 − 12 x2
e dx = e dy e dx ,
x=0 y=0 x=0

162
5.3. Improper double integrals

which allows us to rewrite the problem, in terms of repeated integrals, as


Z ∞ 2 Z ∞ Z ∞  Z ∞Z ∞
− 21 x2 − 12 y 2 − 12 x2 1 2 2
e dx = e e dx dy = e− 2 (x +y ) dx dy.
x=0 y=0 x=0 y=0 x=0

So, ultimately, we need to show that


Z ∞Z ∞
1 2 +y 2 ) π
e− 2 (x dx dy = ,
y=0 x=0 2
and from this the desired result will follow. However, to evaluate this repeated integral,
we need to think of it as a double integral so that we can apply our change of variable
techniques and use polar coordinates. But, this move from a repeated integral to a
double integral requires Fubini’s theorem and this, as stated in Theorem 5.1, is only
valid for finite rectangles rather than the ‘infinite rectangle’ (i.e. the entire positive
quadrant of the xy-plane) that we are considering here. This means that, to avoid the
worry that Fubini’s theorem is not applicable here, we have to approach this problem in
a slightly different way and try to progress by considering finite domains of integration 5
as we do in the next example.

Example 5.15 Suppose that, as illustrated in Figure 5.11(a), Q(R) is the quarter
disk of radius R centred on the origin that lies in the positive xy-plane and that, as
illustrated in Figure 5.11(b), S(R) is the square of side R that has one corner at the
origin and also lies in the positive xy-plane.
1 2 2
Now, consider the case where the integrand is the function e− 2 (x +y ) . As this is
always positive, the volume enclosed by this function will increase as the area of the
base we are considering increases. That is, with a base of Q(R) the volume must be
less than the volume enclosed by this function with a base of S(R) because, as we
can see in Figure 5.12(a), we have Q(R) ⊆ S(R). That is, in terms of double
integrals we can write the relationship between these two volumes as
Z Z Z Z
− 12 (x2 +y 2 ) 1 2 2
e dxdy ≤ e− 2 (x +y ) dxdy.
Q(R) S(R)

But, on the other hand, the volume enclosed by this function with a base√of S(R)
must be less than the volume enclosed by this function with √ a base of Q( 2R)
because, as we can see in Figure 5.12(b), we have S(R) ⊆ Q( 2R). That is, in terms
of double integrals we can write the relationship between these two volumes as
Z Z Z Z
− 21 (x2 +y 2 ) 1 2 2
e dxdy ≤ √
e− 2 (x +y ) dxdy.
S(R) Q( 2R)

Consequently, if we consider the volumes over these three different bases, we have
Z Z Z Z Z Z
− 21 (x2 +y 2 ) − 12 (x2 +y 2 ) 1 2 2
e dxdy ≤ e dxdy ≤ √
e− 2 (x +y ) dxdy,
Q(R) S(R) Q( 2R)

and, using the result √


from Example 5.14 on the double integrals that involve the
regions Q(R) and Q( 2R), we see that
π − 12 R2
 Z Z
− 12 (x2 +y 2 ) π −R2

1−e ≤ e dxdy ≤ 1−e .
2 S(R) 2

163
5. Double integrals

Then, looking at the remaining double integral, we see that Fubini’s theorem
guarantees that
Z Z Z R Z R
− 12 (x2 +y 2 ) 1 2 2
e dxdy = e− 2 (x +y ) dx dy,
S(R) y=0 x=0

as S(R) is a finite rectangle and this repeated integral can now be written as
Z R Z R Z R Z R Z R 2
− 21 (x2 +y 2 ) − 12 y 2 − 12 x2 − 21 x2
e dx dy = e e dx dy = e dx ,
y=0 x=0 y=0 x=0 x=0

as x and y are just dummy variables. Thus, we have


 Z R 2
π − 21 R2 − 12 x2 π 2

1−e ≤ e dx ≤ 1 − e−R ,
2 2
5 x=0

so that, taking the limit as R → ∞, the Sandwich theorem gives us


Z ∞ 2
− 12 x2 π
e dx = ,
x=0 2

as, clearly, we have


π − 12 R2
 π π −R2
 π
lim 1−e = and lim 1−e = .
R→∞ 2 2 R→∞ 2 2
Consequently, bearing in mind that the integral we are interested in must give us a
positive number because its integrand is positive over the interval [0, ∞), we must
have Z ∞ r
− 12 x2 π
e dx = ,
x=0 2
as expected from our earlier argument.

y y

2R

11111111
00000000 111111111
000000000
000000000
111111111
00000000
11111111
R

00000000
11111111
R

000000000
111111111
00000000
11111111 000000000
111111111
000000000
111111111
00000000
11111111
00000000
11111111 000000000
111111111
00000000
11111111 000000000
111111111
000000000
111111111
00000000
11111111
00000000
11111111
O
R
x
000000000
111111111
O
R

2R
x

(a) (b)
Figure 5.12: Following on from Figure 5.11 for Example 5.15. (a) The hatched region,
Q(R), is contained in the√
shaded region, S(R). (b) The hatched region, S(R), is contained
in the shaded region, Q( 2R).

164
5.3. Learning outcomes

∞ ∞
dy
Z Z
Activity 5.11 Consider the repeated integral I = dx .
x=1 y=0 (x2 + y 2 )2

(i) For fixed x ≥ 1, make the substitution z = y/x in the y-integral and hence
show that I is convergent.

(ii) Convert the original integral to polar coordinates with θ as the dependent
variable. Evaluate the θ-integral and hence determine whether the repeated
integral converges.

(iii) Evaluate I.

Learning outcomes
5
At the end of this chapter and having completed the relevant reading and activities, you
should be able to:

state Fubini’s theorem on repeated integration;

find the volume under a surface over a rectangular base;

find the volume under a surface over a general base using bounding curves;

evaluate a repeated integral by changing the order of integration;

find the Jacobian of a transformation and evaluate a double integral by using a


change of variable.

Solutions to activities
Solution to activity 5.1
For (i), we do the x-integral first (with y fixed) to get
4 4
x3 y 2 3/2 y

64 16 y
Z
2 1/2 y
(x y − x e +1) dx = − x e +x = y− e +4,
x=0 3 3 x=0 3 3

and then, integrating with respect to y, we have


2    2
64 16 y 32 2 16 y
Z
y− e +4 dy = y − e +4y
y=1 3 3 3 3 y=1
   
128 16 2 32 16 1
= − e +8 − − e +4
3 3 3 3
16
= 36 − e(e −1).
3

165
5. Double integrals

That is, we have found that


Z 2 Z 4
√ 16
dy (x2 y − x ey +1) dx = 36 − e(e −1),
y=1 x=0 3

is the required value.

For (ii), we do the x-integral first (with y fixed) and, after making the substitution
u = x2 y 2 + 1 so that du = 2xy 2 dx, we get
1 y 2 +1 y2 +1
2xy 3

y
Z Z
dx = du = y ln u = y ln(y 2 + 1),
x=0 x2 y 2 + 1 u=1 u u=1

and then, integrating with respect to y, after making the substitution v = y 2 + 1 so that
dv = 2y dy, we get
5 Z 1 2
1 2
  
1 1 1
Z
2
y ln(y +1) dy = ln v dv = v ln v−v = [2 ln 2−2]−[ln 1−1] = ln 2− .
y=0 2 v=1 2 v=1 2 2

That is, we have found that


Z 1 1
2xy 3 1
Z
dy 2 2
dx = ln 2 − ,
y=0 x=0 x y +1 2

is the required value.

For (iii), we do the y-integral first (with x fixed) and, after making the substitution
u = xy so that du = x dy, we get
Z 1 Z x  x
x du −1
2 2
dy = 2
= tan u = tan−1 x,
y=0 1 + x y u=0 1 + u u=0

and then, integrating with respect to x, we need to use integration by parts on


‘1 · tan−1 x’, to get
Z 1  1 Z 1
−1 −1 x
tan x dx = x tan x − 2
dx
x=0 x=0 x=0 1 + x
 1
−1 1 2
= tan 1 − ln(1 + x )
2 x=0

π 1
= − ln 2.
4 2
That is, we have found that
Z 1 1
x π 1
Z
dx dy = − ln 2,
x=0 y=0 1 + x2 y 2 4 2

is the required value.

166
5.3. Solutions to activities

Solution to activity 5.2


Following on from Example 5.4 where T is the set of points

T = {(x, y) : 0 ≤ x ≤ 1 and 0 ≤ y ≤ x},

we can apply Definition 5.2 to obtain


Z Z Z Z
x−y
e dxdy = ex−y 1T (x, y) dxdy,
T R

and then, Fubini’s theorem allows us to write this as


Z 1 Z 1 Z 1 Z 1
x−y
dy e 1T (x, y) dx = dy ex−y dx,
y=0 x=0 y=0 x=y

where this last integral is most easily explained if we look at Figure 5.4 and see that for
any fixed y in the interval [0, 1], we have (x, y) ∈ T if and only if y ≤ x ≤ 1. 5
Having done this, we see that evaluating the x-integral (holding y fixed) we get
Z 1  1
x−y x−y
e dx = e = e1−y − e0 = e1−y −1,
x=y x=y

so that we have
Z 1 Z 1 1    1−y 1  0   1 
e e e
Z
x−y 1−y
dy e dx = e −1 dy = −y = −1 − −0 ,
y=0 x=y y=0 −1 y=0 −1 −1

and so, once again, e −2 is the final answer for this double integral.

Solution to activity 5.3


The region D, as illustrated in Figure 5.13, is bounded by the curves y = 1 + x2 and
x = 1. Indeed, looking at this illustration, we can see that it is easiest if we think of D
in terms of a fixed 0 ≤ x ≤ 1 so that 0 ≤ y ≤ 1 + x2 . With this in mind, we can then
write the given double integral as a repeated integral to get
Z Z Z 1 Z x2 +1 !
x ey dxdy = x ey dy dx
D x=0 y=0

Z 1  x2 +1
y
= x e dx
x=0 y=0
Z 1  
2 +1
= x ex −x dx
0
" 2
#1
ex +1 x2
= −
2 2
x=0

1 2 
= e − e −1 ,
2
as the final answer.

167
5. Double integrals

y = x2 + 1

D
1
O x
1

5 Figure 5.13: The curves which determine the region D in Activity 5.3.

Solution to activity 5.4


It’s not that hard to do the repeated integral in this question as it stands, but as we
shall see it is easier to do as we are told and change the order of integration. To do this,
we sketch the region of integration, let’s call it R, and split it into the two regions R1
and R2 as illustrated in Figure 5.14. We then have
Z 1 Z 2−x
x x x x
Z Z Z Z Z Z
I= dx dy = dxdy = dxdy + dxdy.
x=0 y=x y R y R1 y R2 y

Now, before, we were supposed to integrate with respect to y first and then integrate
with respect to x and so we now want to change the order of integration and integrate
with respect to x first and then integrate with respect to y. That means, for the double
integral over R1 , we need to fix 0 ≤ y ≤ 1 and take 0 ≤ x ≤ y so that
Z 1 Z y  Z 1  2 y Z 1  2 1
x x x y y 1
Z Z
dxdy = dx dy = dy = dy = = ,
R1 y y=0 x=0 y y=0 2y x=0 y=0 2 4 y=0 4

whereas for the double integral over R2 , we need to fix 1 ≤ y ≤ 2 and take
0 ≤ x ≤ 2 − y so that
Z 2 Z 2−y Z 2  2 2−y Z 2
(2 − y)2

x x x
Z Z
dxdy = dx dy = dy = dy
R2 y y=1 x=0 y y=1 2y x=0 y=1 2y
Z 2 Z 2  2
4 − 4y + y 2 y2
 
2 y
= dy = −2+ dy = 2 ln y − 2y +
y=1 2y y=1 y 2 4 y=1
 
1 5
= (2 ln 2 − 4 + 1) − 2 ln 1 − 2 + = 2 ln 2 − .
4 4
Consequently, we find that
   
x x 1 5
Z Z Z Z
I= dxdy + dxdy = + 2 ln 2 − = 2 ln 2 − 1,
R1 y R2 y 4 4
is the final answer.

168
5.3. Solutions to activities

2
y=x

R2
1
R1
y =2−x
O x
1 2
Figure 5.14: The curves which determine the region in Activity 5.4 and its decomposition
into the regions R1 and R2 which allow us to change the order of integration.
5
Solution to activity 5.5
For (i), as instructed, we want to write p(x, y) as a definite integral of the form
Z x
p(x, y) = f (t) dt,
t=A

for some suitable choice of A and f . Indeed, notice that the question leads us to expect
that p is a function of both x and y and, from the form we are given, the only place
that y can occur is in A. With this in mind, when we partially differentiate p(x, y) with
respect to x, we assume that A only contains instances of y so that we can treat it as a
constant5 and, if this is the case, we can use the FTC to see that
Z x
∂p ∂
= f (t) dt = f (x),
∂x ∂x t=A

which is similar to what we saw in Activity 3.10. However, we are told that

∂p 2
= e−x ,
∂x
and so we have Z x
−x2 2
f (x) = e =⇒ p(x, y) = e−t dt,
t=A

for a suitable choice of A which, as assumed above, should involve instances of y only.
Now, we are also told that p(y, y) = 0 for all y, which means that
Z y
2
p(y, y) = e−t dt = 0,
A

2
and, since this is the area between the positive function e−t , the t-axis and the vertical
lines t = A and t = y, this means that we must take A = y for this to hold. (Of course,
5
As in Activity 3.10, if this assumption turns out to give us nonsense, then we would have to try
something more complicated.

169
5. Double integrals

this is consistent with our earlier assumption about A since, choosing it in this way, it
only involves instances of y!) Consequently, we see that
Z x
2
p(x, y) = e−t dt,
t=y

is the required definite integral.

For (ii), we are asked to fix the value of x so that we can find
Z x Z x Z x
2
p(x, y) dx = e−t dt dy,
y=0 y=0 t=y

and to do this we will need to change the order of integration. In order to do this, we
note that this repeated integral takes a fixed 0 ≤ y ≤ x and then takes y ≤ t ≤ x which
5 gives us the region of integration illustrated in Figure 5.15. Consequently, to change the
order of integration we need to fix 0 ≤ t ≤ x and then take 0 ≤ y ≤ t so that we now
have Z x Z x Z x Z x Z t
−t2 2
p(x, y) dx = e dt dy = e−t dy dt,
y=0 y=0 t=y t=0 y=0

and we can evaluate this new repeated integral by noting that


" #x
x t x t x 2 2
e−t 1 − e−x
Z Z Z   Z
−t2 −t2 −t2
e dy dt = ye dt = te dt = − = .
t=0 y=0 t=0 y=0 t=0 2 2
t=0

Consequently, we have found that


x 2
1 − e−x
Z
p(x, y) dx = ,
y=0 2

which is a function of x only, as it should be.

x
t

t=x
=
y

O x t

Figure 5.15: The region of integration for Activity 5.5.

170
5.3. Solutions to activities

Solution to activity 5.6


In Example 5.12, we found that the inverse transformation was given by
u uv
x(u, v) = and y(u, v) = ,
1+v 1+v
and we can see that this gives us
!
1/(v + 1) −u/(v + 1)2
 
d(x, y) xu xv
= = ,
d(u, v) yu yv v/(v + 1) u/(v + 1)2

where, in particular, we have used the quotient rule to see that


∂y (u)(1 + v) − (uv)(1) u
= 2
= .
∂v (1 + v) (1 + v)2
As such, the Jacobian for the inverse transformation is 5
!
1/(v + 1) −u/(v + 1)2
 
∂(x, y) d(x, y)
= det = det
∂(u, v) d(u, v) v/(v + 1) u/(v + 1)2
u uv u(v + 1) u
= 3
+ 3
= 3
= ,
(v + 1) (v + 1) (v + 1) (v + 1)2
which agrees with what we found in the example. In particular, observe that this
Jacobian is non-zero when u 6= 0 and is well-defined as long as 1 + v 6= 0.

Indeed, if we think about the original transformation, it is interesting to note that

u=x+y means that u 6= 0 ⇐⇒ x + y 6= 0,

and we can also use


y x+y
1+v =1+ = ,
x x
to see that
1 + v 6= 0 ⇐⇒ x + y 6= 0.
That is, the conditions for both Jacobians to be non-zero are the same regardless of
whether we write them in terms of u and v or in terms of x and y.

Solution to activity 5.7


We have already seen, in Example 5.10, how the region D becomes the region ∆ where

∆ = {(u, v) : 1 ≤ u ≤ 2 and 1 ≤ v ≤ 2},

under this transformation and, in Example 5.12, we found its Jacobian. This means
that the only thing left for us to consider when using the given transformation to
evaluate the double integral
y
Z Z
dxdy,
D x
is what happens to the integrand,
y
f (x, y) = ,
x

171
5. Double integrals

but we can easily see that this is just v. Thus, we now have
Z Z  
y u
Z Z
dxdy = v dudv
D x ∆ (1 + v)2
Z 2 Z 2 
uv
= 2
du dv
v=1 u=1 (1 + v)
Z 2  Z 2 
v
= 2
dv u du .
v=1 (v + 1) u=1

Now, the u-integral gives us


2 2
u2 4−1

3
Z
u du = = = ,
u=1 2 u=1 2 2
5 whereas, using the substitution w = 1 + v so that dw = dv, we see that the v-integral
gives us
2 3 Z 3  3
w−1
 
v 1 1 1
Z Z
dv = dw = − dw = ln w +
v=1 (v + 1)2 w=2 w
2
w=2 w w2 w w=2
     
1 1 3 1
= ln 3 + − ln 2 + = ln − .
3 2 2 6

Consequently, we have
      
y 3 1 3 3 3 1
Z Z
dxdy = ln − = ln − ,
D x 2 6 2 2 2 4

as the final answer.

Solution to activity 5.8


At the end of Example 5.13, we had the repeated integral
3 6−v
du
Z Z
dv,
v=2 u=v 2u2 v

and we have been asked to evaluate this. Looking at the u-integral, we see that
6−v  6−v  
du 1 1 1 1
Z
= − = − + ,
u=v 2u2 v 2uv u=v 2 v(6 − v) v 2

and, using partial fractions, we see that


 
1 1 1 1
= + ,
v(6 − v) 6 v 6−v

so that we have
6−v  
du 1 1 1 6
Z
2
= − − + 2 .
u=v 2vu 12 6−v v v

172
5.3. Solutions to activities

This means that we now have


Z 3 Z 6−v Z 3    3
du 1 1 1 6 1 6
2
dv = − − + dv = ln |6 − v| − ln |v| −
v=2 u=v 2u v v=2 12 6 − v v v2 12 v v=2
1 − ln 2
 
1
= (ln 3 − ln 3 − 2) − (ln 4 − ln 2 − 3) = ,
12 12
as ln 4 = 2 ln 2.

Notice, in particular, that the calculation is simpler if we choose v to be the


independent variable because, if we had chosen u to be the independent variable
instead, we would have had to split the region of integration into two sub-regions.
Indeed, referring to Figure 5.9(b), we see that the first of these sub-regions would fix
2 ≤ u ≤ 3 and then take 2 ≤ v ≤ u to give us the repeated integral
Z 3 Z u
dv 5
2
du,
u=2 v=2 2u v

whereas the second would fix 3 ≤ u ≤ 4 and then take 2 ≤ v ≤ 6 − u to give us the
repeated integral Z 4 Z 6−u
dv
2
du,
u=3 v=2 2u v
so that we now have
3 u 4 6−u
dudv dv dv
Z Z Z Z Z Z
= du + du,
∆ 2u2 v u=2 v=2 2u2 v u=3 v=2 2u2 v
and this would require us to do more work!

Solution to activity 5.9


We are given the transformation that maps a point (x, y) ∈ R2 with x > 0 to the point
(u, v) given by
y
u = x2 + y 2 and v= ,
x
and we observe, in particular, that this means that u > 0.

For (i), we can see that y = vx and so


r
u u
u = x2 + y 2 = x2 + v 2 x2 = (1 + v 2 )x2 =⇒ x2 = =⇒ x=± ,
1 + v2 1 + v2
which is well-defined because u > 0. Then, using y = vx again, we see that this gives us
two local inverse transformations, one for x > 0 which is given by
r r
u u
x= and y = v ,
1 + v2 1 + v2
and another for x < 0 which is given by
r r
u u
x=− and y = −v .
1 + v2 1 + v2

173
5. Double integrals

The question tells us that we are interested in points (x, y) where x > 0 and so the
inverse transformation we seek is the first of these.

For (ii), as instructed, we find that the Jacobian of this transformation is

2y 2 y2
     
∂(u, v) ux uy 2x 2y
= det = det =2+ 2 =2 1+ 2 ,
∂(x, y) vx vy −y/x2 1/x x x

and this is well-defined because we are only interested in points where x > 0.

For (iii), the region D is given by

D = {(x, y) : 0 < y ≤ x ≤ 1},

and this tells us that, for any fixed 0 < y ≤ 1, we need to take y ≤ x ≤ 1 in order to be
in the region. As such, a sketch of this region will look like the one illustrated in
5 Figure 5.16(a).

To find ∆, the image of D under the given transformation, we note it is constrained by


the following inequalities.

u > 0 as this is a property of the transformation itself.

0 ≤ v ≤ 1 as, from our description of D, we have x > 0 and


y
0<y≤x =⇒ 0< ≤1 =⇒ 0 ≤ v ≤ 1,
x
if we use the fact that v = y/x from the original transformation.

u ≤ 1 + v 2 as, from our description of D, we have


r
u
x ≤ 1 =⇒ ≤ 1 =⇒ u ≤ 1 + v2,
1 + v2

if we use the inverse transformation found in (i).


Consequently, if we sketch the region in the uv-plane that is given by these inequalities,
we find that ∆ is the region illustrated in Figure 5.16(b).

For (iv), we see that the Jacobian can be written as

y2
 
∂(u, v)
= 2 1 + 2 = 2(1 + v 2 ),
∂(x, y) x

if we use the fact that v = y/x for x > 0. This means that, using our formula, we have
 −1
∂(x, y) ∂(u, v) 1
= = ,
∂(u, v) ∂(x, y) 2(1 + v 2 )

and this is always positive. Consequently, as the integrand

y2
 
y x2 +y2
f (x, y) = e 1+ 2 becomes v eu (1 + v 2 ),
x x

174
5.3. Solutions to activities

if we use the original transformation, we see that the given integral can now be written
as
y2
 
y x2 +y2 dudv 1
Z Z Z Z Z Z
u 2
e 1 + 2 dxdy = v e (1 + v ) 2
= v eu dudv.
D x x ∆ 2(1 + v ) 2 ∆

However, this is just a half times the double integral that we found in Activity 5.3 and
so we find that
y2 e2 − e −1
 
y x2 +y2
Z Z
e 1 + 2 dxdy = ,
D x x 4
is the final answer.

y v

5
1

v=1
1
x
=

x=1
y


D
O u
x 1 2
O u=1+v 2
1
(a) (b)

Figure 5.16: The sketches for Activity 5.9(iii). In (a) we have the region, D, and in (b)
we have its image, ∆.

Solution to activity 5.10


For (i), we are interested in the region in the xy-plane that arises when we take polar
coordinates with a fixed 0 ≤ θ ≤ π/2 and then take 0 ≤ r ≤ 1/ cos θ. However, since
0 ≤ θ ≤ π/2, we know that cos θ ≥ 0 and so we can see that the second of these
inequalities gives us

0 ≤ r ≤ 1/ cos θ =⇒ 0 ≤ r cos θ ≤ 1 =⇒ 0 ≤ x ≤ 1,

as x = r cos θ. Then, as the first of these inequalities specifies that 0 ≤ θ ≤ π/2, we can
also see that we must be in the positive quadrant of the xy-plane. Putting this
information together, we get the region illustrated in Figure 5.17(a).

175
5. Double integrals

For (ii), we are interested in the region


√ in the xy-plane that arises −1 when we take polar
coordinates with a fixed 2 ≤ r ≤ 2 2 and then take π/4 ≤ θ ≤ sin (2/r). However, the
second of these inequalities tells us that we need θ ≥ π/4 which means that we must
have y ≥ x as θ = π/4 gives us the line y = x whereas
 
−1 2 2
θ ≤ sin =⇒ sin θ ≤ =⇒ r sin θ ≤ 2 =⇒ y ≤ 2,
r r

as y = r sin θ. Then as the first of these inequalities specifies that 2 ≤ r ≤ 2 2, we can
also see that we√must be between the two circles, both centred on the origin, that have
radii of 2 and 2 2. Putting this information together,6 we get the region illustrated in
Figure 5.17(b).

y y
5

2 2

x
=
y
2
R R

O x O √ x
1 2 2 2
(a) (b)

Figure 5.17: (a) and (b) are the regions for Activities 5.10(i) and (ii) respectively.

Solution to activity 5.11


For (i), we take a fixed x ≥ 1 and substitute z = y/x so that y = xz and dy = x dz.
This means that the y-integral becomes
Z ∞ Z ∞ Z ∞
dy x dz x dz 1 ∞ dz
Z
2 2 2
= 2 2 2 2
= 4 2 2
= 3 ,
y=0 (x + y ) z=0 (x + x z ) z=0 x (1 + z ) x z=0 (1 + z 2 )2

as x is a constant as far as integration with respect to y is concerned. In particular, we


can see that this is an improper integral of the first kind with an integrand
1 1
f (z) = which behaves like g(z) = ,
(1 + z 2 )2 z4
for large z. As such, if we note that
6
And taking great care when sketching the lines y = x and y = 2 together with the two circles so
that we can appreciate where the points of intersection are!

176
5.3. Solutions to activities

f (z) z4 1
lim = lim 2 2
= lim 2 = 1, and
z→∞ g(z) z→∞ (1 + z ) z→∞ 1
+ 1
z 2

Z ∞ Z ∞
dz
g(z) dz = is convergent,
1 1 z4
then we see that the integral Z ∞
dz
2 2
,
z=0 (1 + z )
must also be convergent by the LCT. Consequently, we can see that the y-integral can
be written as Z ∞
dy 1 ∞ dz k
Z
2 2 2
= 3 2 2
= 3,
y=0 (x + y ) x z=0 (1 + z ) x
where k is the finite value, whatever it is, of the z-integral. This means that the
repeated integral we are given can now be written as
Z ∞ Z ∞
dy
Z ∞
k

k
∞
k
5
I= dx 2 2 2
= 3
dx = − 2 = ,
x=1 y=0 (x + y ) x=1 x 2x x=1 2

and so I is convergent as well since its value is just k/2.

For (ii), we see that the given repeated integral can be written as the double integral
Z ∞
dy dxdy
Z Z
I= 2 2 2
= 2 2 2
,
y=0 (x + y ) R (x + y )

where R is the region illustrated in Figure 5.18. To convert this to polar coordinates we
see that the integrand
1 1
f (x, y) = becomes ,
(x2 + y 2 )2 r4
because the trigonometric identity sin2 θ + cos2 θ = 1 gives us

x2 + y 2 = r2 cos2 θ + r2 sin2 θ = r2 (sin2 θ + cos2 θ) = r2 ,

and, as we have seen, the Jacobian for this transformation is


∂(x, y)
= r.
∂(r, θ)
So, given that the question wants us to take θ as the dependent variable, all we have to
do is figure out what the region R is in terms of polar coordinates by taking r to be the
independent variable. Indeed, by looking at Figure 5.18, we can see that if we fix a value
of 1 ≤ r ≤ ∞, we can then take 0 ≤ θ ≤ cos−1 (1/r) since the largest angle we are
interested in is given by
 
1 −1 1
x = 1 =⇒ r cos θ = 1 =⇒ cos θ = =⇒ θ = cos ,
r r
for each value of r that we consider. This means that we now have
Z ∞ Z cos−1 (1/r) ! Z ∞ Z cos−1 (1/r) !
dxdy r dθ
Z Z
I= 2 2 2
= 4
dθ dr = dr,
R (x + y ) r=1 θ=0 r r=1 θ=0 r3

177
5. Double integrals

and so, evaluating the θ-integral as directed, we find that


∞ cos−1 (1/r) ∞
cos−1 (1/r)

θ
Z Z
I= dr = dr.
r=1 r3 θ=0 r=1 r3

We can then see that, for r ≥ 1, we have

cos−1 (1/r)
 
1 −1 1 π π
0 ≤ ≤ 1 =⇒ 0 ≤ cos ≤ =⇒ 0≤ 3
≤ 3,
r r 2 r 2r

which means that, as the integral



π
Z
dr,
r=1 2r3
is convergent, we can conclude that I must also be convergent by the DCT.

5 For (iii), we note that all of the integrals that we have encountered in this question can
be evaluated to give us the value of I. For instance:

In (i), we found that



k dz
Z
I= where k = ,
2 z=0 (1 + z 2 )2

and, as you can see for yourself, we can find the value of k by using the
substitution z = tan ϕ.

In (ii), we found that



cos−1 (1/r)
Z
I= dr,
r=1 r3
and, as you can see for yourself, we can find the value of I by using the substitution
cos ψ = 1/r.

Alternatively, using the repeated integral given in the question, you can find the
value of I by starting off with the substitution y = x tan µ for fixed x ≥ 1.
However, we will find the value of I by changing the order of integration in our answer
to (ii). To do this, we note that if we take θ to be the independent variable, then for
each fixed 0 ≤ θ ≤ π/2, we need r ≥ 1/ cos θ since the smallest value of r we are
interested in is given by
1
x=1 =⇒ r cos θ = 1 =⇒ r= ,
cos θ
for each value of θ that we consider. This means that we now have
Z π/2 Z ∞ 
dxdy dr
Z Z
I= 2 2 2
= 3
dθ,
R (x + y ) θ=0 1/ cos θ r

and, evaluating the r-integral, we find that


Z ∞ ∞
cos2 θ

dr 1 cos(2θ) + 1
3
= − 2
= = ,
1/ cos θ r 2r 1/ cos θ 2 4

178
5.3. Exercises

if we use the double-angle formula cos(2θ) = 2 cos2 θ − 1. Consequently, we find that


π/2  π/2
cos(2θ) + 1 1 sin(2θ) 1 π  π
Z
I= dθ = +θ = = ,
θ=0 4 4 2 θ=0 4 2 8

and, of course, this is what you should find if you try any of the other methods
mentioned above.

r R

θ
O x
1 r
Figure 5.18: What we need to think about when changing to polar coordinates in
Activity 5.11.

Exercises
Exercise 5.1
By changing the order of integration, evaluate the repeated integral

1 2x−x2
x−1 y
Z Z
dx e dy.
x=0 y=x y−1

[Hint: You will need to sketch the region of integration.]

Exercise 5.2
The transformation
y
u=x+y and v= ,
x+y
is defined for points (x, y) ∈ R2 with x + y 6= 0.

179
5. Double integrals

(i) Find the Jacobian of this transformation.

(ii) Find the inverse transformation.

(iii) Sketch the triangle, T , that is bounded by the lines x = 0, y = 0 and x + y = 1.

(iv) Sketch the image of T under this transformation.


Z Z
Hence evaluate the double integral ey/(x+y) dxdy.
T

Exercise 5.3
Find the volume enclosed by the surface z = x2 + y 2 when x and y satisfy the inequality
x 2 + y 2 ≤ a2 .

[Hint: Use polar coordinates.]


5
Exercise 5.4
Evaluate the double integral
x2
Z Z
dxdy,
D y2
where D is the region bounded by the curves x = 2, x = y and xy = 1.

[Hint: Use the transformation where u = xy and v = y/x for x 6= 0.]

Solutions to exercises
Solution to exercise 5.1
have a repeated integral where, for a fixed 0 ≤ x ≤ 1, we take
Here we √
x ≤ y ≤ 2x − x2 . In particular, the largest value of y we consider satisfies the equation

y = 2x − x2 =⇒ x2 − 2x + y 2 = 0 =⇒ (x − 1)2 + y 2 = 1,

and so this value of y lies on the circle of radius one centred on the point (1, 0).
Consequently, as illustrated in Figure 5.19, we can sketch the region of integration, R.
p change the order of integration, we take a fixed 0 ≤ y ≤ 1 and then take
So, to
1 − 1 − y 2 ≤ x ≤ y since the smallest value of x we are interested in is given by
p p
(x − 1)2 + y 2 = 1 =⇒ x − 1 = ± 1 − y 2 =⇒ x = 1 ± 1 − y 2 ,

and we take the ‘−’ here because we know that x < 1 if we are in R. This means that
we now have

180
5.3. Solutions to exercises


1 2x−x2 1 y
x−1 y x−1 y
Z Z Z Z
dx e dy = dy √ e dx
x=0 y=x y−1 y=0 x=1− 1−y 2 y−1
1 y
(x − 1)2 y
Z 
= e √ dy
y=0 2(y − 1) x=1− 1−y 2

1
(y − 1)2 − (1 − y 2 ) y
Z
= e dy
y=0 2(y − 1)
1
2y(y − 1) y
Z
= e dy
y=0 2(y − 1)
Z 1
= y ey dy
y=0 5
= [y ey − ey ]1y=0

= (e1 − e1 ) − (0 − e0 )
= 1,
where we have used integration by parts to evaluate the y-integral.

y
x
=
y

O x
1

Figure 5.19: A sketch of the region of integration in Exercise 1.

Solution to exercise 5.2


We are given the transformation
y
u=x+y and v= ,
x+y
which is defined for points (x, y) ∈ R2 with x + y 6= 0.

The Jacobian of this transformation is given by


   
∂(u, v) ux uy 1 1 x+y 1
= det = det 2 2 = = ,
∂(x, y) vx vy −y/(x + y) x/(x + y) (x + y)2 x+y

181
5. Double integrals

and this is well-defined since we are told that x + y 6= 0.

For (ii), with x + y 6= 0, we can see that u = x + y gives us u 6= 0 and


y y
v= = =⇒ y = uv,
x+y u

which means that, using u = x + y again, we have

u = x + uv =⇒ x = u − uv = u(1 − v).

That is, the inverse transformation is given by

x = u(1 − v) and y = uv,

where u 6= 0.
5
For (iii), we can see that a sketch of the triangle, T , that is bounded by the lines x = 0,
y = 0 and x + y = 1 will look like the one illustrated in Figure 5.20(a).

For (iv), we see that the triangle T is determined by the inequalities

x ≥ 0, y≥0 and x + y ≤ 1,

and so the image of this triangle under the given transformation will be given by the
following inequalities.

(a) x ≥ 0 implies that u(1 − v) ≥ 0 which means that we can have either have u ≥ 0
and v ≤ 1 or u ≤ 0 and v ≥ 1.

(b) y ≥ 0 implies that uv ≥ 0 which means that we can have either have u ≥ 0 and
v ≥ 0 or u ≤ 0 and v ≤ 0.

(c) x + y ≤ 1 implies that u ≤ 1.


But, a quick sketch should convince us that the inequalities in (a) and (b) only hold
simultaneously if u ≥ 0 and 0 ≤ v ≤ 1 whereas if we take the inequality in (c) into
account as well, we see that we must also have u ≤ 1. That is, putting this all together,
we see that the image of T is the rectangular region Θ illustrated in Figure 5.20(b).

Lastly, we are asked to evaluate the double integral


Z Z
ey/(x+y) dxdy,
T

and, performing the change of variable we see that this is given by


Z Z Z Z
y/(x+y)
e dxdy = ev u dudv,
T Θ

as we have  −1
∂(x, y) ∂(u, v)
= = x + y = u,
∂(u, v) ∂(x, y)

182
5.3. Solutions to exercises

which is never negative if u ≥ 0. Thus, we can see that


Z Z Z 1 Z 1 
y/(x+y) v
e dxdy = e u du dv
T v=0 u=0
Z 1  Z 1 
v
= u du e dv
u=0 v=0
1 1
u2
 
v
= e
2 u=0 v=0

e −1
= ,
2
is the sought after answer.

y v
5

1 1
x
+

Θ
y
=
1

O x O u
1 1
(a) (b)

Figure 5.20: The sketches for Exercise 2. (a) The region, T . (b) The region Θ is the image
of T under the given transformation.

Solution to exercise 5.3


The volume, V , enclosed by the surface z = x2 + y 2 when x and y satisfy the inequality
x2 + y 2 ≤ a2 is given by the double integral
Z Z
V = (x2 + y 2 ) dxdy,
D

where D is the region in the z = 0 plane given by x2 + y 2 ≤ a2 , i.e. it is the disk of radius
a centred on the origin. So, following the hint, we use polar coordinates to see that as
∂(x, y)
x2 + y 2 = r 2 and = r,
∂(r, θ)
we need to evaluate the double integral
Z Z
V = r3 drdθ,

183
5. Double integrals

where ∆ is the image of D under the transformation described by polar coordinates, i.e.
∆ is given by 0 ≤ r ≤ a and 0 ≤ θ ≤ 2π. Thus, we see that

2π a 2π a  2π  4 a  4
πa4
Z  Z  Z 
r a
Z
3 3
V = r dr dθ = dθ r dr = θ = 2π = ,
θ=0 r=0 θ=0 r=0 θ=0 4 r=0 4 2

is the required volume.

Solution to exercise 5.4


To evaluate the double integral
x2
Z Z
dxdy,
D y2
5
where D is the region bounded by the rectangular hyperbola xy = 1 and the straight
lines x = 2 and x = y, we start by noting that D is the region illustrated in
Figure 5.21(a). We then follow the hint and think about the transformation

y
u = xy and v= ,
x

for x 6= 0. The image of D under this transformation, ∆, and this can be found by
noting that D is specified by the inequalities

x ≤ 2, y≤x and xy ≥ 1,

which means that, from the transformation, we can use u = xy and y = vx to see that
u = vx2 so that the first of these inequalities gives us u ≤ 4v whereas the other two
inequalities give us v ≤ 1 and u ≥ 1 straightaway. As such, we can see that ∆ is the
region illustrated in Figure 5.21(b). Furthermore, the Jacobian of the transformation is
given by
   
∂(u, v) ux uy y x y y 2y
= det = det 2 = + = ,
∂(x, y) vx vy −y/x 1/x x x x

and this is well-defined as long as x 6= 0. This means that we have

 −1
∂(x, y) ∂(u, v) x 1
= = = ,
∂(u, v) ∂(x, y) 2y 2v

which is positive given that we are only interested in values of v > 0 here. So, noting
that the integrand can be written as 1/v 2 , we see that under this change of variable, we
have
Z Z 2 Z Z   
x 1 1 dudv
Z Z
2
dxdy = 2
dudv = 3
,
D y ∆ v 2v ∆ 2v

184
5.3. Solutions to exercises

and using our sketch of ∆, we can see that this means that we have
Z Z 2 Z 1 Z 4v 
x du
2
dxdy = 3
dv
D y v=1/2 u=1 2v
Z 1 h
u i4v
= 3 u=1
dv
v=1/2 2v
Z 1
4v − 1
= 3
dv
v=1/2 2v
Z 1  
2 1
= − dv
v=1/2 v 2 2v 3
 1
2 1
= − + 2
v 4v v=1/2 5
   
1
= −2 + − −4+1
4
5
= ,
4
as the final answer.

y v
x
=

2 1
y

D 1
1 2
4v
=

1
xy = 1
u

O x O u
1 2 1 4
(a) (b)

Figure 5.21: The sketches for Exercise 4. (a) The region, D. (b) The region ∆ is the image
of D under the given transformation.

185
5. Double integrals

186
Chapter 6
Manipulation of integrals

Essential reading

(For full publication details, see Chapter 1.)

+ Binmore and Davies (2002) Sections 10.12 and 10.13.

+ Ostaszewski (1991) Sections 18.1–18.5 and 18.11.

Further reading
6
+ Wrede and Spiegel (2010) pp.186–7, 195–6 and 203.

Aims and objectives

The objectives of this chapter are:

to see how proper integrals can be manipulated and be able to justify the use of
these manipulations by establishing the required joint continuity;

to see how improper integrals can be manipulated and be able to justify the use of
these manipulations by establishing the required joint continuity and dominated
convergence;

to see why the manipulation of integrals is useful in mathematics.


Specific learning outcomes can be found near the end of this chapter.

6.1 The manipulation of proper integrals


In this chapter, we are interested in manipulating integrals which involve functions of
two variables. In particular, if we have a function of two variables, say K(t, x), we may
be interested in the integral
Z b
K(t, x) dt,
a

where we have taken one of these variables, in this case t, as the variable of integration
and the other, in this case x, as a parameter whose value does not depend on t. Now,
given the presence of the parameter, we may wonder how this integral changes if we
attempt to manipulate it by performing one of the following operations on it.

187
6. Manipulation of integrals

Z b
Taking a limit: What is lim K(t, x) dt?
x→x0 a
Z b
d
Differentiation: What is K(t, x) dt?
dx a
Z d Z b 
Integration: What is K(t, x) dt dx?
c a
Indeed, not only do we need to know how the integral changes under these operations,
we also need to know when these changes are valid, i.e. when we can actually
manipulate such integrals in these ways.
As we shall see, the ‘when’ will come down to a property of functions of two variables
called joint continuity and we shall consider what this means first of all. Then we shall
consider the ‘how’ by looking at the rules for manipulating such integrals using the
operations listed above. Lastly, we will look at some applications of this material.

6.1.1 Joint continuity


6 Technically, we say that a function of two variables, such as K(t, x), is jointly
continuous at the point (t0 , x0 ) if there is a δ > 0 such that, for every ε > 0, we have

|(t, x) − (t0 , x0 )| < δ =⇒ |K(t, x) − K(t0 , x0 )| < ε,

i.e. if whenever (t, x) is sufficiently close to (t0 , x0 ), we can guarantee that K(t, x) is
sufficiently close to K(t0 , x0 ). Indeed, we shall be concerned with functions, K(t, x),
which are jointly continuous over some rectangle where a ≤ t ≤ b and c ≤ x ≤ d.
However, in this course, we will not really make use of this technical definition. Instead,
we will be able to see whether we have the required joint continuity by considering how
it is related to the continuity of functions of one variable and how it is preserved when
we take combinations of functions of two variables.1
Firstly, we note that we can think of a function of one variable as a function of two
variables and this allows us to link continuity (as we saw in Section 2.2.1) to joint
continuity as follows.
Theorem 6.1 If, for x ∈ [c, d], g(x) is a continuous function of one variable, then

G(t, x) = g(x),

is a jointly continuous function of two variables for t ∈ R and x ∈ [c, d].

Of course, this also means that the continuity of f (t) guarantees the joint continuity of
F (t, x) = f (t). Secondly, if we look at combinations of jointly continuous functions, we
find that we get the following.
Theorem 6.2 If the functions F (t, x) and G(t, x) are jointly continuous at the point
(t0 , x0 ), then so are the functions
1
That is, our discussion of joint continuity will follow the same pattern as our discussion of continuity
in Section 2.2.1.

188
6.1. The manipulation of proper integrals

kF (t, x) where k ∈ R;

F (t, x) + G(t, x);

F (t, x)G(t, x);

F (t, x)
as long as G(t, x) 6= 0.
G(t, x)
Moreover, if F (t, x) is jointly continuous at the point (t0 , x0 ) and g(y) is continuous at
y0 = F (t0 , x0 ), then the function of two variables G(t, x) given by the composition

G(t, x) = g(F (t, x)),


is also jointly continuous at the point (t0 , x0 ).

Or, more generally, if F1 (t, x) and F2 (t, x) are jointly continuous at the point (t0 , x0 )
and G(y, z) is jointly continuous at the point (y0 , z0 ) where y0 = F1 (t0 , x0 ) and
z0 = F2 (t0 , x0 ), then the function of two variables H(t, x) given by the composition

H(t, x) = G(F1 (t, x), F2 (t, x)),


6
is also jointly continuous at the point (t0 , x0 ).

Just as we did when we discussed the results on continuity in Section 2.2.1, we state
these intuitively obvious results without proof and look at a few examples of how they
can be used.

Example 6.1 Explain why the function x2 + t2 is jointly continuous for all t, x ∈ R.

As x2 and t2 are both continuous functions of one variable for all x ∈ R and t ∈ R
respectively, by Theorem 6.1, we can treat them both as jointly continuous functions
of the two variables x and t for points (t, x) ∈ R2 . Then, by Theorem 6.2, as it is the
sum of two jointly continuous functions, the function of two variables given by
x2 + t2 is also jointly continuous for all (t, x) ∈ R2 .

Example 6.2 Explain why the function xt is jointly continuous for points (t, x)
where x > 0 and t ∈ R.

We use the fact that


xt = et ln x ,
and, using Theorems 6.1 and 6.2, we note that:

As t and ln x are continuous functions of one variable for all t ∈ R and x > 0
respectively, we can treat them both as jointly continuous functions of the two
variables x and t for all points (t, x) where t ∈ R and x > 0.

As the product of two jointly continuous functions, the function of two variables
t ln x is also jointly continuous for all points (t, x) where t ∈ R and x > 0.

189
6. Manipulation of integrals

As the function of one variable ey is continuous for all y ∈ R, we can take


y = t ln x and so that the composition et ln x = xt is jointly continuous for all
points (t, x) where t ∈ R and x > 0.
This establishes that the function xt is jointly continuous at all points (t, x) where
x > 0 and t ∈ R, as required.

Activity 6.1 Explain why the function tx is jointly continuous for points (t, x)
where x ∈ R and t > 0.

Lastly, following on from what we saw at the end of our discussion of continuity in
Section 2.2.1, we consider how functions which are undefined at certain points can be
‘repaired’ to give us suitably jointly continuous functions if we take a little more care
with how they are defined.

Example 6.3 Given that 0 ≤ t ≤ π and |x| < 1, show that the function

ln(1 + x cos t)
F (t, x) = ,
cos t
6 is defined and jointly continuous as long as t 6= π/2. Redefine this function so that it
is defined for all 0 ≤ t ≤ π and |x| < 1 and hence establish that this redefined
function is jointly continuous for all such values of t and x.

We first notice that, for 0 ≤ t ≤ π, we have | cos t| ≤ 1 and so, as |x| < 1 too, we can
see that |x cos t| < 1. That is, we have −1 < x cos t < 1 which means that
1 + x cos t > 0, i.e. ln(1 + x cos t) is defined and jointly continuous for all of the given
values of t and x by Theorem 6.2. Then, as long as t 6= π/2, we also have cos t 6= 0
and so we see that F (t, x) is defined and jointly continuous as long as this is also the
case.

To suitably redefine the function when t = π/2 so that we can make progress with
the next part of the question, we note that if we fix a value of x 6= 0 as well, we can
make the substitution u = x cos t to see that
   
ln(1 + x cos t) ln(1 + u)
lim F (t, x) = lim x = lim x = x,
t→π/2 t→π/2 x cos t u→0 u

if we use L’Hôpital’s rule as we did in Example 2.23. In particular, this should


prompt us to be more careful when we define F (t, x) in the sense that, given a
relevant value of x 6= 0, what we really need here is the function
ln(1 + x cos t)
F (t, x) = ,
cos t
for t 6= π/2 together with F (t, x) = x when t = π/2.

So, how do we establish the required joint continuity of this redefined function?
Consider the function, h(u), which (for u > −1) is given by

ln(1 + u)
h(u) = ,
u

190
6.1. The manipulation of proper integrals

if u 6= 0 together with h(0) = 1. Indeed, the important thing to note here is that, as

lim h(u) = 1 = h(0),


h→0

we can certainly assert that h is continuous (for u > −1). With this function, we
then claim that F (t, x) can actually be written as

F (t, x) = xh(x cos t),

for all 0 ≤ t ≤ π and |x| < 1. To see why, note that if x 6= 0, we have
 
ln(1 + x cos t) ln(1 + x cos t)
xh(x cos t) = x = = F (t, x),
x cos t cos t

as required whereas, if x = 0 and t 6= π/2, we have cos t 6= 0 and so

ln 1
F (t, 0) = = 0,
cos t
(using the first part of the redefinition of F ) and, if x = 0 and t = π/2, we have
F (π/2, 0) = 0 (using the second part of the redefinition of F ), i.e. as F (t, 0) = 0 for
6
all 0 ≤ t ≤ π, we can see that

0h(0) = 0(1) = 0 = F (t, 0),

as well. Consequently, as F (t, x) = xh(x cos t) is now both well-defined and the
product of two jointly continuous functions for 0 ≤ t ≤ π and |x| < 1, we have
established that it is a jointly continuous function as well by Theorem 6.2.

6.1.2 The manipulation rules for proper integrals


Now that we understand joint continuity, we can present the rules for manipulating
proper integrals. In particular, we will present the three rules and a simple example to
motivate their use.
The first rule allows us to manipulate proper integrals by taking limits through the
integral sign as follows.
Rule 1: If F (t, x) is jointly continuous at all points (t, x) in the rectangle
[a, b] × [c, d], i.e. at all points where a ≤ t ≤ b and c ≤ x ≤ d, then
Z b Z b
lim F (t, x) dt = lim F (t, x) dt,
x→x0 a a x→x0
when c < x0 < d.
Notice that the joint continuity of F (t, x) over the given rectangle guarantees that we
are manipulating a proper integral on the left-hand side of this rule and it also allows us
to conclude that we will get
lim F (t, x) = F (t, x0 ),
x→x0
when we look at the integrand on the right-hand side of this rule. Let’s look at a simple
example of its use.

191
6. Manipulation of integrals

Z T
Example 6.4 Given that f (t) is continuous for all t ∈ R, find lim f (t) e−xt dt.
x→0 0

As f (t) is continuous for all t ∈ R, it should be clear that f (t) e−xt is jointly
continuous at all points (t, x) ∈ R2 . In particular, using this joint continuity at all
points (t, x) in the rectangle [0, T ] × [−1, 1] for the justification, we can use Rule 1 to
see that Z T Z T Z T
−xt −xt
lim f (t) e dt = lim f (t) e dt = f (t) dt,
x→0 0 0 x→0 0
as here x0 = 0 is such that −1 < x0 < 1.

Activity 6.2 Verify by direct calculation that


Z 2 Z 2  
t t
lim dt = lim dt,
x→1 0 x + 1 0 x→1 x+1

in accordance with Rule 1.


6
The second rule allows us to manipulate proper integrals by taking derivatives through
the integral sign as follows.

Rule 2: If F (t, x) and ∂F/∂x are jointly continuous at all points (t, x) in the
rectangle [a, b] × [c, d], i.e. at all points where a ≤ t ≤ b and c ≤ x ≤ d, then
Z b  Z b
d ∂F
F (t, x) dt = dt,
dx a x=x0 a ∂x
x=x0

when c < x0 < d.


Notice that the joint continuity of F (t, x) over the given rectangle guarantees that we
are manipulating a proper integral on the left-hand side of this rule and the joint
continuity of ∂F/∂x over the given rectangle guarantees that the integral we get on the
right-hand side will also be a proper integral.
In particular, observe that as stated, Rule 2 requires us to test both functions for joint
continuity, the worry here being that partially differentiating the jointly continuous
function F (t, x) with respect to x may produce a function, ∂F/∂x, that is not itself
jointly continuous. That is, Rule 2 would no longer be applicable as we could end up
with an improper integral of the second kind on the right-hand side and this might not
exist.
However, overall, provided we have the required joint continuity, this rule will work for
any x = x0 where c < x0 < d. Let’s look at a simple example of its use.

T
d
Z
Example 6.5 Given that f (t) is continuous for all t ∈ R, find f (t) e−xt dt.
dx 0

As we saw in Example 6.4, the continuity of f (t) for all t ∈ R guarantees that
f (t) e−xt is jointly continuous at all points (t, x) ∈ R2 . In particular, using this joint

192
6.1. The manipulation of proper integrals

continuity at all points (t, x) in the rectangle [0, T ] × [c, d] with c, d ∈ R, we see that
Rule 2 ‘suggests’ that
Z T Z T   Z T
d −xt ∂ −xt
f (t) e dt = f (t) e dt = (−t)f (t) e−xt dt,
dx 0 0 ∂x 0

for any x such that c < x < d. Indeed, this ‘suggestion’ is correct since we also see
that (−t)f (t) e−xt is jointly continuous at all points (t, x) in the given rectangle.

Activity 6.3 Verify by direct calculation that, for any x > −1,
Z 2 Z 2  
d t ∂ t
dt = dt,
dx 0 x + 1 0 ∂x x+1

in accordance with Rule 2.

The third rule allows us to manipulate proper integrals by taking integrals through the
integral sign as follows.

Rule 3: If F (t, x) is jointly continuous at all points (t, x) in the rectangle 6


[a, b] × [c, d], i.e. at all points where a ≤ t ≤ b and c ≤ x ≤ d, then
Z d Z b  Z b Z d 
F (t, x) dt dx = F (t, x) dx dt.
x=c t=a t=a x=c

Notice that the joint continuity of F (t, x) over the given rectangle guarantees that we
are manipulating a proper integral and the fact that we can perform this manipulation
should come as no surprise as it was the basis of our method of evaluating double
integrals by repeated integration. Let’s look at a simple example.

Example 6.6 Find Z 1 Z T 


−xt
f (t) e dt dx,
x=−1 t=0
given that f (t) is continuous for all t ∈ R.

As we saw in Example 6.4, the continuity of f (t) for all t ∈ R, guarantees that
f (t) e−xt is jointly continuous at all points (t, x) ∈ R2 . In particular, using this joint
continuity at all points (t, x) in the rectangle [0, T ] × [−1, 1], we can use Rule 3 to
see that
Z 1 Z T  Z T Z 1 
−xt −xt
f (t) e dt dx = f (t) e dx dt.
x=−1 t=0 t=0 x=−1

Now, looking at the inner integral on the right-hand side for t 6= 0, we have
Z 1 1
e−xt et − e−t

−xt
f (t) e dx = f (t) = f (t) ,
x=−1 −t x=−1 t
and so, if we are not being particularly careful, we can conclude that
Z 1 Z T Z T
et − e−t

−xt
f (t) e dt dx = f (t) dt.
x=−1 t=0 t=0 t

193
6. Manipulation of integrals

Notice, in particular, that this answer looks like it is an improper integral of the
second kind with a problem at t = 0 and, if it was, this would be a significant
problem for Rule 2. But, in Activity 6.4, we will see why this is not the case if we
take a little more care.

Activity 6.4 Following on from Example 6.6, consider the function, g(t), defined by
the integral Z 1
g(t) = f (t) e−xt dx,
x=−1

for t ∈ R. Find g(t) and show that it is continuous for all t ∈ R.

Using this, explain what the answer in Example 6.6 should be if we take a ‘little
more care’ and hence deduce that the answer given is, in fact, acceptable.

Activity 6.5 Verify by direct calculation that, for any M > 1,


Z M Z 2  Z 2 Z M 
t t
dt dx = dx dt,
t=0 x + 1 x=1 x + 1
6 x=1 t=0

in accordance with Rule 3.

6.1.3 Applications of the rules for manipulating proper integrals


Using manipulation to find integrals

One particularly useful application of the rules for manipulating proper integrals,
especially Rule 2, is that they allow us to find proper integrals in cases where our
traditional techniques may not be of much use. Let’s look at an example.

Example 6.7 For |x| < 1, consider the integral


Z π
ln(1 + x cos t)
I(x) = dt.
0 cos t

Find I 0 (x) and hence determine I(x).

Assuming that Rule 2 holds, we would expect to find that


Z π Z π   Z π
0 d ln(1 + x cos t) ∂ ln(1 + x cos t) 1
I (x) = dt = dt = dt,
dx 0 cos t 0 ∂x cos t 0 1 + x cos t

and this works provided that the functions

ln(1 + x cos t) ∂F 1
F (t, x) = and = ,
cos t ∂x 1 + x cos t
are both jointly continuous for 0 ≤ t ≤ π and |x| < 1. Now, we established the
required joint continuity of F (t, x) in Example 6.4 and so we just need to do the
same for ∂F/∂x. To do this, note that as 0 ≤ t ≤ π we have | cos t| ≤ 1 and so, as

194
6.1. The manipulation of proper integrals

|x| < 1 too, we can see that |x cos t| < 1. That is, we have −1 < x cos t < 1 which
means that 1 + x cos t > 0, i.e. as ∂F/∂x is the quotient of two jointly continuous
functions and the denominator is always non-zero for the given values of t and x, we
see that it has the required joint continuity too.

Now, the integral we are left with for I 0 (x) is easier to deal with than the original
one for I(x) since, as you will see in Activity 6.5, the substitution u = tan 2t gives us
Z π
0 1 π
I (x) = dt = √ ,
0 1 + x cos t 1 − x2
and so, integrating both sides with respect to x, we find that

I(x) = π sin−1 x + c,

in terms of the arbitrary constant c. However, we can find the value of c by noting
that, from the definition of I(x) we have
Z π
I(0) = 0 dt = 0 =⇒ c = 0,
0 6
as sin−1 0 = 0 and so, finally, we have

I(x) = π sin−1 x,

for |x| < 1.

Activity 6.6 Following on from Example 6.7, use the substitution u = tan 2t to
show that Z π
1 π
dt = √ .
0 1 + x cos t 1 − x2
[The solution to Exercise 5.4 of 174 Calculus may be useful if you have forgotten
how this substitution works.]

The Leibniz rule for differentiating integrals

Recall that, in Section 3.2.5 we saw that the FTC could be extended to cover the case
where we had to differentiate an integral whose limits of integration were both functions
of x, i.e. we saw that
Z q(x)
d
f (t) dt = q 0 (x)f (q(x)) − p0 (x)f (p(x)).
dx p(x)

However, now that we have Rule 2, we can see how to extend this even further to cover
the case where we want to find
Z q(x)
d
F (t, x) dt,
dx p(x)

and can do this using the Leibniz rule for differentiating integrals.

195
6. Manipulation of integrals

Theorem 6.3 (Leibniz rule) If p(x) and q(x) are differentiable for c < x < d and take
values in some finite interval [a, b], then

q(x) q(x)
d ∂F
Z Z
0 0
F (t, x) dt = q (x)f (q(x), x) − p (x)f (p(x), x) + dt,
dx p(x) p(x) ∂x

provided that both F (t, x) and ∂F/∂x are jointly continuous at all points (t, x) in the
rectangle [a, b] × [c, d].

Of course, given what we saw in Section 3.2.5 and Rule 2, the Leibniz rule should not
come as a great surprise. But, even so, we can easily see where it comes from by using
the chain rule as follows. Suppose that we have the function of three variables given by
the integral Z v
G(u, v, w) = F (t, w) dt,
u

where, quite clearly, we have


v  Z u 
∂G ∂ ∂
Z
6 ∂u
=
∂u u
F (t, w) dt =
∂u

v
F (t, w) dt = −F (u, w),

and
v
∂G ∂
Z
= F (t, w) dt = F (v, w),
∂v ∂v u

if we use the FTC. Moreover, we also have


Z v Z v
∂G ∂ ∂F
= F (t, w) dt = dt,
∂w ∂w u u ∂w

if we use Rule 2. Consequently, using the chain rule from Section 6.3.3 of 174 Calculus,
we have
dG ∂G du ∂G dv ∂G dw
= + + ,
dx ∂u dx ∂v dx ∂w dx
so that, substituting u = p(x), v = q(x) and w = x into the expressions we found above,
and noting that we have the derivatives

du dv dw
= p0 (x), = q 0 (x) and = 1,
dx dx dx

we see that this gives us the Leibniz rule. Of course, this all works because p(x) and
q(x) are assumed to be differentiable for c < x < d and, assuming that these functions
take values in the finite interval [a, b],2 the additional assumption that both F (t, x) and
∂F/∂x are jointly continuous at all points (t, x) in the rectangle [a, b] × [c, d] assures us
that our application of Rule 2 is justified.
Let’s now consider an example of the Leibniz rule in action.

2
So that all of the relevant values of t, i.e. those where p(x) ≤ t ≤ q(x) for c < x < d, are included in
this interval.

196
6.2. The manipulation of improper integrals

Example 6.8 Given that, for x > 0,


Z x2
f (x) = ln(1 + xt) dt,
x

use the Leibniz rule to show that f 0 (1) = ln 2.

Given that Z x2
f (x) = ln(1 + xt) dt,
x
we can apply the Leibniz rule to get
x2 x2
d t
Z Z
0 3 2
f (x) = ln(1 + xt) dt = (2x) ln(1 + x ) − (1) ln(1 + x ) + dt,
dx x x 1 + xt

provided that its use is justified. To see that it is, observe that both x and x2 are
differentiable for x > 0 and, furthermore, the functions
∂F t
F (t, x) = ln(1 + xt) and = , 6
∂x 1 + xt
are both jointly continuous at all points where x > 0 as this guarantees that
1 + xt > 0. (In particular, from the limits of integration, we have 0 < x2 ≤ t ≤ x or
0 < x ≤ t ≤ x2 depending on whether 0 < x ≤ 1 or x ≥ 1 respectively.) Indeed, to be
more precise, both of these functions are jointly continuous at all points (t, x) in any
rectangle of the form [a, b] × [c, d] where 0 < c and 0 < a.

In particular, we find that


1
t
Z
0
f (1) = 2 ln 2 − ln 2 + dt = ln 2,
1 1+t

because the value of the integral here is zero.3

Activity 6.7 Following on from Example 6.8, find f 0 (x) for any x > 0.

6.2 The manipulation of improper integrals


We now consider under what circumstances we can manipulate improper integrals using
the rules given above. In particular, we will only be interested in manipulating improper
integrals of the first kind, i.e. those of the form
Z ∞
K(t, x) dt.
0

3
In particular, notice that we did not need to determine the integral to answer this question, but we
will need to do this to answer the question in Activity 6.7.

197
6. Manipulation of integrals

Of course, by definition, we know that an improper integral of this form is interpreted


as the limit of the proper integrals
Z T
K(t, x) dt,
0

as T → ∞ and so, in order for the manipulations to be valid we will certainly need our
earlier conditions involving joint continuity to be met. The question is, though, what
additional condition(s) will we need to ensure that the manipulations are still valid when
we let T → ∞? That is, when we are manipulating an improper integral of this form?
As we shall see the additional condition will come down to a property of functions of
two variables called dominated convergence and we shall consider what this means first
of all. Then we will be able to state the rules for manipulating such integrals and see
how they can be applied.

6.2.1 Dominated convergence


We say that a function of two variables, such as K(t, x), has dominated convergence for
6 t ≥ a and c ≤ x ≤ d if the following two conditions are met.

Dominance: There is a function, k(t), of t only such that

|K(t, x)| ≤ k(t),

for all t ≥ a and c ≤ x ≤ d.

Convergence: The integral Z ∞


k(t) dt,
a

is convergent.
Here, the dominance condition asserts that there is a function, k(t), which dominates
K(t, x) as it is always greater than |K(t, x)| for the values of x that we are considering.
With this, we can then we can establish that the integral
Z ∞
K(t, x) dt,
a

is convergent for those values of x because, using our test for absolute convergence in
Section 4.2.3, we have
Z ∞ Z ∞ Z ∞
K(t, x) dt ≤ |K(t, x)| dt ≤ k(t) dt,
a a a

and the last of these integrals is convergent because of the convergence condition. In
particular, and most importantly, the [absolute] convergence of this integral is
independent of x because we have managed to bound K(t, x) by a function, k(t), that
has no x-dependence. We will discuss this last point in more detail at the end of this
section as it will be clearer once we understand how to establish that a function does
have dominated convergence.

198
6.2. The manipulation of improper integrals

Establishing dominated convergence

Now that we know what dominated convergence is, we need to see how to show that a
function K(t, x) has it for the appropriate values of t and x. Sometimes this can be
done because the x-dependence in K(t, x) can be easily ‘factored out’ to give us an
appropriate k(t) but, in other cases, we will need to use more sophisticated methods.
Let’s start with such an easy example and then we can move on to the more general
case.

Example 6.9 Show that the function K(t, x) = cos(tx) e−t has dominated
convergence for t ≥ 0 and c ≤ x ≤ d where c, d ∈ R.

In this case, the x-dependence of K(t, x) is wholly contained in the factor cos(tx)
and so, as | cos(tx)| ≤ 1 for all t, x ∈ R, we can easily see that

|K(t, x)| = | cos(tx) e−t | ≤ e−t .

Therefore, if we take k(t) = e−t we have a function of t only which satisfies the
dominance condition. Furthermore, we can easily see that the integral 6
Z ∞ Z ∞  −t ∞
−t e 0−1
k(t) dt = e dt = = = 1,
0 0 −1 0 −1

and so it is convergent, i.e. this choice of k(t) also satisfies the convergence
condition. That is, we have shown that K(t, x) has dominated convergence for t ≥ 0
and c ≤ x ≤ d where c, d ∈ R.4

Activity 6.8 Suppose that, for t ≥ 0, p(t) is a positive continuous function and
that the integral Z ∞
p(t) dt,
0

is convergent. Show that the function p(t) e−xt has dominated convergence for t ≥ 0
and x ≥ 0.

However, more generally, the x-dependence in K(t, x) will not be so easy to deal with
and so it will be harder to find an appropriate k(t). In such cases, we observe that in
order to satisfy the dominance condition, we need to be sure that for all t ≥ a we have

|K(t, x)| ≤ k(t),

for all c ≤ x ≤ d. As such, for each fixed value of t ≥ a, it would make sense to find the
value of x ∈ [c, d], let’s call it x∗ (t), that maximises the function |K(t, x)| and, if we do
this, as |K(t, x∗ (t))| is the maximum value of |K(t, x)| for each fixed t ≥ a, we must
have
|K(t, x)| ≤ |K(t, x∗ (t))| for t ≥ a.
4
Notice that we actually have dominated convergence for any x ∈ R here and so, in particular, we can
certainly guarantee that we have dominated convergence for any value of x such that c ≤ x ≤ d where
c > d are any two real numbers.

199
6. Manipulation of integrals

Indeed, notice that x∗ (t) can not depend on x but, generally speaking, there will be
some explicit dependence on t. Having done this, if we set k(t) = |K(t, x∗ (t))|, we will
have satisfied the dominance condition and, furthermore, if this k(t) satisfies the
convergence condition as well, then we will have established dominated convergence for
the given values of x and t.
Consequently, if we bear this in mind, then the problem of finding x∗ (t), i.e. the value of
x ∈ [c, d] that maximises |K(t, x)| for fixed values of t ≥ a, is simply a one-variable
constrained optimisation problem such as the ones we saw in Section 4.5.1 of 174
Calculus.5 This means that one of three things can happen, the two obvious cases are as
follows:

For fixed t ≥ a and any c ≤ x ≤ d, we find that |K(t, x)| is a decreasing function of
x, e.g. we may find that

|K(t, x)| < 0,
∂x
in which case |K(t, x)| is maximised at x = c, i.e. the left-hand endpoint of the
interval [c, d], giving us a maximum value of K(t, c).
6 For fixed t ≥ a and any c ≤ x ≤ d, we find that |K(t, x)| is an increasing function
of x, e.g. we may find that

|K(t, x)| > 0,
∂x
in which case |K(t, x)| is maximised at x = d, i.e. the right-hand endpoint of the
interval [c, d], giving us a maximum value of K(t, d).
Of course, the third possibility is where, for fixed t ≥ a, the function |K(t, x)| is neither
an increasing nor a decreasing function of x for all values of x in the interval [c, d]. In
this case, we should expect to find that the maximum value of |K(t, x)| will either
occur at one of the endpoints or at a value of x given by a stationary point which is
inside the interval, i.e. a value of x ∈ (c, d) which gives us


|K(t, x)| = 0.
∂x
Of course, in this case we can use what we saw in Section 4.5.1 of 174 Calculus to
decide which of these values of x does indeed maximise |K(t, x)|.
Let’s look at a couple of examples to see how this works in practice.

Example 6.10 Show that the function K(t, x) = t e−t/x has dominated convergence
for t ≥ 0 and 0 < c ≤ x ≤ d where c, d ∈ R.

Here |K(t, x)| = t e−t/x as this function is non-negative for the given values of t and
x. Furthermore, for any fixed value of t > 0, we can use the chain rule to see that

t2 −t/x
   
∂ ∂ −t/x t −t/x
|K(t, x)| = te =t e = e > 0,
∂x ∂x x2 x2
5
Obviously, it is a constrained optimisation problem because we are only interested in values of x that
satisfy the constraint c ≤ x ≤ d.

200
6.2. The manipulation of improper integrals

for all values of x > 0. Thus, for 0 < c < x < d, |K(t, x)| is an increasing function of
x which means that we can guarantee that

|K(t, x)| ≤ |K(t, d)| =⇒ |K(t, x)| ≤ |t e−t/d |,

and so, as we know that the function t e−t/d is non-negative, we can take
k(t) = t e−t/d in order to satisfy the dominance condition. Then, using integration by
parts (see Activity 6.8) we can see that, if 0 < c < x < d, we have d > 0 and so
Z ∞ Z ∞
k(t) dt = t e−t/d dt = d2 ,
0 0

which means that this integral is convergent, i.e. this choice of k(t) also satisfies the
convergence condition. Consequently, we have shown that K(t, x) has dominated
convergence for t ≥ 0 and 0 < c ≤ x ≤ d where c, d ∈ R.

Activity 6.9 Following on from Example 6.10, use integration by parts to show that
Z ∞
t e−t/d dt = d2 ,
0
6
when d > 0.

Activity 6.10 Following on from Example 6.10, explain why taking k(t) = t
satisfies the dominance condition but does not establish the required dominated
convergence.

Example 6.11 Show that the function


x
K(t, x) = ,
1 + x2 t4
has dominated convergence for t > 1 and 0 ≤ x ≤ 1.

Here we have
x
|K(t, x)| = ,
1 + x2 t 4
as this function is non-negative for the given values of t and x. Furthermore, for any
fixed value of t ≥ 1, we can use the quotient rule to see that

(1)(1 + x2 t4 ) − (x)(2xt4 ) 1 − x2 t4
 
∂ ∂ x
|K(t, x)| = = = ,
∂x ∂x 1 + x2 t4 (1 + x2 t4 )2 (1 + x2 t4 )2

and this is zero when x2 t4 = 1. In particular, for fixed t ≥ 1, the function of x given
by |K(t, x)| has a stationary point when x = 1/t2 which is in the interval [0, 1] that
we are interested in.6 What is the nature of this stationary point? Well, given that
t ≥ 1, we can observe that

t4
  
∂ 1 1
|K(t, x)| = −x +x ,
∂x (1 + x2 t4 )2 t2 t2

201
6. Manipulation of integrals

and, for 0 < x < 1/t2 this is positive whereas for 1/t2 < x < 1 it is negative. That is,
using the first-derivative test, x = 1/t2 must be giving us the maximum value of
|K(t, x)|, i.e. we have

1/t2
 
1 1
|K(t, x)| ≤ K t, 2 =

2 2 4
= 2,
t 1 + (1/t ) t 2t

and so, we can take k(t) = 1/(2t2 ) in order to satisfy the dominance condition.
Then, as we also have
Z ∞ Z ∞  ∞
1 1 1
k(t) dt = 2
dt = − = ,
1 1 2t 2t 1 2

we can see that this integral is convergent, i.e. this choice of k(t) also satisfies the
convergence condition. Consequently, we have shown that K(t, x) has dominated
convergence for t > 1 and 0 ≤ x ≤ 1.

Why do we need dominated convergence?


6
Now, returning to our discussion at the beginning of this section, although dominated
convergence for the values of x that we are considering guarantees that the integral
Z ∞
I(x) = K(t, x) dt,
a

is convergent for all x ∈ [c, d], it is actually a stronger condition on K(t, x) as it


establishes more than this. In particular, there are situations where I(x) converges for
all x ∈ [c, d] but K(t, x) does not have dominated convergence. The difference here is
between the following two observations concerning the ‘tails’ of these improper integrals
of the first kind.

The function K(t, x) is such that the integral I(x) converges for all x ∈ [c, d] which
means that, for any ε > 0, there is a T > 0 such that
Z ∞
K(t, x) dt < ε,
T

where the value of T we need to choose will generally depend on the value of x
because of the x-dependence in the integrand.

The function K(t, x) has dominated convergence for x ∈ [c, d] and so, for any ε > 0,
there is a T > 0 such that
Z ∞ Z ∞
K(t, x) dt ≤ k(t) dt < ε,
T T

where the value of T we need to choose is now independent of x as we have


managed to ‘remove’ the x-dependence in the integrand.
6
Of course, there is another stationary point when x = −1/t2 , but for t ≥ 1, this point is not in the
interval [0, 1] that we are interested in and so we can ignore it here.

202
6.2. The manipulation of improper integrals

Consequently, whatever we are doing with the x-dependence in K(t, x) when we


manipulate integrals like I(x), we need to be sure that it is not affecting the
convergence of the integral and this means that we need to be sure that it is not
affecting the behaviour of these ‘tails’. That is, we need K(t, x) to have dominated
convergence because then we know that the behaviour of the ‘tails’ will be unaffected
simply because their behaviour can now be guaranteed in a way that is independent of
x and, indeed, however we may be manipulating it.
Let’s now look at an example that draws out the differences that we have been
discussing.

Example 6.12 Consider the function


x
K(t, x) = ,
1 + x2 t2
for t ≥ 1 and x ∈ R. Show that the integral
Z ∞
I(x) = K(t, x) dt,
1
6
is convergent for all −1 ≤ x ≤ 1 but that the function K(t, x) does not have
dominated convergence for t ≥ 1 and −1 ≤ x ≤ 1.

Using the expression for K(t, x), we clearly have


Z ∞
I(0) = 0 dt = 0,
1

and, for x 6= 0, we have


Z ∞ ∞
1 ∞

x 1 1 π
Z
−1
I(x) = 2 2
dt = 2 2
dt = x tan (xt) = − tan−1 x,
1 1+x t x 1 (1/x) + t x 1 2

which is finite for all x 6= 0. That is, we have shown that the integral I(x) is
convergent for all −1 ≤ x ≤ 1.7

However, for −1 ≤ x ≤ 1, we notice that |K(t, x)| is an even function of x because



−x x
|K(t, −x)| = = = |K(t, x)|,
1 + (−x)2 t2 1 + x2 t2

and so, when looking for the maximum value of |K(t, x)| for −1 < x < 1 it suffices
to consider the case where 0 ≤ x ≤ 1. In particular, for t ≥ 1 and 0 ≤ x ≤ 1, we have
x
|K(t, x)| = ,
1 + x2 t 2
as this function is non-negative for these values of t and x. Furthermore, for any
fixed value of t ≥ 1, we can use the quotient rule to see that

(1)(1 + x2 t2 ) − (x)(2xt2 ) 1 − x2 t2
 
∂ ∂ x
|K(t, x)| = = = ,
∂x ∂x 1 + x2 t2 (1 + x2 t2 )2 (1 + x2 t2 )2

203
6. Manipulation of integrals

and this is zero when x2 t2 = 1. Consequently, for fixed t ≥ 1, the function of x given
by |K(t, x)| has a stationary point when x = 1/t which is in the interval [0, 1] that
we are interested in.8 What is the nature of this stationary point? Well, given that
t ≥ 1, we can observe that

t2
  
∂ 1 1
|K(t, x)| = −x +x ,
∂x (1 + x2 t2 )2 t t

and, for 0 < x < 1/t this is positive whereas for 1/t < x < 1 it is negative. That is,
using the first-derivative test, x = 1/t must be giving us the maximum value of
|K(t, x)|, i.e. we have
 
1 1/t 1
|K(t, x)| ≤ K t, = 2 2 2
= ,
t 1 + (1/t ) t 2t

for 0 ≤ x ≤ 1. However, as observed above, |K(t, x)| is an even function of x, and so


we actually have
1
|K(t, x)| ≤ ,
2t
6 for all −1 ≤ x ≤ 1. In particular, we see that any function k(t) that will satisfy the
dominance condition must satisfy the inequality
1
|K(t, x)| ≤ ≤ k(t),
2t
as, otherwise, we would have values of x, most notably around x = 1/(2t), for which
|K(t, x)| > k(t). Now, if that is the case, we find that any function k(t) that satisfies
the dominance condition gives us
1
k(t) ≥ ,
2t
for t ≥ 1, but we also know that the integral
Z ∞
dt
,
1 2t
is divergent and so, by the DCT, we see that
Z ∞
k(t) dt is divergent.
1

Thus, any k(t) that satisfies the dominance condition can not satisfy the convergence
condition and so K(t, x) does not have dominated convergence for t ≥ 1 and
−1 ≤ x ≤ 1.

We will see why this example is important when it comes to manipulating improper
integrals once we have looked at the rules themselves.
7
In fact, as we can clearly see, this I(x) is finite, and hence convergent, for all x ∈ R.
8
Of course, there is another stationary point when x = −1/t, but for t ≥ 1, this point is not in the
interval [0, 1] that we are interested in and so we can ignore it here.

204
6.2. The manipulation of improper integrals

Activity 6.11 Given any ε > 0, show that:

(i) Following on from Example 6.11, for 0 ≤ x ≤ 1,


Z ∞
x 1
2 4
dt < ε if T > .
T 1+x t 2ε

(ii) Following on from Example 6.12, for x > 0,


Z ∞
x cot ε
2 2
dt < ε if and only if T > .
T 1+x t x

Comment on your results in light of the discussion above.

6.2.2 The manipulation rules for improper integrals


Now that we understand dominated convergence, we can present the rules for
manipulating improper integrals of the first kind. In particular, we will present the
three rules and a simple example to motivate their use.
6
The first rule allows us to manipulate such improper integrals by taking limits through
the integral sign as follows.

Rule 1*: If F (t, x) is jointly continuous and has dominated convergence at all
points (t, x) where t ≥ a and c ≤ x ≤ d, then
Z ∞ Z ∞
lim F (t, x) dt = lim F (t, x) dt,
x→x0 a a x→x0

when c < x0 < d.


Notice that the joint continuity of F (t, x) guarantees that nothing peculiar is happening
at finite values of t (as we saw in Rule 1) and the dominated convergence guarantees
that nothing peculiar is happening as t → ∞ as we know that this behaviour is
independent of x. Of course, the joint continuity of F (t, x) also allows us to conclude
that we will get
lim F (t, x) = F (t, x0 ),
x→x0
when we look at the integrand on the right-hand side of this rule.
The second rule allows us to manipulate improper integrals by taking derivatives
through the integral sign as follows.

Rule 2*: Suppose that F (t, x) and ∂F/∂x are jointly continuous at all points
(t, x) where t ≥ a and c ≤ x ≤ d. If the integral
Z b
F (t, x) dt,
a
is convergent for all c ≤ x ≤ d and ∂F/∂x has dominated convergence for all
points (t, x) where t ≥ a and c ≤ x ≤ d, then
Z ∞  Z ∞
d ∂F
F (t, x) dt
= dt,
dx a x=x0 a ∂x x=x0

205
6. Manipulation of integrals

when c < x0 < d.


Notice that the joint continuity of F (t, x) and ∂F/∂x guarantees that nothing peculiar
is happening at finite values of t (as we saw in Rule 2) and the convergence of the
integrals on the left-hand side guarantees that we have something to differentiate. We
then need the dominated convergence of ∂F/∂x to guarantee that nothing peculiar is
happening as t → ∞ as we know that this behaviour will be independent of x.9 Of
course, overall, provided that the relevant conditions are satisfied, this rule will work for
any x = x0 where c < x0 < d.
The third rule allows us to manipulate proper integrals by taking integrals through the
integral sign as follows.

Rule 3*: If F (t, x) is jointly continuous and has dominated convergence at all
points (t, x) where t ≥ a and c ≤ x ≤ d, then
Z d Z ∞  Z ∞ Z d 
F (t, x) dt dx = F (t, x) dx dt.
c a a c

Notice that the joint continuity of F (t, x) guarantees that nothing peculiar is happening
6 at finite values of t (as we saw in Rule 3) and the dominated convergence guarantees
that nothing peculiar is happening when t → ∞ as we know that this behaviour will be
independent of x.
Let’s now look at these rules in action by considering how they can be used.

6.2.3 Using the rules for manipulating improper integrals


Unsurprisingly, one of the main uses of the rules for manipulating improper integrals,
especially Rule 2*, is that they also allow us to find improper integrals in cases where
our traditional techniques may not be of much use. Let’s consider two examples to see
how this works.

Example 6.13 As you will see in Activity 6.12, the function


Z ∞
sin t −xt
I(x) = e dt,
0 t
is well-defined for all x > 0 and I(x) → 0 as x → ∞. Find I 0 (x) and, by using
integration by parts, show that
1
I 0 (x) = − ,
1 + x2
for x > 0. Hence find I(x).

Given that I(x) is well-defined for all x > 0, we see that


Z ∞ Z ∞   Z ∞
0 d sin t −xt ∂ sin t −xt
I (x) = e dt = e dt = (− sin t) e−xt dt,
dx 0 t 0 ∂x t 0
9
The worry being, of course, that partially differentiating the jointly continuous function F (t, x) with
respect to x may produce a function, ∂F/∂x, which doesn’t have either the joint continuity or the
dominated convergence that is needed for this rule to work!

206
6.2. The manipulation of improper integrals

provided that this manipulation is justified. To see that it is, we first note that the
function
K(t, x) = (− sin t) e−xt ,
is clearly jointly continuous for t ≥ 0 and x > 0. And, secondly, we see that:

As | sin t| ≤ 1 for all t ∈ R and e−xt ≤ e−ct for x > c > 0, we have
|K(t, x)| = | sin t| e−xt ≤ e−ct = k(t),
for t ≥ 0 and x > c > 0.
And, for such a c > 0, we have
Z ∞ Z ∞  −ct ∞
−ct e 0−1 1
k(t) dt = e dt = = = ,
0 0 −c 0 −c c
which means that this integral is convergent.
Thus, as any x > 0 allows us to find a c ∈ R such that 0 < c < x, this establishes
that |K(t, x)| has dominated convergence for all t ≥ 0 and all x > 0.10

Having shown that


0
Z ∞ 6
I (x) = (− sin t) e−xt dt,
0
we can use integration by parts twice to see that, for x > 0, we have
 ∞ Z ∞
0 −xt
I (x) = (cos t) e − (−x cos t) e−xt dt
0
0 ∞ Z ∞ 
−xt −xt
= [0 − 1] + x (sin t) e − (−x sin t) e dt
0 0
 Z ∞ 
−xt
= −1 + x [0 − 0] − x (− sin t) e dt
0
0 2 0
∴ I (x) = −1 − x I (x),
if we use our expression for I 0 (x) again. Consequently, if we rearrange this, we find
that
1
(1 + x2 )I 0 (x) = −1 =⇒ I 0 (x) = − ,
1 + x2
for x > 0, as required.

In particular, if we integrate both sides of this expression, we see that


dx
Z Z
0
I (x) dx = − =⇒ I(x) = − tan−1 x + c,
1 + x2
which tells us what I(x) is up to some arbitrary constant, c. However, given that
I(x) → 0 as x → ∞, we also have
 
−1 π π
lim I(x) = lim − tan x + c =⇒ 0 = − + c =⇒ c = ,
x→∞ x→∞ 2 2
which means that
π
I(x) = − tan−1 x,
2
is the sought after expression for I(x).

207
6. Manipulation of integrals

Activity 6.12 Following on from Example 6.13, show that I(x) is indeed
well-defined if x > 0 and that I(x) → 0 as x → ∞.

Example 6.14 Suppose that, for x ∈ R, we have the function


Z ∞
1 2
f (x) = cos(tx) e− 2 t dt.
0

Find f 0 (x) and use integration by parts to show that f 0 (x) = −xf (x). Hence, using
the result from Example 5.15, find f (x).

Given that f (x) exists for x ∈ R (since otherwise, it would not be defined), we note
that f 0 (x) is given by
Z ∞ Z ∞ Z ∞
d − 12 t2 ∂  − 12 t2
 1 2
cos(tx) e dt = cos(tx) e dt = (−t) sin(tx) e− 2 t dt,
dx 0 0 ∂x 0

provided that this manipulation is justified. To see that it is, we note that the
function
6 1 2
K(t, x) = −t sin(tx) e− 2 t ,
is clearly jointly continuous for t ≥ 0 and x ≥ 0. And, moreover, it has dominated
convergence for t ≥ 0 and x ≥ 0 since | sin(tx)| ≤ 1 gives us
1 2 1 2
|K(t, x)| = | − t sin(tx) e− 2 t | ≤ t e− 2 t = k(t),

for these values of x and t as well as


Z ∞ Z ∞ " 1 2 #∞
1 2 e− 2 t 0−1
k(t) dt = t e− 2 t dt = = = 1,
0 0 −1 −1
0

which means that this integral is convergent. Thus K(t, x) has dominated
convergence for t ≥ 0 and any x ∈ R, i.e. for any c ≤ x ≤ d with c, d ∈ R.

Having shown that Z ∞


1 2
0
f (x) = (−t) sin(tx) e− 2 t dt,
0
we can use integration by parts to see that
h i∞ Z ∞ Z ∞
0 − 12 t2 − 21 t2 1 2
f (x) = sin(tx) e − x cos(tx) e dt = −x cos(tx) e− 2 t dt = −xf (x),
0 0 0
10 −tx
Observe that, in particular, if we were to use the fact that e ≤ 1 for all x > 0, we might have
chosen k(t) to be such that

|K(t, x)| = | sin t| e−xt ≤ | sin t| = k(t) or even |K(t, x)| = | sin t| e−xt ≤ 1 = k(t).

But, although both of these choices for k(t) satisfy the dominance condition, neither of them give us a
finite value when we look at the integral Z ∞
k(t) dt,
0

and so they do not satisfy the convergence condition. That is, neither of these choices for k(t) can
establish the dominated convergence of |K(t, x)|.

208
6.2. The manipulation of improper integrals

as the term in square brackets is zero. This is a separable first-order differential


equation, which we can solve by writing

df x2
Z Z
1 2
= − x dx =⇒ ln |f (x)| = − + c =⇒ f (x) = A e− 2 x ,
f 2
for some arbitrary constant A ∈ R. In particular, we can see that
Z ∞ r
− 12 t2 π
A = f (0) = e dt = ,
0 2
using our result from Example 5.15 and so we find that
r
π − 1 x2
f (x) = e 2 ,
2

is the sought after expression for f (x).

To finish our discussion of the rules for manipulating improper integrals and as a
warning against using them without due consideration of the conditions under which 6
they are valid, we now look at a simple example where Rule 1* fails.

Example 6.15 Following on from Example 6.12, where we had the function
Z ∞
x π
I(x) = 2 2
dt = − tan−1 x,
1 1+x t 2
for 0 < x ≤ 1, show that
∞ ∞  
x x
Z Z
lim dt 6= lim dt,
x→0 1 1 + x2 t2 1 x→0 1 + x2 t2

and conclude that Rule 1* fails here. Comment on this failure in light of the
discussion in Example 6.12 and Activity 6.11.

We see can see straightaway that, for 0 < x ≤ 1, we have


Z ∞
x π
−1
 π
lim I(x) = lim dt = lim − tan x = ,
x→0 x→0 1 1 + x2 t2 x→0 2 2
whereas ∞   ∞
x
Z Z
lim dt = 0 dt = 0,
1 x→0 1 + x2 t2 0

and so we find that Rule 1* fails here because


Z ∞ Z ∞  
x x
lim dt 6= lim dt,
x→0 1 1 + x2 t2 1 x→0 1 + x2 t2

as required.

Of course, we should expect Rule 1* to fail here because, as we saw in Example 6.12,
we don’t have the dominated convergence that is required for its application when

209
6. Manipulation of integrals

we have −1 ≤ x ≤ 1.11 In particular, using Activity 6.11(ii), we know that for any
ε > 0 and x > 0, we have
Z ∞
x cot ε
2 2
dt < ε if and only if T > ,
T 1+x t x
which means that when we try to manipulate the integral by taking the limit as
x → 0 we can not guarantee that the behaviour of the ‘tails’ is unaffected because
their behaviour depends on x.12 Consequently, it is this x-dependence in the
behaviour of the ‘tails’ that causes Rule 1* to fail here and, because it guarantees
that no such x-dependence occurs, this is why dominated convergence is so
important in such cases.

Learning outcomes
At the end of this chapter and having completed the relevant reading and activities, you
should be able to:
6
establish that a function of two variables is jointly continuous;
manipulate proper integrals and justify the validity of such manipulations;
use the Leibniz rule for differentiating integrals and justify its use;
establish that a function of two variables has dominated convergence;
manipulate improper integrals and justify the validity of such manipulations;
find proper and improper integrals by using appropriately justified manipulations.

Solutions to activities
Solution to activity 6.1
Just as we saw in Example 6.2, we use the fact that

tx = ex ln t ,

and, using Theorems 6.1 and 6.2, we note that:

As x and ln t are continuous functions of one variable for all x ∈ R and t > 0
respectively, we can treat them both as jointly continuous functions of the two
variables x and t for all points (t, x) where x ∈ R and t > 0.
As the product of two jointly continuous functions, the function of two variables
x ln t is also jointly continuous for all points (t, x) where x ∈ R and t > 0.
11
Or, indeed, when we have any interval c ≤ x ≤ d which contains x = 0.
12
That is, as x → 0+ , we see that we must take arbitrarily large values of T to be sure that the
contribution of the ‘tails’ to the value of I(x) is less than the ε > 0 that we have chosen.

210
6.2. Solutions to activities

As the function of one variable ey is continuous for all y ∈ R, we can take y = x ln t


and so that the composition ex ln t = tx is jointly continuous for all points (t, x)
where x ∈ R and t > 0.
This establishes that the function tx is jointly continuous at all points (t, x) where
x ∈ R and t > 0, as required.

Solution to activity 6.2


Doing the required calculations, we see that on the left-hand side of the equality we have
Z 2 2
t2

t 2
lim dt = lim = lim = 1,
x→1 0 x + 1 x→1 2(x + 1) x→1 x + 1
0

whereas on the right-hand side of the equality we have


Z 2   Z 2  2 2
t t t
lim dt = dt = = 1,
0 x→1 x+1 0 2 4 0
and these are indeed equal in accordance with Rule 1. In particular, as required by this
rule, notice that the function
t 6
F (t, x) = ,
x+1
is jointly continuous at all points (t, x) in the rectangle [0, 2] × [c, d] where we have, say,
−1 < c < x0 < d with x0 = 1.

Solution to activity 6.3


Doing the required calculations for any x > −1, we see that on the left-hand side of the
equality we have
Z 2 2
t2
  
d t d d 2 2
dt = = =− ,
dx 0 x + 1 dx 2(x + 1) 0 dx x + 1 (x + 1)2
whereas on the right-hand side of the equality we have
Z 2 Z 2 2
−t −t2
  
∂ t 2
dt = 2
dt = 2
=− ,
0 ∂x x+1 0 (x + 1) 2(x + 1) 0 (x + 1)2
and these are indeed equal in accordance with Rule 2. In particular, as required by this
rule, notice that the functions
t ∂F −t
F (t, x) = and = ,
x+1 ∂x (x + 1)2
are jointly continuous at all points (t, x) in the rectangle [0, 2] × [c, d] where we have,
say, −1 < c < x < d.

Solution to activity 6.4


In Example 6.6, we saw that when t 6= 0, we have
Z 1
−xt et − e−t
g(t) = f (t) e dx = f (t) ,
x=−1 t

211
6. Manipulation of integrals

whereas, when t = 0, we have


Z 1  1
f (0) dx = xf (0) = 2f (0),
x=−1 x=−1

and these two results tell us what g(t) is for all t ∈ R. Moreover, we see that g(t) is
clearly continuous for all t 6= 0 and so we only need to consider what is happening
continuity-wise at t = 0. But, using L’Hôpital’s rule we see that
et − e−t et + e−t
lim= lim = 2,
t→0 t t→0 1
because the numerator and the denominator both tend to zero as t → 0 and,
furthermore, we can be sure that
lim f (t) = f (0),
t→0

because f (t) is continuous for all t ∈ R so, taking these two facts together, we see that
et − e−t   et − e−t

lim g(t) = lim f (t) = lim f (t) lim = 2f (0) = g(0),
t→0 t→0 t t→0 t→0+ t
6 i.e. g(t) is continuous at t = 0 too. Consequently, g(t) is continuous for all t ∈ R.

We now see that, if we take a little more care, the answer in Example 6.6 should be
Z 1 Z T  Z T
−xt
f (t) e dt dx = g(t) dt,
x=−1 t=0 t=0

and, as g(t) is continuous for all t ∈ [0, T ], this is not an improper integral of the second
kind. Indeed, as we now know that g(t) is continuous, we can see that its value at t = 0
will contribute nothing to the value of the integral (as it will contribute no ‘area’) and
so we can conclude that we have
Z 1 Z T Z T
et − e−t

−xt
f (t) e dt dx = f (t) dt,
x=−1 t=0 t=0 t
in agreement with what we said in Example 6.6.

Solution to activity 6.5


Doing the required calculations for any M > 1, we see that on the left-hand side of the
equality we have
Z M Z 2 Z M 2 Z M
t2

t 2
dt dx = dx = dx
x=1 t=0 x + 1 x=1 2(x + 1) t=0 x=1 x + 1
 M  
M +1
= 2 ln(x + 1) = 2 ln ,
x=1 2
whereas on the right-hand side of the equality we have
Z 2 Z M  Z 2   Z 2  
t M +1
dx dt = t ln(x + 1) M dt = t ln dt
t=0 x=1 x + 1 t=0 x=1 t=0 2
 2  2  
t M +1 M +1
= ln = 2 ln ,
2 2 t=0 2

212
6.2. Solutions to activities

and these are indeed equal in accordance with Rule 3. In particular, as required by this
rule, notice that the function
t
F (t, x) = ,
x+1
is jointly continuous at all points (t, x) in the rectangle [0, 2] × [1, M ] for any M > 1.

Solution to activity 6.6


The substitution u = tan 2t gives us

2du
du = 12 sec2 2t dt = 1
1 + tan2 t

2 2
dt =⇒ dt = ,
1 + u2
and we also have
cos t cos2 2t − sin2 t
2
1 − tan2 t
2 1 − u2
cos t = = = t = ,
1 cos2 2t + sin2 t
2
1 + tan2 2
1 + u2

so we see that the denominator of the integrand can be written as

1 + x cos t = 1 + x
1 − u2
=
(1 + x) + (1 − x)u2
.
6
1 + u2 1 + u2
Consequently, we find that

2
Z
0
I (x) = du
0 (1 + x) + (1 − x)u2
Z ∞
2 1
= 1+x du
1 − x 0 1−x + u2
"r r !#∞
2 1−x 1 − x
= tan−1 u
1−x 1+x 1+x
0
"r #
2 1 − x π 
= −0
1−x 1+x 2
π
∴ I 0 (x) = √ ,
1 − x2
as required.

Solution to activity 6.7


In Example 6.8, we found that
x2 x2
d t
Z Z
0 3 2
f (x) = ln(1 + xt) dt = 2x ln(1 + x ) − ln(1 + x ) + dt,
dx x x 1 + xt

for x > 0 and, to find f 0 (x) explicitly, all we need to do now is note that as

1 1 + xt − 1
     
t 1 xt 1 1
= = = 1− ,
1 + xt x 1 + xt x 1 + xt x 1 + xt

213
6. Manipulation of integrals

we have
x2 x2  
t 1 1
Z Z
dt = 1− dt
x 1 + xt x x 1 + xt
 x2
1 1
= t − ln |1 + xt|
x x x
   
1 2 1 3 1 2
= x − ln |1 + x | − x − ln |1 + x |
x x x
1 1 + x3

= x − 1 − 2 ln ,
x 1 + x2
so that we find that
1 + x3
 
0 3 1 2
f (x) = 2x ln(1 + x ) − ln(1 + x ) + x − 1 − 2 ln ,
x 1 + x2
as x > 0. Indeed, as you should expect, this still gives us f 0 (1) = ln 2.

6 Solution to activity 6.8


Given that, for t ≥ 0, p(t) is a positive continuous function and that the integral
Z ∞
p(t) dt,
0

is convergent, we can see that the function p(t) e−xt has dominated convergence for
t ≥ 0 and x ≥ 0 as follows.

For t ≥ 0 and x ≥ 0 we have e−xt ≤ 1 and so


|p(t) e−xt | ≤ p(t),
as p(t) is positive. So, if we take k(t) = p(t), we have a function of t only that
satisfies the dominance condition.
We are given that the integral
Z ∞ Z ∞
k(t) dt = p(t) dt,
0 0

is convergent and so this choice of k(t) also satisfies the convergence condition.
Consequently, we have shown that the function p(t) e−xt has dominated convergence for
t ≥ 0 and x ≥ 0 as required.

Solution to activity 6.9


Following on from Example 6.10, we have d > 0 and so, using integration by parts, we
get
Z ∞  −t/d ∞ Z ∞  −t/d ∞
−t/d e e−t/d e 0−1
te dt = t − (1) dt = [0 − 0] − 2
=− 2
= d2 ,
0 −1/d 0 0 −1/d (−1/d) 0 (−1/d)
as required.

214
6.2. Solutions to activities

Solution to activity 6.10


Following on from Example 6.10, we see that given t ≥ 0 and 0 < c ≤ x ≤ d for c, d ∈ R,
we have e−t/x ≤ 1 and so
|K(t, x)| = t e−t/x ≤ t,
which means that if we take k(t) = t we have satisfied the dominance condition.
However, the integral
Z ∞ Z ∞  2 ∞
t
k(t) dt = t dt = = ∞,
0 0 2 0

and so it is divergent. That is, this choice of k(t) doesn’t satisfy the convergence
condition and so it can’t be used to establish the required dominated convergence.

Solution to activity 6.11


For (i), we can see from Example 6.11 that when 0 ≤ x ≤ 1, we have
x 1
0≤ 2 4
≤ 2,
1+x t 2t
for all t ≥ 1 and so this means that we have
6
Z ∞ Z ∞  ∞  
x dt 1 1 1
2 4
dt ≤ 2
= − =0− − = .
T 1+x t T 2t 2t T 2T 2T
That is, given any ε > 0, we have

1 1 x
Z
T > =⇒ ε> ≥ dt,
2ε 2T T 1 + x2 t4
as required.

For (ii), we can see from Example 6.12 that when x > 0, we have
Z ∞
x ∞ π
dt = tan−1 (xt) T = − tan−1 (xT ),

2
1+x t 2 2
T

and, for ε > 0, this gives us


Z ∞
x π −1 −1 π π 
dt = −tan (xT ) < ε ⇐⇒ tan (xT ) > −ε ⇐⇒ xT > tan − ε ,
T 1 + x2 t2 2 2 2
so, as x > 0 and tan( π2 − ε) = cot ε, this holds if and only if
cot ε
T > ,
x
as required.

In these two examples, we see that the given integrals are convergent for the stated
values of x but the behaviour of the ‘tails’ — i.e. what values of T will make the value
of the integral less than ε — is independent of x in (i) whereas it is dependent on x in
(ii). This is what the dominated convergence of the integrand in (i) guarantees and it
also explains why the integrand in (ii) does not have dominated convergence.

215
6. Manipulation of integrals

Solution to activity 6.12


To see that the function in Example 6.13 is well-defined for all x > 0, we note that, for
all t ∈ R, we have
sin t
≤ 1 and so sin t e−xt ≤ e−xt ,


t t
for any fixed x > 0. Consequently, we can see that
Z ∞ Z ∞ Z ∞  −xt ∞
sin t −xt sin t −xt −xt e 1
e dt ≤ e dt ≤
e dt = = ,

0 t 0
t 0 −x 0 x

and so, as |I(x)| ≤ 1/x for all x > 0, it exists. In particular, we can see that this gives us
1 1
− ≤ I(x) ≤ ,
x x
and so, by the Sandwich theorem, I(x) → 0 as x → ∞.13

6 Exercises
Exercise 6.1
For x > 0, write down a function, K(t, x), such that

∂K sin2 t
= .
∂x (1 + x sin2 t)2
Z π/2
Use the substitution u = tan t to find K(t, x) dt.
0
Hence, for x > 0, find
π/2
sin2 t
Z
dt,
0 (1 + x sin2 t)2
justifying any manipulations you make.

Exercise 6.2
1
t2
Z
For 0 < x < 1, evaluate the integral dt.
0 (1 − xt2 )3/2

Exercise 6.3
Z x2
For x > 0, a function, f (x), is defined by f (x) = √
ln(1 + x2 t) dt.
x

Show that f 0 (1) = 32 ln 2.

13
Notice that we can’t use Rule 1* here because we have a limit as x → ∞ and not a limit as x → x0
for c < x0 < d.

216
6.2. Solutions to exercises

Exercise 6.4
Suppose that, for t ≥ 0, the continuous function p(t) is such that |p(t)| ≤ M eγt for some
constants M, γ ≥ 0.

Show that the function p(t) e−xt has dominated convergence for all t ≥ 0 and x > γ.

Exercise 6.5
Show that, for x > 0,

∞ ∞
d dt dt
Z Z
= −2x ,
dx 0 x + t2
2
0 (x2 + t2 )2

justifying any manipulations you make.


Z ∞
dt
Use this to find for any value of x > 0.
0 (x2 + t2 )2

6
Solutions to exercises
Solution to exercise 6.1
For x > 0, we can treat sin2 t as a constant and see that

−1
K(t, x) = ,
1 + x sin2 t

is a function14 such that


∂K sin2 t
= .
∂x (1 + x sin2 t)2

Then, we can find the integral

π/2 π/2
−1
Z Z
K(t, x) dt = dt,
0 0 1 + x sin2 t

by using the substitution u = tan t as directed. As such, we have

dt 1 sin2 t tan2 t u2
= and sin2 t = = = ,
du 1 + u2 sin2 t + cos2 t tan2 t + 1 u2 + 1

so we get

14
Of course, we could take this function ‘plus any arbitrary constant’ where here, because we are
partially differentiating with respect to x, that means we could add on any function, g(t), of t only.
However, we are only asked for one such function and so we make our life easier by taking g(t) = 0.

217
6. Manipulation of integrals

π/2 π/2
−1
Z Z
K(t, x) dt = dt
0 1 + x sin2 t
Z0 ∞
−1 du
= 2
0
u
1 + x 1+u 2
1 + u2

−1
Z
= du
0 1 + (1 + x)u2
∞
−1 √

−1
= √ tan (u 1 + x)
1+x 0
−π
= √ .
2 1+x
Therefore, for x > 0, we can use Rule 2 to see that
Z π/2 Z π/2 Z π/2
sin2 t −π
 
∂K d d π
2 2 dt = dt = K(t, x) dt = √ = ,
0 (1 + x sin t) 0 ∂x dx 0 dx 2 1 + x 4(1 + x)3/2
as long as we can justify this manipulation. However, we can see that this manipulation
6 is justified because the functions
−1 ∂K sin2 t
K(t, x) = and = ,
1 + x sin2 t ∂x (1 + x sin2 t)2
are both jointly continuous for 0 ≤ t ≤ π/2 and x > 0 since these values of t and x
guarantee that 1 + x sin2 t ≥ 1 > 0.

Solution to exercise 6.2


Given that 0 < x < 1, the key here is to spot that, using Rule 2, we have
Z 1 Z 1 Z 1
t2
 
∂ 2 d 2
2 3/2
dt = √ dt = √ dt,
0 (1 − xt ) 0 ∂x 1 − xt2 dx 0 1 − xt2
as long as we can justify this manipulation. However, we can see that this manipulation
is justified because the functions
2 ∂K t2
K(t, x) = √ and = ,
1 − xt2 ∂x (1 − xt2 )3/2
are both jointly continuous for 0 ≤ t ≤ 1 and 0 < x < 1 since these values of t and x
guarantee that 1 − xt2 > 0.

Now, √we can see that if, for fixed 0 < x < 1, we make the substitution u = xt so that
du = xdt, we have
Z 1 Z √x √x


2 2 1 2 2
√ dt = √ p dy = √ sin−1 y = √ sin−1 x,
0 1 − xt2 x 0 1 − y2 x 0 x

as sin−1 0 = 0. And, in turn, this means that if we use what we saw above for 0 < x < 1,
we have
Z 1 Z 1
t2 √ √
 
d 2 d 2 −1 1 −1 1
2 3/2
dt = √ dt = √ sin x = − 3/2
sin x+ √ ,
0 (1 − xt ) dx 0 dx x x x 1−x
1 − xt2

218
6.2. Solutions to exercises

where we have used the product and chain rules in the last step.

Solution to exercise 6.3


Given that Z x2
f (x) = √
ln(1 + x2 t) dt,
x
for x > 0, we can use the Leibniz rule for differentiating integrals to see that
  Z x2
0 4 1 5/2 2xt
f (x) = (2x) ln(1 + x ) − √ ln(1 + x ) + √ 2
dt,
2 x x 1+x t

and we can justify its use by noting that the functions x and x2 are differentiable for
x > 0 and that the functions
∂F 2xt
F (t, x) = ln(1 + x2 t) and = ,
∂x 1 + x2 t
are clearly jointly continuous as 1 + x2 t > 1 > 0 if x > 0 and t > 0.15

Indeed, with this we see that

0 1
Z 1
2xt 3 6
f (1) = 2 ln 2 − ln 2 + 2
dt = ln 2,
2 1 1+x t 2
as required.

Solution to exercise 6.4


Given that, for t ≥ 0, the continuous function p(t) is such that |p(t)| ≤ M eγt for some
constants M, γ ≥ 0, we can see that the function p(t) e−xt is such that
|p(t) e−xt | ≤ M eγt e−xt ≤ M eγt e−ct ,
for any c < x. That is, if we take k(t) = M e−(c−γ)t then we have satisfied the dominance
condition for p(t) e−xt when t ≥ 0 and 0 < c < x. Moreover, we see that the integral
Z ∞ Z ∞
k(t) dt = M e−(c−γ)t dt,
0 0

is convergent as long as c > γ. That is, we have satisfied the convergence condition too
as long as c > γ. Consequently, we can conclude that the function p(t) e−xt has
dominated convergence for t ≥ 0 and γ < x (since γ < c < x) as required.

Solution to exercise 6.5


For x > 0, we want to use Rule 2* to argue that
Z ∞ Z ∞ Z ∞ Z ∞
−2x
 
d dt ∂ 1 dt
2 2
= 2 2
dt = 2 2 2
dt = −2x ,
dx 0 x + t 0 ∂x x + t 0 (x + t ) 0 (x + t2 )2
2

and to justify that this manipulation is valid we need to look at the functions
1 ∂F −2x
F (t, x) = and = 2 ,
x2 + t2 ∂x (x + t2 )2
and see that:
15

As it must be if it takes values between x and x2 with x > 0.

219
6. Manipulation of integrals

They are both jointly continuous for t ≥ 0 and x > 0 because x2 + t2 > 0.

For all x > 0, the integral


Z ∞   ∞
dt 1 −1 t π
2 2
= tan = ,
0 x +t x x 0 2x

is convergent.

For any t ≥ 0 and 0 < c < x < d, we can see that



∂F −2x 2x 2d
∂x = (x2 + t2 )2 = (x2 + t2 )2 ≤ (c2 + t2 )2 ,

and so if we take this last expression to be k(t), we can also see that the integral
Z ∞ Z ∞
2d
k(t) dt = dt,
0 0 (c + t2 )2
2

is convergent by the LCT with g(t) = 1/t4 . That is, we have shown that the
6 function ∂F/∂x has dominated convergence for t ≥ 0 and x > 0.
Of course, having done this, we now see that
Z ∞ Z ∞
dt d dt d π π
−2x 2 2 2
= 2 2
= = − 2,
0 (x + t ) dx 0 x + t dx 2x 2x

and so we can conclude that


Z ∞
dt 1  π  π
2 2 2
= − − 2 = 3,
0 (x + t ) 2x 2x 4x

for any x > 0.

220
Chapter 7
Laplace transforms

Essential reading

(For full publication details, see Chapter 1.)

+ Ostaszewski (1991) Chapter 21.

Further reading

+ Wrede and Spiegel (2010) pp.314–316, 329–30 and 334–335.

Aims and objectives

The objectives of this chapter are:


7
to introduce Laplace transforms and establish their properties;

to see how Laplace transforms can be used in mathematics.


Specific learning outcomes can be found near the end of this chapter.

7.1 What is a Laplace transform?


This chapter will make use of much of the material that we have seen so far in this half
course. We start by defining the Laplace transform of a function and establish some of
its properties. We will also see how it can be used to solve second-order differential
equations with constant coefficients (cf. Section 8.3 of 174 Calculus) and investigate the
relationship between Laplace transforms and a special kind of integral known as a
convolution. In particular, we will see that this last relationship has many interesting
applications.

The definition of the Laplace transform

If f (t) is a continuous function for t ≥ 0, we define the Laplace transform of f (t) to be


the function Z ∞
˜
f (s) = f (t) e−st dt, (7.1)
0

for all values of s for which this integral exists. In particular, observe that the Laplace
transform takes one function, f (t), and returns another function, f˜(s), i.e. we can think

221
7. Laplace transforms

of it as an operation, L, that takes f (t) and returns another function f˜(s) where
L{f (t)}(s) = f˜(s),
which is read as ‘the Laplace transform of f (t) [in terms of s] is f˜(s)’.1 Indeed, it will be
convenient in what follows to use both of these notations for the Laplace transform and,
given that we will always be thinking of Laplace transforms as functions of s, we will
often omit the explicit dependence on this variable in L{f (t)}(s) and simply write it as
L{f (t)}.
To see how this works, let’s start by finding two simple Laplace transforms.

Example 7.1 Suppose that, for t ≥ 0, we have the functions f (t) = 1 and
g(t) = eat for some constant a. Find the Laplace transforms of these functions and
state the values of s for which they are defined.

For f (t) = 1, (7.1) gives us


Z ∞ ∞ ∞
e−st
  
1 1
Z
˜
f (s) = (1) e−st
dt = e −st
dt = =0− − = ,
0 0 −s 0 s s
as long as s > 0. Indeed, we need s > 0 to guarantee that
e−st
lim= 0,
t→∞ −s

7 so that the integral exists. In particular, this means that this Laplace transform only
exists for s > 0 and, of course, we could write this result as
1
L{1} = ,
s
if we wanted.

For g(t) = eat , (7.1) gives us


Z ∞ Z ∞  −(s−a)t ∞  
at −st −(s−a)t e 1 1
g̃(s) = e e dt = e dt = =0− − = ,
0 0 −(s − a) 0 s−a s−a
as long as s > a. Indeed, we need s > a to guarantee that
e−(s−a)t
lim = 0,
t→∞ −(s − a)

so that the integral exists. In particular, this means that this Laplace transform only
exists for s > a and, of course, we could write this result as
1
L{eat } = , (7.2)
s−a
if we wanted.
1
Recall that, so far, we have defined a function to be a rule which returns a unique real number for
each real number in its domain. However, with the introduction of the Laplace transform, we now have
two different notions of ‘function’. One refers to a continuous function on the interval [0, ∞), like f , that
takes numbers from the interval [0, ∞) and returns real numbers as we have seen before. The other is
a function, like L, that takes functions, like f (t), and returns other functions, like f˜(s). To distinguish
the latter use of the word ‘function’ from the former one, we will typically use the word transform or
operation when dealing with the latter, i.e. we will say that L is an operation and we call the result of
this operation the Laplace transform.

222
7.1. What is a Laplace transform?

Activity 7.1 Suppose that, for t ≥ 0, we have the function f (t) = t. Find the
Laplace transform of this function and state the values of s for which it is defined.

7.1.1 Some properties of the Laplace transform


We now investigate some useful properties of the Laplace transform and, to simplify our
discussion, we will restrict our attention to functions f (t) that are continuous for t ≥ 0
and of exponential growth at most γ, i.e. for t ≥ 0 we have

|f (t)| ≤ M eγt ,

for some M, γ ≥ 0. In particular, this means that if f (t) is such a function, we can
guarantee that its Laplace transform, i.e.
Z ∞
L {f (t)} = f (t) e−st dt,
0

will exist for all s > γ because, using direct comparison techniques, we see that the
integrand f (t) e−st satisfies

|f (t) e−st | ≤ |M eγt e−st | = M e−(s−γ)t ,

and the integral Z ∞ 7


−(s−γ)t
Me dt,
0
is convergent provided that s − γ > 0.

Laplace transforms are linear transformations

Suppose that, for t ≥ 0, we have two functions f (t) and g(t) whose Laplace transforms
Z ∞ Z ∞
−st
L{f (t)} = f (t) e dt and L{g(t)} = g(t) e−st dt,
0 0

both exist for s > γ for some γ ∈ R. Then, for any constants α, β ∈ R, the Laplace
transform of the function αf (t) + βg(t) also exists for s > γ and is given by

L{αf (t) + βg(t)} = αL{f (t)} + βL{g(t)}, (7.3)

i.e. the operation of taking Laplace transforms is a linear transformation.

Activity 7.2 Using the properties of the definite integral, show that Laplace
transforms are linear in this way.

The Laplace transform of a derivative

Suppose that, for t ≥ 0, we have a continuous function, f (t), of exponential growth at


most γ for some γ ≥ 0 so that its Laplace transform exists for s > γ. Then, provided

223
7. Laplace transforms

that this function is differentiable for t ≥ 0, we find that another useful property of the
Laplace transform is that

L {f 0 (t)} = sL {f (t)} − f (0), (7.4)

and this also exists for all s > γ. To see why, notice that using (7.1), we have
Z ∞ Z T
0 0 −st
L {f (t)} = f (t) e dt = lim f 0 (t) e−st dt,
0 T →∞ 0

using the definition of an improper integral of the first kind. Now, if we consider this
last integral and integrate it by parts, we get
Z T T
Z T
0 −st
dt = f (t) e−st 0 − f (t)(−s) e−st dt

f (t) e
0 0
Z T
= f (T ) e−sT −f (0) + s f (t) e−st dt.
0

But, of course, this means that as


Z T
0
L {f (t)} = lim f 0 (t) e−st dt,
T →∞ 0
7 we now have
 Z T 
0 −sT −st
L {f (t)} = lim f (T ) e −f (0) + s f (t) e dt
T →∞ 0
Z T
= lim f (T ) e−sT −f (0) + s lim f (t) e−st dt,
T →∞ T →∞ 0

provided that these two limits exist. However, we see that the second limit is just
Z T
lim f (t) e−st dt = L {f (t)} ,
T →∞ 0

which, by the assumption above exists for all s > γ and, since f (t) is of exponential
growth at most γ, we also have

|f (T ) e−sT | ≤ |M eγT e−sT | = M e−(s−γ)T ,

and so, for s > γ, we have

lim M e−(s−γ)T = 0 =⇒ lim f (T ) e−sT = 0,


T →∞ T →∞

by the Sandwich theorem.


We will see many applications of (7.4) in what follows, but a particularly nice one is that
it allows us to deduce the Laplace transform of the function tn for n ∈ N and t ≥ 0.

224
7.1. What is a Laplace transform?

Example 7.2 Suppose that, for t ≥ 0, we have the functions fn (t) = tn for
n = 1, 2, 3, . . .. Use (7.3) and (7.4) to find the Laplace transform of these functions
and state the values of s for which they are defined.

Taking fn (t) = tn for n = 1, 2, 3, . . ., we have fn (0) = 0 and fn0 (t) = ntn−1 . So,
applying (7.4), we find that
n
L{ntn−1 } = sL{tn } − 0 =⇒ L{tn } = L{tn−1 },
s
using (7.3). But, rearranging this, we see that
n
L{tn } = L{tn−1 },
s
and, by repeatedly applying this result, we get

n−1
  
n n
L{t } = L{tn−2 }
s s
n−1 n−2
   
n
= L{tn−3 }
s s s
= ···
n−1
    
n 2
= ··· L{t}.
s s s 7
So, using what we found in Activity 7.1, namely that L{t} = 1/s2 , we have

n−1
     
n n 2 1 n!
L{t } = ··· 2
= n+1 ,
s s s s s

as the Laplace transform of tn for n = 1, 2, 3, . . ..

Indeed, (7.4) can also be used to find the Laplace transform of higher-order derivatives.
For instance, you can see how to do this for second-order derivatives in the next
Activity.

Activity 7.3 Use (7.4) to show that, if f (t) is a function of exponential growth at
most γ for which f 00 (t) exists, then for s > γ, we have

L{f 00 (t)} = s2 L{f (t)} − sf (0) − f 0 (0). (7.5)

(Note: For fear of mixing up the second and third terms on the right-hand side of
this formula, it is best if you learn how to derive it like we do here instead of just
trying to memorise it.)

Using similar reasoning, we can find the Laplace transform of yet higher-order
derivatives. However, a useful application of (7.5) is that it enables us to find the
Laplace transform of the continuous functions sin(at) and cos(at), for some constant
a 6= 0, in a particularly nice way.

225
7. Laplace transforms

Example 7.3 Let f (t) = sin(at) for some constant a 6= 0. Use (7.3) and (7.5) to
find its Laplace transform. For which values of s does this exist?

Taking f (t) = sin(at), we have f 0 (t) = a cos(at) and f 00 (t) = −a2 sin(at) so, using
(7.5), we get
L −a2 sin(at) = s2 L {sin(at)} − s(0) − a,


as f (0) = 0 and f 0 (0) = a. Consequently, applying (7.3), we now have

−a2 L {sin(at)} = s2 L {sin(at)} − a =⇒ (s2 + a2 )L {sin(at)} = a,

and so we find that


a
L {sin(at)} = ,
+ a2 s2
is the Laplace transform of sin(at) for some constant a 6= 0. Of course, this is defined
for all s > 0 since the function sin(at) satisfies | sin(at)| ≤ 1 = e0t and so it is of
exponential growth at most zero.

Activity 7.4 Let f (t) = cos(at) for some constant a 6= 0. Use (7.4) and the result
from Example 7.3 to find its Laplace transform. For which values of s does this
exist?

7 Alternatively, we could have used the definition of the Laplace transform to find the
Laplace transform of the continuous functions sin(at) and cos(at) where a 6= 0 is a
constant. In particular, the next activity gives us a particularly nice way of doing this if
you can handle complex numbers.

Activity 7.5 Find the Laplace transform of the function f (t) = eiat for a ∈ R and,
using the fact that
eiat = cos(at) + i sin(at),
deduce the Laplace transforms of sin(at) and cos(at).

The Laplace transform of a definite integral

Suppose that f is a continuous function on the interval [0, t] and that the function,
F (t), is given by the definite integral
Z t
F (t) = f (u) du,
0

so that F (0) = 0. Using the FTC, we see that F 0 (t) = f (t) and so, using (7.4), we see
that
L {F 0 (t)} = sL {F (t)} − F (0) =⇒ L {f (t)} = sL {F (t)} − 0,
i.e. for s > 0, we have Z t 
1
L f (u) du = L {f (t)} . (7.6)
0 s
This Laplace transform will then exist as long as L {f (t)} does.

226
7.1. What is a Laplace transform?

The Laplace transform of a function multiplied by eat

Suppose that, for t ≥ 0, we have a continuous function, f (t), of exponential growth at


most γ for some γ ≥ 0 so that its Laplace transform exists for s > γ. Then, given that
L {f (t)} = f˜(s),
we find that another useful property of the Laplace transform is that
L f (t) eat = f˜(s − a),

(7.7)
and this exists for all s > a + γ. That is, multiplying a function by the exponential eat
‘shifts’ its Laplace transform by −a. To see why, notice that using (7.1), we have
Z ∞ Z ∞
f (t) e−(s−a)t dt = f˜(s − a),
 at
at
 −st
L f (t) e = f (t) e e dt =
0 0

using (7.1) again. Of course, we can guarantee that the Laplace transform on the
right-hand side exists as long as s − a > γ or, indeed, s > a + γ.

1
Example 7.4 Find the function whose Laplace transform is .
(s − 2)2
We recognise that
1 1
is a ‘shift’ by a = 2 of ,
(s − 2)2 s2
7
and so, as we know from Activity 7.1 that
1
L {t} = = f˜(s),
s2
we can use (7.7) to see that
1 ˜(s − 2) = L t e2t .

= f
(s − 2)2

That is, t e2t is the sought after function.

The Laplace transform of a function with a scaled variable

Suppose that, for t ≥ 0, we have a continuous function, f (t), of exponential growth at


most γ for some γ ≥ 0 so that its Laplace transform exists for s > γ. Then, given that
L {f (t)} = f˜(s),
we find that if we scale the variable t on the left-hand side by some positive number a,
we get
1 ˜ s 
L {f (at)} = f , (7.8)
a a
and this also exists for all s > aγ. To see why, notice that using (7.1) and making the
substitution u = at so that t = u/a and du = a dt, we have
Z ∞
1 ∞ 1 ˜ s 
Z
−st −(s/a)u
L {f (at)} = f (at) e dt = f (u) e du = f ,
0 a 0 a a

227
7. Laplace transforms

using (7.1) again. Of course, we can guarantee that the Laplace transform on the
right-hand side exists as long as s/a > γ or, indeed, s > aγ.
Of course, we can also scale the variable s on the right-hand side of
L {f (t)} = f˜(s),
by some positive number b as you will see in Activity 7.6.

Activity 7.6 Use (7.8) to show that, for b > 0, we also have
  
1 t
f˜(bs) = L f .
b b

For what values of s does this Laplace transform exist?

The Laplace transform of a function multiplied by −t

Suppose that, for t ≥ 0, we have a continuous function, f (t), of exponential growth at


most γ for some γ ≥ 0 so that its Laplace transform exists for s > γ. Then, given that
L {f (t)} = f˜(s),
we find that another useful property of the Laplace transform is that
7 df˜
L {−tf (t)} = , (7.9)
ds
and this also exists for all s > γ. That is, multiplying a function by −t differentiates its
Laplace transform. To see why, notice that using (7.1) we have
df˜ d ∞
Z ∞   Z ∞ 

Z
−st −st
= f (t) e dt = f (t) e dt = −tf (t) e−st dt = L {−tf (t)} ,
ds ds 0 0 ∂s 0

using (7.1) again. Of course, in order to manipulate the integral in this way we need to
be sure that:

The function f (t) e−st is jointly continuous for t ≥ 0 and s > γ, which it clearly is
because f (t) is assumed to be continuous for t ≥ 0.
The function −tf (t) e−st is jointly continuous for t ≥ 0 and s > γ, which it clearly
is because f (t) is assumed to be continuous for t ≥ 0.
The function −tf (t) e−st has dominated convergence for t ≥ 0 and s > γ, which it
does because, using the fact that f (t) is of exponential growth at most γ for some
γ ≥ 0, we have
| − tf (t) e−st | ≤ t(M eγt ) e−st = M t e−(s−γ)t = k(t),
so that, using integration by parts (as you should be able to show), we can see that
Z ∞ Z ∞
k(t) dt = M t e−(s−γ)t dt,
0 0

is convergent as long as s > γ as specified above.

228
7.1. What is a Laplace transform?

Summary

We have now introduced the key properties of the Laplace transform and found the
Laplace transforms of some of our basic functions. Now, using what we have seen so far,
we can easily find the Laplace transform of certain combinations of these functions.

Example 7.5 Find the Laplace transform of the function f (t) = 5 sin(3t) + 7t2 et .

To find the Laplace transform of f (t) we use (7.3) to see that

L {f (t)} = L 5 sin(3t) + 7t2 et = 5L {sin(3t)} + 7L t2 et .


 

Now, using the result from Example 7.3, we see that


3 3
L {sin(3t)} = = ,
s2 + 3 2 s2 + 9
whereas, thinking about (7.7) as we also need the Laplace transform of a function
multiplied by an exponential, we see that as
2! 2
L t2 = 2+1 = 3 ,

s s
using our result from Example 7.2, we have
7
2
L t2 et (s) = L t2 (s − 1) =
 
.
(s − 1)3

Consequently, putting this all together, we see that


   
3 2 15 14
L {f (t)} = 5 2 +7 3
= 2 + ,
s +9 (s − 1) s + 9 (s − 1)3

is the Laplace transform of f (t). Observe, in particular, that we must have s > 1 in
order for this Laplace transform to exist.

Activity 7.7 Find the Laplace transforms of the following functions.

(a) 2 sin(2t) − 3t, (b) (t + 2) et , (c) e−t cos(2t).

(You should do this by using the results concerning Laplace transforms that we have
found so far.)

Moreover, as we saw in Example 7.4, if we have the Laplace transform of a function and
we can write it in a useful way, we can apply this kind of reasoning ‘in reverse’ and find
the original function itself.

229
7. Laplace transforms

s
Example 7.6 Find the function whose Laplace transform is .
s2 +s−2
We need to find the function, f (t), whose Laplace transform is
s
f˜(s) = .
s2 +s−2
The first thing to note is that the denominator factorises to give us
s
f˜(s) = ,
(s − 1)(s + 2)

and, using partial fractions, we can write


s A B
= + ,
(s − 1)(s + 2) s−1 s+2

for some constants A and B. Finding these constants in the usual way, we get
A = 1/3 and B = 2/3 so that we now have
   
1 1 2 1
f˜(s) = + .
3 s−1 3 s+2

But, looking at Example 7.1, we see that this means that


7  t 
1  t 2  −2t e 2
f˜(s) = L e + L e =L + e −2t
,
3 3 3 3

using (7.3). Thus, we find that

et +2 e−2t
f (t) = ,
3
is the sought after function.

Activity 7.8 Find the functions whose Laplace transforms are as follows.

3s + 1 1 + s2 + s3 s
(a) , (b) , (c) .
s2 s3 (1 + s2 ) s2 + 4s + 5

(You should do this by using the results concerning Laplace transforms that we have
found so far.)

7.1.2 Extending our view of Laplace transforms


So far, for t ≥ 0, we have been looking at continuous functions, f (t), which are of
exponential growth at most γ for some γ > 0 so that we could be sure that their
Laplace transform, namely the integral
Z ∞
L {f (t)} = f (t) e−st dt,
0

230
7.1. What is a Laplace transform?

exists for s > γ. In particular, since f (t) is continuous for t ≥ 0, the Laplace transform
is an improper integral of the first kind and so any questions about its existence
ultimately come down to any problems we may have with its convergence as t → ∞.
Indeed, given that f (t) is also of exponential growth at most γ for some γ > 0, this just
comes down to insisting that s > γ as we saw at the beginning of Section 7.1.1.
But now, we want to consider some cases where we seek the Laplace transform of a
function that is no longer continuous for all t ≥ 0. In particular, we will be dealing with
functions which are not continuous at t = 0. Of course, this means that their Laplace
transform is now an improper integral of the third kind with problems at t = 0 and as
t → ∞. So, in order for their Laplace transform to exist we’ll need to be sure that
neither of these problems cause the integral to diverge.
However, when we tackle these more tricky functions, it will be very useful if we can
make use of a new function, called the Gamma function, which we shall now discuss.

The Gamma function

We define the Gamma function, Γ(α), to be the integral


Z ∞
Γ(α) = uα−1 e−u du,
0

and this is defined for all α > 0. In particular, notice that, for 0 < α < 1, this is an
improper integral of the third kind as we have a problem at u = 0 (due to the factor of 7
uα−1 in the integrand) as well as the problem as u → ∞.

Activity 7.9 Show that the Gamma function, Γ(α), is defined if and only if α > 0.

Using the definition we can see that


Z ∞ ∞ ∞
e−u

1
Z
0 −u −u
Γ(1) = u e du = e du = =0− = 1,
0 0 −1 0 −1

and this will be useful in a moment. You can find another value of the Gamma function
in Activity 7.10.

Activity 7.10 Recall from Example 5.15 that


Z ∞ r
− 12 t2 π
e dt = .
0 2
 √
Use the substitution u = 12 t2 to show that Γ 12 = π.

It turns out that there is a nice relationship between the values of Γ(α + 1) and Γ(α)
that arises if we take the former and integrate by parts. In particular, for any real
number α > 0, we have
Z ∞ −u ∞ Z ∞ −u
 
α −u αe α−1 e
Γ(α + 1) = u e du = u − (αu ) du,
0 −1 0 0 −1

231
7. Laplace transforms

if we integrate by parts. But, as the first term on the right-hand side gives us zero at
both limits, this becomes
Z ∞
Γ(α + 1) = α uα−1 e−u du = αΓ(α),
0

i.e. for any real number α > 0, we have


Γ(α + 1) = αΓ(α). (7.10)
Furthermore, if α > 0 is a natural number, say n = 1, 2, 3, . . ., we can apply this result n
times to get
Γ(n + 1) = nΓ(n) = n(n − 1)Γ(n − 1) = · · · = n(n − 1) · · · (2)(1)Γ(1).
So, using the fact that Γ(1) = 1 from above, we see that
Γ(n + 1) = n! (7.11)
provided that n ∈ N.

7

Activity 7.11 Evaluate Γ 2
.

The Laplace transform of an arbitrary power of t

7 We saw in Example 7.2 that, for n = 1, 2, 3, . . ., the Laplace transform of tn is given by


n!
L {tn } = .
sn+1
But, now that we have the Gamma function, we can ask about the Laplace transform of
an arbitrary power of t. That is, what is the Laplace transform of tα where α can be
any real number? Of course, some care will be needed here because when −1 < α < 0,
the function tα is only continuous for t > 0. However, as we will see in Activity 7.12, the
Laplace transform of tα , i.e. the integral
Z ∞
α
L {t } = tα e−st dt,
0

will exist as long as α > −1 and s > 0.

Activity 7.12 Show that the Laplace transform of tα exists if α > 0 and s > 0.

Indeed, we can see that for some fixed s > 0, the substitution u = st gives us du = sdt
so that we have
1 ∞  u α −u
Z ∞
1
Z
α
L {t } = e du = α+1 uα e−u du,
s 0 s s 0

which means that, using the definition of the Gamma function, we have found that
Γ(α + 1)
L {tα } = , (7.12)
sα+1
when α > −1 is a real number.

232
7.2. Using Laplace transforms

Activity 7.13 Verify that (7.12) agrees with the result from Example 7.2 when
α ∈ N.

Activity 7.14 Use (7.4) and (7.12) to obtain an alternative derivation of (7.10).

The Laplace transform of a logarithm

We may also wonder whether the function ln t has a Laplace transform bearing in mind
that this function is only continuous for t > 0. But, as we will see in Activity 7.15, it
turns out that its Laplace transform, i.e. the integral
Z ∞
L {ln t} = (ln t) e−st dt,
0

does exist as long as s > 0.

Activity 7.15 Show that the Laplace transform of ln t exists if s > 0.

In fact, as you will see in Question 4(c) of the sample examination paper in
Appendix A, it can be shown that
1 0
L {ln t} = (Γ (1) − ln s) ,
s 7
and, for reasons beyond the scope of this course, we have Γ0 (1) = −γE where
 
1 1
γE = lim 1 + + · · · + − ln n = 0.57722 . . . ,
n→∞ 2 n
is called Euler’s constant.

7.2 Using Laplace transforms


We will consider two applications of Laplace transforms. The first will involve using
them to solve ordinary differential equations (ODEs). Indeed, as we shall see, they will
give us a relatively simple method for solving ODEs with constant coefficients like the
ones we saw in Section 7.3 of 174 Calculus. The second will involve looking at the
Laplace transform of the convolution of two functions and seeing how this is useful in
several different situations.

7.2.1 Solving ODEs with constant coefficients


We saw in Section 7.3 of 174 Calculus how to solve non-homogeneous second-order
ODEs with constant coefficients, i.e. we saw how to find a function y(t) that satisfies an
ODE of the form
ay 00 (t) + by 0 (t) + cy(t) = f (t),
for constants a, b, c ∈ R and some function, f (t). The method for solving such an ODE
using Laplace transforms is as follows.

233
7. Laplace transforms

Find the Laplace transform of the ODE and get an expression for the Laplace
transform, ỹ(s), of the function, y(t).
Then by using the ideas that we encountered in Example 7.6 and Activity 7.8 we
can find the function, y(t), whose Laplace transform is ỹ(s).
This function, y(t), that we have found will then be the sought after solution to the
ODE.2 Let’s consider an example.

Example 7.7 Find the solution that satisfies the ODE

y 00 (t) − y 0 (t) − 2y(t) = 18 e2t ,

with the initial conditions y(0) = 3 and y 0 (0) = 6.

Using (7.3), we see that the Laplace transform of the ODE will be given by

L {y 00 (t)} − L {y 0 (t)} − 2L {y(t)} = 18L e2t ,




and so, if we let L {y(t)} (s) = ỹ(s) as usual, and use

(7.5) and the initial conditions, to get

L {y 00 (t)} = s2 ỹ(s) − sy(0) − y 0 (0) = s2 ỹ(s) − 3s − 6,

7 (7.4) and the initial conditions, to get

L {y 0 (t)} = sỹ(s) − y(0) = sỹ(s) − 3,

(7.2), to get
1
L e2t =

,
s−2
we see that the Laplace transform of the ODE is
     
2 1
s ỹ(s) − 3s − 6 − sỹ(s) − 3 − 2ỹ(s) = 18 ,
s−2
in terms of ỹ(s). Tidying this up, we then have
18 18 + 3(s + 1)(s − 2)
(s2 − s − 2)ỹ(s) = + 3(s + 1) =⇒ (s − 2)(s + 1)ỹ(s) = ,
s−2 s−2
and so, we find that
3s2 − 3s + 12
ỹ(s) = ,
(s + 1)(s − 2)2
is our expression for ỹ(s). We can now rewrite this using partial fractions, i.e. we
have
3s2 − 3s + 12 A B C
2
= + + ,
(s + 1)(s − 2) s + 1 s − 2 (s − 2)2
2
Now, this may sound a bit circular, but this method works because, if y(t) is a solution to the ODE,
then its Laplace transform, ỹ(s), must satisfy the equation that we get from the Laplace transform of
the ODE. So, if we can figure out which function, y(t), has ỹ(s) as its Laplace transform, we will have
found a solution.

234
7.2. Using Laplace transforms

for some constants A, B and C. Finding these in the usual way, we get A = 2, B = 1
and C = 6 so that we now have
2 1 6
ỹ(s) = + + .
s + 1 s − 2 (s − 2)2

Of course, using (7.2), we have


 
2 1 1
= 2L e−t = L e2t ,
 
=2 and
s+1 s+1 s−2

whereas, using Example 7.4, we see that


 
6 1  2t
= 6 = 6L te ,
(s − 2)2 (s − 2)2

which means that, using (7.3), we have

ỹ(s) = 2L e−t + L e2t + 6L t e2t = L 2 e−t + e2t +6t e2t .


   

Consequently, we see that

y(t) = 2 e−t + e2t +6t e2t ,

is the sought after solution to the ODE. 7


Activity 7.16 Check that the solution that we found in Example 7.7 is correct by
showing that it satisfies the ODE and the given initial conditions.

In particular, observe that when we solved this ODE in Example 7.14 of the subject
guide for 174 Calculus, we found the general solution

y(t) = A e−t +B e2t +6t e2t ,

where A and B are arbitrary constants. Indeed, if we take this general solution and
apply the initial conditions y(0) = 3 and y 0 (0) = 6 that we used in Example 7.7 above,
we would find that the constants A and B have to be A = 2 and B = 1 as you should
expect because these will give us the required particular solution for these initial
conditions. So, more generally, you can see that there is a difference between the two
methods that we now have for solving such ODEs:

The method from Section 7.3 of 174 Calculus gives us a general solution which,
given any initial conditions, can be turned into a particular solution.

The method above using Laplace transforms uses some given initial conditions at
the beginning and so only yields a particular solution.
That is, in general, the method above which uses Laplace transforms is only really
useful if we seek a particular solution for some initial conditions as opposed to a general
solution!

235
7. Laplace transforms

7.2.2 Convolutions
Suppose that we have two continuous functions, f and g. We define the convolution of f
and g, which we denote by f ? g, to be the function given by
Z t
(f ? g)(t) = f (u)g(t − u) du.
0

Note, in particular, that for 0 ≤ u ≤ t we have 0 ≤ t − u ≤ t and so, both of the


functions in the integrand have arguments that take values in the interval [0, t].

Example 7.8 Find the convolution of the functions f (t) = t and g(t) = et .

Using the definition above, we have


Z t Z t
(f ? g)(t) = f (u)g(t − u) du = u et−u du,
0 0

and so, using integration by parts, we get


Z t  t−u t Z t   t−u t
et−u
 0  0
et

t−u e e e e
ue du = u − (1) du = t −0 + = −t+ − ,
0 −1 0 0 −1 −1 −1 0 −1 −1

7 and so, we find that


(f ? g)(t) = −t − 1 + et ,
is the convolution of the functions f (t) = t and g(t) = et .

In particular, it is useful to note that convolutions have the following useful property.

Activity 7.17 Show that convolutions are symmetric, i.e. that

(f ? g)(t) = (g ? f )(t).

[Hint: Use the substitution v = t − u where t is fixed.]

The Convolution theorem

It turns out that convolutions have a particularly nice Laplace transform. In particular,
we have the Convolution theorem which runs as follows.
Theorem 7.1 (The Convolution theorem) If f (t) and g(t) are two continuous
functions of exponential growth at most γ1 and γ2 respectively, then
Z t 
L {(f ? g)(t)} = L f (u)g(t − u) du = f˜(s)g̃(s),
0

and this is defined for all s > max{γ1 , γ2 }. Note that, as always, f˜(s) and g̃(s) are the
Laplace transforms of the functions f (t) and g(t) respectively and these are both
defined for the stated values of s.

236
7.2. Using Laplace transforms

To see why this works, consider that


Z ∞ Z ∞
˜
f (s) = f (t) e−st
dt and g̃(s) = g(t) e−st dt,
0 0

both exist if s > max{γ1 , γ2 } and so, replacing the dummy variable t in these integrals
with the dummy variables u and v respectively, we have
Z ∞  Z ∞ 
˜
f (s)g̃(s) = f (u) e−su
du g(v) e−sv
dv .
0 0

But, we can interpret the product of integrals on the right-hand side as a repeated
integral if we take
Z ∞ Z ∞ 
˜
f (s)g̃(s) = f (u)g(v) e−s(u+v)
dv du.
u=0 v=0

Now, if we fix u and make the substitution w = u + v in the inner integral, we have
v = w − u and dw = dv, which means that
Z ∞ Z ∞ 
f˜(s)g̃(s) = f (u)g(w − u) e−sw
dw du.
u=0 w=u

But, as you will see in Activity 7.18, we can change the order of integration to get
Z ∞ Z w 
˜
f (s)g̃(s) = f (u)g(w − u) du e−sw dw,
w=0 u=0 7
and if we replace the dummy variable w with the dummy variable t, we get
Z ∞ Z t  Z t 
˜
f (s)g̃(s) = f (u)g(t − u) du e −st
dt = L f (u)g(t − u) du ,
t=0 u=0 u=0

which is L {(f ? g)(t)}, as required.

Activity 7.18 Show that


Z ∞ Z ∞  Z ∞ Z w 
−sw
f (u)g(w − u) e dw du = f (u)g(w − u) du e−sw dw,
u=0 w=u w=0 u=0

by changing the order of integration.

It turns out that the Convolution theorem has three immediate applications and we
shall consider these in turn. Firstly, we can see that it gives us another way of finding
convolutions as the next example shows.

Example 7.9 Following on from Example 7.8, use the Convolution theorem to find
the convolution of the functions f (t) = t and g(t) = et .

We seek the convolution (f ? g)(t) and, using the Convolution theorem, we know
that its Laplace transform is given by
  
˜ 1 1
L {(f ? g)(t)} = f (s)g̃(s) = ,
s2 s−1

237
7. Laplace transforms

if we use the results from Activity 7.1 and Example 7.1 respectively. Of course, this
is the Laplace transform of the function we seek and so, by using partial fractions or
simply noting that

1 s2 − (s2 − 1) 1 (s + 1)(s − 1) 1 s+1


= = − = − ,
s2 (s − 1) s2 (s − 1) s−1 s2 (s − 1) s−1 s2

we see that, using the results from Activity 7.1 and Example 7.1 again, we have
1 s+1 1 1 1
L {(f ? g)(t)} = − 2 = − − 2 =⇒ (f ? g)(t) = et −1 − t,
s−1 s s−1 s s
as we found in Example 7.8.

Secondly, it gives us another way of ‘reversing’ Laplace transforms, i.e. finding a


function when we are given its Laplace transform, as the next example shows.

1
Example 7.10 Find the function whose Laplace transform is .
(s − 2)2 (s + 3)
The function we seek, let’s call it h(t), has a Laplace transform given by
  
1 1 1
h̃(s) = = ,
(s − 2)2 (s + 3) (s − 2)2 s+3
7 which is the product of two Laplace transforms, let’s call them
1 1
f˜(s) = 2
and g̃(s) = .
(s − 2) s+3
This means that f (t) and g(t) must be the functions
f (t) = t e2t and g(t) = e−3t ,
if we use the results from Examples 7.1 and 7.4. However, using the Convolution
theorem, we now have
h̃(s) = f˜(s)g̃(s) = L {(f ? g)(t)} ,
and so the function we seek is just the convolution given by
Z t Z t Z t
2u −3(t−u) −3t
h(t) = (f ? g)(t) = f (u)g(t − u) du = (u e )(e ) du = e u e5u du,
0 0 0
−3t
as e is a constant here. Consequently, using parts to find the integral we get,
( )
5u t Z t 5u

e e
h(t) = e−3t u − (1) du
5 0 0 5
(  5u t )
5t
t e −0 e
= e−3t −
5 25 0
 5t
e5t −1

−3t te
=e −
5 25
2t −3t
(5t − 1) e + e
∴ h(t) = ,
25
is the sought after function.

238
7.2. Using Laplace transforms

Thirdly, they allow us to solve certain integral equations.

Example 7.11 In Example 3.10, we considered the integral equation


Z t
f (t) = 1 − f (u) du.
0

Use the Convolution theorem to find f (t).

We first notice that the integral equation we are given involves a convolution, i.e. if
we take g(t) = 1, we have
Z t Z t Z t
f (u) du = f (u)(1) du = f (u)g(t − u) du = (f ? g)(t),
0 0 0

and so, taking the Laplace transform of both sides of the equation we get
1 1
f˜(s) = − f˜(s) · ,
s s
if we use the Convolution theorem and one of the results from Example 7.1. As such,
we can see that the function we seek has a Laplace transform given by
 
1 ˜ 1 s+1 ˜ 1 1
1+ f (s) = =⇒ f (s) = =⇒ f˜(s) = . 7
s s s s s+1

Consequently, using the other result from Example 7.1, we see that f (t) = e−t is the
sought after function in agreement with what we found in Example 3.10.

Our last application of convolutions involves a new function, called the Beta function
and, as this is a bit more involved than the other three applications, we’ll consider it in
more detail.

The Beta function

Another application of convolutions and their Laplace transform is that they allow us to
see how to evaluate certain integrals. In particular, we will look at the Beta function,
B(p, q), which is given by the integral
Z 1
B(p, q) = up−1 (1 − u)q−1 du,
0

and, as we will see in Activity 7.19, this function is defined for p, q > 0.

Activity 7.19 Show that the Beta function, B(p, q), is defined if and only if
p, q > 0.

If we want to evaluate Beta functions, it is useful to make a connection between Beta


functions and convolutions. To do this, we can define the ‘generalised Beta function’,

239
7. Laplace transforms

B(p, q, t), to be the integral


Z t
B(p, q, t) = up−1 (t − u)q−1 du,
0

so that, in particular, we have B(p, q, 1) = B(p, q).

Activity 7.20 Use the substitution u = tv where t is fixed to show that the formula

B(p, q, t) = tp+q−1 B(p, q),

gives us another relationship between B(p, q, t) and B(p, q).

Of course, this is useful because we can see that B(p, q, t) is the convolution of the
functions
f (t) = tp−1 and g(t) = tq−1 ,
so that we have B(p, q, t) = (f ? g)(t). Thus, to find the Laplace transform of this
function, we use (7.12) to get
Γ(p) Γ(q)
f˜(s) = p and g̃(s) = q ,
s s
so that the Convolution theorem gives us
Γ(p) Γ(q)
L {B(p, q, t)} = L {(f ? g)(t)} = f˜(s) · g̃(s) = · q ,
7 sp s
or, indeed,
Γ(p)Γ(q)
L {B(p, q, t)} = . (7.13)
sp+q
Now, using (7.12) again, we also have
Γ(p + q) 1 1
L tp+q−1 = L tp+q−1 ,
 
=⇒ =
sp+q sp+q Γ(p + q)
and so we see that we have
 
Γ(p)Γ(q) Γ(p)Γ(q)  p+q−1 Γ(p)Γ(q) p+q−1
L {B(p, q, t)} = = L t =L t ,
sp+q Γ(p + q) Γ(p + q)
using (7.3). That is, we see that
Γ(p)Γ(q) p+q−1
B(p, q, t) = t ,
Γ(p + q)
allows us to relate the generalised Beta function to the Gamma function. In particular,
if we set t = 1 in this formula, we get
Γ(p)Γ(q)
B(p, q, 1) = ,
Γ(p + q)
and so, as B(p, q, 1) = B(p, q), we see that we have
Γ(p)Γ(q)
B(p, q) = , (7.14)
Γ(p + q)
i.e. we can evaluate the Beta function using the Gamma function.

240
7.2. Learning outcomes

Activity 7.21 Take the Laplace transform of the relationship between B(p, q, t) and
B(p, q) found in Activity 7.20 and then use (7.12) and (7.13) to get an alternative
derivation of (7.14).

Of course, now that we can evaluate Beta functions in terms of Gamma functions
(which, in turn, we can usually evaluate using what we saw when we discussed them in
Section 7.1.2), we can use this to evaluate certain integrals very easily.

1
t2
Z
Example 7.12 Use the Beta function to find I = √ dt.
0 1−t
We see that
Z 1 Z 1 Z 1
t2 2 −1/2
I= √ dt = t (1 − t) dt = t3−1 (1 − t)1/2−1 dt = B(3, 12 ),
0 1 − t 0 0

as a Beta function. So, using (7.14), we have

Γ(3)Γ( 21 ) (2!)Γ( 21 ) 2 16
I= = 1 = 15 = ,
Γ( 72 ) 5
2
3 1
· 2 · 2 Γ( 2 ) 8
15

where we have used (7.10) and (7.11) to simplify the Gamma functions.
7
1
t5/2
Z
Activity 7.22 Use the Beta function to find I = √ dt.
0 1−t

Learning outcomes
At the end of this chapter and having completed the relevant reading and activities, you
should be able to:

define the Laplace transform and establish some of its basic properties;

find the Laplace transform of a given function;

find a function given its Laplace transform;

use Laplace transforms to solve a differential equation;

identify a convolution and use the convolution theorem;

define the Beta and Gamma functions and establish their basic properties.

241
7. Laplace transforms

Solutions to activities
Solution to activity 7.1
For f (t) = t, (7.1) gives us
∞ ∞ ∞
e−st e−st
Z  Z
f˜(s) = te −st
dt = t − (1) dt,
0 −s 0 0 −s

if we integrate by parts. At this point, we need s > 0 to guarantee that

t e−st
lim = 0,
t→∞ −s

so that the integral exists, and if that is the case, we get


1 ∞
Z
˜
f (s) = 0 − 0 + (1) e−st dt,
s 0

where, again using (7.1), we see that the integral on the right-hand side is just L {1}.
Thus, using what we saw in Example 7.1, we have L {1} = 1/s for s > 0 and with this
we get
1 1 1 1
f˜(s) = L {1} = · = 2 ,
s s s s
7 and, as we have seen, this Laplace transform will only exist if s > 0. Of course, we could
write this result as
L{t} = 1/s2 ,
if we wanted.

Solution to activity 7.2


Given that, for t ≥ 0, we have two functions f (t) and g(t) whose Laplace transforms
Z ∞ Z ∞
−st
L{f (t)} = f (t) e dt and L{g(t)} = g(t) e−st dt,
0 0

both exist for s > γ for some γ > 0, we see that for any constants α, β ∈ R, the Laplace
transform of the function αf (t) + βg(t) is given by
Z ∞
αf (t) + βg(t) e−st dt

L{αf (t) + βg(t)} =
0
Z ∞ Z ∞
−st
=α f (t) e dt + β g(t) e−st dt
0 0
= αL{f (t)} + βL{g(t)},

if we use the properties of the definite integral. Of course, the Laplace transform of the
function αf (t) + βg(t) only exists when s > γ as we need the Laplace transform of both
f (t) and g(t) to exist if this is to work.

242
7.2. Solutions to activities

Solution to activity 7.3


Using (7.4) on f 00 (t), we see that
L {f 00 (t)} = sL {f 0 (t)} − f 0 (0),
as f 00 (t) is the first derivative of f 0 (t). Then, using (7.4) on f 0 (t), we find that
L {f 00 (t)} = s sL {f (t)} − f (0) − f 0 (0) = s2 L {f (t)} − sf (0) − f 0 (0),


as required.

Solution to activity 7.4


If we take f (t) = cos(at) for some constant a 6= 0 so that f 0 (t) = −a sin(at), we see that
(7.4) gives us
L {−a sin(at)} = sL {cos(at)} − 1,
as f (0) = 1. Consequently, applying (7.3) and using the result from Example 7.3, we
now have
a2 s2
 
a
−a 2 = sL {cos(at)} − 1 =⇒ sL {cos(at)} = 1 − = ,
s + a2 s 2 + a2 s 2 + a2
and so we find that
s
L {cos(at)} = ,
+ a2 s2
is the Laplace transform of cos(at) for some constant a 6= 0. Of course, this is defined
for all s > 0 since the function cos(at) is continuous and of exponential growth at most
7
zero as | cos(at)| ≤ 1 = e0t .

Solution to activity 7.5


Given that f (t) = eiat for some a ∈ R, its Laplace transform is given by
Z ∞ Z ∞  −(s−ia)t ∞  
˜ iat −st −(s−ia)t e 1 1
f (s) = e e dt = e dt = =0− − = ,
0 0 −(s − ia) 0 s − ia s − ia
where since
| eiat e−st | ≤ | eiat | · | e−st | ≤ e−st ,
as | eiat | = 1, we need s > 0 in order to guarantee that
e−(s−ia)t
lim = 0,
t→∞ −(s − ia)

so that the integral exists.


Now, using the fact that eiat = cos(at) + i sin(at), we have
Z ∞
˜
f (s) = eiat e−st dt
Z0 ∞
cos(at) + i sin(at) e−st dt

=
Z0 ∞ Z ∞
−st
= cos(at) e dt + i sin(at) e−st dt
0 0
= L {cos(at)} + iL {sin(at)} ,

243
7. Laplace transforms

whereas, using the Laplace transform that we found above, we also have
1 s + ia s + ia s a
f˜(s) = = = 2 = + i .
s − ia (s − ia)(s + ia) s + a2 s 2 + a2 s 2 + a2

Consequently, if we equate the real and imaginary parts of these expressions for f˜(s),
we find that
s a
L {cos(at)} = 2 2
and L {sin(at)} = 2 ,
s +a s + a2
as expected.

Solution to activity 7.6


Given some b > 0, we can use (7.8) with a = 1/b > 0 to get
     
t 1 t
L f = bf˜(bs) =⇒ f˜(bs) = L f ,
b b b

if we use (7.3). Of course, if f (t) is of exponential growth at most γ, then (7.8) only
works for s > aγ and so, consequently, this Laplace transform only exists if s > γ/b.

Solution to activity 7.7


For (a), we use (7.3) to see that
7 2 1 4 3
L {2 sin(2t) − 3t} = 2L {sin(2t)} − 3L {t} = 2 −3 2 = 2 − 2,
s2 +2 2 s s +4 s
if we use the results from Example 7.3 and Activity 7.1 respectively.

For (b), we can use (7.3) to see that

1 1 1 2
L (t + 2) et = L t et + 2L et =
  
2
+2 = 2
+ ,
(s − 1) s−1 (s − 1) s−1

where we have used L {t} = 1/s2 from Activity 7.1 and (7.7) with a ‘shift’ by a = 1 to
get the first term as well as one of the results from Example 7.1 for the second term.

For (c), we can use the result from Activity 7.4 to see that

s s s+1
L e−t cos(2t) =

L {cos(2t)} = = =⇒ ,
s2 + 2 2 s2 + 4 (s + 1)2 + 4

if we use (7.7) with a ‘shift’ by a = −1.

Solution to activity 7.8


For (a), we can rewrite this and use one of the results from Example 7.1 and the result
from Activity 7.1 to see that
3s + 1 1 1
2
= 3 + 2 = 3L {1} + L {t} = L {3 + t} ,
s s s
if we use (7.3). Thus, in this case, the sought after function is 3 + t.

244
7.2. Solutions to activities

For (b), we can rewrite this and use the results from Example 7.2 and Activity 7.4 to
see that
1 + s2 + s3 t2
 
1 1 1 2! 1 1 
3 2
= 3+ 2
= · 3+ 2 = L t2 + L {sin t} = L + sin t ,
s (1 + s ) s 1+s 2 s s +1 2 2
if we use (7.3). Thus, in this case, the sought after function is 12 t2 + sin t.

For (c), we can rewrite this as


s s s+2 1
= = − 2 ,
s2 + 4s + 5 (s + 2)2 + 1 (s + 2)2 + 1 (s + 2)2 + 1
and observe that we effectively have the Laplace transforms of cos t and sin t (from
Example 7.3 and Activity 7.4) after a ‘shift’ by a = −2. So, using (7.7), we see that
s  −2t  −2t  −2t −2t

= L e cos t − 2L e sin t = L e cos t − 2 e sin t ,
s2 + 4s + 5
if we use (7.3). Thus, in this case, the sought after function is e−2t (cos t − 2 sin t).

Solution to activity 7.9


We write the Gamma function as
Z ∞ Z 1 Z ∞
α−1 −u α−1 −u
Γ(α) = u e du = u e du + uα−1 e−u du,
0
|0 {z } |1 {z }
I1 I2

so that we can discuss the problems at u = 0 and as u → ∞ separately. 7


For I1 , we consider values of u that are close to zero so that
f (u) = uα−1 e−u ' uα−1 = g(u),
which gives us
f (u) uα−1 e−u
lim+ = lim+ = lim+ e−u = 1 and
u→0 g(u) u→0 uα−1 u→0
Z 1 Z 1
g(u) du = uα−1 du which is convergent if and only if α > 0.
0 0
So, by the LCT, we see that I1 is convergent if and only if α > 0.

For I2 , we again have f (u) = uα−1 e−u and for large u we take
g(u) = e−u/2 ,
which gives us
f (u) uα−1 e−u
lim = lim −u/2
= lim uα−1 e−u/2 = 0 and
u→∞ g(u) u→∞ e u→∞
Z ∞ Z ∞
g(u) du = e−u/2 du is convergent.
1 1
So, by the LCT, we see that I2 is convergent for all α ∈ R.

Consequently, in order for Γ(α) to exist we must have α > 0 as we need both I1 and I2
to be convergent.

245
7. Laplace transforms

Solution to activity 7.10


Using the definition of the Gamma function we have
Z ∞ Z ∞
1 1
−1 −u
1
u− 2 e−u du,

Γ 2 = u 2 e du =
0 0

and the substitution u = 12 t2 gives us du = tdt so that we get


−1/2 √
∞ ∞ √ Z ∞ − 1 t2 √
 r
1 2 2 − 1 t2 π
Z Z
1 − 21 t2

Γ 2
= t e t dt = e 2 t dt = 2 e 2 dt = 2 · ,
0 2 0 t 0 2
1
 √
using the given result. That is, Γ 2
= π as required.

Solution to activity 7.11


Using (7.10) three times we have

Γ 72 = 52 Γ 25 = 52 · 32 Γ 23 = 52 · 32 · 12 Γ 1 15
   
2
= 8
π,
 √
as we know that Γ 21 = π from Activity 7.10.

Solution to activity 7.12


We can write the Laplace transform of tα as
7 Z ∞ Z 1 Z ∞
α −st α −st
α
L {t } = t e dt = t e dt + tα e−st dt,
0
| 0 {z } |1 {z }
I1 I2

so that we can discuss the problems at t = 0 and as t → ∞ separately.

For I1 , we consider values of t that are close to zero so that

f (t) = tα e−st ' tα = g(t),

which gives us

f (t) tα e−st
lim+ = lim+ = lim+ e−st = 1 and
t→0 g(t) t→0 tα t→0

Z 1 Z 1
g(t) dt = tα dt which is convergent if and only if α > −1.
0 0
So, by the LCT, we see that I1 is convergent if and only if α > −1.

For I2 , we again have f (t) = tα e−st and for large t we take

g(t) = e−st/2 ,

which gives us

f (t) tα e−st
lim = lim −st/2 = lim tα e−st/2 = 0 and
t→∞ g(t) t→∞ e t→∞

246
7.2. Solutions to activities

Z ∞ Z ∞
g(t) dt = e−st/2 dt which is convergent if s > 0.
1 1
So, by the LCT, we see that I2 is convergent if s > 0.

Consequently, in order for L {tα } to exist we must have α > −1 and s > 0 as we need
both I1 and I2 to be convergent.

Solution to activity 7.13


Of course, when α ∈ N, we can use (7.11) to write

Γ(α + 1) n!
L {tα } = α+1
= α+1 ,
s s
in agreement with what we found in Example 7.2.

Solution to activity 7.14


Given that α > 0, we can use (7.4) with f (t) = tα to get

L αtα−1 = sL {tα } − 0,


as we have f 0 (t) = αtα−1 and f (0) = 0. Thus, using (7.3), this gives us

Γ(α) Γ(α + 1)
αL tα−1 = sL {tα }

=⇒ α = s α+1 ,
s α s 7
if we use (7.12). Consequently, as the s’s on both sides cancel, we have

Γ(α + 1) = αΓ(α),

which is (7.10) as required.

Solution to activity 7.15


We can write the Laplace transform of ln t as
Z ∞ Z 1 Z ∞
−st −st
L {ln t} = (ln t) e dt = (ln t) e dt + (ln t) e−st dt,
0
|0 {z } |1 {z }
I1 I2

so that we can discuss the problems at t = 0 and as t → ∞ separately.

For I1 , we consider values of t that are close to zero so that

f (t) = (ln t) e−st ' ln t = g(t),

which gives us

f (t) (ln t) e−st


lim+ = lim+ = lim+ e−st = 1 and
t→0 g(t) t→0 ln t t→0

Z 1 Z 1
g(t) dt = ln t dt which is convergent as we saw in Activity 4.8(b).
0 0

247
7. Laplace transforms

So, by the LCT, we see that I1 is convergent for all s ∈ R.

For I2 , we again have f (t) = (ln t) e−st and for large t we take

g(t) = e−st/2 ,

which gives us

f (t) (ln t) e−st


lim = lim = lim (ln t) e−st/2 = 0 and
t→∞ g(t) t→∞ e−st/2 t→∞
Z ∞ Z ∞
g(t) dt = e−st/2 dt which is convergent if s > 0.
1 1
So, by the LCT, we see that I2 is convergent if s > 0.

Consequently, in order for L {ln t} to exist we must have s > 0 as we need both I1 and
I2 to be convergent.

Solution to activity 7.16


Having found that

y(t) = 2 e−t + e2t +6t e2t = 2 e−t +(1 + 6t) e2t ,

7 we see that we get the derivatives

y 0 (t) = −2 e−t +(8 + 12t) e2t and y 00 (t) = 2 e−t +(28 + 24t) e2t ,

where we have used the product rule to differentiate the final term in each case.
Consequently, if we substitute these derivatives into the left-hand side of the given
ODE, we get

y 00 (t) − y 0 (t) − 2y(t) = [2 e−t +(28 + 24t) e2t ] − [−2 e−t +(8 + 12t) e2t ] − 2[2 e−t +(1 + 6t) e2t ],

and this simplifies to give us 18 e2t which is the right-hand side of the ODE. This
verifies that y(t) does indeed satisfy the ODE. Moreover, we see when t = 0 we have

y(0) = 2 + 1 = 3 and y 0 (0) = −2 + 8 = 6,

which verifies that y(t) also satisfies the given initial conditions.

Solution to activity 7.17


For fixed t, the substitution v = t − u gives us dv = −du and so
Z t Z 0 Z t
(f ? g)(t) = f (u)g(t − u) du = f (t − v)g(v)(−dv) = g(v)f (t − v) dv = (g ? f )(t),
0 t 0

as required.

248
7.2. Solutions to activities

Solution to activity 7.18


Given the repeated integral
Z ∞ Z ∞ 
−sw
f (u)g(w − u) e dw du,
u=0 w=u

we see that for a fixed 0 ≤ u < ∞ we have u ≤ w < ∞ and so we get the region of
integration illustrated in Figure 7.1. As such, if we change the order of integration, we
now need a fixed 0 ≤ w < ∞ to see that 0 ≤ u ≤ w and this gives us
Z ∞ Z ∞  Z ∞ Z w 
−sw
f (u)g(w − u) e dw du = f (u)g(w − u) du e−sw dw,
u=0 w=u w=0 u=0

as required. Of course, this is not really a satisfactory argument because we are dealing
with an improper repeated integral here and so Fubini’s theorem is not applicable (cf.
our discussion in Section 5.3), but we could overcome this problem if we wanted to.

7
u
=
w

O u

Figure 7.1: The region of integration in Activity 7.18.

Solution to activity 7.19


We can write the Beta function as
Z 1 Z 1/2 Z 1
p−1 q−1 p−1 q−1
B(p, q) = u (1 − u) du = u (1 − u) du + up−1 (1 − u)q−1 du,
0
|0 {z } | 1/2 {z }
I1 I2

so that we can deal with the problems at u = 0 and u = 1 (which would occur if p < 1
or q < 1) separately.

For I1 , we consider values of u that are close to zero so that

f (u) = up−1 (1 − u)q−1 ' up−1 = g(u),

which gives us

f (u) up−1 (1 − u)q−1


lim+ = lim+ p−1
= lim+ (1 − u)q−1 = 1 and
u→0 g(u) u→0 u u→0

249
7. Laplace transforms

Z 1/2 Z 1/2
g(u) du = up−1 du which is convergent if and only if p > 0.
0 0
So, by the LCT, we see that I1 is convergent if and only if p > 0.

For I2 , we consider values of u that are close to one so that

f (u) = up−1 (1 − u)q−1 ' (1 − u)q−1 = g(u),

which gives us

f (u) up−1 (1 − u)q−1


lim− = lim− = lim− up−1 = 1 and
u→1 g(u) u→1 (1 − u)q−1 u→1
Z 1 Z 1
g(u) du = (1 − u)q−1 du which is convergent if and only if q > 0.
1/2 1.2

So, by the LCT, we see that I1 is convergent if and only if q > 0.

Consequently, B(p, q) exists if and only if p > 0 and q > 0 as we need both I1 and I2 to
be convergent.

Solution to activity 7.20


For fixed t, we make the substitution u = tv so that du = tdv to see that

7
Z t Z 1 Z 1
p−1 q−1 p−1 q−1 p+q−1
B(p, q, t) = u (t−u) du = (tv) (t−tv) (tdv) = t v p−1 (1−v)q−1 dv,
0 0 0

where we can take the factor of tp+q−1 out of the integral because t is fixed and p, q are
constants. But, using the definition of the Beta function, we now observe that
Z 1
v p−1 (1 − v)q−1 dv = B(p, q) =⇒ B(p, q, t) = tp+q−1 B(p, q),
0

as required.

Solution to activity 7.21


Taking the Laplace transform of the relationship between B(p, q, t) and B(p, q) found in
Activity 7.20, we have

L {B(p, q, t)} = L tp+q−1 B(p, q) = B(p, q)L tp+q−1 ,


 

as B(p, q) is a constant as far as t is concerned. Then, using (7.13) and (7.12), this gives
us
Γ(p)Γ(q) Γ(p + q) Γ(p)Γ(q)
p+q
= B(p, q) p+q =⇒ B(p, q) = ,
s s Γ(p + q)
which is (7.14), as required.

Solution to activity 7.22


We note that the given integral is a Beta function because
Z 1 5/2 Z 1
t 7 1
t 2 −1 (1 − t) 2 −1 dt = B 7 1

I= √ dt = ,
2 2
,
0 1−t 0

250
7.2. Exercises

and, as such, we can use (7.14) to see that


15 √
 √
Γ 72 Γ 12 15
 
8
π ( π) 8
π 5π
I= = = = ,
Γ(4) 3! (3)(2)(1) 16
if we use the results from Activities 7.10 and 7.11 as well as (7.11) to see that Γ(4) = 3!.

Exercises
Exercise 7.1
Find the Laplace transforms of the following functions.

(a) 2 e3t −t2 , (b) 3 cos(5t) − sin(5t), (c) (t + 3)2 e−t , (d) et sin t.

(Of course, you should do this by using the results concerning Laplace transforms that
we found in Section 7.1.)

Exercise 7.2
Find the functions whose Laplace transforms are as follows.
s+1 5s + 2 1 + s + s2 s
(a) , (b) , (c) , (d) .
s2 + 4 s3 s2 (1 + s) s2 + 2s + 10 7
(Of course, you should do this by using the results concerning Laplace transforms that
we found in Section 7.1.)

Exercise 7.3
Use Laplace transforms to solve the differential equation

y 00 (t) − y(t) = t,

given the initial conditions y(0) = −1 and y 0 (0) = 0.

Exercise 7.4
Z t
Use Laplace transforms to solve the integral equation f (t) = t + 2 f (u) cos(t − u) du.
0

Exercise 7.5
For which values of a ∈ R does the integral
Z ∞
ta
J(a) = dt,
0 (1 + t)3
converge? By making an appropriate substitution, show that

J(a) = B(2 − a, 1 + a),

whenever the integral J(a) converges.

251
7. Laplace transforms

Solutions to exercises
Solution to exercise 7.1
We use the results from Section 7.1 (omitting references to the more familiar ones).

(a) We have
2 2
L 2 e3t −t2 = 2L e3t − L t2 =
  
− 3.
s−3 s

(b) We have

3s − 5
 
s 5
L {3 cos(5t) − sin(5t)} = 3L {cos(5t)}−L {sin(5t)} = 3 2 2
− 2 2
= 2 .
s +5 s +5 s + 25

(c) We have (t + 3)2 = t2 + 6t + 9 so that


   
2 2 1 1
= L t + 6t + 9 = L t2 + 6L {t} + 9L {1} = 3 + 6
2
  
L (t + 3) 2
+9 ,
s s s

which means that


2 6 9
L (t + 3)2 e−t =

+ ,
7 3
(s + 1) (s + 1) 2 s+1

if we use (7.7) and a ‘shift’ by a = −1.

(d) We have

1 1 1
L et sin t =

L {sin t} = = =⇒ ,
s2 + 1 2 s2 + 1 (s − 1)2 + 1

if we use (7.7) and a ‘shift’ by a = 1.

Solution to exercise 7.2


We use the results from Section 7.1 (omitting references to the, by now, more familiar
ones).

(a) We have
 
s+1 s 1 2
= L {cos(2t)}+ 12 L {sin(2t)} = L cos(2t) + 12 sin(2t) ,

2
= 2 2
+
s +4 s +2 2 s + 22
2

and so this is the Laplace transform of the function cos(2t) + 21 sin(2t).

(b) We have  
5s + 2 1 2  2  2

=5 + = 5L {t} + L t = L 5t + t ,
s3 s2 s3
and so this is the Laplace transform of the function 5t + t2 .

252
7.2. Solutions to exercises

(c) We have

1 + s + s2 1 1  −t  −t

= + = L {t} + L e = L t + e ,
s2 (1 + s) s2 1 + s

and so this is the Laplace transform of the function t + e−t .

(d) We have

(s + 1) − 1
 
s s+1 1 3
2
= 2
= 2 2
− ,
s + 2s + 10 (s + 1) + 9 (s + 1) + 3 3 (s + 1)2 + 32

which, thinking of (7.7), is a shift by a = −1 of


   
s 1 3 1 1
− = L {cos(3t)} − L {sin(3t)} = L cos(3t) − sin(3t) ,
s2 + 3 2 3 s2 + 3 2 3 3

i.e. we have the Laplace transform of cos(3t) − 13 sin(3t) e−t .


 

Solution to exercise 7.3


Using (7.3), we see that the Laplace transform of the ODE will be given by

L {y 00 (t)} − L {y(t)} = L {t} ,

and so, if we let L {y(t)} (s) = ỹ(s) as usual, and use


7
(7.5) and the initial conditions, to get

L {y 00 (t)} = s2 ỹ(s) − sy(0) − y 0 (0) = s2 ỹ(s) − s(−1) − 0 = s2 ỹ(s) + s,

(7.2), to get
1
L {t} = ,
s2
we see that the Laplace transform of the ODE is
 
2 1
s ỹ(s) + s − ỹ(s) = 2 ,
s

in terms of ỹ(s). Tidying this up, we then have

1 1 − s3
(s2 − 1)ỹ(s) = −s =⇒ (s − 1)(s + 1)ỹ(s) = ,
s2 s2
and so, we find that
1 − s3
ỹ(s) = ,
(s − 1)(s + 1)s2
is our expression for ỹ(s). We can now rewrite this using partial fractions or, more
simply, we can just note that

1 − s3 = (1 − s)(1 + s + s2 ),

253
7. Laplace transforms

which means that we have


(1 − s)(1 + s + s2 ) 1 + s + s2
= −L t + e−t ,

ỹ(s) = 2
= − 2
(s − 1)(s + 1)s (s + 1)s

if we use our answer from Exercise 7.2(c). Consequently, we see that

y(t) = −t − e−t ,

is the sought after solution to the ODE.

Solution to exercise 7.4


We first notice that the integral equation we are given involves a convolution, i.e. if we
take g(t) = cos t, we have
Z t Z t
f (u) cos(t − u) du = f (u)g(t − u) du = (f ? g)(t),
0 0

and so, taking the Laplace transform of both sides of the equation we get
1 s
f˜(s) = 2 + 2f˜(s) · 2 ,
s s + 12
if we use the Convolution theorem and the results from Activities 7.1 and 7.4. As such,
7 we can see that the function we seek has a Laplace transform given by

s2 − 2s + 1 ˜ 2
 
2s 1 1 ˜(s) = s + 1 .
1− 2 f˜(s) = 2 =⇒ f (s) = =⇒ f
s +1 s s2 + 1 s2 s2 (s − 1)2
We can now rewrite this using partial fractions or, perhaps more simply, we can just
note that
s2 + 1 (s − 1)2 + 2s
=
s2 (s − 1)2 s2 (s − 1)2
1 2
= 2+
s s(s − 1)2
1 2s − 2(s − 1)
= 2+
s s(s − 1)2
1 2 2
= 2+ 2

s (s − 1) s(s − 1)
1 2 2s − 2(s − 1)
= 2+ 2

s (s − 1) s(s − 1)
1 2 2 2
= 2+ 2
− + .
s (s − 1) s−1 s
That is, we have
1 2 2 2
f˜(s) = 2 + 2
− + =⇒ f (t) = t + 2t et −2 et +2,
s (s − 1) s−1 s
is the sought after function.

254
7.2. Solutions to exercises

Solution to exercise 7.5


For a ∈ R, we see that the integral

ta
Z
J(a) = dt,
0 (1 + t)3

is an improper integral of the third kind with problems at t = 0 (if a < 0) and as
t → ∞. As such, we write it as
1 Z ∞
ta ta
Z
J(a) = dt + dt,
(1 + t)3 (1 + t)3
|0 {z } | 1
{z }
I1 I2

and note that J(a) converges if and only if both I1 and I2 converge. So, considering
these two integrals separately, we have:

For I1 , we note that for t close to zero, the integrand


ta ta
f (t) = ' = ta = g(t),
(1 + t)3 1

which gives us
f (t) ta 1 7
• lim+ = lim+ 3 a
= lim+ = 1 and
t→0 g(t) t→0 (1 + t) t t→0 (1 + t)3
Z 1 Z 1
• g(t) dt = ta dt which is convergent if and only if a + 1 > 0.
0 0
So, by the LCT, I1 is convergent if and only if a > −1.

For I2 , we note that for large t, the integrand


ta ta
f (t) = ' = ta−3 = g(t),
(1 + t)3 t3

which gives us
f (t) ta 1
• lim = lim 3 a−3
= lim 3 = 1 and
t→∞ g(t) t→∞ (1 + t) t t→∞ 1
+ 1
t
Z ∞ Z 1
• g(t) dt = ta−3 dt which is convergent if and only if a − 2 < 0.
1 0
So, by the LCT, I1 is convergent if and only if a < 2.
Consequently, we see that J(a) is convergent if and only if −1 < a < 2.

To show that
J(a) = B(1 + a, 2 − a),
whenever the integral J(a) converges, we can make the substitution

1 1 du
u= =⇒ t= − 1 so that dt = − 2 ,
1+t u u

255
7. Laplace transforms

as this gives us
Z ∞ 0 a     Z 1
ta

1 du
Z
3
J(a) = dt = −1 u − 2 = u1−a (1 − u)a du,
0 (1 + t)3 1 u u 0

so that, using the definition of the Beta function, we have

J(a) = B(2 − a, 1 + a),

as required. Observe, in particular, that this Beta function exists if 2 − a > 0 and
1 + a > 0 in agreement with what we found above about the convergence of J(a).

256
A

Appendix A
Sample examination paper

Important note: This sample examination paper reflects the intended examination
and assessment arrangements for this course in the academic year 2012–2013. The
format and structure of the examination may have changed since the publication of this
subject guide. You can find the most recent examination papers on the VLE where all
changes to the format of the examination are posted.

Further Calculus

Time allowed: TWO hours.

There are FIVE questions on this paper. You may attempt as many as you wish but
only your BEST FOUR answers will count towards your final mark. All questions
carry equal numbers of marks.

Calculators may not be used for this paper.

1. (a) Show that


(1 − eh )
lim+ = −1,
h→0 h
h
and hence find lim+ (1 − e ) ln h.
h→0

(b) Determine the limits


Z t Z t
−t2 s2 −t2 2
lim e e ds and lim t e es ds.
t→∞ 0 t→∞ 0

(c) Determine whether either of the integrals


Z ∞ ∞
(1 + 2x)3
Z
dx and (ln x) e−2x dx,
1 x4 + 5x6 1
is convergent.

(d) Verify that Z 1


sin(xp ) dx,
0
is convergent for all values of p.
[Hint: Consider the cases p > 0, p = 0 and p < 0 separately. The substitution y = xp
may be useful.]

257
A. Sample examination paper
A
2. (a) Find the third-order Taylor series about x = 0 of the function ln(1 + x).

Use Taylor’s theorem to show that, for x > 0,


1
x − x2 < ln(1 + x) < x,
2
and use these inequalities to find

ln(1 + x)
lim+ .
x→0 x
Hence determine the values of q ∈ R for which the integral
1/2
ln(1 + x)
Z
dx,
0 xq
is convergent.

(b) Use the transformation where x = r cos θ and y = r sin θ to evaluate the integral
Z 1 Z √ 2−y 2
dy − ln(x2 + y 2 ) dx.
y=0 x=y
R
[Hint: You may use the facts that ln z dz = z ln z − z + c and that lim+ z ln z = 0.]
z→0

3. (a) By considering the upper and lower estimates for the integral
Z n
dt
,
1 t

with the partition P = {1, 2, ..., n − 1, n} of the interval [1, n], deduce that
1 1 1 1 1 1
1+ + + ··· + ≥ ln n ≥ + + · · · + .
2 3 n−1 2 3 n
By considering the interval [1, 1 + n1 ] and using a similar approach, show that
 
1 1
≤ ln 1 + .
n+1 n

For n = 1, 2, ..., define γn to be


1 1 1
γn = 1 + + + · · · + − ln n,
2 3 n
and deduce that, for these values of n, 1 ≥ γn ≥ 0 and γn ≥ γn+1 .

258
A
(b) Let I be the double integral

dx dy
Z Z
I= ,
D x2 y
where the region D is bounded by the four lines y = x, y = 2x, y = 1 − x and y = 2 − x.

(i) Find the inverse of the transformation


y
u(x, y) = x + y and v(x, y) = ,
x
which is defined for x 6= 0.

(ii) Sketch the region D and write down, in terms of the variables x and y, the
inequalities which determine this region.

(iii) Hence find, in terms of the variables u and v, the inequalities that determine the
image, ∆, of D in the uv-plane under the transformation given in (i). Sketch the
region ∆.
∂(u, v)
(iv) Find the Jacobian, , of the transformation given in (i).
∂(x, y)
(v) Hence show that
v+1
Z Z
I= du dv,
∆ u2 v
and use this to evaluate I.

4. (a) The function f is given by


t2 2
e−tx
Z
f (t) = dx.
1 x

Use the Leibniz rule to find f 0 (t).


2
Hence show that f 0 (1) = and find lim+ f 0 (t).
e t→0

(b) Show that, for β > 0, the function f (t) defined by f (0) = 0 and

f (t) = tβ ln t,

for t > 0 is continuous at t = 0. Also show that this function is of exponential growth at
most γ for each γ > 0.

259
A. Sample examination paper
A Z ∞
(c) The Gamma function is given by Γ(α) = tα−1 e−t dt for α > 0.
0

(i) Use integration by parts to show that



1
Z
Γ(α) = tα e−t dt,
α 0

for α > 0.
(ii) Differentiate the result in (i) with respect to α to show that
Z ∞
0
Γ (1) = (t ln t − t) e−t dt.
0

You should refer to part (b) to justify any manipulations.


Z ∞
0
(iii) Hence use integration by parts to deduce that Γ (1) = (ln u) e−u du.
0

(iv) For s > 0, use the substitution u = st in (iii) to deduce that


ln s Γ0 (1)
L {ln t} = − + .
s s

5. (a) What is the Laplace transform f˜(s) = L{f (t)} of a function f (t) defined for
t ≥ 0? To what class of functions is this transform customarily applied?

(b) Given that


L{f 0 (t)} = sL{f (t)} − f (0),
use f (t) = 1 to deduce that L{1} = 1/s.

Also use this given result to show that


L{f 00 (t)} = s2 L{f (t)} − sf (0) − f 0 (0),
and hence deduce that L{t} = 1/s2 .

(c) Use the results in (b) to solve the differential equation


f 00 (t) + 4f 0 (t) + 4f (t) = t,
given that f (0) = f 0 (0) = 1.
Z 1
(d) The Beta function is given by B(p, q) = up−1 (1 − u)q−1 du for p, q > 0.
0
Use the substitution v = u/t to derive the identity
Z t Z 1
p−1 q−1 p+q−1
u (t − u) du = t v p−1 (1 − v)q−1 dv.
0 0

State the Convolution theorem and hence show that, for p, q > 0,
Γ(p)Γ(q)
B(p, q) = .
Γ(p + q)
[Hint: You may use the fact that L {tα } = s−1−α Γ(1 + α) for α > −1.]

260
Appendix B B
Solutions to the sample examination
paper

Question 1.

(a) For the first limit, since the numerator and the denominator both tend to zero as
h → 0+ , we can use L’Hôpital’s rule to see that
(1 − eh ) − eh
lim+ = lim+ = −1,
h→0 h h→0 1
as required. For the second limit, we can write
(1 − eh )
  
h ln h
(1 − e ) ln h = ,
h 1/h
where in the last term in this product, we observe that the numerator and the
denominator both tend to infinity as h → 0+ and so we can use L’Hôpital’s rule to see
that
ln h 1/h
lim+ = lim+ = lim (−h) = 0.
h→0 1/h h→0 −1/h2 h→0+

This means that, overall, we have


(1 − eh )
  
h ln h
lim (1 − e ) ln h = lim+ = (−1)(0) = 0,
h→0+ h→0 h 1/h
as the value of the second limit.

(b) For the first limit, we can write


Z t
2
Z t es ds
2 2
lim e−t es ds = lim 0
,
t→∞ 0 t→∞ et2
and so, as the numerator and the denominator both tend to infinity as t → ∞, we can
use L’Hôpital’s rule and the FTC to see that
Z t
2
es ds 2
0 et 1
lim 2 = lim 2 = lim = 0.
t→∞ et t→∞ 2t et t→∞ 2t

For the second limit, we have


Z t
2
Z t t es ds
2 2
lim t e−t es ds = lim 0
,
t→∞ 0 t→∞ et2

261
B. Solutions to the sample examination paper

and so, as the numerator and the denominator both tend to infinity as t → ∞, we can
use L’Hôpital’s rule, the product rule and the FTC to see that
B Z t
s2 t2
Z t
s2
 Z t
s2

t e ds te + e ds 1 e ds 
0 0 1  0  = 1 + (0)(0),
lim = lim = lim  + 
t→∞ et2 t→∞ 2t et2 t→∞  2 2t  et2  2

if we use the previous answer. Consequently, we see that


t
1
Z
−t2 2
lim t e es ds = ,
t→∞ 0 2

is the final answer.


(c) For the first integral, we have

1 3 1
3
x3

(1 + 2x)3 x
+2 x
+2
f (x) = 4 = 1
 = 1
,
x + 5x6 x6 x2
+5 x3 x2
+5

and so we take g(x) = 1/x3 . With this, we have

f (x) ( 1 + 2)3 8
lim = lim x1 = and
x→∞ g(x) x→∞ ( 2 + 5) 5
x

∞ ∞
dx
Z Z
g(x) dx = is convergent,
1 1 x3
and so the given integral is also convergent by the LCT.

For (ii), we have f (x) = (ln x) e−2x and so we take g(x) = e−x so that we have

f (x) (ln x) e−2x ln x


lim = lim = lim ,
x→∞ g(x) x→∞ e−x x→∞ ex

where, as the numerator and the denominator both tend to infinity as x → ∞, we can
use L’Hôpital’s rule to see that

ln x 1/x 1
lim x
= lim x = lim = 0.
x→∞ e x→∞ e x→∞ x ex

As such, because the integral


Z ∞ Z ∞
g(x) dx = e−x dx,
1 1

is convergent, so is the given integral by the LCT.

(d) When p = 0 the integrand is constant and so the integral is convergent in this case.
For the cases where p 6= 0, we make the substitution y = xp so that x = y 1/p and
1
dx = p1 y p −1 dy to see that:

262
When p > 0, we have
1 1
1
Z Z
1

0
sin(x ) dx = p
p 0
y p −1 sin y dy. B
We now note that near y = 0, the integrand is positive and, if we take g(y) = y 1/p ,
then we have
f (y) sin y
lim+ = lim+ = 1,
y→0 g(y) y→0 y
and so, as the integral
Z 1 Z 1
g(y) dy = y 1/p dy,
0 0

is convergent (because 1/p > 0), by the LCT, the given integral is convergent when
p > 0.

When p < 0, we have


1 1 ∞
1 1
Z Z Z
1 1
−1
p
sin(x ) dx = y p sin y dy = − y p −1 sin y dy.
0 p ∞ p 1

However, as | sin y| ≤ 1 for all y ∈ R, we have


1 1
p −1
y sin y ≤ y p −1 ,

for all y ≥ 1. Consequently, as p < 0, we see that


Z ∞ h 1 i∞
1
−1
y p dy = py p = p,
1 1

is finite and so the given integral is absolutely convergent when p < 0.


Thus we see that the given integral is indeed convergent for all values of p, as required.

Question 2.

(a) With f (x) = ln(1 + x), we have f (0) = 0 and

1
f 0 (x) = =⇒ f 0 (0) = 1,
1+x
−1
f 00 (x) = =⇒ f 00 (0) = −1,
(1 + x)2
2
f 000 (x) = 3
=⇒ f 000 (0) = 2,
(1 + x)

and so the third-order Taylor series of ln(1 + x) about x = 0 is

1 1
ln(1 + x) = 0 + x − x2 + x3 + · · · .
2 3
If we now fix x > 0, Taylor’s theorem implies that:

263
B. Solutions to the sample examination paper

For some 0 < c < x, we have


1
ln(1 + x) = 0 + x + x2 f 00 (c) < x,
B 2
since f 00 (c) < 0.
For some 0 < d < x, we have
1 1 1
ln(1 + x) = 0 + x − x2 + x3 f 000 (d) > x − x2 ,
2 3 2
since f 000 (d) > 0.
Thus, for x > 0, we have shown that
1
x − x2 < ln(1 + x) < x,
2
as required. But, of course, for x > 0, this gives us
1 ln(1 + x)
1− x< < 1,
2 x
and so, by the Sandwich theorem, we have
ln(1 + x)
lim+ = 1.
x→0 x
We now consider the integral
1/2
ln(1 + x)
Z
dx,
0 xq
whose integrand is
ln(1 + x)
h(x) = ,
xq
which means that, if we take g(x) = x1−q , we have
h(x) ln(1 + x) ln(1 + x)
lim+ = lim+ q 1−q = lim+ = 1,
x→0 g(x) x→0 x x x→0 x
using the limit that we have just found. Furthermore, as the integral
Z 1/2 Z 1/2
g(x) dx = x1−q dx,
0 0

is convergent if and only if 2 − q > 0, we can use the LCT to see that the given integral
converges if and only if q < 2.

(b) The Jacobian of the transformation where x = r cos θ and y = r sin θ is given by
   
∂(x, y) xr xθ cos θ −r sin θ
= det = det = r cos2 θ + r sin2 θ = r,
∂(r, θ) yr yθ sin θ r cos θ
as we saw when we discussed polar coordinates. The domain
p of integration can be found
by taking a fixed 0 ≤ y ≤ 1 and then taking y ≤ x ≤ 2 − y 2 where we note that any
point (x, y) satisfying
p
x = 2 − y 2 must satisfy the equation x2 + y 2 = 2,

264

which means that it lies on a circle of radius 2 centred on the origin. Putting this all
together gives us the region of integration
√ illustrated in Figure B.1 and this is
represented by the inequalities 0 ≤ r ≤ 2 and 0 ≤ θ ≤ π/4 in polar coordinates.
Consequently, if we then write the integrand as
B
x2 + y 2 = r2 cos2 θ + r2 sin2 θ = r2 (sin2 θ + cos2 θ) = r2 ,

the given integral becomes


Z 1 Z √ 2−y 2 Z π/4 Z √
2
!
dy − ln(x2 + y 2 ) dx = −r ln(r2 ) dr dθ.
y=0 x=y θ=0 r=0

To evaluate the r-integral we use the substitution z = r2 so that dz = 2r dr and



2 2
1
Z Z
2
−r ln(r ) dr = − ln z dz = [z ln z − z]2z=0 ,
r=0 2 z=0

if we use the first result in the hint. Now, of course, z ln z is undefined at z = 0 and so
we need to reinterpret this as
 2    
2
[z ln z − z]z=0 = lim+ z ln z − z = 2 ln 2 − 2 − lim+ ε ln ε − ε = 2 ln 2 − 2,
ε→0 z=ε ε→0

if we use the second result in the hint. This leaves us with


Z 1 Z √2−y2 Z π/4   π/4
2 ln 2 − 2

2 2 π
dy − ln(x +y ) dx = − dθ = (1−ln 2) θ = (1−ln 2),
y=0 x=y θ=0 2 θ=0 4

as the final answer.

y

2
x
=
y

π
4
O √ x
2

Figure B.1: A sketch of the region of integration for Question 2(b).

265
B. Solutions to the sample examination paper

Question 3.

(a) We are given the integral


B Z n
dt
,
1 t
and the partition P = {1, 2, ..., n − 1, n} of [1, n]. Here the integrand is 1/t and this is a
decreasing function over the interval [1, n] which means that, possibly with the aid of a
sketch, you should be able to see that the lower estimates of the integral are given by
1 1 1
L(P) = + + ··· + ,
2 3 n
whereas the upper estimates are given by
1 1 1
U (P) = 1 + + + ··· + .
2 3 n−1
Hence, as the definition of the Riemann integral assures us that
Z n
dt
L(P) ≤ ≤ U (P),
1 t
and we know that n  n
dt
Z
= ln t = ln n − ln 1 = ln n,
1 t 1
we can see that we have
1 1 1 1 1 1
+ + · · · + ≤ ln n ≤ 1 + + + · · · + ,
2 3 n 2 3 n−1
as required. Similarly, on the interval [1, 1 + n1 ], we can look at the integral given by
1
1+ n  1+ n1    
dt 1 1
Z
= ln t = ln 1 + − ln 1 = ln 1 + ,
1 t 1 n n
and, as the integrand is still the decreasing function 1/t, the lower estimate is given by
 
1 1 1
L(Q) = 1 = ,
n 1+ n n+1

if we take Q to be the partition {1, 1 + n1 } of the interval [1, 1 + n1 ]. Hence, as the


definition of the Riemann integral again assures us that
Z 1+ 1
n dt
L(P ) ≤ ,
1 t
we can see that we have  
1 1
≤ ln 1 + ,
n+1 n
as required.

We now define γn to be
1 1 1
γn = 1 + + + · · · + − ln n,
2 3 n

266
for n = 1, 2, ... so that, using the inequality
1 1 1 1 1 1
+ + · · · + ≤ ln n ≤ 1 + + + · · · + , B
2 3 n 2 3 n−1
from above, we see that
1 1 1 1 1 1
+ + · · · + ≤ ln n =⇒ + + · · · + − ln n ≤ 0,
2 3 n 2 3 n
so that we get
1 1 1
γn = 1 + + + · · · + − ln n ≤ 1 =⇒ 1 ≥ γn ,
2 3 n
and we also see that
1 1 1 1 1 1
ln n ≤ 1 + + + ··· + =⇒ 0≤1+ + + ··· + − ln n,
2 3 n−1 2 3 n−1
so that we get
1 1 1 1 1 1
≤ 1 + + + ··· + + − ln n = γn =⇒ γn ≥ ≥ 0,
n 2 3 n−1 n n
as n > 0. That is, for n = 1, 2, ..., we have shown that

1 ≥ γn ≥ 0,

as required. And, finally, we saw above that


 
1 1
≤ ln 1 + ,
n+1 n
which means that
 
1 n+1 1
≤ ln = ln(n + 1) − ln n =⇒ − ln(n + 1) ≤ − ln n,
n+1 n n+1

and so, we also have


1 1 1 1 1 1 1
γn+1 = 1 + + + ··· + + − ln(n + 1) ≤ 1 + + + · · · + − ln n = γn ,
2 3 n n+1 2 3 n
which gives us γn ≥ γn+1 , as required.

(b) We are given the double integral


dx dy
Z Z
I= ,
D x2 y
where the region D is bounded by the four lines y = x, y = 2x, y = 1 − x and y = 2 − x.

For (i), we are asked to find the inverse of the transformation


y
u(x, y) = x + y and v(x, y) = ,
x

267
B. Solutions to the sample examination paper

which is defined for x 6= 0. To do this, we see that y = vx and so


u
u = x + y = x + vx = (1 + v)x =⇒ x= ,
B 1+v
as long as v + 1 6= 0. Then, using y = vx again, we see that
uv
y= .
1+v
That is, for v 6= −1, we have found that
u uv
x(u, v) = and y(u, v) = ,
1+v 1+v
is the inverse of the given transformation.

For (ii), as illustrated in Figure B.2(a), we sketch the lines y = x, y = 2x, y = 1 − x and
y = 2 − x so that we can determine the region D. Indeed, using this sketch, we can see
that the inequalities that determine this region are 1 ≤ x + y ≤ 2 and x ≤ y ≤ 2x.

For (iii), we can use the given transformation to see that

1≤x+y ≤2 =⇒ 1 ≤ u ≤ 2,

and that, for x > 0, which is certainly the case in D, we have


y
x ≤ y ≤ 2x =⇒ 1≤ ≤2 =⇒ 1 ≤ v ≤ 2.
x
That is, the image of D in the uv-plane is the region ∆ that satisfies the inequalities

1 ≤ u ≤ 2 and 1 ≤ v ≤ 2,

as illustrated in Figure B.2(b).

For (iv) The Jacobian for the transformation given in (i) is


!
1 1
 
∂(u, v) ux uy 1 y x+y
= det = det = + 2 = ,
∂(x, y) vx vy 2
−y/x 1/x x x x2

and this is well-defined as long as x 6= 0.

For (v), we see that under the given transformation, the integrand
2  2
(v + 1)3

1 v+1 v+1
f (x, y) = 2 becomes = ,
xy u uv u3 v
and, using the Jacobian that we found in (iv), we also have
−1 2  
x2
 
∂(x, y) ∂(u, v) u 1 u
= = = = ,
∂(u, v) ∂(x, y) x+y v+1 u (v + 1)2
which is certainly positive given that u ≥ 1 in ∆. Consequently, applying the change of
change of variable formula, we see that
(v + 1)3
Z Z   
dx dy u v+1
Z Z Z Z
I= 2
= 3 2
dudv = 2
dudv,
D x y ∆ uv (v + 1) ∆ u v

268
as required. Then, of course, using the information about ∆ that we found in (iii), we
see that this gives us
Z 2 Z 2
v+1
 Z 2
du
 Z 2 
1
  B
I= 2
dv du = 2
1+ dv
u=1 v=1 u v u=1 u v=1 v
 2  2   
1 1
= − v + ln v = − +1 [2 + ln 2] − [1 + ln 1]
u u=1 v=1 2
1
= (1 + ln 2),
2
as the value of I.

y v

2
2x

2
y=


x
=
y

1 1
y

D
=
2

x
y
=
1

x

O x O u
1 2 1 2
(a) (b)

Figure B.2: The sketches for Question 3(b). (a) The region of integration, D, for (ii) and
(b) its image under the given transformation for (iii).

Question 4.

(a) Given that


t2 2
e−tx
Z
f (t) = dx,
1 x
the Leibniz rules tells us that
!
2 )2 t2 2 t2
e−t(t e−tx 2
Z Z
0 5 2
f (t) = (2t) + (−x ) 2
dx = e−t − x e−tx dx.
t2 1 x t 1

Of course, as
Z 1
2
x e−tx dx = 0,
1

this means that


1
2
Z
0 −1 2
f (1) = 2 e − x e−x dx = ,
1 e

269
B. Solutions to the sample examination paper

as required. Then, to find the limit of f 0 (t) as t → 0+ , we note that


" #t2
t2 2 5 5
e−tx e−t e−t e−t − e−t
Z
B xe −tx2
dx = = − =− ,
1 −2t −2t −2t 2t
1

so that 5 5
2 5 e−t − e−t 5 e−t − e−t
0
f (t) = e−t + = ,
t 2t 2t
which means that 5
5 e−t − e−t
0
lim+ f (t) = lim+ = ∞,
t→0 t→0 2t
because, as t → 0+ , the numerator tends to four whereas the denominator is positive
and tends to 0.

(b) To show that, for β > 0, the function f (t) defined by f (0) = 0 and

f (t) = tβ ln t,

is continuous, we need to verify that

lim f (t) = f (0).


t→0+

To do this, we note that as β > 0, we have


ln t
lim+ f (t) = lim+ tβ ln t = lim+ ,
t→0 t→0 t→0 t−β
and this is a quotient where both the numerator and denominator tend to infinity [in
magnitude] as t → 0+ . So, applying L’Hôpital’s rule, we have

ln t t−1 tβ
lim+ f (t) = lim+ = lim = lim = 0,
t→0 t→0 t−β t→0+ −βt−β−1 t→0+ −β
as β > 0. That is, we have shown that

lim f (t) = 0 = f (0),


t→0+

and so f (t) is continuous at t = 0 as claimed.

To see that the function f (t) is of exponential growth at most γ for each γ > 0, we need
to show that for any γ > 0, we have

|f (t)| ≤ M eγt ,

for some M > 0. To do this, we first note that, for any γ > 0, we have
tβ ln t tβ
  
ln t
lim = lim γt/2 lim = (0)(0) = 0,
t→∞ eγt t→∞ e t→∞ eγt/2

since, as t → ∞, eγt/2 tends to infinity much faster than either tβ or ln t. This means
that there is some value of t, say T , for which we can guarantee that

tβ ln t < eγt ,

270
provided that t > T . Secondly, as we have established that f (t) is continuous at t = 0,
we can now see that it is, in fact, continuous for all t ≥ 0 since the functions tβ and ln t
are both continuous for t > 0. In particular, this means that for 0 ≤ t ≤ T , we can find
the maximum value of |f (t)| over this interval and denote it by MT . As such, since
B
eγt ≥ 1 for any t ≥ 0, we have
|f (t)| ≤ M eγt ,
where M = max{1, MT }. Consequently, for t ≥ 0, f (t) does indeed have exponential
growth at most γ for any γ > 0.

(c) We are given that the Gamma function is defined by


Z ∞
Γ(α) = tα−1 e−t dt,
0

when α > 0.

For (i), we use integration by parts to see that, for α > 0, we have
Z ∞ ∞ Z ∞ α
1 ∞ α −t
 α
t −t t
Z
α−1 −t −t
Γ(α) = t e dt = e − (− e ) dt = t e dt,
0 α 0 0 α α 0

since, given that α > 0, we have tα e−t = 0 when t = 0 and we also have

lim tα e−t = 0,
t→∞

because, as t → ∞, e−t will tend to zero faster than tα tends to infinity.

For (ii), with α > 0, we differentiate the result in (i) with respect to α to see that

d 1 ∞ α −t
 Z  Z ∞  Z ∞ 
0 1 α −t 1 d α −t
Γ (α) = t e dt = − 2 t e dt + t e dt .
dα α 0 α 0 α dα 0

Now, provided that we can justify the manipulation (see below), the derivative in the
second term here can be written as
 Z ∞  Z ∞   Z ∞
d α −t ∂ α −t
t e dt = t e dt = tα (ln t) e−t dt,
dα 0 0 ∂α 0

since we have tα = eα ln t and, this in turn would then give us


Z ∞
1 ∞ α
Z ∞
1 1
Z
0 α −t −t
Γ (α) = − 2 t e dt + t (ln t) e dt = 2 (α ln t − 1)tα e−t dt,
α 0 α 0 α 0
so that, taking α = 1, this would give us
Z ∞
0
Γ (1) = (t ln t − t) e−t dt,
0

as required.

271
B. Solutions to the sample examination paper

Of course, we still have to provide a justification for the manipulation we have used and
to do this we note that:
B The function tα e−t is clearly jointly continuous for α > 0 and t ≥ 0.

The function tα (ln t) e−t is jointly continuous for α > 0 and t ≥ 0 since, for any
fixed 0 < a < α, we can see that the function
a a
tα (ln t) e−t = tα− 2 t 2 ln t e−t ,


is the product of three functions which are jointly continuous, continuous (using
what we saw in (b) for a fixed value of β given by a/2) and continuous respectively.

The function tα (ln t) e−t has dominated convergence for all t ≥ 0 and α > 0
because, as we saw in (b), the function tα (ln t) with α > 0 is of exponential growth
at most γ for any γ > 0. That is, we have
• |tα (ln t) e−t | ≤ M eγt e−t = M e(γ−1)t = k(t), and
Z ∞ Z ∞
• k(t) dt = M e(γ−1)t dt is convergent if we take 0 < γ < 1,
0 0
and this establishes the required dominated convergence.

For (iii), we note that


   
d 1
t ln t − t = (1) ln t + t − 1 = ln t + 1 − 1 = ln t,
dt t
and so, again using integration by parts, we see that
Z ∞   −t ∞ Z ∞  −t  Z ∞
0 −t e e
Γ (1) = (t ln t−t) e dt = (t ln t − t) − (ln t) dt = (ln t) e−t dt,
0 −1 0 0 −1 0

since, from (b), the continuity of f (t) guarantees that t ln t → 0 as t → 0+ whereas


t ln t − t
lim (t ln t − t) e−t = lim ,
t→∞ t→∞ et
and in this quotient, both the numerator and the denominator tend to infinity as
t → ∞, which means that we can apply L’Hôpital’s rule to see that
ln t
lim (t ln t − t) e−t = lim = lim (ln t) e−t = 0,
t→∞ t→∞ et t→∞

because, as t → ∞, e−t will tend to zero faster than ln t tends to infinity. Consequently,
we see that, replacing the dummy variable t with the dummy variable u in the integral,
we now have Z ∞
0
Γ (1) = (ln u) e−u du,
0
as required.

For (iv), we now take a fixed s > 0 and make the substitution u = st in (iii) so that
du = s dt and Z ∞ Z ∞
0 −st
Γ (1) = ln(st) e s dt = s ln(st) e−st dt,
0 0

272
so we have
∞ ∞ ∞
Γ0 (1)
Z Z Z
−st −st
= (ln s + ln t) e dt = (ln s) e dt + (ln t) e−st dt.
s 0 0 0 B
But, of course, if s > 0, we have
Z ∞  −st ∞
−st e 1 1
e dt = =0− = ,
0 −s 0 −s s

whereas, by the definition of the Laplace transform, we also have


Z ∞
L {ln t} = (ln t) e−st dt,
0

so, putting this all together, we get


Γ0 (1) ln s ln s Γ0 (1)
= + L {ln t} =⇒ L {ln t} = − + ,
s s s s
as required.

Question 5.

(a) The Laplace transform of a function f (t) defined for t ≥ 0 is given by


Z ∞
˜
f (s) = L{f (t)} = e−st f (t) dt,
0

and this transform is customarily applied to functions of exponential growth at most γ.

(b) We are told that


L{f 0 (t)} = sL{f (t)} − f (0),
and, taking f (t) = 1 as instructed, we can see that f (0) = 1 and f 0 (t) = 0 so that
1
L{0} = sL{1} − 1 =⇒ L{1} = ,
s
as L {0} = 0. We can also see that this result also gives us
 
L{f (t)} = sL{f (t)} − f (0) = s sL{f (t)} − f (0) − f 0 (0) = s2 L{f (t)} − sf (0) − f 0 (0),
00 0 0

as required. So that, lastly, if we take f (t) = t so that f (0) = 0, f 0 (t) = 1, f 0 (0) = 1 and
f 00 (t) = 0, we can see that this last result and L {0} = 0 gives us
1
0 = s2 L {t} − s(0) − 1 =⇒ L {t} = ,
s2
as required.

(c) Using the results in (b), we see that the Laplace transform of

f 00 (t) + 4f 0 (t) + 4f (t) = t,

273
B. Solutions to the sample examination paper

with f (0) = 1 and f 0 (0) = 1 is given by


 
1
B s f (s) − s(1) − 1 + 4(sf˜(s) − 1) + 4f˜(s) = 2 .

s
This can be written as
1 s3 + 5s2 + 1 s3 + 5s2 + 1
(s2 +4s+4)f˜(s) = 2 +s+5 =⇒ (s+2)2 f˜(s) = =⇒ f˜(s) = 2 ,
s s2 s (s + 2)2
and this is the Laplace transform of the function f (t) that we are seeking. To find this
function we can use partial fractions to find the constants A, B, C and D which make
s3 + 5s2 + 1 A B C D
= + + + .
s2 (s + 2)2 s s2 s + 2 (s + 2)2
Indeed, this means that we need

s3 + 5s2 + 1 = As(s + 2)2 + B(s + 2)2 + Cs2 (s + 2) + Ds2 ,

and so, straightaway we can see that with s = 0 we get B = 1/4 and with s = −2 we
get D = 13/4. Now, to find the other two constants, we can compare the s coefficients
on both sides to see that 0 = 4A + 4B which means that A = −1/4 and then,
comparing the s3 coefficients on both sides we find that 1 = A + C which means that
C = 5/4. Thus, we have found that
−1/4 1/4 5/4 13/4
f˜(s) = + 2 + + ,
s s s + 2 (s + 2)2
which allows us to see that
1 t 5 13
f (t) = − + + e−2t + t e−2t ,
4 4 4 4
if we use the facts that L {eat } = 1/(s − a) and L {f (t) eat } = f˜(s − a).

(d) We are given that the Beta function is defined by


Z 1
B(p, q) = up−1 (1 − u)q−1 du,
0

for p, q > 0.

To establish the identity given in the question, we start with the integral on the
left-hand side and use the substitution v = u/t, as directed, to see that (for fixed t) we
have u = tv and du = t dv so that
Z t Z 1 Z 1
p−1 q−1 p−1 q−1 p+q−1
u (t − u) du = (tv) (t − tv) t dv = t v p−1 (1 − v)q−1 dv,
0 0 0

as required.

The Convolution theorem states that


Z t 
L f (u)g(t − u) du = f˜(s)g̃(s).
0

274
Now, for the last part of the question, we note that the identity which we derived above
can be written as Z t
up−1 (t − u)q−1 du = tp+q−1 B(p, q),
0
B
for p, q > 0 and so, if we take the Laplace transform of both sides, we get
Z t 
u (t − u) du = B(p, q)L tp+q−1 ,
p−1 q−1

L
0

as B(p, q) does not depend on t. This means that, applying the Convolution theorem to
the left-hand side, we have
  
 p−1  q−1  p+q−1 Γ(p) Γ(q) Γ(p + q)
L t L t = B(p, q)L t =⇒ p q
= B(p, q) p+q ,
s s s

if we use the fact given in the hint which says that, for α > −1,

Γ(1 + α)
L {tα } = .
s1+α
Of course, this can then be rearranged to give us

Γ(p)Γ(q)
B(p, q) = ,
Γ(p + q)

for p, q > 0, as required.

275
B. Solutions to the sample examination paper

276
Comment form
We welcome any comments you may have on the materials which are sent to you as part of your
study pack. Such feedback from students helps us in our effort to improve the materials produced
for the International Programmes.
If you have any comments about this guide, either general or specific (including corrections,
non-availability of Essential readings, etc.), please take the time to complete and return this form.

Title of this subject guide:

Name
Address

Email
Student number
For which qualification are you studying?

Comments

Please continue on additional sheets if necessary.

Date:

Please send your completed form (or a photocopy of it) to:


Publishing Manager, Publications Office, University of London International Programmes,
Stewart House, 32 Russell Square, London WC1B 5DN, UK.

Potrebbero piacerti anche