Mathematics

UNDERSTANDING MATHEMATICAL PROOF

The notion of proof is central to mathematics yet it is one of the most difficult
aspects of the subject to master. Understanding Mathematical Proof
describes the nature of mathematical proof, explores the various techniques
that mathematicians adopt to prove their results, and offers advice and
strategies for constructing proofs. It will improve your ability to understand
proofs and construct correct proofs of your own. UNDERSTANDING
The first chapter of the text introduces the kind of reasoning that mathemati-
cians use when writing their proofs and gives some example proofs to set
the scene. The book then describes basic logic to enable an understand-
ing of the structure of both individual mathematical statements and whole
MATHEMATICAL
mathematical proofs. It also explains the notions of sets and functions and
dissects several proofs with a view to exposing some of the underlying fea-
tures common to most mathematical proofs. The remainder of the book
PROOF
delves further into different types of proof, including direct proof, proof using
contrapositive, proof by contradiction, and mathematical induction. The au-
thors also discuss existence and uniqueness proofs and the role of counter
examples.

Features
• Presents a rigorous yet accessible introduction to one of the defining
concepts of mathematics
• Explores the structure of mathematical proof
• Classifies different types of proof
• Develops all the standard techniques for proving mathematical
theorems
• Provides hints and guidance on finding the appropriate proof
• Includes many elementary examples and, in addition, develops
proofs in linear algebra, group theory, and real analysis that will be of
Taylor • Garnier John Taylor
particular interest to undergraduate mathematicians
Rowan Garnier

K15018

K15018_Cover.indd 1 2/4/14 10:37 AM

 UNDERSTANDING
MATHEMATICAL
PROOF

K15018_FM.indd 1 2/3/14 5:33 PM

K15018_FM.indd 2 2/3/14 5:33 PM
 UNDERSTANDING
MATHEMATICAL
PROOF

John Taylor
Rowan Garnier

K15018_FM.indd 3 2/3/14 5:33 PM

CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2014 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Version Date: 20140131

International Standard Book Number-13: 978-1-4665-1491-1 (eBook - PDF)

This book contains information obtained from authentic and highly regarded sources. Reasonable
efforts have been made to publish reliable data and information, but the author and publisher cannot
assume responsibility for the validity of all materials or the consequences of their use. The authors and
publishers have attempted to trace the copyright holders of all material reproduced in this publication
and apologize to copyright holders if permission to publish in this form has not been obtained. If any
copyright material has not been acknowledged please write and let us know so we may rectify in any
future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information stor-
age or retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access www.copy-
right.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222
Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that pro-
vides licenses and registration for a variety of users. For organizations that have been granted a pho-
tocopy license by the CCC, a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
Contents

Preface ix

List of Figures xi

List of Tables xv

List of Symbols xvii

1 Introduction 1
1.1 The need for proof . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The language of mathematics . . . . . . . . . . . . . . . . . . 3
1.3 Reasoning . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Deductive reasoning and truth . . . . . . . . . . . . . . . . . 8
1.5 Example proofs . . . . . . . . . . . . . . . . . . . . . . . . . 11

2 Logic and Reasoning 15

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Propositions, connectives, and truth tables . . . . . . . . . . 16
2.3 Logical equivalence and logical implication . . . . . . . . . . 34
2.4 Predicates and quantification . . . . . . . . . . . . . . . . . . 45
2.5 Logical reasoning . . . . . . . . . . . . . . . . . . . . . . . . 66

3 Sets and Functions 83

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.2 Sets and membership . . . . . . . . . . . . . . . . . . . . . . 83
3.3 Operations on sets . . . . . . . . . . . . . . . . . . . . . . . . 92

v
vi

3.4 The Cartesian product . . . . . . . . . . . . . . . . . . . . . 103

3.5 Functions and composite functions . . . . . . . . . . . . . . . 107
3.6 Properties of functions . . . . . . . . . . . . . . . . . . . . . 117

4 The Structure of Mathematical Proofs 129

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.2 Some proofs dissected . . . . . . . . . . . . . . . . . . . . . . 130
4.3 An informal framework for proofs . . . . . . . . . . . . . . . 135
4.4 Direct proof . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.5 A more formal framework . . . . . . . . . . . . . . . . . . . . 161

5 Finding Proofs 175

5.1 Direct proof route maps . . . . . . . . . . . . . . . . . . . . . 175
5.2 Examples from sets and functions . . . . . . . . . . . . . . . 189
5.3 Examples from algebra . . . . . . . . . . . . . . . . . . . . . 200
5.3.1 Group theory . . . . . . . . . . . . . . . . . . . . . . . 200
5.3.2 Linear algebra . . . . . . . . . . . . . . . . . . . . . . 209
5.4 Examples from analysis . . . . . . . . . . . . . . . . . . . . . 226
5.4.1 Sequences . . . . . . . . . . . . . . . . . . . . . . . . . 227
5.4.2 Limits of functions . . . . . . . . . . . . . . . . . . . . 232

6 Direct Proof: Variations 239

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
6.2 Proof using the contrapositive . . . . . . . . . . . . . . . . . 240
6.3 Proof of biconditional statements . . . . . . . . . . . . . . . 246
6.4 Proof of conjunctions . . . . . . . . . . . . . . . . . . . . . . 251
6.5 Proof by contradiction . . . . . . . . . . . . . . . . . . . . . 257
6.6 Further examples . . . . . . . . . . . . . . . . . . . . . . . . 260
6.6.1 Examples from algebra . . . . . . . . . . . . . . . . . . 260
6.6.2 Examples from analysis . . . . . . . . . . . . . . . . . 263
 vii

7 Existence and Uniqueness 269

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
7.2 Constructive existence proofs . . . . . . . . . . . . . . . . . . 270
7.3 Non-constructive existence proofs . . . . . . . . . . . . . . . 276
7.4 Counter-examples . . . . . . . . . . . . . . . . . . . . . . . . 286
7.5 Uniqueness proofs . . . . . . . . . . . . . . . . . . . . . . . . 295

8 Mathematical Induction 303

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
8.2 Proof by induction . . . . . . . . . . . . . . . . . . . . . . . . 305
8.3 Variations on proof by induction . . . . . . . . . . . . . . . . 316

Hints and Solutions to Selected Exercises 333

Bibliography 385

Index 387
Preface

The notion of ‘proof’ is central to mathematics; indeed, it is one of its unique

features that distinguishes mathematics from other disciplines. In recent years,
highly sophisticated software products have been developed, particularly so-
called computer algebra systems and a variety of visualisation tools. As a
result, some people suggested that mathematics is becoming more ‘experimen-
tal’ in nature and the notion of proof less salient. This has provoked vigorous
denials from the mathematical community, which continues to maintain that
proof is one of the key concepts that characterises the discipline. It is also one
of the most difficult aspects of the subject both to teach and to master.
Our aims in this book are to describe the nature of mathematical proof, to
explore the various techniques that mathematicians adopt in proving their
results, and to offer advice and strategies for constructing proofs. Undergrad-
uate mathematics students often experience difficulties both in understanding
those proofs they come across in their courses and books and, more particu-
larly, in constructing their own proofs. When first asked to write mathematical
proofs, students of mathematics often struggle to know where to begin and
where to end. In other words, students often find it challenging to know how
to start an appropriate line of reasoning and knowing when they have ‘done
enough’ to complete it. In short, constructing proofs may seem to a student
to be something of a ‘black art’ known only to their teachers and the authors
of their textbooks.
This book is intended primarily for undergraduate students of mathematics
and other disciplines where reasoned arguments need to be developed and
understood, such as engineering and computer science. We hope that this book
will help improve students’ ability to understand those proofs that they meet
and will enhance their facility for constructing correct proofs of their own.
Any mathematical proof has a definite logical structure and will be written
with a certain audience in mind and with a particular style. In the ‘zoo’ of
mathematical proofs, we may categorise the ‘animals’ (i.e., proofs themselves)
by their logical types: direct proofs are placed in one enclosure, proofs by
contradiction in another, proofs by mathematical induction in a third, and so
on. To understand a proof, one needs to know which enclosure it belongs to; in
other words, its underlying logical structure. One of our tasks is to describe the
different ‘enclosures’ by cataloguing different types of proof. When it comes
to writing proofs, a knowledge of the different kinds of proofs is important but

ix
x

is not, by itself, enough. So we also provide some hints and guidance on some
of the approaches the proof writer may adopt to find an appropriate proof.
The book is structured as follows. Chapter 1 introduces the kind of reasoning
that mathematicians use when writing their proofs and gives some example
proofs to set the scene. In chapter 2, we introduce some basic logic in order
to understand the structure both of individual mathematical statements and
whole mathematical proofs. Much of the language of mathematics is based on
the notions of sets and functions, which we introduce in chapter 3. In chapter 4,
we dissect some proofs with a view to exposing some of the underlying features
common to most mathematical proofs. In the last four chapters of the book, we
explore in more detail different types of proof; to continue the analogy above,
we explore in more detail each of the enclosures in the zoo of mathematical
proofs.

RG and JT
September 2013
List of Figures

3.1 A Venn-Euler diagram . . . . . . . . . . . . . . . . . . . . . . 93

3.2 Venn-Euler diagram: subset . . . . . . . . . . . . . . . . . . . 93
3.3 Venn-Euler diagram: intersection . . . . . . . . . . . . . . . . 94
3.4 Venn-Euler diagram: union . . . . . . . . . . . . . . . . . . . 94
3.5 Venn-Euler diagram: disjoint sets . . . . . . . . . . . . . . . . 95
3.6 Venn-Euler diagram: difference . . . . . . . . . . . . . . . . . 96
3.7 Venn-Euler diagram: complement . . . . . . . . . . . . . . . . 96
3.8 Illustrating a set equality . . . . . . . . . . . . . . . . . . . . 98
3.9 The Cartesian product of two sets . . . . . . . . . . . . . . . 104
3.10 Cartesian product and intersection of sets . . . . . . . . . . . 105
3.11 A function machine . . . . . . . . . . . . . . . . . . . . . . . . 108
3.12 Diagram of a function showing the domain and codomain . . 108
3.13 The image of a function . . . . . . . . . . . . . . . . . . . . . 109
3.14 Illustrating the function in example 3.11.1 . . . . . . . . . . . 109
3.15 The image of the square function . . . . . . . . . . . . . . . . 110
3.16 The image of a function R → R . . . . . . . . . . . . . . . . . 111
3.17 Attempting to define the square root function . . . . . . . . . 113
3.18 Illustrating the composite function . . . . . . . . . . . . . . . 114
3.19 Composite of function machines . . . . . . . . . . . . . . . . . 114
3.20 Different elements of the domain with the same image . . . . 117
3.21 A surjective function . . . . . . . . . . . . . . . . . . . . . . . 118
3.22 Horizontal line test for functions R → R . . . . . . . . . . . . 120
3.23 Reversing arrows in example 3.11.1 . . . . . . . . . . . . . . . 122

xi
xii

3.24 Defining the inverse function . . . . . . . . . . . . . . . . . . 122

3.25 The graph of a linear function . . . . . . . . . . . . . . . . . . 123
3.26 Reversing arrows when the function is not injective . . . . . . 124
3.27 Reversing arrows when the function is not surjective . . . . . 124
3.28 Defining an inverse when the function is not surjective . . . . 125

4.1 Modulus function . . . . . . . . . . . . . . . . . . . . . . . . . 146

4.2 Illustrating cases in a proof . . . . . . . . . . . . . . . . . . . 148
4.3 Euclidean metric . . . . . . . . . . . . . . . . . . . . . . . . . 153
4.4 Manhattan metric . . . . . . . . . . . . . . . . . . . . . . . . 153
4.5 Spherical metric . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.6 Triangle inequality . . . . . . . . . . . . . . . . . . . . . . . . 155
4.7 Closed ball in the Manhattan metric . . . . . . . . . . . . . . 158
4.8 Relations between discs in R2 . . . . . . . . . . . . . . . . . . 158
4.9 Euclid’s parallel axiom . . . . . . . . . . . . . . . . . . . . . . 163

1
5.1 Possible starting points in the proof of x + ≥ 2 (x ∈ R+ ) . 177
x
1
5.2 Working backwards from x + ≥ 2 (x ∈ R+ ) . . . . . . . . 178
x
5.3 The Smarandache function . . . . . . . . . . . . . . . . . . . 185
5.4 Injective and surjective functions with finite domain and
codomain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
5.5 Symmetries of a rectangle . . . . . . . . . . . . . . . . . . . . 203
5.6 Composing symmetries of a rectangle . . . . . . . . . . . . . . 204
5.7 Symmetries of an equilateral triangle . . . . . . . . . . . . . . 205
5.8 Vector in R . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2
210
5.9 Addition and scalar multiplication of vectors in R2 . . . . . . 210
5.10 A plane in R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
5.11 Eigenspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
5.12 Possible behaviour of sequences . . . . . . . . . . . . . . . . . 228
5.13 Convergent sequence . . . . . . . . . . . . . . . . . . . . . . . 229
 xiii

5.14 Limit of a function . . . . . . . . . . . . . . . . . . . . . . . . 233

6.1 Finding a supremum . . . . . . . . . . . . . . . . . . . . . . . 265

7.1 Interior angle of an n-sided polygon . . . . . . . . . . . . . . 278

7.2 Finding a set-theory counter-example . . . . . . . . . . . . . . 290
7.3 Counter-examples to theorem 5.7 . . . . . . . . . . . . . . . . 291

8.1 Convex and non-convex polygons . . . . . . . . . . . . . . . . 321

8.2 Illustrating the inductive step of an angle sum theorem about
polygons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
List of Tables

2.1 Logical equivalence laws . . . . . . . . . . . . . . . . . . . . . 39

3.1 Set theory laws . . . . . . . . . . . . . . . . . . . . . . . . . . 99

4.1 Comparing conditional deduction rule with method of direct

proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
4.2 Comparing formal structure and real proof using cases . . . . 174

xv
List of Symbols

Symbol Description Page

Logic
P ⇔Q Biconditional connective: P if and only if Q 23
P ⇒Q Conditional connective: if P then Q 21
P ∧Q Conjunction: P and Q 18
Γ T Deduction in an axiom system:
T can be deduced from the statements in Γ 168
P1 , . . . , Pn ⊢ Q Deducible from: Q is deducible from P1 , . . . , Pn 69
P YQ Exclusive disjunction: P or Q but not both 20
P ∨Q Inclusive disjunction: P or Q 19
∃x • P (x) Existential quantifier: there exists x such that P (x) 49
P1 ≡ P2 Logical equivalence: P1 is logically equivalent to P2 35
P1  P2 Logical implication: P1 logically implies P2 42
¬P Negation of a proposition P : not P 17
∀x • P (x) Universal quantifier: for all x, P (x) 49

Sets and Functions

|A| Cardinality of set A 86
A×B Cartesian product of sets:
A × B = {(x, y) : x ∈ A and y ∈ B} 103
B(a, r) Closed ball in a metric space:
B(a, r) = {x ∈ X : d(x, a) ≤ r} 157
[a, b] Closed interval {x ∈ R : a ≤ x ≤ b} 90
Ā Complement of a set A 95
g◦f Composite of function f and function g 113
A−B Difference of sets A and B 95
m|n Divides: m divides n 187

xvii
xviii

x∈A Element of a set: x is an element of A 84

x ̸∈ A Not an element of a set: x is not an element of A 84
∅ Empty set 85
⌊x⌋ Floor or integer part of a real number x 230
f :A→B Function f with domain A and codomain B 107
im f Image of a function f 108
f : a 7→ b Image of an element: f maps a to b 108
A∩B Intersection of sets A and B 94
d : X × X → R Metric on a set X 155
dM Metric: Manhattan 153
dS Metric: sphere 153
(a, b) Open interval {x ∈ R : a < x < b} 90
(x, y) Ordered pair of elements x and y 103
ϕ:Z →Z + +
Phi function or totient function 182
P(A) Power set of A 99
B⊂A Proper subset: B is a proper subset of A 87
C Set of complex numbers 90
Z Set of integers 90
N Set of natural numbers 90
Q Set of rational numbers 90
R Set of real numbers 90
S:Z →Z + +
Smarandache function 184
B⊆A Subset: B is a subset of A 87
A⊇B Superset: A is a superset of B 87
A∗B Symmetric difference of sets A and B 102
A∪B Union of sets A and B 94
U Universal set 86

Group Theory
x∗y Binary operation: the result of combining x and y 201
(G, ∗) Group: G is a group with binary operation ∗ 201
e Identity element of a group 201
x−1 Inverse of an element x in a group 201
ker θ Kernel of a morphism θ 262
 xix

×n Multiplication modulo n 275

θ:G→H Morphism or homomorphism between groups 262

Linear Algebra
det X Determinant of a matrix X 297
S⊕T Direct sum of subspaces 251
E(λ) Eigenspace of an eigenvalue λ 221
In Identity n × n matrix 225
im T Image of a linear transformation 226
ker T Kernel of a linear transformation 226
T :U →V Linear transformation between vector spaces 226
∥a∥ Magnitude of a vector a 209
−v Negative of a vector v in a vector space 211
N (A) Null space of a matrix A 226
S+T Sum of two subspaces S and T or a vector space 225
XT Transpose of a matrix X 294
0 Zero matrix 246
0 Zero vector in a vector space 211

Analysis
inf X Infimum of a set X or real numbers 264
lim f (x) = ℓ Limit of a function f as x tends towards a 233
x→a
lim an = ℓ Limit of a sequence: (an ) converges to limit ℓ 228
n→∞
(an ) Sequence a1 , a2 , a3 , a4 , . . . of real numbers 227
sup X Supremum of a set X or real numbers 264
Chapter 1
Introduction

1.1 The need for proof

Mathematics has many different aspects, properties, and features, so it is

probably impossible to answer succinctly the question ‘what is mathematics?’
There are, however, properties of the subject that spring readily to mind: most
would agree that mathematics is abstract, rigorous, precise, and formal. No
doubt, many would also add ‘difficult’ to that list. Mathematics is abstract
because it deals with concepts rather than ‘real’ objects. For example, math-
ematics concerns itself with the number 3 rather than three apples or three
buttons; it deals with ‘idealised’ triangles rather than imperfectly drawn ‘real’
ones, and so on. It is rigorous and precise because it does not accept vague
or emotional arguments; rather, ‘facts’ about mathematical objects are only
established by logical reasoning. The sum of angles in a Euclidean triangle
is 180◦ , not because that would be a nice thing to have or because we want
it to be so; we know the angle sum is 180◦ because we can prove this to be
the case from the basic assumptions of Euclidean geometry. Mathematics is
formal because mathematical objects are given explicit and precise definitions
and mathematical reasoning has a definite and systematic structure.
Of course, people learn and understand mathematics at different ‘levels’, not
all of which conform to the description given in the previous paragraph. When
a young child learns about numbers, he or she may appreciate that a collec-
tion of two sweets when combined with a collection of three sweets produces
a collection of five sweets without having any real concept of the abstract
mathematical equation 2 + 3 = 5. There are various theories that describe
the different ‘levels’ of understanding mathematics. For example, the mathe-
matics educator David Tall has described ‘three worlds of mathematics’ which
he calls the ‘embodied’, ‘proceptual’ and ‘formal’ worlds [13]. The embodied
world builds directly on our perceptions of the world, both the physical world
and our mental world. Our concept of an idealised triangle, as an abstraction
of real, physical triangles, belongs to this world. The proceptual world1 is the
world where symbols represent concepts and we apply processes to those sym-
1 Tall coined this term from the words ‘process’ and ‘concept’.

1
2

bols. Thus, for example, the representation of fractions as a/b and the rule
a/b + c/d = (ad + bc)/bd belong to this world. The formal world is one where
formally expressed properties are used to specify mathematical structures.
For example, some readers may be familiar with the mathematical notions of
‘group’, ‘vector space’, ‘field’, and so on; these belong to the formal world.
Properties of these mathematical objects are deduced by formal proof rather
than from direct experience. In this book, we will be operating (mostly) in the
formal world, for it is here that the concept of mathematical proof belongs.
For mathematicians, Tall’s formal world, where proof plays a central role, is
the world of ‘their’ mathematics. In 1929, the eminent English mathematician
G. H. Hardy articulated the special role that proof plays in mathematics when
he wrote, ‘It is generally held that mathematicians differ from other people
in proving things, and that their proofs are in some sense grounds for their
beliefs’ [8]. Hardy is discussing the formal world of mathematics and he makes
a critical observation that mathematicians differ from others in two crucial
respects: in what they do (they prove things) and, more deeply, in their criteria
for belief (the existence of proofs). In this way, the notion of proof is central
to the formal world of mathematics.
If the notion of a mathematical proof is one of the defining features of mathe-
matics, for those learning the subject, it is also one of its greatest challenges.
In fact, there are (at least) three challenges for students of mathematics asso-
ciated with the notion of proof. The first is to understand why mathematicians
are ‘obsessed’ with their proofs. For instance, why is it that (pure) mathemat-
ics textbooks are peppered with proofs? A second challenge is to understand
what constitutes a proof and, just as important, to understand when an argu-
ment does not constitute a proof. The final — and most difficult — challenge
is to develop an ability to construct one’s own proofs. In this book we aim to
address each of these challenges.
A simple, if incomplete, description of a mathematical proof is ‘a precise logi-
cal argument of a certain type that establishes a conclusion’. One of our aims
is to explore this further to provide a fuller and more detailed understanding
of what constitutes a proof. Of course, proofs are written by human beings,
so there are many different styles and approaches adopted. Proofs are written
for a variety of different audiences in various cultures and in many languages.
A proof that may be appropriate for an audience at one level of mathematical
development may not be appropriate for an audience with a deeper under-
standing of the subject. Nevertheless, we believe that there is something that
is shared by (almost) all mathematical proofs. We shall attempt to delve be-
neath surface features such as style and language to examine the structure of
the argument that lies beneath. This aspect of the book could be summarised
as ‘proofs dissected, examined, and their structure revealed.’
Our second goal is probably more difficult to achieve. Put simply, it is to show
how to construct proofs. On one level, such a goal is unattainable. There is
 3

no magic formula which, if learned, will enable us to construct a proof of

any result we care to choose. However, there are a variety of methods of proof
that are appropriate in different contexts and any aspiring ‘proof writer’ needs
to know these. We will also provide hints and guidance on how to approach
the challenging task of finding a proof. Many teachers of mathematics will
agree with the following advice to students. ‘How do you stay in physical
shape? Do exercise. How do you stay in mathematical shape? Do exercises!’
For this reason, we present a variety of worked proofs and we provide numerous
exercises for the reader to do (to stay in mathematical shape).

1.2 The language of mathematics

Like any specialised area of human activity, mathematics has its own language
and commonly used terms that make it less accessible to the outsider. For
example, two mathematicians in discussion may be heard uttering phrases
similar to the following.
‘I have a counter-example to your recent conjecture published in the
Journal of Esoteric and Obscure Pure Mathematics.’
‘Ah, but your proof relies on Zorn’s Lemma. Can you find a proof
that is independent of the Axiom of Choice? ’
‘If you factor out the nilpotent elements, your group is torsion-free.’
Whereas these statements might make perfect sense to a professional mathe-
matician, to an outsider they are almost certainly completely opaque. In this
section, we introduce informally some of the language that is frequently used
in mathematics texts. Many of the terms we introduce here will be considered
in more depth later in the book.

Statements are the basic sentences of mathematics; they may be expressed

in words or symbols or, frequently, using a combination of both. A statement
may be true or false, or its truth may depend on the values of some variables
that it contains. However, a statement does not express an opinion, emotion,
or question. In this sense, we are using the term ‘statement’ to have a specific
meaning 2 which differs from its common usage. In everyday language, the
expression of an opinion, such as ‘this rose is beautiful ’, would often be referred
to as a statement. The following are statements in the sense that we are using
the term.

2 We will define this more precisely in chapter 2; by a statement we mean either a

proposition — see page 16 — or a propositional function — see page 46.

4
‘17 is a prime number ’; true.
√
‘ 2 is rational ’; false — see section 6.5.
‘2n − 1 is prime’; truth depends on the value of n — see section 7.4.
‘n(n + 1) is even for every integer n’; true – see example 1.4.

Mathematical arguments have hypotheses and a conclusion. The hypothe-

ses are statements that are assumed to be true. A mathematical argument
then proceeds to make deductions from the hypotheses until the final state-
ment, the conclusion, is reached. As we shall discover, the hypotheses of an
argument are assumed true for a variety of reasons. A hypothesis may be a
statement that has previously been proved or it may be a basic assumption,
or axiom, that underpins a whole branch of the subject. Alternatively, the ar-
gument may concern a conditional statement of the form ‘if statement 1, then
statement 2’. The argument may then have ‘statement 1’ as the hypothesis
and ‘statement 2’ as the conclusion. Readers familiar with statistics, may have
come across the idea that a hypothesis is a statement that is ‘tested’ based on
some data; a statistical test will lead us to accept or reject the hypothesis with
a certain level of confidence. This is not the case in mathematics; hypotheses
are not tested, they are just assumed.

As in most disciplines, a branch of mathematics will have its own specialist

terms. In mathematics, these will usually be precisely stated as definitions.
A definition just gives a precise description of a word or phrase in terms of
other, more basic words or phrases. Thus the phrase ‘a prime number is an
integer greater than 1 that has no factors other than 1 and itself ’ provides
a definition of ‘prime number’. It is really saying that we may use ‘prime
number’ as shorthand for ‘an integer greater than 1 that has no factors other
than 1 and itself’. Of course, we need to know what the more basic terms in a
definition mean in order for the definition itself to make sense. For example, we
would not expect most readers to obtain much meaning from the definition: a
simply connected 3-manifold is a topological space that is locally homeomorphic
to R3 and whose fundamental group is trivial. The ‘problem’ here is that the
more basic terms such as ‘topological space’, ‘locally homeomorphic’ and so
on are likely to be unfamiliar. Definitions are necessary in practice in order
to keep the language of statements manageable. Consider, for example, the
statement: for all positive integers n, if 2n − 1 is prime, then n is prime.
(We will prove this later – see theorem 6.3.) Without the definition of ‘prime
number’, we would need to write this in a much more cumbersome way, as
follows: for all positive integers n, if 2n − 1 is an integer greater than 1 that
has no factors other than 1 and itself, then n is an integer greater than 1
that has no factors other than 1 and itself. Clearly, the first way of stating
the theorem is to be preferred. There is a danger with definitions, of course.
Having too many definitions will make a piece of writing inaccessible as the
reader may have to keep looking up the definitions of terms used and not be
able to follow the ‘flow’ of the argument itself.
 5

In mathematics, the truth of statements is established by providing a proof.

A statement that has a proof is called a theorem. (We will not discuss here
what constitutes a proof; that is for consideration elsewhere in this chapter
and, indeed, elsewhere in this book.) Thus ‘17 is a prime number ’ is a theorem,
albeit a rather trivial one, because it has a proof — we simply need to verify
that each of the whole numbers 2, 3, . . . , 16 is not a factor of 17. A slightly
more interesting theorem is the statement ‘for all integers n, if n is odd, then
n2 is odd ’. In example 1.3, we give a proof of this statement and we are
therefore justified in calling it a theorem.
A statement may fail to be a theorem for one of two reasons. It might be
false and therefore no proof is possible. For example, the statement ‘2256 − 1
is prime’ is false (see section 7.4) so, no matter how hard we try, we will not
be able to find a proof. The other possibility is that we simply may not know
whether or not the statement has a proof. A statement for which no proof or
disproof is known is called a conjecture. An example of such a statement is:
‘every even integer greater than 2 can be expressed as the sum of two primes’.
To date, no-one has been able either to prove the statement or to find an even
integer greater than 2 that cannot be expressed as the sum of two primes.
Actually, this is a very famous conjecture called Goldbach’s Conjecture — see
section 2.2.
Conjectures are not generally arbitrarily chosen statements for which a proof
is not known. Most often, a conjecture is only stated as such if there is evidence
to support it, usually in the form of examples where it is known to be true. To
state something as a conjecture really means ‘I think this is true but, so far,
I have been unable to find a proof’. The status of a conjecture may change
over time. Probably the most famous example of this is Fermat’s Last Theo-
rem. Pierre de Fermat (1601 – 1665) claimed to have discovered a ‘remarkable
proof’ of the statement: ‘for all integers n > 2, the equation xn + y n = z n has
no solutions where x, y, and z are integers’. However, Fermat never commu-
nicated his proof, so the statement should, more properly, have been known
as Fermat’s Conjecture. This remained a conjecture until the 1990s when it
was proved by Andrew Wiles.3 With the completion of the proof after over
300 years, the status of the statement changed from ‘conjecture’ to ‘theorem’.
Sometimes a theorem may be called a ‘lemma’ or a ‘corollary’. Although
these are still theorems (a lemma or a corollary must still have a proof), the
terminology indicates a slightly different ‘status’. The proofs of some theorems
are long and are best split into separate parts or phases. It may be sensible to
prove some preliminary results first, separate from the main proof; then these
preliminary results may be used within the main proof itself. In situations
like this, the preliminary results are usually referred to as lemmas. Labelling
3 The proof was originally announced in 1993, but a gap in the reasoning was subse-

quently found. Wiles and a colleague, Richard Taylor, were able to ‘plug the gap’ and
complete the proof in 1994.
6

a theorem as a lemma probably indicates that the result is not particularly

interesting in its own right and its principal purpose is as a ‘stepping stone’
towards what might be called the ‘main’ theorem. There are exceptions, of
course; for example, Zorn’s Lemma, mentioned at the beginning of this section,
is an important result in its own right. In contrast to a lemma, a corollary
generally follows a main theorem and is a relatively simple consequence of it.
A corollary may simply be a special case of the preceding theorem or it may
follow from it by a simple deduction. For example, a corollary of Pythagoras’
theorem about right-angled
√ triangles is that the diagonal of a square, with
side length 1, is 2.

1.3 Reasoning

Human beings are constantly reasoning about their environment in order to

make decisions and to guide actions. Mostly this reasoning will be informal
and very often it will even be subconscious. Here are a few examples.

• I know it takes me an hour to prepare, cook, and eat this evening’s meal.
I need to leave home at 8.30 pm to go out with my friends. So I need to
start preparing my evening meal by 7.30 pm at the latest.
• The car approaching in that direction is signaling to turn off the road.
The car approaching from the other direction is going slowly. Therefore
it is safe for me to cross the road.
• When people catch a cold they cough and sneeze. If I am in close prox-
imity to someone with a cold, I am likely to catch it. Mike is coughing
and sneezing. So I should stay away from him.
• I know that whenever I drop an object, it will fall to the ground. If a glass
falls to the ground, it will smash. So I must be very careful carrying this
tray of glasses.

Most human reasoning is based on experience, either our own direct experience
or some shared, collective experience. For example, consider the following
statement, related to the last example above.

If I drop this glass, then it will hit the floor and smash.

At some time or other, most of us will either have dropped a glass and seen it
smash or observed someone else do so. Our reasoning here is therefore based
on direct experience. Experience tells us that when we drop glasses they (a)
 7

fall to the floor and (b) when they hit the floor, they smash. Hence we deduce
that if I drop this glass, then it will hit the floor and smash.
The kind of reasoning based on experience and experimentation is called in-
ductive reasoning. Individual or collective experience tells us that a certain
action (for example, dropping a glass) results in a particular consequence (the
glass hits the floor and smashes) because:
• we have observed ‘many’ times that the certain action results in the
particular consequence; and
• we have never — or almost never — observed the certain action not to
result in the particular consequence.
This ‘reasoning from experience’ will not always be from direct, personal ex-
perience but may be based on collective experience of a group (or indeed, the
collective experience of human kind). For example, neither of the authors of
this book has ever sky-dived although we know people who have. Nevertheless,
we believe that the action of sky-diving results in the sky-diver experiencing
an adrenalin rush. This belief is based on the collective experience of those
who have engaged in the activity of sky-diving. Similarly, imagine a scenario
where a young child is playing with a pair of scissors. The child’s mother might
say:
Don’t cut the lamp flex! If you cut the flex, then you will
get electrocuted and it will hurt you.
In this case, we hope that the statement is not based on the direct experi-
ence of the mother or child. However, the mother still ‘knows’ that cutting
a live electric flex causes electrocution and that hurts! This is part of our
shared collective experience; some people have experienced this directly and
this experience is then shared more widely.
Inductive reasoning — reasoning based on (direct or indirect) experience and
observation — is probably the basis for most human knowledge and belief. But
it does not provide absolute certainty. (The next time I drop a glass it might
not break.) This is why, for example, in criminal law, the test of guilt is ‘proof
beyond reasonable doubt’ rather than proof beyond all doubt. In assembling
a web of evidence — forensic, eye-witness, expert opinion, circumstantial, etc.
— the criminal prosecutor can rarely, if ever, achieve absolute certainty of
guilt.
Inductive reasoning is also the basis for acceptance of scientific theories. A
scientific theory can be thought of as an intellectual framework (which is
frequently mathematical in nature) that explains existing observations and
predicts new ones. The more experiments that produce data consistent with
the theory, the stronger our belief in the theory becomes, but no amount
of experimental data will prove a theory to be correct. A single experiment
(conforming to the established scientific criteria of reliability and repeatabil-
ity) that conflicts with the theoretical framework can invalidate it. Indeed,
8

scientific breakthroughs are frequently the result of experiments that do not

‘fit’ with existing theories. In this way, it is sometimes said that scientific
theories are falsifiable but never provable.
The type of reasoning used in mathematics is of a different kind altogether.
In mathematics, the framework for belief is based on deductive reason-
ing, where facts are deduced from others by a process of logical deduction.
As we mentioned in section 1.1, mathematicians’ criteria for belief is based
on mathematical proof: a correct logical argument that proceeds by making
logical deductions from a collection of premises to a conclusion. For a math-
ematician, a proof eliminates doubt: if we accept that the premises are true,
then inevitably we must accept the conclusion to be true also. Consider, for
example, Pythagoras’ theorem:
In any right-angled triangle, the area of the square on the hypotenuse
equals the sum of the areas of the squares on the other two sides.
For mathematicians, this is true because there is a proof of it. Even if we
have constructed many hundreds, or even thousands, of right-angle triangles,
measured the lengths of the sides, and verified Pythagoras’ Theorem for each
triangle (and have never found a right-angled triangle for which the theorem
fails), this evidence would not be sufficient for a mathematician. There are
numerous results that are true for many thousands of examples, but which
are not true in general. For the mathematician, therefore, only rigorous logical
deduction is sufficient to guarantee the truth of a result in all cases. We explore
this kind of deductive reasoning in more detail in the next section.
Although only deductive reasoning is allowed in establishing mathematical
facts, inductive reasoning still has an important place in mathematics. In gen-
eral, theorems begin life as conjectures and conjectures are formulated through
inductive reasoning. Consider Pythagoras’ theorem about right-angled trian-
gles. The formulation of the theorem almost certainly came about by observing
examples of right-angled triangles, calculating the squares of the lengths of
the sides, and observing that, in each case, the square of the length of the
hypotenuse equals the sum of the squares of the lengths of the other sides.
Only once the statement of a theorem — more properly, the statement of a
conjecture — has been formulated, can the search for a proof begin.

1.4 Deductive reasoning and truth

In mathematics, we require a standard of reasoning that is both precise —
there should be no ambiguity — and logically rigorous — we may only draw
conclusions that ‘follow inevitably’ from what we already know. Let us explore
 9

in a bit more detail what it means to say that one statement follows inevitably
from others.

Example 1.1
Consider the following piece of reasoning.

The square of any even integer is also even.

1234567893 is odd.
Therefore (1234567893)2 is odd.

Is this reasoning valid? Firstly, we need to explain the notation used. In a

fragment of reasoning presented in this way, the (implicit) claim is that the
statement below the line follows logically from the statements above the line.
To ask whether the reasoning is valid is not to ask whether or not the indi-
vidual statements are true. Rather, it is asking whether the conclusion, which
is the statement below the line following ‘Therefore . . . ’, necessarily follows
from the previous statements. In this case, each of the three individual state-
ments is true, but the reasoning itself is not valid. This is important but quite
subtle. The point is that last statement ‘(1234567893)2 is odd’ is true but its
truth does not follow from the truth of the previous statements. To see this,
suppose we replace the term ‘odd’ with ‘prime’ and, to be consistent, we also
replace the term ‘even’ (meaning ‘not odd’) with ‘composite’ (meaning ‘not
prime’). Then the reasoning is the following.

The square of any composite integer is also composite.

1234567893 is prime.
Therefore (1234567893)2 is prime.

In this case the first two statements are true as before but the final statement
is false. An obvious requirement of any piece of valid reasoning must be that,
if the initial (above the line) statements are true, then the (below the line)
conclusion must also be true. Hence this second reasoning fragment is clearly
invalid. Each of the two reasoning fragments has the following form.

If any integer has property P , then its square also has property P .
A particular integer n does not have property P .
Therefore n2 does not have property P .

The second instance of this form of reasoning shows that it is not valid. The
purported conclusion does not ‘follow inevitably’ from the given initial state-
ments. The important point to note here is that whether a piece of reasoning
is valid (or not) depends on the form of the reasoning fragment, not on the
truth of the individual statements. In some cases, it will not be immediately
10

or intuitively obvious whether a reasoning fragment is valid, so we will explore

this further in chapter 2. For example, the following form of reasoning, which
is closely related to that above, is valid.

If any integer has property P , then its square also has property P .
For a particular integer n, its square n2 does not have property P .
Therefore n does not have property P .

Example 1.2
Now consider the following piece of reasoning.

For any prime integer n, the integer 2n − 1 is also prime.

23 is prime.

Therefore 223 − 1 = 8 388 607 is prime.

In this case, the conclusion is false because 223 − 1 factorises as follows

223 − 1 = 8 388 607 = 47 × 178481.

Does this mean that the reasoning itself is invalid? In fact, the real reason
why the conclusion is false is because one of the ‘initial’ statements is false.
The statement

‘if n is a prime integer then 2n − 1 is also prime’

is false. Indeed, the example of n = 23 shows that this statement is false.

There are, in fact, many examples of prime numbers n for which 2n − 1 is
not prime. However, this does not mean that the reasoning itself is incorrect.
Were it to be the case that each of the statements
for any prime positive integer n, the integer 2n − 1 is also prime
and 23 is prime
was true, then it would necessarily be true that 223 − 1 would be prime.
The structure of the reasoning fragment may be described as follows.

For any n, if n has property P , then n has property Q.

A particular k has property P .

Therefore k has property Q.

This is, indeed, a valid piece of reasoning. Again, this example emphasises that
whether a piece of reasoning is valid depends on the structure of the fragment
and not the truth of the individual statements.
 11

As we have noted, establishing truth in mathematics depends on making valid

logical deductions. However, the two previous examples show that the validity
of a piece of reasoning does not, by itself, guarantee the truth of the conclusion.
Equally, the truth of a conclusion does not mean that the reasoning that gave
rise to it was valid. The important situation is when the initial statements
are true and the reasoning is valid; then the conclusion is guaranteed to be
true also. This is the situation in a mathematical proof which starts with
true statements and makes a sequence valid of logical deductions from these
until a conclusion is reached. Provided the starting statements are true and
the reasoning is valid, then the conclusion must also be true. In the following
section, we introduce two proofs to give a flavour of what is to come in the
rest of the book.

1.5 Example proofs

We conclude this chapter by giving proofs of two simple results about the
integers or whole numbers. In each case, readers will be able convince them-
selves of the truth of the result ‘beyond reasonable doubt’ by considering a
selection of example integers and checking the result in each case. Our proofs
will be short deductive proofs that, we hope, establish the results ‘beyond all
doubt’.

We shall examine the structure of mathematical proofs in more detail later but,
in broad terms, a proof of a theorem must show that, given the truth of certain
‘precursor’ statements that will be assumed as part of the reader’s ‘background
knowledge’, the truth of the theorem inevitably follows. Normally, such a proof
takes the form of a sequence of statements whose truth is guaranteed either
because it is assumed background knowledge or because it follows from earlier
statements in the proof itself. In other words, the truth of each statement
in the proof is established using valid deductive reasoning from the truth of
other statements whose truth is known or assumed.

Example 1.3
Consider the following theorem. In this example, we will give a proof of the
theorem (which is why it is legitimate to call it a theorem). Although we will
try not to labour this too much, we will examine the statement of the theorem
and its proof in rather more detail than is normal in a mathematics text.

Theorem 1.1
For all integers n, if n is odd, then n2 is odd.
12

Before we think about proving this theorem, we first need to understand

precisely
(a) the overall logical structure of the statement and
(b) the meaning of the terms used.

We will explore the structure of individual statements such as this more fully
in chapter 2. For now, it is sufficient to note that this is a conditional statement
of the form
if (statement 1), then (statement 2).

We will explore conditional statements in more detail in chapter 2. For now,

we need to note that a conditional statement is true provided statement 2 is
true in all circumstances in which statement 1 is true. This means that we
need to establish that ‘n2 is odd ’ is true in all circumstances that ‘n is odd ’
is true.
The term ‘integer’ just means ‘whole number’. What does it mean to say
that an integer is odd ? This probably seems like a curious question since we
anticipate that all readers will know that the odd integers are
. . . , −7, −5, −3, −1, 1, 3, 5, 7, . . . .
However, knowing this list of the odd integers is not sufficient for our purposes.
We need to establish that, whenever n is in this list of odd integers, its square
n2 is also in the list. What we really need is a property that defines what it
means (for an integer) to be odd. Then we can establish that whenever n has
this ‘oddness property’, then n2 also has the ‘oddness property’. The property
we are seeking is the following.

Oddness property: An integer n is odd precisely when it can be expressed as

n = 2m + 1 for some (other) integer m.

For example, 17 is odd because 17 = 2 × 8 + 1; −243 is odd because −243 =

2×(−122)+1. Similarly, 26 is not odd because it cannot be expressed as 2m+1
where m is an integer. (We can express 26 = 2m + 1, but then m = 12.5 is
not an integer.)
Now that we have understood both the structure of the statement and the
meaning of the terms used, we can give a proof.

Proof. Let n be an odd integer.

Then, by the oddness property, n = 2m + 1 for some integer m. Now
n2 = (2m + 1)2
= 4m2 + 4m + 1
= 2(2m2 + 2m) + 1
= 2M + 1 where M = 2m2 + 2m.
 13

We have shown that n2 = 2M + 1 where M = 2m2 + 2m is an integer. Hence,

by the oddness property, n2 is odd.

Example 1.4
In this example, we will give alternate proofs of the following theorem.

Theorem 1.2
For all integers n, n2 + n is even.

As in the previous example, we need the following property that defines what
it means for an integer to be even.

Evenness property: An integer n is even precisely when it can be expressed as

n = 2m for some (other) integer m.

Our first proof considers separately the two possibilities for n: either it is even
or it is odd. Then it uses the evenness or oddness properties for n, together
with a little algebra, to deduce the evenness property for n2 + n.

Proof 1.
There are two cases to consider: n is even or n is odd.
Case 1: n is even.
Then n = 2m for some integer m. Therefore

n2 + n = (2m)2 + 2m
= 4m2 + 2m
= 2(2m2 + m), where 2m2 + m is an integer.

Therefore n2 + n is even in this case.

Case 2: n is odd.
Then n = 2m + 1 for some integer m. Therefore

n2 + n = (2m + 1)2 + 2m + 1
= 4m2 + 4m + 1 + 2m + 1
= 4m2 + 6m + 2
= 2(2m2 + 3m + 1), where 2m2 + 3m + 1 is an integer.

Therefore n2 + n is even in this case also.

In both cases n2 + n is even, which completes the proof.
14

This proof is probably the most obvious way of establishing the given result,
especially given our proof in the previous example. It is not unusual to break
a proof down into separate cases and consider each separately. The only pre-
requisite knowledge for the proof is the evenness and oddness properties and
some basic algebra. Compare this with the following, much shorter proof.

Proof 2.
Let n be an integer. Then n2 + n = n(n + 1).
Now n and n + 1 are consecutive integers, so one of them is even and the
other is odd (but we don’t know which is which). Hence their product is the
product of an even and an odd integer, which is even.
Hence n2 + n = n(n + 1) is even.

This second proof is clearly much shorter, avoids the need for separate cases
and requires much less algebra. However, it does require the clever observation
that factorising n2 + n provides a helpful first step. This proof also assumes
an additional piece of background knowledge that the product of an even and
an odd integer is even. If we accept this as fact, then surely the second proof
is simpler and, arguably, more elegant. Some readers might reasonably object
that the proof is not complete as it has not established the ‘missing’ piece of
background information. To overcome these objections, we could replace the
penultimate sentence with the following.

The product of an even and an odd integer is of the form 2p × (2q +

1) = 2(p(2q + 1)), for some integers p and q, which is even.

Exactly how much background information may be assumed and how much
must be included explicitly in a proof will always be a matter of judgement
by the person constructing the proof.
No doubt some readers will prefer our first proof and some will think the
second one is better. That is as it should be — there is no ‘right answer’ to
the question, which is the better proof?
Chapter 2
Logic and Reasoning

2.1 Introduction

In this chapter, we consider in more detail the notions of ‘statements’ and

‘deductive reasoning’ that were introduced in chapter 1. We aim to understand
the structure of both individual statements and whole logical arguments. Since
mathematical proofs are structured, logical arguments, a working knowledge
of basic logic and logical reasoning will be important in understanding the
structure and correctness of mathematical proofs.
As we have seen, some statements are about specific objects, such as ‘17 is
prime’. Others concern whole classes of objects; for example, ‘for all positive
integers n, if 2n − 1 is prime, then n is prime’. Some statements are ‘atomic’
and cannot be broken down further; others are ‘compound ’ and are built
from simpler statements into more complicated structures. An example of the
latter type of statement is: ‘if n is even and m is odd, then the product mn
is even’. Understanding how statements are put together and how this affects
their truth is one of the goals of this chapter. Similarly, recognising when and
why a statement is not true is important. Consider the following imaginary
dialogue, for example.
Person A: I’ve noticed that, whenever x + y is positive,
xy is also positive.
Person B: No, sorry, you’re wrong because, for example, when
x = 1 and y = −2 both x + y and xy are negative.
Who, if anyone, is correct? In this case, neither person is correct. Person A’s
statement is certainly false. Although Person B’s example is correct, this is
not the negation of Person A’s statement and so is not the reason why Person
A is wrong.
In the previous chapter, we considered reasoning fragments that purported
to show that one statement ‘follows inevitably’ from a collection of other
statements. Later in this chapter we will examine both what it means to say
one statement ‘follows logically’ from other statements and also when a piece
of reasoning establishes this connection.

15
16

2.2 Propositions, connectives, and truth tables

Propositions are statements that are either true, T, or false, F. The truth
or falsity of a proposition is called its truth value. Since there are only two
truth values, a proposition that is not true is regarded as being false and
a proposition that is not false is regarded as being true. We will generally
use upper case letters, P , Q, R, etc. to denote propositions. Here are some
examples of propositions together with their truth values where these can be
determined.

1. The Potomac river flows into the Chesapeake bay. T

2. Brighton is the capital city of the United Kingdom. F
3. 2 + 2 = 4. T
4. Today is my birthday. depends
5. The digit in the 7th position after the decimal point T
in the decimal expansion of π is 6.
6. The digit in the 200 millionth position after the dec- ?
imal point in the decimal expansion of π is 6.
7. Every even integer greater than 2 can be expressed as unknown
the sum of two primes.

Note that in some cases we will know the truth value of a proposition, but in
others we may not. For example, in the first two statements above, we need to
know some geography (of North America and the British Isles respectively) in
order to know their truth values. The truth value of proposition 4, ‘today is my
birthday’, depends on who utters the statement and when. If the statement is
true when I utter it, then it will be false if I utter the statement 24 hours later.
The two sentences involving the decimal expansion of π are propositions since
they are definitely true or false. We can identify that proposition 5 is true using
a calculator, spreadsheet, or other software to evaluate π = 3.14159265 . . ..
Proposition 6 is a little more tricky. Although the 200 millionth decimal digit
of π is known,1 we (the authors) do not know its value. So, although we know
that statement 6 is definitely true or false, we don’t actually know which it is.
Statement 7 is a famous unsolved problem in mathematics. It is known as the
Goldbach Conjecture and was first stated in 1742 by the German mathemati-
cian Christian Goldbach in correspondence with Leonhard Euler. Although
it has been verified that every even integer up some vast number (at least
1 In fact, π has been evaluated to several trillion digits.
 17

1017 ) is indeed expressible as the sum of two primes, it is still currently un-
known whether every even integer greater than 2 can be so expressed. Hence,
although statement 7 is definitely either true or false, as yet no one knows
which.2
Sentences that cannot be viewed as true or false are not propositions. These in-
clude instructions, questions, exclamations, exhortations, demands, opinions,
and so forth. Hence the following are not propositions.

1. Keep off the grass. an instruction

2. Formal methods are vital for software development. an opinion
3. Will you come to my party? a question

Negation

Given a proposition P , adding the prefix ‘It is not the case that . . . ’ or inserting
‘not’ in an appropriate place in the sentence results in another proposition with
the reverse truth value. The resulting proposition is called the negation of
P , which we will denote ¬P . This is usually read as ‘not P ’. Other notations
for negation are P̄ and ∼ P .

Example 2.1
Let P be the proposition ‘Today is my birthday.’ Then the negation of P is
the proposition ‘It is not the case that today is my birthday.’ This is a rather
unnatural sentence and a more natural way of expressing ¬P is ‘Today is not
my birthday.’

In example 2.1, we noted that there is more than one way to express negation.
Whilst adding the prefix ‘It is not the case that . . . ’ to P will always produce
the negation ¬P , there is usually a more natural-sounding sentence for the
negation. From the point of view of logic, what is important is that ¬P is
true when P is false and ¬P is false when P is true. We summarise this in
the following table, called the truth table for negation.
2 In 2013, a related conjecture called Goldbach’s Ternary Conjecture was proved by
Harald Helfgott, a Peruvian mathematician working in Paris. This states that every odd
integer greater than 5 is the sum of three primes. It is a weaker conjecture than Goldbach’s
original conjecture: if every even integer greater than 2 can be expressed as the sum of two
primes, then adding 3 to each sum of two primes proves Goldbach’s ternary conjecture. Even
the weaker ternary conjecture was extraordinarily difficult to prove. Not only did it involve
some highly sophisticated number theory, but it also required approximately 150,000 hours
(17 years) of computer processing. In fact, this was part of a larger calculation that took
approximately 440,000 hours (50 years) of computer processing. The experts still believe
that a proof of the original Goldbach conjecture is a long way off.
18

P ¬P
T F
F T

Connectives and compound propositions

The examples of propositions given above are simple propositions because

they make a single statement. By contrast, the proposition
Today is my birthday and I am having a party
is not a simple proposition since it makes two distinct statements. We view
this proposition as being composed of the two simple propositions ‘today is
my birthday’ and ‘I am having a party’ conjoined using the word ‘and’. It is an
example of a compound proposition, and its truth value depends on the truth
values of the two component simple propositions. For example, if today is not
my birthday, but nevertheless, I am still having a party, then the compound
proposition is false.
In general, we may combine simple propositions together using connectives
(such as ‘and’ and ‘or’) to form compound propositions. The truth value of
a compound proposition depends on the truth values of its component parts as
well as the connective(s) used. For example, the proposition above is only true
when both the simple propositions ‘today is my birthday’ and ‘I am having
a party’ are true; in all other cases, the compound proposition is false. There
are five connectives that we will consider: conjunction, inclusive disjunction,
exclusive disjunction, conditional, and biconditional.

Conjunction

The conjunction of two propositions is generally expressed by using ‘and ’

between the two propositions. If P and Q are propositions, then their con-
junction is the proposition ‘P and Q’. The conjunction is denoted P ∧ Q and
this is usually read as ‘P and Q’. The truth values of the conjunction P ∧ Q
are defined by the following truth table.

P Q P ∧Q
T T T
T F F
F T F
F F F
 19

Note that there are four rows to the truth table because there are four possible
combinations of truth values of P and Q. It is important to realise that the
truth table defines what we mean by conjunction. Thus the conjunction P ∧ Q
has truth value T only when both the component propositions P and Q,
sometimes called the conjuncts, have truth value T; in all other cases P ∧ Q
has truth value F. There are other ways of expressing conjunction in English
other than using the word ‘and’. Consider the following propositions.

1. Sue went to the party last night but Jamie didn’t.

2. Hamid passed the module whereas Mike did not.
3. Jim played football although he was suffering from a cold.

In each case, the proposition is true only when both of its component simple
propositions are true; hence, each proposition is the conjunction of its two
components. The English sentences each have slightly different nuances com-
pared with the corresponding sentences using ‘and’ to form the conjunction.
For example, the third proposition hints at a little surprise that Jim played
football while suffering from a cold. Any natural language, such as English,
can express a vast range of subtlety, emotion, and nuance that we are unable
to capture in logical expressions. However natural languages are also prone
to ambiguity. It is for this reason that legal documents, for example, which
seek to eliminate ambiguity, are written in a rather formulaic manner that
does not follow the way most people speak. Although logical expressions are
unable to capture subtlety, provided they are properly formulated, they can
express statements precisely and unambiguously.

Disjunction

The disjunction, or inclusive disjunction, of two propositions is generally

expressed by using ‘or ’ between the two propositions. If P and Q are propo-
sitions, then their disjunction is the proposition ‘P or Q’. The disjunction is
denoted P ∨Q and this is read as ‘P or Q’. The truth values of the conjunction
P ∨ Q are defined by the following truth table.

P Q P ∨Q
T T T
T F T
F T T
F F F

Again, we regard the truth table as defining the meaning of disjunction. Thus
P ∨ Q is false only when both component propositions (called the disjuncts)
20

are false; if either or both of the disjuncts are true, then the disjunction is
true. In particular, note that P ∨ Q is true when both P and Q are true. For
example, the disjunction
Today is my birthday or I am having a party
should surely be regarded as true when I am enjoying my birthday party. This
is the so-called inclusive use of disjunction and it is the default interpretation
in logic.
There is an alternative form of disjunction, called exclusive disjunction,
that is denoted P Y Q. In this case, the (exclusive) disjunction of P and Q is
false when both P and Q are true. The truth values of P Y Q are defined by
the following truth table.

P Q P YQ
T T F
T F T
F T T
F F F

For example, the statement

Next week, I will work at home or visit a friend in Paris
suggests an exclusive interpretation of ‘or’. The two disjuncts, ‘I will work at
home’ and ‘I will visit a friend in Paris’, appear to be mutually exclusive. In
this case, we may interpret the proposition as true only when just one of the
two disjuncts is true. However, even in this case, one could imagine a scenario
where both conjuncts are true — suppose I live in Paris, for example. In
logic, the default interpretation of ‘or’ is the inclusive disjunction. If we wish
to ensure that a disjunction is given the exclusive interpretation, we would
normally use the construction
‘P or Q, but not both’.

This often leads to sentences which are somewhat unnatural in English. For
example, you will receive a fine or you have diplomatic immunity but not both.
However, since disjunctive statements can often be interpreted inclusively, we
will add ‘but not both’ whenever the exclusive interpretation is the intended
one.

Conditional

A conditional proposition is commonly denoted in English by a phrase of the

form
‘if . . . then . . . ’.
 21

Here are some typical examples.

1. If the backspace key is pressed, then the cursor moves one place to the
left.
2. If I leave my umbrella at home, then it will rain.
3. If you break the speed limit, you will receive a fine.
4. If n is a prime number, then n is not a perfect square.

Note that the word ‘then’ is sometimes omitted from the phrase ‘if . . . then
. . . ’. This is illustrated in the third proposition above.
If P and Q are propositions, then the conditional ‘if P then Q’ is denoted
P ⇒ Q. Here the proposition P is called the antecedent and the proposition
Q is called the consequent of the conditional proposition. Note that P ⇒ Q
may also be expressed as ‘P implies Q’. The truth table for P ⇒ Q is given
below.

P Q P ⇒Q
T T T
T F F
F T T
F F T

Note that a conditional proposition P ⇒ Q is true in all cases except where

the antecedent P is true and the consequent Q is false. It conforms to normal
usage that the phrase ‘if P , then Q’ should be regarded as false in the case
where P is true and Q is false. For example, suppose that, on a particular
day, I leave my umbrella at home but it does not rain that day. In this case
the statement
If I leave my umbrella at home, then it will rain
is certainly false. What is usually less easily appreciated is why the conditional
P ⇒ Q is true in each of the last two rows; in other words, P ⇒ Q is true
when P is false regardless of the truth value of Q. This is best understood
by considering a suitable example. Consider the first conditional proposition
listed above
If the backspace key is pressed, then the cursor moves one
place to the left
and consider the cases when the backspace key (on a computer keyboard) is
not pressed. The proposition makes no claims about what happens in this
case. If the cursor remains in the same place, then the conditional proposition
is true. Equally, if the cursor does move one place to the left (possibly caused
22

by a mouse or touchpad, for example) then the conditional statement is also

true (because it cannot be reasonably be regarded as false).

Similar reasoning can be applied to the other examples listed above. Consider,
for example, the situation where I carry my umbrella with me when I leave
home. Regardless of whether or not it then rains, the statement

If I leave my umbrella at home, then it will rain

cannot be regarded as false, so must therefore be considered true.

There are other ways of expressing conditional propositions. As noted above,

the phrase ‘P implies Q’ is also denoted by P ⇒ Q, although this is used
more frequently in mathematics than in everyday speech. Thus

n is a prime number implies n is not a perfect square

is a natural-sounding sentence in mathematics, whereas

I leave my umbrella at home implies it will rain

would sound a little curious in everyday conversation. It is for this reason that
P ⇒ Q is often read as ‘if P , then Q’ instead of ‘P implies Q’. In mathematics,
if P ⇒ Q we also say that P is a sufficient condition for Q and that Q is a
necessary condition for P .

It is also worth noting that P ⇒ Q is sometimes denoted by the phrase ‘Q if

P ’ where the order of P and Q in the sentence is reversed. Thus

You will receive a fine if you break the speed limit

is an alternative way of expressing ‘If you break the speed limit, then you will
receive a fine.’

Unlike conjunction and disjunction, conditional statements are not symmetric

in their component propositions. Conjunction and disjunction are symmetric
in the sense that P ∧ Q and Q ∧ P have the same meaning as do P ∨ Q and
Q ∨ P . However, P ⇒ Q and Q ⇒ P are not equivalent logically. For example,
the conditional propositions

If you break the speed limit, then you will receive a fine

and

If you receive a fine, then you will break the speed limit

have quite different meanings.

We can illustrate this generally by comparing the truth tables for P ⇒ Q and
Q ⇒ P as shown below.
 23

P Q P ⇒Q Q⇒P
T T T T
T F F T
F T T F
F F T T

This illustrates that P ⇒ Q and Q ⇒ P only have the same truth value
when P and Q are either both true or both false. The conditional proposition
Q ⇒ P is called the converse of P ⇒ Q.

Biconditional

The biconditional connective is usually expressed in English by the phrase

‘. . . if and only if . . . ’ usually between the two component propositions. An
example is:
The backspace key is pressed if and only if the cursor
moves one place to the left.
If P and Q are propositions, then the biconditional ‘P if and only if Q’ is
denoted by P ⇔ Q. The truth table for the biconditional is given below. Note
that the biconditional P ⇔ Q is true when P and Q have the same truth
values, both true or both false, and P ⇔ Q is false when P and Q have
opposite truth values.

P Q P ⇔Q
T T T
T F F
F T F
F F T

As we have seen with other connectives, there are other ways of expressing a
biconditional proposition. An alternative to ‘P if and only if Q’ is to use the
phrase ‘If and only if P , then Q’ as in the following example.
If and only if the backspace key is pressed, then the cursor moves
one place to the left.
However, since the truth values of P ⇔ Q are symmetric in P and Q, we
usually prefer the more symmetric way of expressing P ⇔ Q as ‘P if and only
if Q’. In the case where P ⇔ Q we also say that P is a necessary and sufficient
condition for Q (and also that Q is a necessary and sufficient condition for
P ). The biconditional is not frequently used in everyday discourse, which is
one reason why the linguistic constructions for it appear a little odd.
24

Truth tables

Using negation and the five connectives defined above, we can build more
complicated compound propositions such as

(P ∧ Q) ⇒ ¬R, ¬(P ∨ Q) ∧ (P ∨ S) and (P ∨ ¬Q) ⇒ (P ∧ Q).

In order to obtain the truth values for these more complicated propositions,
we may build their truth tables column by column as we now illustrate.

Examples 2.2

1. Construct a truth table for (P ∨ ¬Q) ⇒ (P ∧ Q).

Solution
Firstly, note that there are two simple propositions, P and Q, involved
in the compound proposition so, to account for the possible truth values,
we will require four rows to the truth table. For the sake of consistency,
we will always list the truth values in the following order.

P Q ...
T T ...
T F ...
F T ...
F F ...

Before we can find the truth values of this conditional proposition, we

need to obtain the truth values of the antecedent P ∨ ¬Q and the con-
sequent P ∧ Q. The antecedent also requires the negation of Q, so we
build the next three columns of the truth table as follows.

P Q ¬Q P ∨ ¬Q P ∧ Q
T T F T T
T F T T F
F T F F F
F F T T F

Finally, we can complete the table using the truth values in the last two
columns to evaluate the truth values of (P ∨ ¬Q) ⇒ (P ∧ Q). We can
do this in two ways.
Firstly, referring to the truth table of the conditional P ⇒ Q on page 21,
note that a conditional proposition is true in all cases where the an-
tecedent is false. In our example, that situation only occurs in row 3
 25

where P ∨ ¬Q is false; hence in row 3, (P ∨ ¬Q) ⇒ (P ∧ Q) has truth

value T. Referring again to the truth table of P ⇒ Q, note that, when
the antecedent is true, the truth values of a conditional agree with the
truth values of its consequent. In our example, the antecedent is true in
rows 1, 2 and 4 so, in these rows, the truth values of (P ∨¬Q) ⇒ (P ∧Q)
follow those of its consequent P ∧ Q.

P Q ¬Q P ∨ ¬Q P ∧ Q (P ∨ ¬Q) ⇒ (P ∧ Q)
T T F T T T
T F T T F F
F T F F F T
F F T T F F

The alternative way of obtaining the final column of truth values is

to note that the only occasion when a conditional is false is when its
antecedent is true and its consequent is false. This occurs in rows 2
and 4, so (P ∨ ¬Q) ⇒ (P ∧ Q) is false in these rows and true otherwise.
Note that the vertical lines between columns are not significant to the
meaning of the truth table. They may be inserted appropriately to break
up the table and make it easier to read.

2. Construct a truth table for (P ∨ Q) ⇒ (Q ∧ R).

Solution
Since there are three simple propositions P , Q, and R involved in the
compound proposition, we will need eight rows in the truth table to
capture all the possible combinations of truth values. Again, for the
sake of consistency and to ensure that we always list all eight different
possibilities, we will always list the truth values in the order shown
below.

P Q R ...
T T T ...
T T F ...
T F T ...
T F F ...
F T T ...
F T F ...
F F T ...
F F F ...

To complete the truth table for (P ∨ Q) ⇒ (Q ∧ R), we first need to

26

construct columns containing the truth values of P ∨ Q and Q ∧ R. Then

we can use the reasoning outlined in the previous example to obtain the
truth values for the conditional (P ∨ Q) ⇒ (Q ∧ R). In this way, we
obtain the following truth table.

P Q R P ∨Q Q∧R (P ∨ Q) ⇒ (Q ∧ R)
T T T T T T
T T F T F F
T F T T F F
T F F T F F
F T T T T T
F T F T F F
F F T F F T
F F F F F T

3. Construct a truth table for (P ∧ ¬R) ⇔ (Q ∧ (P ∨ R)).

Solution
Building up the truth table column by column, we obtain the truth table
shown below. Note that, so that the table does not become unwieldy, we
have labelled P ∧ ¬R as A (column 5) and Q ∧ ¬(P ∨ R) as B (column
7). This allows us to express the compound proposition (P ∧ ¬R) ⇔
(Q ∧ (P ∨ R)) simply as A ⇔ B.
Observing the truth table for the biconditional proposition P ⇔ Q on
page 23, we may note that a biconditional is true when its component
simple propositions have the same truth value and is false when its
component simple propositions have the opposite truth values. This ob-
servation allows us to complete the final column of the truth table.

A B
P Q R ¬R P ∧ ¬R P ∨R Q ∧ (P ∨ R) A⇔B
T T T F F T T F
T T F T T T T T
T F T F F T F T
T F F T T T F F
F T T F F T T F
F T F T F F F T
F F T F F T F T
F F F T F F F T
 27

Logic and natural language

As we have seen, natural language is more expressive than propositional logic

as it is able to express nuances that we cannot capture using logical expres-
sions. For example, each of the two English sentences
Max played for the first team and he scored the wining
try
and
Max played for the first team although he was nursing
an injury
is logically a conjunction. However, the different way the sentences are ex-
pressed may carry nuances that would not be captured simply by saying that
each is a conjunction P ∧ Q.
As we have noted, natural language is also prone to ambiguity. It is for this
reason, for example, that the language of legal documents, such as contracts,
often seems complicated and contrived as it attempts to eliminate ambigu-
ity. Some sentences in English may have more than one completely different
meaning. Consider, for example, the meaning of the sentence, ‘She hit the
man with a walking stick.’ Similarly, small changes in punctuation can change
the meaning of English sentences. A well-known example is the phrase, ‘Eats,
shoots and leaves’ 3 , which has quite different meanings with and without the
comma. Of course, many comedians use the ambiguity of natural language to
great humorous effect.
It should be clear, then, that it is not always straightforward ‘translating’ be-
tween logic and natural language. Using brackets, logical sentences can con-
vey quite complicated propositions which natural language may sometimes
struggle to express. For example, consider how we might distinguish in En-
glish between two similar sentences with logical structures (P ∧ Q) ∨ R and
P ∧ (Q ∨ R) respectively.

Examples 2.3
1. Let P : The sun is shining.
Q: It is raining.
R: There is a rainbow.
Translate each of the following propositions into symbolic form.
(i) If the sun is shining and it is raining, then there is a rainbow.
(ii) It is not raining and the sun is shining.
(iii) If there is no rainbow, then the sun is not shining.
3 This was taken by Lynne Truss as the title for her best-selling book on the importance

of correct punctuation.
28

(iv) Either it is not raining or, if the sun is shining, then there is a
rainbow.
(v) There is not a rainbow and it is raining if and only if the sun is
not shining.

Solution

(i) There are two connectives in the sentence: a conditional ‘if . . . then
. . . ’ and a conjunction ‘. . . and . . . ’. However the ‘top-level’ struc-
ture of the sentence is a conditional and the conjunction only ap-
pears in the antecedent of the conditional proposition:
If (the sun is shining and it is raining), then (there
is a rainbow ).
Therefore we may symbolise the proposition as

(P ∧ Q) ⇒ R.

(ii) This proposition is a conjunction where the first conjunct is the

negation of Q. Hence we may symbolise it as

(¬Q) ∧ P.

There is a convention in writing logic sentences where negation

‘binds more tightly’ than the other connectives. This means that
the negation symbol ¬ only negates what is immediately to its
right, so we may omit the brackets and symbolise the proposition
as
¬Q ∧ P.
Had we wanted to negate the whole conjunction Q∧P , it is raining
and the sun is shining, then we would need to use brackets as
follows: ¬(Q ∧ P ). This convention in logic is similar to one in
arithmetic or algebra where the negative also binds more tightly
than the binary operations of addition and multiplication. Thus,
for example, −3 + 5 is interpreted as (−3) + 5 = 2 and not as
−(3 + 5) = −8.
(iii) Again, the highest-level structure of the sentence is a conditional,
where the antecedent is the negation of R and the consequent is the
negation of P . Hence, using the convention mentioned in part (ii)
above, we may symbolise this as

¬R ⇒ ¬P.

(iv) The sentence contains both a disjunction ‘(either) . . . or . . . ’ and

a conditional ‘if . . . then . . . ’. In order to symbolise the sentence,
 29

we need to decide two things: at the highest level, is the sentence

a disjunction or a conditional and is the disjunction inclusive or
exclusive? Firstly, it should be clear that the basic structure of the
sentence is a disjunction with the conditional forming the second
disjunct. The use of ‘either . . . or . . . ’ — and also knowledge of
when rainbows actually occur — may indicate that the disjunction
is exclusive, in which case the sentence is symbolised as

¬Q Y (P ⇒ R).

Alternatively, we may regard the use of ‘either . . . or . . . ’ simply as

a device to emphasise that the top level structure is a disjunction
rather than a conditional. In this case, we would apply the conven-
tion that disjunctions are interpreted as inclusive unless they are
explicitly denoted as exclusive and symbolise the sentence as

¬Q ∨ (P ⇒ R).

Either of these is a reasonable interpretation of the English lan-

guage sentence and we leave the reader to decide which is their
preferred interpretation. This again emphasises that natural lan-
guage may sometimes be ambiguous and open to more than one
interpretation.
(v) Clearly this sentence contains a conjunction ‘. . . and . . . ’ and a
biconditional ‘. . . if and only if . . . ’. However it is not obvious, from
the English sentence, how to bracket the simple propositions, so

¬R ∧ (Q ⇔ ¬P ) and (¬R ∧ Q) ⇔ ¬P

are equally valid interpretations. This is a little problematic since

we have no reason to choose one interpretation over the other and
the two interpretations are logically different.

2. Let P : Jo is on holiday.
Q: Jo goes swimming.
R: Jo studies logic.
Translate each of the following propositions into naturally sounding En-
glish sentences.
(i) P Y Q
(ii) Q ⇒ ¬R
(iii) R ∨ (P ⇒ Q)
(iv) (R ∧ ¬Q) ⇒ ¬P
(v) (P ∧ Q) ∨ R
30

Solution

(i) This is an exclusive disjunction, so we need to ensure that this

is made explicit in the natural language sentence. Hence we may
express P Y Q in English as:
Jo is on holiday or she goes swimming but not both.
Note that, to keep the sentence as natural-sounding as possible, we
have used the pronoun ‘she’ rather than repeat the proper noun
‘Jo’ twice in the same sentence.
(ii) This is a conditional where the only complication is that the con-
sequent is a negated proposition. It can be expressed as:
If Jo goes swimming, then she doesn’t study logic.

(iii) The basic structure of the logic sentence is a disjunction where

the second disjunct is a conditional. We will use the construction
‘either . . . or . . . ’ to emphasise that the top-level structure of the
sentence is a disjunction and write it as
Either Jo studies logic or, if she is on holiday, she
goes swimming.
Note that we have also omitted the word ‘then’ from the conditional
statement that forms the second disjunct in order to emphasise
further that the main connective in the sentence is a disjunction.
Some people will prefer to omit ‘either’ as this hints at an exclusive
disjunction and write the sentence instead as:
Jo studies logic or, if she is on holiday, she goes
swimming.
(iv) Here we have a conditional where the antecedent is a conjunction.
We write this as:
If Jo studies logic and doesn’t go swimming, then she
is not on holiday.
To emphasise that the principal structure of the sentence is condi-
tional, we have omitted the pronoun ‘she’ in the second part of the
conjunction ‘Jo studies logic and she doesn’t go swimming.’
(v) It can be tricky in written language to indicate the positioning of
the brackets in sentences such as (P ∧ Q) ∨ R. One way we can hint
at the placement of the brackets in this case is to omit the pronoun
‘she’ in the conjunction and write
Jo is on holiday and goes swimming or she studies
logic.
Of course, this would not have been possible had the three sim-
ple propositions involved in the sentence (P ∧ Q) ∨ R referred to
 31

three different objects or individuals. Another option is to use the

construction ‘either . . . or’ to indicate the bracketing, as follows.
Either Jo is on holiday and she goes swimming or
she studies logic.
This is effective at indicating the bracketing, but there is a down-
side: some people are inclined to interpret ‘either . . . or’ as an ex-
clusive disjunction.

Tautologies and contradictions

Consider the truth table for the compound proposition (P ∧ Q) ⇒ (P ∨ Q)

given below.

P Q P ∧Q P ∨Q (P ∧ Q) ⇒ (P ∨ Q)
T T T T T
T F F T T
F T F T T
F F F F T

Notice that, no matter what the truth values of the component simple propo-
sitions P and Q, the compound proposition (P ∧ Q) ⇒ (P ∨ Q) is always true.
This means that the compound proposition is true by virtue of the structure
of the proposition itself and not its content. We can see this by substituting
particular propositions for P and Q. For example, suppose

P : Today is my birthday
Q: I am happy.

Then (P ∧ Q) ⇒ (P ∨ Q) is the proposition

If today is my birthday and I’m happy, then today is my birth-

day or I’m happy,

which is clearly true irrespective of whether or not it really is my birthday or

whether or not I really am happy. Such a compound proposition that is true
by virtue of its structure is called a ‘tautology’.

Clearly, the negation of a tautology will always have truth value F. Any com-
pound proposition with this property is called a ‘contradiction’. An example
of a contradiction is the proposition (¬P ∧ Q) ∧ (P ∨ ¬Q). This is shown in
the following truth table.
32

P Q ¬P ¬Q ¬P ∧ Q P ∨ ¬Q (¬P ∧ Q) ∧ (P ∨ ¬Q)
T T F F F T F
T F F T F T F
F T T F T F F
F F T T F T F

Definition 2.1
A compound proposition is a tautology if it is true for all combinations of
truth values of its component (simple) propositions.
A compound proposition is a contradiction if it is false for all combinations
of truth values of its component (simple) propositions.

Exercises 2.1

1. Construct truth tables for each of the following propositions.

(i) ¬(P ∨ ¬Q)

(ii) (P ∧ Q) ⇔ ¬Q
(iii) P Y (Q ⇒ ¬P )
(iv) (P ⇒ Q) ∧ (Q ⇒ P )
(v) (P ⇒ R) ∧ (R ⇒ Q)
(vi) (P ⇒ ¬Q) ∧ R
(vii) P ∨ ¬(R ⇔ Q)
(viii) ¬(P ∧ Q) ⇒ (P ∧ ¬R)
(ix) P ∨ (Q ⇔ R)
(x) (P ∧ R) ∨ (¬P ∧ ¬Q)

2. The four propositions S, W , R, and T are defined as follows.

S: The sun shines.
W: The wind blows.
R: The rain falls.
T: The temperature rises.
Translate each of the following propositions into natural-sounding En-
glish sentences.

(i) W ⇒ (¬S ∨ R)
 33

(ii) (W ∧ R) ⇔ ¬S
(iii) (W ∨ R) ∧ ¬T
(iv) ¬(S ∧ W (⇒ (R ∨ ¬T )
(v) (¬R Y T ) ⇒ (S ∧ ¬W )
3. With the four propositions S, W , R, and T defined as in question 1
above, symbolise each of the following propositions.
(i) The wind blows but the rain doesn’t fall.
(ii) If the sun shines and the wind doesn’t blow, then the temperature
rises.
(iii) The temperature rises if and only if the sun shines and the rain
doesn’t fall.
(iv) Either the sun shines and the temperature rises or the wind blows
and the rain falls.
(v) If the sun doesn’t shine or the wind blows with rain falling, then
the temperature doesn’t rise.
4. Determine whether each of the following propositions is a tautology, a
contradiction, or neither.
(i) P ∧ ¬P
(ii) ¬P ∨ P
(iii) P ⇒ (P ∨ Q)
(iv) (P ∧ Q) ∧ ¬(P ∨ Q)
(v) (P ⇒ Q) ∨ (Q ⇒ R)
(vi) (P ⇒ Q) ∧ (Q ⇒ R)
(vii) (P ∧ ¬P ) ⇒ (Q ∨ R)
(viii) (P ⇒ ¬Q) ∧ (¬R ⇒ P )
5. A compound proposition A is called a substitution instance of a com-
pound proposition B if A may be obtained from B by replacing each
occurrence of one or more of its simple constituent propositions by an-
other proposition; in making the substitution, every occurrence of a
given simple proposition must be replaced by the same proposition.
For example, each of the following is a substitution instance of the propo-
sition (P ∧ R) ⇒ (Q ∨ R):
Substitution instance Substitution
(P ∧ S) ⇒ (Q ∨ S) R −→ S
(P ∧ ¬R) ⇒ (Q ∨ ¬R) R −→ ¬R
(P ∧ (S ∨ T )) ⇒ (Q ∨ (S ∨ T )) R −→ S ∨ T
(¬P ∧ R) ⇒ ((P ∧ Q) ∨ R) P −→ ¬P, Q −→ (P ∧ Q)
34

(i) Which of the following propositions is a substitution instance of

the compound proposition (P ⇒ Q) ∧ ¬(P ∨ R)? If the proposition
is a substitution instance, state the substitution that connects it to
(P ⇒ Q) ∧ ¬(P ∨ R).
(a) (¬P ⇒ Q) ∧ ¬(¬P ∨ R)
(b) (P ⇒ R) ∧ ¬(P ∨ Q)
(c) (P ⇒ Q) ∧ ¬(S ∨ R)
(d) (P ⇒ ¬Q) ∧ ¬(P ∨ (R ⇔ S))
(e) (P ⇒ Q) ∨ ¬(P ∧ R)
(f) ((R ⇒ S) ⇒ S) ∧ ¬((R ⇒ S) ∨ R)
(ii) Explain why, if B is a tautology, then any substitution instance of
B is also a tautology and similarly if B is a contradiction, then any
substitution instance of B is also a contradiction.
(iii) Using part (ii) and the results of question 4, determine whether
each of the following propositions is a tautology, a contradiction,
or neither.
(a) (P ∨ Q) ∧ ¬(P ∨ Q)
(b) ¬(P ⇒ Q) ∨ (P ⇒ Q)
(c) P ⇒ (P ∨ (Q ∧ R)
(d) (P ⇒ (Q ∧ S)) ∨ ((Q ∧ S) ⇒ R)

2.3 Logical equivalence and logical implication

On page 25, we derived the truth table for (P ∨ ¬Q) ⇒ (P ∧ Q). A simplified
version of the truth table, without the columns that show how the truth values
are built up, is repeated below.

P Q ... (P ∨ ¬Q) ⇒ (P ∧ Q)
T T ... T
T F ... F
F T ... T
F F ... F

Looking at the table, we notice a curious fact: the truth values for (P ∨
¬Q) ⇒ (P ∧Q) are the same as those for Q. So, although the two propositions
((P ∨ ¬Q) ⇒ (P ∧ Q) and Q) are structurally quite different, they have the
same logical outcome in terms of their truth values. In this case we say that
the two propositions are ‘logically equivalent’.
 35

Definition 2.2
Two propositions P1 and P2 are logically equivalent, written P1 ≡ P2 , if
they have the same truth values for all combinations of truth values of their
component simple propositions.

With this definition, we may write

[(P ∨ ¬Q) ⇒ (P ∧ Q)] ≡ Q.

Note that logical equivalence ≡ is a relationship between propositions (having

the same truth values) and is not a new logical connective, since P ≡ Q is not
a new proposition.
Intuitively, we can think of logically equivalent propositions as, in a sense,
saying the same thing in a different way. This will be very important when we
come to proving mathematical propositions. As we shall see, if we need to prove
a proposition P1 , sometimes it will be easier to prove a logically equivalent
proposition P2 because the structure of P2 is easier to work with. But this
is legitimate since P1 and P2 are true under exactly the same circumstances.
Hence it will be important for us to establish certain logical equivalences that
we will be able to use, where appropriate, when constructing mathematical
proofs.

Examples 2.4
1. Show that P ⇒ Q is logically equivalent to ¬P ∨ Q.
Solution
We first construct a truth table giving the truth values for both P ⇒ Q
and ¬P ∨ Q.

P Q ¬P ¬P ∨ Q P ⇒Q
T T F T T
T F F F F
F T T T T
F F T T T

The fourth and fifth columns of the table show that ¬P ∨ Q and P ⇒ Q
have the same truth values for all combinations of truth values of P and
Q. Therefore
(¬P ∨ Q) ≡ (P ⇒ Q).
There is a danger that expressions like this stating a logical equivalence
can become complicated through the use of brackets. From now on we
will generally omit the brackets around P1 and P2 in the expression
P1 ≡ P2 .
36

2. Show that [P ⇒ (Q ∧ R)] ≡ [(P ⇒ Q) ∧ (P ⇒ R)].

Solution
As before, we construct a truth table showing the truth values of both
P ⇒ (Q ∧ R) and (P ⇒ Q) ∧ (P ⇒ R).

(P ⇒ Q)
P Q R Q∧R P ⇒ (Q ∧ R) P ⇒Q P ⇒R
∧(P ⇒ R)
T T T T T T T T
T T F F F T F F
T F T F F F T F
T F F F F F F F
F T T T T T T T
F T F F T T T T
F F T F T T T T
F F F F T T T T

Comparing columns 5 and 8 of the truth table, we see that P ⇒ (Q ∧ R)

and (P ⇒ Q) ∧ (P ⇒ R) have the same truth values. Hence
P ⇒ (Q ∧ R) ≡ (P ⇒ Q) ∧ (P ⇒ R).

There is a simple relationship between logically equivalent propositions and

the notion of a tautology. Suppose that two compound propositions P1 and
P2 are logically equivalent, P1 ≡ P2 . Recall that a biconditional proposition
is true precisely when the truth values of its two components are the same
(page 23). Since the truth values of P1 and P2 are identical, it means that
the biconditional proposition P1 ⇔ P2 will always have truth value T and is
therefore a tautology. In other words, if we know that P1 ≡ P2 , then it follows
that the biconditional proposition P1 ⇔ P2 is a tautology. It works the other
way round as well. Suppose that P1 ⇔ P2 is a tautology. Then all the truth
values of P1 ⇔ P2 are T. Since a biconditional proposition is true precisely
when its two components have the same truth values, this means that P1 and
P2 are logically equivalent. In summary, we have established the following.

Theorem 2.1
Let P1 and P2 be two (compound) propositions. Then P1 ≡ P2 if and only if
P1 ⇔ P2 is a tautology.

Example 2.5
Show that the proposition ¬(P ∧ Q) ⇔ (¬P ∨ ¬Q) is a tautology. Deduce that
¬(P ∧ Q) is logically equivalent to ¬P ∨ ¬Q.
 37

Solution
As always we begin by constructing the appropriate truth table.

A B
P Q ¬P ¬Q P ∧Q ¬(P ∧ Q) ¬P ∨ ¬Q A⇔B
T T F F T F F T
T F F T F T T T
F T T F F T T T
F F T T F T T T

The last column, containing the truth values of ¬(P ∧ Q) ⇔ (¬P ∨ ¬Q), only
has truth values T. Hence ¬(P ∧ Q) ⇔ (¬P ∨ ¬Q) is a tautology. It therefore
follows from theorem 2.1 that

¬(P ∧ Q) ≡ ¬P ∨ ¬Q.

This logical equivalence, which could be paraphrased as ‘the negation of a

conjunction is the disjunction of the negations’, is one of two equivalences
known collectively as ‘De Morgan’s Laws’. It should be noted that, if we only
need to establish the logical equivalence, ¬(P ∧ Q) ≡ ¬P ∨ ¬Q, it is simpler
to do so directly by comparing their truth values — in columns 6 and 7 of the
truth table above. Establishing that the biconditional ¬(P ∧ Q) ⇔ (¬P ∨ ¬Q)
is a tautology required an additional column in the truth table.

Standard logical equivalences

In ordinary algebra (or arithmetic), we are familiar with rules that allow us
to simplify expressions. Here are some simple examples:

x + 0 = x, x − (−y) = x + y, and x(y + z) = xy + xz.

There are similar logical equivalences that will allow us to manipulate logical
expressions much as we might manipulate algebraic ones. Some of the logical
rules are comparable to their algebraic counterparts and some are unique to
logic, as the following examples demonstrate.

Examples 2.6

1. Distributive law of conjunction over disjunction

Show that P ∧ (Q ∨ R) ≡ (P ∧ Q) ∨ (P ∧ R).
Note that this law parallels the distributive law of multiplication over
addition: x(y + z) = (xy) + (xz).
38

Solution
Consider the following truth table.

(P ∧ Q)
P Q R Q∨R P ∧ (Q ∨ R) P ∧Q P ∧R
∨(P ∧ R)
T T T T T T T T
T T F T T T F T
T F T T T F T T
T F F F F F F F
F T T T F F F F
F T F T F F F F
F F T T F F F F
F F F F F F F F

Since the truth values in columns 4 and 7 are identical, we have

P ∧ (Q ∨ R) ≡ (P ∧ Q) ∨ (P ∧ R).

2. Absorption law of conjunction over disjunction

Show that P ∧ (P ∨ Q) ≡ P .
Note that this law has no counterpart in ordinary algebra.

Solution
The truth values for P ∧ (P ∨ Q) are given in the following truth table.

P Q P ∨Q P ∧ (P ∨ Q)
T T T T
T F T T
F T T F
F F F F

The truth values for P ∧ (P ∨ Q) in column 4 are the same as those for
P in column 1 so P ∧ (P ∨ Q) ≡ P .

In table 2.1 below we list a number of standard logical equivalence laws,

together with their usual names. Each of these may be established using truth
tables as in our previous examples. In the table, ‘true’ stands for any tautology
and ‘false’ stands for any contradiction.
As we noted above, these laws can be used to manipulate logical expressions
 39

Logical Equivalence Laws

Idempotent Laws P ∧P ≡P
P ∨P ≡P

Commutative Laws P ∧Q≡Q∧P

P ∨Q≡Q∨P

Associative Laws (P ∧ Q) ∧ R ≡ P ∧ (Q ∧ R)
(P ∨ Q) ∨ R ≡ P ∨ (Q ∨ R)

Absorption Laws P ∧ (P ∨ Q) ≡ P
P ∨ (P ∧ Q) ≡ P

Distributive Laws P ∧ (Q ∨ R) ≡ (P ∧ Q) ∨ (P ∧ R)
P ∨ (Q ∧ R) ≡ (P ∨ Q) ∧ (P ∨ R)

Involution Law ¬(¬P ) ≡ P

De Morgan’s Laws ¬(P ∧ Q) ≡ ¬P ∨ ¬Q

¬(P ∨ Q) ≡ ¬P ∧ ¬Q

Identity Laws P ∨ false ≡ P

P ∧ true ≡ P
P ∨ true ≡ true
P ∧ false ≡ false

Complement Laws P ∨ ¬P ≡ true

P ∧ ¬P ≡ false
¬false ≡ true
¬true ≡ false

TABLE 2.1: Logical equivalence laws.

40

without having to resort to truth tables. This is illustrated in the following

example where we compare the solution using the laws with that using truth
tables. Note that for logic expressions that involve, say, four or more simple
propositions, truth tables are large and somewhat unwieldy.

Example 2.7
Show that P ∧ [(Q ∨ (P ∧ R)] ≡ P ∧ (Q ∨ R).

Solution 1
For our first solution, we manipulate the expression on the left-hand side using
the standard logical equivalences.
P ∧ [(Q ∨ (P ∧ R)] ≡ P ∧ [(Q ∨ P ) ∧ (Q ∨ R)] Distributive law
≡ [P ∧ (Q ∨ P )] ∧ (Q ∨ R) Associative law
≡ [P ∧ (P ∨ Q)] ∧ (Q ∨ R) Commutative law
≡ P ∧ (Q ∨ R) Absorption law

Solution 2
As an alternative, we could draw up the following truth table.
P Q R P ∧ R Q ∨ (P ∧ R) P ∧ [Q ∨ (P ∧ R)] Q ∨ R P ∧ (Q ∨ R)
T T T T T T T T
T T F F T T T T
T F T T T T T T
T F F F F F F F
F T T F T F T F
F T F F T F T F
F F T F F F T F
F F F F F F F F

The truth values in columns 6 and 8 are identical, so P ∧ [(Q ∨ (P ∧ R)] ≡

P ∧ (Q ∨ R).

The standard logical equivalence laws listed in the table above describe what
we might call the ‘algebra’ of propositions. In addition to these laws, we now
introduce three more logical equivalence laws that will be used when we come
to consider mathematical proofs.

Contrapositive Law P ⇒ Q ≡ ¬Q ⇒ ¬P .
The following truth table establishes the logical equivalence since the truth
values in columns 5 and 6 are identical. The proposition ¬Q ⇒ ¬P is called
 41

the contrapositive of the conditional P ⇒ Q. Hence the contrapositive law

says that a conditional is logically equivalent to its contrapositive.

P Q ¬P ¬Q P ⇒ Q ¬Q ⇒ ¬P
T T F F T T
T F F T F F
F T T F T T
F F T T T T

Contradiction Law (¬P ⇒ false) ≡ P .

Recall that false denotes any contradiction; that is, any compound proposi-
tion that only has truth value F. The truth table that establishes the logical
equivalence only requires two rows, as follows.

P ¬P false ¬P ⇒ false
T F F T
F T F F

The truth values in the last column are the same as those for P in the first
column so (¬P ⇒ false) ≡ P .

Biconditional Law P ⇔ Q ≡ (P ⇒ Q) ∧ (Q ⇒ P ).

Recall that the proposition Q ⇒ P is called the converse of P ⇒ Q (page 23).

The biconditional law therefore states that a biconditional proposition is log-
ically equivalent to the conjunction of the corresponding conditional and its
converse. The following truth table establishes the law since the truth values
in columns 3 and 6 are the same.

P Q P ⇔Q P ⇒Q Q⇒P (P ⇒ Q) ∧ (Q ⇒ P )
T T T T T T
T F F F T F
F T F T F F
F F T T T T
42

Logical implication

When constructing mathematical proofs, we will want to know that each state-
ment in a proof ‘follows logically’ from previous statements. In other words,
we will wish to ensure that the truth of a statement in a proof is guaranteed
by the truth of the statements that come before it in the proof. When dealing
with propositions, this notion is called ‘logical implication’.
Let P1 and P2 be two (possibly compound) propositions such that P2 is
true in all cases where P1 is true. Then we say that P1 logically im-
plies P2 , written P1  P2 . This notion may be extended to several proposi-
tions. Let P1 , P2 , . . . , Pn and Q be propositions. We say that the propositions
P1 , P2 , . . . , Pn logically imply the proposition Q, written P1 , P2 , . . . , Pn 
Q, provided Q is true in all those cases where each of the propositions
P1 , P2 , . . . , Pn is true. For the sake of clarity of notation, we will also write
{P1 , P2 , . . . , Pn }  Q in this case. Since the conjunction P1 ∧ P2 ∧ . . . ∧
Pn is true precisely when each of the propositions P1 , P2 , . . . , Pn is true,
{P1 , P2 , . . . , Pn }  Q is equivalent to saying P1 ∧ P2 ∧ . . . ∧ Pn  Q.

Examples 2.8
1. Show that (P ⇒ Q) ∧ (Q ⇒ R)  (P ⇒ R).
Solution
A truth table giving the truth values of both (P ⇒ Q) ∧ (Q ⇒ R) and
P ⇒ R is given below.

P Q R P ⇒Q Q⇒R (P ⇒ Q) ∧ (Q ⇒ R) P ⇒R
T T T T T T T
T T F T F F F
T F T F T F T
T F F F T F F
F T T T T T T
F T F T F F T
F F T T T T T
F F F T T T T

There are four rows where (P ⇒ Q) ∧ (Q ⇒ R) is true — rows 1, 5,

7, and 8. In this case, these rows have been highlighted by emboldening
the truth value T although we will not routinely do this. In each of
these rows, P ⇒ R is also true. Hence P ⇒ R is true in all cases where
(P ⇒ Q) ∧ (Q ⇒ R) is true so
(P ⇒ Q) ∧ (Q ⇒ R)  (P ⇒ R).
 43

Note that P ⇒ R is also true in some situations where (P ⇒ Q) ∧ (Q ⇒

R) is false, namely those cases given by rows 3 and 6. However we only
require P ⇒ R to be true in all cases where (P ⇒ Q) ∧ (Q ⇒ R) is true;
what happens when (P ⇒ Q) ∧ (Q ⇒ R) is false is of no significance
here.
2. Show that {P ∨ Q, P ⇒ R, Q ⇒ R}  R.
Solution
We need to show that in those cases where P ∨ Q, P ⇒ R and Q ⇒ R
are all true, then so too is R. The following truth table gives the truth
values of P ∨ Q, P ⇒ R and Q ⇒ R.

P Q R P ∨Q P ⇒R Q⇒R
T T T T T T
T T F T F F
T F T T T T
T F F T F T
F T T T T T
F T F T T F
F F T F T T
F F F F T T

The propositions P ∨ Q, P ⇒ R and Q ⇒ R are all true only in rows 1,

3, and 5. In each of these rows, R is also true. Therefore R is true in all
those cases where each of the propositions P ∨ Q, P ⇒ R and Q ⇒ R
is true, so
{P ∨ Q, P ⇒ R, Q ⇒ R}  R.

Exercises 2.2
1. Using truth tables, establish each of the following logical equivalences.

(i) ¬(P ∨ Q) ≡ ¬P ∧ ¬Q
(ii) P ∨ (P ∧ Q) ≡ P
(iii) P ⇒ ¬Q ≡ ¬(P ∧ Q)
(iv) (P ∧ Q) ∨ R ≡ (P ∨ R) ∧ (Q ∨ R)
(v) (P ∨ Q) ⇒ R ≡ (P ⇒ R) ∧ (Q ⇒ R)
(vi) P ⇒ (Q ⇒ R) ≡ (P ∧ Q) ⇒ R
(vii) ((P ⇒ Q) ⇒ R) ≡ ((¬P ⇒ R) ∧ (Q ⇒ R)) .
44

(viii) ¬(P ∧ Q) ∧ R ≡ (¬P ∧ R) ∨ (¬(Q ∨ ¬R))

(ix) (P ∧ Q) ⇒ R ≡ (P ⇒ R) ∧ (Q ⇒ ¬P )
(x) (P ⇔ Q) ∨ R ≡ [(P ⇒ (Q ∨ R)) ∧ (Q ⇒ (P ∨ R))]

2. Using truth tables, establish each of the following logical implications.

(i) (P ∧ Q)  (P ∨ Q)
(ii) (P ⇔ Q) ∧ Q  P
(iii) {P ⇒ Q, P ∨ ¬Q}  ¬(P ∨ Q)
(iv) {P ⇒ (Q ∨ R), P ∧ R}  (P ∨ R)
(v) {P ∨ (Q ∧ R), R ⇒ P }  (¬Q ⇒ R)
(vi) {P ⇒ Q, (P ∧ Q) ⇒ R}  (P ⇒ R)
(vii) {P ⇒ R, Q ⇒ R, ¬P ⇒ Q}  (¬P ∨ R)

3. Use the logical equivalence laws on page 39 to establish each of the

following logical equivalences.

(i) ¬P ∨ (P ∧ Q) ≡ ¬P ∨ Q
(ii) ¬(P ∨ Q) ∨ (¬P ∧ Q) ≡ ¬P
(iii) ¬(¬P ∧ Q) ∧ (P ∨ ¬Q) ≡ (P ∨ ¬Q)
(iv) P ∨ (Q ∧ (P ∨ ¬Q)) ≡ P
(v) P ∧ [(P ∧ Q) ∨ ¬P ] ≡ (P ∧ Q)
(vi) (P ∧ Q) ∨ (¬P ∨ R) ≡ (¬P ∨ Q) ∨ R
(vii) ¬(P ∧ ¬(Q ∧ R)) ≡ (¬P ∨ Q) ∧ ¬(P ∧ ¬R)
(viii) [(P ∧ Q) ∨ (R ∧ S)] ≡ [(P ∨ R) ∧ (P ∨ S) ∧ (Q ∨ R) ∧ (Q ∨ S)]

4. Show that P ⇒ Q is logically equivalent to ¬P ∨ Q. This is sometimes

called the Law of Material Implication.
Using the law of material implication and the other logical equivalence
laws on page 39, establish each of the following.

(i) ¬Q ∧ (P ⇒ Q) ≡ ¬(P ∧ Q)
(ii) P ⇒ (P ∧ Q) ≡ (P ⇒ Q)
(iii) (P ∧ Q) ⇒ R ≡ P ⇒ (Q ⇒ R)
(iv) (P ∨ Q) ⇒ R ≡ (P ⇒ R) ∧ (Q ⇒ R)
(v) P ⇒ (Q ∨ R) ≡ (P ⇒ Q) ∨ (P ⇒ R)
 45

2.4 Predicates and quantification

In propositional logic, we cannot represent the fact that several propositions

may be referring to the same object or have the same subject matter. For
example, we would need to denote the propositions

Sarah is an athlete
Sarah is physically fit
All athletes are physically fit

using different letters (P , Q, R say), and then the fact that they are making
statements about the same individuals or properties of individuals would be
lost.
A proposition of the form ‘Sarah is an athlete’ may be regarded as having two
components. One is the object or individual — ‘Sarah’ in this case. The other
is the property which the object or individual is said to possess — being an
athlete, in this example. If we wish to be able to capture the fact that two
propositions refer to the same object or refer to the same property, then we
will need a notation for propositions that refer specifically to the objects and
their properties.
A predicate describes a property of one or more objects or individuals. We
will use uppercase letters to denote predicates. The following are examples of
predicates.

R : . . . is red.
A : . . . is an athlete.
F : . . . is physically fit.
T : . . . cannot be tolerated under any circumstances.

A predicate can be ‘turned into’ a proposition by supplying an appropriate

object or individual. With the predicates defined above, we have the following
propositions.

R(this rose) : This rose is red.

A(Sarah) : Sarah is an athlete.
F (Sarah) : Sarah is physically fit.
T (John): John cannot be tolerated under any circumstances.

The general notation for representing propositions using objects and predi-
cates should be clear from these examples. If P is a predicate and a is an
46

object or individual, then we use P (a) to represent the proposition ‘The ob-
ject or individual a has the property P ’. More compactly,

P (a) : a has property P.

We may negate propositions denoted in this way and build them into more
complex propositions using connectives as we have done previously. Using the
notation above, here are some examples.

¬R(this rose) : This rose is not red.

A(Sarah) ∧ A(Dan) : Sarah is an athlete and Dan is an athlete.
(or Sarah and Dan are athletes)
A(Sarah) ⇒ F (Sarah) : If Sarah is an athlete, then she is physically fit.

So far in this section we have simply introduced a slightly more expressive

notation for propositions that allows us to represent separately the object
being referred to and the property that it is claimed to possess. There are
various ways in which we can exploit this new notation. Suppose we replace
a specific object or individual by a variable. Then we obtain expressions like
the following.

R(x) : x is red.
A(x) : x is an athlete.
F (y) : y is physically fit.

These expressions are not propositions since they cannot be declared true or
false. They are called ‘propositional functions’ by analogy with ‘numerical’
functions. Readers familiar with functions (see chapter 3) will recognise an
expression such as f (x) = x2 + 2 as (partly) defining a function. Here, x is a
numerical variable. Substituting a value for x gives a value for the expression;
for example, when x = 3 we have f (3) = 32 + 2 = 11. It is similar with
propositional functions. For the propositional function A(x) : x is an athlete,
substituting an object such as ‘Sarah’ for the variable x gives a proposition
A(Sarah): Sarah is an athlete.
More formally, a propositional function is a logical expression which de-
pends on the value of one or more variables. Let P be some predicate which
we may read as ‘. . . has property P ’. Then P (x) is the propositional function
‘x has property P ’.
As we have seen, one way to convert a propositional function P (x) into a
proposition is to substitute (the name of) a particular object or individual
for the variable x. If a is a particular object or individual, then P (a) is a
proposition. Thus the propositional function P (x) represents a whole family
of possible propositions, one for each possible object or individual which may
 47

be substituted for the variable x. The relationship between predicates, propo-

sitional functions, and propositions is illustrated in the following example.

Predicate: M : . . . is my friend.
Propositional function: M (x) : x is my friend.
Propositions: M (Elaine) : Elaine is my friend.
M (Jack) : Jack is my friend.

It is also important to note that propositional functions behave in the similar

way to propositions. We can negate propositional functions and use all the
standard connectives between propositional functions. Using the predicates
defined previously, we have the following examples.

¬R(x) : x is not red.

R(x) ∧ R(y) : x is red and y is red.
(or x and y are red.)
A(x) ⇒ F (x) : If x is an athlete, then x is physically fit.
R(x) ∨ A(y) : x is red or y is an athlete.

Example 2.9
The following predicates are defined.

M : . . . is my friend.
R : . . . is rich.
F : . . . is famous.
B : . . . is boring.
Symbolise each of the following propositions.

(i) Jimmy is my friend but he’s not rich.

(ii) Sue and Mark are rich.
(iii) If Peter is my friend, then he is not boring.
(iv) Mark is rich or famous but not both.
(v) If Jane is rich and not boring, then she is my friend.

Solution

(i) Recall that ‘but’ signifies a conjunction (see page 19). Hence the propo-
sition is symbolised as
M (Jimmy) ∧ ¬R(Jimmy).
Note that we only use connectives between propositions (or proposi-
tional functions) and not between predicates themselves. Thus ‘(M ∧
¬R)(Jimmy)’ is not a well-formed proposition.
48

(ii) This is a conjunction of the propositions Sue is rich and Mark is rich
and so is:
R(Sue) ∧ R(Mark).
In the same way that we do not use connectives between predicates,
so it is also incorrect to use connectives between objects. Hence the
expression ‘R(Sue ∧ Mark)’ is also not a well-formed proposition.

(iii) This is a conditional where both the antecedent and the consequent refer
to Peter:
M (Peter) ⇒ ¬B(Peter).

(iv) This is a (rare) example of an exclusive disjunction:

R(Mark) Y F (Mark).

(v) The top-level structure of the proposition is that of a conditional:

(R(Jane) ∧ ¬B(Jane)) ⇒ M (Jane).

Quantifiers

Using predicates, or propositional functions, we can identify when propositions

refer to the same individual or the same property. However, we are still unable
to represent (except using a single letter) propositions such as the following.

All athletes are physically fit.

Some roses are red.

In general, if P (x) is a propositional function, then the following statements

are propositions since each can be declared true or false.

For all x, P (x) or All x have property P .

For some x, P (x) or Some x have property P .

These propositions are quantified propositional functions and the phrases

‘for all’ and ‘for some’ are called quantifiers. A proposition that asserts that
all objects (in a certain class or category) have a specified property is said
to be universally quantified. Universally quantified propositions may be
recognised by phrases such as ‘for all . . . ’, ‘for every . . . ’ or ‘for each . . . ’.
A proposition that asserts that some objects (in a certain class or category)
have a specified property is said to be existentially quantified. Existentially
quantified propositions may be recognised by phrases such as ‘some . . . ’, ‘there
exists . . . ’ or ‘at least one . . . ’.
As usual, in natural language there are different ways of expressing quantified
 49

propositions. The important point is that, however it is expressed, a univer-

sally quantified proposition is one that is true when all the specified objects
have the specified property. Similarly, an existentially quantified proposition
is one that is true when at least one of the specified objects has the specified
property. Existentially quantified statements are often expressed in the plural:
some roses are red. This could have been expressed as: some rose is red. In
either case, the proposition is true when there exists an example of a rose that
is red. Here are some further examples of quantified propositions expressed in
a variety of ways.

Universally quantified propositions

All fish swim in water.
Every bird can fly.
Each apple in this batch is bruised.

Existentially quantified propositions

Some birds cannot fly.
There exist black swans.
At least one prime number is even.

We need a way of symbolising quantified propositions. The only missing pieces

in the notation we have so far developed are symbols for the quantifiers ‘all’
and ‘some’ which we now introduce.
The universal quantifier ‘for all’ is denoted ∀ and a universally quantified
propositional function is denoted ∀x • P (x) or just ∀x P (x). Here the ‘bullet’
is being used simply as punctuation to separate the quantified variable ‘for
all x’ from the propositional function P (x) being quantified. We will use this
notation, but many authors omit the bullet. The following summarises how
we symbolise a universally quantified proposition and the different ways that
the symbols may be read.

∀x • P (x) : for all x, P (x)

for every x, P (x)
for each x, P (x)

The existential quantifier ‘there exists’ is denoted ∃ and an existentially

quantified propositional function is denoted ∃x • P (x) or just ∃x P (x). Again
the bullet is an optional, but useful, piece of punctuation. The symbol ∃ is
most commonly read as ‘there exists’ and, in this case, we may read the bullet
as ‘such that’. This is illustrated in the following summary of an existentially
quantified proposition and the different ways that the symbols may be read.

∃x • P (x) : there exists x, such that P (x)

for some x, P (x)
for at least one x, P (x)
50

Examples 2.10
1. The following predicates are defined.
M : . . . is my friend.
P : . . . is Peter’s friend.
U : . . . is prone to unruly behaviour.
S: . . . should be shunned.
Symbolise each of the following propositions.
(i) All of Peter’s friends are prone to unruly behaviour.
(ii) Some of my friends are also friends of Peter’s.
(iii) If all of Peter’s friends are prone to unruly behaviour, then some
of them should be shunned.
(iv) Some of Peter’s and my friends are prone to unruly behaviour.
(v) All of my friends are prone to unruly behaviour and some of them
should be shunned.
(vi) None of Peter’s friends are my friends.
Solution

(i) We first paraphrase this into a ‘half-way house’ between English

and symbolic logic:
For all x, if x is Peter’s friend, then x is prone to
unruly behaviour.
From this we can readily symbolise the proposition as:

∀x • P (x) ⇒ U (x).

Most people find the somewhat unnatural-sounding half-way house

to be a useful device in understanding the structure of the quan-
tified sentence. With some practise, however, it is usually possible
to be able to symbolise many sentences without resorting to this.
(ii) In this case, the ‘half-way house’ between English and symbolic
logic might be expressed as:
For some x, x is my friend and (also) x is Peter’s
friend.
Hence the symbolic version is:

∃x • M (x) ∧ P (x).

(iii) In this case the proposition is a conditional ‘if . . . then . . . ’ where

the antecedent is universally quantified and the consequent is ex-
istentially quantified. We begin by symbolising the antecedent and
consequent separately.
 51

Antecedent
Half-way house: For all x, if x is Peter’s friend, then
x is prone to unruly behaviour.
Symbolic expression: ∀x • P (x) ⇒ U (x)

Consequent
Half-way house: For some x, x is Peter’s friend and x
should be shunned.
Symbolic expression: ∃x • P (x) ∧ S(x)
Now we can symbolise the whole sentence as

(∀x • P (x) ⇒ U (x)) ⇒ (∃x • P (x) ∧ S(x)) .

(iv) This is similar to (ii) above in the sense that the sentence may be
paraphrased as saying some objects with the first property also have
the second property. In this case, the first property is a conjunction
so that the half-way house ‘pseudo-English’ statement is:
For some x, (x is my friend and x is Peter’s friend)
and x is prone to unruly behaviour.
Hence the symbolic version is

∃x • (M (x) ∧ P (x)) ∧ U (x).

By the associative law for conjunction, (P ∧ Q) ∧ R ≡ P ∧ (Q ∧ R)

(see page 39), so the brackets around M (x) ∧ P (x) may be omitted
and we can write this more simply as

∃x • M (x) ∧ P (x) ∧ U (x).

(v) This is similar to (iii) above in the sense that the proposition
contains two ‘sub-propositions’, one universally quantified and the
other existentially quantified. However, in this case the two com-
ponent quantified propositions are conjoined. If we separately sym-
bolise ‘all of my friends are prone to unruly behaviour’ and ‘some
of my friends should be shunned’ and then form their conjunction,
we obtain the following.

(∀x • M (x) ⇒ U (x)) ∧ (∃x • M (x) ∧ S(x))

(vi) The phrase ‘None of . . . ’ signifies the negation of ‘Some of . . . ’; in

other words, ‘none of . . . ’ is equivalent to ‘it is not the case that
some of . . . ’. So we first symbolise the sentence ‘some of Peter’s
friends are my friends’ and then negate the result. This gives:

¬∃x • P (x) ∧ M (x).

52

It is worth noting that there is a convention being applied here:

to negate a quantified propositional function, we simply write the
negation symbol in front of the quantifier
¬ (∃x • P (x)) ≡ ¬∃x • P (x)
¬ (∀x • P (x)) ≡ ¬∀x • P (x).

2. Symbolise each of the following propositions.

(i) All politicians wear suits and talk too much.
(ii) Not everybody approves of capital punishment.
(iii) Everyone was curious but no one was prepared to ask.
Solution
In these examples the predicates are not already defined so we need, in
each case, to define the predicates that we wish to use.
(i) There are three predicates involved in the proposition which we
need to define.
Let P : . . . is a politician
S : . . . wears a suit
T : . . . talks too much.
Then the half-way house is:
For all x, if x is a politician, then (x wears a suit
and x talks too much).
Hence we symbolise the proposition as
∀x • P (x) ⇒ (S(x) ∧ T (x)).
(ii) This is the negation of the proposition ‘everybody approves of cap-
ital punishment’. The term ‘everybody . . . ’ is a little awkward be-
cause it really means ‘all people . . . ’. Hence, one of the predicates
we will need will be ‘. . . is a person’.
Let P : . . . is a person
A : . . . approves of capital punishment.
Then the half-way house for the proposition ‘everybody approves
of capital punishment’ is
For all x, if x is a person then x approves of capital
punishment,
which we would symbolise as ∀x • P (x) ⇒ A(x). Hence the origi-
nal negated proposition ‘not everybody approves of capital punish-
ment’ is
¬∀x • P (x) ⇒ A(x).
 53

(iii) In this sentence, ‘everyone’ and ‘no one’ seem to be referring to

a specific group of individuals This may have already been made
clear in the context in which the sentence was uttered. One of the
predicates we will need, therefore, is ‘. . . is a person (belonging to
the specific group being referred to)’.
The phrase ‘no one . . . ’ represents ‘it is not the case that some-
one . . . ’; in other words, ‘no one . . . ’ signals the negation of the
existential quantifier.
Let P : . . . is a person (belonging to the specific group being
referred to)
C : . . . was curious
A : . . . was prepared to ask.
Then we can symbolise ‘everyone was curious’ as ∀x • P (x) ⇒
C(x). We may also symbolise ‘someone was prepared to ask’ as
∃x•P (x)∧A(x). Hence the given proposition ‘everyone was curious
but no one was prepared to ask’ may be represented as follows.

(∀x • P (x) ⇒ C(x)) ∧ (¬∃x • P (x) ∧ A(x))

Note that the previous examples all build on the following two general patterns
for symbolising universally and existentially quantified propositions.

All P s are Qs : ∀x • P (x) ⇒ Q(x)

Some P s are Qs : ∃x • P (x) ∧ Q(x)

Universe of Discourse

Consider again the proposition from example 2.10.2 (iii) above.

Everyone was curious but no one was prepared to ask.

When symbolising this, we noted that it appeared that the proposition was
most likely about some specific group of people. Furthermore, we needed a
predicate that referred to membership of the particular group. There is an al-
ternative approach in cases like this where the proposition is referencing some
known collection of objects or individuals. Instead of defining the predicate

P : . . . is a person (belonging to the specific group being referred to).

we could specify that the variable(s) will only range over the people in the
specific group. To do this we define the ‘universe of discourse’ to be the col-
lection of people in the specific group. Then any variable x will refer only
54

to people in the group. This gives a more natural and more efficient way of
representing the proposition, as follows.

Let Universe : people in the specific group being referred to

C : . . . was curious
A : . . . was prepared to ask.

Then we may symbolise the proposition as

(∀x • C(x)) ∧ (¬∃x • A(x)) .

Compare this with the previous symbolisation

(∀x • P (x) ⇒ C(x)) ∧ (¬∃x • P (x) ∧ A(x)) .

The saving, in both space and mental effort, comes from defining the particular
group of people once (as the universe) rather than having to refer to it (via a
predicate) every time we wish to refer to elements of the group.
In general, for a given propositional function P (x), the universe of discourse
is the set from which we may select an object or individual to substitute for x.
As we have seen, defining a universe of discourse can simplify the symbolisation
of quantified propositional functions. If a universe of discourse is not defined,
then we must assume that any object or individual may be substituted for x.
In principle, we may define separate universes for each propositional variable.
In practice, however, we will usually have a single universe for all the propo-
sitional variables under consideration. However, when there are two different
kinds of objects under consideration, it may make sense to define two separate
universes.

Example 2.11
Defining suitable universes as appropriate, symbolise each of the following
propositions.

(i) All of my friends are either rich or famous.

(ii) All athletes who can lift heavy weights cannot run fast.
(iii) Everybody was too hot and someone fainted.
(iv) If everyone was generous, then no one would go hungry.

Solution

(i) We define the universe to be ‘my friends’ and define predicates:

R : . . . is rich
A : . . . is famous.
 55

Then we may paraphrase the proposition as:

For all x (who is my friend ), x is rich or x is famous.
Hence we may symbolise as: ∀x • R(x) ∨ F (x).
(ii) Let Universe : athletes
H : . . . can lift heavy weights
R : . . . can run fast.
Then we may paraphrase the proposition as:
For all (athletes) x, if x can lift heavy weights, then
x cannot run fast.
Symbolically, we have: ∀x • H(x) ⇒ ¬R(x).
(iii) Again it appears that the sentence refers to a specific group of people,
so we define this group as the universe.
Let Universe : people in the specific group being referred to
H : . . . was too hot
F : . . . fainted.
Then the sentence is a conjunction of two propositions:
For all x, x was too hot and for some x, x fainted.
Therefore the symbolic expression is: (∀x • H(x)) ∧ (∃x • F (x)) .
(iv) Let Universe : people
G : . . . was generous
H : . . . goes hungry.
The proposition is a conditional where the antecedent and consequent
are both (separately) quantified propositions. The antecedent, ‘everyone
was generous’, is straightforward to symbolise: ∀x • G(x).
Recall that ‘no one . . . ’ signifies the negation of ‘someone . . . ’. Hence
the consequent may be symbolised as: ¬∃x • H(x).
Putting these together, we may symbolise the given proposition as:

(∀x • G(x)) ⇒ (¬∃x • H(x)) .

Two-place predicates

The predicates we have considered so far are one-place predicates. To con-

vert each such predicate to a proposition requires the name of a single member
of the universe of discourse or quantification over a single variable.
56

There are predicates which are relational in nature and require two objects to
convert them to propositions. Consider, for example, the predicate ‘belongs
to’. This describes the relationship of ‘belonging’ and requires two objects to
become a proposition: ‘this cat belongs to me’, for example. Predicates like this
that require two objects to convert them to propositions are called two-place
predicates or, sometimes, relational predicates.
When denoting two-place predicates, we will follow a similar notation to one-
place predicates and place the objects after the predicate symbol. However,
this time there will be a pair of objects. Thus if we define

B: ‘. . . belongs to . . . ’

then the proposition ‘this cat belongs to me’ is denoted

B(this cat, me).

There are two key points to note in relation to the objects required for a
two-place predicate. Firstly, their order is important. For example, the two
propositions

‘this cat belongs to me’ and ‘I belong to this cat’

have quite different meanings. Similarly, changing the order of the objects can
change the truth value of the resulting proposition. For example, ‘17 is greater
than 3’ is a true proposition, whereas ‘3 is greater than 17’ is false.
The second point to note is that the two objects may be drawn from different
universes. For example, consider the predicate

C : ‘. . . is the capital city of . . . ’

To convert this into a meaningful proposition, the first object needs to be a

city and the second object should be a country (or other entity, such as a state
or region, that has a capital city). Thus

Paris is the capital city of France

Brighton is the capital city of the United Kingdom
Little Rock is the capital city of Arkansas

are all properly defined propositions (of which two are true and one is false).
However
this banana is the capital city of the Eiffel tower, and
a herring is the capital city of Sarah

are simply not meaningful propositions.

Given a two-place predicate P , we may form a two-variable propositional
 57

function P (x, y) by supplying two variables x and y. We have seen that

replacing both variables with objects or individuals gives a proposition. Re-
placing just one of the variables with an object produces a single-variable
propositional function. For example,

B(x, y) : x belongs to y

is a two-variable proposition. Replacing one of the variables gives single-

variable propositional functions such as

B(this cat, y) : this cat belongs to y

B(x, me) : x belongs to me.

We may now quantify the remaining variable to obtain a proposition. For

example, assuming a universe of people for the second variable:

∃y • B(this cat, y) : this cat belongs to someone

∀x • B(x, me) : everything belongs to me.

In general, if P (x, y) is a two-variable propositional function and a and b are

objects (of the appropriate type), then each of the following are propositions:

P (a, b)
∀x • P (x, b)
∃x • P (x, b)
∀y • P (a, y)
∃y • P (a, y)

Finally, we may also convert a two-variable propositional function P (x, y) into

a proposition by quantifying over both of the variables. There are eight ways
in which this may be done.

∀x ∀y • P (x, y) ∀y ∀x • P (x, y)
∃x ∀y • P (x, y) ∀y ∃x • P (x, y)
∀x ∃y • P (x, y) ∃y ∀x • P (x, y)
∃x ∃y • P (x, y) ∃y ∃x • P (x, y)

Note that, when different quantifiers are used, their order is significant. For
example, consider the following propositions where the universe for both vari-
ables is the set of positive integers Z+ .

∀m ∃n • m < n ∃n ∀m • m < n

The first of these, ∀m ∃n • m < n, asserts that, for every positive integer
m, there is a positive integer n that is greater than it. This is clearly a true
58

proposition because there is no largest positive integer. The second proposi-

tion, ∃n ∀m•m < n, asserts that there exists a positive integer n that is larger
than every positive integer m. This is just as clearly a false proposition again
because there is no largest positive integer. In this case, changing the order of
the quantifiers has changed the truth value of the proposition.

Examples 2.12

1. The following one-variable and two-variable propositional functions are

given.
B(x, y) : x belongs to y
D(x, y) : x detests y
L(x, y) : x loves y
C(x) : x is a cat
F (x) : x is fierce
P (x) : x is a person
Write each of the propositions symbolised below as natural-sounding
English sentences.

(i) ∀x • C(x) ⇒ L(Max, x)

(ii) ∃x • C(x) ∧ F (x) ∧ B(x, Chloe)
(iii) ∀x ∃y • P (x) ⇒ (C(y) ∧ L(x, y))
(iv) ∀x ∀y • (C(x) ∧ F (x)) ⇒ (P (y) ⇒ D(y, x)
(v) ∃x ∀y • P (x) ∧ ((C(y) ∧ B(y, x)) ⇒ L(x, y))
(vi) ∃x ∃y • P (x) ∧ C(y) ∧ ¬F (y) ∧ D(x, y)

Solution

(i) The literal translation of the proposition is ‘for all x, if x is a cat,

then Max loves x’. An idiomatic version of this is ‘Max loves all
cats’ or simply ‘Max loves cats’.
(ii) The ‘pseudo English’ translation of the proposition is ‘for some x, x
is a cat and x is fierce and x belongs to Chloe’. A natural-sounding
sentence for this would be ‘there is a fierce cat that belongs to Chloe’
or, better, ‘Chloe has a fierce cat’.
(iii) Since the first propositional function P (x) does not involve the
variable y, the quantifier ∃y may be moved past P (x). In other
words,
∀x ∃y • P (x) ⇒ (C(y) ∧ L(x, y))
≡ ∀x • P (x) ⇒ (∃y • C(y) ∧ L(x, y))
 59

and the second logical sentence is easier to translate. Literally it

means ‘for all x, if x is a person, then there exists y such that y is
a cat and x loves y’. As a natural-sounding sentence we might say
‘everyone loves some cat’.
(iv) The literal ‘half-way house’ translation is ‘for all x and for all y,
if x is a cat and x is fierce, then if y is a person, y detests x. In
other words, ‘all people detest all fierce cats’ or, more naturally,
‘everyone detests fierce cats’.
(v) Again, since this is quite a complicated expression, we may move
the quantifier ∀y past the propositional function P (x) since it does
not involve the variable y and re-write the logical sentence as:

∃x ∀y • P (x) ∧ ((C(y) ∧ B(y, x)) ⇒ L(x, y))

≡ ∃x • P (x) ∧ (∀y • (C(y) ∧ B(y, x)) ⇒ L(x, y)) .

This second sentence can be translated as ‘for some x, x is a person

and, for all y, if y is a cat and y belongs to x, then x loves y’. As
an English sentence, we might write ‘there is someone who loves
all the cats that belong to them’ or, more naturally, ‘someone loves
all their cats’.
(vi) Literally, this translates into ‘pseudo English’ as ‘there exists x and
there exists y such that x is a person and y is a cat and y is not
fierce and x detests y’. As an idiomatic sentence: ‘someone detests
a cat that is not fierce’.

2. Let the universe of discourse (for both variables) be ‘people’ and define
the following two-variable propositional functions.
K(x, y) : x knows y
T (x, y) : x is taller than y
Y (x, y) : x is younger than y

Symbolise each of the following propositions.

(i) Jess knows everyone.

(ii) Someone knows everyone.
(iii) Everyone knows Ahmed.
(iv) Everyone knows someone.
(v) Everyone knows someone who is taller than them.
(vi) Everyone Misha knows is taller than him.
(vii) Sam is taller than everyone (who is) younger than him.
(viii) Everyone is taller than all those people who are younger than them.
60

Solution
(i) We may rewrite the sentence as, ‘for all y, Jess knows y’. Hence we
may symbolise it as: ∀y • K(Jess, y).
Note that the name of the variable is not important; we chose y as
it appeared in the second place in the propositional function, but
any variable name would be acceptable. Thus each of the following
are also acceptable, even if the third is not one we would normally
choose!
∀x • K(Jess, x), ∀t • K(Jess, t), ∀ wallaby • K(Jess, wallaby).
(ii) This proposition is similar to the previous one except that we are
now asserting the existence of someone who knows everyone rather
than stating that this ‘all knowing’ person is Jess. This means that
we simply need to take the previous proposition and replace ‘Jess’
with an existentially quantified variable.
In pseudo-English we have: there exists x such that for all y, x
knows y. Hence in symbols the proposition is: ∃x ∀y • K(x, y).
Note that this example is much easier to symbolise once we have
already attempted part (i). This provides a useful tool in symbolis-
ing propositions involving two quantifiers: first replace one quanti-
fier with a named object or individual and symbolise the resulting
proposition that now only has a single quantifier. Secondly, replace
the named individual with an appropriately quantified variable.
(iii) This is similar to the first proposition except that the named indi-
vidual is the person ‘known’ rather than the person ‘knowing’. We
may symbolise as ∀x • K(x, Ahmed).
(iv) Following on from example (iii), we simply need to replace ‘Ahmed’
with an existentially quantified variable ∃y. Thus we may symbolise
the proposition as ∀x ∃y•K(x, y). In the original sentence, ‘everyone
knows someone’, the universal quantification precedes the existen-
tial quantification and this is mirrored in the symbolic version.
There is, however, an important point to make in relation to the
proposition in part (iii). In that case, the proposition asserts that
everyone knows the same specific individual, namely Ahmed. For
the proposition ‘everyone knows someone’ to be true, it is not neces-
sarily the case that everyone knows the same person. For example,
although I might know Ahmed, you may not, but you might know
Julie instead. So long as everyone in the universe has someone that
they know, the proposition is true.
(v) This proposition is an extension of that in part (iv). We simply
need to add that the person y, who is known by x, is taller than x.
Hence we may symbolise as:
∀x ∃y • K(x, y) ∧ T (y, x).
 61

(vi) In our half-way house language we may write this as: ‘for all x,
if Misha knows x, then x is taller than Misha’. Hence the fully
symbolised version is: ∀x • K(Misha, x) ⇒ T (x, Misha).
(vii) Firstly, we may write in literal form as ‘for all x, if x is younger
than Sam, then Sam is taller than x’. Therefore we may symbolise
the proposition as: ∀x • Y (x, Sam) ⇒ T (Sam, x).
(viii) Comparing this with the previous proposition, we just need to re-
place ‘Sam’ with a universally quantified variable. Hence the propo-
sition is: ∀y ∀x • Y (x, y) ⇒ T (y, x).

3. Let the universe of discourse be ‘people’ and consider the two-variable

propositional function
L(x, y): ‘x loves y’.
Write down English sentences representing each of the following possible
ways of doubly quantifying L(x, y).

(i) ∀x ∀y • L(x, y)
(ii) ∀x ∃y • L(x, y)
(iii) ∃y ∀x • L(x, y)
(iv) ∃x ∀y • L(x, y)
(v) ∀y ∃x • L(x, y)
(vi) ∃x ∃y • L(x, y)

Solution

(i) Everyone loves everyone.

(ii) In our half-way house language, this would be ‘for all x there exists
y such that x loves y’ or just ‘every x loves some y’. In natural-
sounding language we would say ‘everyone loves someone’.
(iii) Again, in the half-way house language, this is ‘there exists y such
that, for all x, x loves y’. In order to preserve the order in which the
variables are mentioned, we note that ‘x loves y’ can be replaced
with ‘y is loved by x’, so that the half-way house expression becomes
‘some y is loved by every x’. More naturally, this is ‘someone is loved
by everyone’.
(iv) This time the ‘unnatural’ expression is ‘there exists x such that,
for all y, x loves y or ‘some x loves all y’. More naturally, we might
say ‘someone loves everyone’.
(v) Replacing ‘x loves y’ with ‘y is loved by x’ as in part (iii), we have
‘for all y, there exists x such that y is loved by x’ or ‘every y is
loved by some x’. More naturally, ‘everyone is loved by someone’.
62

(vi) Someone loves somebody.

Note that (ii) and (iii) are logically different propositions and they differ
only in the ordering of the quantifiers. The same is true of are proposi-
tions (iv) and (v). This illustrates the fact we have noted previously —
when we have different quantifiers, the order in which they are written
is significant.

Negation of Quantified Propositional Functions

At first glance, the proposition ‘no athletes are physically fit’ might suggest the
negation of the proposition ‘all athletes are physically fit’. However, recall that
the negation of a proposition is false when the proposition is true and is true
when the proposition is false. Therefore ‘no athletes are physically fit’ is not
the negation of ‘all athletes are physically fit’ because, in the case where some
athletes are fit and some are not, both propositions are false. The proposition
‘all athletes are physically fit’ is false when there is at least one athlete who
is not physically fit. Hence the negation of ‘all athletes are physically fit’ is
‘some athletes are not physically fit’. If we define the universe to be ‘athletes’
and we let P (x) denote ‘x is physically fit’, then, in symbols, the negation of
∀x • P (x) is ∃x • ¬P (x),

¬∀x • P (x) ≡ ∃x • ¬P (x).

Similarly, the existentially quantified proposition ‘some athletes are physically

fit’ is false in the unlikely event that all athletes are not physically fit. In other
words, the negation of ‘some athletes are physically fit’ is the proposition ‘all
athletes are not physically fit’. In symbols,

¬∃x • P (x) ≡ ∀x • ¬P (x).

Note that, in each case, to negate the quantified proposition, the quantifier
changes (from ∀ to ∃ or vice versa) and the propositional function itself is
negated. Thus, each of the two logical equivalences above conform to the
general description (or ‘meta-rule’) that says,

to negate a quantified propositional function, change the

quantifier and negate the propositional function.

There is nothing special about the context — that is, athletes and their fitness
— of the previous examples. So the two logical equivalences above, or the
meta-rule, give the following general rule for negating quantified propositions.
 63

Rule for negating quantified propositions

¬∀x • P (x) ≡ ∃x • ¬P (x).

¬∃x • P (x) ≡ ∀x • ¬P (x).

This rule is particularly helpful for negating quantified statements where more
than one quantifier is involved. Consider, for example, the proposition ‘some-
one knows everyone’. In exercise 2.12.2 (ii), we symbolised this as

∃x ∀y • K(x, y)

where K(x, y) is ‘x knows y’ defined over the universe of people. We may

ask: what is the negation of this proposition? To answer this, we move the
negation through the quantifiers in two stages, applying the rule for negating
quantified propositions at each stage, as follows.

¬∃x ∀y • K(x, y) ≡ ∀x ¬∀y • K(x, y)

≡ ∀x ∃y • ¬K(x, y).

Turning this back into English, we can first write is as; ‘for all x there exists
y such that x does not know y’. In idiomatic form, we may write ‘everyone
has someone they don’t know ’.
On page 53, we introduced the following two general forms of quantified propo-
sitions.
All P s are Qs : ∀x • P (x) ⇒ Q(x)
Some P s are Qs : ∃x • P (x) ∧ Q(x)
We conclude this chapter by considering the negation of each of these propo-
sitions. Consider first, the negation of ‘all P s are Qs’: ¬∀x • P (x) ⇒ Q(x).
Applying the rule for negating quantified propositions, we have:

¬∀x • P (x) ⇒ Q(x) ≡ ∃x • ¬(P (x) ⇒ Q(x)).

To understand what this means, we need to consider the propositional func-

tion ¬(P (x) ⇒ Q(x)). In exercise 2.2.4 we established the following logical
equivalence (called the Law of Material Implication).

P ⇒ Q ≡ ¬P ∨ Q.

Therefore, using De Morgan’s law and the involution law, we have

¬(P ⇒ Q) ≡ ¬(¬P ∨ Q) ≡ P ∧ ¬Q.

64

Applying this to the quantified proposition above shows that

¬∀x • P (x) ⇒ Q(x) ≡ ∃x • P (x) ∧ ¬Q(x).

Now we may paraphrase ∃x • P (x) ∧ ¬Q(x) as ‘some P s are not Qs’. Hence
the negation of ‘all P s are Qs’ is ‘some P s are not Qs’.
Now consider the negation of ‘some P s are Qs’: ¬∃x • P (x) ∧ Q(x). Again
applying the rule for negating quantified propositions, we have:

¬∃x • P (x) ∧ Q(x) ≡ ∀x • ¬(P (x) ∧ Q(x)).

Applying De Morgan’s law and the law of material implication given in exer-
cise 2.2.4, we have

¬(P ∧ Q) ≡ ¬P ∨ ¬Q ≡ P ⇒ ¬Q.

Applying this to the quantified proposition shows that

¬∃x • P (x) ∧ Q(x) ≡ ∀x • P (x) ⇒ ¬Q(x).

Now we may paraphrase ∀x • P (x) ⇒ ¬Q(x) as ‘all P s are not Qs’. Hence the
negation of ‘some P s are Qs’ is ‘all P s are not Qs’ or ‘every P is not a Q’.

Exercises 2.3

1. Translate each of the following propositions into symbolic form using

one-place or two-place predicates. Define the predicates used and, where
appropriate, define a universe of discourse.

(i) All students like to party.

(ii) Some footballers are overpaid but are not talented.
(iii) Everyone who went to the auction bought something.
(iv) Someone shouted ‘Fire!’ and everyone panicked.
(v) All celebrities give interviews but not all of them participate in re-
ality TV.
(vi) Not all athletes can run fast.
(vii) No one likes people who are rude.
(viii) If all Graham’s friends came to the party, then some of them would
have to stand.
(ix) All of my friends believe in global warming but some of them drive
large cars.
(x) Everyone applauds someone who is courageous.
 65

(xi) No one who went to the races was cold but some of them lost money.
(xii) If all politicians are honest, then none of them would receive illegal
payments.

2. The following propositional functions are defined. The universe of dis-

course is ‘people’.
C(x) : x is clever
H(x) : x is honest
L(x) : x is likeable
S(x) : x is smiles a lot
D(x) : x is prone to depression
Write each of the following propositions in natural-sounding English
sentences.

(i) S(Julie) ∧ D(Julie)

(ii) (C(Pete) ∧ H(Pete)) ⇒ L(Pete)
(iii) ∀x • L(x) ⇒ S(x)
(iv) ∃x • C(x) ∧ ¬H(x)
(v) ∀x • (¬L(x) ∨ ¬S(x)) ⇒ D(x)
(vi) ¬∀x • L(x) ∧ H(x)
(vii) ¬∀x • D(x) ⇒ S(x)
(viii) ¬∃x • L(x) ∧ ¬H(x)

3. The following two-variable propositional functions are defined. The uni-

verse of discourse for the variable x is ‘students’ and the universe of
discourse for the variable y is ‘courses’.
T (x, y) : x takes y.
E(x, y) : x enjoys y.
P (x, y) : x passes y.
Symbolise each of the following propositions.

(i) Carl passes every course he takes.

(ii) Every student who takes Statistics enjoys it.
(iii) Some students who take Statistics do not pass.
(iv) There are students who take courses which they do not enjoy.
(v) Some students pass every course they take.
(vi) If Gemma passes Statistics, then any student who takes Statistics
passes it.
66

(vii) There are some courses that are passed by all students who take
them.
(viii) If all students take courses that they don’t enjoy, then no student
passes every course they take.

4. This question uses the same universes of discourse and propositional

functions as defined in question 3. For each of the following propositions
written symbolically:

(a) write the proposition as a natural-sounding English sentence;

(b) negate the symbolic form and apply the rule for negating quantified
propositions to move the negation through the quantifier(s);
(c) write the resulting negated proposition as a natural-sounding En-
glish sentence.

(i) ∀x • E(x, Logic)

(ii) ∃x • ¬P (x, Logic)
(iii) ∀y • T (Poppy, y) ⇒ E(Poppy, y)
(iv) ∃y • T (Poppy, y) ∧ ¬E(Poppy, y)
(v) ∀x ∃y • P (x, y)
(vi) ¬∃x ∀y • P (x, y)
(vii) ∃x ∀y • T (x, y) ⇒ P (x, y)
(viii) (∃x ∀y • ¬T (x, y)) ⇒ (¬∀x ∃y • P (x, y))

2.5 Logical reasoning

In chapter 1, we noted that the kind of reasoning that is deployed in math-

ematics is, or should be, both precise and logically rigorous. In section 2.3,
we introduced the notion of logical implication which captures what we mean
by ‘logically rigorous’. Recall that a collection of propositions P1 , P2 , . . . , Pn
logically implies another proposition Q, denoted {P1 , P2 , . . . , Pn }  Q, pro-
vided Q is true in all those cases where each of the propositions P1 , P2 , . . . , Pn
is true. We were able to verify the relation of logical implication using truth
tables. All of our examples were restricted to situations where there were at
most three simple propositions involved. There was an obvious reason for this:
a truth table for a compound proposition that involves n simple propositions
has 2n rows. So when there are more than three or four simple propositions
involved, evaluating a truth table becomes unwieldy and impractical.
 67

In this section we give an alternative way of establishing logical implica-

tion. We give a set of simple rules for deducing a ‘conclusion’ proposition
from ‘premise’ propositions such that the premise propositions logically imply
the conclusion proposition. To illustrate this, using truth tables we can show
P, P ⇒ Q  Q ∨ R. Using our ‘deduction rules’ we will be able to establish
this in two steps:

• one deduction rule allows us to deduce the conclusion Q from premises

P and P ⇒ Q;

• a second deduction rule allows us to deduce the conclusion Q ∨ R from

the premise Q.

We will summarise this deduction as a sequence of propositions where each

proposition is either a premise or can be deduced from earlier propositions in
the list using a deduction rule.

1. P premise
2. P ⇒Q premise
3. Q from 1,2 using first deduction rule above
4. Q∨R from 3 using second deduction rule above

A deduction rule will be written in two lines, with the premises of the rule on
the first line and the conclusion on the second. Thus the two rules used in the
deduction above are written as follows.

P, P ⇒ Q and P
Q P ∨Q

Note that the second of these is written in terms of P and Q rather than Q and
R, which were actually used in the deduction above. In applying a deduction
rule we may substitute any propositions for those in the rule provided, of
course, that the same proposition is substituted throughout for a particular
proposition that appears in the rule. For example, given propositions P ∧ Q
and (P ∧ Q) ⇒ (R ∨ S), we may apply the first deduction rule above to deduce
R ∨ S.

We now list a set of rules that will allow us to make the deductions that we
need. In each case the premise propositions logically imply the conclusion. As
an illustration, in example 2.8.1, we showed that {P ⇒ Q, Q ⇒ R}  (P ⇒
R), and this is the basis for the deduction rule called hypothetical syllogism.
This means that, as we apply the rules in making deductions, the conclusion
of a sequence of deductions will always be logically implied by the premises.
68

Deduction rules for propositional logic

Simplification P ∧Q
P
Addition P
P ∨Q
Conjunction P, Q
P ∧Q
Disjunctive syllogism P ∨ Q, ¬P
Q
Modus ponens P, P ⇒ Q
Q
Modus tollens P ⇒ Q, ¬Q
¬P
Hypothetical syllogism P ⇒ Q, Q ⇒ R
P ⇒R
Resolution (Res) P ∨ Q, ¬P ∨ R
Q∨R
Constructive dilemma P ⇒ Q, R ⇒ S, P ∨ R
Q∨S

In addition to these rules, we also allow the replacement of a proposition with

any logically equivalent proposition. For example, if we have ¬(P ∧ Q) in
a deduction, then we may add the proposition ¬P ∨ ¬Q since ¬(P ∧ Q) ≡
¬P ∨ ¬Q (De Morgan’s rules, page 39).
These deduction rules are not independent of one another, and they are not
a ‘minimal set’ of rules. Equally, there are other rules that we could have
included — see exercise 2.4.2. What is important about the rules is that
they are sufficient to be able to deduce any logical implication. The following
definition will help make this a little more precise.

Definition 2.3
Let P1 , P2 , . . . , Pn , Q be propositions. We say that Q is deducible from
 69

P1 , P2 , . . . , Pn , written P1 , P2 , . . . , Pn ⊢ Q or {P1 , P2 , . . . , Pn } ⊢ Q, if there

is a list of propositions, the last of which is Q, such that each proposition
is either one of the P1 , P2 , . . . , Pn or is obtained from earlier propositions in
the list by applying one of the deduction rules or is logically equivalent to an
earlier proposition in the list.
The list of propositions, together with justifications for each proposition be-
longing to the list, is called a deduction of Q from P1 , P2 , . . . , Pn . We refer
to the propositions P1 , P2 , . . . , Pn as the premises of the deduction and the
proposition Q as the conclusion of the deduction.

The fact that, in each of our deduction rules the premises logically imply
the conclusion, means that we will only be able to derive propositions that
are logically implied by the premises. In other words, if Q is deducible from
P1 , P2 , . . . , Pn , then P1 , P2 , . . . , Pn logically imply Q:
if {P1 , P2 , . . . , Pn } ⊢ Q then {P1 , P2 , . . . , Pn }  Q.
Logicians refer to this property as soundness and we can say that our deduc-
tion rules are sound. In addition, our rules are sufficient to be able to deduce
any logical implication. In other words, if P1 , P2 , . . . , Pn logically imply Q,
then Q is deducible from P1 , P2 , . . . , Pn using our deductions rules:
if {P1 , P2 , . . . , Pn }  Q then {P1 , P2 , . . . , Pn } ⊢ Q.
Logicians refer to this property as completeness, and we can say that our
deduction rules are complete. Completeness is, in fact, rather rare in logic
and it is only for simple systems such as propositional logic that we can find
a complete set of deduction rules.

Examples 2.13
1. Show that {(P ⇒ Q) ∧ (R ⇒ S), R}  Q ∨ S.
Solution
By the soundness property, it is sufficient to show that Q ∨ S can be
deduced from the propositions P ⇒ Q, R ⇒ S, and R; symbolically,
{(P ⇒ Q) ∧ (R ⇒ S), R} ⊢ Q ∨ S.
Here is one way of completing the deduction.

1. (P ⇒ Q) ∧ (R ⇒ S) premise
2. R premise
3. (R ⇒ S) ∧ (P ⇒ Q) 1. Equivalence: commutative law
4. R ⇒ S 3. Simplification
5. S 2, 4. Modus ponens
6. S ∨ Q 5. Addition
7. Q ∨ S 6. Equivalence: commutative law
70

Note that the inclusion of each proposition in the deduction has been
justified either by noting that it is a premise or by explaining how it
follows from previous propositions using the deduction rules. In the case
where we introduce a proposition that is logically equivalent to an ex-
isting proposition in the deduction, we indicate this by ‘equivalence’
followed by a reason for the logical equivalence.
In each case in this deduction, the logical equivalence was the commu-
tative law (for ∧ or ∨) — see page 39. The reason we need this is that
our deduction laws are quite precise. For example, from P ∧ Q the Sim-
plification deduction rule allows us to deduce P but not Q so, if we wish
to deduce Q, we first need to deduce the logically equivalent proposi-
tion Q ∧ P . An alternative approach would have been to introduce two
versions of the simplification rule, one where we may deduce P and one
where we may deduce Q.
Is is frequently the case that there are several different ways of deducing
a conclusion from a collection of premises. The following is an alterna-
tive deduction to the one given above. In this case we have given the
abbreviated form of the rules used at each step.

1. (P ⇒ Q) ∧ (R ⇒ S) premise
2. R premise
3. (R ⇒ S) ∧ (P ⇒ Q) 1. Equivalence: commutative law
4. P ⇒ Q 1. Simplification
5. R ⇒ S 3. Simplification
6. R ∨ P 2. Addition
7. P ∨ R 3. Equivalence: commutative law
8. Q ∨ S 4, 5, 7. Constructive Dilemma

2. Show that {(P ∨ Q) ∧ (Q ∨ R), ¬Q}  P ∧ R.

Solution
As in the previous example, we give a deduction of P ∧ R from the two
premises (P ∨ Q) ∧ (Q ∨ R) and ¬Q.

1. (P ∨ Q) ∧ (Q ∨ R) premise
2. ¬Q premise
3. P ∨ Q 1. Simplification
4. Q ∨ P 3. Equivalence: commutative law
5. P 2, 4. Disjunctive syllogism
6. (Q ∨ R) ∧ (P ∨ Q) 1. Equivalence: commutative law
7. Q ∨ R 6. Simplification
8. R 2, 7. Disjunctive syllogism
9. P ∧ R 5, 8. Conjunction
 71

Arguments in Propositional Logic

In section 1.4, we considered briefly the nature of logical arguments. We are

now in a position to be more precise about what constitutes a logical argument.
An argument in propositional logic comprises a collection of propositions
P1 , P2 , . . . , Pn , called premises, and a proposition Q called the conclusion.
An argument with premises P1 , P2 , . . . , Pn and conclusion Q is valid if the
premises logically imply the conclusion

{P1 , P2 , . . . , Pn }  Q.

A formal deduction, or formal proof, of the validity of an argument is a

deduction that shows
{P1 , P2 , . . . , Pn } ⊢ Q.

Examples 2.14
1. Consider the validity of the following argument.

If Mark is correct, then unemployment will rise, and if Ann is correct

then, there will be a hard winter. Ann is correct. Therefore unemploy-
ment will rise or there will be a hard winter.

Solution
Before we can test the validity of the argument, we need to understand
the structure of the argument. We first symbolise the propositions in-
volved.
Let M: Mark is correct
U: Unemployment will rise
A: Ann is correct
H: There will be a hard winter.
There are two premises of the argument: (M ⇒ U ) ∧ (A ⇒ H) and
A.4 The conclusion is signalled by the word ‘therefore’; in symbols, the
conclusion is U ∨ H.
To show that the argument is correct, we need to establish that

{(M ⇒ U ) ∧ (A ⇒ H), A}  U ∨ H.

In example 2.13.1, we established that

{(P ⇒ Q) ∧ (R ⇒ S), R}  Q ∨ S
4 It may be worth noting that, had the first sentence been split in two, If Mark is correct,

then unemployment will rise. If Ann is correct, then there will be a hard winter, then we
would regard the argument as having three premises M ⇒ U , A ⇒ H, and A. This does
not affect the validity or otherwise of the argument because from P ∧ Q we may deduce
both P and Q and, conversely, from both P and Q we may deduce P ∧ Q.
72

by giving a deduction of Q ∨ S from (P ⇒ Q) ∧ (R ⇒ S) and R. This is

the same as the deduction that we wish to establish except that M , U ,
A, and H have been relabelled P , Q, R, and S, respectively. Clearly, the
labelling of the propositions does not matter so the deduction given in
example 2.13.1 can also be used to show that {(M ⇒ U )∧(A ⇒ H), A} 
U ∨ H, so the argument is valid. For future reference, translating the
deduction given in example 2.13.1 gives the following.

1. (M ⇒ U ) ∧ (A ⇒ H) premise
2. A premise
3. (A ⇒ H) ∧ (M ⇒ U ) 1. Equivalence: commutative law
4. A ⇒ H 3. Simplification
5. H 2, 4. Modus ponens
6. H ∨ U 5. Addition
7. U ∨ H 6. Equivalence: commutative law

2. Provide a formal deduction of the validity of the following argument.

If Sam had taken my advice or he’d had his wits about him, then he
would have sold his house and moved to the country. If Sam had sold
his house, Jenny would have bought it. Jenny did not buy Sam’s house.
Therefore Sam did not take my advice.

Solution
We first symbolise the propositions as follows.
Let A: Sam took my advice
W: Sam had his wits about him
H: Sam sold his house
C: Sam moved to the country
J: Jenny bought Sam’s house.
Then the argument has premises: (A ∨ W ) ⇒ (H ∧ C), H ⇒ J and ¬J.
The conclusion is ¬A.
When attempting to discover a deduction, it often helps both to work
forwards from the premises and also to work backwards from the con-
clusion. In other words, we may ask ‘in order to be able to deduce the
conclusion, what do we first need to deduce?’
In this example, we may deduce ¬A if we can first deduce ¬A ∧ ¬W ≡
¬(A ∨ W ). Now we may deduce ¬(A ∨ W ) from the first premise using
modus tollens, provided that we can deduce ¬(H ∧ C) ≡ ¬H ∨ ¬C.
The full deduction is the following where the first few steps establish
¬H ∨ ¬C.
 73

1. (A ∨ W ) ⇒ (H ∧ C) premise
2. H ⇒ J premise
3. ¬J premise
4. ¬H 2, 3. Modus tollens
5. ¬H ∨ ¬C 4. Addition
6. ¬(H ∧ C) 5. Equivalence: De Morgan’s law
7. ¬(A ∨ W ) 1, 6. Modus tollens
8. ¬A ∧ ¬W 7. Equivalence: De Morgan’s law
9. ¬A 8. Simplification

3. Consider the validity of the following argument.

If Mark is correct, then unemployment will rise and if Ann is correct,

then there will be a hard winter. Anne is correct. Therefore, unemploy-
ment will rise and there will be a hard winter.

Solution
This is almost the same argument as the one given in Example 1 above.
The only difference is that the conclusion has been strengthened to be
a conjunction. In other words, this argument concludes that unemploy-
ment will rise and there will be a hard winter, whereas the argument in
example 1 claimed that only one of these events would occur.
With the symbolism used in example 1, the premises are (M ⇒ U ) ∧
(A ⇒ H) and A (as before) and the conclusion is U ∧ H. If we look
more closely at the derivation in example 1, we can see that the reason
why we were able to conclude U ∨ H was that we earlier deduced H.
However this will not allow us to conclude U ∧ H.
At this point we may start thinking that the argument is not valid; in
other words, that the premises do not logically imply the conclusion. In
general, to show that P does not logically imply Q, P ̸ Q, we need to
show that Q is not true in all cases where P is true. This means we need
to find a situation where P is true and Q is false.
Returning to the argument, to show that it is not valid, we need to
find a situation where all of the premises are true but the conclusion
is false. Now A must be true since it is a premise and H must be true
since the deduction in Example 1 shows that H may be deduced from
the premises. In order to make the conclusion false, we need to suppose
that U is false. Consider the premise (M ⇒ U ) ∧ (A ⇒ H). The second
conjunct A ⇒ H is true since both A and H are true. If U is false, the
first conjunct will be true only if M is also false.
So, suppose that:
74

M is false: Mark is not correct

U is false: unemployment will not rise
A is true: Ann is correct
H is true: there will be a hard winter.
Then both premises are true but the conclusion is false. Therefore the
argument is not valid.
4. Consider the validity of the following argument.

If Rory didn’t get the job, then he didn’t become a lawyer. Rory didn’t
get the job and he took up golf. Rory became a lawyer or he didn’t take
up golf. Therefore Rory won the Open Championship.

Solution
We first symbolise the propositions as follows.
Let J: Rory got the job
L: Rory become a lawyer
G: Rory took up golf
C: Rory won the Open Championship.
Then the argument has premises: ¬J ⇒ ¬L, ¬J ∧ G, and L ∨ ¬G. The
conclusion is C.
At this stage we may think that there is something strange about the ar-
gument as the conclusion C does not feature at all in the premises. Thus,
were we able to provide a deduction of the argument, we would also be
able to provide a deduction for any argument with these premises and
any conclusion whatsoever. In fact, it is possible to provide a deduction
as follows.
1. ¬J ⇒ ¬L premise
2. ¬J ∧ G premise
3. L ∨ ¬G premise
4. ¬J 2. Simplification
5. ¬L 1, 4. Modus ponens
6. ¬G 3, 5. Disjunctive syllogism
7. G ∧ ¬J 2. Equivalence: commutative law
8. G 7. Simplification
9. G∨C 8. Addition
10. C 6, 9. Disjunctive syllogism
This deduction gives further grounds for concern since we have deduced
both G (line 8) and its negation ¬G (line 6). The reason for this is that
the premises are inconsistent in the sense that it is not possible for them
all to be true simultaneously. This could be verified by drawing up a
truth table for the premises and showing that there is no combination
 75

of truth values of J, L, and G for which the premises are all true — see
exercise 2.4.5.
Whenever we have a situation like this, we may deduce any proposition
in the manner shown using addition (line 9 above) followed by disjunc-
tive syllogism (line 10 above). Beginning with inconsistent premises, it
is possible to deduce any proposition from them.

Arguments in Predicate Logic

It is relatively easy to extend the idea of a formal deduction to arguments

in predicate logic. An argument in predicate logic will still have premises
P1 , P2 , . . . , Pn and conclusion Q, but now some or all of these may be quanti-
fied propositional functions. In addition to the deduction rule in propositional
logic, we also need rules that allow us to eliminate or introduce quantifiers.
For example, suppose that ∀x • P (x) is a premise of an argument. Then we
may deduce the proposition P (a) for any object a in the appropriate universe.
We call this the ‘quantifier elimination’ rule for the universal quantifier ∀. The
‘quantifier introduction’ rule for the universal quantifier is a little more subtle.
This is because we will usually not be able to deduce P (a) for each and every
element a of the universe. The quantifier introduction rule says that, if we
have established P (a) for an arbitrary element of the universe, then we may
deduce ∀x • P (x). By ‘arbitrary’ we just mean that a has no special attributes
other than being a member of the particular universe.
The corresponding rules for elimination and introduction of the existential
quantifier are similar; the essential difference is that the object a of the uni-
verse is a specific one rather than an arbitrary element.

Deduction rules for predicate logic

Universal quantifier elimination (∀-elimination)

From ∀x • P (x), we may deduce P (a) for any a in the universe.
Universal quantifier introduction (∀-introduction)
From P (a) where a is an arbitrary element of the universe, we
may deduce ∀x • P (x).
Existential quantifier elimination (∃-elimination)
From ∃x • P (x), we may deduce P (a) for some particular a in the
universe.
Existential quantifier introduction (∃-introduction)
From P (a) where a is a particular element of the universe, we may
deduce ∃x • P (x).
76

With these rules, in addition to the deduction rules for propositional logic, we
can give formal deductions of arguments in predicate logic. The basic strategy
is to apply quantifier elimination rules at the beginning of the deduction to ob-
tain non-quantified propositions; then apply appropriate deduction rules from
propositional logic; finally, apply appropriate quantifier introduction rules,
where necessary. In the last step, we will need to be careful to note whether
we have deduced Q(a), say, for an arbitrary or a particular element a of the
universe. In the former case, we may deduce ∀x • Q(x) and in the latter case
we may deduce ∃x • Q(x).

Examples 2.15

1. Give a formal deduction of the validity of the following argument.

All athletes are physically fit. Sarah is an athlete. Therefore Sarah is
physically fit.

Solution
We first define the universe to be ‘people’ and symbolise the predicates
as follows.
Let A: . . . is an athlete
F : . . . is physically fit.
Then the argument has premises: ∀x • A(x) ⇒ F (x) and A(Sarah). The
conclusion is F (Sarah).

1. ∀x • A(x) ⇒ F (x) premise

2. A(Sarah) premise
3. A(Sarah) ⇒ F (Sarah) 1. ∀-elimination
4. F (Sarah) 2, 3. Modus ponens

Note that, in this case the conclusion is a non-quantified proposition, so

we do not need a quantifier introduction step at the end.

2. Give a formal deduction of the validity of the following argument.

All elephants are mammals. Some elephants are playful. Therefore some
mammals are playful.

Solution
We first define the universe to be ‘living beings’ and symbolise the pred-
icates as follows.
Let E: . . . is an elephant
M : . . . is a mammal
P : . . . is playful.
 77

Then the argument has premises: ∀x • E(x) ⇒ M (x), ∃x • E(x) ∧ P (x).

The conclusion is ∃x • M (x) ∧ P (x).
At the beginning of the argument after listing the premises, we may
apply quantifier elimination to each premise. However, we need to
be careful. If we apply ∀-elimination to the first premise, we deduce
E(a) ⇒ M (a) for an arbitrary element of the universe. However we
cannot assume that this element a is also one for which E(a) ∧ P (a) is
true. Applying ∃-elimination to the premise ∃x • E(x) ∧ P (x) allows us
to deduce E(b) ∧ P (b) for some particular element b of the universe, but
we don’t know that it is the element a in the proposition E(a) ⇒ M (a).
The way around this is to apply ∃-elimination first to deduce E(a) ∧
P (a) for some particular element a of the universe. Then applying ∀-
elimination to the first premise, we can infer E(a) ⇒ M (a) for this a.
However we need to be aware that a is not arbitrary as it was ‘obtained’
from an ∃-elimination step; therefore, at the end of the argument, we
may only apply ∃-introduction and not ∀-introduction.
The full deduction is the following.

1. ∀x • E(x) ⇒ M (x) premise

2. ∃x • E(x) ∧ P (x) premise
3. E(a) ∧ P (a) 2. ∃-elimination
4. E(a) ⇒ M (a) 1. ∀-elimination
5. E(a) 3. Simplification
6. M (a) 4, 5. Modus ponens
7. P (a) ∧ E(a) 3. Equivalence: commutative law
8. P (a) 7. Simplification
9. M (a) ∧ P (a) 6, 8. Conjunction
10. ∃x • M (x) ∧ P (x) 9. ∃-introduction

3. Give a formal deduction of the validity of the following argument.

Everyone supports a football team or plays hockey. Everyone enjoys ex-
ercise or they don’t play hockey. Therefore everyone who doesn’t support
a football team enjoys exercise.

Solution
We first define the universe to be ‘people’ and symbolise the predicates
as follows.
Let F : . . . supports a football team
H: . . . plays hockey
E: . . . enjoys exercise
Then the argument has premises: ∀x • F (x) ∨ H(x), ∀x • E(x) ∨ ¬H(x).
The conclusion is ∀x • ¬F (x) ⇒ E(x).
78

In this case, when applying ∀-elimination to the premises we obtain non-

quantified propositions with an arbitrary element of the universe. Hence,
at the end of the deduction, we are able to apply ∀-introduction. The
full deduction is the following.

1. ∀x • F (x) ∨ H(x) premise

2. ∀x • E(x) ∨ ¬H(x) premise
3. F (a) ∨ H(a) 1. ∀-elimination
4. E(a) ∨ ¬H(a) 2. ∀-elimination
5. H(a) ∨ F (a) 3. Equivalence: commutative law
6. ¬H(a) ∨ E(a) 4. Equivalence: commutative law
7. F (a) ∨ E(a) 5, 6. Resolution
8. ¬(¬F (a)) ∨ E(a) 7. Equivalence: involution law
9. ¬F (a) ⇒ E(a) 8. Equivalence: material implication law
10. ∀x • ¬F (x) ⇒ E(x) 9. ∀-introduction

Exercises 2.4

1. Find a deduction from the premises to the conclusion in each of the

following cases.

(i) {P ⇒ Q, P ∧ R} ⊢ Q
(ii) {P ⇒ ¬Q, Q ∨ R} ⊢ P ⇒ R
(iii) {P ⇔ Q, ¬(P ∧ Q)} ⊢ ¬P ∧ ¬Q
(iv) (P ∧ Q) ⇒ (S ∧ T ), Q ∧ P } ⊢ S
(v) {(Q ∨ S) ⇒ R, Q ∨ P, ¬R} ⊢ P
(vi) {Q, ¬S, (P ∧ R) ⇒ S} ⊢ Q ∨ ¬R
(vii) {(P ∨ Q) ⇒ (R ∧ S), P } ⊢ R
(viii) {P ∨ Q, R ⇒ ¬Q, ¬P, (¬R ∧ Q) ⇒ S} ⊢ S
(ix) {P ⇒ ¬Q, Q ∨ (R ∧ S)} ⊢ P ⇒ (R ∧ S)
(x) {(P ⇒ Q) ∧ (R ⇒ Q), S ⇒ (P ∨ R), S} ⊢ Q

2. Show that each of the following deductions can be made from the deduc-
tion rules given on page 67. In each case the premises of the deduction
are the propositions above the line and the conclusion of the deduction
is the proposition below the line.
Note that this means that each of these rules could also be considered
a deduction rule and could therefore be added to our list of rules.
 79

(i) Rule of biconditional introduction

P ⇒ Q, Q ⇒ P
P ⇔Q
(ii) Rule of biconditional elimination
P ⇔Q
P ⇒Q
(iii) Rule of case analysis 1
P ∨ Q, P ⇒ R, Q ⇒ R
R
(iv) Rule of case analysis 2
P ⇒ R, Q ⇒ R
(P ∨ Q) ⇒ R

3. Give formal deductions of the validity of each of the following arguments.

(i) If Mary drinks wine or eats cheese, then she gets a headache. She
is drinking wine and eating chocolate. Therefore Mary will get a
headache.
(ii) If you get a degree or you get a good job, then you will be successful
and happy. You get a good job. Therefore you will be happy.
(iii) If the battery is flat or the car is out of petrol, then it won’t start
and I’ll be late for work. The car is out of petrol or the battery is
flat. Therefore I’ll be late for work.
(iv) Either the project wasn’t a success or Sally did not invest her inher-
itance. If she were sensible, then Sally would invest her inheritance.
The project was a success. If Sally wasn’t sensible and didn’t invest
her inheritance, then she is broke. Therefore Sally is broke.
(v) The murder was committed either by A or by both B and C. If
A committed the murder, then the victim was poisoned. Therefore
either C committed the murder or the victim was poisoned.
(vi) Peter is either brave or brainy and he is either brainy or bald. Peter
is not brainy. Therefore he is brave and bald.
(vii) If it is useful, then I’ll keep it, and if it is valuable, then I’ll keep
it. If it belonged to Ben, then it is useful or valuable. It belonged
to Ben. So I’ll keep it.
(viii) If it doesn’t rain, then I’ll go shopping. If I go shopping, then if I
don’t take an umbrella, it will rain. If I go by car, then I won’t take
an umbrella. So it will rain or I won’t go by car.
(ix) If ghosts are a reality, then there are spirits roaming the Earth, and
if ghosts are not a reality, then we do not fear the dark. Either we
80

fear the dark or we have no imagination. We do have an imagination

and ghosts are a reality. Therefore there are spirits roaming the
Earth.
(x) If Tim committed the crime, then he’ll flee the country and we’ll
never see him again. If we see Tim again, then he is not Tom’s
friend. Hence, if Tim committed the crime or he’s Tom’s friend,
then we’ll never see him again.

4. Show that each of the following arguments are not valid.

(i) If Mary drinks wine and eats cheese, then she gets a headache.
She is drinking wine or eating chocolate. Therefore Mary will get
a headache.
(ii) Either the project wasn’t a success or Sally invested her inheritance.
If she were sensible, then Sally would invest her inheritance. The
project was a success. If Sally wasn’t sensible and didn’t invest her
inheritance, then she is broke. Therefore Sally is broke.
(iii) If it doesn’t rain, then I’ll go shopping. If I go shopping, then if I
don’t take an umbrella, it will rain. If I go by car, then I won’t take
an umbrella. So it will rain and I won’t go by car.
(iv) If ghosts are a reality, then there are spirits roaming the Earth or if
ghosts are not a reality, then we do not fear the dark. Either we fear
the dark or we have no imagination. We do have an imagination or
ghosts are a reality. Therefore there are spirits roaming the Earth.
(v) If Tim committed the crime, then he’ll flee the country and we’ll
never see him again. If we see Tim again, then he is not Tom’s
friend. Hence if he’s Tom’s friend, then we’ll never see Tim again.

5. Draw up a truth table, with simple propositions labelled J, L, and G,

giving the truth values of ¬J ⇒ ¬L, ¬J ∧ G, and L ∨ ¬G.
Hence show that the premises in example 2.14.4 are inconsistent.

6. Provide a formal deduction of the validity of each of the following argu-

ments in predicate logic.

(i) Some people are good-looking and rich. Everyone who is rich is
dishonest. Therefore there are people who are good-looking and
dishonest.
(ii) Some people are good-looking and rich. Everyone who is rich is
dishonest. Therefore not everyone who is good-looking is honest.
(iii) All even numbers are rational and divisible by two. Some even
numbers are divisible by four. Hence some numbers are divisible
by two and four.
 81

(iv) All numbers which are integers are even or odd. All numbers
which are integers are even or non-zero. Some numbers are inte-
gers. Therefore there are numbers that are either even or they are
odd and non-zero.
(v) All animals with feathers are not aquatic. There are aquatic animals
that live in the sea. So there are animals that live in the sea and
don’t have feathers.
(vi) Some functions are continuous and differentiable. All functions
which are continuous are defined for all values of x. Therefore some
functions which are defined for all values of x are differentiable.
(vii) Everything which is enjoyable and cheap is harmful to one’s health.
All holidays are enjoyable. There are holidays which are not harmful
to one’s health. Therefore some things are not cheap.
(viii) There are no polynomials which are not differentiable functions. All
differentiable functions are continuous. Therefore all polynomials
are continuous.
Chapter 3
Sets and Functions

3.1 Introduction

In the previous chapter, we discussed the elements of logic which we may re-
gard as providing a framework for constructing and understanding mathemat-
ical statements. Much of the language of mathematics, however, is described
in terms of sets and functions, which we introduce in this chapter. If logic
provides the ‘backbone’ of mathematics, then sets and functions put flesh on
the bone.
We imagine that many of our readers will have met sets and functions previ-
ously, but possibly will not have covered all the material that we will introduce
here or maybe not as formally as our approach. The material of this chapter
will be assumed as background knowledge in the rest of the book. Some read-
ers who have met sets and functions may choose to ignore this chapter on first
reading, referring back to it as and when necessary.
In the next three sections, we first introduce sets, how they may be defined
and their properties. We then look at operations on sets and building new sets
from old. Finally, we introduce an important construction called the Cartesian
product. In the last two sections of the chapter, we consider functions and some
of their properties.

3.2 Sets and membership

The notion of a ‘set’ is one of the basic concepts of mathematics — some

would say the basic concept. We will not give a precise, formal definition of
a set but simply describe a set as a well-defined collection of objects called
elements. The elements contained in a given set need not have anything in
common (other than the obvious common attribute that they all belong to the
given set). Equally, there is no restriction on the number of elements allowed

83
84

in a set; there may be an infinite number, a finite number, or even no elements

at all.
There is, however, one restriction we insist upon: given a set and an object,
we should be able to decide (in principle at least — it may be difficult in
practice) whether or not the object belongs to the set. This is what we meant
above by a ‘well-defined’ collection: given an object a and a set A, it must be
unambiguous whether or not a belongs to A. Thus, ‘the set of all tall people’
is not a set, whereas ‘the set of all people taller than 2 m’ is a set.
Clearly, a concept as general as this has many familiar examples as well as
many frivolous ones. For example, a set could be defined to contain the Mona
Lisa, London Bridge, and the number e (although what ‘use’ this set has is
hard to envisage). This is a finite set because it contains a finite number
of elements. The set containing all the positive, even integers is clearly well-
defined. This is an infinite set as it contains infinitely many elements.

Notation
We shall generally use uppercase letters to denote sets and lowercase letters
to denote elements. It is not always possible to adhere to this convention
however; for example, when the elements of a particular set are themselves
sets. The symbol ∈ denotes ‘belongs to’ or ‘is an element of’. Thus

a ∈ A means (the element) a belongs to (the set) A

and
a∈
/ A means ¬(a ∈ A) or a does not belong to A.

Defining Sets
Sets can be defined in various ways. The simplest is by listing the elements
enclosed between curly brackets or ‘braces’ { }.

Examples 3.1
1. A = {Mona Lisa, London Bridge, e}.
This is the rather odd set containing three elements described above.
2. B = {2, 4, 6, 8, . . .}
This is the infinite set described above. We clearly cannot list all the
elements. Instead we list enough elements to establish a pattern and use
‘. . . ’ to indicate that the list continues indefinitely.
3. C = {1, {1, 2}}
This set has two elements, the number 1 and the set {1, 2}. This illus-
trates that sets can themselves be elements of other sets.
 85

4. D = {1, 2, . . . , 100}
This set contains the first 100 positive integers. Again we use ‘. . . ’ to
indicate that there are elements in the list which we have omitted, al-
though in this case only finitely many are missing.
5. E = { }
This set contains no elements. It is called the empty set or null set.
The empty set is usually denoted ∅.

When listing elements, the order in which elements are listed is not important.
Thus, for example {1, 2, 3, 4} and {4, 3, 2, 1} define the same set. Also, any
repeats in the listing are ignored so that {1, 2, 2, 3, 3, 3, 4, 4, 4, 4} also describes
the same set. To be precise, two sets are equal if and only if they contain the
same elements; that is, A = B if ∀x • x ∈ A ⇔ x ∈ B is a true proposition,
and conversely.
Listing the elements of a set is impractical except for small sets or sets where
there is a pattern to the elements such as B and D in examples 3.1 above. An
alternative is to define the elements of a set by a property or predicate (see
section 2.4). More precisely, if P (x) is a single-variable propositional function,
we can form the set whose elements are all those objects a (and only those)
for which P (a) is a true proposition. A set defined in this way is denoted
A = {x : P (x)}.
This is read: the set of all x such that P (x) (is true).
Note that ‘within A’ — that is, if we temporarily regard A as the universe of
discourse — the quantified propositional function ∀x•P (x) is a true statement.

Examples 3.2
1. The set B in example 3.1 above could be defined as
B = {n : n is an even, positive integer}, or
B = {n : n = 2m, where m > 0 and m is an integer},
or, with a slight change of notation,
B = {2m : m > 0 and m is an integer}.
Note that although the propositional functions used are different, the
same elements are generated in each case.
2. The set D in example 3.1 above could be defined as
D = {k : k is an integer and 1 ≤ k ≤ 100}.
3. The set {1, 2} could alternatively be defined as {x : x2 − 3x + 2 = 0}.
We sometimes refer to the set {1, 2} as the solution set of the equation
x2 − 3x + 2 = 0.
86

4. The empty set ∅ can be defined in this way using any propositional
function P (x) which is true for no objects x. Thus, for example,
∅ = {x : x ̸= x}.

Note that, if P (x) and Q(x) are propositional functions which are true for
precisely the same objects x, then the sets they define are equal, {x : P (x)} =
{x : Q(x)}. For example, the two solution sets {x : (x − 1)2 = 4} and {x :
(x + 1)(x − 3) = 0} are equal, since the two propositional functions P (x) :
(x − 1)2 = 4 and Q(x) : (x + 1)(x − 3) = 0 are true for precisely the same
values of x, namely −1 and 3.

Universal set
Sometimes we wish to ‘build’ our sets only from some larger ‘context set’
called the universal set, denoted U . The universal set is simply the set of
all objects being considered at the current time. The universal set plays much
the same role as the universe of discourse does in logic. In chapter 2, we noted
that we could define the universe of discourse to be what was convenient for
the particular context. So it is with sets. For example, if we are interested
in studying the properties of soccer players in a particular league, we might
define
U = {soccer players in the particular league}.
Then the set
A = {x : x has scored at least three goals}
denotes only those players in the particular league who have scored at least
three goals.
Sometimes the role of the universe is made explicit when defining sets using
predicates with the notation
{x ∈ U : P (x)},
which may be read as ‘the set of all x in U such that P (x)’.

Cardinality
If A is a finite set, its cardinality, |A|, is the number of (distinct) elements
which it contains. If A has an infinite number of elements, we say it has
infinite cardinality, and write |A| = ∞.1
Other notations commonly used for the cardinality of A are n(A), #(A), and
Ā¯.
1 There is a more sophisticated approach to cardinality of infinite sets which allows

different infinite sets to have different cardinality. Thus ‘different sizes’ of infinite sets can
be distinguished. In this theory the set of integers has different cardinality from the set of
numbers, for example. For further details, see Garnier and Taylor [6], for example.
 87

Examples 3.3
1. Let A = {0, 1, 2, 3, 4}. Then |A| = 5.
2. Let A = {1, 2, 2, 3, 3, 3, 4, 4, 4, 4}. Then |A| = 4. This is because we ignore
repeats when listing the elements of sets, so A = {1, 2, 3, 4}.
3. |∅| = 0 since the empty set ∅ contains no elements.
4. Generalising from Example 1, if A = {0, 1, . . . , n}, then |A| = n + 1.
5. Let A be the set of positive even integers, A = {2, 4, 6, 8, . . .}. Then
|A| = ∞.
6. Let A = {1, 2, {1, 2}}. Then |A| = 3 because A contains three elements:
the number 1, the number 2, and the set {1, 2}.
Determining the cardinality of sets, some of whose elements are also sets,
can be tricky. In this context, it may help to think of sets as some kind
of abstract paper bag. In this example, the set A is a paper bag that,
when we look inside, contains two numbers and another paper bag; that
is, three elements in total.
7. Let A = {{1, 2, 3, 4}}. Then |A| = 1 because A contains a single element,
namely the set {1, 2, 3, 4}.
Using the paper bag analogy, the set A is a paper bag that contains, as
its single element, another paper bag.
8. Let A = {{1, 2}, {3, 4}, {1, 2}}. In this case, we have |A| = 2.
Initially, we might think that |A| = 3 since A contains three sets (paper
bags). However two of these are the same set, {1, 2}, so we must ignore
the repeated element. Hence A = {{1, 2}, {3, 4}}, which clearly contains
two elements, so |A| = 2.

Subset
The set B is a subset of the set A, denoted B ⊆ A, if every element of B is
also an element of A. Symbolically, B ⊆ A if ∀x • x ∈ B ⇒ x ∈ A is true, and
conversely. If B is a subset of A, we say that A is a superset of B, and write
A ⊇ B.
Note that every set A is a subset of itself, A ⊆ A. Any other subset of A is
called a proper subset of A. The notation B ⊂ A is used to denote ‘B is a
proper subset of A’. Thus B ⊂ A if and only if B ⊆ A and B ̸= A.
It should also be noted that ∅ ⊆ A for every set A. This is because the
definition above is satisfied in a trivial way: the empty set has no elements,
so certainly each of them belongs to A. Alternatively, for any object x, the
proposition x ∈ ∅ is false, which means that the conditional (x ∈ ∅) ⇒ (x ∈
A) is true.
88

Examples 3.4

1. {2, 4, 6, . . .} ⊆ {1, 2, 3, . . .} ⊆ {0, 1, 2, . . .}. Of course, we could have used

the proper subset symbol ⊂ to link these three sets instead.

2. Similarly: {women} ⊆ {people} ⊆ {mammals} ⊆ {creatures};

{Catch 22} ⊆ {novels} ⊆ {works of fiction};
{Mona Lisa} ⊆ {paintings} ⊆ {works of art}; etc.
Again, in each of these we could have used ⊂ instead.

3. Let X = {1, 2, {3, 4}}. Then {1, 2} ⊆ X because each element in {1, 2}
is also an element of X. However {3, 4} is not a subset of X, which
we can denote by {3, 4} * X, because the elements of {3, 4} — 3 and
4 — are not themselves elements of X. Using the paper bag analogy
introduced in example 3.3.6, the set X is a paper bag containing two
numbers, 1 and 2, and the paper bag {3, 4}. Thus {3, 4} is an element
of X, {3, 4} ∈ X, but not a subset of X, {3, 4} ̸⊆ X. However, since
{3, 4} is an element of X, the set containing {3, 4} as its only element
is a subset of X, {{3, 4}} ⊆ X.
Care clearly needs to be taken to distinguish between set membership
and subset, particularly when a set has elements which are themselves
sets. In particular, note that a ∈ A is true if and only if {a} ⊆ A is true.

Recall that two sets are equal, A = B, if and only if they contain the same
elements: for all x, x ∈ A if and only if x ∈ B. In section 2.3 (page 41), we
introduced the following logical equivalence, which we called the Biconditional
Law:
P ⇔ Q ≡ (P ⇒ Q) ∧ (Q ⇒ P ).
Thus, the statements

x ∈ A ⇔ x ∈ B and (x ∈ A ⇒ x ∈ B) ∧ (x ∈ B ⇒ x ∈ A)

are true for precisely the same objects x. Saying that the first of these is true
for all x means A = B. Saying that the second of these is true for all x means
both A ⊆ B (from x ∈ A ⇒ x ∈ B) and B ⊆ A (from x ∈ B ⇒ x ∈ A).
Therefore A = B means the same thing as A ⊆ B and B ⊆ A. We summarise
this as a theorem.

Theorem 3.1
Two sets A and B are equal if and only if both A ⊆ B and B ⊆ A.

Examples 3.5

1. Show that {x : 2x2 + 5x − 3 = 0} ⊆ {x : 2x2 + 7x + 2 = 3/x}.

 89

Solution
Let A = {x : 2x2 + 5x − 3 = 0} and B = {x : 2x2 + 7x + 2 = 3/x}.
We need to show that every element of A is an element of B. The
} 2x + 5x − 3 = 0 has solutions x = 2 and x = −3, so A =
2 1
equation
{
2 , −3 .
1

When x = 21 , 2x2 + 7x + 2 = 1
2 + 7
2 + 2 = 6 = 3/x, so 1
2 ∈ B.
When x = −3, 2x2 + 7x + 2 = 18 − 21 + 2 = −1 = 3/x, so −3 ∈ B.
Therefore every element of A is an element of B, so A ⊆ B.
2. Let A = {{1}, {2}, {1, 2}} and let B be the set of all non-empty subsets
of {1, 2}. Show that A = B.
Solution
A ⊆ B since each of the three elements of A is a non-empty subset of
{1, 2} and therefore an element of B.
B ⊆ A since every non-empty subset of {1, 2} (i.e., every element of B)
is contained in A.
Using theorem 3.1, we conclude that A = B.
3. Suppose that A ⊆ B and let C be the set defined by

C = {x : x ∈ A ∨ x ∈ B}.

Show that C = B.
Solution
To show that C and B are equal, we will show that B ⊆ C and C ⊆ B.
Consider x ∈ B. Then x ∈ A ∨ x ∈ B is true, so x ∈ C. Thus every
element of B also belongs to C, so B ⊆ C.
Now consider x ∈ C. Then, by the definition of C, either x ∈ A or x ∈ B
(or both). However, if x ∈ A, then it follows that x ∈ B also, because
A ⊆ B. Therefore, in either case we can conclude x ∈ B. This shows
that every element of C also belongs to B, so C ⊆ B.
We have now shown B ⊆ C and C ⊆ B, so theorem 3.1 allows us to
conclude that B = C.

Special sets of numbers

The following is a list of some special sets of numbers that are frequently used
as universal sets.

N = {0, 1, 2, 3, . . .} the set of natural numbers.

Z = {. . . , −2, −1, 0, 1, 2, . . .} the set of integers.
90

Q = {p/q : p, q ∈ Z and q ̸= 0} the set of rational numbers.

R = the set of real numbers; real numbers can be thought of as
corresponding to points on a number line or as numbers
written as (possibly infinite) decimals.
C = {x+iy : x, y ∈ R and i2 = −1} the set of complex numbers.

Clearly the following subset relations hold amongst these sets:

N ⊆ Z ⊆ Q ⊆ R ⊆ C.

Also frequently used are Z+ , Q+ , and R+ , the sets of positive integers, rational
numbers, and real numbers, respectively. Note that N is not equal to Z+ since
0 belongs to the former but not the latter. In addition, we shall sometimes
use E and O to denote the sets of even and odd integers, respectively:

E = {2n : n ∈ Z} = {. . . , −4, −2, 0, 2, 4, . . .}

O = {2n + 1 : n ∈ Z} = {. . . , −3, −1, 1, 3, 5, . . .}.

Finally, we will need a notation for intervals. Let a and b be real numbers
such that a < b. The open interval (a, b) is (a, b) = {x ∈ R : a < x < b}.
The closed interval [a, b] is [a, b] = {x ∈ R : a ≤ x ≤ b}. The notation also
extends to ‘half-open’ intervals [a, b) = {x ∈ R : a ≤ x < b} and (a, b] = {x ∈
R : a < x ≤ b}.

Exercises 3.1

1. List the elements of each of the following sets, using the ‘. . . ’ notation
where necessary:

(i) {x : x is an integer and − 2 < x < 5}

(ii) {x : x is a positive (integer) multiple of three}
(iii) {x : x = y 2 and y is an integer}
(iv) {x : (3x − 1)(x + 2) = 0}
(v) {x : x ≥ 0 and (3x − 1)(x + 2) = 0}
(vi) {x : x is an integer and (3x − 1)(x + 2) = 0}
(vii) {x : x is a positive integer and (3x − 1)(x + 2) = 0}
(viii) {x : 2x is a positive integer}

2. Define each of the following sets using a predicate; that is, write each
set in the form {x : P (x)} for some suitable predicate P .
 91

(i) {3, 4, 5, 6, 7, 8, 9, 10}

(ii) {2, 4, 6, 8, . . . , 100}
(iii) The set of all odd integers.
(iv) {1, 4, 9, 16, 25, 36, 49, . . .}
(v) {1, 2, 4, 8, 16, 32, 64, . . .}
(vi) {2, 7, 12, 17, 22, 27, 32, 37, . . .}
(vii) {a, b, c, d, e, f, . . . , x, y, z}
(viii) The set of integers which can be written as the sum of the squares
of two integers.
(ix) The set of positive integers that are powers of 2.
(x) The set of positive integers that are powers of some prime number.

3. State whether each of the following statements is true or false.

(i) {1, 2} ∈ {1, 2, 3, 4}

(ii) {1, 2} ⊆ {1, 2, 3, 4}
(iii) {1, 2} ∈ {{1, 2}, {3, 4}}
(iv) {1, 2} ⊆ {{1, 2}, {3, 4}}
(v) 0 ∈ ∅
(vi) ∅ ∈ {{1, 2}, {3, 4}}
(vii) ∅ ⊆ {{1, 2}, {3, 4}}
(viii) ∅ ∈ {∅, {∅}}

4. Determine the cardinality of each of the following sets:

(i) {2, 4, 6, 8, 10}

(ii) {x : x is an integer and 2/3 < x < 17/3}
√
(iii) {x : x is an integer}
(iv) {x ∈ Z : x2 ≤ 2}
(v) {x ∈ R : x2 ≤ 2}
(vi) {2, 4, {6, 8}, 10}
(vii) {2, 4, {{6, 8}, 10}}
(viii) {{2}, {4}, {6}, {8}, {10}}
(ix) {1, {1}, {{1}}, {{{1}}}}
(x) {∅, {∅}, {{∅}}}.
92

5. In each of the following cases, state whether x ∈ A, x ⊆ A, both, or

neither.
(i) x = {1}; A = {1, 2, 3}
(ii) x = {1}; A = {{1}, {2}, {3}}
(iii) x = {1}; A = {1, 2, {1, 2}}
(iv) x = {1, 2}; A = {1, 2, {1, 2}}
(v) x = {1}; A = {{1, 2, 3}}
(vi) x = 1; A = {{1}, {2}, {3}}

6. Given that X = {a, b, c, d}, list the elements of each of the following
sets:

(i) {A : A ⊆ X and |A| = 3}

(ii) {A : A ⊆ X and |A| = 2}
(iii) {A : A is a proper subset of X}
(iv) {A : A ⊆ X and b ∈ A}.

7. Let U = {x : x is an integer and 1 ≤ x ≤ 12}. In each of the following

cases, determine whether A ⊆ B, B ⊆ A, both, or neither.

(i) A = {x : x is odd} B = {x : x is a multiple of 5}

(ii) A = {x : x is even} B = {x : x2 is even}
(iii) A = {x : x is even} B = {x : x is a power of 2}
(iv) A = {x : 3x + 1 > 10} B = {x : x2 > 20}
√
(v) A = {x : x ∈ Z} B = {x : x is a power of 2 or 3}
√
(vi) A = {x : x ≤ 3} B = {x : x is a perfect square}

3.3 Operations on sets

Venn-Euler diagrams 2 are a useful visual representation of sets. In these

diagrams, sets are represented as regions in the plane and elements which
belong to a given set are placed inside the region representing it. Frequently, all
the sets in the diagram are placed inside a box which represents the universal
2 These diagrams are more commonly called just ‘Venn diagrams’ after John Venn, the

nineteenth-century English mathematician. In fact, diagrams such as figure 3.2 are more
properly called ‘Euler diagrams’ after Leonhard Euler who first introduced them in 1761.
Although both Venn and Euler had precise rules for constructing their diagrams, today
the term ‘Venn diagram’ is used informally to denote any diagram that represents sets by
regions in the plane.
 93

set U . If an element belongs to more than one set in the diagram, the two
regions representing the sets concerned must overlap and the element is placed
in the overlapping region.

Example 3.6
Let U = {1, 2, 3, . . . , 12},
A = {n : n is even} = {2, 4, 6, 8, 10, 12},
B = {n : n is prime} = {2, 3, 5, 7, 11}.
Figure 3.1 is a Venn-Euler diagram representing these sets and their elements.

U
A 3 B
4
6 8 5
2
7
10
12 11
1 9

FIGURE 3.1: A Venn-Euler diagram.

Venn-Euler diagrams are very good at representing the relationships between

sets. If A ⊆ B, then the region representing A will be enclosed inside the
region representing B, so that every element in the region representing A is
also inside that representing B. This is illustrated in figure 3.2 where the set
A is represented by the shaded region.

B
U

FIGURE 3.2: Venn-Euler diagrams representing subset.

Let A and B be sets.

The intersection of A and B, denoted A ∩ B, is the set of all elements that
belong to both A and B:

A ∩ B = {x : x ∈ A and x ∈ B}.
94

A B
U

FIGURE 3.3: Venn-Euler diagram showing A ∩ B.

Figure 3.3 is a Venn-Euler diagram representing the intersection; the shaded

region of the diagram represents A ∩ B.
The union of A and B, denoted A ∪ B, is the set of all elements that belong
to A or to B or to both:

A ∪ B = {x : x ∈ A or x ∈ B}.

Figure 3.4 is a Venn-Euler diagram representing the union; the shaded region
of the diagram represents A ∪ B.

A B
U

FIGURE 3.4: Venn-Euler diagram showing A ∪ B.

There are obvious connections between intersection of sets and conjunction of

propositions, and between union of sets and (inclusive) disjunction of proposi-
tions. Suppose that A and B are sets defined by propositional functions P (x)
and Q(x), respectively:

A = {x : P (x)} and B = {x : Q(x)}.

Then

A ∩ B = {x : P (x) ∧ Q(x)} and A ∪ B = {x : P (x) ∨ Q(x)}.

The definitions of intersection and union extend to more than two sets. Let
A1 , A2 , . . . , An be sets.
 95

Their intersection is:

∩
n
Ar = A1 ∩ A2 ∩ · · · ∩ An
r=1
= {x : x ∈ A1 and x ∈ A2 and . . . and x ∈ An }
= {x : x belongs to each set Ar , for r = 1, 2, . . . , n}.

Their union is:

∪
n
Ar = A1 ∪ A2 ∪ · · · ∪ An
r=1
= {x : x ∈ A1 or x ∈ A2 or . . . or x ∈ An }
= {x : x belongs to at least one set Ar , r = 1, . . . , n}.

Two sets A and B are said to be disjoint if they have no elements in common;
that is, if their intersection is the empty set, A ∩ B = ∅. In a Venn-Euler
diagram, we represent disjoint sets by drawing the regions representing them
to be separated and non-overlapping. This is illustrated in figure 3.5.

A B
U

FIGURE 3.5: Venn-Euler diagram representing disjoint sets.

The difference of A and B, denoted A − B, is the set of all elements that

belong to A but do not belong to B:

A − B = {x : x ∈ A and x ̸∈ B}.

The difference A − B is sometimes denoted A \ B. Figure 3.6 is a Venn-

Euler diagram representing the difference; the shaded region of the diagram
represents A − B.
The complement of a set A, denoted Ā, is the set of all elements that do not
belong to A:
Ā = {x : x ̸∈ A} = {x ∈ U : x ̸∈ A}.
The complement of A is sometimes denoted A′ or Ac . Note that Ā = U − A.
Figure 3.7 is a Venn-Euler diagram representing the complement; the shaded
region of the diagram represents Ā.
96

A B
U

FIGURE 3.6: Venn-Euler diagram representing A − B.

U
A

FIGURE 3.7: Venn-Euler diagram representing Ā.

Examples 3.7

1. Let U = {1, 2, 3, . . . , 11, 12},

A = {n : n is a multiple of 3},
B = {n : n is a factor of 12}
and C = {n : n is a multiple of 4}.
List the elements of each of the following sets.
(i) A ∩ (B ∪ C)
(ii) (A ∩ B) ∪ (A ∩ C)
(iii) A − (B ∩ C)
(iv) (A − B) ∪ (A − C)
(v) A∪C
(vi) Ā ∩ C̄

Solution
It is helpful to have the elements of A, B, and C listed first:

A = {3, 6, 9, 12}
B = {1, 2, 3, 4, 6, 12}
C = {4, 8, 12}.

(i) B ∪ C comprises the elements that belong to B or to C or to both,

 97

so B ∪ C = {1, 2, 3, 4, 6, 8, 12}. The intersection of this set with A

comprises those elements common to both A and B ∪ C; this gives
A ∩ (B ∪ C) = {3, 6, 12}.
(ii) Firstly, we evaluate A ∩ B = {3, 6, 12} and A ∩ C = {12}. The
union of these two sets is (A ∩ B) ∪ (A ∩ C) = {3, 6, 12}.
(iii) Firstly, B∩C = {4, 12}. To form A−(B∩C), we take the elements of
A = {3, 6, 9, 12} and remove those that are also in B ∩ C = {4, 12}.
This gives A − (B ∩ C) = {3, 6, 9}.
(iv) First note that A − B contains those elements of A that are not
also in B; in fact, 9 is the only such element, so A − B = {9}.
Similarly, A − C contains those elements of A that are not also in
C, so A − C = {3, 6, 9}. Forming the union of these two sets gives
(A − B) ∪ (A − C) = {3, 6, 9}.
(v) First we find A ∪ C = {3, 4, 6, 8, 9, 12}. The complement of this
set contains all those elements of U that do not belong to A ∪ C.
Hence A ∪ C = {1, 2, 5, 7, 10, 11}.
(vi) Since Ā = {1, 2, 4, 5, 7, 8, 10, 11} and C̄ = {1, 2, 3, 5, 6, 7, 9, 10, 11},
their intersection is Ā ∩ C̄ = {1, 2, 5, 7, 10, 11}.

Note that, in this example, the sets are equal in pairs. For these sets,
we have

A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C),
A − (B ∩ C) = (A − B) ∪ (A − C),
and A ∪ C = Ā ∩ C̄
It is natural to ask whether these are properties of these particular sets
or whether they are true for all sets. We will explore this for one of the
pairs of sets in the next example.

2. The equality A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C) does, in fact, hold for all

sets, and we can use Venn-Euler diagrams to illustrate this. In figure 3.8,
we have identified the regions in a Venn-Euler diagram representing the
two sets A ∩ (B ∪ C) and (A ∩ B) ∪ (A ∩ C).
In the left-hand diagram, we have separately shaded the sets A and
B ∪ C using diagonal lines (of different types). Then the intersection
A ∩ (B ∪ C) has both diagonal line shadings, which we have emphasised
as the grey-shaded region.
In the right-hand diagram, we have separately shaded the sets A∩B and
A ∩ C using diagonal lines (again, of different types). Then the union
of these sets (A ∩ B) ∪ (A ∩ C) has either or both of the diagonal line
shadings and again we have emphasised as the grey-shaded region.
The grey-shaded region in each diagram is the same, which illustrates
98

that A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C). We should emphasise that the

diagrams in figure 3.8 do not prove that A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C)
for all sets A, B, and C.

A B U A B U

C C

A BÈC AÇ B AÇC

A Ç (B È C ) ( A Ç B) È ( A Ç C )

FIGURE 3.8: Illustrating A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C).

We have previously noted the connection between intersection of sets and

conjunction of propositions and between unions of sets and disjunctions. This
connection between sets and logic extends further. The complement of a set
A , Ā, contains all those elements of the universe not in A, so complement
corresponds to negation. To be a little more precise, let A = {x : P (x)}
be a set defined by a predicate P . Then the complement of A is defined by
Ā = {x : ¬P (x)}.
In section 2.3, we used true to denote a tautology (a proposition that is always
true) and false to denote a contradiction (a proposition that is always false).
If we imagine these as defining sets then {x : true} gives the universal set U
since true is satisfied for all elements x ∈ U . Similarly, {x : false} gives the
empty set ∅ since false is satisfied for no elements x.
In table 2.1 on page 39, we gave a list of standard logical equivalences. Each
of these has its counterpart as an equality involving sets. For example, the
identity illustrated in figure 3.8,
A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C),
is the counterpart of a distributive law for propositions
P ∧ (Q ∨ R) = (P ∧ Q) ∨ (P ∧ R).
The set theory identities corresponding to the logical equivalences in table 2.1
are listed in the table below.
 99

Set Theory Laws

Idempotent Laws A∩A=A

A∪A=A

Commutative Laws A∩B =B∩A

A∪B =B∪A

Associative Laws (A ∩ B) ∩ C = A ∩ (B ∩ C)
(A ∪ B) ∪ C = A ∪ (B ∪ C)

Absorption Laws A ∩ (A ∪ B) = A
A ∪ (A ∩ B) = A

Distributive Laws A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C)
A ∪ (B ∩ C) = (A ∪ B) ∩ (A ∪ C)

Involution Law Ā¯ = A

De Morgan’s Laws A ∪ B = Ā ∩ B̄
A ∩ B = Ā ∪ B̄

Identity Laws A∪∅=A

A∩U =A
A∪U =U
A ∩ Ā = ∅

Complement Laws A ∪ Ā = U
A ∩ Ā = ∅
∅
¯ =U
U¯ = ∅

TABLE 3.1: Set theory laws.

Power set

Let A be any set. Then we can define a set comprising all the subsets of A.
This is called the power set of A, denoted P(A).
100

Examples 3.8
1. The following is a list of sets A and their power sets P(A) for some small
sets.
A=∅ P(A) = {∅}
A = {a} P(A) = {∅, {a}}
A = {a, b} P(A) = {∅, {a}, {b}, {a, b}}
A = {a, b, c} P(A) = {∅, {a}, {b}, {c}, {a, b}, {a, c}, {b, c}, {a, b, c}}
A = {a, b, c, d} P(A) = {∅, {a}, {b}, {c}, {d}, {a, b}, {a, c}, {a, d},
{b, c}, {b, d}, {c, d}, {a, b, c}, {a, b, d},
{a, c, d}, {b, c, d}, {a, b, c, d}}
Note that, for every set A, the empty set ∅ and the set A itself both
belong to the power set P(A).
Also note that the power set of the empty set is not empty; P(∅) = {∅}
which is a set containing a single element, namely the empty set. Using
the analogy introduced in example 3.3.6, P(∅) = {∅} is a paper bag
that contains another paper bag (albeit an empty one); this is different,
of course, from an empty paper bag.
2. Let A = {a, b, c} and B = {a, b}. Determine whether each of the follow-
ing is true or false and give a brief justification.

(i) B ∈ P(A)
(ii) A ∈ P(A)
(iii) A ⊆ P(A)
(iv) {{a, c}, B} ⊆ P(A)
(v) {a, c} ∈ P(A) ∩ P(B)
(vi) ∅ ⊆ P(A)

Solution

(i) B ∈ P(A) is true since B = {a, b} is one of the elements listed in

P({a, b, c}) in Example 1.
(ii) A ∈ P(A) is true. For any set A, we have A ⊆ A, which means that
A is an element of the power set P(A).
(iii) This is false. Recall that X ⊆ Y means each element of X is also
an element of Y . In this case, the elements of A are letters (a, b,
c), whereas the elements of P(A) are sets. Hence A ̸⊆ P(A).
(iv) The statement is true. Each of the sets {a, c} and B = {a, b}
is a subset of A and hence is an element of P(A). Therefore
{{a, c}, B} ⊆ P(A).
 101

(v) This is false. The set {a, c} is not a subset of B = {a, b}, and so
does not belong to P(B). Hence {a, c} ̸∈ P(A) ∩ P(B).
(vi) The statement ∅ ⊆ P(A) is true because the empty set is a subset
of every set.

Exercises 3.2
1. For each of the following, draw a single Venn-Euler diagram and shade
the region representing the given set.
(i) (A − B) ∪ (B − A)
(ii) (A ∪ B) − (A ∩ B)
(iii) (A ∩ B) ∪ C
(iv) A ∩ (B ∪ C)
(v) (A − B) ∪ C
2. Let U = {n : n ∈ Z and 1 ≤ n ≤ 10}, A = {n : n is even}, B = {n :
n is prime}, and C = {1, 4, 9}.
Define each of the following sets by listing their elements.
(i) A∩B
(ii) A∪B
(iii) A−B
(iv) B∩C
(v) Ā ∩ B
(vi) A ∩ (B ∪ C)
(vii) A∪C
(viii) (A − C) − B
3. Consider the sets A, B, C, D, and E represented by the following Venn-
Euler diagram. (The sets C and E are represented by shaded regions.)
For each of the following pairs of sets X and Y , state whether X ⊆ Y ,
Y ⊆ X, X ∩ Y = ∅ (that is, X and Y are disjoint) or none of these.
102

(i) X = A ∪ B, Y =C
(ii) X = A ∩ B, Y =D
(iii) X = A ∩ B, Y =C
(iv) X = E, Y =A∩D
(v) X = B ∩ C, Y =C ∪D
(vi) X = A ∩ E, Y =D∪E
(vii) X = C ∪ E, Y =A∪D
(viii) X = C − B, Y =D∪E
(ix) X = A ∪ D, Y =B∩E
(x) X = A − E, Y =A−D

4. (i) Show that if A, B, and C are sets such that C ⊆ A and C ⊆ B,

then C ⊆ A ∩ B.
(ii) Show that if A, B, and C are sets such that A ⊆ C and B ⊆ C,
then A ∪ B ⊆ C.

5. The symmetric difference A ∗ B of two sets A and B is defined by

A ∗ B = (A − B) ∪ (B − A).

(i) Let A = {2, 4, 6, 8, 10}, B = {4, 8, 12, 16}, and C = {1, 4, 16, 64}.
List the elements of each of the following sets.
(a) A∗B
(b) A∗C
(c) A ∗ (B ∩ C)
(d) A ∗ (B ∪ C)
(e) (A ∗ B) ∩ (A ∗ C)
(f) (A ∗ B) ∪ (A ∗ C)

(ii) For each of the following, draw a single Venn-Euler diagram and
shade the region representing the given set.
(a) A ∗ (B ∩ C)
(b) A ∗ (B ∪ C)
(c) (A ∗ B) ∩ (A ∗ C)
(d) (A ∗ B) ∪ (A ∗ C)

6. List the elements of P(A) in each of the following cases.

(i) A = {1, 2, 3}
(ii) A = {1, {2, 3}}
 103

(iii) A = {{1}, {2, 3}}

(iv) A = {{1, 2, 3}}
(v) A = P({1, 2})

7. Let A = {a, b}, and B = {b, c, d}.

Evaluate P(A), P(B), P(A ∩ B), and P(A) ∩ P(B).
What do you notice about the relationship between P(A∩B) and P(A)∩
P(B)?
Will this relationship always hold? If so, can you explain why; if not,
can you explain why not?

3.4 The Cartesian product

The order in which the elements of a finite set are listed is immaterial; in par-
ticular, {x, y} = {y, x}. In some circumstances, however, order is significant.
For instance, in coordinate geometry the points with coordinates (1, 2) and
(2, 1), respectively, are distinct. We therefore wish to define, in the context of
sets, something akin to the coordinates of points used in analytical geometry.
An ordered pair (x, y) of elements satisfies:

(x, y) = (a, b) if and only if x = a and y = b.

With this definition it is clear that (x, y) and (y, x) are different (unless x = y),
so the order is significant. It could be argued, with justification, that we have
not really defined the ordered pair, but merely listed a property which we
desire of it. In fact, (x, y) can be defined in terms of (unordered) sets, but this
is not particularly instructive. What is important about ordered pairs is the
property above.
Let A and B be sets. Their Cartesian product 3 denoted A × B is the set
of all ordered pairs where the first element is drawn from the set A and the
second element is drawn from the set B:

A × B = {(x, y) : x ∈ A and y ∈ B}.

When A = B we write A2 for the Cartesian product A×A. Thus, for example,
the Cartesian plane is R2 = R × R = {(x, y) : x ∈ R and y ∈ R}.
3 Named after the French mathematician and philosopher René Descartes (1596–1650),

the founder of analytical geometry.

104

Since Venn-Euler diagrams have proved to be a useful representation of sets,

we would like to have a diagrammatic representation of the Cartesian product
A × B that relates it to the sets A and B. The idea for the diagrammatic
representation comes from the coordinate plane R2 with perpendicular axes.
For arbitrary sets, we represent A and B as horizontal and vertical line seg-
ments (rather than regions in the plane). Then A × B is represented as the
rectangular region lying above A and to the right of B — see figure 3.9. An
element (a, b) ∈ A × B is represented as the point lying above a and to the
right of b.

B A´ B

(a,b)
b

a A

FIGURE 3.9: Diagram representing the Cartesian product A × B.

Examples 3.9

1. Let A = {1, 2, 3, 4} and B = {a, b, c}. Then

A × B = {(1, a), (1, b), (1, c), (2, a), (2, b), (2, c),
(3, a), (3, b), (3, c), (4, a), (4, b), (4, c)}.

2. Let A = {1, 2, 3}, B = {2, 3, 4} and X = {a, b, c}

We will investigate the interaction between intersection and the Carte-
sian product.
Since

A × X = {(1, a), (1, b), (1, c), (2, a), (2, b), (2, c), (3, a), (3, b), (3, c)}

and

B × X = {(2, a), (2, b), (2, c), (3, a), (3, b), (3, c), (4, a), (4, b), (4, c)},

it follows that their intersection is

(A × X) ∩ (B × X) = {(2, a), (2, b), (2, c), (3, a), (3, b), (3, c)}.
 105

Now A ∩ B = {2, 3}. Hence

(A ∩ B) × X = {(2, a), (2, b), (2, c), (3, a), (3, b), (3, c)}.

Therefore, for these sets,

(A × X) ∩ (B × X) = (A ∩ B) × X.

An obvious question is whether this equation is satisfied for all sets A,

B and X or whether it was a particular property of these sets.
In fact, the equation holds for all sets, and this is illustrated in fig-
ure 3.10. The figure shows both A × X and B × X with different di-
agonal line shading. Their intersection, (A × X) ∩ (B × X), has both
line shadings and this is emphasised by the gray shading. However, this
grey-shaded region is the region lying above A ∩ B, so it also represents
(A ∩ B) × X.
The diagram in figure 3.10 is not a proof of this identity, but it does
give a powerful visualisation of it.

A´ X B´ X

A
AÇ B B

A´ X B´ X

FIGURE 3.10: Diagram representing the interaction between Cartesian

product and intersection.

We can extend the definition of Cartesian product to n sets. Ordered pairs

readily generalise to ordered triples (x, y, z), ordered quadruples (w, x, y, z),
or more generally, ordered n-tuples (x1 , x2 , . . . , xn ) with the property that

(x1 , x2 , . . . , xn ) = (a1 , a2 , . . . , an ) ⇔ x1 = a1 , x2 = a2 , . . . , xn = an
⇔ xr = ar , for r = 1, 2 . . . , n.

Let A1 , A2 , . . . An be sets. Their Cartesian product is the set of all ordered

106

n-tuples where the first element lies in the first set A1 , the second element lies
in the second set A2 , and so on:

A1 × A2 × . . . × An = {(a1 , a2 , . . . , an ) : a1 ∈ A1 , a2 ∈ A2 , . . . , an ∈ An }.

Example 3.10
Let A = {1, 2, 3}, B = {a, b, c} and C = {α, β}. Then

A × B × C = {(1, a, α), (1, a, β), (1, b, α), (1, b, β), (1, c, α), (1, c, β),
(2, a, α), (2, a, β), (2, b, α), (2, b, β), (2, c, α), (2, c, β),
(3, a, α), (3, a, β), (3, b, α), (3, b, β), (3, c, α), (3, c, β)}.

Exercises 3.3
1. Let A = {1, 2, 3}, B = {2, 3, 4}, X = {a, b, c}, and Y = {c, d, e}. Evalu-
ate each of the following sets by listing their elements.

(i) A × (X ∩ Y )
(ii) (A ∩ B) × (X ∩ Y )
(iii) (A × X) ∩ (B × Y )
(iv) (A × Y ) ∩ (B × X)
(v) (A − B) × X
(vi) (A × X) − (B × X)
(vii) (A ∪ B) × (X ∩ Y )
(viii) (A ∩ X) × Y

2. This question explores how the Cartesian product interacts with the
subset relation.

(i) Choose sets A, B, X, and Y satisfying A ⊆ X and B ⊆ Y .

Evaluate A × B and X × Y . What do you notice?
(ii) Draw a diagram, similar to that in figure 3.10, to show that, if
A ⊆ X and B ⊆ Y , then A × B ⊆ X × Y .
(iii) Is every subset of X × Y of the form A × B for some A ⊆ X and
B ⊆ Y ? Explain your answer.
(iv) Suppose sets A, B, X, and Y are such that A × B ⊆ X × Y . Does
it necessarily follow that A ⊆ X and B ⊆ Y ?

3. This question considers the following identity:

(A ∩ B) × (X ∩ Y ) = (A × X) ∩ (B × Y ).
 107

(i) Verify this identity for the sets A = {1, 2, 3, 4}, B = {2, 4, 6}, X =
{x, y}, and Y = {y, z}.
(ii) Draw two diagrams similar to that in figure 3.10, one showing

(A ∩ B) × (X ∩ Y )

and one showing

(A × X) ∩ (B × Y ),
to illustrate the identity.

4. The aim now is to find and illustrate an identity, similar to that in

question 3, for the set (A ∪ B) × (X ∪ Y ).

(i) Using the same sets as in question 3 (i), A = {1, 2, 3, 4}, B =

{2, 4, 6}, X = {x, y}, and Y = {y, z}, show that

(A ∪ B) × (X ∪ Y ) ̸= (A × X) ∪ (B × Y ).

Note that you only need to show the two sets are different; it should
not be necessary to list all the elements of both sets.
(ii) By drawing similar diagrams to those in question 3 (ii), conjecture
a general expression for

(A ∪ B) × (X ∪ Y )

as a union of Cartesian products.

3.5 Functions and composite functions

In this section we introduce another of the central concepts of modern math-
ematics, that of a function or mapping. Although functions have been used
in mathematics for several centuries, it is only comparatively recently that
a rigorous and generally accepted definition of the concept has emerged. We
will actually give two definitions of a function. The first definition, below, is a
little informal but captures the idea of a function. Later, we will give a more
formal definition based on Cartesian products of sets that makes a little more
precise one aspect of the informal definition. We will use the terms ‘function’
and ‘mapping’ interchangeably.

Definition 3.1 (Informal definition of a function)

Let A and B be sets. A function or mapping f : A → B is a rule which
associates, to each element a ∈ A, a unique element b = f (a) ∈ B.
108

This is a very general definition — A and B are any sets and f is any rule with
the appropriate property. It is quite common to visualise the function rule as
being encapsulated in a ‘function machine’. This is a ‘black box’, illustrated
in figure 3.11, which has the property that if an element a ∈ A is fed into the
machine, it produces as output the associated element f (a) ∈ B.

a f f(a)

FIGURE 3.11: A function machine.

An alternative visualisation, which shows f linking the two sets A and B, is

given in figure 3.12.

f
A B

a f(a)

FIGURE 3.12: Diagram of a function showing the sets involved.

Both representations of a function are useful in different contexts, although

figure 3.12 is probably more satisfactory as is shows the three different com-
ponents of the function: the sets A and B and the rule associating elements
of A with elements of B.
Let f : A → B be a function. Then the set A is called the domain of f and
the set B is called the codomain of f . For a ∈ A, the element b = f (a) ∈ B
is called the image of a and we also write a 7→ b in this case. The set of all
images of elements of A is called the image of f , im f:

im f = {b ∈ B : b = f (a) for some a ∈ A} = {f (a) : a ∈ A}.

The image of f is illustrated in figure 3.13.

Examples 3.11
1. Let A = {a, b, c, d, e} and B = {1, 2, 3, 4, 5, 6}. A function f : A → B is
defined by
a 7→ 2, b 7→ 4, c 7→ 2, d 7→ 5, e 7→ 1.

This function is illustrated in figure 3.14.

 109

f
A B

a
f(a) im f

FIGURE 3.13: Diagram showing the image of a function.

f
A B

a 1
b 2
c 3
d 4
e 5
6

FIGURE 3.14: Illustrating the function in example 3.11.1.

Note that it is possible to have two different elements of A with the

same image in B; in this example we have both f (a) = 2 and f (c) = 2.
Also, it is possible to have elements of B that are not the image of any
element of A; in this example, 3 is not the image of any a ∈ A.
The image of the function is im f = {1, 2, 4, 5}.
2. Frequently, authors of textbooks will use phrases such as ‘the function
f (x) = (x + 1)2 ’ or ‘the function g(x) = sin x/x’. However, the expres-
sion itself does not define a function because the domain and codomain
have not been specified. There is usually an implicit assumption that a
numerical variable x ranges over the real numbers or a subset of it.
Thus the first expression defines a function
f : R → R, f (x) = (x + 1)2 .
However the second function does not define a function R → R because
the expression sin x/x is not defined when x = 0. Usually, in such a
situation, we would wish to define the function to have as large a domain
as possible. Hence the second expression defines a function
sin x
g : R − {0} → R, g(x) = .
x
110

The domain R − {0} is the set of non-zero real numbers which is fre-
quently denoted
R∗ = {x ∈ R : x ̸= 0}.

3. Find the image of each of the following functions.

(i) f : R → R, f (x) = x2 .
2x
(ii) g : R → R, g(x) = 2 .
x +1

Solution

(i) For the square function, we have f (x) = x2 ≥ 0 for all (x ∈) R.

√ √
Also, for any y ≥ 0 taking x = y ∈ R, we have f (x) = f y =
(√ )2
y = y. Hence the image of f is

im f = {y ∈ R : y ≥ 0}.

The image of f is illustrated in figure 3.15.

y
y = x2
im f

FIGURE 3.15: The image of the square function R → R, x 7→ x2 .

2x
(ii) Finding the image of g : R → R, g(x) = is more difficult
+1 x2
since we cannot immediately ‘see’ what the image is.
Now y ∈ im g if and only if

2x
y= for some x ∈ R.
x2 + 1
Now this is equivalent to

yx2 + y = 2x

or
yx2 − 2x + y = 0.
 111

Regarding this as a quadratic equation in x and using the quadratic

formula, we have, provided y ̸= 0,
√
2 ± 4 − 4y 2
x= .
2y
In order that this has a real solution we require y ̸= 0 and

4 − 4y 2 ≥ 0.

Hence
y 2 ≤ 1 (and y ̸= 0),
which means
−1 ≤ y ≤ 1 (and y ̸= 0).
Therefore, provided −1 ≤ y ≤ 1 and y ̸= 0, there exists a real
number x such that y = g(x). The value y = 0 is a special case,
but clearly g(0) = 0, so 0 ∈ im g.
Therefore im g is the interval

im g = [−1, 1] = {y ∈ R : −1 ≤ y ≤ 1}.

A sketch of the graph of g showing im g is given in figure 3.16.

y
2x
1 y= 2
im f x +1

-1

2x
FIGURE 3.16: The image of the function g : R → R, g(x) = .
x2 + 1

The definition of a function f : A → B given in definition 3.1 raises an im-

portant question: what constitutes a ‘rule’ ? Sometimes this will be a formula,
like the functions defined in example 3.11.2 above. However the ‘rule’ may
simply be describing explicitly which element is associated to which as in ex-
ample 3.11.1 above. In this case, we can describe the ‘rule’ compactly just by
listing, as pairs, the elements and their images:

(a, 2), (b, 4), (c, 2), (d, 5), (e, 1).
112

This list defines a subset of the Cartesian product A × B = {(a, b) : a ∈ A, b ∈

B} comprising all the pairs (a, f (a))

{(a, f (a)) : a ∈ A}.

This is the key to the formal definition of a function. Instead of trying to

define what we mean by a rule (that associates elements of A with elements
of B), instead we simply define the set of those pairs (a, f (a)) using any of
the ways described earlier for defining sets.

Definition 3.2 (Formal definition of a function)

Let A and B be sets. A function or mapping f : A → B is a subset of the
Cartesian product, f ⊆ A × B, that satisfies:

for each a ∈ A there exists a unique b ∈ B such that (a, b) ∈ f.

Examples 3.12

1. The function f : A → B in example 3.11.1 above is defined by the set

f = {(a, 2), (b, 4), (c, 2), (d, 5), (e, 1)} ⊆ A × B.

2. Similarly, the two functions in example 3.11.2 are defined by the sets
{( ) }
f = x, (x + 1)2 : x ∈ R ⊆ R2
{( ) }
sin x
and g = x, : x ∈ R − {0} ⊆ R − {0} × R.
x

3. The square function was defined informally in example 3.11.3 (i) as f :

R → R, f (x) = x2 . We may define this formally as

f = {(x, y) ∈ R2 : y = x2 }.

Suppose we attempt to define the square root function g in a similar

way as the set
g = {(x, y) ∈ R × R : x = y 2 }.
There is a problem with this, however. Figure 3.17 shows this set as a
subset of R2 . The set g does not satisfy the condition

for each x ∈ R there exists a unique y ∈ R such that (x, y) ∈ g

in definition 3.5 in two respects.

Firstly, for some x ∈ R, there does not exist any y ∈ R such that
(x, y) ∈ g. For example, there is no element (−1, y) ∈ g. Secondly, for
some x ∈ R, there exist two values of y ∈ R such that (x, y) ∈ g. For
example, both (4, 2) ∈ g and (4, −2) ∈ g.
 113

FIGURE 3.17: Attempting to define the square root function.

In order to overcome these problems and properly define a square root

function, we need to restrict the domain and codomain. If we define

R≥0 = {x ∈ R : x ≥ 0}.

then we may define this square root function informally as

√
g : R≥0 → R≥0 , g(x) = x

or formally as
√
g = {(x, y) ∈ R≥0 × R≥0 : x = y 2 } = {(x, y) ∈ R≥0 × R≥0 : y = x}.
√
Note that x means ‘the non-negative square root of x’.

Composition of functions

Let f : A → B and g : B → C be two functions. If x ∈ A, then y = f (x)

belongs to B so we can ‘apply’ the function g to y to get z = g(y) = g(f (x))
which belongs to C. In symbols:

x ∈ A ⇒ y = f (x) ∈ B ⇒ z = g(y) = g(f (x)) ∈ C.

This association x 7→ g(f (x)) defines a function A → C, called the composite

of f and g. The composite function is denoted g ◦ f :

g ◦ f : A → C, (g ◦ f )(x) = g(f (x)).

The composite function g ◦ f is illustrated in figure 3.18.

If we think of the functions f and g being represented by ‘function machines’,
then the composite g ◦f has a function machine that is obtained by connecting
the output of f to the input of g. This is represented in figure 3.19.
114

f g
A B C

x y z

gof

FIGURE 3.18: Illustrating the composite function.

f (a)
a f g ( g o f )(a )

gof

FIGURE 3.19: Composite of function machines.

Examples 3.13
1. Let f : R → R, f (x) = 4x − 1 and g : R → R, g(x) = x2 + 1 be two
functions. Evaluate:
(i) g ◦ f (2)
(ii) f ◦ g(2)
(iii) g ◦ f (x)
(iv) f ◦ g(x)

Solution

(i) Since g ◦ f (2) = g(f (2)), we first need to evaluate f (2) and then
‘apply’ g to the result. We have f (2) = 4 × 2 − 1 = 7, so

g ◦ f (2) = g(f (2)) = g(7) = 72 + 1 = 50.

(ii) This time we evaluate g(2) = 22 + 1 = 5 and then ‘apply’ f to the

result:
f ◦ g(2) = f (g(2)) = f (5) = 4 × 5 − 1 = 19.
(iii) Repeating the process in part (i) but with a general x gives

g ◦ f (x) = g(f (x)) = g(4x − 1) = (4x − 1)2 + 1 = 16x2 − 8x + 2.

 115

(iv) Similarly, applying the function f to g(x) = x2 + 1 gives

f ◦ g(x) = f (g(x)) = f (x2 + 1) = 4(x2 + 1) − 1 = 4x2 + 3.

2. Let f : R → R, f (x) = 2x + 1. Define ‘multiple’ composite functions as

follows:

f [2] = f ◦ f, f [3] = f ◦ (f ◦ f ) = f ◦ f [2] , f [4] = f ◦ f [3] . . . .

Conjecture an expression for f [n] (x).

Solution
We have
f [2] (x) = f ◦ f (x) = f (2x + 1) = 2(2x + 1) + 1 = 4x + 3,
f [3] (x) = f ◦ f [2] (x) = f (4x + 3) = 2(4x + 3) + 1 = 8x + 7,
f [4] (x) = f ◦ f [3] (x) = f (8x + 7) = 2(8x + 7) + 1 = 16x + 15,
f [5] (x) = f ◦ f [4] (x) = f (16x + 15) = 2(16x + 15) + 1 = 32x + 31.

From these examples we conjecture that

f [n] (x) = 2n x + (2n − 1).

In chapter 8 we will develop a method of proof that will allow us to

prove this conjecture.

Exercises 3.4

1. Three functions f , g, and h are defined as follows.

f : R → R, f (x) = |x − 1|
2x
g : Z → R, g(x) = 2
x −3
h : R → R, h(x) = 2x + 1

For each of the following, either evaluate the expression or explain why
it is not defined.
( )
(i) f 12 (v) (g ◦ g)(3)
(1)
(ii) g 2 (vi) (h ◦ h)(x)
(1)
(iii) (g ◦ h) 2 (vii) (f ◦ h)(x)
(iv) (f ◦ f )(−2) (viii) (f ◦ g)(x)
116

2. Three functions f , g, and h are defined as follows.

f : R → R, f (x) = 2x + 1
1
g : R → R, g(x) = 2
x +1
√
h : R → R, h(x) = x2 + 1

Find expressions for each of the following.

(i) (f ◦ g)(x) (iv) (f ◦ f )(x))
(ii) (g ◦ f )(x) (v) ((f ◦ g) ◦ h)(x)
(iii) (g ◦ h)(x) (vi) (f ◦ (g ◦ h))(x)

3. Let f : A → B and g : C → D be two functions.

What is the most general situation in which the composite function g ◦ f
is defined? Draw a diagram to illustrate this situation.

4. Find im f, the image of f , for each of the following functions.

(i) A = P({a, b, c, d}), the power set of {a, b, c, d},

f : A → Z, f (C) = |C|.
(ii) f : Z → Z, f (n) = n2 .
(iii) A = {countries of the world}, B = {cities of the world}
f : A → B, f (X) = the capital city of X.
(iv) A = P({a, b, c, d}), f : A → A, f (X) = X ∩ {a}.
(v) A = P({a, b, c, d}), f : A → A, f (X) = X ∪ {a}.

5. Determine the image of each of the following functions.

(i) f : R → R, x 7→ x2 + 2
(ii) f : R → R, x 7→ (x + 2)2
(iii) f : R → R, x 7→ 1/(x2 + 2)
(iv) f : R → R, x 7→ 4x/(x2 + 5)
√
(v) f : R → R, x 7→ x2 + 1

6. Let f : A → B and g : B → C be two functions. Give the formal

definition of the composite function g ◦ f : A → C as a subset of A × C.
 117

3.6 Properties of functions

A function f : A → B is injective or one-one if different elements have
different images; in other words if we never have the situation given in fig-
ure 3.20.

f
A B

a1
b
a2

FIGURE 3.20: Different elements of the domain with the same image.

The statement that different elements have different images is:

for all a1 , a2 ∈ A, a1 ̸= a2 ⇒ f (a1 ) ̸= f (a2 ).
Recall from section 2.3, page 40, that a conditional statement P ⇒ Q is
logically equivalent to its contrapositive ¬Q ⇒ ¬P . For any a1 , a2 ∈ A the
two statements
a1 ̸= a2 ⇒ f (a1 ) ̸= f (a2 ) and f (a1 ) = f (a2 ) ⇒ a1 = a2
are contrapositives of one another and hence are logically equivalent. In math-
ematics, it is generally easier to work with equals rather than not equals, so
we will usually use the contrapositive of the conditional statement:
for all a1 , a2 ∈ A, f (a1 ) = f (a2 ) ⇒ a1 = a2 .

Definition 3.3
Let f : A → B be a function. Then f is injective (or one-one) if:
for all a1 , a2 ∈ A, f (a1 ) = f (a2 ) ⇒ a1 = a2 .

We say that a function f : A → B is surjective or onto if every element of

B is the image of some element of A; more colloquially, f is surjective if every
element of B is ‘hit’ by an arrow coming from A. This is equivalent to saying
that the image of f is ‘all of’ the codomain b, im f = B. This is illustrated in
the diagram in figure 3.21.
118

f
A B

im f = B

FIGURE 3.21: A surjective function.

Definition 3.4
Let f : A → B be a function. Then f is surjective (or onto) if:

for every b ∈ B, there exists a ∈ A such that f (a) = b.

An injective function is also called an injection and a surjective function is

also called a surjection.

Examples 3.14

1. Consider the function f : {a, b, c, d, e} → {1, 2, 3, 4, 5, 6} defined by

a 7→ 2, b 7→ 4, c 7→ 2, d 7→ 5, e 7→ 1.

This function is illustrated in figure 3.14.

The function is not injective because f (a) = 2 = f (c) (but a ̸= c). Also
f is not surjective because 3 in the codomain is not the image of any
element of the domain, 3 ̸∈ im f. Note that 6 ̸∈ im f either, but only one
element not in im f is needed to show that the function is not surjective.

2. Consider the square function f : R → R, f (x) = x2 . The graph of f is

given in figure 3.15.
Clearly it is possible for different elements of the domain to have the
same image; for example, f (2) = 4 = f (−2). Therefore f is not injective.
Also, since x2 ≥ 0 for all x ∈ R, no negative real number is in the
image of f ; for example, −1 ̸= im f. Therefore f is not surjective. In
example 3.11.3 (i), we showed that im f = R≥0 = {y ∈ R : y ≥ 0}. In
this case, we don’t need to determine im f completely to show that f is
not surjective; all we require is an element in the codomain R that is
not in the image of the function im f.
 119

Now consider the function g : R≥0 → R≥0 , g(x) = x2 . In other words,

g has the same ‘rule’ as f but its domain and codomain are different.
If x ≥ 0 and y ≥ 0 are such that x2 = y 2 , then x and y are equal.
Hence g is injective. Similarly, if y ≥ 0, then y ∈ im g (as shown in
example 3.11.3 (i)). So g is also surjective.
Comparing the functions f and g, we can see that the definitions of the
domain and codomain are crucially important in terms of the properties
of the functions. The functions f and g both are defined by the same
rule, although one function is both injective and surjective but the other
is neither.

3. In the previous example, we saw that f was neither injective nor sur-
jective and g was both injective and surjective. To emphasise that the
properties of injectivity and surjectivity are independent of one another,
define functions that are:

(i) injective but not surjective

(ii) surjective but not injective.

Solution
Simple solutions are provided using small finite sets for the domain and
codomain which we do first in each case. A little more challenging is to
find examples A → B where A and B are both subsets of R and we do
this secondly in each case.

(i) To construct a function A → B that is injective but not surjective

we require that different elements of A have different images and
also that there are elements of B that are not the image of any
element of A.
Let A = {a, b, c} and B = {1, 2, 3, 4} and define f : A → B by

f (a) = 1, f (b) = 2, f (c) = 3.

Clearly, different elements of A have different images, so f is injec-

tive. Also 4 ∈ B is not the image of any element of A so 4 ̸∈ im f.
Hence f is not surjective.

For an example where the domain and codomain are subsets of R,

let
g : R → R, g(x) = ex .
If x ̸= y, then ex ̸= ey so g is injective. Also ex > 0 for all x ∈ R
so, for example, −1 ̸∈ im g. Hence g is not surjective.
(ii) To construct a function A → B that is surjective but not injective,
we require that each element of B is the image of some element of
120

A and also that (at least) a pair of elements of A have the same
image.
Let A = {a, b, c, d} and B = {1, 2, 3} and define f : A → B by

f (a) = 1, f (b) = 2, f (c) = 3, f (d) = 3.

Clearly, every element in B is the image of some element in A, so g

is surjective. Also g(c) = 3 = g(d) but c ̸= d, so g is not injective.
For an example where the domain and codomain are subsets of R,
we can use the square function. Let

g : R → R≥0 , g(x) = x2 .

Every y ≥ 0 is the image of some x ∈ R so g is surjective, Since

g(2) = 4 = g(−2) but 2 ̸= −2, for example, it follows that g is not
injective.

Reflecting on the examples we have considered of functions f : A → B where

A and B are subsets of R, we now describe how we can tell from its graph
whether a function is injective and/or surjective.
Let f : A → B be a function where A and B are subsets of R. Suppose, first,
that f is not injective. Then there are distinct elements a1 and a2 in A such
that f (a1 ) = f (a2 ) = b, say. This means that the horizontal line at height b
meets the graph at points corresponding to x = a1 and x = a2 on the x-axis.
This situation is illustrated in figure 3.22. If, on the other hand, f is injective,
then this situation never occurs. In other words, a horizontal line through any
point in B on the y-axis will not meet the graph in more than one point.

b¢

y = f ( x)
B
b

A a1 a2

FIGURE 3.22: Horizontal line test for functions R → R.

Suppose now that f is not surjective. Then there is an element b′ ∈ B which

is not the image of any element of A. In terms of the graph, this means that
the horizontal line through b′ does not meet the graph at any point. Again,
figure 3.22 illustrates this case. If, on the other hand, f is surjective, then this
 121

situation never occurs. In other words, a horizontal line through any point
b ∈ B on the y-axis will meet the graph in at least one point.
These considerations are summarised in the following theorem.

Theorem 3.2 (Horizontal line test)

Let f : A → B be a function, where A and B are subsets of R. Then:

(i) f is injective if and only if every horizontal line through a point of B on

the y-axis meets the graph of f at most once;
(ii) f is surjective if and only if every horizontal line through a point of B
on the y-axis meets the graph of f at least once.

Bijective functions and inverse functions

We now turn our attention to a particularly important class of functions —

those that are both injective and surjective.

Definition 3.5
A function f : A → B is bijective, or is a bijection, if it is both injective
and surjective.

Given a function f : A → B, ‘reversing the arrows’ does not necessarily

define a function B → A. Consider, for example, the function defined in
example 3.11.1:

f : {a, b, c, d, e} → {1, 2, 3, 4, 5, 6}, a 7→ 2, b 7→ 4, c 7→ 2, d 7→ 5, e 7→ 1.

The arrow diagram for f was given in figure 3.14. Reversing the arrows gives
the diagram shown in figure 3.23. This diagram is not that of a function
B → A for two reasons. Firstly, the elements 3 and 6 are not associated with
element of A. Secondly, the element 2 in B is associated with two elements,
a and c in A. Thus the arrow diagram in figure 3.23 fails both aspects of the
definition of a function given in definition 3.1.
One of the reasons why bijective functions are important is that if f : A → B is
a bijection, then neither of these problems occurs when reversing the arrows.
Hence, if f : A → B is a bijection, then ‘reversing the arrows’ does define
a function B → A. The resulting function B → A is called the inverse
function of f and is denoted f −1 . By reversing the arrows we mean that if f
gives an arrow from a to b, then f −1 gives an arrow from b to a. Symbolically, if
f (a) = b, then f −1 (b) = a. We summarise these considerations in the following
definition.
122

A B

a 1
b 2
c 3
d 4
e 5
6

FIGURE 3.23: Reversing the arrows in example 3.11.1.

Definition 3.6
Let f : A → B be a bijective function. The function f −1 : B → A defined by

f −1 (b) = a if and only if b = f (a)

is called the inverse function of f .

The definition of the inverse function of a bijection is illustrated in figure 3.24.

f
A B
f -1

a = f -1(b) b = f(a)

FIGURE 3.24: Defining the inverse function.

Examples 3.15

1. In example 3.14.2, we showed that the square function

g : R≥0 → R≥0 , g(x) = x2

√
is bijective. If y = x2 = g(x), then x = y = g −1 (y). Hence the inverse
function is
√
g −1 : R≥0 → R≥0 , g −1 (y) = y.
Of course the variable name is not significant
√ when defining a function,
so we could have defined g −1 by g −1 (x) = x.
 123

2. Let f : R → R be defined by f (x) = 3x − 1.

The graph of f is a straight line — see figure 3.25. Therefore, by the
horizontal line test, theorem 3.2, f is both injective and surjective.
To find the inverse function, we let y = f (x) and then rearrange to
obtain x = f −1 (y):

y = f (x) ⇒ y = 3x−1 ⇒ 3x = y+1 ⇒ x = 13 (y+1) ⇒ f −1 (y) = 13 (y+1).

Hence the inverse function, defined in full, is

f −1 : R → R, f −1 (y) = 13 (y + 1).

y
y = 3x - 1

x
1
3
-1

FIGURE 3.25: The graph of f : R → R, f (x) = 3x − 1.

We have seen that if a function f : A → B is not bijective, then the process

of reversing the arrows does not define a function B → A. Let us examine
this in a little more detail by considering separately the impact of the failure
of a function to be injective and surjective when trying to define an inverse
function.
If f : A → B is not injective, then there exists at least two distinct elements
a1 , a2 ∈ A with the same image. This is illustrated in figure 3.26. Reversing
the arrows does not define a function B → A because there is an element in
B that is associated with more than one element of A. This ‘problem’ is not
easily rectified. It would involve identifying some elements of the domain A to
be removed, although there is not an obvious way to do this. If f (a1 ) = f (a2 ),
we would need to remove either a1 or a2 or both from the domain; the situation
would be more complicated if there were several elements of A all with the
same image. Thus a failure of injectivity may be regarded as a ‘killer blow’ to
defining an inverse function.
Now consider the case f : A → B is injective but not surjective. This is
illustrated in figure 3.27. In this case, reversing the arrows does not define a
124

f
A B A B

f is not injective not a function

FIGURE 3.26: Reversing arrows when the function is not injective.

f
A B A B

f is not surjective not a function

FIGURE 3.27: Reversing arrows when the function is not surjective.

function B → A because there are elements of B that are not associated with
any element of A.
This problem is more easily fixed than a failure of injectivity. If we ‘throw
away’ the elements of B that are not hit by arrows, then reversing the arrows
does define a function. Throwing away the elements not hit by arrows just
leaves the image of the function im f. So, if f : A → B is injective but not
surjective, then changing the codomain to im f gives a bijective function f :
A → im f. This bijective function does have an inverse f −1 : im f → A defined
by
f −1 (b) = a if and only if f (a) = b for all b ∈ im f.
This inverse function is illustrated in figure 3.28.

Exercises 3.5

1. For each of the following functions, determine whether or not the func-
tion is: (a) injective; (b) surjective.
 125

f f -1
A B A im f

im f

FIGURE 3.28: Defining an inverse when the function is not surjective.

(i) f : R → R, f (x) = 12 x − 7.
(ii) E = {2n : n ∈ Z} = {even integers}
f : Z → E, f (n) = 2n + 4.
(iii) R≥0 = {x ∈ R : x ≥ 0}
f : R≥0 → R≥0 , f (x) = |x − 1|.
(iv) A = {countries of the world}, B = {cities of the world}
f : A → B, f (X) = capital city of X.
(v) A = {countries of the world }, B = {cities of the world}
f : B → A, f (X) = country containing X.
 n

 if n is even
2
(vi) f : Z → Z, f (n) =
 n − 1 if n is odd.

2
{
n + 1 if n ≥ 0
(vii) f : Z → Z, f (n) =
n if n < 0.

(viii) A = P({1, 2, 3, 4})

f : A → A, f (X) = X ∩ {1, 2}.
(ix) A = P({1, 2, 3, 4})
f : A → A, f (X) = X ∪ {1, 2}.
(x) A is any non-empty set
f : A → P(A), f (a) = {a}.

2. Each of the following is the graph of a function A → B where A and B

are subsets of R.
Determine whether or not each function is: (a) injective; (b) surjective.
126

y y

B B
A
x x
A

(i) (ii)

y y

B B
A
x x
A

(iii) (iv)

y y

B x x
A A

(v) (vi)

3. Show that each of the following functions is a bijection and find its
inverse.
5x + 3
(i) f : R → R, f (x) = .
8
3x
(ii) f : R − {−1} → R − {3}, f (x) = .
x+1
(iii) f : [1, 3] → [−2, 2], f (x) = 2x − 4.
4
(iv) f : R+ → (0, 2), f (x) = .
x+2
(v) f : R2 → R2 , f (x, y) = (y, x).
(vi) f : R2 → R2 , f (x, y) = (2x − 1, 5y + 3).
 127
 n

 if n is even
2
(vii) f : Z+ → Z, f (n) =

 1 − n
if n is odd.
2
4. Let f : A → B be a function that is neither injective nor surjective.

(i) Show that it is possible to define a surjective function f : A → B ′

by replacing B with a suitable subset B ′ .
(ii) Show that it is possible to define an injective function f : A′ → B
by replacing A with a suitable subset A′ .
Chapter 4
The Structure of Mathematical Proofs

4.1 Introduction

In this chapter we introduce the key features of mathematical proof. It is

very hard, some would say impossible, to define precisely what is meant by
a mathematical proof as mathematicians themselves write them. The notion
of proof in certain logical systems may be defined quite precisely, but this
does not really describe what mathematicians do when constructing their
proofs. The following professional mathematician’s definition of proof, given
by Slomson [11], serves as a useful starting point.
A mathematical proof is a correct and convincing mathematical
argument. Whether an argument is correct is a matter of logic: the
conclusion must be a logical consequence of the premises. What
counts as convincing will vary from person to person, and has
changed with time.
In the following sections we will explain some of the general features that
make a ‘correct and convincing mathematical argument’. We will begin by
returning to the work we began in section 1.5 and looking at some example
proofs in rather more detail than is usual. From this we will extract some gen-
eral features that are common to most mathematical proofs and so develop an
informal framework for constructing proofs. In particular, we will explore the
deductive method that is at the heart of mathematical reasoning. We will also
consider how we approach universally quantified propositional functions of the
form ∀x • P (x) where the universe for x is infinite. Returning to Slomson’s
requirement that a proof must be convincing, we will also discuss what knowl-
edge the proof writer may assume the proof reader possesses. In section 4.4,
we describe the method of direct proof which underpins most mathematical
proofs. Our discussion in these sections will be geared towards developing an
understanding of the principal features of mathematical proofs as they are
usually written. In the final section, we explore in more depth Slomson’s cri-
terion for correctness: ‘the conclusion must be a logical consequence of the
premises’. We will develop a more formal framework for understanding what
is a proof, more closely based on the notion of formal proof that we considered

129
130

in section 2.5. Any mathematical proof will have a context and an intended
audience, which our formal framework will need to take into account.

4.2 Some proofs dissected

In this section we look in some detail at some proofs of elementary properties

of the positive integers. Our consideration of these proofs will be rather more
detailed than is usual for a mathematics text. For each example, we will pro-
vide both a proof and a commentary on the proof. In the following section we
will use these proofs to extract some general features and principals that will
apply to many proofs.
Building on the proof, given in section 1.5 (page 11), that the square of an
odd integer is odd, we begin by proving two similar results about evenness,
oddness and divisibility.

Theorem 4.1
The sum and difference of two odd integers are even.

Before we begin to prove this, we need to understand precisely what the

result is saying. Firstly, note that there is implicit universal quantification in
the proposition:

The sum and difference of any two odd integers are

always even.

We may express this more formally using universal quantification as:

For all integers m and n, if m and n are odd, then

both m + n and m − n are even.

Proof. Suppose that m and n are odd integers.

Then there exist integers a and b such that m = 2a + 1 and n = 2b + 1.
Hence m + n = (2a + 1) + (2b + 1) = 2a + 2b + 2 = 2(a + b + 1) where a + b + 1
is an integer. Therefore m + n is even.
Similarly, m − n = (2a + 1) − (2b + 1) = 2a − 2b = 2(a − b) where a − b is an
integer. Therefore m − n is even.
 131

Commentary
The original statement of the result, ‘the sum and difference of two odd inte-
gers are even’, did not obviously give the logical structure of the proposition.
We first needed to rewrite it so that the universal quantification became evi-
dent, ‘for all integers m and n, if m and n are odd, then both m + n and m − n
are even’. If we define the following predicates over the universe of integers

O : . . . is odd
E : . . . is even,

then we could symbolise the proposition as

∀m ∀n • (O(m) ∧ O(n)) ⇒ (E(m + n) ∧ E(m − n)).

In the proof, we assumed that m and n are odd, O(m) ∧ O(n), and from this
deduced that m + n is even and m − n is even, E(m + n) ∧ E(m − n).
In fact, we proved m + n is even separately from the proof that m − n is
even. Essentially, we gave two separate proofs: one for (O(m) ∧ O(n)) ⇒
E(m+n) and one for (O(m)∧O(n)) ⇒ E(m−n). However, from example 2.4.2
(page 36), we know that

[P ⇒ (Q ∧ R)] ≡ [(P ⇒ Q) ∧ (P ⇒ R)].

This means that, to prove a statement of the form P ⇒ (Q ∧ R), it is sufficient

to prove both P ⇒ Q and P ⇒ R, which is precisely what we have done in
this case.

Theorem 4.2
For all integers m and n, if m and n are odd, then m2 − n2 is divisible by 4.

Perhaps the most obvious proof of this statement would be to follow a similar
strategy to that used in the previous example. This would produce the follow-
ing proof (which starts in exactly the same way as the proof in the previous
example).

Proof 1. Suppose that m and n are odd integers.

Then there exist integers a and b such that m = 2a + 1 and n = 2b + 1.
Therefore m2 − n2 = (2a + 1)2 − (2b + 1)2
= (4a2 + 4a + 1) − (4b2 + 4b + 1)
= 4a2 + 4a − 4b2 − 4b
= 4(a2 + a − b2 − b) where a2 + a − b2 − b ∈ Z.

Hence m2 − n2 is divisible by 4.
132

This proof is perfectly correct and, we hope, easy to follow. There is an al-
ternative proof, however, that builds on the result of example 4.1 above. We
give this alternative proof below.

Proof 2. Suppose that m and n are odd integers.

Then both m + n and m − n are even, so there exist integers a and b such
that m + n = 2a and m − n = 2b. Therefore

m2 − n2 = (m − n)(m + n) = (2a)(2b) = 4ab where ab ∈ Z.

Hence m2 − n2 is divisible by 4.

Commentary
There are several observations to make regarding these proofs. Perhaps the
most obvious is that this example indicates that proofs are not unique. There
will frequently be several essentially different ways of proving a result; that
is, proofs which differ in approach or substance, and not just in the style of
presentation. Which of several proofs is to be preferred will be a matter of
judgement and will depend on various factors. For example, some people will
prefer shorter, more sophisticated proofs that may be harder to understand;
others may prefer proofs that are perhaps longer but do not rely on so much
knowledge on the part of the reader.
The two proofs given above illustrate this. The first proof is more elementary
in the sense that it assumes nothing other than the meaning of even and odd
and some algebraic manipulation. The second proof is shorter and, one might
argue, more elegant but relies on the results proved in the previous example
and so is assuming greater ‘background knowledge’ on behalf of the reader.
In fact, this occurs all the time — the proofs of some results rely on other
propositions that need to have been proved earlier. We shall explore this idea
— that proofs build on proofs that in turn build on other proofs etc. — in
more detail in section 4.5 below.
The prime factorisation theorem says that every integer greater than 1 can
be factored into prime numbers, and is part of the so-called ‘Fundamental
Theorem of Arithmetic’. The full Fundamental Theorem goes on to state that
the factorisation for a given integer is unique apart from the ordering of the
prime factors. However, the proof of the uniqueness part is more sophisticated
and can wait until later (see chapter 7).

Theorem 4.3 (The Prime Factorisation Theorem)

Every integer greater than 1 can be expressed as a product of prime numbers.

Before embarking on a proof of the theorem, we need to understand precisely

 133

what a prime number is. A simple definition is that a prime number is an

integer greater than 1 which is not divisible by any positive integer except 1
and itself. Thus 5 is a prime number since it is not divisible by any positive
integer except 1 and 5, whereas 6 is not a prime number since it is divisible
by 2, for instance, which is different from both 1 and 6 itself. (Notice that,
according to the definition, 1 is not a prime number.)
Now that we understand the term ‘prime number’, we could embark on a
search for a proof of the theorem. It is usually best though to ensure that we
first understand thoroughly what the theorem is really saying. Often the most
effective way of achieving this is to look at some examples.
Consider the integer 24. We can write 24 = 2 × 12, which expresses 24 as the
product of a prime number (2) and a non-prime number (12). Since 12 is not
prime, we can now look for its factors. We continue in this way as follows:

24 = 2 × 12
=2×3×4
= 2 × 3 × 2 × 2.

We have now expressed 24 as a product of prime numbers: 2 × 3 × 2 × 2. Of

course, there are other such expressions, for example 2 × 2 × 2 × 3, but this
just contains the same prime numbers written in a different order.
We may carry out this process on any positive integer. For example, carry-
ing out the process on 1234 567 890 takes rather longer and is trickier but
eventually produces the following expression:

1234 567 890 = 10 × 123 456 789

= 2 × 5 × 3 × 41 152 263
= 2 × 5 × 3 × 3 × 13 717 421
= 2 × 3 × 3 × 5 × 3607 × 3803.

In this case, the last step is the hardest: factorising 13 717 421 = 3607 × 3803
and knowing that these factors are prime so that no further factorisation is
possible. The theorem says that we can obtain such an expression for any
integer bigger than 1.
Now that we have some intuitive ‘feel’ for the theorem, we can begin the search
for a proof. In fact, the basis of the proof is already contained in the examples
above. There, for example, we found the prime factors of 24 by first finding
two factors (2 and 12) and then finding factors of these where possible, and
so on. Since this process can be applied to any integer greater than 1, we can
construct a proof of the general result. (A shorter, more sophisticated proof of
this result will be given later — see chapter 8.) We can now give an informal
proof of the Prime Factorisation Theorem.
134

Proof. Let n be any integer greater than 1. If n is prime then there is nothing
to prove as n itself is already expressed as a ‘product’ of primes, albeit in a
rather trivial way.
If n is not prime, then there exist factors n1 and n2 , each greater than 1, such
that n = n1 × n2 .
Now consider n1 and n2 in turn. If n1 is composite (that is, not prime),
then it can be expressed as a product of two integers each greater than 1, say
n1 = m1 ×m2 . Similarly, either n2 is prime or it can be expressed as a product
n2 = m3 × m4 where m3 and m4 are greater than 1. At this stage we have
expressed n in one of the following four ways:

n = n1 × n2 (if both n1 and n2 are prime),

n = m 1 × m 2 × n2 (if n1 is composite and n2 is prime),
n = n1 × m3 × m4 (if n1 is prime and n2 is composite),
n = m1 × m2 × m3 × m4 (if n1 and n2 are both composite).

Next consider each mi in turn and continue the process. At every step in
the process, each factor is either prime or is split into two smaller factors.
Therefore, this ‘subdivision’ process must eventually stop. When the process
stops, the result is an expression of the form

n = p1 × p2 × . . . × pk

where each pi is prime. Therefore we have shown that n can be expressed as

a product of primes.

Commentary
Our treatment of this theorem and its proof is more detailed than is usual for
a mathematics text. There are several reasons for this. One is our desire to
give some indication at least of how the proof might be discovered, rather than
just presenting the proof itself. If we are to learn how to construct proofs, it is
clearly desirable to gain some insight into how a proof evolves from underlying
ideas. It will not be sufficient just to study completed proofs. Another reason is
to indicate the importance of precisely defined terms, such as ‘prime number’.
Perhaps the most important lesson to learn from this example is that any
mathematical proof is an exercise in communication. A correct but incompre-
hensible proof is of little use to anyone (and we hope our proof does not fall
into that category). In writing proofs, clarity and comprehensibility, as well
as correctness, are important goals.
 135

Exercises 4.1
1. Prove each of the following.
(i) The sum of any two consecutive integers is odd.
(ii) For all integers n, if n is prime, then n2 + 5 is not prime.
(iii) For all integers n, if n − 2 is divisible by 4, then n2 − 4 is divisible
by 16.
2. (i) Prove that the sum of any three consecutive positive integers is
divisible by 3.
(ii) Prove that the sum of any five consecutive positive integers is di-
visible by 5.
3. (i) Prove that the product of any three consecutive positive integers
is divisible by 6.
(ii) Prove that the product of any five consecutive positive integers is
divisible by 24.
4. (i) Prove that the sum of the squares of any two consecutive positive
integers is odd.
(ii) Prove that the sum of the squares of any three consecutive positive
integers is one less than a multiple of 3.

4.3 An informal framework for proofs

In this section we give an informal description of one of our principal proof
techniques, the method of direct proof. By looking in a bit more detail at
simple examples of proofs, we identify some general features that will appear
in many other proofs. In other words, we will extract some of the essential
structure of our examples that will provide a general framework for many
proofs.
To begin with, we look a little more closely at the structure of one half of the
proof in theorem 4.1. We repeat the result and its proof below.

Theorem 4.1
The sum of two odd integers is even.

Proof. Suppose that m and n are odd integers.

Then there exist integers a and b such that m = 2a + 1 and
n = 2b + 1.
Hence m+n = (2a+1)+(2b+1) = 2a+2b+2 = 2(a+b+1)
where a + b + 1 is an integer.
Therefore m + n is even.
136

Recall that the statement being proved is a universally quantified propositional

function ‘for all integers m and n, if m and n are odd, then m + n is even’. In
symbols,
∀m ∀n • (O(m) ∧ O(n)) ⇒ E(m + n)
where the universe is the integers, O denotes the predicate ‘. . . is odd ’, and E
denotes the predicate ‘. . . is even’.
There are two important features of our proof.

• To prove the universally quantified statement, we actually proved

(O(m) ∧ O(n)) ⇒ E(m + n) for arbitrary integers m and n. By ‘ar-
bitrary’ we mean that m and n were not assumed to have any special
characteristics (such as being equal to 347, being greater than 17 or be-
ing prime or whatever) other than the property of ‘being an odd integer’.
In other words, m and n could have been any odd integers.
Although the statement was universally quantified, we did not prove it
for every odd integer m and every odd integer n individually. Clearly,
this is impossible as there are infinitely many odd integers. Instead, we
proved this for arbitrary ‘representative’ odd integers m and n.
This distinction between universally quantified variables ‘∀m’ and ‘∀n’
on the one hand and arbitrary integers m and n on the other, is quite
subtle. In part, the distinction is masked by the fact that we are using
the same letter to denote both the universally quantified variable and
its arbitrary representative. Some people prefer to make this distinction
more explicit by using different letters, as follows.

Theorem 4.1
For all integers m and n, if m and n are odd, then m + n is
even.

Proof. Suppose that k and ℓ are odd integers.

Then there exist integers a and b such that k = 2a + 1 and
ℓ = 2b + 1.
Hence k + ℓ = (2a + 1) + (2b + 1) = 2a + 2b + 2 = 2(a + b + 1)
where a + b + 1 is an integer.
Therefore k + ℓ is even.

• The second feature is that, to prove the conditional statement

if m and n are odd, then m + n is even
(for our arbitrary integers m and n), we assumed m and n were odd
and from this we made a sequence of deductions concluding with the
statement m + n is even. Symbolically, to prove
(O(m) ∧ O(n)) ⇒ E(m + n)
 137

we assumed the antecedent O(m)∧O(n) and, from this, made a sequence

of deductions to establish the consequent E(m + n).

These two features of our proof are examples of general principles which we
now describe.

Principle of Universal Generalisation

To prove a universally quantified statement of the form ∀x•P (x), it is sufficient
to prove P (k) where k is an arbitrary element of the universe for x.

This principle allows us to prove statements over infinite universes. If a propo-

sitional function is universally quantified over a finite universe, it is possible in
principle to prove it by verifying the statement individually for each element
of the universe. Of course, if the universe is finite but very large, this approach
may not be feasible.
For example, suppose we wish to prove

if m and n are odd then m + n is even

over the universe for m and n of positive integers less than 6. We could do
this as follows.

Proof. Let: n=1 m = 1: then m + n = 2 which is even;

n=1 m = 3: then m + n = 4 which is even;
n=1 m = 5: then m + n = 6 which is even;
n=3 m = 1: then m + n = 4 which is even;
n=3 m = 3: then m + n = 6 which is even;
n=3 m = 5: then m + n = 8 which is even;
n=5 m = 1: then m + n = 6 which is even;
n=5 m = 3: then m + n = 8 which is even;
n=5 m = 5: then m + n = 10 which is even.

This verifies the result for each pair (m, n) where m and m are odd positive
integers less than 6, so the result follows.

When the universe is infinite, this approach is clearly impossible and the
Principle of Universal Generalisation is required. The crucial point to note
about the arbitrary element k of the universe is not whether it is labelled n
or k, but rather that it is not assumed to possess any attribute that is not
possessed by every element of the universe. In other words, it is a ‘general’
element of the universe.
138

The second feature of our proof that is a general principle is that we began with
an initial proposition P and then made a sequence of deductions culminating
in a final proposition Q. We call this the ‘deductive method’. Essentially,
this says that mathematical proofs use deductive reasoning as described in
section 1.4 rather than inductive reasoning, described in section 1.3, which is
the basis of most human knowledge and belief.

Deductive Method
A proof starts with some initial statement or statements and proceeds by
making a sequence of deductions that follow logically until a final statement
is obtained.

As we shall see, different types of proof have different initial and final state-
ments. The important aspect of the deductive method is that a proof comprises
a sequence of statements that follow logically from earlier statements in the
proof. Later in this chapter, we will examine in more detail what we mean by
‘follows logically’; see section 4.5.
There is one further important observation that we wish to make regarding our
proof of theorem 4.1. Although the proof itself is straightforward, it assumes
a certain familiarity with the integers and their properties. In particular, the
following properties are among those assumed, either explicitly or implicitly.

• The oddness property.

An integer n is odd if and only if it can be expressed as n = 2a + 1 for
some integer a.
• The evenness property.
An integer n is even if and only if it can be expressed as n = 2a for some
integer a.
• Associativity and commutativity of addition.
By equating (2a + 1) + (2b + 1) with 2a + 2b + 2, we are assuming that:
◦ we can ignore the brackets and regroup terms as necessary so that
we can rewrite (2a + 1) + (2b + 1) as 2a + (1 + 2b) + 1
◦ we can equate 1 + 2b with 2b + 1.
Of these assumptions, the first relies on the associative property of ad-
dition
for all a, b, c ∈ Z, a + (b + c) = (a + b) + c,
and the second relies on the commutative property of addition
for all a, b ∈ Z, a + b = b + a.
 139

• Distribution law of multiplication over addition.

By equating 2a + 2b + 2 with 2(a + b + 1) we are also assuming that
‘factorisation works’. This relies on the distributive property of multipli-
cation over addition:

for all a, b, c ∈ Z, a(b + c) = ab + ac.

Of course, we would not normally consider a simple algebraic re-arrangement

(2a+1)+(2b+1) = 2a+2b+2 = 2(a+b+1) in such minute detail. Indeed, we
would normally complete the re-arrangement automatically without giving a
second thought to the algebraic processes and steps required. This is because
we are very familiar with the basic properties of addition and multiplication of
integers, so much so that they feel like ‘second nature’ to us. However, for the
current discussion, the important point is not the details of which properties
of the integers we are using, but rather the fact that there are some properties
that the proof uses.
This is quite typical. In almost any proof, there will be properties about the
system under consideration that will be assumed implicitly but not referred to
explicitly. The previous discussion highlighted some properties of the ‘system’
of integers (and their addition and multiplication) that were implicit in the
proof in theorem 4.1. We refer to all of these properties as our ‘background
knowledge’ of the system under consideration.

Background knowledge
In any proof, the properties of the system under consideration that are as-
sumed implicitly but not referred to explicitly, will be referred to as our back-
ground knowledge of the system.

By its very nature, this background knowledge is somewhat vague and ill-
defined and it will vary from context to context. For example, two specialists
in a particular branch of mathematics will have a very extensive background
knowledge of their field that will not be shared by a layperson or novice in
the field. Although it is ill-defined, it is nevertheless important. A proof will
only be understandable by a reader if they share the background knowledge
that the proof writer assumed when writing the proof.
Anyone writing a proof will need to consider how much background knowl-
edge they may assume or wish to assume. This will depend on who is the
intended audience for the proof as well as the preferences of the proof writer.
In section 4.2, we gave two different proofs of theorem 4.2.

For all integers m and n, if m and n are odd, then m2 − n2

is divisible by 4.
140

The first proof assumed less prior knowledge than the second proof, but was
longer and, perhaps, less elegant. This example illustrates in a small way
that there is often a trade-off between ‘elementary’ proofs, those that assume
less background knowledge, and more ‘sophisticated’ proofs. Since elementary
proofs have less to draw on, they can sometimes be longer and more complex
in their reasoning. Set against this, there is a potential danger with more
sophisticated proofs in that the reader may not possess all the knowledge that
is assumed of them.
We conclude this section with a brief consideration of how the previous dis-
cussion relates to the following proof in set theory. In this example, we first
prove the following result about sets.

Theorem 4.4
For all sets A and B, A ∩ B ⊆ A ∪ B.

Proof. Let A and B be sets.

Let x ∈ A ∩ B. Then x ∈ A and x ∈ B. Hence it follows that x ∈ A or x ∈ B,
so x ∈ A ∪ B.
We have shown that: x ∈ A ∩ B ⇒ x ∈ A ∪ B. Therefore A ∩ B ⊆ A ∪ B.

How does this example relate to our previous discussion? The first point to
note is the use of the Principle of Universal Generalisation. The statement to
be proved is universally quantified (‘for all sets A and B . . . ’) and the proof
starts by letting A and B be arbitrary sets. In other words, although it is
not mentioned explicitly, the proof implicitly used the Principle of Universal
Generalisation.
Secondly, the background knowledge used in the proof draws on section 3.3.
In particular, it explicitly uses the meaning of

• intersection: x ∈ X ∩ Y if and only if x ∈ X and x ∈ Y ;

• union: x ∈ X ∪ Y if and only if x ∈ X or x ∈ Y ;
• subset: X ⊆ Y if and only if, for all x, x ∈ X ⇒ x ∈ Y .

Thirdly, the proof follows the deductive method. The initial statements are
‘A is a set’, ‘B is a set’, and ‘x ∈ A ∩ B’. Deductions are then made firstly to
obtain the statement x ∈ A ∪ B and then, using the background knowledge
of the meaning of subset, the final statement A ∩ B ⊆ A ∪ B is deduced.
In summary, the framework for proofs that we have developed has the following
components and each of our proofs has exhibited these.
 141

• Principle of Universal Generalisation: to prove a universally quan-

tified statement ∀x • P (x), the statement P (k) is proved for an arbitrary
element of the universe k.

• Use of background knowledge: in any proof, certain properties of

the system under consideration will be assumed, implicitly or explicitly.

• Deductive method: a proof starts with some initial statement or state-

ments and proceeds by making a sequence of deductions that follow
logically until a final statement is obtained.

Exercises 4.2

1. For each of the proofs in exercise 4.1:

(a) identify how the proof uses the Principle of Universal Generalisa-
tion;
(b) identify explicitly some of the background knowledge that is used
in the proof;
(c) describe how the proof uses the deductive method — what are the
initial statements and what is the final statements?

4.4 Direct proof

In the previous section we described a proof as following the deductive method:

start with an initial statement or statements and make a sequence of deduc-
tions until a final statement is obtained. However, we did not describe what
comprises the initial and final statements. In this section, we consider this
in more detail and describe the most common structure of proof, known as
‘direct proof’. In fact, each of the proofs given earlier in this chapter have
been direct proofs.

We begin by looking again at theorem 4.1, the first theorem and proof in
section 4.3. The following summarises the overall structure of the proof but
omits the details of the actual deductions made.
142
Theorem 4.1
For all integers m and n, if m and n are odd, then m + n is
even.

Proof. Suppose that m and n are odd integers.

..
.
Therefore m + n is even.

The structure of the proof is to assume that m and n are arbitrary odd integers
— these are the initial statements — and to deduce that m+n is even — this is
the final statement. With the universe being the integers and with predicates
O for ‘odd’ and E for ‘even’, the statement being proved is the conditional

(O(m) ∧ O(n)) ⇒ E(m + n).

The structure of the proof was to assume the antecedent O(m) ∧ O(n) and
deduce the consequent E(m + n). This general structure is called the ‘method
of direct proof’, which we now define.

Method of Direct Proof (for P ⇒ Q)

To prove a proposition of the form P ⇒ Q, it is sufficient to assume P and,
from this, deduce Q.

It is easy to see why this method works. Recall the truth table for the condi-
tional P ⇒ Q, given on page 21 and repeated below.

P Q P ⇒Q
T T T
T F F
F T T
F F T

The only situation when P ⇒ Q is false is when P is true and Q is false, so

if we wish to establish the truth of P ⇒ Q, it is this case that we must rule
out. In other words, we must show that when P is true, Q cannot be false. We
do not need to consider the situation when P is false because P ⇒ Q is true
in this case, regardless of the truth or falsity of Q. Assuming P to be true at
the beginning of the proof has the effect of restricting attention to the first
two rows of the truth table. Deducing that Q is true in this case, rules out
the second row as a possibility and hence shows that the conditional P ⇒ Q
is true.
In section 4.2, we gave two different proofs of the following theorem.
 143

Theorem 4.2
For all integers m and n, if m and n are odd, then m2 − n2 is divisible by 4.

The statement being proved is a conditional P ⇒ Q where P is ‘m and n

are odd’ and Q is ‘m2 − n2 is divisible by 4’. Both proofs followed the same
structure, described as follows.

Suppose that m and n are odd integers.

..
.
Therefore m2 − n2 is divisible by 4.

Each proof assumed P and from this deduced Q; in other words, each proof
was a direct proof. The two proofs differed in the sequence of deductions used
to obtain Q from P , but not in their overall structure.
We have also seen two other proofs, in theorems 4.3 and 4.4, where the state-
ments being proved are not naturally expressed as a conditional, P ⇒ Q.

• Prime Factorisation Theorem 4.3: every integer greater than 1 can be

expressed as a product of prime numbers.
• Theorem 4.4: for all sets A and B, A ∩ B ⊆ A ∪ B.

The method of direct proof of a conditional, assume P and then deduce Q,

does not really apply in these cases as the proposition being proved is not
(obviously) of the form P ⇒ Q. If we examine those proofs, we see that
they start by considering arbitrary elements of the universe and then assum-
ing some piece of background knowledge about the appropriate system: the
integers or sets. In the case of the Prime Factorisation Theorem, the proof
considers an arbitrary integer greater than 1 and starts with the property
that every such integer is either prime or can be factorised in some way. For
the set theory result, the background knowledge that gets the proof going is
the knowledge that, to prove that one set is a subset of another, we must
show that every element of the first set is an element of the second set. So
each of these proofs have the same overall structure: assume some background
knowledge and proceed to make deductions until the final desired conclusion
is reached. We still regard these as direct proofs because we proceed directly
to the appropriate conclusion.
In each of these cases, the statement of the theorem is of the general form
∀x • Q(x) or ∀x ∀y • Q(x, y). The proofs also apply the Principle of Universal
Generalisation which reduces the statement to be proved to a proposition Q
involving an arbitrary element or elements of the appropriate universe. In
each case, the proof begins with some background knowledge and proceeds by
making deductions until the statement Q is obtained. So each of the proofs
followed the following method of direct proof but started with some aspect of
144

background knowledge rather than the antecedent P of a conditional P ⇒ Q.

We can summarise this as follows.

Method of Direct Proof (for Q)

To prove the proposition Q, it is sufficient to assume appropriate background
knowledge and, from this, deduce Q.

The distinction we have made between theorems expressed as conditional

statements P ⇒ Q and those that are expressed as a non-conditional propo-
sition Q is somewhat less clear cut than we have indicated. Suppose we
have a proposition of the form ∀x • Q(x) where x belongs to some universe
U . By making the universe explicit, we could express this as a conditional
∀x • x ∈ U ⇒ Q(x). For example, theorem 4.4 could have been expressed as
a conditional as follows: ‘if A and B are sets, then A ∩ B ⊆ A ∪ B’.
We now look at some further examples of direct proofs of the two different
types; that is, direct proofs of P ⇒ Q and direct proofs of Q.

Example 4.1
In this example, we consider the interaction between composition of functions,
described in section 3.5, with the properties of injectivity and surjectivity,
defined in section 3.6. Two obvious questions to ask are:
Is the composite of two injective functions an injective function?
Is the composite of two surjective functions a surjective function?
Let f and g be two functions such that the composite g ◦ f is defined —
figure 3.18 on page 114 illustrates the setup. Recall that, informally, a function
is injective if different elements always have different images. Suppose this is
true for both f and g and consider the composite g ◦ f , which has rule ‘apply
f then apply g’. Starting with different elements, applying f gives different
elements since f is injective; then applying g still gives different elements
since g is injective. This suggests that the composite of two injections is also
an injection.
Now suppose that f and g are surjective. Then each element of their respec-
tive codomain is the image of some element in the domain; informally, every
element of the codomain is ‘hit by an arrow’ coming from some element of
the domain. Again consider g ◦ f . The codomain of g ◦ f is the same as the
codomain of g. Since g is surjective, every element of this codomain is hit by
a ‘g-arrow’. But the starting point of this g-arrow lies in the codomain of f ,
so it is hit by an ‘f -arrow’ as f is surjective. Putting the two arrows together,
we see that every element of the codomain of g ◦ f is hit by a composite ‘f
followed by g arrow’. This suggests that the composite of two surjections is
also an surjection.
The following theorem summarises the preceding discussion.
 145

Theorem 4.5
Let f : A → B and g : B → C be two functions with composite function
g ◦ f : A → C.

(i) If f and g are both injective then so, too, is g ◦ f .

(ii) If f and g are both surjective then so, too, is g ◦ f .

Proof.
(i) Suppose that f : A → B and g : B → C are both injective functions.
Let a1 , a2 ∈ A and let b1 = f (a1 ) and b2 = f (a2 ). To show that g ◦ f is
injective, we need to show that, if (g ◦ f )(a1 ) = (g ◦ f )(a2 ), then a1 = a2 .
Now
(g ◦ f )(a1 ) = (g ◦ f )(a2 )
⇒ g(f (a1 )) = g(f (a2 ))
⇒ g(b1 ) = g(b2 ) (since f (a1 ) = b1 , f (a2 ) = b2 )
⇒ b1 = b2 (since g is injective)
⇒ f (a1 ) = f (a2 ) (since f (a1 ) = b1 , f (a2 ) = b2 )
⇒ a1 = a2 (since f is injective).

Hence g ◦ f is injective.

(ii) Suppose that f : A → B and g : B → C are both surjective functions.

Let c ∈ C. To show that g ◦ f is surjective, we need to show that
(g ◦ f )(a) = c for some a ∈ A.
Since g is surjective, there exists b ∈ B such that g(b) = c. Also, f is
surjective, there exists a ∈ A such that f (a) = b. Therefore, there exists
a ∈ A such that

(g ◦ f )(a) = g(f (a)) = g(b) = c.

Hence g ◦ f is surjective.

Commentary
Each of the two parts of the proof is a direct proof of a conditional P ⇒ Q.
We started each proof by assuming P : f and g are both injective or both
surjective functions. Then each proof made a number of deductions to obtain
Q: g ◦ f is injective or surjective.
Note also that each proof uses implicitly the Principal of Universal Generalisa-
tion: each part proves the result for arbitrary injective or surjective functions,
146

as appropriate. Indeed, as we have seen previously, even the universal quan-

tification in the statement of the theorem is implicit. The structure of the
results is as follows, although the universal quantification was not explicit in
the statement of the theorem:

for all functions f and g, if f and g are injective/surjective,

then g ◦ f is injective/surjective.

Finally, the principal background knowledge used in each case is the meaning
of the terms ‘injective’ and ‘surjective’, both in terms of what we may assume
about f and g and also the property that must be established for g ◦ f .

Examples 4.2 (Properties of the modulus function)

The modulus |x| of a real number x ∈ R is defined by
{
x if x ≥ 0
|x| =
−x if x < 0.

This defines a function | · | : R → R, x 7→ |x| which has graph given in

figure 4.1.

f ( x)

f ( x) = x

FIGURE 4.1: The modulus function.

The modulus function is important in mathematics. For example, it is an

important part of definitions (and hence proofs) of limits in real analysis; we
will consider some examples of these proofs in section 5.4. In the following
examples, we will explore some of the properties of the modulus function.

1. The first property we will prove is the following inequality, known as the
triangle inequality for the modulus function.

Theorem 4.6
For all x, y ∈ R, |x + y| ≤ |x| + |y|.

Proof. Without loss of generality, we may assume that x ≤ y.

 147

Since the values of |x| and |y| depend on whether x and y are positive
or negative, we will split the proof into four cases:
Case 1: 0 ≤ x ≤ y
Case 2: x < 0 ≤ y and x + y ≥ 0
Case 3: x < 0 ≤ y and x + y < 0
Case 4: x ≤ y < 0

Case 1. In this case, |x| = x, |y| = y and, since x + y ≥ 0, we also have

|x + y| = x + y. Therefore

|x + y| = x + y = |x| + |y|,

so the result holds in this case.

Case 2. In this case |x| = −x, which means that x < 0 < −x. Also
|y| = y and |x + y| = x + y. Therefore

|x + y| = x + y < −x + y = |x| + |y|,

so the result holds in this case.

Case 3. Again we have |x| = −x. Also |y| = y, which means that
−y < 0 < y. Since x + y is negative, we have |x + y| = −(x + y).
Therefore

|x + y| = −(x + y) = −x − y < −x + y = |x| + |y|,

so the result holds in this case.

Case 4. In this case we have |x| = −x and |y| = −y. Since x + y < 0
also, we have |x + y| = −(x + y). Therefore

|x + y| = −(x + y) = −x − y = |x| + |y|,

so the result holds in this case.

We have shown that the result holds in all four cases. Hence, for all
x, y ∈ R, |x + y| ≤ x + y.

Commentary
The opening phrase, ‘without loss of generality we may assume that . . . ’,
is quite common in proofs. What this means is that there are several
cases — two in our example — where the reasoning for each case is
essentially the same. Therefore we will present the reasoning in only one
of the cases. In our example, the two cases are x ≤ y and y ≤ x. Since
the statement we are proving, |x + y| ≤ |x| + |y|, is symmetric in x and
148

y, we could obtain the proof in the case y ≤ x from the proof when
x ≤ y simply by swapping the roles of x and y. Hence there is no loss of
generality by making the additional assumption x ≤ y.
Since the definition of the modulus function involves two cases (x ≥ 0
and x < 0), the proof naturally splits into four cases. These comprise
three ‘main’ cases, one of which has two ‘sub-cases’, as illustrated in
figure 4.2. To complete the proof, we need to show that the required
result holds in each of the four cases. The argument in each case is
straightforward and just requires careful consideration of which terms
are positive and which are negative.

0£ x£ y x<0£ y x£ y<0

x+ y ³0 x+ y <0

x+ y £ x + y

FIGURE 4.2: Illustrating the four cases in the proof of |x + y| ≤ |x| + |y|.

When it is sensible to split a proof into cases is often a matter of judge-

ment. On the one hand, the individual cases themselves are usually more
straightforward to deal with because we will have made some additional
assumptions for each of the cases. Balanced against this is the additional
work required to prove the result for each case separately.
2. In this example, we will explore the relationship between the values of
the modulus of various expressions. There are (at least) two reasons for
doing so. Firstly, a good facility for arguments of this kind is essential
for an understanding of those parts of real analysis that consider conver-
gence of sequences, limits of functions, continuity and differentiability
of functions, and so forth. From our perspective, however, a second rea-
son for considering these examples, is that they illustrate the interplay
between universal and existential quantification that we considered in
section 2.4.
(i) The first result we consider is the following.
There exists a real number δ > 0 such that
 
|x + 1| < δ ⇒ 3x2 − x − 4 < 1.
 149

The first point to note here is that, although the existential quan-
tification is explicit, there is a hidden universal
 quantification:
 it is
implicit that the implication |x + 1| < δ ⇒ 3x2 − x − 4 < 1 holds
for all real numbers x.
We can think of the proposition
 as asserting that we may control
the size of 3x2 − x − 4, in the sense of ensuring it is less than 1,
provided we are permitted to control the size of |x + 1|, in the sense
of ensuring that it is less than some specified δ. So, we begin by
considering the expression that we are trying to control,
 2 
3x − x − 4 = |(x + 1)(3x − 4)| = |x + 1| |3x − 4|.

Here we have first factorised the expression and then used the prop-
erty, |ab| = |a||b|, whose proof we leave as an exercise; see exer-
cise 4.3.1. Of these two terms, we have control over |x + 1| as we
can ensure that it is less than some δ > 0, but we need to consider
|3x − 4| more carefully. The ‘trick’ is to rewrite 3x − 4 in terms of
x + 1 and then apply the triangle inequality proved in the previous
example, as follows:

|3x − 4| = |3(x + 1) − 7|
≤ |3(x + 1)| + | − 7| (by the triangle inequality)
= 3|x + 1| + 7. (simplifying)

Recall that we are trying to establish the existence of δ > 0 such

that
 
|x + 1| < δ ⇒ 3x2 − x − 4 < 1.

Suppose, for a moment, that δ = 1. Then, for |x + 1| < 1, we have

|3x − 4| ≤ 3|x + 1| + 7 < 3 + 7 = 10.

From this we have,

 2 
3x − x − 4 = |x + 1| |3x − 4| < 10|x + 1| (since |3x − 4| ≤ 10),

which will be less than 1 provided |x + 1| < 1/10. So if we now take

δ = 1/10, then both the required inequalities, |3x − 4| ≤ 10 and
|x + 1| < 1/10, will hold when |x + 1| < δ. We can now piece this
together and write as a coherent proof of the result.
There exists a real number δ > 0 such that
 
|x + 1| < δ ⇒ 3x2 − x − 4 < 1.
150

Proof. Let δ = 1
10 . For all x satisfying |x + 1| < δ, we have
 2 
3x − x − 4 = |(x + 1)(3x − 4)|

= |x + 1| |3x − 4|
= |x + 1| |3(x + 1) − 7|
≤ |x + 1|(3|x + 1| + 7) (triangle inequality)
< 10|x + 1| (since |x + 1| < δ < 1)
≤ 10 × 1
10 =1 (since |x + 1| < δ = 1
10 ).

Hence
 2 we have shown that, when δ = 1
10 , |x + 1| < δ ⇒
3x − x − 4 < 1.

Commentary
The structure of the statement being proved is an existentially
quantified proposition: ∃δ • P (δ). (We also note that P (δ) involved
a universal quantification over x, but that does not alter the over-
all form of the statement as an existentially quantified proposi-
tion.) We proved the proposition by finding an explicit value for δ
(δ = 1/10) which ‘worked’. The bulk of the proof is concerned with
showing that the chosen value of δ has the required property. Later,
in chapter 7, we will see occasions where existence of an object is
established without an explicit value being found.
This example also illustrates that the process of finding a proof
is sometimes quite different from how the proof is eventually pre-
sented. In this case, the realisation that δ = 1/10 is sufficient comes
at the end of what we might call the ‘discovery process’. However,
in the proof itself, the choice of δ = 1/10 comes right at the begin-
ning.
It is also worth noting that the choice of δ = 1/10 is not unique.
Recall that we established
 2 
3x − x − 4 ≤ |x + 1|(3|x + 1| + 7).

Suppose, in the bracket, we chose |x + 1| ≤ 1/3. Then 3|x + 1| + 7 ≤

8,
 so that 3x 2 
 − x − 4 ≤ 8|x + 1| and we can then ensure that
3x − x − 4 < 1 by ensuring, in addition, that |x + 1| ≤ 1/8. To
2

ensure both inequalities hold, we require δ < 1/3 and δ < 1/8.
Therefore, an alternative choice for δ which makes the proof work
is δ = 1/8.
(ii) In this example we consider the following result.
 151
For all ε > 0, there exists δ > 0 such that
 
|x + 2| < δ ⇒ x2 − 4 < ε.
This is similar to the previous example, but with another level of
quantification.
 In the previous example we were required to show
that 3x2 − x − 4 was less than some specific value,
 namely
 1, by
choosing |x + 1| < δ. Now we have to show that x2 − 4 is less than
any arbitrary ε > 0 again by choosing |x + 2| < δ.
{ ε}
Proof. Given ε > 0, let δ = min 1, .
5
Then δ > 0 and |x + 2| < δ implies both that |x + 2| < 1 and that
ε
|x + 2| < . Now
5
 2 
x − 4 = |(x + 2)(x − 2)|
= |x + 2| |(x + 2) − 4|
≤ |x + 2|(|x + 2| + 4) (triangle inequality)
< 5|x + 2| (since |x + 2| < 1)
ε ε
≤ 5× =ε (since |x + 2| < ).
5 5

Hence
 2  have shown that, for the given choice of δ, |x + 2| < δ ⇒
we
x − 4 < ε.

Commentary
This time we did not show the discovery process. Hence the choice
of δ = min {1, ε/5} at the start of the proof seems to come out of
nowhere.
What we actually did was to work through the first part of the
reasoning:
 2 
x − 4 = |(x + 2)(x − 2)|
= |x + 2| |(x + 2) − 4|
≤ |x + 2|(|x + 2| + 4) (triangle inequality).

We then realised that, provided |x + 2| < 1, the term in the bracket

would be less than 5. Hence we could then ensure the whole ex-
pression was less than ε provided, in addition, that |x + 2| < ε/5.
To ensure that both of these steps work, we then recognised that
we could choose δ = min {1, ε/5}. In other words, this choice of δ
comes after a careful discovery process even though, in the proof
itself, this is not shown.
It is also worth remarking on the structure of the statement in this
152

example. As we noted at the start, it is similar to the previous ex-

ample but with an extra quantification: ∀ε ∃δ • P (ε, δ) where the
universe of discourse for both ε and δ is the positive real numbers.
As noted in section 2.4 (see examples 2.12, questions 2 and 3), the
order of quantification is significant. In this case, because the uni-
versal quantification of ε comes before the existential quantification
of δ, the value of δ may depend on ε. In other words, for each value
of ε there must be an appropriate value of δ, but different values
of ε may have different values of δ.

A case study: metric spaces

In the next few pages, we introduce a topic that, we suspect, will be unfamiliar
to many readers. Our aim here is to show how a new piece of abstract math-
ematics is built from a few basic assumptions together with what we might
call a ‘standard toolkit of mathematics’. The basic assumptions will usually
be chosen to capture key properties that are common to many examples, pos-
sibly in diverse areas, that will provide a common framework worthy of study.
We do not wish to specify precisely what is contained in the ‘standard toolkit
of mathematics’, but it will include elements of propositional and predicate
logic, some set theory, the simple properties of integers, rational numbers and
real numbers, the properties of ‘equals’ and inequalities, and the idea of func-
tions and their basic properties. For the purposes of this discussion, we will
assume the material covered in chapters 2 and 3. Those readers who wish to,
may omit this discussion and go straight to section 4.5.
We aim to define a ‘metric space’ that will capture different circumstances in
which we have some space where we can measure distance. In other words,
we may think of a metric space as providing a general framework for ‘space
where we can measure distances’. An important observation is that the notion
of ‘distance’ — or ‘metric’ — is not universal but will vary according to context
as the following examples illustrate.

Examples 4.3
1. As a familiar example, the straight line distance between points P (x1 , y1 )
and Q(x2 , y2 ) in the plane is given by
√
d(P, Q) = (x1 − x2 )2 + (y1 − y2 )2 .

Mathematically, this is often referred to as the Euclidean distance or

Euclidean metric. When applied to points on a map, for example, this
is sometimes referred to as the ‘distance the crow flies’ – see figure 4.3.
The formula for d(P, Q) follows by applying Pythagoras’ theorem to the
right-angled triangle in the figure.
 153

d ( P, Q ) = ( x1 - x2 ) 2 + ( y1 - y2 ) 2

Q
y1

y2 P

x2 x1

FIGURE 4.3: The Euclidean metric.

2. Residents of Manhattan in New York, for example, would not find the
‘crow flies’ distance very useful simply because this is not way they are
able to travel around the city. They may refer instead to the distance
between locations in the city as being ‘three blocks west and two blocks
north’, for example. The mathematical version of this is what we shall
call the Manhattan distance or Manhattan metric dM between
points P (x1 , y1 ) and Q(x2 , y2 ) in the plane defined by

dM (P, Q) = |x1 − x2 | + |y1 − y2 |.

The dM distance is the sum of the horizontal and vertical distances

between the points — see figure 4.4.

Q
y1

| y1 - y2 |
P
y2
| x1 - x2 |

x2 x1

FIGURE 4.4: The Manhattan metric.

3. It is well known that the paths aeroplanes take when flying between cities
are arcs of great circles. (A great circle on a sphere is the intersection of
the sphere with a plane that passes through the centre of the sphere.)
Aeroplanes take these paths because they are the shortest distances
between points on the sphere. We call this great circle distance the
sphere distance or sphere metric dS between the points P and Q —
see figure 4.5.
154

Q
d S ( P, Q )
P

FIGURE 4.5: The spherical metric.

4. Our final example is not geometric in nature. A binary word is simply

a string of 0s and 1s, such as 0111000110101110. Binary words are used
to store and communicate data electronically. The distance between two
(equal length) binary words is the number of positions, or bits, in which
the words differ. We refer to this as the binary word distance or
binary word metric. For example, if x = 10110101 and y = 11010111
are two 8-bit binary words, then their distance is d(x, y) = 3 because
they differ in the second, third, and seventh bits.

Each of the previous examples gives a sensible interpretation of ‘distance’

between ‘points’ where, in the last example, we need to interpret points as
binary words. If we are to study ‘spaces where we can measure distances’ in
a general context, then we need to extract the key properties that a function
must possess in order to be a (sensible) distance function. Let’s suppose we
have a set of ‘points’ X; these could be points in the plane or on a sphere or
something less geometric such as binary words. The following are reasonable
properties that we would want any distance function, or metric, to possess.

• The distance function d should take a pair of points (x, y) as ‘input’ and
return a result d(x, y) that is a real number.

• The distance d(x, y) should be non-negative, d(x, y) ≥ 0.

• The distance d(x, y) should only be zero when x = y.

• It should not matter whether we measure distance from x to y or from

y to x: d(x, y) = d(y, x).

• The last property is a little less obvious but arises from thinking about
the lengths of sides of triangles. If we have three points x, y, and z,
then the (direct) distance from x to z should be no greater than the
distance that goes via y: d(x, z) ≤ d(x, y) + d(y, z). This is illustrated in
figure 4.6.
 155

It turns out that these properties are just those needed to define a sensible
distance function for which (a) we can build a useful theory and (b) there are
many examples in different contexts. The formal definition is the following.

Definition 4.1
Let X be a non-empty set. A metric on X is a function d : X × X → R
satisfying the following three conditions.

(M1) For all x, y ∈ X, d(x, y) ≥ 0 and d(x, y) = 0 if and only if x = y.

(M2) For all x, y ∈ X, d(x, y) = d(y, x).

(M3) For all x, y, z ∈ X, d(x, z) ≤ d(x, y) + d(y, z).

A set X together with a metric is called a metric space.

The condition (M3) is called the triangle inequality for the metric d.

d ( x, z ) z
x
d ( y, z )
d ( x, y )
y

FIGURE 4.6: The triangle inequality.

Although it is not immediately obvious, each of the distance functions in

examples 4.3 defines a metric. In each case, it is clear that we have defined
the appropriate function d : X × X → R. To verify that the function defines a
metric we need to prove that it satisfies the three conditions (M1), (M2), and
(M3) in definition 4.1. We will establish this for the Manhattan metric in the
next theorem.

Theorem 4.7
The function

d : R2 × R2 → R, d((x1 , y1 ), (x2 , y2 )) = |x1 − x2 | + |y1 − y2 |

defines a metric on R2 , called the Manhattan metric.

Proof. We show that d satisfies the three properties in definition 4.1.

156

(M1) Let (x1 , y1 ), (x2 , y2 ) ∈ R2 .

Then d((x1 , y1 ), (x2 , y2 )) = |x1 − x2 | + |y1 − y2 | ≥ 0 since the modulus
function is non-negative.
Also d((x1 , y1 ), (x2 , y2 )) = 0 ⇔ |x1 − x2 | + |y1 − y2 | = 0
⇔ |x1 − x2 | = 0 and |y1 − y2 | = 0
⇔ x1 = x2 and y1 = y2
⇔ (x1 , y1 ) = (x2 , y2 ).
Hence d satisfies (M1).

(M2) To establish (M2), we will need the following property of modulus

|a − b| = | − (a − b)| = |b − a|. (*)
Let (x1 , y1 ), (x2 , y2 ) ∈ R .
2

Then d((x1 , y1 ), (x2 , y2 )) = |x1 − x2 | + |y1 − y2 |

= |x2 − x1 | + |y2 − y1 | (using (*))
= d((x2 , y2 ), (x1 , y1 )).
Hence d satisfies (M2).

(M3) To establish (M3), we will need the corresponding property for the mod-
ulus function,
|a − c| ≤ |a − b| + |b − c|. (**)
This follows from the triangle inequality for the modulus function, ex-
ample 4.2.4.6, page 146, by letting x = a − b and y = b − c.
Let (x1 , y1 ), (x2 , y2 ), (x3 , y3 ) ∈ R2 .
Then d((x1 , y1 ), (x2 , y2 )) + d((x2 , y2 ), (x3 , y3 ))
= (|x1 − x2 | + |y1 − y2 |) + (|x2 − x3 | + |y2 − y3 |)
= (|x1 − x2 | + |x2 − x3 |) + (|y1 − y2 | + |y2 − y3 |)
≥ |x1 − x3 | + |y1 − y3 | (using (**))
= d((x1 , y1 ), (x3 , y3 )).
Hence d satisfies (M3).

Since d satisfies the three conditions (M1), (M2), and (M3), it is a metric on
R2 .

In order to give a flavour of how an abstract theory of metric spaces might

be developed that is both independent of individual examples but applying
to all metric spaces, we consider the generalizations of discs and spheres in
2-dimensional and 3-dimensional spaces.
A disc in R2 with centre a ∈ R2 and radius r ∈ R+ is the set of those points x
 157

whose distance from the centre is less than or equal to the radius. We denote
the disc with centre a and radius r as
D(a, r) = {x ∈ R2 : d(x, a) ≤ r}.
In three-dimensional space, the corresponding concept is the sphere S(a, r)
with a ∈ R3 and radius r ∈ R+ given by
S(a, r) = {x ∈ R3 : d(x, a) ≤ r}.
Clearly these definitions rely only on the notion of distance and therefore they
generalise to any metric space. The corresponding concept in a metric space
is that of a ‘closed ball’, which we now define.

Definition 4.2
Let X be metric space with metric d. Let a ∈ X and r ≥ 0 be a real number.
The (closed) ball, centre a and radius r in X is
B(a, r) = {x ∈ X : d(x, a) ≤ r}.

Example 4.4
Using the Manhattan metric for R2 , closed balls are diamond-shaped rather
than disc-shaped as in the Euclidean metric. For example, consider the unit
ball (that is, unit radius) centred at the origin 0 = (0, 0),
B(0, 1) = {x ∈ R2 : d(x, 0) ≤ 1} = {(x1 , x2 ) ∈ R2 : |x1 | + |x2 | ≤ 1}.
The ball, shown in figure 4.7, is bounded by four lines:

x1 + x2 = 1 in the first quadrant where x1 ≥ 0, x2 ≥ 0

−x1 + x2 = 1 in the second quadrant where x1 ≤ 0, x2 ≥ 0
−x1 − x2 = 1 in the third quadrant where x1 ≤ 0, x2 ≤ 0
x1 − x2 = 1 in the fourth quadrant where x1 ≤ 0, x2 ≥ 0.

Examples 4.5
In these examples, we see how we can use what is familiar — discs in R2
coming from the Euclidean metric — to motivate results in the general context.
However, we also sound a note of caution. Although the familiar context can
motivate the more general case, we also need to be aware that any proofs must
only rely on the general context and must not assume anything that may be
true in the familiar context but not true more generally.

1. We first use our intuition about when discs in R2 are disjoint or subsets of
one another to state and prove corresponding results for general metric
spaces. Motivated by figure 4.8, it should be clear that, for two discs
D(a, r) and D(b, s) in R2 ,
158

1
B(0,1)

-1 1

-1

FIGURE 4.7: A closed ball in the Manhattan metric.

(i) if d(a, b) > r + s, then the discs are disjoint, D(a, r) ∩ D(b, s) = ∅;
(ii) if d(a, b) + r < s, then D(a, r) ⊆ D(b, s).

D (a, r )
D (a, r ) r
D (b, s )
r s d (a,b ) a

a d (a,b ) b s b

D (b, s )
(i)
(ii)

FIGURE 4.8: Relationships between discs in R2 .

Motivated by the situation in R2 with the Euclidean metric, we now

state and prove the corresponding results for a general metric space.

Theorem 4.8
Let X be a metric space with metric d, and let B(a, r), B(b, s) be two
closed balls in X.

(i) If d(a, b) > r + s, then the balls are disjoint, B(a, r) ∩ B(b, s) = ∅.
(ii) If d(a, b) + r ≤ s, then B(a, r) ⊆ B(b, s).

Proof.
(i) Let B(a, r), B(b, s) be two closed balls in X and suppose d(a, b) >
r + s.
To prove that the balls are disjoint, we will show that if an element
 159

x ∈ X belongs to one of the balls, then it does not belong to the

other.
So suppose that x ∈ B(a, r). This means that d(x, a) ≤ r. Note
that, by (M2), we will use d(x, a) and d(a, x) interchangeably.
By the triangle inequality (M3) we have d(a, x) + d(x, b) ≥ d(a, b).
Therefore d(x, b) ≥ d(a, b) − d(a, x).
Since d(a, b) > r + s and d(a, x) ≤ r, we have

d(x, b) ≥ d(a, b) − d(a, x) > (r + s) − r = s

so x ̸∈ B(b, s).
Hence the balls are disjoint, B(a, r) ∩ B(b, s) = ∅.
(ii) Let B(a, r) and B(b, s) be two closed balls in X and suppose that
d(a, b) + r ≤ s.
To prove that B(a, r) is a subset of B(b, s) we need to show that,
if x ∈ B(a, r), then x ∈ B(b, s).
So suppose x ∈ B(a, r). Then d(x, a) ≤ r.
Note that, since d(a, b) + r ≤ s we have d(a, b) ≤ s − r.
Again, the triangle inequality (M3) gives d(x, b) + d(b, a) ≤ d(x, a).
Therefore

d(x, b) ≤ d(x, a) − d(a, b) ≤ r − (r − s) = s

since d(a, b) ≤ s − r.
Therefore x ∈ B(b, s). We have proved that:

x ∈ B(a, r) ⇒ x ∈ B(b, s).

Hence B(a, r) ⊆ B(b, s).

2. In example 1 (i) above, we showed that if the distance between ball

centres is greater than the sum of the radii, then the balls are disjoint.
For the Euclidean metric in R2 , the converse is also true: if the balls
(discs, in this case) are disjoint, then the distance between their centres
must be greater than the sum of the radii.
Surprisingly, perhaps, this is not true generally for metric spaces. To
show this, we give an example of a metric space and two balls which
are disjoint but where the distance between their centres is less than the
sum of their radii.
Let X be any non-empty set. We need X to contain at least two elements
160

because we need to define two different balls. We leave it as an exercise

(see exercise 4.1.4) to show that the function
{
0 if x = y
d : X × X → R, d(x, y) =
̸ y
1 if x =

defines a metric on X. This is called the discrete metric on X.

( )
Now let X = R with the discrete metric and consider a ball B a, 43 with
center a ∈(R and ) radius 4 . If x ̸= a then, by definition, d(x, a) = 1 > 4 ,
3 3

so (x ̸∈ )B a, 4 . Therefore the only element of X that lies in the ball

3

B a, 34 is a itself (as d(a, a) = 0, of course). Hence

( )
B a, 43 = {a}.
( ) ( )
Now consider the two balls B 0, 34 = {0} ( and) B (1, 43 ) = {1}. Clearly
the balls have no elements in common, B 0, 43 ∩B 1, 43 = ∅. However,
the distance between their centres d(0, 1) = 1 is less than the sum of
radii 43 + 34 = 23 . Therefore the converse of theorem 4.8 (i) is false for the
discrete metric: if B(a, r) ∩ B(b, s) = ∅ it does not necessarily follow
that d(a, b) > r + s.

Exercises 4.3

1. Prove each of the following properties of the modulus function.

(i) For all x ∈ R, x ≤ |x|.

(ii) For all x, y ∈ R, |xy| = |x||y|.
(iii) For all x, y ∈ R, ||x| − |y|| ≤ |x − y|.
Hint: begin by applying the triangle inequality for | · | (page 146)
to the expression |(x − y) + y|.

2. (i) Prove that there

 exists a real number δ > 0 such that |x − 1| <
δ ⇒ x3 − 1 < 2.
(ii) Prove that, for all real numbers ε > 0, there exists
 a real number
δ > 0 such that |x − 1| < δ ⇒ 2x2 + 3x − 5 < ε.
(iii) Prove that, for all real numbers ε > 0, there exists a real number
δ > 0 such that |x + 2| < δ ⇒ x3 + x2 + 4 < ε.
(iv) Prove that, for all real numbers ε > 0, there exists a real number
x − 1
δ > 0 such that |x − 1| < δ ⇒   < ε.
x 
 161

3. Verify that the binary word metric, defined in example 4.3.4, does indeed
define a metric on the set of all binary words of some fixed length.

4. Let X be any non-empty set and let d be defined by

{
0 if x = y
d : X × X → R, d(x, y) =
1 if x ̸= y.

(i) Show that d is a metric on X. This is called the discrete metric

on X and (X, d) is called a discrete metric space.
(ii) Describe the closed balls in the discrete metric space (X, d).
(iii) In example 4.5.2, we showed that the converse of theorem 4.8 (i) is
false for the discrete metric.
Investigate whether theorem 4.8 (ii) is true or false for the discrete
metric.

4.5 A more formal framework

In this section, we provide a more formal discussion of the nature of proof.

Those readers who are mainly interested in various techniques of proof, rather
than an underlying framework for proof, may omit this section.
To understand more fully what mathematicians mean by proof, we first need
to look a little more closely at the nature of mathematics itself. From a for-
mal standpoint, mathematics operates according to something known as the
‘axiomatic method’. This was first introduced by Euclid over two thousand
years ago and has subsequently evolved, particularly during the last two hun-
dred years, into the current modus operandi of mathematics. We begin this
section by giving a brief, if somewhat incomplete, description of the axiomatic
method which governs the development of any mathematical theory.
In outline, a branch of mathematics starts with a set of premises and pro-
ceeds by making deductions from these assumptions using the methods of
logic described in chapter 2. The premises are called ‘axioms’, the statements
deduced from them are the theorems, and the sequences of deductions them-
selves are the proofs of the theorems. The case study on metric spaces given
in section 4.4 was intended to illustrate in microcosm the development of a
mathematical theory in this way. In the remainder of this section, we shall
expand upon and make more precise this overview of the formal description
of the mathematics, referring to the metric spaces example as appropriate.
Of course, mathematics is a discipline engaged in by human beings and, in
162

practice, it does not develop in quite such a precise and orderly manner as we
have just indicated. In the remaining chapters of this book, we shall consider in
more detail how mathematicians really go about exploring their mathematical
landscapes.
We begin by considering the Prime Factorisation Theorem 4.3: every integer
greater than 1 can be expressed as a product of prime numbers. To compre-
hend the statement, we needed precise meanings for the terms ‘divisible’,
‘prime number’, and so on. This shows that any mathematical theory will
need precisely stated definitions. However, it is not possible to define all the
terms used in a given mathematical theory. A little thought should indicate
why this is so. Consider the definition of a prime number given in section 4.2:
‘a prime number is an integer greater than 1 which is not divisible by any
positive integer except 1 and itself’. This relates the term ‘prime number’ to
more basic concepts such as ‘integer’, ‘positive’, the number ‘1’ and ‘divisible’.
Any definition is like this — it relates the term being defined to other terms.
Some or all of these other terms may then be defined using yet more terms,
and so on. Clearly, this process of definition must stop somewhere or we would
spend all our time defining yet more and more terms and never get round to
doing any mathematics proper. Therefore, some terms must be left undefined.
This discussion indicates what should be the first ingredients in any axiomatic
mathematical theory. They are the undefined or primitive terms.
In the same way that we cannot define every term, so we cannot prove every
result. For example, in our proof of the Prime Factorisation Theorem, we used
(implicitly as well as explicitly) various properties of the integers and prime
numbers. For the most part, we assumed that these properties were familiar
and did not need referring to explicitly. If we were required to prove these
properties, the proofs would need to be based on some other statements about
the integers, and so on. Again, to avoid an infinite regression, we are forced
to have some statements which will not be proved.1 These are the axioms of
the theory.
In the example of a metric space given in definition 4.1, the axioms are the
properties listed as (M1), (M2), and (M3). In this case, the undefined terms are
harder to identify since the notion of a metric space builds on other concepts
such as ‘set’, ‘function’, and ‘real number’.
As we have mentioned, Euclid is generally recognised as the first person to
state axioms explicitly in around 300 BC. Just five axioms were the basis
for his famous development of geometry. To Euclid, however, axioms did not
require proof because they were basic statements about the real physical world
1 The Greek philosopher Aristotle (384–322 BC) was well aware of this. In his Meta-

physics, Aristotle wrote, ‘Now it is impossible that there should be demonstration of abso-
lutely everything, for there would be an infinite regress, so that even then there would be
no proof.’ Indeed, he went on to say of those who took the contrary view, ‘Such a man, as
such, is no better than a vegetable.’ See [3] for further details.
 163

which he took to be self-evidently true. (The Greek word axioma — αξιωµα

— means ‘that which is thought fitting’.) Although mathematicians no longer
view axioms in this way, the Euclidean perspective still lingers in our culture.
In non-mathematical discourse or writing we may come across the phrase ‘it is
axiomatic that . . . ’ meaning that what follows is not open to question. To see
why mathematicians were forced to abandon the Euclidean view of axioms, we
indulge in a brief digression to describe the birth of non-Euclidean geometry.
One of Euclid’s axioms, the parallel axiom, states that, for every line ℓ and
every point P not lying on ℓ, there exists a line m containing P which is
parallel to ℓ in the sense that the two lines never meet; see figure 4.9.

P
m

FIGURE 4.9: Euclid’s parallel axiom.

To claim that this statement is self-evidently true is problematic; the problem

lies in the word ‘never’ in the statement that the lines ℓ and m never meet.
This means that, no matter how far they are extended, the lines will not meet.
Since it is not possible to extend the lines forever, to claim that the parallel
axiom is self-evidently true seems at best to be overstating the case. There
was enough doubt about this axiom for the mathematical community to spend
some two thousand years attempting to show that it could be deduced from
Euclid’s other axioms. If this could have been achieved, then there would have
been no need to include the proposition as an axiom because it would have
been a theorem. Eventually, however, it was discovered that the axiom could
not be deduced from the remaining Euclidean axioms.
In the first half of the nineteenth century, two young mathematicians, the Hun-
garian János Bolyai and the Russian Nikolai Lobachevsky,2 independently of
one another found a geometry in which the parallel axiom is false. This new
geometry shares the remaining Euclidean axioms and its discovery or inven-
tion (depending on your point of view) showed finally that the parallel axiom
could not be deduced from the remaining Euclidean axioms. The reason for
this is quite simple. If the parallel axiom were deducible from the remaining
axioms, then it would be a theorem and so it would not be possible to con-
struct a consistent geometry where the parallel axiom property was false. Since
2 It is highly probable that the great German mathematician Carl Friedrich Gauss shared

this discovery. Although he published relatively little, Gauss was the foremost mathemati-
cian of his time — many would say the greatest ever — and he was acutely aware of the
controversy which would inevitably result from the discovery of non-Euclidean geometry.
164

Bolyai and Lobachevsky found a geometry in which the parallel axiom was
contradicted, it follows that it is not possible to deduce the axiom from the
other Euclidean axioms. This geometry of Bolyai and Lobachevsky is usually
referred to simply as ‘non-Euclidean geometry’ even though there are many
geometries — for example, the geometry of a sphere — that are different from
Euclid’s geometry.
The existence of two geometries, one in which the parallel axiom is true and
one in which it is false, has certain implications. In particular, it is not pos-
sible for both the parallel axiom and its negation to be true, self-evidently or
otherwise. Mathematicians were therefore forced to re-think their views of the
nature of axioms.
Today, we no longer regard axioms as self-evident truths, but simply as state-
ments about the undefined terms which are taken as assumptions to serve as
the basic building blocks of the theory. It is not necessary for axioms to reflect
any perceived property of the ‘real world’, such as the geometry of points and
lines in a plane. In principle, we are free to choose any consistent set of axioms
as the starting point for a mathematical theory. The requirement of consis-
tency, though, is vitally important. A set of axioms is consistent if it is not
possible to deduce from the axioms some proposition P as well as its negation
¬P . If it were possible to infer both P and ¬P , then the axioms contain a hid-
den self-contradiction which make the system useless. Recall from section 2.5
that if an argument has inconsistent premises, then it is automatically valid no
matter what the conclusion — see example 2.14.4 on page 74. When applied
to axiom systems, this means that it is possible to deduce any proposition
whatsoever from an inconsistent set of axioms. The modern perspective has
replaced self-evidence by consistency as the paramount criterion for an axiom
system.3
We have said that, in principle, any consistent set of axioms can serve as the
framework for a mathematical theory. In practice, though, mathematicians do
not choose their axiom systems arbitrarily. Some sets of axioms are tailor-made
for a particular purpose and others are studied because they have interesting
and far-reaching applications. The reasons for studying a particular axiom
system lie outside the system itself and relate to possible interpretations of
the system. For example, the reason why metric spaces are studied in the
abstract (that is, via the axioms given in definition 4.1) is that there are
many examples of systems where the axioms are satisfied. Any result we may
prove about abstract metric spaces will then apply to every example. For
instance, the results in theorem 4.8 apply to closed balls in all metric spaces.
3 Although crucially important, consistency is somewhat elusive. In 1931, the Austrian

logician Kurt Gödel showed that any set of axioms for the arithmetic of the positive integers
could not formally be proved to be consistent. Since elementary arithmetic is fundamental
to just about all of mathematics, this is a rather depressing state of affairs. Although we
know that axiom systems must be consistent, we will frequently be unable to prove them
to be so.
 165

These properties of balls are used in practice when designing error-correcting

codes, where the set of words within a certain distance of codewords (using the
binary word metric described in example 4.3.4) is important in determining
the error-correcting properties of the code.
We will take a slightly informal view of the notion of an ‘axiom system’. In
describing and axiom system, we will not require that everything is ‘axioma-
tised’; in other words, we will allow ‘standard’ mathematical symbols and
terminology to be used in describing the system without these having to be
described by their own set of axioms. In particular, we will generally assume
language of sets and functions as part of this background knowledge that we
can draw on. In the definition of a metric space, for example, we used the
notions of sets and set membership, the Cartesian product, functions, the
real numbers etc. Roughly speaking, we will assume the logical framework
introduced in chapter 2 and the theory of sets and functions introduced in
chapter 3 as forming this ‘standard’ mathematical background that we will
draw on freely in describing axiom systems.

Definition 4.3
An axiom system comprises a collection of undefined terms, which may be
words, phrases or symbols, and a collection of statements, called axioms,
involving the undefined terms (as well as ‘standard’ mathematical symbols
and terminology).

A mathematical theory can now be described as the evolution of an axiom

system by the use of deductive reasoning to prove theorems about the terms
of the system. Definitions can be, and in practice always are, introduced to
smooth the flow of the theory. They serve to simplify notation. In principle,
definitions are unnecessary. In practice, we could never get very far if we had
only the language of the undefined terms to use. For example, once we have
introduced the definition of a prime number we can use this concise term freely
without having to refer constantly to an ‘integer greater than 1 which is not
divisible by any positive integer other than 1 and itself’. The basic core of
the theory is its theorems and their proofs which we consider in more detail
below.
There is an analogy for an axiom system which may prove helpful here. We
could loosely compare the development of an axiomatic mathematical theory
with the construction of a building from, say, bricks and mortar. The raw
materials — sand, cement, clay and so on — are like the undefined terms and
symbols of the system. The first layer of bricks forming the foundations of the
building represents the axioms. It is vitally important that this first layer of
bricks is laid properly if any building constructed on top is not to collapse.
This is analogous to the consistency requirement of the axioms — if the axioms
are inconsistent, then any theory developed from them will ‘collapse’. At this
stage, of course, there is no building but only foundations together with raw
166

materials and rules which will permit a building to be constructed. So it is

with an axiom system — the system itself is just the basic framework from
which a theory can be developed. A building rises from its foundations by
brick being laid on top of brick using mortar to hold the structure in place. In
the mathematical context, each individual brick could be likened to a theorem
and the mortar holding it firmly in place is its proof.
When defining an axiom system, we often have an interpretation of the axioms
in mind; that is, a situation or situations where the undefined terms are given
meanings such that the axioms are true propositions. Before defining a metric
space, we introduced four specific situations of a set together with a distance
function. Each of these may be regarded as an interpretation of the abstract
notion of a metric space. In theorem 4.7, we proved this for the Manhattan
metric by showing that each of the axioms (M1), (M2), and (M3) was satisfied
in this case. The following definition captures this idea.

Definition 4.4
An interpretation of an axiom system is a situation where the undefined
terms of the system are given a meaning. An interpretation is called a model
of the axiom system if the axioms, when interpreted according to the given
meanings, are true propositions.

Note that it is in an interpretation of the system where meaning is introduced

and we may think of models as ‘correct’ interpretations. Models are the raison
d’être of axiom systems. The reason that axiom systems are useful is because
they provide information about their models, which is where meaning resides
and hence is where our interest lies. Indeed, it is the models which really
determine which axioms systems are studied. We have said that, in principle,
any consistent axiom system is just as valid or worthy of study as any other.
In practice though, some axiom systems are more important and hence more
deeply studied than others. The importance of any axiom system lies in its
models and not in some intrinsic property of the system itself.
Axiom systems studied by mathematicians fall into one of two main categories
which serve separate purposes. Some axiom systems, like those of a metric
space, have many different models. In addition to metric spaces, examples
which some readers may have encountered are the axiom systems for various
kinds of algebraic objects such as groups, rings, fields, vector spaces, Boolean
algebras, monoids, and the like. In each of these cases there are many examples
of the particular algebraic structure. Each example is a model of the axiom
system. An important advantage in studying the axiom system in such cases
is that of economy of labour. If we can prove some theorem directly from the
axioms, then it must be the case that the theorem becomes a true proposition
in every model of the axiom system. Thus we will know that every example of
the particular algebraic structure will possess whatever property is described
by the theorem. For instance, using the group theory axioms (see definition 5.4
 167

on page 201 below) it is not too difficult to prove that inverses are unique.
(This proof is given in chapter 7; see theorem 7.15 on page 298.) From this
we know that, in every example of a group, inverses are unique and we do not
need to prove this fact for each and every group.
The second category of axiom systems commonly studied comprise those
which have essentially only one model. In other words, all models are for all
practical purposes the same. (The notion of two models being ‘essentially the
same’ is one which can be made completely precise. The word mathematicians
use for this notion is isomorphic, which is derived from Greek and means
literally ‘having the same shape or form.’ The details of how ‘being essentially
the same’ is given a precise meaning need not concern us here.) Usually in
these cases, the model is a familiar structure such as the set of integers or the
set of real numbers or the Euclidean geometry of two- or three-dimensional
space. Here, the purpose of using axiom systems is rather different. The ax-
ioms represent a few basic properties of the given structure from which it is
possible to deduce many other properties of the structure (the theorems of
the system). For example, there is an axiom system with thirteen axioms, de-
scribing an algebraic object called a ‘complete ordered field’. It can be shown
that there is essentially only one example of a complete ordered field in the
sense that all models of the axiom system are equivalent in a very precise way.
‘The’ example is the set R of real numbers together with the operations of
addition and multiplication as well as the usual less-than-or-equal-to ordering,
≤, of real numbers. What this means is that all the usual properties of the
real numbers can be deduced from just thirteen axioms. The advantage of the
axiomatic approach is that we need assume only a limited number of prop-
erties, as the remainder can be rigorously deduced from these. It also means
that, in a sense, the thirteen axioms define what we mean by the system of
real numbers, i.e., the axioms characterise the real number system.
We now turn to a consideration of how a mathematical theory based on an
axiom system develops by proving theorems using the rules of logic. We need
first to define precisely what we mean by a theorem. Consider, for example,
the theorem known to anyone who has studied elementary geometry: the sum
of the angles of any triangle is 180◦ . In fact, this is a theorem of Euclidean
geometry, but is not a theorem of the non-Euclidean geometry described by
Bolyai and Lobachevsky. Non-Euclidean geometry has a different set of axioms
from Euclidean geometry; Euclid’s parallel axiom is replaced in non-Euclidean
geometry by a different axiom concerning parallel lines. Hence the two axiom
systems have different sets of theorems. In fact, the corresponding theorem
in the non-Euclidean geometry described by Bolyai and Lobachevsky states:
all triangles have angle sum less than 180◦ . The important point here is that
what is or is not a theorem depends on the particular axiom system.
Informally, a theorem in an axiom system is a result that ‘can be deduced
from’ the axioms and any background knowledge that is assumed. We now
wish to consider this informal notion in a little more detail. In some systems,
168

such as propositional logic, the deduction rules can be made very precise.
Indeed, we have given a description of this in section 2.5. Below we describe,
much less formally than for propositional logic, some of the deduction rules
that we may apply when reasoning in an axiom system.
We will use K to denote our background knowledge. We do not wish to make
precise exactly what is included in K as this will depend on the context.
However, it will usually include things like the properties of equality (such as,
if a = b and b = c, then a = c), simple properties of sets (such as, if A ⊆ B and
B ⊆ A, then A = B), simple properties of the integers (such as a + b = b + a)
and so forth.
In section 2.5, we used the notation {P1 , P2 , . . . , Pn } ⊢ Q to denote that the
proposition Q can be deduced from the propositions P1 , P2 , . . . , Pn using the
deduction rules for propositional logic. We now describe something akin to
this for axiom systems. Let Γ be a set of statements and let T be another
statement within the current axiom system. We use Γ T to denote that
T ‘can be deduced from’ the background knowledge K and the statements in
Γ. This is not standard mathematical notation. We will read this ‘squiggly’
arrow as ‘can be deduced from’ or simply ‘follows from’. Shortly we will
outline some of the deduction rules that will apply in axiom systems. First
we wish to introduce the notions of theorem and proof. A theorem is simply a
proposition that can be deduced in the axiom system and a proof is a sequence
of deductions that shows this.

Definition 4.5
Let A = {A1 , A2 , . . . , Ak } be a collection of axioms that comprises an axiom
system and let K be the assumed background knowledge within which the
axiom system sits.

(i) A theorem in the axiom system is a statement T that may be deduced

from the axioms of the system, A T.

(ii) A proof of a theorem T is a sequence P1 , P2 , . . . , Pn of statements such

that, Pn = T and, for r = 1, . . . , n, either Pr is an axiom or it is deducible
from the previous statements in the proof:

Pr ∈ A or A, {P1 , . . . , Pr−1 } Pr .

Note that, according to this definition, the first step in a proof will always be
an axiom because there are no prior statements from which it may be deduced.
This is analogous to our formal deductions in proposition logic where the first
propositions to appear in the deductions were always premises. Later, we will
modify this notion slightly, to allow theorems that we have previously proved
to appear as statements in the proof rather than having always to begin right
back with the axioms.
 169

In section 2.5 (page 67), we introduced deduction rules for propositional logic
that allow us to establish logical implication of propositions. We now consider
a similar collection of deduction rules for axiom systems. We may regard these
rules as describing some of the properties of the ‘follows from’ relation . The
rules are described in an analogous way to the deduction rules for propositional
logic. Each rule has two lines; on the first line we have a collection of ‘follows
from’ statements and on the second line is a single ‘follows from’ statement.
The rule is intended to capture the fact that, if we can establish each of the
‘follows from’ statements in the first line, then we may deduce the ‘follows
from’ statement of the second line. As we shall see, these rules underpin some
of the methods of proof that are considered elsewhere in this text.

Conjunction Rule
This rule says that if we can separately deduce both P and Q from the same
set of statements Γ, then we may deduce their conjunction P ∧ Q.
Γ P; Γ Q
Γ P ∧Q
We used this rule in our proof in theorem 4.1 where, to prove
the sum and difference of two odd integers is even,
we first separated this out into two separate statements
the sum of two odd integers is even
and the difference of two odd integers is even.
We then proved each of these statements individually.

Conditional Rule
This rule says that if we can deduce Q from Γ and P , then we can deduce the
conditional P ⇒ Q from Γ.
Γ, P Q
Γ (P ⇒ Q)
This rule underpins the method of direct proof of a conditional P ⇒ Q. It
says that, to deduce P ⇒ Q from a collection of statements Γ, we can instead
add P to our assumptions and from this deduce Q. This was precisely the
method of direct proof of P ⇒ Q: assume P and deduce Q.

Modus Ponens Rule

The next rule is one that allows us to make step-by-step deductions in a proof.
It states that if we can deduce P from Γ and we can separately deduce Q from
Γ and P , then we can deduce Q from Γ.
170

Γ P ; Γ, P Q
Γ Q
We have called this the Modus Ponens rule although it does not quite mirror
Modus Ponens for propositional logic.
P; P ⇒ Q
Q
However, using the Conditional rule, from Γ, P Q we may deduce Γ
(P ⇒ Q). Were we to replace Γ, P Q with Γ (P ⇒ Q) in the statement
of the rule, then it would look like the corresponding rule in propositional
logic.
Γ P; Γ (P ⇒ Q)
Γ Q
In exercise 4.4.1 we show that the ‘converse’ of the Conditional rule also holds.
In other words, if we can deduce P ⇒ Q from Γ, then we can deduce Q from
Γ and P . Hence these two deductions are essentially equivalent.

Adding Assumptions Rule

This rule says that if we can deduce P from Γ, then we can also deduce P
from Γ together with some additional statements Γ1 .
Γ P
Γ, Γ1 P

Equivalence Rule
This rule says that if P and Q are logically equivalent, and if we can deduce
P from Γ, then we can deduce Q from Γ.
Γ P, P ≡ Q
Γ Q
This rule is intuitively obvious because, if P ≡ Q, then P and Q are true under
exactly the same circumstances. In fact, this rule underpins several methods
of proof that will be considered in later chapters.

Cases Rule
This rule says that if we can deduce T separately from Γ and P and also from
Γ and Q, then we can deduce T from Γ, and P ∨ Q.
Γ, P T ; Γ, Q T
Γ, P ∨ Q T
We may think of P and Q in the rule as covering two separate cases; for
 171

example, x ≥ 0 and x ≤ 0 in the case of real numbers. The rule then says that
if we can deduce a theorem T in each of the two cases, then we can deduce T
in the ‘combined’ case P ∨ Q.
We used this (with four cases rather than two) in example 4.2.1, in the proof
of the triangle inequality for the modulus function |x + y| ≤ |x| + |y|. We
actually gave separate mini-proofs for each of four separate cases (each of
which also assumed x ≤ y). Letting T stand for |x+y| ≤ |x|+|y|, we separately
established
1. (0 ≤ x ≤ y) T
2. (x < 0 ≤ y ∧ x + y ≥ 0) T
3. (x < 0 ≤ y ∧ x + y < 0) T
4. (x ≤ y < 0) T.

Note that some of the rules above mirror precisely deduction rules for propo-
sitional logic. For example, compare the Conjunction rules in propositional
logic and axiom systems.
P, Q Γ P; Γ Q
P ∧Q Γ P ∧Q
Similarly, compare the Cases rule above with the Rule of case analysis 2 given
in exercise 2.4.2 (iv).
P ⇒ R, Q ⇒ R Γ, P T ; Γ, Q T
(P ∨ Q) ⇒ R Γ, P ∨ Q T
Rather than giving an ‘axiom system version’ for each of the rules of deduction
in propositional logic individually, instead we give a generic rule that allows
any of the propositional logic deduction rules to be mirrored in an axiom
system.

Propositional Logic Rule

Suppose that there is a propositional logic deduction rule of the following
form.
P1 , P2 , . . . , Pn
Q
Then the following is a deduction rule in an axiom system.
Γ P1 , Γ P2 , . . . , Γ Pn
Γ Q
We have described a proof of a theorem T in an axiom system A as a sequence
of propositions P1 , P2 , . . . , Pn = T where each is either an axiom, Pr ∈ A,
or can be deduced from earlier propositions in the proof {P1 , . . . , Pr−1 }
172

Pr . However, a glance at any of the proofs in this chapter, will reveal that
mathematical proofs are not set out as a sequence of numbered steps in the
way we did for deductions in propositional logic. Similarly, they do not always
start with axioms; indeed, rarely do proofs begin with axioms. Instead, proofs
are written much less formally and the starting point is usually some piece
of background knowledge or some theorem that has already been proved.
We have allowed for this in our description of the deduction rules above by
allowing a non-specified collection of statements Γ (together with background
knowledge) in each of the deduction rules.
Given a collection of axioms A, the first theorems proved will, of necessity,
depend directly on the axioms. In other words, for the first theorem(s) proved
we will have Γ = A. However, thereafter, any theorems that we have already
proved can be added to Γ. In this way we may regard Γ as being (selected
from) the axioms A and all of the previously proved theorems.
We conclude this chapter with a brief consideration of how our deduction rules
shape the structure of some proofs. Consider the Modus Ponens rule:
Γ P ; Γ, P Q
Γ Q

We may regard the notation Γ P as meaning that there is a sequence of

steps that means we can deduce P from Γ.
Γ
..
. (sequence of steps 1)
P
Similarly Γ, P Q means there is a second sequence of deduction steps
beginning with Γ and P ending with Q.
Γ
P
..
. (sequence of steps 2)
Q
We can then splice these two sequences of deduction steps together to establish
Γ Q.
Γ
..
. (sequence of steps 1)
P
..
. (sequence of steps 2)
Q
 173

This really says that we may continue to make deductions that build on earlier
parts of the proof. If we have deduced P in the first half of the proof, we may
use it in any deductions we may make in the second half of the proof.
As we have noted, the Conditional rule underpins the method of direct proof of
P ⇒ Q. Suppose that there is a sequence of deductions that shows Γ, P Q.
Then the Conditional rule gives Γ (P ⇒ Q). Table 4.1 illustrates the
Conditional rule alongside how a direct proof of P ⇒ Q might be structured.

Formal structure ‘Real’ proof

Γ
P Assume P
.. ..
. (sequence of steps) . (sequence of steps)
Q Q
P ⇒ Q (conditional rule) Hence P ⇒ Q

TABLE 4.1: Comparing Conditional rule with direct proof of P ⇒ Q.

Finally, we consider the Cases rule

Γ, P T ; Γ, Q T
Γ, P ∨ Q T
and relate this to a ‘real’ proof using cases. The notation Γ, P T means
there is a sequence of deduction steps starting with Γ and P and ending with
T.
Γ
P
..
. (sequence of steps 1)
T
Similarly, Γ, Q T means there is a sequence of deduction steps starting with
Γ and Q and ending with T .
Γ
Q
..
. (sequence of steps 2)
T
Combining these together, one after the other, together with the Cases rule
allows us to deduce Γ, P ∨ Q T . This is illustrated in the left-hand column
of Table 4.2; the right-hand column shows how a ‘real’ proof with two cases
might be structured.
174

Formal structure ‘Real’ proof

Γ There are two cases.
P Case 1: P
.. ..
. (sequence of steps 1) . (sequence of steps 1)
T T
Q Case 2: Q
.. ..
. (sequence of steps 2) . (sequence of steps 2)
T T
(P ∨ Q) ⇒ T (proof by Cases rule). Hence in both cases T .

TABLE 4.2: Comparing formal structure and real proof using cases.

Exercises 4.4

1. Show that the ‘converse’ of the Conditional deduction rule is also a

deduction rule. In other words, show that if Q can be deduced from Γ
and P , then P ⇒ Q can be deduced from Γ.
Γ (P ⇒ Q)
Γ, P Q
2. Using the deduction rules in this section, establish the following Modus
Tollens rule.
Γ ¬Q; Γ, P Q
Γ ¬P
3. Construct a table, similar to Tables 4.1 and 4.2, that compares the
formal version of a proof of P ∧ Q using the Conjunction rule with the
structure of a real proof of a proposition of the form P ∧ Q.
Chapter 5
Finding Proofs

5.1 Direct proof route maps

In section 4.3, we developed the notion of a direct proof of a conditional
proposition P ⇒ Q. According to the method, described on page 142, we
assume P and from this deduce Q. In practice, of course, the challenge is to
find the chain of deductions that takes us from P to Q.
We will use the analogy of taking a car journey, say, from point A to point
B when we are not completely familiar with the road network. Before setting
out on the journey we need to plan the route from A to B. There are various
techniques that we might deploy in deciding which route to take. Suppose
we have a road atlas but the locations A and B are on different pages so
that we cannot ‘see’ the whole journey in a single place. We might find some
intermediate location, C say, and then plan two ‘sub-journeys’: one from A
to C and the other from C to B. Each of these sub-journeys might also be
subdivided further. In deciding which location C to select, we might reason
along the following lines: ‘if I can find a route to C, then I can see how I can
complete the journey because I can see how to get from C to my destination
B’. We will see that these techniques are analogous to those we may apply in
finding proofs.
With the advent of modern technology, there are other aids that help us find a
route from A to B. For example, we might have a satellite navigation system,
either fitted to the car or as a mobile device. Although stories abound of ‘sat
nav’ systems taking motorists on inappropriate routes, many people rely on
them. There are also mapping services available online, such as Google maps,
that will also find a route from A to B (at least when A and B are on the
same land mass).
Returning to the notion of proof, we can describe the process of finding a
direct proof of P ⇒ Q as finding a ‘route of deductions’ from P to Q. We will
know the starting point of the journey, P , and the destination, Q, but the
route itself will be unknown. Worse still, we will not have at our disposal a sat
nav or Google maps that will find a route ‘automatically’ and we will probably
not even have a complete road atlas. To develop proof-finding skills, we need

175
176

to develop the equivalent of our own ‘sat nav skills’ but in the world of logical
deductions rather than that of a road network. We might call this developing
our ‘proof navigation’ skills, or ‘proof nav’, skills. There is no algorithm or
process that will always find a route of deductions from P to Q. Instead, we
will need to develop strategies and techniques that we can deploy in different
situations and, crucially, we will need to develop a lot of experience of finding
proofs in different circumstances.
In the situation where we are trying to find a direct proof of a proposition Q (as
opposed to a conditional proposition P ⇒ Q), there is a further complication:
we know the destination Q but we are not given the starting point of the
journey of deductions. For a direct proof of Q, described on page 144, we
must start with any appropriate background knowledge and then proceed to
deduce Q. So, in addition to finding a route that ends at Q, we must also select
a suitable starting point for the journey. Often this can be difficult, especially
when there is no obvious piece of background knowledge that will enable us
to start the chain of deductions.

Example 5.1
Consider the following statement.
1
For all positive real numbers x, x + ≥ 2.
x
The obvious questions to ask are: Is the statement true? If so, how do we
prove it? To get a feel for the first question, we try some examples:
1
x = 1: x+ =1+1=2≥2
x
1
x = 2: x+ =2+ 1
2 ≥2
x
1 1
x = 32 : x + = 2
3 + 2 = 2
3 + 3
2 ≥2
x 3
1 1
x = π: x+ = π + > π ≥ 2.
x π

In each of these (few) examples, the result is true. Since there does not seem to
be anything special about the examples, we may start to have some confidence
that the result is true and turn our attention to the second question.
As we described earlier, we need to find a route, comprising statements that
follow logically from one another, from the starting point ‘x is a positive real
number’ to the desired conclusion ‘x + 1/x ≥ 2’. The problem here is the
generality of the starting point. Some of the initial
√ deductions we could make
from x > 0 are, for example, 1/x > 0, 2x > 0 or x > 0. These are illustrated
 177

in figure 5.1. However, it is not very clear that any of these is a useful starting
point.

x Î R, x > 0

1 x>0 2x > 0 x >0

0< x£ 1 1 <x<2
2 2 x³2

1
x+ ³2
x

FIGURE 5.1: Some possible starting points in the proof.

Another possibility, also illustrated in figure 5.1, is that we could consider

cases. If x ≥ 2, then x + 1/x ≥ 2. If 0 < x ≤ 12 , then 1/x ≥ 2 so again
x + 1/x ≥ 2. It might be tempting, therefore, to separate the proof into three
cases:
0 < x ≤ 12 , 12 < x < 2, and x ≥ 2.
As we have indicated, in the first and third cases it is easy to complete the
proof. However, it is not so obvious how to proceed in the second case. Indeed,
it is not clear that this middle case, 12 < x < 2, is any easier to deal with than
the general case x > 0. Hence the consideration of cases is probably not very
helpful either.
At this stage, we might be starting to become concerned as we cannot see a
route of deductions from x > 0 to x + 1/x ≥ 2. A useful technique in such
cases is to ‘work backwards’ from the final result. In other words, think of
a proposition R such that, if we could prove R, then we could deduce the
final result. If we can do that, can we then find another proposition S such
that, if we could prove S, then we could deduce R and hence complete the
proof? Continuing in this way, we would hope to be able to work backwards
far enough that we could see how to connect up our starting position, x > 0,
with where we have reached working backwards.
In our example, as a first step, if we could show that x2 + 1 ≥ 2x, then since
x is positive, we could divide through by x to obtain x + 1/x ≥ 2. This is
illustrated in figure 5.2. In fact, the way we make this ‘backwards step’ is
probably to reason in the wrong direction by making the deduction
1
x+ ≥ 2 ⇒ x2 + 1 ≥ 2x.
x
178

If we do this, we must always be careful to check that the reasoning ‘goes the
other’ way as well. In our case, this works. The implication x + 1/x ≥ 2 ⇒
x2 + 1 ≥ 2x follows because we can multiply the inequality by the positive
quantity x. The reverse implication x2 + 1 ≥ 2x ⇒ x + 1/x ≥ 2 follows
similarly: we can multiply the inequality by the positive quantity 1/x.

x Î R, x > 0

x2 + 1 ³ 2 x

(valid since x > 0)

1
x+ ³2
x

FIGURE 5.2: Working backwards from the final result.

We continue working backwards. Firstly, note that

x2 + 1 ≥ 2x ⇒ x2 − 2x + 1 ≥ 0

and this implication also reverses. Now the left-hand side x2 −2x+1 = (x−1)2
and we know that the square of any real number is greater than or equal to
zero. In other words, we have found our starting deduction, the first step in
the proof, (x − 1)2 ≥ 0. By working backwards, we have obtained a statement
that we know to be true. We can now work forwards to construct the proof
itself as follows.

Proof. Let x be a positive real number. Then x − 1 is also a real number, so

(x − 1)2 ≥ 0. Hence we may proceed as follows.

(x − 1)2 ≥ 0 ⇒ x2 − 2x + 1 ≥ 0
⇒ x2 + 1 ≥ 2x
1
⇒ x+ ≥2 (dividing by x > 0)
x

The actual proof itself is straightforward. It just uses a little algebra and the
 179

facts that (a) the square of a real number is non-negative and (b) multiply-
ing or dividing an inequality by a positive quantity maintains the inequality.
However, what is not at all obvious is that the first step in the proof should be
the observation that (x − 1)2 ≥ 0. We can only really see that this is a helpful
first step using the process of working backwards from the desired conclusion.

Example 5.2
Investigate the truth or otherwise of the statement:

for all positive integers k, the sum of any k consecutive

positive integers is divisible by k.

If the statement is true, give a proof. If the statement is not true, describe the
most general situation where it is true and provide a proof in this case.
Solution
We begin investigating this situation by considering some examples.

k=3 We consider examples of sums of three consecutive integers:

2 + 3 + 4 = 9 which is divisible by 3,
3 + 4 + 5 = 12 which is divisible by 3,
5 + 6 + 7 = 18 which is divisible by 3,
17 + 18 + 19 = 54 which is divisible by 3.
Hence, for these examples, the result holds.
k=4 Since 2 + 3 + 4 + 5 = 14, which is not divisible by 4, the result is
false in general for k = 4.
k=5 We consider examples of sums of five consecutive integers:
2 + 3 + 4 + 5 + 6 = 20 which is divisible by 5,
3 + 4 + 5 + 6 + 7 = 25 which is divisible by 5,
5 + 6 + 7 + 8 + 9 = 35 which is divisible by 5,
17 + 18 + 19 + 20 + 21 = 95 which is divisible by 5.
Hence, for these examples, the result holds.
k=6 Since 2 + 3 + 4 + 5 + 6 + 7 = 27, which is not divisible by 6, the
result is false in general for k = 6.

At this stage, we know that the result is not true for all positive integers k
and we may believe that, when k is odd, the sum of any k consecutive positive
integers is divisible by k. However, we need to be a little careful: the examples
above suggest that the result holds for the odd prime numbers k = 3 and 5 but
not for the even composite (non-prime) numbers k = 4 and 6. So perhaps the
result depends on whether or not k is a prime rather than just any odd integer.
To test this, we consider further examples: the only even prime number, 2,
and a non-prime odd number 9.
180

k=2 Since 2 + 3 = 5, which is not divisible by 2, the result is false in

general for k = 2.
k=9 We consider examples of sums of nine consecutive integers:
2 + 3 + 4 + 5 + 6 + 7 + 8 + 9 + 10 = 54 which is divisible by 9,
3 + 4 + 5 + 6 + 7 + 8 + 9 + 10 + 11 = 63 which is divisible by 9,
5 + 6 + 7 + 8 + 9 + 10 + 11 + 12 + 13 = 81 which is divisible by 9,
17 + 18 + 19 + 20 + 21 + 22 + 23 + 24 + 25 = 189 which is divisible
by 9.
Hence, for these examples, the result holds.

These examples are consistent with our conjecture that the result holds for
odd positive integers only. We can now state our conjecture:

for all positive integers k, if k is odd, then the sum of any

k consecutive positive integers is divisible by k.

We will give two proofs of this drawing on different background knowledge.

The first proof builds directly on the examples we have considered where k
is odd. This is very common: careful consideration of well-chosen examples
can often lead to an argument that applies generally. Our first proof uses the
observation, which the sharp-eyed reader will have spotted from our examples,
that in any sum of an odd number of consecutive integers, there is a middle
integer, and those either side of the middle integer may be combined into pairs
with the same sum. For example, the middle integer of

17 + 18 + 19 + 20 + 21 + 22 + 23 + 24 + 25

is 21 and, either side of 21, working outwards, we have 20 + 22 = 42, 19 + 23 =

42, 18 + 24 = 42, and 17 + 25 = 42.

Proof 1. Let k be an odd positive integer.

The case k = 1 is trivial since every positive integer is divisible by 1, so we

suppose k ≥ 3.

Then k = 2m + 1 for some positive integer m.

Any sum S of k = 2m + 1 consecutive positive integers has a middle integer

n, say, and m integers either side of this, thus:

S = (n − m) + · · · + (n − 1) + n + (n + 1) + · · · + (n + m).

Pairing off each n − r less than n with the corresponding n + r greater than

 181

n gives

S = (n − m) + · · · + (n − 1) + n + (n + 1) + · · · + (n + m)
= [(n − m) + (n + m)] + · · · + [(n − 1) + (n + 1)] + n
= 2n + · · · + 2n + n
= 2mn + n (since there are m ‘paired’ terms)
= (2m + 1)n
= kn.

Therefore S = kn is divisible by k.

Our second proof is both more straightforward and more sophisticated. It is

more straightforward because it does not rely on spotting that a sum of 2m+1
consecutive positive integers has a ‘middle’ integer together with integers on
either side that may be paired off appropriately. It is more sophisticated since
it assumes more background knowledge; in particular, is uses the following
formula for the sum of the first r positive integers

1 + 2 + · · · + r = 21 r(r + 1). (∗)

Proof 2. First note that any sum of k consecutive integers can be written as
follows, where n is the first (that is, smallest) integer in the sum:

n + (n + 1) + · · · + (n + k − 1).

Let k be an odd positive integer. Then k = 2m + 1 for some positive integer

m.
A sum S of k = 2m + 1 consecutive positive integers can be written as follows
(by replacing k with 2m + 1 in the expression above).

S = n + (n + 1) · · · + (n + 2m)
= (n + n + · · · + n) + (1 + 2 + · · · + 2m) (rearranging)
= kn + × 2m × (2m + 1)
1
2 (using (∗) above)
= kn + m(2m + 1) (simplifying)
= k(n + m) (since k = 2m + 1).

Therefore S = k(n + m) is divisible by k.

Euler’s phi function and the Smarandache function

In this section we explore proof-finding techniques using two functions defined

on the set of positive integers. The definition of each function depends on
182

the notion of divisibility: recall that, for m, n ∈ Z+ , we say m divides n if

n = km for some integer k; in this case we also say that m is a factor of
n. If r ∈ Z+ divides both m and n, then r is a common factor of m and
n. The greatest common divisor, or highest common factor, of m and
n is the largest integer that divides them both; we use gcd(m, n) to denote
the greatest common divisor of m and n. Finally, we say that m and n are
coprime if their greatest common divisor is 1, gcd(m, n) = 1. For example,
the greatest common divisor of 18 and 45 is 9, gcd(18, 45) = 9; in particular
18 and 45 are not coprime. However, 18 and 35 are coprime as they have no
common divisors greater than 1.

Definition 5.1
Euler’s phi function, also known as Euler’s totient function, ϕ : Z+ → Z+
is defined by
ϕ(n) = number of integers 1 ≤ a ≤ n coprime with n
= |{a ∈ Z+ : 1 ≤ a ≤ n and gcd(a, n) = 1}|.

Here are some examples:

ϕ(6) = 2 since there are 2 integers coprime with 6, namely 1 and 5,
ϕ(7) = 6 since there are 6 integers coprime with 7, namely 1, 2, 3, 4, 5, 6,
ϕ(8) = 4 since there are 4 integers coprime with 8, namely 1, 3, 5, 7,
ϕ(9) = 6 since there are 6 integers coprime with 9, namely 1, 2, 4, 5, 7, 8.
The function was introduced by Euler in the eighteenth century and has been
much studied since. Indeed, the properties of the function play a crucial role
in the so-called RSA public key encryption system that is used, for example,
to encrypt data sent via the Internet. We will consider two properties of the
function, the first of which is very easy to prove.

Theorem 5.1
If p is prime, then ϕ(p) = p − 1.

Before embarking on the proof, let us consider some examples. In the examples
above, there was only one prime: ϕ(7) = 6 because the integers 1, 2, 3, 4, 5 and
6 are all coprime with 7. Similarly, all the integers 1, 2, 3, . . . , 10 are coprime
with 11, so ϕ(11) = 10. The general pattern is probably clear now: if p is
prime, then all of the integers 1, 2, . . . , p − 1 are coprime with p. Hence we are
ready to write the proof.

Proof. Let p be prime. Then the only (positive) factors of p are 1 and p.
Therefore, in the range, 1 ≤ a ≤ p, all of the integers 1, 2, . . . p − 1 are coprime
with p. There are p − 1 integers in this list so ϕ(p) = p − 1.
 183

Theorem 5.2
If p and q are distinct primes, then ϕ(pq) = (p − 1)(q − 1) = ϕ(p)ϕ(q).

The proof of theorem 5.1 became clear from considering one or two examples,
so let’s begin again by considering examples.

p = 3, q = 5 Here pq = 15. Consider the integers 1, 2, 3, . . . , 15.

The multiples of 3 — 3, 6, 9, 12, 15 — are not coprime with
15. Similarly the multiples of 5 — 5, 10, 15 — are not coprime
with 15. All the remaining integers 1, 2, 4, 7, 8, 11, 13, 14 are
coprime with 15. There are 8 of these.
Hence ϕ(15) = 8 = 2 × 4 = (3 − 1) × (5 − 1), as required.

p = 5, q = 7 Here pq = 35. Consider the integers 1, 2, 3, . . . , 35.

The multiples of 5 — 5, 10, 15, 20, 25, 30, 35 — are not co-
prime with 35. Similarly the multiples of 7 — 7, 14, 21, 28, 35
— are not coprime with 35. All the remaining integers 1, 2,
3, 4, 6, 8, 9, 11, 12, 13, 16, 17, 18, 19, 22, 23, 24, 26, 27, 29,
31, 32, 33, 34 are coprime with 35.
There are 24 of these, so ϕ(35) = 24 = 4×6 = (5−1)×(7−1),
as required.

These examples hold the clue to the proof — it is difficult to see a pattern
in the integers that are coprime with 35, but it is easier to see which integers
are not coprime with 35. In other words, instead of counting the integers that
are coprime with pq, a better strategy will be to count the number of integers
that are not coprime with pq. It is this strategy that allows us to complete
the proof.

Proof. Let p and q be distinct primes and consider the list of integers
1, 2, 3, . . . , pq.
In the list, there are q integers that are multiples of p, namely p, 2p, 3p, . . . , qp.
Any pair of these has a common factor p, so that kp and pq are not coprime for
k = 1, 2, . . . , q. Similarly, there are p integers that are multiples of q, namely
q, 2q, 3q, . . . , pq. Again, any pair of these has a common factor q so that ℓq
and pq are not coprime for ℓ = 1, 2, . . . , p. These two lists combined contain a
total of p + q − 1 integers since pq appears in both lists and is the only integer
belonging to both.
Since p and q are prime, all of the remaining integers are coprime with pq.
Therefore, the total number of integers coprime with pq is
pq − (p + q − 1) = pq − p − q + 1 = (p − 1)(q − 1).
The last part of the inequality, (p − 1)(q − 1) = ϕ(p)ϕ(q), follows from theo-
rem 5.1, which gives ϕ(p) = p − 1 and ϕ(q) = q − 1.
184

We now turn our attention to another function, known as the Smarandache

function, whose definition relies on the notion of divisibility. This was first
considered by Édouard Lucas, a French mathematician, in 1883 and subse-
quently rediscovered by Florentin Smarandache, a Romanian mathematician
just under a hundred years later in the late 1970s.

Definition 5.2
The Smarandache function S : Z+ → Z+ is defined by

S(n) = the smallest positive integer m such that n divides m!

To help get a feel for the function, let’s consider some examples.
n=3 3 does not divide 1! = 1 or 2! = 2 but 3 does divide 3! = 6.
Hence S(3) = 3.
n=4 4 does not divide 1! = 1, 2! = 2 or 3! = 6 but 4 does divide 4! = 24.
Hence S(4) = 4.
n=5 5 does not divide 1! = 1, 2! = 2, 3! = 6 or 4! = 24. However 5 does
divide 5! = 120.
Hence S(5) = 5.
n=6 6 does not divide 1! = 1 or 2! = 2, but 6 does divide 3! = 6.
Hence S(6) = 3.

When considering the first few examples above, we might have begun to sus-
pect that the function just gives S(n) = n, but this is not the case as the
example S(6) = 3 shows. Were we to have continued, we would have found
S(7) = 7, S(8) = 4, S(9) = 6, S(10) = 5 and so on.
Our aim is to explore the function and prove some of its simple properties.
The sequence of values S(1), S(2), S(3), . . . , S(n), . . . is available as sequence
number A002034 of the On-Line Encyclopedia of Integer Sequences.1 A plot
of the first 100 terms of the sequence is given in figure 5.3. From this it is
clear both that the values of S(n) exhibit some complex behaviour but also
that there are some identifiable patterns.
In figure 5.3 it is clear that all the values lie on or below the line y = x and
there are a number of values that appear on the line y = x. In other words,
S(n) ≤ n for all n and S(n) = n for some values of n. From the limited number
of examples above, we may guess that the values of n for which S(n) = n are
the prime numbers. This is our first result about the Smarandache function.

Theorem 5.3
For all n ∈ Z+ , S(n) ≤ n and S(n) = n if n is prime.
1 Available at http://oeis.org/A002034.
 185

100

90

80

70

60

50

40

30

20

10

0
0 10 20 30 40 50 60 70 80 90 100

FIGURE 5.3: The Smarandache function.

This is really two theorems, or at least two separate statements about S(n):
for all n, S(n) ≤ n and for all n, if n is prime, then S(n) = n. Let us try to
understand both parts by considering S(11). We are seeking the first term in
the sequence 1!, 2!, 3!, 4!, . . . that is divisible by 11:
11 does not divide 1! = 1
11 does not divide 2! = 1 × 2
11 does not divide 3! = 1 × 2 × 3
11 does not divide 4! = 1 × 2 × 3 × 4
..
.
11 does not divide 10! = 1 × 2 × 3 × 4 × . . . × 9 × 10
11 does divide 11! = 1 × 2 × 3 × 4 × . . . × 9 × 10 × 11.
This example is probably sufficiently general to indicate how the proof of each
part of the theorem can be constructed.

Proof. Let n be any positive integer.

Consider the sequence of factorials 1!, 2!, 3!, 4!, . . . , n!, (n + 1)!, (n + 2)! . . ..
Clearly n divides n! = 1 × 2 × 3 . . . × (n − 1) × n. Therefore the smallest
m! that is divisible by n must either be n! itself or something smaller than n!.
Hence S(n) ≤ n.
Now suppose that n is equal to some prime p, say, and consider the sequence
of factorials 1!, 2!, 3!, 4!, . . . , p!.
186

It is a property of primes that says if p divides a product ab, then p divides

one of the factors, p divides a or p divides b or both. Since all of the factors
of 1!, 2!, 3!, 4!, . . . , (p − 1)! are less than p, it follows that p does not divide any
of these factorials.
Hence p! is the smallest factorial that is divisible by p, so S(p) = p.

For our second theorem about S, we mirror theorem 5.2 and evaluate S(pq)
where p and q are distinct primes. Again we begin by considering some exam-
ples.
p = 3, q = 5 Then pq = 15, so we need to find the smallest factorial that
is divisible by 15.
Consider a factorial expression for n ≥ 5:

1 × 2 × 3 × 4 × 5 × . . . × n.

Clearly 5! contains both 3 and 5 as factors, so 15 divides 5!.

However any factorial smaller that 5! will not be divisible
by 5.
Hence S(15) = 5.
p = 5, q = 11 Then pq = 77, so we need to find the smallest factorial that
is divisible by 55.
Consider a factorial expression for n ≥ 11:

1 × 2 × 3 × 4 × 5 × 6 × . . . × 10 × 11 × . . . × n.

Again 11! contains both 5 and 11 as factors, so 77 divides

11!. However any factorial smaller that 11! will not be di-
visible by 11.
Hence S(55) = 11.

As before these examples are probably sufficiently general for us to be able to

identify both the correct theorem for S(pq) and its proof.

Theorem 5.4
If p and q are distinct primes, then S(pq) = max{p, q}, the maximum of p and
q.

Proof. Without loss of generality 2 we may suppose that p < q. Consider q!:

q! = 1 × 2 × 3 × . . . × p × . . . × q.
2 We discussed the use of the phrase ‘without loss of generality . . . ’ in proofs in the

commentary to the proof, in example 4.2.1, of the triangle inequality for the modulus
function on page 147.
 187

It is clear that both p and q are factors of q! so that pq divides q!. By theo-
rem 5.3, q will not be a factor of any factorial less than q!. Hence pq does not
divide any factorial less than q!.
Therefore q! is the smallest factorial that has pq as a factor, so S(pq) = q =
max{p, q}.

Exercises 5.1

1. (i) Let a be a real number. Prove that for all non-zero real numbers x,

a2 2a
1+ ≥ .
x2 x

(ii) Prove that, if x and y are non-negative real numbers, then x2 ≥

y 2 ⇒ x ≥ y.
Hint: consider the expression x2 − y 2 .
(iii) Prove that, for all real numbers x and y, x2 ≥ y 2 ⇒ |x| ≥ |y|.
(iv) Prove that, for all real numbers x and y, if x ̸= 0 or y ̸= 0, then
x2 + xy + y 2 > 0.
2. (i) Prove that, for all positive integers n, n3 − n is divisible by 3.
(ii) Prove that, for all integers m and n, if m + n is even, then mn
is the difference of two squares; in other words, mn = a2 − b2 for
some integers a and b.
(iii) Prove that, for all integers n ≥ 2, the sequence of n − 1 integers

n! + 2, n! + 3, n! + 4, . . . , n! + n

are all composite.

√ √ √ √ √
3. Prove that, for all a ≥ 1, a + a + a − a ≤ 2 a.
4. Firstly, we introduce some notation. For positive integers m and n, let
m|n denote ‘m divides n’.
Prove each of the following properties of the divisibility relation on the
set Z+ .

(i) For all m, n, r ∈ Z+ , if m|n and n|r, then m|r.

(ii) For all m, n, r, s ∈ Z+ , if m|n and r|s, then mr|ns.
(iii) For all a, m, n, r, s ∈ Z+ , if a|m and a|n, then a|(rm + sn).

5. Prove each of the following about Euler’s phi function.

188

(i) If p is prime, then ϕ(p2 ) = p(p − 1).

(ii) More generally, if p is prime and a is a positive integer, then ϕ(pa ) =
pa − pa−1 .
(iii) If m and n are coprime, then ϕ(mn) = ϕ(m)ϕ(n).
Note that this result, which generalises theorem 5.2, is quite tricky
to prove. One approach is to arrange the integers 1, 2, 3, . . . , mn
into an array with n rows and m columns as follows.

1 2 3 ... m
m+1 m+2 m+3 ... 2m
.. .. .. ..
. . . ... .
(n − 1)m + 1 (n − 1)m + 2 (n − 1)m + 3 . . . nm

Then count which columns contain (some) integers that are coprime
with mn and, finally, for each such column, count how many entries
are coprime with mn.
αk
(iv) If n = pα 1 α2
1 p2 . . . pk is the factorisation of n, where p1 , p2 , . . . , pk
are distinct primes and α1 , α2 , . . . , αn are positive integers, then
( )( ) ( )
1 1 1
ϕ(n) = n 1 − 1− ... 1 − .
p1 p2 pk

In your proof, you may use any of the results above. Note that
αk
the existence of the expression n = pα 1 α2
1 p2 . . . pk follows from the
Prime Factorisation Theorem, page 132.

6. Let S be the Smarandache function.

(i) Prove that, for all n ∈ Z+ , S(n!) = n.

(ii) Prove that, if p is prime then S(p2 ) = 2p.
(iii) Evaluate S(2k ) for k = 1, 2, . . . 10.
(iv) Extend the argument given in the proof of theorem 5.4 to show
that, if n = p1 p2 . . . pk where p1 , p2 , . . . , pk are primes and p1 <
p2 < · · · < pk , then S(n) = pk .
( )
a b
7. Let A = be a 2 × 2 real matrix such that A2 = 0 where
( )c d
0 0
0= is the 2 × 2 zero matrix.
0 0
Prove that d = −a.
8. (i) Consider the proof of theorem 5.2. Try to identify (some of) the
background knowledge that is assumed in the proof.
(ii) Repeat part (i) for the proof of theorem 5.4.
 189

5.2 Examples from sets and functions

In this section we apply some of the techniques of direct proof to explore
further the properties of sets and functions. We assume familiarity with the
material introduced in chapter 3. In particular, we consider the interaction
between the notion of cardinality firstly with power sets and secondly with
the injective and surjective properties of functions.

Cardinality and power sets

Let A be a finite set. Recall from chapter 3, page 86, that the cardinality of
A, denoted |A|, is the number of elements in A. We may also consider the
power set of A, denoted P(A), which is the set of all the subsets of A — see
page 99. In this section we explore the interaction between these concepts and
the notions of subset, intersection, etc.
In example 3.8.1, we gave the following list of sets and their power sets.
A=∅ P(A) = {∅}
A = {a} P(A) = {∅, {a}}
A = {a, b} P(A) = {∅, {a}, {b}, {a, b}}
A = {a, b, c} P(A) = {∅, {a}, {b}, {c}, {a, b}, {a, c}, {b, c}, {a, b, c}}
A = {a, b, c, d} P(A) = {∅, {a}, {b}, {c}, {d}, {a, b}, {a, c}, {a, d},
{b, c}, {b, d}, {c, d}, {a, b, c}, {a, b, d},
{a, c, d}, {b, c, d}, {a, b, c, d}}
If we now list only the cardinalities of the sets rather than their elements, we
obtain the following table.
|A| |P(A)|
0 1
1 2
2 4
3 8
4 16
From this table, the obvious conjecture to make is the following:
if |A| = n then |P(A)| = 2n .
How might we prove this conjecture? A good starting point would be to con-
sider in more detail one of the examples above. It is usually sensible to consider
190

an example that is not too simple, so that it is not sufficiently general, and
not too complicated, so that it becomes unmanageable. In this case we will
consider the set with 4 elements, A = {a, b, c, d}.
We aim to count the elements of its power set P(A) in such a way that we
can ‘see’ a general argument that will work for all sets. An obvious strategy
is to split the power set into smaller pieces, according to the cardinality of its
elements as follows.
Size of subset Subsets Count
empty set ∅ 1
1-element sets {a}, {b}, {c}, {d} 4
2-element sets {a, b}, {a, c}, {a, d}, {b, c}, {b, d}, {c, d} 6
3-element sets {a, b, c}, {a, b, d}, {a, c, d}, {b, c, d} 4
4-element sets {a, b, c, d} 1
16

The next step is to consider how easy this will be to generalise to a set with n
elements (for an arbitrary n). Suppose |A| = n. Then its power set obviously
contains one empty set and one set with n elements, namely A itself. It is
also easy to realise that there are n subsets of A containing a single element
— there is a single element subset {a} for each a ∈ A. However, now the
reasoning gets a little more tricky. We need to be able to calculate the number
of two-element subsets, the number of three-element subsets, and so on, for our
set with n elements. This can be done but requires background knowledge of
permutations and combinations. In particular, to form a subset of A containing
r elements, say, we need to choose exactly r of its elements to go into the
subset. Thus to complete the proof along these lines we would need to know,
in general, how many ways there are of choosing r objects from n (distinct)
objects. Some readers will no doubt know that this is the binomial coefficient
( )
n n n!
Cr = = .
r (n − r)!r!
To complete the proof, we also need to be able to add the numbers of subsets
with r elements for each r = 0, 1, . . . , n. This requires another piece of back-
ground knowledge, the Binomial Theorem. We leave this as an exercise — see
exercise 5.2.2 — and instead pursue a different line of reasoning here.
Returning to the example A = {a, b, c, d}, instead of splitting the subsets
into collections of equal cardinality, we consider which elements belong to the
subsets. For example, the element a belongs to the following eight subsets:

{a}, {a, b}, {a, c}, {a, d}, {a, b, c}, {a, b, d}, {a, c, d}, {a, b, c, d}.

In other words, a belongs to exactly half of the subsets of A. The same is true
 191

of each element. For example, the element c belongs to the following eight
subsets
{c}, {a, c}, {b, c}, {c, d}, {a, b, c}, {a, c, d}, {b, c, d}, {a, b, c, d}.
Why is this? If we think about an element of A and a subset of A, there
are only two options: either the element belongs to the subset or it does not.
Thus, to form a subset of A, we may consider each element in turn and make
a choice: either include it in the subset or don’t. Each set of choices gives
a different subset. For example, the set of choices ‘include, exclude, include,
include’ gives rise to the subset {a, c, d}. Counting the number of different sets
of choices therefore counts the number of subsets and this, finally, gives us an
elementary way of proving our conjecture. By ‘elementary’, we mean a way of
proving the conjecture relying on minimal background knowledge.

Theorem 5.5
Let A be a finite set. If A has cardinality n, then its power set has cardinality
2n .

Proof. Let A be a set with cardinality n. Then we may write A =

{a1 , a2 , . . . , an }.
To form a subset of A, consider each element ar in turn and either include it in,
or exclude it from, the subset. For each element there are two choices, include
or exclude, and the choice for each element is independent of the choices for
the other elements. Therefore there are 2 × 2 × · · · × 2 (n times) = 2n choices
in total.
Each set of choices gives a unique subset of A and every subset arises in this
way. Therefore there are 2n subsets of A so |P(A)| = 2n .

We have noted that there is an alternative proof that involves counting the
number of r-element sets for each r and adding these using the binomial theo-
rem. For those readers who possess the appropriate background knowledge, we
leave this as an exercise. It is interesting to note that our two different ways
of analysing the example where A = {a, b, c, d} lead to these two different
methods of proving the result. In fact, there is another approach to proving
the theorem, based on the Principle of Mathematical Induction, which we will
consider in chapter 8.
We now turn attention to how the relation of subset interacts with the con-
cept of power set. As is common, to get a feel for the situation, we consider
examples. In fact, the examples given in example 3.8.1 and repeated above
will be sufficient to make a conjecture. Considering the list of sets and their
power sets on page 189 above, it is clear that if A is a subset of B, then P(A)
is also a subset of P(B). For example, taking A = {a, b} and B = {a, b, c, d},
so that A ⊆ B, we have
192

P(A) = {∅, {a}, {b}, {a, b}}

and P(B) = {∅, {a}, {b}, {c}, {d}, {a, b}, {a, c}, {a, d},
{b, c}, {b, d}, {c, d}, {a, b, c}, {a, b, d},
{a, c, d}, {b, c, d}, {a, b, c, d}}

and clearly P(A) ⊆ P(B). Thus we may make the following conjecture where
A and B are sets:
if A ⊆ B then P(A) ⊆ P(B).

In order to see how we might prove this, we need to understand why the
conjecture is true. Recall that, for any sets X and Y , X is a subset of Y
precisely when every element of X is also an element of Y — see page 87.
Symbolically,

X ⊆ Y if and only if x ∈ X ⇒ x ∈ Y for all x.

In the case of power sets, X ∈ P(A) means X ⊆ A. (This is just the definition
of power set of A: it is the set that contains all the subsets of A.) Thus, to
prove that P(A) ⊆ P(B), we need to show that X ⊆ A implies X ⊆ B. For
this, we will use a direct proof.

Theorem 5.6
For all sets A and B, if A ⊆ B, then P(A) ⊆ P(B).

Proof. Suppose that A ⊆ B. To prove that P(A) ⊆ P(B), we need to show

that X ∈ P(A) ⇒ X ∈ P(B) for all sets X.

Let X ∈ P(A). This means that X ⊆ A. We need to show that X ⊆ B, so

that X ∈ P(B).

Let x ∈ X. Since X ⊆ A it follows that x ∈ A. However, we also have that

A ⊆ B, so it now follows that x ∈ B. We have shown that x ∈ X ⇒ x ∈ B
for all elements x, which means that X ⊆ B. Hence X ∈ P(B).

Overall we have now shown that, for all sets X, X ∈ P(A) ⇒ X ∈ P(B).
Hence P(A) ⊆ P(B) as required.

This proof is not long and none of the individual reasoning steps are hard.
Nevertheless many readers will find this proof quite difficult both to under-
stand and, more especially, to construct. The reason for this is that the proof
is ‘multi-layered’. We may illustrate the layers in the proof as follows.
 193

Assume A ⊆ B
Need to deduce P(A) ⊆ P(B))
P(A) ⊆ P(B) means X ∈ P(A) ⇒ X ∈ P(B)
Assume X ∈ P(A)
Need to deduce X ∈ P(B)
X ∈ P(A) means X ⊆ A
X ∈ P(B) means X ⊆ B so need to deduce X ⊆ B
X ⊆ B means x ∈ X ⇒ x ∈ B
Assume x ∈ X
..
.
Deduce x ∈ B
Hence X ⊆ B
Hence X ∈ P(B)
Hence P(A) ⊆ P(B)

The structure of the proposition to be proved is a conditional, A ⊆ B ⇒

P(A) ⊆ P(B). At the highest level, the structure of the proof is that of a
direct proof of a conditional:
assume A ⊆ B and deduce P(A) ⊆ P(B).
However, the required deduction P(A) ⊆ P(B) is really another conditional
statement X ∈ P(A) ⇒ X ∈ P(B), so we use another direct proof
assume X ∈ P(A) and deduce X ∈ P(B)
nested inside the top-level direct proof. Again the required deduction, X ∈
P(B) is equivalent to another conditional statement, x ∈ X ⇒ x ∈ B, which
requires a further nested direct proof,
assume x ∈ X and deduce x ∈ B.

When we have multi-layered proofs such as this, is is often sensible to see

whether the proof may be ‘modularised’ by pulling out part of the proof into
a separate result that may be proved independently of the main theorem. If
we can do this, we will often refer to the separate ‘interim’ results as lemmas.

In the example of theorem 5.6, there is an interim ‘stand-alone’ result, called

the transitive property of subset, that we can prove separately. We describe
this approach below. Although there are now two separate proofs to be con-
structed, ‘modularising’ the proof in this way almost certainly provides a more
understandable route to the proof of theorem 5.6.
194

Lemma 1 (Transitive property of subset)

For all sets X, Y , and Z, if X ⊆ Y and Y ⊆ Z, then X ⊆ Z.

The statement to be proved is a conditional (X ⊆ Y ∧ Y ⊆ Z) ⇒ X ⊆ Z and

we will use the method of direct proof: assume X ⊆ Y ∧ Y ⊆ Z and deduce
X ⊆ Z.

Proof. Let X, Y , and Z be sets such that X ⊆ Y and Y ⊆ Z.

We need to show that X ⊆ Z. In other words, we must show that, for all
elements x, if x ∈ X, then x ∈ Z.
So suppose that x ∈ X. Since X ⊆ Y it follows that x ∈ Y . But Y ⊆ Z also,
so it follows that x ∈ Z.
We have shown that x ∈ X ⇒ x ∈ Z, which means that X ⊆ Z as required.

Alternative proof of theorem 5.6 using lemma 1. Suppose that A ⊆ B.

To prove that P(A) ⊆ P(B), we need to show that X ∈ P(A) ⇒ X ∈ P(B) for
all sets X.
Suppose that X ∈ P(A). This means that X ⊆ A. Then we have both X ⊆ A
and A ⊆ B so we may deduce X ⊆ B, by lemma 1. Hence X ∈ P(B).
Overall we have now shown that, for all sets X, X ∈ P(A) ⇒ X ∈ P(B).
Hence P(A) ⊆ P(B) as required.

Injective and surjective functions and cardinality

For our next collection of example proofs, we consider injective and surjective
functions and what the injectivity and surjectivity properties tell us about
the cardinalities of the domain and codomain. The injective and surjective
properties are given in definitions 3.3 and 3.4, respectively. Informally, recall
that an injective function is one where different elements of the domain have
different images, and a surjective function is one where every element of the
codomain is the image of something in the domain.

Example 5.3
We begin by considering the function with rule
3x + 5
f (x) =
x+4
where x is a real number. Note that f (x) is not defined when x = −4, so we
 195

will take the domain for f to be the set R − {−4} = {x ∈ R : x ̸= −4}. A

little thought should reveal that f (x) cannot equal 3. One way of seeing this
is to re-write the expression for f (x) as

3(x + 5/3)
f (x) = ;
x+4

since x + 5/3 ̸= x + 4, it is clear that f (x) ̸= 3. As a result, we will take the

codomain for f to be the set R − {3} = {x ∈ R : x ̸= 3}. Our aim is to prove
that the resulting function f is both injective and surjective. We capture this
as the following claim.

Claim
The function
3x + 5
f : R − {−4} → R − {3} defined by f (x) =
x+4
is both injective and surjective.

There are two things to prove here, so we give separate proofs that f is injective
and f is surjective. Recalling the definitions, for injective we need to prove

f (x) = f (y) ⇒ x = y for all x, y ∈ R − {−4}

and for surjective we need to prove

for all y ∈ R − {3} there exists x ∈ R − {−4} such that f (x) = y.

For injectivity, we will use a direct proof of a conditional, so we will assume

f (x) = f (y) and deduce x = y. For surjectivity, given y ∈ R − {3}, we need to
find an appropriate x ∈ R−{−4} and then show that f (x) = y. The discovery
of an appropriate x is obviously vital to the proof, but is not part of the proof
itself, so we will consider this first. Given y ∈ R − {3}, we are seeking x such
that
3x + 5
f (x) = =y
x+4
so to find x we just need to solve this equation for x, as follows:

3x + 5 4y − 5
= y ⇒ 3x + 5 = y(x + 4) ⇒ 3x − xy = 4y − 5 ⇒ x = .
x+4 3−y

4y − 5
It is this value of x, x = , that we will use in the proof itself.
3−y

Proof. Firstly we show that f is injective.

196

Let x, y ∈ R − {−4}. Then

3x + 5 3y + 5
f (x) = f (y) ⇒ =
x+4 y+4
⇒ (3x + 5)(y + 4) = (3y + 5)(x + 4)
⇒ 3xy + 12x + 5y + 20 = 3xy + 5x + 12y + 20
⇒ 12x + 5y = 5x + 12y
⇒ 7x = 7y
⇒ x = y.

We have shown that f (x) = f (y) ⇒ x = y for all x and y in the domain of f .
Therefore f is injective.

Secondly, we show that f is surjective.

4y − 5
Let y ∈ R − {3}. Then y ̸= 3, so let x = .
3−y
To show x ∈ R − {−4}, we need to show that x ̸= −4. Writing x as x =
4(y − 5/4)
and noting that y − 5/4 ̸= y − 3, it follows that x ̸= −4.
−(y − 3)
We now have
( )
4y − 5
f (x) = f
3−y
( )
4y − 5
3 +5
3−y
= ( )
4y − 5
+4
3−y
3(4y − 5) + 5(3 − y)
= (multiplying through by 3 − y)
(4y − 5) + 4(3 − y)
12y − 15 + 15 − 5y
=
4y − 5 + 12 − 4y
7y
= = y.
7
We have shown that, for all y in the codomain of f there exists x in the domain
such that f (x) = y. Therefore f is surjective.

Two-phase proof discovery

It is worth reflecting further on the nature and proof of the surjectivity con-
dition. There are two phases to finding a proof: the discovery phase and the
 197

proof phase. In the discovery phase, given an element y in the codomain, we

need to discover an appropriate x in the domain such that f (x) = y. This
discovery phase is not part of the proof itself. During the proof phase, we only
need to verify that the value of x found in the discovery phase does indeed
satisfy the appropriate property, namely f (x) = y.
This two-phase proof process is common when proving statements with two
quantifiers which have the following structure
∀a ∃b • P (a, b).
The statement that a function f : A → B is surjective has this structure:
for all y ∈ B there exists x ∈ A such that f (x) = y.
For the discovery phase, we assumed that y was an arbitrary element of the
codomain. This is because y is universally quantified so that our y needs to
be arbitrary. Then we found an appropriate x in the domain, that depended
on y and satisfied the required property.
Generalising this process to the generic statement ∀a ∃b • P (a, b), the two
phases in finding the proof are described as follows.

Discovery phase. For an arbitrary a, find some b, which is likely

to depend on a, such that P (a, b) is true.
Proof phase. In the proof, let a be an arbitrary element of its
domain. Let b have the value found during the discovery phase.
Verify that, for the chosen value of b, the statement P (a, b) is true.

We will see this two-phase process in operation in section 5.4 where we prove
properties of limits which also involve doubly quantified propositional func-
tions of the form ∀a ∃b • P (a, b).
We conclude this section with a consideration of the relationship between the
properties of injectivity and surjectivity and the cardinalities of the domain
and codomain (where these are finite sets). Figure 5.4 shows the arrow map-
pings for a typical injective and a typical surjective function, each with a small
finite domain and codomain.
For the injective function, there needs to be at least as many elements in the
codomain as the domain in order that different elements of the domain can
have different images in the codomain. For the surjective function, there needs
to be at least as many elements in the domain as the codomain in order that
each element in the codomain can be ‘hit’ by an arrow; more formally, so
that each element in the codomain can be the image of some element in the
domain. The following theorem expresses these observations more formally.
Each of the two parts of the proof is a conditional proposition and we use the
method direct proof in each case.
198

f f
A B A B

Injective function Surjective function

FIGURE 5.4: Injective and surjective functions with finite domain and
codomain.

Theorem 5.7
Let f : A → B be a function, where A and B are finite sets.
(i) If f is injective, then |A| ≤ |B|.
(ii) If f is surjective, then |A| ≥ |B|.

Proof. Let A and B be finite sets, say |A| = n and |B| = m. Then we may write
A = {a1 , a2 , . . . , an } and B = {b1 , b2 , . . . , bm } where there are no repeated
elements listed in each case.
(i) Suppose that f is injective. Then different elements of A have different
images so that the elements f (a1 ), f (a2 ), . . . , f (an ) of B are all distinct.
Hence B contains at least n elements, so |B| ≥ n = |A|, as required.
(ii) Suppose that f is surjective. Then the images of the elements of A,
f (a1 ), f (a2 ), . . . , f (an ), must contain all the elements of B. However,
this list may contain repeats, so B contains at most n elements. Hence
|B| ≤ n = |A|, as required.

There are some results that follow ‘almost immediately’ from a particular
theorem. It is common to refer to such a result as a corollary of the particular
theorem. In the case of theorem 5.7 there is an obvious corollary concerning
bijective functions obtained by putting the two parts of the theorem together.
The proof is a simple direct proof.

Corollary
Let f : A → B be a bijective function, where A and B are finite sets. Then A
and B have the same cardinality, |A| = |B|.
 199

Proof. Suppose that f : A → B is a bijective function.

Then f is injective, so |A| ≤ |B| by theorem 5.4 (i). Also f is surjective, so
|A| ≥ |B| by theorem 5.4 (ii).
Hence |A| = |B|.

Exercises 5.2

1. Prove that, for all finite sets A and B, if A ⊆ B then |A| ≤ |B|.
2. Give an alternative proof of theorem 5.5 by counting the number of
subsets of A with no elements, one element, two elements, and so on
and then using the Binomial Theorem.
3. Describe the ‘layers’ in the proof of lemma 1 and the alternative proof
of theorem 5.6 in a similar way to the description of the original proof
of theorem 5.6 given on page 192.
4. Prove that for all sets A and B, P(A ∩ B) = P(A) ∩ P(B).
5. Prove that, for all sets A and B, P(A ∪ B) ⊆ P(A) ∪ P(B).
Are there any circumstances where P(A ∪ B) = P(A) ∪ P(B)? If so, can
you prove it?
6. This question explores how the Cartesian product of sets interacts with
the subset relation and with intersections and unions of sets.

(i) Prove that, for all sets A, B, X, and Y , if A ⊆ X and B ⊆ Y , then

A×B ⊆X ×Y.
(ii) Prove that, for all sets A, B, X, and Y , (A ∩ B) × (X ∩ Y ) =
(A × X) ∩ (B × Y ).
(iii) Prove that, for all sets A, B, X, and Y , (A ∪ B) × (X ∪ Y ) =
(A × X) ∪ (A × Y ) ∪ (B × X) ∪ (B × Y ).

7. The symmetric difference A ∗ B of two sets A and B is defined by

A ∗ B = (A − B) ∪ (B − A).

Using the Set Theory Laws, page 99, prove each of the following identi-
ties.
(i) For all sets A, A ∗ ∅ = A and A ∗ A = ∅.
(ii) For all sets A and B, A ∗ B = (A ∪ B) − (A ∩ B).
(iii) For all sets A, B, and C, A ∩ (B ∗ C) = (A ∩ B) ∗ (A ∩ C).
200

8. Show that each of the following functions is a bijection.

5x + 3
(i) f : R → R, f (x) =
8
3x
(ii) f : R − {−1} → R − {3}, f (x) =
x+1
(iii) f : [1, 3] → [−2, 2], f (x) = 2x − 4
4
(iv) f : R+ → (0, 2), f (x) = .
x+2
(v) f : R2 → R2 , f (x, y) = (2x − 1, 5y + 3).
 n

 if n is even
2
(vi) f : Z → Z, f (n) =
+
 1 − n if n is odd.

2
9. Let f : A → B be a function. Prove each of the following.
(i) For all subsets C1 and C2 of A, if C1 ⊆ C2 , then f (C1 ) ⊆ f (C2 ).
(ii) For all subsets D1 and D2 of A, if D1 ⊆ D2 , then f −1 (D1 ) ⊆
f −1 (D2 ).
(iii) For all subsets C of A, C ⊆ f −1 (f (C)).
(iv) If f is injective, then for all subsets C of A, C = f −1 (f (C)).
(v) If f is surjective, then for all subsets D of B, f (f −1 (D)) = D.
10. A function f : R → R is said to be even if f (−x) = f (x) for all x ∈ R.
Similarly f is odd if f (−x) = −f (x) for all x ∈ R.
(i) Let g : R → R be any function. Prove that the function f : R → R
defined by f (x) = g(x) + g(−x) is even.
(ii) Prove that any function f : R → R may be written as f = g + h
where g is an even function and h is an odd function.

5.3 Examples from algebra

5.3.1 Group theory

An important algebraic construction in mathematics is that of a ‘group’.3

Groups are important because the concept is simple and has examples and
3 The notion of a group evolved gradually in nineteenth-century mathematics. The French

mathematician Évariste Galois (1811–1832) first coined the term ‘groupe’ and he used the
permutation group of the roots of polynomial equations to show that polynomials of degree
at least 5 are not solvable with radicals.
 201

applications in a wide variety of contexts. A group is simply a collection of

objects together with a rule for combining those objects satisfying three simple
conditions. The rule for combining objects takes two objects and ‘combines’
them to produce another object of the same kind. For example, addition is
a rule for combining integers: given two integers m and n, their sum m + n
is another integer. A rule for combining is called a binary operation and the
formal definition is the following.

Definition 5.3
Let S be a set. A binary operation on S is a mapping, or function, µ :
S × S → S. For x, y ∈ S, we will write µ(x, y) as x ∗ y.

The domain of the binary operation is the Cartesian product S × S which

contains all ordered pairs of elements of S. The term binary operation re-
flects the fact that the domain comprises pairs of elements. The fact that the
codomain of the operation is also the set S means that ‘combining’ x and y in
S produces another element x ∗ y that is also an element of S. Some authors
refer to this as the closure property of the binary operation.
Familiar examples of binary operations are addition and multiplication of
integers as µ(m, n) = m + n and µ(m, n) = mn, each of which defines a
mapping Z × Z → Z. Notice that division µ(m, n) = m/n does not define a
binary operation on Z for two reasons. Most obviously, the codomain is not
Z because, in general, m/n ̸∈ Z; for example 3/2 is not an integer. Also the
domain is not Z × Z either because division by zero is not defined, so that m/0
is meaningless. We could define integer division as a mapping Z × Z∗ → Q,
where Z∗ denotes the set of non-zero integers, but even this does not define a
binary operation according to definition 5.3.

Definition 5.4
A group is a pair (G, ∗), where G is a set and ∗ is a binary operation on G
that satisfies the following three properties.

(G1) Associative property.

For all x, y, z ∈ G, x ∗ (y ∗ z) = (x ∗ y) ∗ z.
(G2) Existence of an identity element.
There is an element e ∈ G, called an identity element for ∗, such that
e∗x=x=x∗e for all x ∈ G.

(G3) Existence of inverses.

For each x ∈ G there is an element x−1 ∈ G called an inverse for x,
such that
x ∗ x−1 = e = x−1 ∗ x.
202

The properties (G1) – (G3) are often referred to as the ‘group axioms’ as they
are the basic properties that define what we mean by a group. Whenever we
refer to ‘a group (G, ∗)’ we may assume that these properties are satisfied.
Interested readers may refer to section 4.5 for a more extensive discussion of
the nature of axioms. Some authors include a fourth property or axiom called
the ‘closure property’ of the binary operation: this says that, if x and y belong
to G, then so, too, does x ∗ y. However, we have included this in the definition
of a binary operation on G as a mapping whose codomain is the set G.
Note that we do not assume the commutative property:

x∗y =y∗x for all x, y ∈ G.

If a group (G, ∗) also satisfies the commutative property, it is called an

Abelian group4 (or commutative group).

Examples 5.4

1. The set of integers Z with binary operation of addition forms a group

(Z, +). The identity element is 0, since 0 + n = n = n + 0 for all n, and
the inverse of n is −n, since n + (−n) = 0 = (−n) + n for all n. The
associative property of addition,

m + (n + p) = (m + n) + p for all m, n, p ∈ Z,

is a familiar property (whose proof depends on giving a precise definition

of the set of integers and the operation of addition).
Note that the set of integers Z with the binary operation of multipli-
cation does not form a group. Multiplication is an associative binary
operation on Z and 1 is the identity element. However, most elements
do not have an inverse. For example, 2 does not have an inverse since
there is no integer n such that 2n = 1. In fact, 1 and −1 are self-inverse
(since 1 × 1 = 1 and (−1) × (−1) = 1) but no other integers have an
inverse under multiplication.

2. If we wish to create a group under multiplication, the previous example

indicates that we will need to include fractions in the set. So we might
conjecture that the set of rational numbers Q forms a group under mul-
tiplication. As in the previous example, multiplication is an associative
binary operation and 1 is the identity element. The inverse of p/q ∈ Q
is q/p since
p q
× = 1.
q p
4 Named after the Danish mathematician Niels Abel (1802–1829), one of pioneers of

group theory.
 203

However q/p is only defined when p ̸= 0. In other words, every non-zero

rational number has an inverse under multiplication. Let
{ }
p
Q∗ = {q ∈ Q : q ̸= 0} = : p, q ∈ Z and p, q ̸= 0
q

denote the set of non-zero rational numbers. Since the product of non-
zero rational numbers is again a non-zero rational number, multiplica-
tion is also a binary operation on the set Q∗ . Since every element of Q∗
has an inverse,
( )−1
p 1 q
= = ,
q p/q p
it follows that (Q∗ , ×) is a group.
In a similar way, the set of non-zero real numbers

R∗ = {x ∈ R : x ̸= 0}

is also a group under multiplication, again with inverses given by with

x−1 = 1/x.
3. A symmetry of a plane figure F is a mapping F → F such that F
looks the same before and after the mapping. The symmetries of a (non-
square) rectangle are:
• the identity mapping that ‘does nothing’, this is denoted e;
• r = a rotation anticlockwise by π radians (or 180◦ ) about the cen-
tre;
• v = a reflection in a vertical line though the centre;
• h = a reflection in a horizontal line though the centre.
These symmetries are illustrated in figure 5.5.

r
h

FIGURE 5.5: Illustrating the symmetries of a rectangle.

We can define composition of symmetries: if a and b are symmetries, we

define ab to be the symmetry ‘a followed by b’. This is just composition of
204

functions but with a slight change of notation. If f and g are mappings,

their composite ‘f followed by g’ is denoted g ◦ f (see page 113). In
the context of groups, we regard symmetries more as algebraic entities
(rather than functions), which is why the composite ‘a followed by b’ is
denoted ab. However, since composition of functions in general is known
to be associative, (f ◦ g) ◦ h = f ◦ (g ◦ h) when either side is defined, we
will assume that composition of symmetries is associative.
Figure 5.6 illustrates the composite rv, r followed by v. In order to keep
track of what is happening to the rectangle, we have labelled its corners.
From the figure we can see that rv has the same effect on the rectangle
as the horizontal reflection, so we write rv = h.

B A D C C D

r v

C D A B B A

FIGURE 5.6: Composing symmetries of a rectangle.

More generally, figure 5.6 also illustrates that composing two symme-
tries of the rectangle produces another symmetry of the rectangle. Thus
if we let S(@A) = {e, r, v, h} denote the set of symmetries of the rectan-
gle, then composition defines a binary operation on S(@A). In order to
determine whether or not S(@A) forms a group under composition, we
first show all of the possible composites of symmetries in the following
Cayley table.5 The rows represent the first element and the columns
represent the second element of the composite. Hence the h in the row
labelled r and column labelled v represents rv = h as illustrated in
figure 5.6.

e r v h
e e r v h
r r e h v
v v h e r
h h v r e

We can now see that S(@A) is a group under the operation of compo-
sition of symmetries. As mentioned above, composition is associative.
5 Named after the English mathematician Arthur Cayley (1821–1879) who was the first

person to introduce the concept of an abstract group.

 205

Clearly the identity symmetry e is the identity element for the binary
operation. Also each element is self-inverse because ss = e for every sym-
metry s ∈ S(@A). Note also that the operation is commutative — the
Cayley table is symmetric about the top-left to bottom-right diagonal
— so this is an Abelian group. This group is called the Klein 4-group
after the German mathematician Felix Klein (1849–1925).

4. Let S(△) denote the symmetries of an equilateral triangle. The symme-

tries are the following, illustrated in figure 5.7.
e the identity symmetry
r1 rotation by 2π
3 = 120◦ anti-clockwise
r2 rotation by 4π
3 = 240◦ anti-clockwise
s reflection in the line ℓ1
t reflection in the line ℓ2
u reflection in the line ℓ3

l3

t s

l1 l2
u

FIGURE 5.7: Symmetries of an equilateral triangle.

With the binary operation of composition of symmetries, as in example 2,

the Cayley table for S(△) = {e, r1 , r2 , s, s, u} is the following.

e r1 r2 s t u
e e r1 r2 s t u
r1 r1 r2 e t u s
r2 r2 e r1 u s t
s s u t e r2 r1
t t s u r1 e r2
u u t s r2 r1 e
206

As in the previous example, composition is an associative binary oper-

ation with identity element e. Since r1 r2 = e = r2 r1 (from the Cayley
table), it follows that r1−1 = r2 and r2−1 = r1 . Each reflection x satisfies
xx = e and hence is self-inverse. Thus every element has an inverse. (It
is always the case that the identity is self-inverse e−1 = e.) Therefore
S(△) is a group under composition of symmetries.
Note that the group is non-Abelian. For example, from the Cayley table
we have
r1 s = t but sr1 = u.

In fact, this is the smallest example of a non-Abelian group.

Having defined a group and given some examples, it is now time to turn our
attention to proving some elementary results about groups.

When solving equations in ‘ordinary’ algebra, by which we mean the algebra

of the real numbers, the following Cancellation Property is used frequently:

ax = ay ⇒ x = y.

For example, if 2x = 2y, then we may deduce that x = y. The cancellation

property is only valid for a ̸= 0. For example, 0x = 0y is true for all real
numbers x and y. The Cancellation Property should be stated more precisely
as:
(a ̸= 0 and ax = ay) ⇒ x = y.

In fact, this property is a group theory property. It holds for the non-zero real
numbers R∗ because (R∗ , ×) is a group; see example 5.4.2. In a group, we do
not assume the binary operation is commutative, so there are two cancellation
properties, which are called left cancellation and right cancellation.

Theorem 5.8 (Cancellation properties for groups)

Let (G, ∗) be a group and let g, x, y ∈ G.

(i) Left cancellation: g ∗ x = g ∗ y ⇒ x = y.

(ii) Right cancellation: x ∗ g = y ∗ g ⇒ x = y.

Thinking about the proof, to proceed from g ∗ x = g ∗ y to x = y we need to

find a way of ‘removing’ g from both sides, noting that we may only assume
the properties of a group as given in definition 5.4. Since g −1 ∗ g = e (G3)
and e ∗ x = x (G2), we may ‘remove’ g from both sides of g ∗ x = g ∗ y by
combining both sides on the left by g −1 .
 207

Proof.

(i) Let G be a group and let g, x, y ∈ G. Then

g∗x=g∗y ⇒ g −1 ∗ (g ∗ x) = g −1 ∗ (g ∗ y)
⇒ (g −1 ∗ g) ∗ x = (g −1 ∗ g) ∗ y (by (G1))
⇒ e∗x=e∗y (by (G3))
⇒ x=y (by (G2)).

(ii) The proof of right cancellation is similar and is left as an exercise.

Our second elementary property of groups concerns the interaction between

the binary operation and taking inverses, and gives an expression for the
inverse of x ∗ y. Since the cancellation properties of the previous example were
motivated by the cancellation properties of the real numbers, we might again
look to the real numbers where
1 1 1
(xy)−1 = = × = x−1 y −1 .
xy x y

From this we might be tempted to conjecture that, in a group, (x ∗ y)−1 =

x−1 ∗ y −1 . However this is incorrect. We can see this by considering the group
S(△) in example 5.4.4. In this group

(r1 s)−1 = t−1 = t but r1−1 s−1 = r2 s = u.

The problem with our conjectured generalisation is that multiplication of real

numbers is commutative but, in general, the binary operation in a group is
not. The correct result for groups is the following.

Theorem 5.9 (‘Shoes and Socks’ theorem for groups)

Let (G, ∗) be a group and let x, y ∈ G. Then

(x ∗ y)−1 = y −1 ∗ x−1 .

Before considering the proof of the theorem, we first reflect on its name.
When putting on a pair of shoes and socks, one first puts on the socks and
then the shoes. When reversing this process, the order is reversed: the shoes
are removed before the socks. So it is when taking the inverse of x ∗ y: the
inverses of x and y are combined in the opposite order.
Turning to the proof, we first need to understand what we mean by the inverse
(x ∗ y)−1 . The inverse of an element, as defined in (G3) in definition 5.4, is
208

the element that combines with it (each way around) to produce the identity
e. Thus the inverse of x ∗ y is whatever element g ∈ G satisfies the property
(x ∗ y) ∗ g = e and g ∗ (x ∗ y) = e. So the proof of the theorem just involves
checking that g = y −1 ∗ x−1 satisfies this property.

Proof. Let (G, ∗) be a group and let x, y ∈ G. Then

(x ∗ y) ∗ (y −1 ∗ x−1 ) = x ∗ (y ∗ (y −1 ∗ x−1 )) (by (G1))

−1 −1
= x ∗ ((y ∗ y )∗x ) (by (G1))
−1
= x ∗ (e ∗ x ) (by (G3))
−1
= x∗x (by (G2))
= e (by (G3)).

The proof that (y −1 ∗ x−1 ) ∗ (x ∗ y) = e is similar and is left as an exercise.

Therefore, by the definition of inverse (G3),

(x ∗ y)−1 = y −1 ∗ x−1 .

Our last result about groups in this chapter concerns groups where each ele-
ment x satisfies x ∗ x = e. An example of such a group is the Klein 4-group
given in example 5.4.3.
Before we consider the theorem, we wish to simplify our notation for groups.
In the previous two examples, it has been a little cumbersome to write the
binary operation as x ∗ y. It is very common to omit the binary operation
symbol and write x ∗ y = xy. This is sometimes referred to as writing the
group operation ‘multiplicatively’. We may extend this idea and use a power
notation as follows:

x2 = x ∗ x, x3 = x ∗ x2 = x ∗ (x ∗ x), . . . .

We will also simplify the way we refer to a group and use the phrase ‘a group
G’ as shorthand for ‘a group (G, ∗)’.

Theorem 5.10
Let G be a group such that x2 = e for every element x ∈ G. Then G is
Abelian.

Recall that an Abelian group is one that satisfies the commutative property:
xy = yx for all x, y ∈ G. To obtain a direct proof of the theorem, we need to
assume
x2 = e for every element x ∈ G
 209

and from this deduce

xy = yx for all x, y ∈ G.
There are two key observations that allow us to construct an appropriate
chain of deductions. The first is the observation that, if x2 = e, then x is
self-inverse x−1 = x. This is because the inverse of x is precisely the element
y satisfying xy = e and yx = e. The second key observation is the ‘Shoes and
socks’ theorem 5.9 which, in our simplified notation, gives (xy)−1 = y −1 x−1 .
If every element satisfies x2 = e, then every element is self-inverse so the
equation (xy)−1 = y −1 x−1 simplifies to xy = yx, which is what we needed to
deduce. We can now organise these considerations into a coherent argument
as follows.

Proof. Let G be a group such that x2 = e for every element x ∈ G.

Let x, y ∈ G. Then

x2 = e, y 2 = e and (xy)2 = e

so each of these elements is self-inverse

x−1 = x, y −1 = y and (xy)−1 = xy.

Hence xy = (xy)−1 (since xy is self-inverse)

−1 −1
= y x (by theorem 5.9)
= yx (since x and y are self-inverse).
Therefore G is Abelian.

5.3.2 Linear algebra

The area of mathematics known as linear algebra studies objects called vec-
tor spaces which generalise the notion of vectors in two dimensions. Linear
algebra is closely related to the theory of matrices. A vector in two dimensions
is a quantity that has magnitude and direction; it can be represented as an
ordered pair of real numbers or as a column vector,
( )
a1
a = (a1 , a2 ) =
a2

as shown in figure 5.8. The magnitude of a, which is denoted ∥a∥, is repre-

sented by the length of the line and, by Pythagoras’ theorem, is given by
√
∥a∥ = a21 + a22 .
210

a2

a1

FIGURE 5.8: A vector in R2 .

There are two natural operations that can be performed on vectors in R2 .

Two vectors a and b may be added according to the rule
( ) ( ) ( )
a1 b a1 + b1
a+b= + 1 = .
a2 b2 a2 + b2

Also, a vector a may be multiplied by a scalar λ, which is just a real number,

according to the rule
( ) ( )
a λa1
λa = λ 1 = .
a2 λa2

Geometrically, these operations of addition and scalar multiplication are illus-

trated in figure 5.9.

b a+b
2a
a a

Addition Scalar multiplication

FIGURE 5.9: Addition and scalar multiplication of vectors in R2 .

The vectors in R2 satisfy various properties. For example, addition is commu-

tative: a + b = b + a for all vectors a, b. If we collect a ‘reasonable’ collection
of properties that the vectors in R2 satisfy, we obtain the definition of a vector
space. In essence, a vector space is any set of objects, which we call ‘vectors’,
that may be added and multiplied by scalars (real numbers) and which sat-
isfy properties similar to those of vectors in R2 . The formal definition is the
following.
 211

Definition 5.5
A vector space comprises:
• a set V whose elements are called vectors;
• a rule of addition of vectors: if u, v ∈ V , then u + v ∈ V ;
• a rule for multiplying vectors by real numbers (called scalars): if v ∈ S
and λ ∈ R, then λv ∈ V
such that the following properties are satisfied.
Addition properties
(A1) For all u, v ∈ V , u + v = v + u.
(A2) For all u, v, w ∈ V , (u + v) + w = u + (v + w).
(A3) There is a zero vector 0 ∈ V such that, for all v ∈ V ,
0 + v = v = v + 0.
(A4) For every v ∈ V there is a vector −v ∈ V , called the negative of x,
such that v + (−v) = 0.
Scalar multiplication properties
(M1) For all v ∈ V , λ, µ ∈ R, λ(µv) = (λµ)v.
(M2) For all u, v ∈ V , λ ∈ R, λ(u + v) = λu + λv.
(M3) For all v ∈ V , λ, µ ∈ R, (λ + µ)v = (λv) + (µv).
(M4) For all v ∈ V , 1v = v.

The properties (A1) – (A4) and (M1) – (M4) are often referred to as the vector
space axioms. Readers who are familiar with group theory, or who have read
the previous section, may have noticed that addition of vectors is a binary
operation on V and that the conditions (A1) to (A4) define an Abelian group
— see definition 5.4. In other words, forgetting about scalar multiplication, the
pair (V, +) is an Abelian group. This means that anything we can prove about
groups will apply to V under the operation of addition. Scalar multiplication
is not a binary operation as it combines two different kinds of things: real
numbers (scalars) and elements of V (vectors). In fact, scalar multiplication
defines a mapping,
R × V → V, (λ, x) 7→ λx.

We now consider some examples of vector spaces and then prove some fairly
straightforward properties of vector spaces.

Examples 5.5
1. The set {( ) }
x1
R =
2
: x1 , x2 ∈ R
x2
is a vector space with addition and scalar multiplication defined above.
212

Similarly, vectors in R3 , which have three components, also form a vec-

tor space with similar definitions for addition and scalar multiplication.
More generally, define
  

 x1 


 x2  

 
R =  .  : x1 , x2 , . . . , xn ∈ R
n

  .  

 .
 

xn
to be the set of n-dimensional vectors. Algebraically, there is nothing
very special about having vectors with n components rather than two
or three. Geometrically, it is harder to visualise n-dimensional space,
of course. Addition and scalar multiplication are defined in Rn in the
obvious way. If
   
x1 y1
 x2   y2 
   
x =  .  , y =  .  ∈ Rn and λ ∈ R,
 . 
.  .. 
xn yn
then    
x1 + y1 λx1
 x2 + y2   λx2 
   
x+y = ..  and λx =  .  .
 .   .. 
xn + yn λxn
With these definitions, R is a vector space. The zero vector is
n
 
0
0
 
0 = .
 .. 
0
and the negative of x is
 
−x1
 −x2 
 
−x =  .  .
 .. 
−xn

2. Let {( ) }
a b
M2×2 (R) = : a, b, c, d ∈ R
c d
be the set of all 2 × 2 real matrices (matrices with real number entries).
Matrices in M2×2 (R) may be added and multiplied by scalars:
( ) ( ) ( )
a11 a12 b b a11 + b11 a12 + b12
+ 11 12 =
a21 a22 b21 b22 a21 + b21 a22 + b22
 213

and ( ) ( )
a a12 λa11 λa12
λ 11 = .
a21 a22 λa21 λa22
With these definitions, M2×2 (R) forms a vector space. (We may also
multiply matrices in M2×2 (R), but this is not relevant so far as forming
a vector space is concerned.)

3. Let F = {functions R → R} be the set of all functions R → R. We may

add two functions and multiply a function by a scalar. Given functions
f and g, define f + g by

f +g :R→R by (f + g)(x) = f (x) + g(x).

Similarly, given f ∈ F and λ ∈ R, we define the function λf by

λf : R → R, (λf )(x) = λf (x).

For example, if f and g are defined by f (x) = x2 and g(x) = sin x, then
(f + g)(x) = f (x) + g(x) = x2 + sin x, (3f )(x) = 3f (x) = 3x2 , and so
on.
With these definitions F is a vector space. The ‘vectors’ are the func-
tions in F . The zero function, satisfying axiom (A3) in definition 5.5, is
zero ∈ F defined by

zero : R → R, zero(x) = 0.

Similarly, the negative of f ∈ F is the function −f ∈ F defined by

−f : R → R, (−f )(x) = −f (x).

We begin by considering some elementary properties of vector spaces. The

following properties of vector spaces give simple interactions between addition
and scalar multiplications of vectors.

Theorem 5.11 (Elementary properties of a vector space)

Let V be a vector space. For all v ∈ V and λ ∈ R, we have:

(i) 0v = 0
(ii) λ0 = 0
(iii) (−1)v = −v
(iv) λv = 0 ⇒ λ = 0 or v = 0 or both.

Consider property (i). This says that taking the scalar multiple of 0 ∈ R with
214

any vector v ∈ V gives the zero vector 0 ∈ V . This might seem very obvious,
especially if we are thinking of R2 where
( ) ( )
x 0
0 1 = = 0.
x2 0

This is not a proof, of course, because it refers to a specific vector space. So,
how do we prove property (i)?
By axiom (A3) of the definition of a vector space, the zero vector satisfies

0 + v = v.

We first show that 0v also satisfies this property:

0v + v = 0v + 1v (by axiom (M4))

= (0 + 1)v (by axiom (M2))
= 1v (property of R)
= v (by axiom (M4)).

Now, if we knew that the zero vector was unique, then the proof is just about
complete. In other words, if we knew that the zero vector was the only vector
satisfying 0 + v = v, then we could deduce 0v = 0. In fact, the zero vector is
unique, but unfortunately we have not proved it. We will consider uniqueness
proofs in section 7.5. This means we need to work a little harder to complete
the proof. The idea is to add −v to both sides of the equation 0v + v = v
established above, as this will give the zero vector on the right-hand side. We
have:
(0v + v) − v = v − v
⇒ 0v + (v − v) = v − v (by axiom (A2))
⇒ 0v + 0 = 0 (by axiom (A4))
⇒ 0v = 0 (by axiom (A3)).
We have now found a chain of reasoning that establishes the result in part (i)
above. We can now present it as a coherent proof as follows.

Proof of (i). Let V be a vector space and let v ∈ V . Then

0v + v = 0v + 1v (by axiom (M4))
= (0 + 1)v (by axiom (M2))
= 1v (property of R)
= v (by axiom (M4)).
Now, adding −v to both sides gives
 215

(0v + v) − v = v − v
⇒ 0v + (v − v) = v − v (by axiom (A2))
⇒ 0v + 0 = 0 (by axiom (A4))
⇒ 0v = 0 (by axiom (A3)).
Therefore 0v = 0, as required.

We will leave the proof of parts (ii) and (iii) as exercises and consider, instead,
part (iv).
To prove λv = 0 ⇒ λ = 0 or v = 0 by the method of direct proof, we assume
λv = 0 and, from this, deduce λ = 0 or v = 0. To establish the conclusion
we show that if λ ̸= 0, then we must have v = 0. In other words, what we
actually prove is
(λv = 0 and λ ̸= 0) ⇒ v = 0.
Essentially we are using the logical equivalence
P ⇒ (Q ∨ R) ≡ (P ∧ Q) ⇒ R,
which is easily established using truth tables.
So suppose λv = 0 and λ ̸= 0. Since λ is non-zero, it has a reciprocal 1/λ.
The idea is to multiply both sides of λv = 0 by 1/λ to obtain v = 0. Having
found the overall structure of the proof, we can fill in the precise details as
follows.

Proof of (iv). Let V be a vector space and let v ∈ V and λ ∈ R be such that
λv = 0.
1
Suppose λ ̸= 0. Then ∈ R, so:
λ
1 1
λv = 0 ⇒ (λv) = 0
λ λ
( )
1 1
⇒ λ v = 0 (by axiom (M1))
λ λ
1
⇒ 1v = 0 (property of R)
λ
1
⇒ v = 0 (by axiom (M4))
λ
⇒ v = 0 (by property (ii)).

In order to look at some more advanced — and more interesting — aspects

of vector spaces, we will first need to define some additional terminology. We
have chosen to explore the properties of subspaces. A subspace of a vector
space V is simply a subset of V that is also itself a vector space.
216

Definition 5.6
Let V be a vector space and let S be a subset of V , S ⊆ V . Then S is
a subspace of V if S is itself a vector space with the same operations of
addition and scalar multiplication that are defined in V .

If we regard R2 as being the subset of R3 comprising those vectors with final

coordinate equal to 0,
  
 x1 
R 2 = x 2  : x 1 , x 2 ∈ R ,
 
0

then R2 is a subspace of R3 since we have noted, in example 5.5.1, that R2 is

a vector space.

Example 5.6
Show that   
 s−t 
S = −s − t : s, t ∈ R
 
2t
is a subspace of R3 . The set S comprises the points lying on a plane in R3
passing through the origin, illustrated in figure 5.10.

x3

x2

x1

FIGURE 5.10: A plane in R3 passing through the origin.

Proof. To show that S is a subspace of R3 , we need to verify that S is a vector

space with the usual operations of addition and scalar multiplication of vectors
in R3 . Before we start checking the conditions (A1)–(A4) and (M1)–M4) in
 217

definition 5.5, we must verify that addition and scalar multiplication are valid
operations in S. For addition, this means that adding two vectors in S gives
a vector that also lies in S.
Let v, w ∈ S. Then, for some real numbers p, q, s, t we have
   
p−q s−t
v = −p − q  and w = −s − t .
2q 2t
Then
   
p−q s−t
v + w = −p − q  + −s − t
2q 2t
   
p−q+s−t (p + s) − (q + t)
= −p − q − s − t = −(p + s) − (q + t) .
2p + 2t 2(q + t)
Hence v + w belongs to S because is can be expressed as
 
a−b
v + w = −a − b ,
2b
where a = p + s ∈ R and b = q + t ∈ R. Hence addition is defined as an
operation on the set S.
Next we verify that scalar multiplication is a well-defined operation on S; in
other words, if v ∈ S and λ ∈ R, then λv ∈ S. With v defined as above, we
have    
p−q λp − λq
λv = λ −p − q  = −λp − λq  ,
2q 2λq
which is an element of S since λp, λq ∈ R.
Having shown that addition and scalar multiplication are bona fide operations
on S, we now need to verify that the axioms (A1)–(A4) and (M1)–M4) in
definition 5.5 are satisfied.

(A1) The equation u + v = v + u is satisfied for all vectors in S because is is

satisfied for all vectors in R3 and S ⊆ R3 .
(A2) Similarly, (u + v) + w = u + (v + w) is satisfied for all vectors in S
because is is satisfied in R3 .
(A3) The zero vector belongs to S since
 
0−0
0 = −0 − 0 .
2×0
218

Clearly, 0 + v = v = v + 0 is satisfied for vectors in S because it is

satisfied for vectors in R3 .
(A4) If v ∈ S is defined as above, then its negative
   
p−q −p − (−q)
−v = − −p − q  = −(−p) − (−q) ∈ S
2q 2(−q)

since −p, −q ∈ R. Then it follows that v + (−v) = 0 because this

equation holds in R3 .
(M1) In a similar way to condition (A1), the equation λ(µv) = (λµ)v holds
for vectors in S (and scalars in R) because it holds in R3 .
(M2) For all u, v ∈ S, λ ∈ R, λ(u + v) = λu + λv, since this holds in R3 .
(M3) Similarly, for all v ∈ S, λ, µ ∈ R, (λ + µ)v = (λv) + (µv), since this
holds in R3 .
(M4) Finally, for all v ∈ S, 1v = v, again because this holds in R3 .

Since addition and scalar multiplication are defined in S and all the axioms
in definition 5.5 are satisfied, S is a vector space. Therefore S is a subspace
of R3 .

Much of the proof that S is a subspace of R3 in the previous example can be

categorised as ‘this condition holds in the subset S because it holds in R3 ’.
Essentially, once we had established that addition and scalar multiplication
were well defined in S, the vector space axioms follow for S because they hold
in the larger space R3 . This is formalised in the following theorem, which we
will not prove. Any vector space must be non-empty since it has to contain
the zero vector, and this is the reason for the first condition in the theorem.

Theorem 5.12 (Subspace Test)

Let S ⊆ V where V is a vector space. If S is

(i) non-empty: S ̸= ∅,
(ii) closed under addition: v, w ∈ S ⇒ v + w ∈ S,
(iii) closed under scalar multiplication: v ∈ S and λ ∈ R ⇒ λv ∈ S,

then S is a subspace of V .

The second and third properties are usually summarised by stating that S is
closed under addition and scalar multiplication. We will now use theorem 5.12
to explore the properties of subspaces.
 219

Theorem 5.13 (Intersection of subspaces)

Let S and T be subspaces of a vector space V . Then their intersection S ∩ T
is a subspace of V .

Theorem 5.12 provides a generic framework for proving that a subset of a

vector space is a subspace. If we can verify that the conditions (i), (ii), and
(iii) of the theorem are satisfied for a particular set, then it follows that the
set is a subspace.

Proof. Let S and T be subspaces of a vector space V . Recall that S ∩ T =

{v : v ∈ S and v ∈ T }.

(i) Each subspace contains the zero vector, 0 ∈ S and 0 ∈ T . Hence 0 ∈

S ∩ T , from which it follows that S ∩ T is non-empty, S ∩ T ̸= ∅.
(ii) Let v, w ∈ S ∩ T .
Then v, w ∈ S and hence, as S is a vector space, v + w ∈ S. Similarly,
v, w ∈ T and so, as T is a vector space, v + w ∈ T .
Therefore v + w ∈ S ∩ T .
(iii) Let v ∈ S ∩ T and λ ∈ R.
Then v ∈ S and hence, as S is a vector space, λv ∈ S. Similarly, v ∈ T
and so, as T is a vector space, λv ∈ T .
Therefore λv ∈ S ∩ T .

It now follows, from theorem 5.17, that S ∩ T is a subspace of V .

Example 5.7
Let A be an n × n square matrix and let λ ∈ R. Consider the equation
Ax = λx,
where x ∈ Rn . For most values of λ, the equation will only have the zero
solution, x = 0.
( )
3 −1
For example, let A = and let λ = 5. Then
4 −2
( )( ) ( ) {
3 −1 x1 x 3x1 − x2 = 5x1
Ax = =5 1 ⇒
4 −2 x2 x2 4x1 − 2x2 = 5x2
{
2x1 − x2 = 0 (i)
⇒
4x1 − 7x2 = 0 (ii)
⇒ 5x2 = 0 (2 × (i) − (ii))
⇒ x2 = 0
⇒ x1 = 0 (from (i)).
220

Hence, when λ = 5, the only solution to Ax = λx is x = 0.

However there may be particular values for λ for which the equation Ax = λx
has non-zero solutions x ̸= 0. For example,
( )( ) ( ) ( )
3 −1 1 2 1
= =2
4 −2 1 2 1
so, for λ = 2 there is a non-zero solution to the equation Ax = λx, namely
( )
1
x= .
1
Here the value λ = 2 is called an eigenvalue of the matrix A and any non-
zero vector satisfying Ax = 2x is called an eigenvector corresponding to
eigenvalue 2.
In fact, any vector of the form
()
t
x=
t
where t ∈ R and t ̸= 0 is an eigenvector corresponding to λ = 2 since it also
satisfies Ax = 2x:
( )( ) ( ) ()
3 −1 t 2t t
= =2 .
4 −2 t 2t t

The matrix A has another eigenvalue. Since

( )( ) ( ) ( )
3 −1 1 −1 1
= =− ,
4 −2 4 −4 4
it follows that λ = −1 is an eigenvalue and
( )
1
x=
4
is a corresponding eigenvector. Again, it is easy to verify that any vector of
the form ( )
t
x=
4t
where t ∈ R and t ̸= 0 is an eigenvector corresponding to λ = −1 since it also
satisfies Ax = −x. Although we shall not prove this, the values 2 and −1 are
the only eigenvalues for the matrix A; for any other value of λ, the equation
Ax = λx only has the zero solution x = 0.
Note that, for each eigenvector λ, the set E(λ), of all eigenvectors (together
with 0) is a line in R2 through the origin:
{( ) } {( ) }
t t
E(2) = :t∈R and E(−1) = :t∈R .
t 4t
These sets are shown in figure 5.11.
 221

E(–1)
E(2)

FIGURE 5.11: The sets of eigenvectors E(2) and E(−1).

Recall that the plane passing through the origin given in example 5.6 was a
subspace of R3 . It is straightforward to use theorem 5.12 to show that each
of the sets E(2) and E(−1) is a subspace of R2 ; this is left as an exercise.
Instead, we consider the corresponding result for any eigenvalue of any square
matrix.

Theorem 5.14 (Space of eigenvectors)

Let A be an n × n matrix and let λ be an eigenvalue of A. In other words,
the equation Ax = λx has non-zero solutions for x ∈ Rn . Then the set of all
eigenvectors corresponding to λ (together with 0),

E(λ) = {x ∈ Rn : Ax = λx},

is a subspace of Rn .

The set E(λ) is called the eigenspace corresponding to λ.

The Subspace Test, theorem 5.12, again provides the framework for the proof.
This says that, to show E(λ) is a subspace of Rn , we need to show that it
is non-empty, closed under addition, and closed under scalar multiplication.
Consider the condition of closure under addition: if x, y ∈ E(λ), then x + y ∈
E(λ). We will give a direct proof of this, so begin by assuming x, y ∈ E(λ).
We need to find a chain of deductions ending with x + y ∈ E(λ):

x, y ∈ E(λ) ⇒ ···
⇒ x + y ∈ E(λ).

The first step in the deduction is to understand what it means to say x and y
belong to E(λ). It is often the case that the first step in the deduction amounts
to an ‘unpacking’ of the initial assumption. By the definition of E(λ), this
means that Ax = λx and Ay = λy. To prove that the sum x + y ∈ E(λ),
we need to show that x + y also satisfies the same equation defining E(λ):
222

A(x + y) = λ(x + y). This follows by adding the two equations for x and y
and using the following properties of matrices and vectors:
Ax + Ay = A(x + y) and λx + λy = λ(x + y).
These considerations allow us to construct a simple chain of deductions from
x, y ∈ E(λ) to x + y ∈ E(λ), which we include in the proof below.
The proof that E(λ) is closed under scalar multiplication is similar to that
for addition, so we include it directly in the proof without the preamble given
here.

Proof. Let A be an n × n matrix and let λ be an eigenvalue of A.

First note that A0 = 0 = λ0, so 0 ∈ E(λ). Hence E(λ) ̸= ∅.
Let x, y ∈ E(λ). Then Ax = λx and Ay = λy. Hence
A(x + y) = Ax + Ay (property of matrix multiplication)
= λx + λy (since Ax = λx and Ay = λy)
= λ(x + y) (property of scalar multiplication).
It now follows that x + y ∈ E(λ), so E(λ) is closed under addition.
Now let x ∈ E(λ), as above, and let α ∈ R. We wish to show that αx ∈ E(λ):
A(αx) = α(Ax) (property of matrix multiplication)
= α(λx) (since Ax = λx)
= λ(αx) (property of scalar multiplication).
Hence αx ∈ E(λ) so E(λ) is closed under scalar multiplication.
We have shown that E(λ) is non-empty, closed under addition, and closed
under scalar multiplication. Therefore, by the Subspace Test, theorem 5.12,
E(λ) is a subspace of Rn .

Note that, in showing E(λ) is closed under addition and scalar multiplica-
tion, we are using the properties of a vector space for Rn . For addition, the
equation λx + λy = λ(x + y) is just condition (M2) of definition 5.5. For
scalar multiplication, our proof left out the details of establishing the equa-
tion α(λx) = λ(αx). It would be quite common to leave out the details but,
for completeness, here is the full deduction:
α(λx) = (αλ)x (property (M1) of a vector space)
= (λα)x (commutativity of multiplication in R)
= λ(αx) (property (M1) of a vector space).

We conclude this section by proving one further result about eigenvalues and
eigenvectors.
 223

Theorem 5.15 (Eigenspace of the square of a matrix)

Let A be an n × n matrix and let λ be an eigenvalue of A. Then λ2 is an
eigenvalue of the matrix A2 .

Furthermore, the eigenspace of λ for the matrix A, EA (λ), is a subspace of

the eigenspace of λ2 for the matrix A2 , EA2 (λ2 ).

Before we consider the proof, it is worth remarking that we have had to

extend our previous notation for eigenspaces to take account of the fact that
there are two different matrices involved here. Since we need to consider the
eigenspaces of different matrices, we need to extend the previous notation for
the eigenspace to refer also to the matrix.

For the first part of the theorem, we need to find a chain of deductions from ‘λ
is an eigenvalue of A’ to ‘λ2 is an eigenvalue of the matrix A2 ’. Now ‘λ is an
eigenvalue of A’ means that Ax = λx for some non-zero vector x. Similarly,
‘λ2 is an eigenvalue of A2 ’ means that A2 y = λ2 y for some non-zero vector
y. We have used different letters for the vectors x and y as we cannot assume
that the same vector satisfies both equations. Thus a partial proof has the
following structure:
λ is an eigenvalue of A
⇒ Ax = λx for some non-zero vector x
⇒ ···
⇒ A2 y = λ2 y for some non-zero vector y
⇒ λ2 is an eigenvalue of A2 .

To complete the missing steps, consider A2 x. Without, at this stage, worrying

too much about the detailed justifications, we have

A2 x = A(Ax) = A(λx) = λ(Ax) = λ(λx) = λ2 x. (*)

Provided we can justify each of these equations, this completes the deduction
chain.

We now consider the second part of the theorem. The eigenspaces are

EA (λ) = {x ∈ Rn : Ax = λx},
EA2 (λ2 ) = {x ∈ Rn : A2 x = λ2 x}.

From the previous theorem, theorem 5.14, we know that each of these is a
subspace of Rn . Hence, to show that EA (λ) is a subspace of EA2 (λ2 ), we only
need to show that EA (λ) is a subset of EA2 (λ2 ), EA (λ) ⊆ EA2 (λ2 ).

Recall that, for two sets A and B, A ⊆ B means that, for all x, x ∈ A ⇒ x ∈ B.
224

Taking into account the definitions of the eigenspaces, this gives the following
structure to the proof:

x ∈ EA (λ) ⇒ Ax = λx
⇒ ···
⇒ A2 x = λ2 x
⇒ x ∈ EA2 (λ2 ).

However, the missing part of the reasoning is captured by the equations (*)
above. We can now piece together the different parts of the reasoning and give
a proof of the theorem.

Proof. Let A be an n × n matrix and let λ be an eigenvalue of A. Then, by

definition, there exists an eigenvector x ∈ Rn which is a non-zero vector such
that Ax = λx. Therefore

A2 x = A(Ax)
= A(λx) (since Ax = λx)
= λ(Ax) (properties of matrix multiplication)
= λ(λx) (since Ax = λx)
2
= λ x.

This shows that x is also an eigenvector corresponding to λ2 for the matrix

A2 . In particular, λ2 is an eigenvalue of A2 , which is the first part of the
theorem.
The reasoning above shows that

Ax = λx ⇒ A2 x = λ2 x,

so EA (λ) ⊆ EA2 (λ2 ). Since each of these sets is a subspace of Rn (by theo-
rem 5.14), it follows that EA (λ) is a subspace of EA2 (λ2 ), which is the second
part of the theorem.

Exercises 5.3

1. Prove theorem 5.8 (ii), the Right Cancellation Property for groups.

2. Prove the identity missing from the proof of theorem 5.9: for elements
x and y of a group (G, ∗), (y −1 ∗ x−1 ) ∗ (x ∗ y) = e.
 225

3. Similar to the definition of a subspace of a vector space, definition 5.6,

we may define a subgroup of a group (G, ∗) to be a subset H ⊂ G that
is also a group under the same binary operation; (H, ∗) is a group.
Under this definition, both ({e}, ∗) and (G, ∗) are subgroups of (G, ∗).
Any other subgroup different from these is called a proper subgroup
of (G, ∗).
Prove the following subgroup test.

Theorem 5.16 (Subgroup Test)

Let (G, ∗) and let H be a subset of G that satisfies:
(i) H is non-empty: H ̸= ∅;
(ii) H is closed under ∗: for all x, y ∈ H, x ∗ y ∈ H;
(iii) H is closed under taking inverses: for all x ∈ H, x−1 ∈ H.
Then (H, ∗) is a subgroup of (G, ∗).

4. Let H and K be two subgroups of a group G. Prove that their intersec-

tion H ∩ K is a subgroup of G.
Hint: you may use the subgroup test, theorem 5.16, from the previous
example.
5. A set S has a binary operation ∗ that satisfies the following two axioms.
(A1) There is an identity element e ∈ S.
(A2) For all x, y, z ∈ S, (x ∗ y) ∗ z = x ∗ (z ∗ y).
Show that ∗ is both commutative and associative on S.
6. Prove properties (ii) and (iii) in theorem 5.11.
7. Let S and T be two subspaces of a vector space U . Define the sum of
S and T to be

S + T = {x + y : x ∈ S and y ∈ T }.

Prove that S + T is a subspace of U .

8. Using theorem 5.12, show that each of the sets E(2) and E(−1), given
in example 5.7, is a subspace of R2 .
9. Let λ be an eigenvalue of an n × n matrix A.
(i) Prove that λ + 1 is an eigenvalue of the matrix A + In .
The matrix In is the n × n matrix with 1s along the leading top-left
to bottom-right diagonal and 0s elsewhere. It is called the n × n
identity matrix as it satisfies the property that, for any n × n
matric M, MIn = In M = M.
226

(ii) Let EA (λ) denote the eigenspace of λ for the matrix A and let
EA+In (λ + 1) denote the eigenspace of λ + 1 for the matrix A + In .
Show that EA (λ) = EA+In (λ + 1).
(iii) Let k be a positive integer. Generalise the result of part (i) for the
eigenvalue λ + k. Prove your generalisation.

10. Let A be an m × n matrix. The null space of A is N (A) =

{x ∈ Rn : Ax = 0}.
Show that the N (A) is a subspace of Rn .

11. Let A be an n × n matrix. Writing x ∈ Rn as a column vector, we may

define a mapping

T : Rn → Rn by T (x) = Ax.

The mapping T is called a linear transformation;6 it satisfies the

following two properties:
(LT1) for all x, y ∈ Rn , T (x + y) = T (x) + T (y);
(LT2) for all x ∈ Rn and α ∈ R, T (αx) = αT (x).

(i) Prove that the set ker T = {x ∈ Rn : T (x) = 0} is a subspace of

Rn . The space ker T is called the kernel of T .
(ii) Prove that the image of T , im T = {T (x) : x ∈ Rn }, is a subspace
of Rn .

5.4 Examples from analysis

In this section we will consider two important topics in real analysis: conver-
gence of sequences and limits of functions. The key notion of a limit is similar
in each of the two contexts. The definition of a limit involves a statement with
two quantifiers which is of the form

∀x ∃y • P (x, y).

It is probably the interaction between the two quantifiers that makes proofs
involving limits somewhat tricky. As we discussed in section 5.2, when consid-
ering the proof of surjectivity of a function, proofs of statements of this form
with a universally and existentially quantified variable, often require some
6 Any mapping between arbitrary vector spaces that satisfies the properties (LT1) and

(LT2) is called a linear transformation.

 227

preliminary ‘discovery work’ before the proof itself can commence. As this
discovery work is not part of the proof itself, it is sometimes omitted in an
explanation of the proof and, when this occurs, some of the details of the proof
itself look somewhat arbitrary or even mysterious.

5.4.1 Sequences

Intuitively a sequence of real numbers is an infinite list of real numbers

a1 , a2 , a3 , a4 , . . .. It is often helpful to capture the idea of an infinite list as
a single mathematical concept. With this in mind, consider how we might
define a function f : Z+ → R. We might have a rule for f (n) but, if not, we
would need to specify each of the images f (1), f (2), f (3), . . .. In other words,
defining a function Z+ → R is akin to specifying an infinite list of real num-
bers. Hence it is usual to define a sequence more formally as a function, as
follows.

Definition 5.7
A sequence (of real numbers) is a function a : Z+ → R. The image of n ∈ Z+ ,
a(n), is frequently denoted an . The whole sequence is often abbreviated as
(an ).

Often, we are interested in the long-term behaviour of sequences. The graph

of a sequence a : Z+ → R just plots the points an ; it is often referred to as a
sequence diagram. Figure 5.12 shows some possible behaviours of sequences.

• Sequence 1 is periodic: it endlessly cycles between a collection of values,

in this case 21 , 12 , 1, 12 , 12 , 1, 12 , 12 , 1, . . ..

• Sequences 2 and 4 converge: the values get closer and closer to some
limiting value, 1 in the case of Sequence 2 and 21 in the case of Sequence 4.

• Sequence 3 is chaotic: there is no discernible pattern to its values.

We aim to prove some properties of convergent sequences. In order to do so,

we first need a precise definition of convergent. Intuitively, a sequence (an )
converges to a limit ℓ if the terms eventually get ‘as close as possible’ to ℓ.
To make this notion precise we need to quantify ‘eventually’ and ‘as close as
possible’. Saying that the terms come ‘as close as possible’ to ℓ means that
given any positive quantity ε (no matter how small) we can make the distance
between an and ℓ less than ε. Since |an − ℓ| represents the distance between
an and ℓ, this means that ‘eventually’ we have |an − ℓ| < ε. By ‘eventually’ we
just mean that this inequality is satisfied provided we go beyond some point
N in the sequence. These considerations lead to the following definition.
228

1 1

0.5 0.5

0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30

Sequence 1 Sequence 2

1 1

0.5 0.5

0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30

Sequence 3 Sequence 4

FIGURE 5.12: Possible behaviour of sequences.

Definition 5.8
The sequence (an ) converges to limit ℓ if, given any ε > 0, there exists
N ∈ Z+ such that
n > N ⇒ |an − ℓ| < ε.
In this case we write either an → ℓ as n → ∞ or lim an = ℓ.
n→∞

Definition 5.8 is illustrated in figure 5.13. The inequality |an −ℓ| < ε is satisfied
provided ℓ − ε < an < ℓ + ε; in other words, when an lies in a ‘band’ of width ε
either side of ℓ. The figure is intended to represent the fact that we can ensure
that the terms of the sequence lie in the given the ‘ε-band’ about the limit ℓ
provided we take n greater than the specified value N .

Example 5.8
3n2 − 4n
A sequence (an ) is defined by an = for n ∈ Z+ . Show that
(n + 1)(n + 2)
lim an = 3.
n→∞

Solution
Here the limit is ℓ = 3. Let ε > 0 be an arbitrary positive real number. We
 229

FIGURE 5.13: Illustrating the definition of a convergent sequence.

need to find a positive integer N such that, for all n > N ,

 
 3n2 − 4n 

|an − ℓ| =  − 3 < ε.
(n + 1)(n + 2)

The question is: how can we identify a suitable integer N ?

To answer this, we manipulate the expression for |an − ℓ|, making various
choices for n until we can be assured that the expression is less than any
positive ε. We proceed as follows:
   2 
 3n2 − 4n   
 − 3  =  3n − 4n − 3(n + 1)(n + 2) 
 (n + 1)(n + 2)   (n + 1)(n + 2) 
 2 
 3n − 4n − (3n2 + 9n + 6) 
=  

(n + 1)(n + 2)
 
 −13n − 6 
=  
(n + 1)(n + 2) 
13n + 6
= .
(n + 1)(n + 2)

Now we wish to show that the expression

13n + 6
(n + 1)(n + 2)
is less than or equal to some simpler fraction which can still be made less than
ε. For example, we could replace the numerator 13n + 6 with 13n + 6n = 19n
since n ≥ 1. Just as replacing the numerator with a larger value increases the
size of the expression, so does replacing the denominator with a smaller value.
Hence we have:
13n + 6 19n 19n 19
≤ ≤ 2 = .
(n + 1)(n + 2) (n + 1)(n + 2) n n
230

We can ensure that the 19/n is as small as we please by choosing n sufficiently

large. In particular, to ensure 19/n < ε we may take n > 19/ε.
It is tempting to believe that we have now found the required value of N ,
namely 19/ε. However there is one further complication. Definition 5.8 requires
the value of N to be a positive integer and there is no guarantee that 19/ε
will be an integer.
However, this is easy to fix: we take N to be the integer part or floor of
19/ε. This is defined to be the largest integer less than or equal to 19/ε and
is denoted ⌊ ⌋
19
.
ε
Now that we have found a suitable value for N , we can at last proceed to the
proof itself. The proof amounts to checking that definition 5.8 is satisfied for
the given sequence.
⌊ ⌋
19
Proof. Let ε > 0. Now let N = ∈ Z+ . Then, for all integers n > N , we
ε
have    2 
 3n2 − 4n   3n − 4n − 3(n + 1)(n + 2) 
   
 (n + 1)(n + 2) − 3 =  (n + 1)(n + 2) 
 2 
 3n − 4n − (3n2 + 9n + 6) 
=  

(n + 1)(n + 2)
 
 −13n − 6 
=  
(n + 1)(n + 2) 
13n + 6
=
(n + 1)(n + 2)
13n + 6n
≤
(n + 1)(n + 2)
19n
≤
n2
19
=
n
19
< ε since n > .
ε
3n2 − 4n
Therefore lim = 3.
n→∞ (n + 1)(n + 2)

It is worth remarking that, taken in isolation, the beginning of the proof

⌊ ⌋
19
Let ε > 0. Now let N = ∈ Z+
ε
 231

does look strange. Without the algebraic manipulation before the proof, the
choice of N = ⌊19/ε⌋ looks curious in the extreme. This is often the case in
proofs of this type. If we do not show the ‘discovery phase’, which is not part
of the proof itself, then the choices made in the proof can seem somewhat
arbitrary.
Notice that the overall structure of the proof is simply to verify that the con-
ditions of definition 5.8 for a convergent sequence are satisfied by the given
sequence. As we have noted, the definition of convergence involved two quan-
tifiers and is of the form
∀ε ∃N • P (ε, N )
where P (ε, N ) is a propositional function and the universes for ε and N are
R+ and Z+ , respectively. The structure of the proof starts with an arbitrary
ε (in its universe) and then selects a particular N (in its universe) which
may depend on ε. To complete the proof we show that the propositional
function P (ε, N ), which is itself a conditional, is satisfied for the arbitrary
ε and particular N . The principal difficulty in the proof is in selecting an
appropriate N , depending on ε, but this occurs outside the proof itself in the
preliminary discovery phase of the work.

We now prove a general theorem about limits of sequences. This theorem says,
roughly, that limits ‘behave well’ with respect to addition of sequences. Given
two sequences (an ) and (bn ), we may form their sum (an + bn ), which is also
a sequence. More precisely, the theorem says that if (an ) and (bn ) are both
convergent, then their sum is also convergent and, furthermore, the limit of
(an + bn ) is the sum of the limits of (an ) and (bn ).

Theorem 5.17 (Sum of convergent sequences)

Let (an ) and (bn ) be two convergent sequences with lim an = ℓ and lim bn =
n→∞ n→∞
m.
Then the sum (an + bn ) is convergent and lim an + bn = ℓ + m.
n→∞

As in the previous example, we will need to carry out some preliminary work
to understand how the proof will proceed. According to definition 5.8, we need
to ensure that, given any ε > 0,

|(an + bn ) − (ℓ + m)| < ε

provided we take n greater than some positive integer N . The triangle in-
equality for the modulus function (theorem 4.6, page 146) gives

|(an + bn ) − (ℓ + m)| = |(an − ℓ) + (bn − m)| ≤ |an − ℓ| + |bn − m|.

Since limn→∞ an = ℓ and limn→∞ bn = m, we are able to control the sizes of

232

both |an − ℓ| and |bn − m| by taking n sufficiently large. In particular, we can

ensure each of these terms is less than ε/2 by ‘going far enough’ along the
sequences (an ) and (bn ):
ε ε
n > N1 ⇒ |an − ℓ| < and n > N2 ⇒ |bn − m| <
2 2
for some N1 , N2 ∈ Z+ . However, we need to ensure both of these inequalities
hold, so we may take our integer N to be the larger of N1 and N2 . We are
now ready to embark on the proof.

Proof. Suppose that (an ) and (bn ) are two convergent sequences with
lim an = ℓ and lim bn = m.
n→∞ n→∞

Let ε > 0.
Then ε/2 > 0, so there exist N1 , N2 ∈ Z+ such that
ε ε
n > N1 ⇒ |an − ℓ| < and n > N2 ⇒ |bn − m| < .
2 2
Let N = max{N1 , N2 }. Then

n>N ⇒ n > N1 and n > N2

ε ε
⇒ |an − ℓ| < and |bn − m| <
2 2
⇒ |(an + bn ) − (ℓ + m)| = |(an − ℓ) + (bn − m)|
≤ |an − ℓ| + |bn − m|
ε ε
< + = ε.
2 2
Therefore (an + bn ) is convergent and lim an + bn = ℓ + m.
n→∞

5.4.2 Limits of functions

The notion of a limit of a function has similarities with the convergence of

a sequence. Suppose that f : A → B is a function where the domain and
codomain are subsets of the real numbers, A ⊆ R and B ⊆ R. Let a ∈ A be
an element of the domain.7 Intuitively, we say that f converges to a limit ℓ
as x tends to a provided we may make f (x) come ‘as close as we choose’ to
ℓ by taking x ‘sufficiently close’ to a. Just as for sequences, saying that the
value of f (x) is ‘as close as we choose’ to ℓ means ensuring that |f (x) − ℓ| is
less than some specified positive quantity ε. Similarly, taking x ‘sufficiently
7 In fact, it is not strictly necessary that a ∈ A. What is required is that the domain

contains open intervals on either side of a: (p, a) ∪ (a, q) ⊆ A for some p < a and q > a.
 233

close’ to a means requiring that |x − a| is less than some (other) positive

value, traditionally labelled δ. There is one small modification, however. In
the limiting process as x tends towards a, we will not allow x actually to be
equal to a; indeed, as we indicated in the footnote, it may be that f is not
actually defined at a. These considerations lead to the following definition.

Definition 5.9
Let f : A → B be a function where A ⊆ R and B ⊆ R. Let a ∈ A. Then f (x)
tends towards a limit ℓ as x tends towards a if, given any ε > 0, there exists
δ > 0 such that
0 < |x − a| < δ ⇒ |f (x) − ℓ| < ε.
In this case, we either write f (x) → ℓ as x → a or we write lim f (x) = ℓ.
x→a

The values of x satisfying 0 < |x − a| < δ lie in the union of open intervals
either side of a:
(a − δ, a) ∪ (a, a + δ).
This set is often referred to as a punctured neighbourhood of a. Defini-
tion 5.9 is illustrated in figure 5.14. In order to ensure that the values of f (x)
lie in the horizontal band of width ε either side of the limit ℓ, we need to take
x to lie in the punctured neighbourhood of width δ either side of a.

FIGURE 5.14: Illustrating the definition of a limit of a function.

Example 5.9
Show that lim 2x2 − 5x = −2.
x→2

Solution
In this example, we need to show that definition 5.9 is satisfied for f (x) =
234

2x2 − 5x, ℓ = −2 and a = 2. As with the proofs for sequences, we need to do

some preliminary ‘discovery work’. We need to ensure that

|f (x) − ℓ| = |2x2 − 5x + 2|

is less than any specified positive ε by taking x such that 0 < |x − 2| < δ.
Hence we begin by considering |2x2 − 5x + 2|:

|2x2 − 5x + 2| = |(x − 2)(2x − 1)| = |x − 2||2x − 1|.

We will be able to control the size of the term |x − 2| since we can choose δ
such that |x − 2| < δ. The term |2x − 1| causes a little more difficulty, but we
can also control its size if we re-write it in terms of |x − 2| using the triangle
inequality (theorem 4.6, page 146) as follows.

|2x − 1| = |2(x − 2) + 3|
≤ |2(x − 2)| + |3| (by the triangle inequality)
= 2|x − 2| + 3.

Therefore

|2x2 − 5x + 2| = |x − 2||2x − 1| ≤ |x − 2|(2|x − 2| + 3)

and we consider each of the terms |x − 2| and 2|x − 2| + 3 separately. Provided

|x − 2| < 1, say, the second term 2|x − 2| + 3 < 5. (Taking |x − 2| to be less
than 1 here is somewhat arbitrary; we really just need it to be no greater than
some value. Later we will consider other choices at this step.) If, in addition,
|x − 2| < ε/5, then the product |x − 2|(2|x − 2| + 3) will be less than ε, which
is what we are trying to ensure. We can now embark on the proof itself.
{ ε}
Proof. Let ε > 0. Now let δ = min 1, .
5
Then

0 < |x − 2| < δ
ε
⇒ |x − 2| < 1 and |x − 2| <
5
⇒ |2x2 − 5x + 2| = |x − 2||2x − 1|
≤ |x − 2|(2|x − 2| + 3) (triangle inequality)
< 5|x − 2| (since |x − 2| < 1)
( ε)
< ε since |x − 2| < .
5

Therefore lim 2x2 − 5x = −2.

x→2
 235

Just as with example 5.8, the first line of the proof would look mysterious
without having seen the pre-proof ‘discovery’ process to find a suitable choice
for δ. In fact, because there are two requirements for δ, there are other choices
that will work. For example, suppose that, at the stage where we had the
inequality |2x2 − 5x + 2| = |x − 2||2x − 1| ≤ |x − 2|(2|x − 2| + 3), we had taken
|x−2| < 12 (instead of |x−2| < 1). Then the term 2|x−2|+3 < 2× 12 +3 = 4, so
we have |2x2 − 5x + 2| < 4|x − 2|, which means we also require
{ |x
} − 2| < ε/4.
This means that an alternative choice for δ is δ = min 21 , ε/4 . With this
choice, the last part of the proof becomes

|2x2 − 5x + 2| = |x − 2||2x − 1|
≤ |x − 2|(2|x − 2| + 3) (triangle inequality)
( )
< 4|x − 2| since |x − 2| < 12
( ε)
< ε since |x − 2| < .
4

In Theorem 5.17 we proved that the limit of sequences behaves well with re-
spect to addition; more specifically, the limit of the sum of sequences is the
sum of their limits. There is a similar result for limits of functions and the
proof follows a similar pattern. We leave this as an exercise. Instead we prove
that the limit of a function behaves well with respect to taking multiples.

Theorem 5.18 (Limit of a multiple of a function)

Let f be a function such that lim f (x) = ℓ and let α ∈ R. Then the function
x→a
αf satisfies lim αf (x) = αℓ.
x→a

Again we will carry out some preliminary work. In the proof we will need to
show that |αf (x) − αℓ| < ε and, because limx→a f (x) = ℓ, we can control the
size of |f (x) − ℓ|. Now |αf (x) − αℓ| = |α||f (x) − ℓ|, which will be less than ε
provided |f (x) − ℓ| < ε/|α|. Since ε/|α| is only defined when α ̸= 0, we will
consider α = 0 as a special case.
We are now ready to embark on the proof.

Proof. Suppose lim f (x) = ℓ and let α ∈ R.

x→a

If α = 0, then αf (x) = 0 for all x ∈ R. Hence, for any ε > 0,

|αf (x) − αℓ| = 0 < ε

for all x in the domain of f . Therefore lim αf (x) = 0 = αℓ in this case.

x→a

Now suppose α ̸= 0.
236
ε
Let ε > 0. Then > 0.
|α|
Since lim f (x) = ℓ it follows that there exists δ > 0 such that
x→a

ε
0 < |x − a| < δ ⇒ |f (x) − ℓ| < .
|α|
ε
Hence 0 < |x − a| < δ ⇒ |f (x) − ℓ| <
|α|
ε
⇒ |αf (x) − αℓ| = |α||f (x) − ℓ| < |α| × = ε.
|α|
Therefore lim αf (x) = αℓ in this case too.
x→a

In fact, there is a slightly different proof that does not require separate con-
sideration of the case where α = 0. Note that |α| + 1 > 0 for all α. Thus, if
we replace ε/|α| by ε/(|α| + 1) in the second part of the proof, we can avoid
the special case. This leads to the following alternative proof.

Alternative Proof. Suppose lim f (x) = ℓ and let α ∈ R.

x→a
ε
Let ε > 0. Then > 0.
|α| + 1
Since lim f (x) = ℓ it follows that there exists δ > 0 such that
x→a

ε
0 < |x − a| < δ ⇒ |f (x) − ℓ| < .
|α| + 1

Hence 0 < |x − a| < δ

ε
⇒ |f (x) − ℓ| <
|α| + 1
ε
⇒ |αf (x) − αℓ| = |α||f (x) − ℓ| < |α| × < ε.
|α| + 1
Therefore lim αf (x) = αℓ.
x→a

Exercises 5.4

1. For each of the following sequences, give an ‘ε-N ’ argument to prove

lim an = ℓ.
n→∞
( )
3n + 4
(i) (an ) = ; ℓ=3
5n + 1
 237
( )
n3 + 3n
(ii) (an ) = ;ℓ = 14
4n3 + 7n2
( )
2n2 + 4 2
(iii) (an ) = ; ℓ= 3
(n + 1)(3n + 2)
( 4 )
2n − 6
(iv) (an ) = ; ℓ=2
n4 − 2n

2. Let (an ) be a convergent sequence with lim an = ℓ. Also let λ ∈ R.

n→∞
Prove that (λan ) converges and lim λan = λℓ.
n→∞

3. Use the ‘ε-δ’ definition to prove each of the following limits.

1 1
(i) lim 3x − 7 = −4 (v) lim =
x→1 x→2 x 2
√ √
(ii) lim x − 4x = −3
2
(vi) lim x = 2
x→3 x→2
2
(iii) lim 2x + x = 1 (vii) lim x3 = a3
x→−1 x→a

(iv) lim 3x2 + 5x = 8 (viii) lim x sin x = 0

x→1 x→0

4. Prove that the limit of functions behaves well with respect to addition.
More precisely, suppose that f and g are two functions such that

lim f (x) = ℓ and lim g(x) = m.

x→a x→a

Prove that lim f (x) + g(x) = ℓ + m.

x→a

5. In this question we will prove that, if f and g are two functions such
that
lim f (x) = ℓ and lim g(x) = m,
x→a xaraa

then lim f (x)g(x) = ℓm.

x→a

(i) Show that |f (x)g(x) − ℓm| ≤ |g(x)||f (x) − ℓ| + |ℓ||g(x) − m|.

(ii) Show that there exists δ1 > 0 such that

0 < |x − a| < δ1 ⇒ |g(x)| < 1 + |m|.

(iii) Show that, given ε > 0, there exists δ2 > 0 such that
ε
0 < |x − a| < δ2 ⇒ |f (x) − ℓ| < .
2(1 + |m|)

(iv) Suppose ℓ ̸= 0. Show that, given ε > 0, there exists δ3 > 0 such
that
ε
0 < |x − a| < δ3 ⇒ |g(x) − m| < .
2|ℓ|
238

(v) Use parts (i) to (iv) to prove lim f (x)g(x) = ℓm in the case where
x→a
ℓ ̸= 0.
(vi) Modify your proof in part (v) for the case where ℓ = 0.
Chapter 6
Direct Proof: Variations

6.1 Introduction

In some situations it is not possible, or not easy, to find a direct proof of a

theorem. The nature of the statement of the theorem may make it difficult
to find a chain of deductions ending with the appropriate conclusion. A sim-
ple example will illustrate this. In example 1.3 (page 11), we considered the
statement ‘for all integers n, if n is odd, then n2 is odd ’ and we gave what we
now recognise as a direct proof. Consider now the converse statement (where,
for simplicity, we have restricted the universe to be the positive integers):

for all positive integers n, if n2 is odd, then n is odd.

How might we prove this? A direct proof would start by assuming ‘n2 is odd’
for an arbitrary positive integer n and, from this, attempt to deduce ‘n is
odd’. Thus a direct proof would have the following structure.

Proof. Suppose that n is a positive integer such that n2 is

odd. Then n2 = 2k + 1 for some integer k (where k ≥ 0).
..
.
some chain of deductions
..
.
Therefore n is odd.

The problem here is that it is not at all clear what might be a sensible de-
duction from the statement ‘n2 = 2k + 1’. Knowing that we need to make
a conclusion about n itself (namely n is odd), we might be tempted to take
square roots and write
√
n = 2k + 1 for some integer k ≥ 0.
√
We have now hit a dead end: there is no obvious way of showing that 2k + 1
is an odd positive integer. The problem here is not so much in our ability to
make appropriate deductions as the overall structure of the attempted proof.

239
240

This is because it is much more difficult to deduce information about an integer

n (such as n is odd) from information about its square than it is the other
way around. In other words, starting from ‘n2 is odd’ it is very difficult to find
a chain of deductions that allows us to deduce ‘n is odd’.
In this chapter, we will introduce methods of proof that may be useful when
the method of direct proof appears problematic. In each case, the method will
follow the pattern:

instead of proving a statement P , we may instead prove a

statement P ′ that is logically equivalent to P .

Recall that two propositions are logically equivalent if they are true under the
same conditions. Thus is it perfectly reasonable to prove a logically equivalent
statement if this is easier to achieve.

6.2 Proof using the contrapositive

Recall the Contrapositive Law from section 2.3, page 40, that states that a
conditional proposition P ⇒ Q is logically equivalent to its contrapositive
¬Q ⇒ ¬P :
P ⇒ Q ≡ ¬Q ⇒ ¬P.
This means that, to prove a conditional proposition P ⇒ Q, we may instead
prove its contrapositive ¬Q ⇒ ¬P if this is easier. The proof of the contra-
positive itself will be a direct proof:
assume ¬Q and, from this, deduce ¬P .

The following examples illustrate the method. The first of these proves the
result that we discussed in the previous section.

Examples 6.1

1. In the introduction, we indicated that a direct proof of the following

theorem is problematic because it is not clear what sensible deductions
we can make from the statement ‘n2 is odd’. This makes it a prime
candidate for a proof using the contrapositive. We have also removed
the restriction that n should be positive.

Theorem 6.1
For all integers n, if n2 is odd, then n is odd.
 241

Proof. We will prove the contrapositive: if n is even, then n2 is even.

Let n be an even integer. Then n = 2m for some integer m.
Hence n2 = (2m)2 = 4m2 = 2(2m2 ) where 2m2 is an integer. Therefore
n2 is even.
We have proved that, for all integers n, if n is even, then n2 is even.
Hence the contrapositive follows: for all integers n, if n2 is odd then n
is odd.

Commentary
There are a few things worth noting about this proof. Firstly, it is good
practice to signal at the beginning that the proof is using the contrapos-
itive. Without this first sentence, the reader may not see why the main
body of the proof starts by assuming that n is even. Secondly, we have
not explicitly mentioned that the negation of ‘n is odd’ is ‘n is even’ as
this was felt to be sufficiently obvious. Sometimes, when the negation of
one or both of the component propositions of the conditional statements
is less obvious, then it would be sensible to explain the negation in the
proof itself.
Finally, note that the proposition being proved is a universally quantified
propositional function

∀n • O(n2 ) ⇒ O(n)

where O is the predicate ‘is odd’ and where the universe is Z+ . We have
used the Principle of Universal Generalisation given in section 4.3 to
reduce the proof requirement to O(n2 ) ⇒ O(n) for an arbitrary positive
integer and then applied the Contrapositive Law, to change the proof
requirement to ¬O(n) ⇒ ¬O(n2 ) for the arbitrary integer n. In other
words, we have used both the Principle of Universal Generalisation and
the Contrapositive Law but it is only the Contrapositive Law that is
explicitly referred to in the proof itself.

2. In this example, we prove a similar result to the previous one. Since

the negation of one of the statements in the conditional is a little less
obvious, the proof explicitly ‘unravels’ the negation.

Theorem 6.2
For all integers n and m, if mn is even, then m is even or n is even.

Proof. We will prove the contrapositive.

By De Morgan’s Law, the negation of ‘m is even or n is even’ is ‘m is
odd and n is odd’.
242

So suppose that m and n are both odd integers. Then m = 2k + 1 and

n = 2ℓ + 1 for some integers k and ℓ.
Hence mn = (2k + 1)(2ℓ + 1) = 4kℓ + 2k + 2ℓ + 1 = 2(2kℓ + k + ℓ) + 1
where 2kℓ + k + ℓ is an integer. Therefore mn is odd.
We have proved: if m and n are both odd, then mn is odd. Therefore
the contrapositive follows: if mn is even, then m is even or n is even.

3. The next result we consider is somewhat trickier to prove than those in

the two previous examples. It will also require a little more background
knowledge. In particular, we will assume the following formula for the
sum of a geometric progression. If a is a real number and a ̸= 1, then
for all positive integers n,
an − 1
1 + a + a2 + · · · + an−1 = .
a−1
We will give a proof of this result in chapter 8; see example 8.1.4.

Theorem 6.3
For all positive integers n, if 2n − 1 is prime, then n is prime.

Before embarking on the proof, consider, for a moment, how a direct

proof of this result might progress. We would firstly need to assume
that 2n − 1 is prime and, from this, deduce that n itself is prime. The
difficulty here is that it is not clear what deductions can usefully be made
from the assumption 2n − 1 is prime. Certainly 2n − 1 would satisfy any
known properties of prime numbers, but it is not clear how this might
provide useful information about n itself.
For example, assuming that 2n − 1 is prime we might deduce that 2n − 1
is either equal to 2 or is an odd integer (not necessarily prime). The first
case gives 2n = 3, which is not possible where n is an integer. Hence
2n − 1 is an odd integer, which implies that 2n is even — but we knew
this anyway! At this stage, it appears that a direct proof is not leading
anywhere.
We therefore turn our attention to the contrapositive: if n is not prime,
then 2n − 1 is not prime. The statement ‘n is not prime’, or is ‘n is
composite’, means that n = rs for some positive integers r and s such
that 1 < r < n and 1 < s < n. (Note that these two statements
are frequently combined into a single statement 1 < r, s < n which is
more compact but, strictly speaking, is an abuse of notation.) Thus a
proof using the contrapositive will assume n is composite and, from this,
deduce that 2n − 1 is composite.
The geometric progression above with a = 2 just gives
2n − 1 = 1 + 2 + 22 + · · · + 2n−1 ,
 243

which is not very helpful. But this is for an arbitrary integer n and we
are assuming that n is composite. In this case n = rs so we have

2n − 1 = 2rs − 1 = (2r )s − 1

and we can apply the geometric progression formula with a = 2r and

n = s. This gives
(2r )s − 1 2n − 1
1 + 2r + (2r )2 + · · · + (2r )s−1 = = .
2r − 1 2r − 1
Rearranging this equation, we have

2n − 1 = (2r − 1)(1 + 2r + (2r )2 + · · · + (2r )s−1 ),

which expresses 2n − 1 as a product and hence shows that 2n − 1 is

composite, which is what we were seeking to achieve. Having now found
a chain of reasoning from ‘n is composite’ to ‘2n − 1 is composite’, we
can now organise this into a coherent proof.

Proof. We will prove the contrapositive.

Suppose that n is not prime. Then n = rs where 1 < r < n and
1 < s < n. Then 2n = 2rs = (2r )s .
We will use the result
an − 1
1 + a + a2 + · · · + an−1 =
a−1
with a = 2r and n = s. This gives
(2r )s − 1 2n − 1
1 + (2r ) + (2r )2 + · · · + (2r )s−1 = = r
2 −1
r 2 −1
so
(2r − 1)(1 + (2r ) + (2r )2 + · · · + (2r )s−1 ) = 2n − 1.
This last equation expresses 2n −1 as the product of two integers, namely
2r − 1 and 1 + (2r ) + (2r )2 + · · · + (2r )s−1 . Hence 2n − 1 is not prime.
This completes the proof of the contrapositive: n is not prime ⇒ 2n − 1
is not prime. Hence the result follows.

As is always the case with mathematical proof, we are unable to give precise
and foolproof rules concerning when a particular method of proof will be ap-
plicable. However, considering the structure of the propositions P ⇒ Q that
we proved in the three previous examples, we might informally characterise
P as containing ‘compound information’ but Q as containing ‘simple infor-
mation’ of a similar type. In example 6.1.1, P contained information about
the parity (evenness/oddness) of a square of an integer, whereas Q contained
244

information about the parity of the integer itself. example 6.1.2 was similar:
P was a statement about the parity of a product whereas Q was a statement
about the parity of the component integers. Finally, in example 6.1.3, P was
a statement about the primality of 2n − 1 whereas Q was simply a statement
about the primality of n itself. In each case, the statement P has given ‘com-
pound’ information (about a square, a product of an expression involving a
power) whilst Q has provided the corresponding information about an integer
or integers.
It is often simpler to ‘build up’ to more complex, or ‘compound’, information
by combining simpler pieces of information than it is to go in the opposite
direction by ‘splitting down’ compound information into constituent parts. It
is in situations such as these that a proof using the contrapositive might be
considered because the ‘direction’ of the implication is reversed.
To illustrate this, consider the following theorem. Each part is a conditional
statement P ⇒ Q where P refers to a composite function, and so represents
‘compound’ information, and Q refers to similar information but about just
one of the functions. Given our previous discussion, this suggests that a proof
using the contrapositive may be fruitful. For example, in part (i), to prove the
contrapositive we assume that f is not injective and then deduce that the
composite g ◦ f is not injective.

Theorem 6.4
Let f : A → B and g : B → C be functions.
(i) If the composite g ◦ f is injective, then so, too, is the function f .
(ii) If the composite g ◦ f is surjective, then so, too, is the function g.

We will prove part (i) and leave part (ii) as an exercise (see exercise 6.1.5).
We first need to define what it means for a function to be injective. Recall from
section 3.6 that a function is injective if different elements (of the domain)
have different images. This was formalised in definition 3.3 which says that a
function f is injective if, for all elements of its domain,

f (a1 ) = f (a2 ) ⇒ a1 = a2 .

Since we are aiming to prove the contrapositive (f is not injective ⇒ g ◦ f is

not injective) we will need the negation of this statement. In section 2.4, when
considering the negation of quantified propositional functions, we established
the following general logical equivalence (see page 64)

¬∀x • P (x) ⇒ Q(x) ≡ ∃x • P (x) ∧ ¬Q(x).

Thus, saying that f is not injective means that there exist elements a1 and
a2 in its domain such that f (a1 ) = f (a2 ) but a1 ̸= a2 . Thus, if we assume
 245

that f is not injective, it follows that there are distinct (that is, not equal)
elements a1 , a2 ∈ A such that f (a1 ) = f (a2 ). It then follows immediately that
g(f (a1 )) = g(f (a2 )) for these distinct elements, so the composite function g ◦f
is also not injective. We can now organise these thoughts into a quite a short
and straightforward proof.
Proof of (i). We will prove the contrapositive: f is not injective ⇒ g ◦ f is not
injective. So suppose that f : A → B and g : B → C are functions and that f
is not injective.
Then there exist elements a1 , a2 ∈ A such that a1 ̸= a2 but f (a1 ) = f (a2 ).
It follows immediately that g(f (a1 )) = g(f (a2 )), which means (g ◦ f )(a1 ) =
(g ◦ f )(a2 ).
Hence g◦f is not injective, which completes the proof of the contrapositive.

Exercises 6.1
1. Prove each of the following using the contrapositive. Is it possible to
prove any of these results (easily) without using the contrapositive?

(i) For all integers n, if 5n + 6 is even, then n is even.

(ii) For all integers n and all odd integers k, if kn + (k + 1) is even,
then n is even.
(iii) For all integers n, if n2 is divisible by 3, then n is divisible by 3.
(iv) For all integers n, if n2 is not divisible by 7, then n is not divisible
by 7.
(v) For all integers m and n, if mn is divisible by 3, then m is divisible
by 3 or n is divisible by 3.

2. Prove each of the following using the contrapositive.

(i) If m and n are positive integers such that mn = 100, then m ≤ 10

or n ≤ 10.
(ii) If m, n and k are positive integers such that mn = k 2 , then m ≤ k
or n ≤ k.

3. Prove that, if a is an odd integer, then the quadratic equation x2 −x−a =

0 has no roots which are integers.
√
4. In section 6.5, we will prove that 2 is irrational.
Use the Prime Factorisation Theorem 4.3 to√ prove that, if n is a positive
integer that is not a perfect square, then n is irrational.
5. Prove part (ii) of theorem 6.4.
246

6. Prove that, if x is a real number such that 0 < x < 1, then x > x2 .
7. Prove each of the following using the contrapositive. Are any of the
following statements easily proved directly?

(i) For all sets A, B, X, and Y , if A × B ⊆ X × Y then A ⊆ X or

B ⊆Y.
(ii) For all non-empty sets A, B, X, and Y , if A × B ⊆ X × Y , then
A ⊆ X and B ⊆ Y .
(iii) For all sets A and B, if P(A) ⊆ P(B), then A ⊆ B.
 
a11 a12 a13
8. Let A = a21 a22 a23  be a 3 × 3 real matrix.
a31 a32 a33
Prove that, if A2 = 0, the 3 × 3 zero matrix, then at least one of the
‘upper diagonal’ elements, a11 , a12 , a13 , a22 , a23 , a33 is non-zero.

6.3 Proof of biconditional statements

Many theorems in mathematics are expressed in the form of biconditional

statements ‘P if and only if Q’. Here are a few examples.
1. For all integers n, n2 is odd if and only if n is odd.
2. The line y = mx − 2 intersects the parabola y = 3x2 + 1 if and only
|m| ≥ 6.
3. For all sets A and B, P(A) ⊆ P(B) if and only if A ⊆ B.
4. A group G is Abelian if and only if, for all g, h ∈ G, (gh)−1 = g −1 h−1 .

In section 2.3, we established the following logical equivalence, which we called

the Biconditional Law.

P ⇔ Q ≡ (P ⇒ Q) ∧ (Q ⇒ P ).

This provides a method for proving a biconditional statement of the form P ⇔

Q: prove each of the conditional statements P ⇒ Q and Q ⇒ P separately.
For each of the ‘sub-proofs’, P ⇒ Q and Q ⇒ P , we will typically use either
a direct proof or a proof using the contrapositive.
The examples above are clearly signalled as biconditional statements through
the phrase ‘if and only if’. Of course, the biconditional connective is ‘symmet-
ric’, in the sense that P ⇔ Q ≡ Q ⇔ P , so that it does not matter whether
 247

we write ‘P if and only if Q’ or ‘Q if and only if P ’. Sometimes ‘if and only

if’ is abbreviated as ‘iff’ so that the first example above may be expressed as
‘for all integers n, n2 is odd iff n is odd’. Another way of signalling P ⇔ Q is
to use the expression ‘P is a necessary and sufficient condition for Q’. Thus
the third example above could have been expressed as ‘for all sets A and B,
A ⊆ B is a necessary and sufficient condition for P(A) ⊆ P(B)’. Regardless
of how we express a biconditional P ⇔ Q, the standard proof technique is to
prove separately P ⇒ Q and its converse Q ⇒ P .

Occasionally it will be possible to prove P ⇔ Q by providing a sequence of

intermediate propositions which can be linked using the biconditional

P ⇔ P1 ⇔ P2 ⇔ · · · ⇔ Q.

However, this is unusual and separate proofs of P ⇒ Q and Q ⇒ P are far

more common.

Examples 6.2

1. For our first proof, we will combine two proofs that we have given pre-
viously. In example 1.3, we gave what we can now recognise as a direct
proof of the proposition ‘for all integers n, if n is odd, then n2 is odd’.
In example 6.1.1 we used the contrapositive to prove ‘for all integers n,
if n2 is odd, then n is odd’. We now combine those proofs to give a proof
of the following theorem.

Theorem 6.5
For all integers n, n is odd if and only if n2 is odd.

Proof.
(⇒) Let n be an odd integer. Then n = 2m + 1 for some integer m. Now

n2 = (2m + 1)2
= 4m2 + 4m + 1
= 2(2m2 + 2m) + 1
= 2M + 1 where M = 2m2 + 2m is an integer.

Therefore n2 is odd.
(⇐) We will prove the contrapositive of the converse. Since the converse
is ‘if n2 is odd, then n is odd’, its contrapositive is ‘if n is even,
then n2 is even’.
Let n be an even integer. Then n = 2m for some integer m. Now
248

n2 = (2m)2
= 4m2
= 2(2m2 )
= 2M where M = 2m2 is an integer.

Therefore n2 is even.
This completes the proof of the contrapositive. Hence, for all inte-
gers n, if n2 is odd, then n is odd.

Commentary
Compared with the previous proofs, we have modified slightly how we
have written the two parts of the proof to be consistent with one another.
However, we have not changed the structure of the previous proofs. Note
that the proof is laid out clearly in two parts with each part signalled
either by (⇒) or by (⇐). Although it is not necessary to lay out the
proof in this way, it is good policy to identify clearly each of the two
halves of the proof.

2. Our second example comes from set theory. For the proof we will need
to recall from section 3.2 the meaning of the subset relation: X ⊆ Y if
and only if, for all x, x ∈ X ⇒ x ∈ Y . The following theorem is easily
visualised by drawing a Venn-Euler diagram.

Theorem 6.6
For all sets A, B, and C, A ⊆ (B − C) if and only if both A ⊆ B and
A ∩ C = ∅.

Proof. Firstly, suppose that A, B, and C are sets such that A ⊆ (B −C).
We need to establish both A ⊆ B and A ∩ C = ∅.
Let x ∈ A. Then x ∈ B − C since A ⊆ (B − C). This implies both
that x ∈ B and that x ̸∈ C. We have shown x ∈ A ⇒ x ∈ B; therefore
A ⊆ B. But we have also shown x ∈ A ⇒ x ̸∈ C so that A and C have
no elements in common, A ∩ C = ∅.
Conversely, suppose that A, B, and C are sets such that A ⊆ B and
A ∩ C = ∅.
We need to show that A ⊆ (B − C), so let x ∈ A. Since A ⊆ B it
follows that x ∈ B. We also know that A and C have no elements in
common, so x ̸∈ C. Therefore x ∈ B − C (since we have shown x ∈ B
and x ̸∈ C). We have proved that x ∈ A ⇒ x ∈ B − C; it follows that
A ⊆ (B − C).
 249

Commentary
In this proof, we have not used the feature in the previous example
of labelling each half of the proof with (⇒) or (⇐). Instead we have
signalled the two halves of the proof using ‘Firstly, suppose . . . ’ and
‘Conversely, suppose . . . ’.
As with many proofs about sets, we need to ‘drill down’ to considering
elements. The overall structure of the statement to be proved is P ⇔
(Q∧R). However the propositions P and Q identify subset relations and
hence each is equivalent to a conditional statement involving elements.
For example, the statement P is A ⊆ (B − C), which is equivalent to
x ∈ A ⇒ x ∈ B − C (for an arbitrary element x).
3. In this example, we illustrate that it is sometimes more natural (or more
efficient) not to separate the proof of a biconditional into two parts.
Consider the second of our initial examples.

Theorem 6.7
The line y = mx − 2 intersects the parabola y = 3x2 + 1 if and only
|m| ≥ 6.

Before we embark on the proof, there are two pieces of background

knowledge that we will need. Although we anticipate that each of these
will be familiar to our readers, it is probably helpful to identify them
explicitly. Firstly, to say that two curves (and a line is an example of a
curve) intersect means that there is at least one point that lies on each.
This is true if and only if the pair of simultaneous equations defining
the curves has a solution. The second piece of background knowledge
concerns quadratic equations: the quadratic equation ax2 + bx + c = 0,
where a ̸= 0, has a (real) solution if and only if b2 − 4ac ≥ 0. This of
course follows from the quadratic formula
√
−b ± b2 − 4ac
x= .
2a
Proof. The line intersects the parabola if and only the pair of simulta-
neous equations y = mx − 2, y = 3x2 + 1 has a solution. Now
}
y = mx − 2
has a solution ⇔ 3x2 + 1 = mx − 2
y = 3x2 + 1
⇔ 3x2 − mx + 3 = 0.
The quadratic equation 3x2 − mx + 3 = 0 has a solution if and only if
its discriminant ‘b2 − 4ac’ is non-negative. Hence
3x2 − mx + 3 = 0 has a solution ⇔ m2 − 4 × 3 × 3 ≥ 0
⇔ m2 ≥ 36
⇔ |m| ≥ 6.
250

Hence the line y = mx − 2 intersects the parabola y = 3x2 + 1 if and

only |m| ≥ 6.

Commentary
Rather than proving two separate conditional propositions, in this case
we have linked the two statements ‘the line y = mx − 2 intersects the
parabola y = 3x2 + 1’ and ‘|m| ≥ 6’ directly via a sequence of interme-
diate statements. We have been careful to ensure that each statement in
the chain is connected to those before and after it by the biconditional ‘if
and only if’ connective. Thus the structure of the proof can be described
as follows.
The line y = mx − 2 intersects the parabola y = 3x2 + 1
⇔ ...
⇔ . . . [a sequence of intermediate statements]
⇔ ...
⇔ |m| ≥ 6.

It is possible in this example to write the proof as two separate sub-

proofs:

y = mx − 2 intersects y = 3x2 + 1 ⇒ |m| ≥ 6

and
|m| ≥ 6 ⇒ y = mx − 2 intersects y = 3x2 + 1.

However, in this case, the two sub-proofs would essentially be the same
argument in reverse. It is therefore both more efficient to write and easier
to understand if the two sub-proofs are combined as we have done.

Exercises 6.2

1. Let x, y ∈ R. Prove that |x + y| = |x| + |y| if and only if xy ≥ 0.

2. Prove each of the following properties of the divisibility relation on Z+ .
Recall from exercise 5.1.4 that m|n denotes ‘m divides n.

(i) For all positive integers m and n, m|n and n|m if and only if m = n.
(ii) For all positive integers a, m, and n, m|n if and only if am|an.

3. Prove that, for all integers m and n and all primes p, p is a factor of the
product mn if and only if p is a factor of m or p is a factor of n.
 251

4. Prove each of the following about a function f : A → B.

(i) f is injective if and only if, for all subsets C1 and C2 of A, f (C1 ∩
C2 ) = f (C1 ) ∩ f (C2 ).
(ii) f is injective if and only if, for all subsets C of A, C = f −1 (f (C)).
(iii) f is surjective if and only if, for all subsets D of B, f (f −1 (D)) = D.

5. Let A and B be subsets of R. A function f : A → B is said to be

increasing if, for all x, y ∈ A, x < y ⇒ f (x) ≤ f (y). The function f is
strictly increasing if, for all x, y ∈ A, x < y ⇒ f (x) < f (y). Similarly,
we say that f is decreasing if, for all x, y ∈ A, x < y ⇒ f (x) ≥ f (y)
and f is strictly increasing if, for all x, y ∈ A, x < y ⇒ f (x) > f (y).
Prove each of the following for functions A → B where A and B are sub-
sets of R. For which of the statements is it possible to replace ‘increasing’
with ‘strictly increasing’ and ‘decreasing’ with ‘strictly decreasing’ (as
appropriate)?

(i) The function f is increasing if and only if the function −f is de-

creasing. The function −f : A → B is defined by (−f )(x) = −f (x).
(ii) For all α ∈ R+ , f is increasing if and only if αf is increasing. The
function αf : A → B is defined by (αf )(x) = αf (x).
(iii) Suppose that f (x) > 0 for all x ∈ A. Then f is increasing if and
only if the function 1/f is decreasing. The function 1/f : A → B
is defined by (1/f )(x) = 1/f (x).

6. Prove that, for all positive integers n, n has exactly 5 factors (including
1 and n itself) if and only if n = p4 for some prime p.

7. Let S and T be subspaces of a vector space U . Recall from exercise 5.3.7

that S + T = {x + y : x ∈ S and y ∈ T } is also a subspace of U .
Suppose that V is a subspace of U with the property that every v ∈ V
can be written uniquely as v = x + y where x ∈ S and y ∈ T . Then we
write V = S ⊕ T and say that V is the direct sum of S and T .
Suppose V = S + T . Prove that V = S ⊕ T if and only if S ∩ T = {0}.

6.4 Proof of conjunctions

We begin this section with the simple observation that, to prove a conjunction
P ∧ Q, it is sufficient to prove P and Q separately. This is intuitively obvious
but may be formalised using the conjunction deduction rule from propositional
252

logic given in section 2.5. This states that if we know P and we know Q, then
we may deduce P ∧ Q.
Suppose we now combine this observation with the logical equivalence

P ⇒ (Q ∧ R) ≡ (P ⇒ Q) ∧ (P ⇒ R)

that we established in example 2.4.2 (page 36). We may then observe that, to
prove a proposition of the form P ⇒ (Q∧R), it is sufficient to prove separately
the two conditional propositions P ⇒ Q and P ⇒ R.

Examples 6.3

1. Consider the following theorem and its (simple) proof.

Theorem 6.8
For all real numbers x, |x| ≥ x and x + |x| ≥ 0.

Proof. Let x be a real number. We will consider two cases: x ≥ 0 and

x < 0.
Suppose that x ≥ 0. Then |x| = x so, the inequalities |x| ≥ x and
x + |x| ≥ 0 both follow.
Now suppose that x < 0. Then |x| = −x > 0, so again |x| > 0 > x and
x + |x| = x + (−x) = 0.
In each case we have established |x| ≥ x and x + |x| ≥ 0, completing the
proof.

Commentary
We could regard this theorem as two separate theorems: ‘for all real
numbers x, |x| ≥ x’ and ‘for all real numbers x, x + |x| ≥ 0’. In the
interests of efficiency, it is often sensible to combine simple related the-
orems (and their proofs). In this example, each of the separate proofs
would still require two cases, x ≥ 0 and x < 0, so it is more efficient to
combine the two proofs.

2. In this example, the result to be proved is of the form P ⇒ (Q ∧ R), so

the two separate subproofs will be P ⇒ Q and P ⇒ R.

Theorem 6.9
For all sets A and B, if A and B are disjoint, then A ⊆ B̄ and B ⊆ Ā.

Proof. Suppose that A and B are disjoint sets. By definition, this means
their intersection is empty, A ∩ B = ∅.
 253

To prove that A ⊆ B̄, we need to show that, for all x, x ∈ A ⇒ x ∈ B̄.

Let x ∈ A. Since A ∩ B = ∅, this implies that x ̸∈ B. Hence x ∈ B̄. We
have shown that, for all x, if x ∈ A, then x ∈ B̄. Therefore A ⊆ B̄.
The proof of B ⊆ Ā follows from the previous argument by interchanging
A and B.

Commentary
The two sub-proofs here are:

A ∩ B = ∅ ⇒ A ⊆ B̄
and
A ∩ B = ∅ ⇒ B ⊆ Ā.

Since A∩B = B ∩A, the second statement can be obtained from the first
by interchanging A and B. Hence the proof of the second statement can
be obtained by repeating the proof of the first and also interchanging A
and B. Of course, there is little to be gained by actually carrying out
this re-writing, which is why we have simply indicated how the proof
may be constructed.

Set equality

There is a particular situation where the proof of a conjunction P ∧ Q is

commonly achieved by proving each of P and Q separately, and that is in
establishing the equality of two sets. Recall from theorem 3.1 that two sets A
and B are equal if and only if A ⊆ B and B ⊆ A. Thus, to prove A = B, it
is common to prove separately A ⊆ B and B ⊆ A. For sets X and Y , X ⊆ Y
means that, for all x, x ∈ X ⇒ x ∈ Y . Piecing these considerations together
gives the following template for a proof that A = B for two sets A and B.

To prove A = B:

1. A ⊆ B: to do this, assume x ∈ A and deduce x ∈ B

2. B ⊆ A: to do this, assume x ∈ B and deduce x ∈ A.

Examples 6.4
1. For our first example, we prove one of De Morgan’s Laws for sets. This
was one of the Set Theory Laws introduced in section 3.3; see page 99.

Theorem 6.10
For all sets A and B, A ∪ B = Ā ∩ B̄.
254

We will give three proofs of this theorem to illustrate the different pos-
sible styles of proof. As always, anyone writing a proof has to decide the
style they wish to adopt, how much detail to give, how much background
knowledge to assume, and so forth.
Our first proof is quite descriptive and ‘wordy’. The second proof is very
symbolic but follows the same overall structure as the first. We expect
that some readers will prefer the style of the first proof and some the
second. Our third proof combines the two parts of the symbolic second
proof into a single short proof.

Proof 1. First, let x ∈ A ∪ B. This means that x ̸∈ A ∪ B. Now A ∪ B

contains all elements in A or in B or in both, so x ̸∈ A ∪ B means that
x ̸∈ A and x ̸∈ B. Hence x ∈ Ā ∩ B̄. We have shown that x ∈ A ∪ B ⇒
x ∈ Ā ∩ B̄. Therefore A ∪ B ⊆ Ā ∩ B̄.

Next we need to establish the subset relation the other way around, so
let x ∈ Ā ∩ B̄. Then x ∈ Ā and x ∈ B̄, so x ̸∈ A and x ̸∈ B. Since x
belongs neither to A nor to B, we have x ̸∈ A ∪ B, so x ∈ A ∪ B. We
have shown that x ∈ Ā ∩ B̄ ⇒ x ∈ A ∪ B. Therefore Ā ∩ B̄ ⊆ A ∪ B.

Since each set is a subset of the other, it follows that Ā ∩ B̄ = A ∪ B.

Our second proof follows the same structure as the first proof, but the
reasoning is expressed symbolically. Perhaps the drawback of this proof
is that it does not explain in detail why each of the statements ‘x ̸∈ A∪B’
and ‘x ̸∈ A and x ̸∈ B’ imply each other.

Proof 2. Firstly, x ∈ A ∪ B ⇒ x ̸∈ A ∪ B
⇒ x ̸∈ A and x ̸∈ B
⇒ x ∈ Ā and x ∈ B̄
⇒ x ∈ Ā ∩ B̄.

Therefore A ∪ B ⊆ Ā ∩ B̄.
Also, x ∈ Ā ∩ B̄ ⇒ x ∈ Ā and x ∈ B̄
⇒ x ̸∈ A and x ̸∈ B
⇒ x ̸∈ A ∪ B
⇒ x ∈ A ∪ B.

Therefore Ā ∩ B̄ ⊆ A ∪ B.

Since each set is a subset of the other, it follows that Ā ∩ B̄ = A ∪ B.

 255

Having written the proof symbolically, it is clear that the steps in the
reasoning in the second half of the proof are just the reverse of the steps
in the first half. In other words, each of the implications in the first half
of the proof reverses. We may therefore combine the two halves of the
proof using the ‘if and only if’ connective. This gives our third proof.

Proof 3. Note that, x ∈ A ∪ B ⇔ x ̸∈ A ∪ B

⇔ x ̸∈ A and x ̸∈ B
⇔ x ∈ Ā and x ∈ B̄
⇔ x ∈ Ā ∩ B̄.

Therefore A ∪ B = Ā ∩ B̄.

2. In exercise 3.2.5 we defined the symmetric difference A ∗ B of two sets

A and B to be A ∗ B = (A − B) ∪ (B − A). In this example we show
that we could equally well have defined the symmetric difference to be
(A ∪ B) − (A ∩ B) and, indeed, some authors take this as the definition.

Theorem 6.11
For all sets A and B, (A − B) ∪ (B − A) = (A ∪ B) − (A ∩ B).

Proof. Firstly, let x ∈ (A − B) ∪ (B − A). Then x ∈ A − B or x ∈ B − A.

If x ∈ A − B, then x ∈ A and x ̸∈ B. Hence x ∈ A ∪ B (since x ∈ A)
and x ∉ A ∩ B (since x ̸∈ B). Therefore x ∈ (A ∪ B) − (A ∩ B).
If x ∈ B − A, then x ∈ B and x ̸∈ A. Hence x ∈ A ∪ B (since x ∈ B)
and x ∉ A ∩ B (since x ̸∈ A). Therefore x ∈ (A ∪ B) − (A ∩ B).
In either case, x ∈ (A ∪ B) − (A ∩ B). We have proved that x ∈ (A − B) ∪
(B−A) ⇒ x ∈ (A∪B)−(A∩B), so (A−B)∪(B−A) ⊆ (A∪B)−(A∩B).

Next we need to establish the subset relation the other way around, so
let x ∈ (A∪B)−(A∩B). Then x ∈ A∪B and x ̸∈ A∩B. Since x ∈ A∪B
we know that x ∈ A or x ∈ B. If x ∈ A then, since x ̸∈ A ∩ B, we know
that x ̸∈ B. Hence x ∈ A − B. If x ∈ B then, again since x ̸∈ A ∩ B,
we know that x ̸∈ A. Hence x ∈ B − A. We have shown that either
x ∈ A − B or x ∈ B − A. Therefore x ∈ (A − B) ∪ (B − A).
We have proved that x ∈ (A ∪ B) − (A ∩ B) ⇒ x ∈ (A − B) ∪ (B − A),
so (A ∪ B) − (A ∩ B) ⊆ (A − B) ∪ (B − A).

Finally, since each set is a subset of the other, (A − B) ∪ (B − A) =

(A ∪ B) − (A ∩ B).
256

Exercises 6.3
1. Prove each of the following.
(i) If x and y are positive real numbers such that x + y = 1, then
( )2 ( )2
1 1
xy ≤ 14 and x + + y+ ≥ 25
2 .
x y
(ii) If the roots of the quadratic equation x2 + ax + b = 0 are odd
integers, then a is an even integer and b is an odd integer.
(iii) If x ∈ R is such that −1 ≤ x ≤ 4, then −9 ≤ x2 − 6x ≤ 7 and
−5 ≤ x2 + 6x ≤ 40.
2. Prove each of the following set identities. In each case, the results are
true for all sets.
(i) A ∪ (B ∩ C) = (A ∪ B) ∩ (A ∪ C)
(ii) Ā ∪ B̄ = A ∩ B
(iii) (A ∪ B) − C = (A − C) ∪ (B − C)
(iv) (A ∩ B) × (X ∩ Y ) = (A × X) ∩ (B × Y )
(v) (A ∪ B) × C = (A × C) ∪ (B ∪ C)
3. Prove each of the following identities involving the symmetric difference
of two sets. From theorem 6.11, the symmetric difference A ∗ B may be
taken to be either (A − B) ∪ (B − A) or (A ∪ B) − (A ∩ B), whichever
is most convenient. In each case the result is true for all sets.
(i) (A ∗ B) ∩ C = (A ∩ C) ∗ (B ∩ C)
(ii) (A ∪ C) ∗ (B ∪ C) = (A ∗ B) − C
(iii) (A ∗ B) × C = (A × C) ∗ (A × B)

4. In each of the following cases, prove that the sets X and Y are equal.
(i) X = {× ∈ R : 2x2 + 3x − 1 = x2 + 8x − 5}, Y = {1, 4}
{ }
x
(ii) X = : x ∈ R and x ̸= −1 , Y = {x ∈ R : x > −1}
x+1
{ }
x { }
(iii) X = 2
: x ∈ R and x ≥ 1 , Y = x ∈ R : 0 < x ≤ 12
x +1
5. Let n be a positive integer. The sets A1 , A2 , . . . , An are closed intervals
defined by Ar = [r, r + n] for r = 1, 2, . . . , n.
Prove that A1 ∩ A2 ∩ . . . ∩ An = [n, n + 1].
6. Two elements x and y of a group G are said to commute if xy = yx.
Prove that if x and y commute, then so do both x−1 and y −1 and g −1 xg
and g −1 yg for all elements g ∈ G.
 257

6.5 Proof by contradiction

Recall the Contradiction Law from section 2.3:

(¬P ⇒ false) ≡ P

where ‘false’ denotes any contradiction. This means that to prove a statement
P , it is sufficient to prove the conditional ¬P ⇒ false instead. The condi-
tional is proved using direct proof: assume ¬P and, from this, deduce false.
In other words, to prove P we may start by assuming that P is false (that is,
assume ¬P ) and then deduce a contradiction.1 This is the method of proof
by contradiction, which is also known by the Latin reductio ad absurdum
(‘reduction to absurdity’). Initially it appears a slightly odd method of proof:
to prove P we start by assuming that P is false! However, we only assume the
falsity of P in order to obtain a contradiction. The contradiction obtained is
usually of the form Q ∧ ¬Q for some proposition Q; in other words, we deduce
both Q and its negation ¬Q.
Before discussing
√ the method further, we first give one of the classic proofs by
contradiction: 2 is irrational. Recall from section 3.2 that a rational number
is one that can be expressed as a fraction; that is, a rational number is a ratio
of integers m/n, where n ̸= 0. Therefore, to say that a number is irrational
means that it cannot be expressed as the ratio of two integers.

Theorem
√ 6.12
2 is irrational.

Proof. The proof is by contradiction.

√ √
Suppose that 2 is rational. Then 2 can be expressed as a fraction in its
lowest terms; that
√ is, where numerator and denominator have no common
factors. Hence 2 may be expressed as
√ m
2= (*)
n

where m and n have no common factors. Squaring both sides of the equation
gives 2 = m2 /n2 so m2 = 2n2 which means that m2 is even. It follows from

1 In [9], the great mathematician G.H. Hardy described proof by contradiction as ‘one

of a mathematician’s greatest weapons’. He went on to say: ‘It is a far finer gambit than
any chess gambit: a chess player may offer the sacrifice of a pawn or even a piece, but the
mathematician offers the game.’
258

theorem 6.5 that m itself is even.2 Thus m = 2r for some integer r. Substi-
tuting m = 2r into m2 = 2n2 gives 4r2 = 2n2 so n2 = 2r2 . This shows that
r2 is even, so again, by theorem 6.5, it follows that n itself is even.
We have established that both m and n are even, so they have 2 as a common
factor. This contradicts the equation√(*) in which m and n had no common
factors. Hence, the assumption that 2 is rational is false.
√
Therefore 2 is irrational.

Commentary
As always, there are different styles of expressing this proof. For example,
we
√ could have expressed some of the reasoning more symbolically such as
‘ 2 = m/n ⇒ 2 = m2 /n2 ⇒ m2 = 2n2 ⇒ m2 is even’. One aspect of
this proof that is important, however, is to signal at the beginning that the
proof uses the √
method of contradiction. Without this, the initial assumption
‘suppose that 2 is rational’ would look odd indeed.
We may think of theorem 6.12 as being a ‘non-existence
√ theorem’: it asserts
that there do not exist integers m and n such that 2 = m/n. We will consider
‘existence theorems’ in the next chapter; here we reflect on the difficulty of
proving a ‘non-existence theorem’, at least in the case where the universe is
infinite. There √are, of course, infinitely many rational numbers. To prove the
irrationality of 2, we need to rule out the√possibility that any of the infinitely
many rational numbers could be equal to 2 and this seems a daunting task.
Indeed, it is hard to imagine how we might go about constructing a direct
proof of theorem 6.12 (although such direct proofs do exist).
The method of proof by contradiction will be a useful tool to have when
proving non-existence where the universe is infinite. We illustrate this with
our next theorem which asserts that there exist infinitely many prime numbers.
This is not obviously a ‘non-existence theorem’ until we reflect that it is
equivalent to asserting that there does not exist a largest prime number. The
proof of theorem 6.13 is also regarded as a classic proof by contradiction. The
proof is commonly attributed to Euclid, over 2000 years ago.

Theorem 6.13
There exist infinitely many prime numbers.

Before embarking on the proof itself, we begin by exploring the key idea
using some small examples. Consider the list of the first few prime numbers:
2, 3, 5, 7, 11, 13, 17, 19, . . .. If we take the product of all of the primes up to
some point and add 1, what happens?
2 Theorem 6.5 says that ‘n2 is odd, if and only if n is odd’ but this is clearly equivalent

to ‘n2 is even if and only if n is even’.

 259

2+1=3 prime
2×3+1=7 prime
2 × 3 × 5 + 1 = 31 prime
2 × 3 × 5 × 7 + 1 = 211 prime
2 × 3 × 5 × 7 × 11 + 1 = 2311 prime

In each case, the resulting number is a prime which is obviously greater than
each of the primes used in its construction. This simple idea lies behind the
proof which we can now give.

Proof. The proof is by contradiction.

Suppose that there are only finitely many primes. Then we may list all of
them: p1 , p2 , p3 , . . . , pn . Now consider the integer N that is one larger than
the product of all of the primes:

N = p1 p2 p3 . . . pn + 1.

Since N is not equal to any of the primes in the list p1 , p2 , . . . , pn , and since
we have assumed the list contains all primes, it follows that N is not prime.
Therefore, by the Prime Factorisation Theorem 4.3, N has a prime factor. In
other words, for at least one of the primes pk in the list, pk is a factor of N .
However, p1 p2 p3 . . . pn = pk (p1 . . . pk−1 pk+1 . . . pn )
= pk S where S = p1 . . . pk−1 pk+1 . . . pn
⇒ N = pk S + 1 where S ∈ Z.

Thus N gives a remainder of 1 when divided by pk , so that pk is not a factor

of N .
This is a contradiction and completes the proof.

Exercises 6.4
1. Let x and y be real numbers such that x is rational and y is irrational,
x ∈ Q and y ̸∈ Q. Prove that:
(i) x + y is irrational;
(ii) xy is irrational.
2. Prove by contradiction that the set of positive integers Z+ is infinite.
3. Let ax2 + bx + c = 0 be a quadratic equation where a, b, and c are
odd integers. Prove by contradiction that the equation has no rational
solutions.
260
√
4. Using the proof of theorem 6.12 as a guide, prove that 3 is irrational.
√
5. ̸ 0, m + 2n is
(i) Prove that, for all integers m and n where n =
irrational.
(ii) More generally, prove that if α is an irrational number then, for all
integers m and n where n ̸= 0, m + αn is irrational.
6. Prove that there is no rational number r such that 2r = 3.
7. Prove that there is no smallest real number that is strictly greater than
1
2.

8. Prove that if the mean of four distinct integers is n ∈ Z, then at least

one of the integers is greater than n + 1.
9. Prove the following result, which can be used for ‘primality testing’.
If n is an√integer greater than 1 which has no prime factor p satisfying
2 ≤ p ≤ n, then n is prime.
Hint: use a proof by contradiction to prove the contrapositive.
10. Suppose there are n ≥ 3 people at a party. Prove that at least two people
know the same number of people at the party.
Assume that if A knows B, then B knows A.
{ }
n−1
11. Prove that the set A = :n∈Z +
does not have a largest ele-
n
ment.

6.6 Further examples

In this section, we apply the methods introduced in this chapter to the contexts
introduced in chapter 5, sections 5.3 and 5.4. In other words, we consider
proofs in algebra and analysis that are variations on the method of direct
proof.

6.6.1 Examples from algebra

Theorem 6.14 (Alternative Subspace Test)

Let S be a non-empty subset of a vector space V . Then S is a subspace of V
if and only if it satisfies the following condition:

for all v, w ∈ S and λ, µ ∈ R, λv + µw ∈ S. (*)

 261

We use theorem 5.12, the Subspace Test proved in section 5.3.

Proof. Let S be a non-empty subset of a vector space V .

(⇒) Suppose that S is a subspace of V . By definition 5.6, this means that

S is itself a vector space with respect to the operations of addition of
vectors and multiplication by scalars that are defined in V .
Let v, w ∈ S and λ, µ ∈ R. Then, since scalar multiplication is a well-
defined operation in S, we have λv ∈ S and µw ∈ S. Then, since
addition of vectors is a well-defined operation in S, we have λv+µw ∈ S.
Therefore S satisfies condition (*).

(⇐) Suppose that S satisfies condition (*). We need to show that S is a

subspace of V and to do this we use our previous Subspace Test, theo-
rem 5.12.
First note that we are given that S ̸= ∅, so S satisfies condition (i) of
theorem 5.12.
Let v, w ∈ S. Then, using condition (*) with λ = µ = 1 gives

1v + 1w = v + w ∈ S

since 1v = v and 1w = w by axiom (M4) of a vector space (see defini-

tion 5.5). Hence S satisfies condition (ii) of theorem 5.12.
Now let v ∈ S and λ ∈ R. Using condition (*) with µ = 1 and w = 0
gives
λv + 10 = λv + 0 = λv ∈ S
using axioms (M4) and (A3) of a vector space. Hence S satisfies condition
(iii) of theorem 5.12.
Since S satisfies all three conditions of theorem 5.12, it follows that S
is a subspace of V , completing the proof.

In exercise 5.3.7, we introduced the idea of the sum of two subspaces of a

vector space. If S and T are subspaces of V , then their sum S + T is the
subspace of V defined by S + T = {x + y : x ∈ S and y ∈ T }.

Theorem 6.15
Let S and T be subspaces of a vector space V . Then S + T = T if and only
if S is a subspace of T .

Proof. Let S and T be subspaces of a vector space V .

262

(⇒) Suppose that S + T = T .

We need to show that S is a subspace of T . Since S and T are both
vector spaces with the same operations — they are subspaces of V —
to say that S is a subspace of T just means that S is a subset of T .
Let v ∈ S. Then we have v = v + 0 ∈ S + T since v ∈ S and 0 ∈ T .
This shows that v ∈ S ⇒ v ∈ T . Therefore S is a subset, and hence a
subspace, of T .
(⇐) Now suppose that S is a subspace of T . We need to show that S +T = T .
As usual, to prove that these two sets are equal, we show that each is a
subset of the other.
Let v + w ∈ S + T where v ∈ S and w ∈ T . Since S is a subspace of
T , it follows that v ∈ T . Hence we have v, w ∈ T so v + w ∈ T . This
shows that v + w ∈ S + T ⇒ v + w ∈ T , so S + T ⊆ T .
Now let v ∈ T . Therefore v = 0 + v ∈ S + T since 0 ∈ S and v ∈ T .
This shows that v ∈ T ⇒ v ∈ S + T , so T ⊆ S + T .
Therefore S + T = T , as required.

Whenever we have a particular kind of algebraic structure, we may consider

the mappings between them that ‘preserve’ or ‘respect’ the algebraic struc-
ture. For example, part of the structure of a vector space involves addition of
vectors. To say that a mapping f between vector spaces ‘preserves’ addition
of vectors means that f (x + y) = f (x) + f (y). The left-hand side of this equa-
tion adds vectors x and y and then applies the mapping f ; the right-hand
side first applies f to each vector and then adds the result. Thus, to say that
f ‘preserves addition’ simply means that it does not matter whether we apply
f before or after carrying out an addition of vectors. Of course, vector spaces
have two operations — addition of vectors and multiplication by scalars. A
mapping that preserves the algebraic structure of a vector space will need to
preserve both operations. These are called linear transformations and their
definition is given in exercise 5.3.11.
Our last example from the realm of algebra involves the corresponding idea
from group theory. For groups, there is a single operation so the appropriate
mapping just needs to preserve this. Such a mapping is called a ‘morphism’,
or ‘homomorphism’, between groups. The formal definition is the following.
Since there are two groups involved, we use different symbols for the binary
operation in each group.

Definition 6.1
Let (G, ∗) and (H, ◦) be two groups. A mapping θ : G → H is a morphism,
or homomorphism, if, for all x, y ∈ G,
θ(x ∗ y) = θ(x) ◦ θ((y).
 263

The kernel of θ, ker θ, is the set of all elements of G that map to the identity
element of H,
ker θ = {g ∈ F G : θ(g) = e}.

For any morphism θ : G → H, its kernel is a subgroup of G. We leave it

as an exercise to show this — see exercise 6.5.1. Our theorem gives a simple
condition for a morphism to be injective.

Theorem 6.16
Let (G, ∗) and (H, ◦) be groups and let θ : G → H be a morphism.
Then θ is injective if and only if ker θ = {e}.

Proof. Suppose that θ is injective.

Let x ∈ ker θ. Then θ(x) = e. But θ(e) = e (exercise 6.5.1 (i)), so θ(x) = θ(e).
Since θ is injective, we have x = e. In other words, e is the only element of
ker θ, so ker θ = {e}.

Conversely, suppose that ker θ = {e}.

Let x, y ∈ G. Then
−1
θ(x) = θ(y) ⇒ θ(x) (θ(y)) =e
−1
⇒ θ(x)θ(y )=e (exercise 6.5.1 (ii))
−1
⇒ θ(xy )=e (θ is a morphism)
−1
⇒ xy ∈ ker θ
−1
⇒ xy =e (since ker θ = {e})
⇒ x = y.

We have shown that θ(x) = θ(y) ⇒ x = y, which is precisely the condition

that θ is injective.

6.6.2 Examples from analysis

We will apply some of the techniques from this chapter to some proofs in
analysis. First, however, we introduce two further concepts: the ‘infimum’ and
‘supremum’ of a non-empty set of real numbers. Let X be a non-empty subset
of R. An upper bound for X is a real number M such that x ≤ M for all
x ∈ X. If a set has an upper bound it is said to be bounded above.
For example, let X = {x ∈ R : 0 < x < 1} be the interval comprising all real
numbers strictly between 0 and 1; this set is frequently denoted X = (0, 1).
An upper bound for X is 2 since x ≤ 2 for all x ∈ X. There are clearly many
264
√
other upper bounds, including 2, π, 27.25, and so on. Note that saying that
2 is an upper bound for (any set) X tells us more about the set than saying
that 27.25 is an upper bound. (If you had to guess a number to win a prize,
say, would you prefer to be told that the number was less than 2 or less than
27.25? Clearly the former is preferable as it would save you ‘wasting’ any
guesses between 2 and 27.25.) For our set X = {x ∈ R : 0 < x < 1}, the upper
bound that gives most information about X is 1. This is its smallest upper
bound or ‘least upper bound’; it is also called the ‘supremum’ of X, which we
now define formally.

Definition 6.2
Let X be a non-empty subset of R. A real number M satisfying:

(i) M is an upper bound for X: x ≤ M for all x ∈ X and

(ii) if a < M , then a is not an upper bound for X:3 if a < M , then there
exists x ∈ X such that x > a.

is called a supremum or least upper bound for X; it is denoted M = sup X.

Suppose we reverse the inequalities in the previous discussion. This gives,

first, the definition of a lower bound for a set: a lower bound for X is a
real number m such that m ≤ x for all x ∈ X. If a set has a lower bound
it is said to be bounded below; a set that is bounded above and below is
said to be bounded. We can then model definition 6.2 to define an infimum
or greatest lower bound for a set X as a lower bound with the additional
property that any larger real number is not a lower bound. We leave it as an
exercise to complete the definition.

Examples 6.5

1. For our first example, we will prove that 1 is, as expected, the supremum
of the interval X = (0, 1) = {x ∈ R : 0 < x < 1},

sup{x ∈ R : 0 < x < 1} = 1.

We need to show that M = 1 satisfies both the properties that define

the supremum given in definition 6.2. The first property is clear: 1 is an
upper bound for X since x ≤ 1 for all x ∈ X. For the second property,
let a < 1; we need to show that a is not an upper bound for X, which
means finding an element of X that is larger than a. This is illustrated
in figure 6.1. Probably the most obvious choice of x ∈ X is the average
3 Note that this is another instance where there is an implicit universal quantification.

A more precise expression of this condition would be: for all a ∈ R, if a < M , then a is not
an upper bound for X.
 265

of a and 1, (a+1)/2, which will lie strictly between a and 1 and therefore
appears to belong to the set X. We need to be a little careful, however.
The only restriction on a is that is must be a real number less that
1. Suppose a = −3, say, then the average of a and 1 is (a + 1)/2 =
(−3 + 1)/2 = −1 does not belong to X. (Recall that we are seeking
x ∈ X that is greater than a.) In this case, of course, every element of X
is greater than a = −3. To cover all eventualities, we could take x to be
the larger of 0.5, say, and (a + 1)/2. However, it is probably simpler to
consider the cases a > 0 and a ≤ 0 separately, and this is the approach
we will adopt. We are now in a position to commence the proof.

0 a x 1

FIGURE 6.1: Finding the supremum of X = {x ∈ R : 0 < x < 1}.

Proof. Let X = {x ∈ R : 0 < x < 1}.

First note that 1 is an upper bound for X since, by definition, x < 1 for
all x ∈ X.
Now let a < 1. We consider two cases: a > 0 and a ≤ 0.
Suppose a > 0 and let x = (a + 1)/2. Now
a+1
0<a<1 ⇒ 1<a+1<2 ⇒ 1
2 < < 1.
2
In other words, 12 ≤ x < 1, so x ∈ X. Also, since a < 1, we have
x = (a + 1)/2 > (a + a)/2 = a. We have shown that x ∈ X and x > a.
Therefore a is not an upper bound for X in this case.
Now suppose a ≤ 0 and let x = 12 . Then x ∈ X and x > a, so again, a
is not an upper bound for X in this case.
Hence sup{x ∈ R : 0 < x < 1} = 1.
{ }
n2
2. Let X = : n ∈ Z+ . Show that X does not have a supremum.
n+1
As we noted in section 6.5, a proof of non-existence is frequently accom-
plished using proof by contradiction, so we will attempt such a proof
here: assume that X does have a supremum, sup X = M say, and de-
duce a contradiction. Although we will need to work with the general
assumption, sup X = M , in order to see the kind of argument we need,
it may be helpful first to understand why some particular value is not a
supremum.
266

So, suppose sup X = 500. Then, according to definition 6.2, 500 is an

upper bound and any value less than 500 is not an upper bound. To
show that this is false, only one of the two conditions needs to fail. If
500 is an upper bound for X then, for all n ∈ Z+ ,

n2
≤ 500 ⇒ n2 ≤ 500n + 500
n+1
⇒ n2 − 500n ≤ 500
⇒ n(n − 500) ≤ 500.

Clearly this last inequality is not true for all n ∈ Z+ ; take n = 501,
for example. This reasoning is sufficiently generic that we can attempt a
proof by contradiction. However, we just need to be a little careful not
to assume that the supposed supremum is an integer (as we have done
in this example).

{ }
n2
Proof. Let X = : n ∈ Z+ and suppose that sup X exists; call
n+1
it M .
Then M is an upper bound for X so that, for all n ∈ Z+ ,

n2
≤M ⇒ n2 ≤ M n + M
n+1
⇒ n2 − M n ≤ M
⇒ n(n − M ) ≤ M.

This is a contradiction since, for example, when n is any positive integer

satisfying n ≥ M + 1, then n − M ≥ 1, so n(n − M ) ≥ n > M .
Therefore X does not have a supremum.

Note that we have actually proved that X does not have an upper bound
and this is the reason why it does not have a supremum.

3. Each of the two previous examples have been about particular sets.
For our last example, we prove a result about infima and suprema for
arbitrary (non-empty, bounded) sets.

Theorem 6.17
Let A and B be non-empty bounded 4 subsets of R such that A ⊆ B.
Then inf A ≥ inf B and sup A ≤ sup B.

4 In fact, we only need to assume that B is bounded. If B is bounded and A ⊆ B, then

A is bounded — see exercise 6.5.7 (ii).

 267

First note that the logical structure of the theorem is

A ⊆ B ⇒ (inf A ≥ inf B) ∧ (sup A ≤ sup B),

universally quantified over all non-empty bounded subsets of R. As we

noted in section 6.4, P ⇒ (Q ∧ R) is logically equivalent to (P ⇒
Q) ∧ (P ⇒ R). This means that the theorem is equivalent to

(A ⊆ B ⇒ inf A ≥ inf B) ∧ (A ⊆ B ⇒ sup A ≤ sup B)

and we may prove each of the two conjuncts, A ⊆ B ⇒ inf A ≥ inf B

and A ⊆ B ⇒ sup A ≤ sup B separately.
Although we will give the complete structure of the proof below, we will
only prove in detail one of the two conjuncts, leaving the other as an
exercise.

Proof. Let A and B be non-empty bounded sets of real numbers such

that A ⊆ B. We need to show both inf A ≥ inf B and sup A ≤ sup B.
We leave the first of these as an exercise — see exercise 6.5.7 (i).
Let x ∈ A. Then x ∈ B, since A ⊆ B. Now sup B is an upper bound for
B, so x ≤ sup B. We have proved that

x ∈ A ⇒ x ≤ sup B,

which means that sup B is an upper bound for A.

Since the supremum is the least upper bound and sup B is an upper
bound (for A), it now follows that sup A ≤ sup B.

Exercises 6.5

1. Let (G, ∗) and (H, ◦) be groups and let θ : G → H be a morphism.

Prove that:

(i) θ(eG ) = eH where eG is the identity element of G and eH is the

identity element of H,
( ) −1
(ii) θ x−1 = (θ(x)) for all x ∈ G,
(iii) ker θ is a subgroup of G, and
(iv) the image of θ, im θ = {θx : x ∈ G} is a subgroup of H.

2. Let G be a group (written multiplicatively). Prove that G is Abelian if

and only if (gh)−1 = g −1 h−1 for all elements g, h ∈ G.
268

3. Let T : V → W be a linear transformation between vector spaces: see

exercise 5.3.11. Prove that T is injective if and only if ker T = {0}.

4. Let T : V → V be a linear transformation from a vector space V to

itself. Prove that T ◦ T = zero if and only if im T ⊆ ker T .
Here T ◦T is the usual composition of mappings and zero is the mapping
V → V that sends every vector to the zero vector, zero(x) = 0 for all
x∈V.

5. Using definition 6.2 as a model, define the infimum or greatest lower

bound of a non-empty set X of real numbers.

6. For each of the following sets of real numbers, prove that the supremum
is as stated.
{ }
x−1
(i) X = : x ∈ R+ , sup X = 1.
x
{ }
1
(ii) X = x ∈ R : > 2 , sup X = 12 .
x
{ }
1 1
(iii) X = x ∈ R : 2 > 2 , sup X = √ .
x 2
{ }
n { }
(iv) X = : n ∈ Z+ = 12 , 23 , 34 , 45 , . . . , sup X = 1.
n+1
{m }
(v) X = : m, n ∈ Z+ and m < 2n , sup X = 2.
n
7. (i) Complete the proof of theorem 6.17; that is, prove that if A and
B are non-empty bounded subsets of R such that A ⊆ B, then
inf A ≥ inf B.
(ii) Prove the result in the footnote to theorem 6.17: if A and B are
non-empty subsets of R such that B is bounded and A ⊆ B, then
A is bounded.
8. Let f : R → R be a function and let a ∈ R. Note that, if (an ) is a
sequence, then ‘applying’ the function f gives another sequence (f (an )).
Prove that lim f (x) = ℓ if and only if, for every sequence (an ) such that
x→a
an ̸= ℓ for all n ∈ Z+ and lim an = a, the sequence (f (an )) converges
x→∞
to ℓ, lim f (an ) = ℓ.
x→∞
Hint: for the converse, suppose that for every sequence (an ) such that
an ̸= ℓ for all n ∈ Z+ and lim an = a, the sequence (f (an )) converges
x→∞
to ℓ, lim f (an ) = ℓ. Then, to establish lim f (x) = ℓ use a proof by
x→∞ x→a
contradiction: suppose lim f (x) = ℓ and obtain a contradiction.
x→a
Chapter 7
Existence and Uniqueness

7.1 Introduction
So far in this book we have concerned ourselves with the proof of theorems
which are propositions P or universally quantified propositional functions such
as ∀x • P (x). For much of the current chapter we shall turn our attention to
proofs of existence theorems; that is, theorems which assert the existence
within the universe of an object or objects with a certain property, P . We can
symbolise such a theorem by ∃x • P (x).
Some examples of theorems of this form are the following.
(i) Some prime numbers are of the form 32n + 1, where n is an integer.
(ii) Some quadratic equations do not have real roots.
(iii) Not all real numbers are rational.
(iv) There exist sets which have the same cardinality as some of their proper
subsets.
(v) There exist non-Abelian simple groups.

Typically, existence theorems are stated using the phraseology ‘Some ...’ or
‘There exist ...’. Notice, however, that example (iii) above is expressed rather
differently, as the negation of a universally quantified propositional function
¬∀x • P (x). The rule for negating quantified propositions from section 2.4,
page 63, tells us that this is logically equivalent to ∃x • ¬P (x), which is an ex-
istence theorem. In our example, the equivalent existentially quantified state-
ment may be expressed as: ‘some real numbers are irrational’ or ‘there exist
irrational real numbers’. The manner in which these theorems are expressed
seems to suggest that they are asserting the existence of several objects of
the required type. However, this is merely convention and each of the theo-
rems could be expressed in the form ‘There exists at least one object with the
required property.’ To prove a theorem of this type it is sufficient to demon-
strate the existence of a single object of the appropriate type whether or not
there actually exist many such objects. In section 7.5, we shall consider how
we might prove that there is only one object with the required property if this
is the case.

269
270

7.2 Constructive existence proofs

The most obvious way to prove a theorem of the form ∃x • P (x) is to find
a specific object a in the universe for which the proposition P (a) is true.
This method of proof is called proof by construction because we construct
(or find) a specific object a with the required property. How we actually go
about finding or constructing the desired object a will, of course, depend
on the particular theorem under consideration. The proof of the proposition
P (a) may employ any of the methods we have discussed in the previous two
chapters.

Examples 7.1
1. We begin by considering the first of the examples given in the introduc-
tion.

Theorem 7.1
Some prime numbers are of the form 32n + 1, where n is an integer.

Suppose we define the universe of discourse to be the integers and we

define propositional functions
P (x): x is prime
Q(x): x = 32n + 1 for some integer n.
The theorem asserts that there exists an integer satisfying both prop-
erties; this can be symbolised as ∃x • P (x) ∧ Q(x). We need, therefore,
to find a specific integer a for which both P (a) and Q(a) are true. The
simplest approach is to list (some of) the integers a for which Q(a) is
true and then find one of these for which P (a) is also true. Alternatively,
we could list integers a for which P (a) is true — the prime numbers —
and find one of these for which Q(a) is true. However, it is easier to list
integers of the form 32n + 1 than it is to list primes, so we adopt the
former approach.
The positive integers of the form 32n+1 are: 1, 33, 65, 97, 129, 161, 193, . . ..
Which, if any, of these are also prime? We begin:
1 is not prime (by definition),
33 = 3 × 11, so 33 is not prime,
65 = 5 × 13, so 65 is not prime.
However, on testing 97 for factors we find that 97 is, indeed, prime.
Using the result of exercise 6.4.9, page 260, this is straightforward; we
 271

only need to show that none of the prime numbers 2, 3, 5, and 7 is a factor
of 97. Therefore we have found an object with the desired properties and
we can now proceed to the proof itself.

Proof. Consider 97. First note that 97 = 32 × 3 + 1. None of the primes

2, 3, 5, and 7 is a factor of 97 so 97 is prime.
Hence 97 is a prime number of the form 32n + 1 for some integer n.

The proof itself is, of course, very straightforward. It is often the case
with constructive proofs that the proofs themselves are relatively sim-
ple. The hard work frequently lies in finding the required object a, but
this does not show in the final proof. In our example, we have merely
exhibited an object a = 97 with the desired properties.

2. For our second example, we prove the third example introduced in the
previous section.

Theorem 7.2
Not all real numbers are rational.

We noted above that the theorem is equivalent to the proposition: there

exists a real√number which is irrational. In section 6.5, we proved the-
orem 6.12: 2 is irrational. With this theorem now included as part of
our background knowledge of the real numbers, we have the following
one-line proof of the current theorem.
√
Proof. The real number 2 is irrational, by theorem 6.12.
√
The hard work is first discovering√ that 2 is a suitable real number to
consider, and then proving that 2 is indeed irrational. Of course, this
is hidden from view in our one-line proof.
There is also a nice non-constructive proof of this theorem; that is, a
proof which does not actually produce any specific irrational real num-
bers. However, the proof does rely on rather more background knowledge
than the proof given here; see example 7.3.2.

3. When dealing with multiplication (of real numbers for example), we tend
to take the commutative law for granted. This says that, for all x and
y in R, xy = yx. It is used in establishing familiar identities such as
(xy)2 = x2 y 2 and (x + y)2 = x2 + 2xy + y 2 . However, just because it is a
familiar property for multiplication of real numbers does not mean that
it is necessarily satisfied in other contexts. For example, with matrix
multiplication we have the following theorem.
272

Theorem 7.3
Matrix multiplication is not commutative. To be more precise, for any
integer n ≥ 2, multiplication of n × n matrices is not commutative.

The statement of the theorem means that it is not always the case that
AB = BA for n×n matrices A and B. Of course, there may be examples
of some matrices where AB = BA but, to prove the theorem, we must
find two matrices A and B with the property that AB ̸= BA. Two such
matrices are easily found using a little trial and error.
For simplicity, we consider 2 × 2 matrices. Suppose we try
( ) ( )
1 2 −1 1
A= and B = .
3 4 3 2
( )( ) ( )
1 2 −1 1 5 5
Then AB = =
3 4 3 2 9 11
( )( ) ( )
−1 1 1 2 2 2
and BA = = ,
3 2 3 4 9 14
so AB ̸= BA. Since we have found a suitable example, we can proceed
directly to a proof.
( ) ) (
1 2 −1 1
Proof. Let A = and B = .
3 4 3 2
( )( ) ( )
1 2 −1 1 5 5
Then AB = =
3 4 3 2 9 11
( )( ) ( )
−1 1 1 2 2 2
and BA = = ,
3 2 3 4 9 14
so AB ̸= BA.

In this example, trial and error easily produces elements of the universe
with the required property. Indeed, a random choice of matrices is quite
likely to give the desired result. However, this will not always be the
case and sometimes we really do need to be careful when constructing
a suitable example. Staying with 2 × 2 matrices for simplicity, we could
do this here by first considering products of arbitrary matrices and then
making simple choices for the entries as follows.
First note that
( )( ) ( )
a b u v au + bw av + bx
=
c d w x cu + dw cv + dx

and ( )( ) ( )
u v a b ua + vc ub + vd
= .
w x c d wa + xc wb + xd
 273

We wish to choose the entries of the two matrices so that the two prod-
ucts are different. Recall that two matrices are not equal if they differ in
at least one entry. Now, since au = ua, we can ensure that the top-left
entry of the products are different if bw ̸= vc. A simple choice would
be b = w = 0, v = c = 1. With these choices and any choice of a, d,
u and x, we can construct matrices with the required property. Taking
a = d = u = x = 1, for example, gives matrices
( ) ( )
1 0 1 1
A= and B = ,
1 1 0 1

such that AB ̸= BA.

4. For our final example, we show that the set of rational numbers Q is
‘dense’ in the set of real numbers R. This means that, between any two
real numbers x and y, there is a rational number q. We may express this
formally as follows.

Theorem 7.4 (Denseness of the rational numbers)

For all x, y ∈ R such that x < y, there exists a rational number q ∈ Q
such that x < q < y.

To begin to understand how we might find q ∈ Q (given √ real numbers

√
x < y), suppose we try to find a rational number between 2 and 2.1.
The obvious first step is to evaluate these real numbers as decimals:
√ √
2 = 1.41421 . . . and 2.1 = 1.44913 . . . .

Now let √
q = 1.42. Then
√ q is rational, since q = 142/100, and it clearly
satisfies 2 < q < 2.1.
Now let’s set a slightly harder task: find a rational number between

x = 0.12399987 . . . and y = 0.124 . . . .

Again, we may proceed as before: find a terminating decimal that lies

between the two real numbers. Let q = 0.1239999. Then q is rational
since q = 1239999/107 and 0.12399987 . . . < q < 0.124 . . . as required.
The essence of a constructive proof is contained in these examples and
we are now able to proceed to a proof.

Proof. Let x, y ∈ R be such that x < y. Consider the decimal expansions

of x and y and let x = x0 · x1 x2 . . . xn xn+1 . . . xn+k xn+k+1 . . . where
• xn is first position where expansions of x and y differ,
• xn+1 . . . xn+k is a (possibly empty) sequence of 9s, and
• xn+k+1 ̸= 9.
274

Now let q = x0 · x1 x2 . . . xn xn+1 . . . xn+k 9; thus q has the same decimal

expansion as x up to the (n + k)th decimal place, but then terminates
with 9 in the (n + k + 1)st decimal place. Firstly, since q is a terminating
decimal, it is rational, q ∈ Q. Since xn+k+1 < 9 it follows that x < q.
Since xn is the first position where the decimal expansions of x and y
differ, the decimal expansion of y is y = x0 · x1 x2 . . . xn−1 yn . . . where
yn > xn ; therefore q < y.
We have constructed the decimal expansion of a rational number q sat-
isfying x < q < y.

Commentary
Although the proof is constructive in nature, it does not provide an ac-
tual rational number. Of course, the proof could not provide a specific
rational number q because the selection of q must depend on the real
numbers x and y. The proof actually gives a general method of con-
struction of the rational number q depending on the reals x and y. We
can think of the construction given in the proof as an algorithm for
producing q ∈ Q given an ‘input’ of real numbers x and y satisfying
x < y.

Exercises 7.1
1. Prove each of the following existence theorems.

(i) Not all prime numbers are odd.

(ii) There exists an integer k such that k, k + 2 and k + 4 are all prime.
(iii) There exist consecutive positive integers n and n+1 which are both
the sum of squares of two positive integers, i.e., n = a2 + b2 and
n + 1 = c2 + d2 for some a, b, c, d ∈ Z+ .
(iv) There exists a triple of consecutive positive integers n, n + 1, n + 2
each of which is the sum of squares of two positive integers a2 + b2
(a, b ∈ Z+ ).
(v) There exists a complex number z such that z 4 = −1.
(vi) There exists an irrational number x such that x2 is also irrational.
(vii) There exist positive integers n which can be expressed as the sum
of two squares (of positive integers) in two distinct ways: that is,
n = a2 + b2 = c2 + d2 where a, b, c, d ∈ Z+ and {a, b} =
̸ {c, d}.
(viii) There exist integers n > 1 which are both the square of a positive
integer and the cube of a positive integer: n = a2 and n = b3 for
some a, b ∈ Z+ .
 275

(ix) There exist positive integers n which can be expressed as the sum
of two cubes (of positive integers) in two distinct ways: that is,
n = a3 + b3 = c3 + d3 where a, b, c, d ∈ Z+ and {a, b} =
̸ {c, d}.
(x) Let a and b be rational numbers such that a < b. There exists a
rational number q such that a < q < b.
(xi) Let n be an odd positive integer. Then there exists a natural num-
ber m such that n2 = 8m + 1.
(xii) Let x, y ∈ R where x ̸= 0. Prove that there exists n ∈ Z+ such that
n|x| > y.

2. Prove that there exist functions f : A → B and g : B → C satisfying

the following conditions.

(i) The function g is not injective, but the composite function g ◦ f is

injective.
(ii) The function f is not surjective, but the composite function g ◦ f
is surjective.

3. Prove each of the following existence theorems about matrices.

(i) There exists a matrix A not equal to the n × n zero matrix, 0n×n ,
for any n such that A2 = 0n×n .
(ii) There exists an n × n matrix A not equal to In such that A2 = In .
(iii) There exists a matrix A not equal either to 0n×n or to In for any
n such that A2 = A.
(iv) There exist matrices A, B such that AB = In but BA ̸= Im for
any m.
( )
2 1
(v) Let A = .
1 2
( )
6 0
There exists a 2 × 2 matrix B such that BAB = .
0 2

4. This question refers to the set Z8 = {0, 1, 2, 3, 4, 5, 6, 7} with the opera-

tion ×8 , called multiplication modulo 8, defined by:

n ×8 m = remainder when nm is divided by 8.

For example, 2 ×8 6 = 4, 3 ×8 7 = 5, etc.

A multiplicative inverse of an element n ∈ Z8 is an element m such
that n ×8 m = 1.
Prove each of the following.

(i) In Z8 there exists a multiplicative inverse of 5.

276

(ii) Not all elements of Z8 have multiplicative inverse.

(iii) The equation 3 ×8 x = 2 has a solution in Z8 .
(iv) The equation x ×8 x = 1 has a solution in Z8 .
(v) There exist elements x and y in Z8 , both different from 1, such that
x ×8 y = 7.

5. Prove each of the following existence theorems from group theory.

(i) Not all groups are cyclic.

A group G is cyclic if there exists an element g ∈ G such that
every element x ∈ G can be expressed as a power of g, x = g n
for some n ∈ Z. Any such element g is called a generator for G.
For finite groups we may restrict n to be a positive integer. More
generally, we interpret g 0 as being equal
( )to the identity element e,
n
g 0 = e, and we interpret g −n to be g −1 for n ∈ Z+ .
(ii) There exist cyclic groups with more than one generator.
(iii) There exist cyclic groups with only one generator.
(iv) There exist non-Abelian groups.
(v) There exist groups which have no proper subgroups.
Proper subgroups of a group were defined in exercise 5.3.3.
(vi) There exist non-Abelian groups with the property that all of their
proper subgroups are Abelian.

7.3 Non-constructive existence proofs

There are methods of proof of an existence theorem ∃x • P (x) which do not

identify any specific element a in the universe of discourse which has the prop-
erty defined by the predicate P . Any such proof is termed a non-constructive
existence proof or an indirect existence proof. The methods used to
provide non-constructive existence proofs are generally either proof by con-
tradiction or the use of counting arguments.
In a proof by contradiction, we assume the negation of the theorem, that is
¬∃x • P (x), and show that this leads to a contradiction. From section 2.4, we
know that
¬∃x • P (x) ≡ ∀x • ¬P (x).
Thus we assume that ¬P (x) for all x in the universe and show that this leads
to a contradiction.
 277

Examples 7.2

1. Our first example is closely related to Euclid’s theorem about infinitely

many primes, theorem 6.13, which also provides the key result used in
the proof.

Theorem 7.5
There exists a prime number greater than 10100 .

Before embarking on a non-constructive proof, it is worth reflecting on

the difficulties associated with a constructive proof. The number 10100
is enormous; it is greater than the number of atoms in the observable
universe, which is usually estimated to be around 1080 .
Despite the obvious difficulties, there are constructive proofs of the the-
orem. In other words, there are specific known primes larger than 10100 .
For example, several very large so-called Mersenne numbers of the form
2p − 1 (where p is prime) are known to be prime. The smallest of these
that is greater than 10100 and which is also prime is 2521 − 1, which has
157 digits in its decimal notation. The largest of the currently known
primes have millions of digits in their decimal expansions.1
The proof that any one of these extremely large integers is prime requires
a computer (a fast supercomputer in the case of the largest) to perform
all the necessary calculations. This should be contrasted with our non-
constructive proof which follows readily from Euclid’s theorem 6.13. It
is also worth noting that Euclid’s theorem itself had a non-constructive
proof.

Proof. The proof is by contradiction.

Suppose that every prime number p satisfies 2 ≤ p ≤ 10100 . Then, since
there are only finitely many integers between 2 and 10100 , there can be
only finitely many primes. This contradicts theorem 6.13 that there are
infinitely many primes.
Therefore our initial supposition is incorrect and hence there are prime
numbers greater than 10100 .

2. For our second example, we consider the following theorem which is not
expressed as a ‘pure’ existence theorem in the sense that it is not simply
asserting the existence of a particular object with a particular property.
The statement of the theorem involves both universal and existential
quantification and may be symbolised as ∀n ∃θ • P (n, θ) for suitably
defined P and universes for n and θ.
1 In fact, there is an Internet community called GIMPS — the Great Internet Mersenne

Prime Search — dedicated to finding these large primes: see www.mersenne.org.

278

Theorem 7.6
In any n-sided polygon there is an interior angle θ such that
( )
n−2
θ≤ π.
n

Figure 7.1 shows an 8-sided polygon with an interior angle θ marked.

FIGURE 7.1: Interior angle of an n-sided polygon.

The statement of the theorem has two quantifiers: ‘for all n-sided poly-
gons, there exists an interior angle . . . ’. Of course our standard first step
to prove a universally quantified proposition is to consider an arbitrary
object of the appropriate type. In our case, we will start by assuming
P is an arbitrary n-sided polygon. Then the result is a simple existence
theorem about P , namely P has an interior angle of the appropriate
type.
In the proof, we will need the following result:
the sum of the interior angles in any n-sided polygon is (n − 2)π.
This result may be proved using the method of mathematical induction,
which is the subject of the next chapter; see example 8.3.2. For the
current proof, we regard this as part of our background knowledge.

Proof. Let P be an arbitrary n-sided polygon.

The proof is by contradiction, so suppose that every interior angle θ of
P satisfies ( )
n−2
θ> π.
n
Since P has n interior angles, the sum of interior angles satisfies
∑ ( )
n−2
interior angles > n × π = (n − 2)π.
n
This contradicts the result given above as background knowledge. There-
fore, at least one of the interior angles must satisfy
( )
n−2
θ≤ π.
n
 279

3. The proof of our next example does not rely on the method of proof by
contradiction. It is a quite ingenious proof of the existence of an irra-
tional number ab , where a and b are irrational. The proof considers two
numbers of the form ab and proves that one of them is the appropriate
irrational number but without showing which one it is.
We have included the proof for two reasons. Firstly it shows that not
every non-constructive existence proof follows the method of proof by
contradiction or the use of counting arguments described below. As we
have previously indicated, we will never be able to specify precise rules
for how to prove results of a particular type. The second reason is to
indicate that, however many techniques we introduce and guidelines we
give, sometimes there is no substitute for a flash of insight or brilliance.

Theorem 7.7
There exist irrational numbers a and b such that ab is rational.

Let’s begin by considering what might be suitable candidates for a and b.

√ b
√ √2
Suppose we take a = b = √ 2 and consider a = 2 . We do not know
√ 2
(immediately) whether 2 is rational or irrational, even though it
√ √2
probably seems unlikely that it is rational. However, if √ 2 is rational,
√ 2
then the proof will be complete. On the other hand, if 2√ is irrational,
√ 2 √
then we may take √ √
this to be the value of a: let a = 2 and b = 2.
√ 2 √ √√ √
Then ab = ( 2 ) 2 = ( 2)( 2 2) = ( 2)2 = 2, which is certainly
rational.
√ √ √2
So, although we do not know which of a = 2 or a√ = 2 will provide
a suitable example, when taken together with b = 2, we do know that
one of them will. So we may now proceed to the proof.

√
√ 2 √ √2
Proof. Consider 2 . Either 2 is rational or irrational.
√ √2 √
If 2 is rational, then we have proved the theorem: take a = b = 2
which we know to be irrational by theorem 6.12.
√ √2 √ √2 √
However, if 2 is irrational, then let a = 2 and b = 2. Then
√
( √ ) 2 (√ )(√2√2) (√ )2
b
√ 2
a = 2 = 2 = 2 = 2,

which is rational but a and b are irrational.

280

Use of counting arguments

Although counting is apparently an elementary activity, many quite advanced

theorems can be proved using counting arguments. In fact, counting can be
a complex task and there is a branch of mathematics, called enumeration
theory, devoted to techniques of counting. Our aim in this section is to show
how some simple counting theorems can form the basis of non-constructive
existence proofs.

Theorem 7.8 (The Subset Counting Theorem)

If A and B are finite sets such that A ⊆ B and |A| ̸= |B|, then there exists
an element of B which does not belong to A.

Proof. Let A and B be finite sets.

Clearly, if A = B, then |A| = |B|. The contrapositive is: |A| ̸= |B| ⇒ A ̸= B.
Now suppose that A ⊆ B and |A| ̸= |B|. From the contrapositive above, we
have A ̸= B. Hence A ⊆ B and A ̸= B, which means that A is a proper subset
of B. Therefore, there exists at least one element of B that is not an element
of A.

Theorem 7.8 provides a basis for non-constructive existence proofs. If we can

establish A ⊆ B and |A| ̸= |B| for two sets A and B, then the theorem
ensures the existence of an element of B − A without of course identifying any
specific element. We illustrate the use of the Subset Counting Theorem in the
following examples.

Examples 7.3
1. For our first application of the Subset Counting Theorem we prove that
any group with an even number of elements must have an element that
is its own inverse.

Theorem 7.9
Let (G, ∗) be a finite group with an even number of elements. Then there
exists an element g ∈ G such that g ̸= e and g is self-inverse, g −1 = g.

The idea for the proof is to compare the set G with its subset comprising
the identity e together with those elements that are not self-inverse,
g −1 ̸= g: X = {x ∈ G : x = e or x−1 ̸= x}. Providing we can show
that the subset X has a different number of elements, it will follow from
the Subset Counting Theorem that there exists an element of G that is
not in X. Any such element satisfies the required conditions: g ̸= e and
g −1 = g.
 281

Proof. Let (G, ∗) be a finite group with an even number of elements.

Define X ⊆ G to be the identity element together with those elements
that are not self-inverse,

X = {x ∈ G : x = e or x−1 ̸= x}.

Apart from e, all the other elements in X may be grouped in pairs, x

and x−1 . There are clearly an even number of elements that may be
grouped together in pairs like this. Since X also contains the identity
element, it must contain an odd number of elements in total.
However, G contains an even number of elements, so |G| ̸= |X|. There-
fore, there exists an element g ∈ G that does not belong to X. Any such
element g ̸∈ X is self-inverse, g −1 = g and is not equal to the identity
element e.

Commentary
It is worth remarking that the proof assumes that each element of the
group has a unique inverse. This allows the elements to be paired off
with their inverses, g and g −1 . We will consider uniqueness proofs in
section 7.5, where we will also prove that the inverse of each element of
a group is unique; see theorem 7.15.

2. For our second example, we give a non-constructive existence proof of

the existence of irrational numbers, theorem 7.2. Part of our motivation
for this is to highlight that the Subset Counting Theorem applies equally
well to infinite sets, although a proper understanding of the cardinality
of infinite sets is required for this. Recall the corollary to theorem 5.7
which states that if there is a bijective function between two finite sets
then they have the same cardinality.
In fact this can be used as the definition of ‘same cardinality’ for arbi-
trary sets (whether finite or infinite): A and B are defined to have the
same cardinality, |A| = |B|, if there exists a bijection A → B. This
simple definition leads to a sophisticated theory of cardinality of infi-
nite sets, originally developed in the 1870s and 1880s by Georg Cantor,
but which is beyond the scope of this book. However, we mention two
aspects of this general theory. Firstly, unlike a finite set, it is possible
for an infinite set to have the same cardinality as a proper subset; see
exercise 7.2.5. Secondly, the sets of rational and real numbers have dif-
ferent cardinality, |Q| ̸= |R|. The proof of this uses a clever argument,
now known as ‘Cantor’s diagonal argument’; see Garnier and Taylor [6]
for example.
We are now in a position to give an alternative proof of theorem 7.2 that
asserts the existence of irrational real numbers.
282

Alternative proof of Theorem 7.2. First note that Q ⊆ R. As noted

above, Q and R have different cardinalities, |Q| ̸= |R|. Therefore, by the
Subset Counting Theorem, there exists an element in R that is not in
Q. In other words, there exists an irrational real number.

There is a second ‘counting theorem’ that we will use to provide the basis
for non-constructive existence proofs. This is commonly referred to as the
‘Pigeonhole Principle’ as this provides an easy context to visualise the theo-
rem. From a more mathematical point of view, the theorem is more naturally
expressed as a property of functions. We have chosen to state the theorem
using the more colloquial language of pigeonholes, but use functions to give
the proof.

Theorem 7.10 (Pigeonhole Principle)

If k objects are placed in n pigeonholes where k > n, then some pigeonhole
contains more than one object.

Proof. Let A denote a set of k pigeonholes and B denote a set of n objects

placed in the pigeonholes. Suppose k > n.
Define a function f : A → B by

f (object) = pigeonhole in which it is placed.

Recall theorem 5.7 (i) which says that if f : A → B is injective, then |A| ≤ |B|.
The contrapositive of this asserts that if |A| > |B|, then f : A → B is not
injective. Since k > n we may deduce that the function f defined above is not
injective. This means that there exists objects a1 , a2 ∈ A such that a1 ̸= a2 ,
but f (a1 ) = f (a2 ). In other words, there are distinct objects a1 and a2 that
are placed in the same pigeonhole, proving the theorem.

Examples 7.4

1. For our first application of the Pigeonhole Principle, we prove the fol-
lowing simple theorem.

Theorem 7.11
Let A be a set of 6 distinct positive integers. Then there exists a pair of
elements in A whose difference is a multiple of 5.

Proof. Let A = {a1 , a2 , . . . , a6 } be a set of six distinct positive integers.

For each ak , let rk be the remainder when ak is divided by 5. Formally,
rk is the unique integer satisfying ak = 5qk + rk where qk ∈ N and
0 ≤ rk ≤ 4. This follows from the Division Algorithm; see exercise 7.4.5.
 283

There are 6 remainders, r1 , r2 , . . . , r6 , but only five possible values for

the remainders, 0, 1, 2, 3, 4. Therefore, by the Pigeonhole Principle, at
least two of the remainders are equal, say ri = rj . Hence the difference
ai − aj = (5qi + ri ) − (5qj − rj ) = 5(qi − qj ) is a multiple of 5, as
required.

Commentary
Note that we appealed to the Pigeonhole Principle without explicitly
referring to ‘objects’ and ‘pigeonholes’. It is quite common to do this,
assuming that the reader can, if necessary, relate the argument to the-
orem 7.10. If we wanted to refer explicitly to the Pigeonhole Princi-
ple, we would have five pigeonholes labelled by the possible remainders
0, 1, 2, 3, 4. Then each integer ak would be placed in the pigeonhole la-
belled m if its remainder rk = m.
Alternatively, we could have proved the theorem without any reference
to theorem 7.10. Instead we could have modelled the argument in terms
of functions given in the proof of the Pigeonhole Principle. Thus we
would first define a function f : A → {0, 1, 2, 3, 4} by defining f (ak ) to
be the remainder when ak is divided by 5. Then use the cardinalities of
the sets to deduce that f is not injective and so on.

2. Our second example is most naturally a theorem in graph theory. We

have formulated a statement of the theorem in non-graph-theoretic
terms in order to avoid introducing more terminology. Any reader who
has studied the basics of graph theory should have no trouble in refor-
mulating the following theorem in graph theory terminology.

Theorem 7.12
Let S be a network of bus stations and bus routes where each bus route
connects exactly two bus stations. Suppose that there are n bus stations
and m bus routes where m > 12 n(n − 1). Then there exists a pair of bus
stations connected by at least two distinct bus routes.

Proof. Since each bus route connects exactly two stations, we need to
compare the number of bus routes with the number of pairs of bus
stations.
Given that there are n bus stations, the number of pairs is 12 n(n − 1).
However the number of bus routes m is greater than this. Therefore, by
the Pigeonhole Principle, there exists a pair of bus stations connected
by more than one bus route.
284

Commentary
As in the previous example, we have referred to the Pigeonhole Prin-
ciple without explicitly saying what are the objects (bus routes) and
pigeonholes (pairs of bus stations).
We have assumed, as background knowledge, that the number of pairs of
objects that can be formed from n distinct objects is 12 n(n − 1). Readers
may be familiar with the result that says the number of ways of selecting
a subset of r objects from a set of n objects is the so-called binomial
coefficient ( )
n n n!
Cr = = .
r (n − r)!r!
The particular result we need then follows by setting r = 2. For readers
not familiar with the general result, we may reason as follows. To select
a pair of objects from a set of n distinct objects, there are n choices for
the first selected object and then there are n − 1 choices for the second
selected object, giving n(n−1) selections in total. However, each possible
pair is counted twice: the pair {a, b} is counted once when selecting a
first and b second and once when selecting b first and a second. Hence
the total number of pairs is 12 n(n − 1) as claimed.

Exercises 7.2
1. A polygon is convex if every interior angle θ is such that θ < π.
Prove that, in any n-sided non-convex polynomial there is an interior
angle θ such that ( )
n−1
θ< π.
n−3
This theorem appears to assert that every non-convex triangle has a
negative internal angle. Explain this apparent contradiction.
∑
n
2. (i) Let {a1 , a2 , . . . , an } be a set of non-zero integers such that ak <
k=1
n.
Prove that at least one of the integers in the set is negative.
∑
n
(ii) Let {b1 , b2 , . . . , bn } be a set of integers such that b2k < n.
k=1
Prove that at least one of the integers in the set is zero.
3. A tennis club has 2n + 1 members, where n is a positive integer. During
one week, n + 1 matches were played between members.
Prove that some member played more than once during the week.
 285

4. Let x be an irrational number, x ∈ R − Q. Show that, in the decimal

expansion of x, at least one digit n ∈ {0, 1, 2, . . . , 9} occurs infinitely
many times.

5. Using the definition of ‘same cardinality’ given in example 7.3.2, prove

that there is an infinite set that has the same cardinality as a proper
subset. In other words, prove that there is an infinite set A and a proper
subset B, B ⊂ A, such that |A| = |B|.

6. Prove that, in any collection of 12 distinct integers selected from the set
{1, 2, 3, . . . , 30}, there exists a pair with common factor greater than 1.

7. (i) Prove that, for any set of five points located in a rectangle of di-
mensions 6 units by 8 units, there exists a pair which are no more
than 5 units apart.
(ii) Prove that, for any set of n2 + 1 points located√in a square of side
n, there exists a pair which are no more than 2 units apart.
(iii) Prove that, for any set of n2 + 1 points located in an equilateral
triangle of side n, there exists a pair in which the points are no
more than 1 unit apart.

8. Let (G, ∗) be a group with n elements, |G| = n, and let g be an element

of G. Prove that g k = e for some positive integer k ≤ g.

9. Prove the following generalisation of the Pigeonhole Principle.

Theorem (Generalised Pigeonhole Principle)

Let k, n and r be positive integers. If k distinct objects are placed in n
pigeonholes where k > rn, then some pigeonhole contains more than r
objects.

10. Use the Generalised Pigeonhole Principle (given in the previous exercise)
to prove each of the following.

(i) In any set of 750 people, there exist three people with the same
birthday (day and month but not necessarily year of birth).
(ii) If a pair of dice is rolled 45 times, there is a score that occurs at
least 5 times.
(iii) In a certain lottery, six distinct numbers are drawn randomly each
week from the set {1, 2, 3, . . . , 49}. Prove that, in a year of lottery
draws, some number was drawn on at least seven occasions.

11. The following theorem is given as background knowledge.

286

Theorem (Interval theorem)

Let f : [a, b] → R be a continuous function where [a, b] is a closed
interval. Then the image of f is also a closed interval, im f = [c, d] for
some c, d ∈ R.

Use the Interval Theorem to prove the following existence theorem.

Theorem (Intermediate Value Theorem)

Let f : [a, b] → R be a continuous function where [a, b] is a closed
interval. Suppose that k lies between f (a) and f (b); in other words, k
satisfies f (a) ≤ k ≤ f (b) or f (b) ≤ k ≤ f (a). Then there exists c ∈ [a, b]
such that f (c) = k.

12. The Intermediate Value Theorem itself (given in the previous exercise)
can be used as the basis for many non-constructive existence proofs.
Prove each of the following using the Intermediate Value Theorem.

(i) The polynomial p(x) = x3 − 5x + 1 has a root between x = 0 and

x = 1.
2
(ii) The equation ex = x + 10 has a solution between x = 1 and
x = 2.
(iii) Let f and g be two functions, each continuous on [a, b]. Suppose
that f (a) < g(a) and f (b) > g(b). Then f (c) = g(c) for some
c ∈ (a, b).
(iv) Let [f be]continuous on [0, 1] be
( such)that f (0) = f (1). There exists
c ∈ 12 , 1 such that f (c) = f c − 12 .
( )
Hint: consider the function g defined by g(x) = f (x) − f x − 21 .

7.4 Counter-examples

Our principal concern in this book has been with finding and understanding
proofs of theorems. Of course, given a particular proposition, we will not know
whether it really is a theorem until a proof has been found. Suppose we are
presented with a proposition of the form ∀x • P (x), which may or may not be
a theorem. If it turns out that the proposition is not a theorem, then all our
techniques and strategies for finding a proof are bound to fail for the glaringly
obvious reason that no proof exists! Unfortunately, there is no way of showing
that a proposition is a theorem in advance of finding a proof — finding a proof
is precisely how a proposition is shown to be a theorem.
 287

Consider, for example, the proposition:

n
for all non-negative integers n, the integer Fn = 22 + 1 is prime.

In 1640, Pierre de Fermat asserted his belief that this proposition was a theo-
rem, although he was unable to supply a proof. These numbers are now called
Fermat numbers in his honour. Was Fermat correct in his belief? The first
stage in investigating the question is to look at some of the smaller examples.
0
F0 = 22 + 1 = 21 + 1 = 3,
1
F1 = 22 + 1 = 22 + 1 = 5,
2
F2 = 22 + 1 = 24 + 1 = 17,
3
F3 = 22 + 1 = 28 + 1 = 257,
4
F4 = 22 + 1 = 216 + 1 = 65 537,
5
F5 = 22 + 1 = 232 + 1 = 4294 967 297,
6
F6 = 22 + 1 = 264 + 1 = 18 446 744 073 709 551 617.

It is clear that F0 , F1 , and F2 are prime and we can fairly quickly verify that
F3 is prime. With rather more work, F4 can be shown to be prime but even
with a standard scientific calculator, this would be a lengthy and tedious task.
Beyond F4 the Fermat numbers grow very rapidly indeed. In exercise 6.4.9, we
gave a method for testing the primality of a √ positive integer n: test whether
n has a prime factor p in the range 2 ≤ p ≤ n. We cannot imagine anyone
wishing to use this method to test whether or not F5 is prime aided only by a
pocket calculator. Indeed, it was not until 1732, nearly one hundred years after
Fermat proposed the conjecture, that Euler established that F5 is composite
by showing that

F5 = 4 294 967 297 = 641 × 6 700 417.

Of course, this factorisation of F5 shows that Fermat’s conjecture is not a

theorem.2 The factorisation provides a ‘counter-example’ to the proposition
above.
Let us consider again the general situation: suppose we are presented with a
proposition which is a universally quantified propositional function ∀x • P (x).
If we can find a single specific member a of the universe such that P (a) is
false, then ∀x • P (x) is not a theorem. Any element a in the universe such
2 As an aside, it is interesting to note that what took the mathematical com-

munity nearly 100 years to achieve now takes a modest desktop computer no
more than a few seconds. There are various computer algebra packages which will
obtain these factors in a fraction of a second. Indeed, the factorisation of the
next three Fermat numbers, F6 , F7 , and F8 , can be obtained in at most a few
minutes. This is quite impressive since, for example, F8 = 1238926361552897 ×
93461639715357977769163558199606896584051237541638188580280321.
288

that P (a) is false is called a counter-example to the proposition ∀x • P (x).

The method of finding the appropriate element a and showing P (a) is false is
often called proof by counter-example. Since the existence of a counter-
example establishes that ∀x • P (x) is not a theorem, perhaps ‘disproof by
counter-example’ would be a better term.
Given a proposition ∀x • P (x), which may or may not be a theorem, we
are faced with a dilemma. Do we search for a proof or do we try to find a
counter-example? If ∀x • P (x) is a theorem and we opt to search for a counter-
example, then our quest is bound to fail. On the other hand, if ∀x•P (x) is not
a theorem, then any search for a proof will inevitably be unsuccessful. The
choice of which path to take — proof or counter-example — is often based
on experience, intuition, or pure instinct. In practice, the situation is not as
bad as it appears. As we have seen, the first step in the search for a proof
is frequently to look at some examples, and during this initial phase we may
come across a counter-example anyway.
Actually, there is a third possibility which is rather disturbing. It may be
impossible to find a proof of, or a counter-example to, the proposition ∀x •
P (x). Essentially, we have defined a theorem to be a proposition which is
provable from the axioms. There are some situations when neither ∀x • P (x)
nor its negation ¬∀x • P (x) is provable from the axioms. In other words, the
given axiom system is not sufficiently powerful to determine the ‘truth’ of
∀x • P (x). In this case, we say that ∀x • P (x) is undecidable from the given
axioms. Fortunately, such situations are rare and tend to crop up only in the
more esoteric areas of mathematics.

Examples 7.5
1. In example 7.2.1 we introduced the Mersenne prime numbers; that is,
prime numbers of the form 2p − 1 where p is prime. In example 6.1.3,
we proved theorem 6.3, which asserts that ‘p is prime’ is a necessary
condition for ‘2p − 1 is prime’. We now consider whether it is a sufficient
condition; in other words, is the statement ‘for all p, if p is prime, then
2p − 1 is prime’ a theorem?
We begin by testing some small examples:
p=2: 2p − 1 = 2 2 − 1 = 3
p=3: 2p − 1 = 2 3 − 1 = 7
p=5: 2p − 1 = 25 − 1 = 31
p=7: 2p − 1 = 27 − 1 = 127
p = 11 : 2p − 1 = 211 − 1 = 2047

Of these, the first three are clearly prime. It is easy to check that 27 −1 =
127 is prime using the method given in exercise 6.4.9: test whether n
 289
√
has a prime factor p in the range 2 ≤ p ≤ n. Using this method to test
211 −1 = 2047 takes a little longer, but we soon find that 211 −1 = 2047 =
23 × 89, so that we have found a counter-example to the proposition ‘for
all primes p, the integer 2p − 1 is prime’.

2. Find a counter-example to the proposition:

for all x, y ∈ R, if x ≤ y then|x| ≤ |y|.

Solution
We need to find specific real numbers x and y such that the conditional
‘x ≤ y ⇒ |x| ≤ |y|’ is false; in other words. we need to find x and y where
x ≤ y ie true but |x| ≤ |y| is false. If a is non-negative then |a| = a; hence
any counter-example must have at least one of x and y being negative.
Taking both to be negative gives simple counter-examples. For example,
if x = −2 and y = −1, then x < y. However, |x| = 2 > 1 = |y|, so
|x| ≤ |y| is false. We have found our counter-example.

3. Either find a proof or find a counter-example to the proposition

for all sets A, B and C, A ∪ (B − C) = (A ∪ B) − C.

Solution
In this case, we do not initially know whether we should be seeking a
proof or a counter-example. We could start by considering some specific
examples of sets A, B, and C and calculate both A ∪ (B − C) and
(A ∪ B) − C. However, for sets, we have a powerful visualisation in
the form of Venn-Euler diagrams. Although one or more Venn-Euler
diagrams will not constitute a proof of a theorem,3 they can be extremely
useful in pointing the way towards a proof or counter-example. Figure 7.2
represents the sets A ∪ (B − C) and (A ∪ B) − C in two separate Venn-
Euler diagrams.
From the diagrams in figure 7.2, we can see that the ‘difference’ between
the two sets is that A ∪ (B − C) contains A ∩ C as a subset whereas
(A ∪ B) − C does not. This indicates that a counter-example will require
the set A ∩ C to be non-empty.

Counter-example
Let A = {1, 2, 3, 4, 5}, B = {2, 4, 6}, and C = {2, 3, 5}.
3 In fact, this is not strictly true. In her seminal 1994 book [10], Sun-Joo Shin devel-

ops two logical systems related to Venn diagrams where rigorous proofs can be carried out
entirely diagrammatically. Since then, other diagrammatic logics have been developed, in-
cluding those for Venn-Euler diagrams, where proofs are purely diagrammatic. However,
our informal use of Venn-Euler diagrams only points the way towards proofs and they do
not, in themselves, constitute a proof.
290

A B A B

C C

A È (B – C) (A È B) – C

FIGURE 7.2: Finding a set-theory counter-example.

Then B − C = {4, 6}, so A ∪ (B − C) = {1, 2, 3, 4, 5, 6}.

However, A ∪ B = {1, 2, 3, 4, 5, 6}, so (A ∪ B) − C = {1, 4, 6}.
Therefore A ∪ (B − C) ̸= (A ∪ B) − C for these sets.

Commentary
In terms of proving a particular proposition to be false, any counter-
example is as good as any other. However, simpler counter-examples are
to be preferred to more complicated ones. A complicated example may
be difficult to understand and can obscure the underlying reason why
the particular proposition is false. A simpler example is more likely to
come close to the heart of why the proposition is false and thus provide
greater insight.
For instance, another counter-example to the proposition given above
is provided by the sets A = R+ = {positive real numbers}, B = Z =
{integers}, and C = R − Q = {irrational real numbers}. (We leave it
as an exercise to verify that these sets do, indeed, provide a counter-
example; see exercise 7.3.3.) However, it is more difficult to evaluate the
various sets involved here and we may wonder whether the reason that
A ∪ (B − C) ̸= (A ∪ B) − C has something to do with the fact that the
sets are infinite or involve irrational numbers.

4. Find counter-examples to the converse of each part of theorem 5.7. The

theorem states that, for any function f : A → B between finite sets A
and B:

(i) if f is injective, then |A| ≤ |B|;

(ii) if f is surjective, then |A| ≥ |B|.

Solution
 291

In part (i), the converse is: if |A| ≤ |B|, then f is injective. For a counter-
example, we need to find a function f : A → B where |A| ≤ |B| but
where f is not injective. Recall that a function is not injective if there
exist different elements of the domain A with the same image in the
codomain B; figure 3.20 (page 117) gives a diagrammatic representation
of this situation.
In part (ii), the converse is: if |A| ≥ |B|, then f is surjective. For a
counter-example, we need to find a function f : A → B where |A| ≥ |B|
but where f is not surjective. For a function to fail to be surjective,
there must exist elements of the codomain B that do not belong to the
image of f , im f; figure 3.13, page 109, illustrates this situation.
We can achieve these properties with functions like those shown in fig-
ure 7.3. The function in figure 7.3 (i) is not injective but has |A| ≤ |B|,
and the function in figure 7.3 (ii) is not surjective but has |A| ≥ |B| In
fact, the figure essentially defines the counter-examples we are seeking.

f f
A B A B

(i)
(ii)

FIGURE 7.3: Counter-examples to theorem 5.7.

Counter-examples

(i) Let A = {a1 , a2 } and B = {b1 , b2 , b3 } and let f : A → B be defined

by f (a1 ) = f (a2 ) = b2 .
Then |A| = 2 < 3 = |B|. Note that f is not injective since a1 ̸= a2
but f (a1 ) = f (a2 ).
(ii) Let A = {a1 , a2 , a3 } and B = {b1 , b2 } and let f : A → B be defined
by f (a1 ) = f (a2 ) = f (a3 ) = b1 .
Then |A| = 3 < 2 = |B|. Also f is not surjective since b2 ∈ B is
not the image of any element of A, b1 ̸∈ im f.

Commentary
These counter-examples certainly satisfy the ‘simplicity test’ that we
292

discussed in the previous example. Some readers may be surprised that

we have not defined the functions using formulae. However, recall from
definition 3.1 that a function is any rule which associates, to each element
a ∈ A, a unique element b = f (a) ∈ B. Explicitly defining f (ai ) for each
ai ∈ A defines such a rule.
Each of the counter-examples does, however, make implicit assumptions
which, one might argue, should really be made explicit. These are that
the elements of A are distinct and the elements of B are distinct. An
alternative that avoids this issue is to define the sets A and B to contain
familiar elements (rather that ‘arbitrary’ elements a1 , a2 and so on). For
example, in the counter-example in part (i), we could take A = {1, 2} and
B = {1, 2, 3} and define f (1) = f (2) = 2; this is essentially equivalent
to our example.

Exercises 7.3
1. Let f (n) = n2 + n + 41. Then f (0) = 41, f (1) = 43, f (2) = 47, f (3) =
53, f (4) = 61, . . . are all prime.
Find a counter-example to the proposition: for all non-negative integers
n, f (n) is prime.
This formula, which does produce a long sequence of primes, was dis-
covered by Euler. In fact, amongst all expressions of the form n2 +an+b
where a and b are non-negative integers less than 10, 000, there is none
which produces a longer sequence of primes.
2. Find a counter-example to each of the following propositions.
(i) For all real numbers a, b, c, and d, if a > b and c > d, then (a − c) >
(b − d).
(ii) For all positive integers a, b, and c, if c is a factor of a + b, then c
is a factor of a or c is a factor of b.
(iii) f (n) = n2 − n + 17 is prime for all positive integers n.
(iv) 6n + 4n4 is divisible by 5 for all positive integers n.
(v) 3n < 4n4 for all even positive integers n.
(vi) n4 + 1 is prime for all even positive integers n.
3. Verify that the sets A = R+ = {positive real numbers}, B = Z =
{integers}, and C = R − Q = {irrational real numbers} provide a
counter-example to the proposition: A ∪ (B − C) = (A ∪ B) − C for
all sets A, B, C, as claimed in example 7.5.3.
4. Prove or disprove each of the following propositions.
 293

(i) If a and b are rational numbers, then ab is a rational number.

(ii) If a and b are irrational numbers, then ab is an irrational number.
(iii) If a and b are rational numbers and b ̸= 0, then a/b is a rational
number.
(iv) If a and b are irrational numbers, then a/b is an irrational number.
(v) (Compare with exercise 6.4.5 (ii).) Let α and β be irrational num-
bers. Then, for all non-zero integers m and n, αm+βn is irrational.
5. Find a counter-example to each of the following propositions where A,
B, and C are 2 × 2 matrices with real number entries.
(i) If AB = AC and A is not a zero matrix, then B = C.
( )
2 0 0
(ii) The only matrices satisfying the equation A = A are 0 =
( ) 0 0
1 0
and I2 = .
0 1
( ) ( )
1 0 1 0
(iii) If A4 = , then A2 = .
0 1 0 1
(iv) If A and B are distinct (
matrices
) such that( AB)= BA, then either
0 0 1 0
A or B is equal to 0 = or I2 = .
0 0 0 1
6. Prove or disprove each of the following propositions.
(i) For all positive integers a, b and n, if a and b are factors of n, and
ab < n, then ab is a factor of n.
(ii) There exists a positive integer n such that 8n + 1 is prime.
7. For each of the following statements about functions A → B where A
and B are subsets of R, either find a proof or find a counter-example to
the statement.
The definitions of increasing and decreasing functions are given in exer-
cise 6.2.5.

(i) The function f + g is increasing if and only if both f and g are

increasing functions.
The function f + g : A → B is defined by (f + g)(x) = f (x) + g(x).
(ii) The function f g is increasing if and only if both f and g are in-
creasing functions.
The function f g : A → B is defined by (f g)(x) = f (x)g(x).
(iii) Suppose that g(x) ̸= 0 for all x ∈ A. Then the function f /g is
increasing if and only if f is increasing and g is decreasing.
The function f /g : A → B is defined by (f /g)(x) = f (x)/g(x).
294

8. (Compare with theorem 5.13.)

Find a counter-example to the following proposition.
Let S and T be subspaces of a vector space V . Their union S ∪ T is also
a subspace of V .

9. The notion of a subgroup of a group was introduced in exercise 5.3.3.

Find a counter-example to each of the following propositions.

(i) Let H and K be subgroups of a group G. Their union H ∪ K is

also a subgroup of G.
(ii) If every proper subgroup of a group is Abelian, then the group itself
is Abelian.

10. Prove or disprove each of the following propositions.

(i) If A, B, and C are sets such that A ⊆ B and B ⊆ C, then A ⊆ C.

(ii) If A, B, and C are sets such that A ̸⊆ B and B ̸⊆ C, then A ̸⊆ C.
(The symbol ̸⊆ means ‘is not a subset of’.)
(iii) If A and B are n × n matrices such that both A and AB are
symmetric, then B is symmetric.
A symmetric matrix is one that remains unchanged when its
rows and columns are interchanged. The simplest way of formalising
this is to first define the transpose of a matrix. Given an m × n
matrix X, its transpose XT is the n × m matrix whose rows are
the columns of X written in the same order. Then a matrix X is
symmetric if and only if it equals its transpose, X = XT .
(iv) If G is a group and g, h ∈ G, then (gh)n = g n hn for all positive
integers n.
(v) If A, B, and C are sets such that C ⊆ A × B, then C = X × Y for
some X ⊆ A and Y ⊆ B.
Informally, this says that every subset of a Cartesian product is
itself a Cartesian product.

11. (For readers who know about the elementary properties of continuous
and differentiable functions.)
Find a counter-example to each of the following propositions.

(i) Every continuous function f : A → B, where A and B are subsets

of R, is differentiable.
(ii) Every continuous function f : (a, b) → R is bounded, where (a, b)
is an open interval.
A real-valued function f is said to be bounded if its image, im f,
is bounded. Referring to the definition of a bounded set given in
 295

section 6.6, this means that there exist real numbers m and M such
that m ≤ f (x) ≤ M for all x in the domain of the function.
Note: there is a theorem which states that every continuous func-
tion f : [a, b] → R is bounded, where [a, b] is a closed interval. Your
counter-example shows that this theorem does not extend to open
intervals.
(iii) If f : R → R is twice differentiable and f has a local maximum at
x = a, then f ′′ (a) < 0.
Similarly, if g : R → R is twice differentiable and g has a local
minimum at x = a, then g ′′ (a) > 0.
Note: your counter-examples indicate the limitations of what is
frequently called the second derivative test for local maxima and
minima.

7.5 Uniqueness proofs

Sometimes in mathematics, we wish to prove not only that an object with

certain properties exists but also that there is only one such object; in other
words, that the object is unique. The existence part of such a proof was dis-
cussed in sections 7.2 and 7.3. In this section, we focus on the uniqueness part.
To see what is required in such a proof, suppose we define the natural number
N to be the number of objects with the required property. An existence proof
amounts to showing N ≥ 1; in other words, that there is at least one such
object. Given this, to establish uniqueness as well we must show that N = 1,
so that there is exactly one object of the required type.
The method of proving uniqueness is to show that N ≤ 1. This is generally
done by assuming that there are two objects with the required property and
then showing that they are equal. This sounds like a proof by contradiction
but it is not (quite!). In everyday language, when we say ‘I have two objects
of a particular type’, two apples say, we mean ‘I have two different objects of
the particular type’. In our proofs we will assume that there are two objects
of the required type, a and b say, but we will not usually assume that they are
different. The proof is completed by showing that a = b, so that there really
is (at most) one object of the kind under consideration.

Examples 7.6

1. For our first example we will prove the uniqueness part of the following
theorem.
296

Theorem 7.13
For every real number a, the equation x3 = a has a unique real solution.

The existence part of the proof is both subtle and difficult√and amounts
to proving that every real number a has a cube root √3 a. Consider,
for example, how we might prove that the real number 3 2 exists? The
answer to this question goes to the heart of what we mean by the real
numbers. There are two approaches to describing the real numbers R. We
may start with a simpler system, usually the natural numbers N, and
from this ‘build up’ to the real numbers by constructing the integers
Z, the rational numbers Q, and then the real numbers R. The natural
numbers themselves are generally described axiomatically by a system of
five axioms called the Peano axioms; we shall consider one of the Peano
axioms, called the Axiom of Induction, in section 8. The last step in
this process, constructing the real numbers from the rational numbers,
is subtle and there is more than one way of achieving this.
Alternatively, as we mentioned in chapter 4.5, the real numbers may
be described axiomatically as a complete ordered field; thirteen
√ axioms
are required for this. The existence of real numbers such as 3 2 follows
from the Completeness Axiom for the real numbers, which says that
every non-empty subset of R that is bounded above has a supremum.
See definition 6.2 for the definition of the supremum of a non-empty
subset of R.
For the current theorem, we √will not concern ourselves with how the ex-
istence of the real number 3 a is established — either through construc-
tion or via axioms. Instead, we will regard this as part of our background
knowledge of the real numbers and concentrate only on the uniqueness
part of the theorem.

Proof of uniqueness. Suppose that x and y are real numbers such that
x3 = a and y 3 = a. Then x3 = y 3 , so we may reason as follows:
x3 = y 3 ⇒ x3 − y 3 = 0
⇒ (x − y)(x2 + xy + y 2 ) = 0 (factorising)
⇒ x−y =0 2 2
or x + xy + y = 0
⇒ x=y or x2 + xy + y 2 = 0.

We are required to show that x = y. This would now follow if we could

show that x2 + xy + y 2 ̸= 0. Unfortunately, we cannot show this because
it is not true! However, from exercise 5.1.1 (iii) we know that
(x ̸= 0 or y ̸= 0) ⇒ x2 + xy + y 2 ̸= 0.
Using De Morgan’s rule (see Table 2.1 on page 39), the contrapositive
of this is
x2 + xy + y 2 = 0 ⇒ (x = 0 and y = 0).
 297

In particular, x2 + xy + y 2 = 0 ⇒ x = y. From the argument above, it

now follows that:

(x − y)(x2 + xy + y 2 ) = 0 ⇒ x=y or x2 + xy + y 2 = 0
⇒ x = y.

Therefore the equation x3 = a has a unique real solution.

Commentary
The proof follows the structure that we outlined above. That is, we
assumed that x and y are both solutions to the equation and then we
deduced that x = y. The proof itself is elementary in the sense that it
uses little more than factorisation. However, there is a little complication
in the middle where we need to show that ‘if x2 +xy +y 2 = 0, then x = 0
and y = 0’, which requires us to be careful with the logical structure
as we need to understand both the contrapositive and the negation of a
disjunction using De Morgan’s rule.

2. In this example, we first prove the following theorem about inverses of

2 × 2 matrices. For those readers who know group theory or who have
read section 5.3, we then relate this to the uniqueness of inverses in any
group.

Theorem 7.14
Let A be a 2 × 2 matrix. If det A ̸= 0, then A has a unique inverse.

The determinant of a 2 × 2 matrix is given by

( )
a b
det = ad − be.
c d

This time we shall prove both the existence and uniqueness part of the
theorem. The existence part is proved by construction, that is, given A
we find a matrix B such that AB = BA = I2 . We shall simply define
the matrix B and show that it is the inverse of A. (For an explanation
of where the matrix B comes from, a textbook covering basic matrix
theory may be consulted.)
( )
a b
Proof. Let A = and suppose that det A = ad − bc ̸= 0. Define
c d
B to be the 2 × 2 matrix
( ) ( )
1 d −b d/(ad − bc) −b/(ad − bc)
B= = .
ad − bc −c a −c/(ad − bc) a/(ad − bc)
298

Then ( ) ( )
a b 1 d −b
AB = ×
c d ad − bc −c a
( )( )
1 a b d −b
=
ad − bc c d −c a
( )
1 ad − bc −ab + ab
=
ad − bc cd − cd ad − bc
( )
1 ad − bc 0
=
ad − bc 0 ad − bc
( )
1 0
= = I2 .
0 1

Therefore AB = I2 . Verifying that BA = I2 is similar. Hence B is an

inverse of A, B = A−1 . This completes the existence part of the proof.

We now turn to the uniqueness part of the proof. We shall need to

assume, as background knowledge, the associative property of multipli-
cation of 2 × 2 matrices which states that X(YZ) = (XY)Z for all 2 × 2
matrices X, Y and Z.
Suppose that B and C are 2 × 2 matrices such that AB = BA = I2
and AC = CA = I2 . Then

B = BI2 (property of I2 )
= B(AC) (since AC = I2 )
= (BA)C (associative property of matrix multiplication)
= I2 C (since BA = I2 )
= C (property of I2 ).

Therefore B = C, so the inverse of A is unique.

The proof of uniqueness clearly relies on the associative law for matrix
multiplication. In fact, the uniqueness proof is valid for any associative
binary operation with an identity element. In particular, the proof can
be used to show that in any group the inverse of each element is unique,
which we now do.

Theorem 7.15
The inverse of any element g of a group (G, ∗) is unique.

Proof. Let G be a group which, for simplicity we will write multiplica-

tively; see section 5.3.
Let g ∈ G. Suppose that x and y are inverses of g. By definition, this
 299

means that gx = xg = e and gy = yg = e where e is the identity element

of the group. Then

x = xe (since e is the identity)

= x(gy) (since y is an inverse of g)
= (xg)y (associative property in G)
= ey (since x is an inverse of g)
= y (since e is the identity).

Therefore x = y, so the inverse of g is unique.

Note that the proof of theorem 7.15 is essentially identical to the unique-
ness part of the proof of theorem 7.14. In fact, the set of all 2 × 2 invert-
ible matrices — that is, matrices that have inverses — forms a group
under the operation of matrix multiplication. Once we have established
this, then theorem 7.14 is just a special case of theorem 7.15. This is
an example of the economy of labour that we mentioned in section 4.5
as one advantage of using axiom systems. Having proved that inverses
are unique in an (abstract) group, we may deduce this for any structure
that is a group.
3. For our next example, we prove the uniqueness part of the Fundamental
Theorem of Arithmetic. The proof of the existence part was outlined in
section 4.2 and will be dealt with more rigorously in chapter 8.

Theorem 7.16 (Fundamental Theorem of Arithmetic)

Every integer greater than 1 can be expressed as a product of prime
numbers in a manner which is unique apart from the ordering of the
prime factors.

Proof. Let a be an integer greater than 1.

The existence of a prime factorisation for a was given in the Prime
Factorisation Theorem 4.3 (page 132). We therefore only need to prove
the uniqueness part.
Let
a = p1 p2 . . . pm = q1 q2 . . . qn (*)

be two factorisations of a into prime factors. Without loss of generality,

we may suppose that m ≥ n.
Now qn is a factor of a, so qn divides the product p1 p2 . . . pm . If a prime
number divides a product, it must divide one of the factors. (See exer-
cise 6.2.3 for the case of two factors and exercise 8.1.7 for the extension
to several factors.) Hence qn divides one of the p’s; suppose qn divides
300

pk . However pk is prime, so it has no factors other than 1 and pk itself.

Since qn ̸= 1 we have qn = pk .
Dividing the equation (*) by qn = pk and renumbering the p’s (if neces-
sary) gives
p1 p2 . . . pm−1 = q1 q2 . . . qn−1 .
Now consider qn−1 and repeat the same argument that we have just
given: qn−1 must be equal to one of the p’s, then divide by it. Continuing
in this way using qn−1 , qn−2 , . . . q2 in turn produces

p1 p2 . . . pk = q1 , (**)

where k = m − n + 1 ≥ 1 (since m ≥ n).

As before, q1 divides the product p1 p2 . . . pk and so must divide one of
the p’s. If k > 1 so that there is more than one factor on the left-hand
side of (**), then dividing by q1 would leave a product of prime numbers
equal to 1. This is impossible since every prime is strictly greater than
1. Therefore k = 1. This implies that n = m so (**) is just p1 = q1 .
In summary, we have shown that with a suitable re-numbering of the
p’s if necessary,

m = n and p1 = q1 , p2 = q2 , . . . , pm = qn .

Therefore the prime factorisation is unique apart from the ordering of

the factors.

4. For our final example, we show that, for a non-empty subset of R that is
bounded above, its supremum is unique. The definition of the supremum
is given in definition 6.2.

Theorem 7.17 (Uniqueness of supremum)

Let X be a non-empty subset of R that is bounded above. Then the
supremum of X, sup X, is unique.

Proof. Let X be a non-empty subset of R that is bounded above. Let

A and B be two suprema for X. This means that A and B both satisfy
the conditions (i) and (ii) of definition 6.2.
Since A is a supremum for X and B is an upper bound for X, condi-
tion (ii) of the definition ensures that A ≤ B.
Now reverse the roles of A and B. Since B is a supremum for X and
A is an upper bound for X, condition (ii) of the definition ensures that
B ≤ A.
We have shown that A ≤ B and B ≤ A, so A = B, as required.
 301

Exercises 7.4

1. (i) Prove that the equation ax = b, where a and b are fixed real num-
bers and a ̸= 0, has a unique solution.
(ii) Prove that, if a is a positive real number, then the equation x2 = a
has a unique positive solution.
√
As in the example 7.6.1, you may assume the existence of a for
any a > 0.
√
(iii) Assuming the existence of a fifth root 5 a of any a ∈ R, prove that,
for every real number a, the equation x5 = a has a unique solution.

2. Prove that, if a, b, c, and d are real numbers such that ad − bc ̸= 0,

then for all real numbers s, t there exists a unique solution (x, y) to the
simultaneous equations
ax + by = s
cx + dy = t.

3. Prove that every integer a > 2 can be expressed uniquely as a = 2n b

where n is an integer and b is an odd integer.

4. Prove that there is a unique prime number p for which p2 + 2 is also

prime.

5. Prove the following theorem. The theorem makes precise the idea that
dividing a positive integer n by a positive integer m gives a ‘quotient’ q
and ‘remainder’ r. For example, dividing 131 by 9 gives quotient 14 and
remainder 5, which we can express as 131 = 14 × 9 + 5.

Theorem (The Division Algorithm)

Let m and n be positive integers. Then there exist unique q, r ∈ N such
that n = qm + r and 0 ≤ r < m.

Hint: for the existence part, consider the set {n − qm : q ∈ N} and

reason that this set must have a smallest non-negative element r.

6. Let f : A → B be a bijective function. Prove that the inverse function

f −1 : B → A is unique.

7. Let (G, ∗) be a group. Prove each of the following uniqueness results.

(i) Uniqueness of the identity: the identity element e is unique.

(ii) Uniqueness of the solution of equations (1): for all g, h ∈ G, the
equation gx = h has a unique solution for x ∈ G.
302

(iii) Uniqueness of the solution of equations (2): for all g, h ∈ G, the

equation xg = h has a unique solution for x ∈ G.

8. Let V be a vector space. Prove each of the following uniqueness results.

(i) Uniqueness of the zero vector: 0 is unique.

(ii) Uniqueness of negatives: for all v ∈ V the negative, −v is unique.

9. (i) Prove that x = 0 is the only real number that satisfies |x| < ε for
all ε ∈ R+ .
(ii) Let (an ) be a convergent sequence. Prove that its limit is unique.

10. Let f : A → B be a function where A and B are subsets of R. Suppose

that lim f (x) exists, where a ∈ A. Prove that the limit is unique.
x→a
Chapter 8
Mathematical Induction

8.1 Introduction
In this chapter we consider in some detail one particular method of proof,
called proof by mathematical induction, which may be used to prove state-
ments about the natural numbers or positive integers. In other words, the
method will apply to the proofs of propositions of the form ∀n • P (n) where
the universe for n is the set of natural numbers N or the set of positive inte-
gers Z+ . The following are examples of the kinds of results that we may prove
using the method of mathematical induction.

1. The sum of the first n positive integers is 12 n(n + 1); symbolically,

for all n ∈ Z+ , 1 + 2 + 3 + · · · + n = 12 n(n + 1).

2. For every n ∈ N, the expression 5n + 3 is divisible by 4.

3. For all finite sets A, if A has cardinality n, then its power set P(A) has
cardinality 2n .

The first of these is clearly of the form ∀n ∈ Z+ • P (n) and the second is
of the form ∀n ∈ N • P (n). However, the third statement is not obviously of
one of these two forms. Note that, to say that a set A is finite means it has
cardinality |A| = n where n is some natural number. This means we could
rephrase the third statement as ‘for all n ∈ N, if A is a set with cardinality
|A| = n, then its power set has cardinality |P(A)| = 2n ’ and this is of the form
∀n ∈ N • P (n). This is not unusual. Sometimes a statement is not obviously
one about N or Z+ , but is more naturally about some other mathematical
object such as a set, function, group, graph or whatever. But, by rethinking
the statement, it may be rephrased to be one about say the positive integers;
the advantage of this is that it brings into play the possibility of using the
powerful method of mathematical induction.
Despite its name, any proof by mathematical induction uses deductive rea-
soning, which is the basis of all mathematical proofs; it does not rely on the
kind of inductive reasoning described in section 1.3.

303
304

To motivate the idea behind the method, suppose we are given the following
instruction.

• If you write down an integer k, then you must also write down the integer
k + 1.

What do we need to do to satisfy the instruction? The simple answer is:

nothing at all! The statement is a conditional of the form W (k) ⇒ W (k + 1)
where W (k) stands for ‘you write down k’. A conditional statement is true
whenever its antecedent is false, so the statement above is true when you write
nothing at all. Now let’s add another instruction. Consider the following two
instructions.

1. Write down the number 1.

2. If you write down an integer k, then you must also write down the integer
k + 1.

What do we now need to do to satisfy these instructions? Firstly, to obey the

first instruction, we need to write down ‘1’. Then the second instruction ‘kicks
in’: because we have now written down ‘1’ we then need to write down ‘2’. But
once we have written down ‘2’, to follow the second instruction, we also need
to write down ‘3’. Then we have written ‘3’ so, to follow the second instruction,
we need to write ‘4’, then ‘5’, then ‘6’, and so on. In other words, to follow
both instructions, we would need to write down all the positive integers. Of
course, since Z+ is infinite, we cannot actually do this. So, although we cannot
actually carry out both instructions, they carry within them a ‘process’ for
‘generating’ all positive integers.
This idea is formalised within the natural numbers by the Axiom of Induction,
which we give below. The natural numbers can be defined by five relatively
simple axioms, now called the Peano Axioms, after the 19th-century Italian
mathematician Giuseppe Peano, who presented them in a book in 1889. All
of the properties of the arithmetic of the natural numbers may be developed
from the five Peano axioms. The Axiom of Induction is one of Peano’s axioms.

Axiom of Induction
Let S be a subset of N such that:

(i) 0 ∈ S, and
(ii) for all k, k ∈ S ⇒ (k + 1) ∈ S.

Then S contains all the natural numbers, S = N.

 305

The Axiom of Induction gives a way of establishing that some property P (n)
holds for all natural numbers. Given a property P (n), define S = {n ∈ N :
P (n)} to be the set containing all those natural numbers for which P (n)
is true. Thus n ∈ S if and only if P (n) is true. Hence the two conditions
in the Axiom of Induction become (i) P (0) is true and (ii) if P (k) is true,
then P (k + 1) is true. This formulation is frequently called the Principle of
Mathematical Induction.

Principle of Mathematical Induction

Let P (n) be a propositional function with universe N. Suppose
that:

(i) P (0) is true, and

(ii) for all k ∈ N, if P (k) is true, then P (k + 1) is true.
Then P (n) is true for all n ∈ N; that is, ∀n ∈ N • P (n) is true.

8.2 Proof by induction

The Principal of Mathematical Induction provides a scheme for proving the-

orems that may be formulated as ∀n ∈ N • P (n). There are two stages in the
proof. Firstly, prove that P (0) is true; in other words, prove that the result
under consideration is true when n = 0. This is called the ‘Base case’ and will
frequently amount to a simple checking of some condition of equation. For the
second part, we need to prove that the conditional statement P (k) ⇒ P (k +1)
is true for all k ∈ N. This is called the ‘Inductive step’ and will be proved using
the method of direct proof; that is, we assume P (k) and, from this, deduce
P (k + 1). Assuming P (k) is called the ‘Inductive hypothesis’ and it appears to
be a slightly odd assumption to make. The reason is that we are attempting
to prove P (n) for all n ∈ N, so assuming P (k), for an arbitrary k, appears
at first sight to be assuming what we are trying to prove. However, the as-
sumption of P (k) is only made in order to deduce P (k + 1). In other words,
we only assume the truth of the result for some integer k in order to deduce
the result for the ‘next’ integer k + 1. We summarise this discussion in the
following scheme for proving results about the natural numbers.
306

Method of Proof by Mathematical Induction

Let P (n) be a propositional function with universe N.
Base case Prove that P (0) is true.
Inductive step Prove that, for all k ∈ N, P (k) ⇒ P (k + 1).
To do this:
assume P (k); this is the inductive hypothesis,
then deduce P (k + 1).
Conclusion:
P (n) is true for all n ∈ N; that is, ∀n ∈ N • P (n) is true.

There is an obvious modification to the method if the universe is the positive

integers Z+ rather than the natural numbers N. In this situation the base
case will be for the smallest positive integer rather than the smallest natural
number; in other words, the base case will be to prove that P (1) is true. We
now illustrate the method, initially by proving the three results mentioned in
the previous section.

Examples 8.1
1. One of the common contexts in which proof by induction is first encoun-
tered is to prove results that involve the summation of a finite number
of terms. We illustrate the approach by proving that the sum of the first
n positive integers is 12 n(n + 1).

Theorem 8.1
For all n ∈ Z+ , 1 + 2 + 3 + · · · + n = 21 n(n + 1).

The base case is straightforward. For the inductive step we need to

assume the result for n = k; that is, we assume

1 + 2 + 3 + · · · + k = 12 k(k + 1).

From this assumption, we then need to deduce the result for n = k + 1

1 + 2 + 3 + · · · + k + (k + 1) = 12 (k + 1)(k + 2).

The left-hand side of this second equation is the same as the left-hand
side of the previous equation, but with the addition of a single extra
term (k + 1). Hence, to deduce the result for n = k + 1 we will take the
equation for n = k, add (k + 1) to both sides and then reorganise the
new right-hand side. This right-hand side is 21 k(k + 1) + (k + 1), which
 307

has a common factor of k + 1. In fact, it is better to regard the common

factor as being 21 (k + 1) by writing the expression as
1
2 k(k + 1) + (k + 1) = 12 k(k + 1) + 2 × 12 (k + 1) = 21 (k + 1) (k + 2) .

This final expression is what we are seeking for the right-hand side of
the equation for k + 1. We are now in a position to organise these con-
siderations into a proof.

Proof. The proof is by mathematical induction.

Base case. When n = 1, LHS = 1 and RHS = 1
2 × 1 × 2 = 1. Hence
the result holds when n = 1.
Inductive step. Assume that 1 + 2 + 3 + · · · + k = 12 k(k + 1); this is the
inductive hypothesis. Adding (k + 1) to both sides gives

1 + 2 + 3 + · · · + k + (k + 1) = 1
2 k(k + 1) + (k + 1)
= 1
2 k(k + 1) + 2 × 12 (k + 1)
1
= 2 (k + 1) (k + 2)
1
= 2 (k + 1) ((k + 1)k + 1) .

This is the result for n = k + 1 and hence completes the inductive step.
Therefore, for all n ∈ Z+ , 1+2+3+· · ·+n = 12 n(n+1), by induction.

Commentary
There are a number of points to make about the organisation and struc-
ture of the proof. Firstly, as with some other methods of proof such as
proof by contradiction, it is usually a good idea to signal at the begin-
ning that the proof uses the method of mathematical induction. The
reader then knows to expect a particular structure to the proof.
To establish the base case, we just need to check that the equation
is satisfied when n = 1. To verify an equation, it is good practice to
evaluate separately the left- and right-hand sides. In this way, it will be
clear what is being evaluated. Less clear would be to write ‘1 = 12 ×1×2’
because it does not explicitly link this (trivial) equation with the result
under consideration.
Finally, at the end of the proof, it is also good practice to conclude with
the result that has been established and the reason that the conclusion
is justified; the phrase ‘by induction’ signals this.
It is quite common to use the letter n, rather than k, within the body of
the inductive step. Of course, the letter used is not important. We have
chosen to use k rather than n to keep the reasoning within the inductive
308

step slightly separate from the conclusion. Using n makes the inductive
hypothesis look identical to the result that we are trying to establish.
However, the choice of letter is a matter of personal taste on behalf of
the proof writer.
2. In this example, we prove the following result which was the second
example mentioned in the previous section. Here the universe is the
natural numbers, so the base case will be n = 0.

Theorem 8.2
For every n ∈ N, the expression 5n + 3 is divisible by 4.

Before commencing the proof, we just need to remind ourselves what

it means to say that ‘5n + 3 is divisible by 4’. This means that the
expression is some integer multiple of 4, so there exists an integer, a say,
such that 5n + 3 = 4a.

Proof. The proof is by mathematical induction.

Base case. When n = 0, 5n + 3 = 50 + 3 = 1 + 3 = 4, which is clearly
divisible by 4. Hence the result holds when n = 0.
Inductive step. Assume that 5k + 3 is divisible by 4; then 5k + 3 = 4a
for some integer a. Hence 5k = 4a − 3. Now
5k+1 + 3 = 5 × 5k + 3
= 5(4a − 3) + 3 from the inductive hypothesis
= 20a − 12
= 4(5a − 3) where 5a − 3 ∈ Z.
Therefore 5k+1 + 3 is divisible by 4. This completes the inductive step.
Therefore, for all n ∈ N, the expression 5n + 3 is divisible by 4, by
induction.

3. In section 5.2 we gave a direct proof of the following theorem. We also

indicated a second proof using the Binomial Theorem; see exercise 5.2.2.

Theorem 5.5
For all finite sets A, if A has cardinality n, then its power set P(A) has
cardinality 2n .

As we indicated in the previous section, we may regard this theorem as

being a candidate for proof by induction by writing it as ‘for all n ∈ N,
if A is a set with cardinality |A| = n, then its power set has cardinality
|P(A)| = 2n ’. In our proof, we signal this at the beginning by stating
that the proof uses induction ‘on the cardinality of the set’.
 309

Proof. The proof is by mathematical induction on the cardinality of A.

Base case. Let A be a set with cardinality 0; in other words A is the
empty set, A = ∅. Then P(A) = {∅}, which is a set with one element
(the empty set); see exercise 3.8.1. Hence, |P(a)| = 1 = 20 , so the result
holds when n = 0.
Inductive step. Assume that, for all finite sets, if the set has cardinality
k, then its power set has cardinality 2k .
Let A be a set with cardinality k + 1. Select an element a∗ ∈ A. We may
divide the subsets of A into two types: those that contain a∗ and those
that don’t.
Any subset B ⊆ A that does not contain a∗ is simply a subset of A−{a∗ },
which is a set with k elements. Hence, by the inductive hypothesis, there
are 2k such subsets B.
Let C ⊆ A be a subset that does contain a∗ ; then C = B ∪ {a∗ }, where
again B is a subset of A − {a∗ }. As before, A − {a∗ } is a set with k
elements, so there are 2k such subsets B, by the inductive hypothesis.
Hence there are 2k subsets C = B ∪ {a∗ } that do contain a∗ .
Therefore the total number of subsets of A is 2k + 2k = 2k+1 . This
completes the inductive step.
Therefore, for all n ∈ N, if A is a set with cardinality |A| = n, then its
power set has cardinality |P(A)| = 2n , by induction.

Commentary
The two previous proofs of theorem 5.5 involved counting arguments
of one sort or another. The proof we gave on page 191 was based on
counting the number of ways a subset of A could be ‘built’ by choosing
whether to include or exclude each element of A from the subset. The
proof outlined in exercise 5.2.2 proceeds by counting the number of r-
element subsets for r = 0, 1, 2, . . . , n. The proof we have given here is
also based on a counting argument: counting the number of subsets that
include a particular element a∗ and also counting the number of subsets
that exclude a∗ .
There is one aspect of the proof that was implicit, which we now wish
to make explicit. In concluding that there are 2k subsets C that contain
the element a∗ , we are implicitly assuming that there is a bijective cor-
respondence between the sets C = B ∪ {a∗ } and the subsets B ⊆ A that
do not contain a∗ . The bijection is, of course, defined by B ∪ {a∗ } 7→ B,
but we have not explicitly proved that this defines a bijection. We have
also used implicitly the corollary to theorem 5.7 (page 198) to deduce
that there are the same number of subsets C = B ∪ {a∗ } containing a∗
as there are subsets B ⊆ A − {a∗ } that do not contain a∗ .
310

4. In example 6.1.3, we used the following theorem in the proof of theo-

rem 6.3 which asserted that, for n ∈ Z+ , if 2n − 1 is prime, then n is
prime.
A geometric progression is a sequence of terms where each term is in
a fixed constant ratio to the previous term; thus each term is obtained
from the previous term by multiplying by a fixed real number, called the
common ratio. If the first term is b and the common ratio is a, then
the sequence of terms is b, ba, ba2 , ba3 , . . .. In the theorem we evaluate
the sum of the first n terms of a geometric progression with first term
1. Taking the first term equal to 1 is not a significant restriction as the
sum of the general progression can be obtained from the theorem by
multiplying both sides of the equation by b.

Theorem 8.3
For all real numbers a ̸= 1 and n ∈ Z+ ,
an − 1
1 + a + a2 + . . . + an−1 = .
a−1

Proof. The proof is by mathematical induction on n.

a1 − 1
Base case. When n = 1, LHS = 1 and RHS = = 1. Hence the
a−1
result holds when n = 1.
ak − 1
Inductive step. Assume that 1 + a + a2 + . . . + ak−1 = .
a−1
Adding the next term of the progression, ak , to both sides gives

ak − 1
1 + a + a2 + . . . + ak−1 + ak = + ak
a−1
ak − 1 ak (a − 1)
= +
a−1 a−1
a −1+a
k k+1
− ak
=
a−1
ak+1 − 1
= .
a−1

This is the result for k + 1 and hence completes the inductive step.
an − 1
Therefore, for all n ∈ N, 1+a+a2 +. . .+an−1 = , by induction.
a−1

5. Consider the following ‘theorem’ and ‘proof’. Since the claim in the
‘theorem’ is patently false, there must be an error in the proof. The
question is: What is wrong with the proof?
 311

‘Theorem’
All rectangles have the same area.

‘Proof.’ Let P (n) be the statement: in any set of n rectangles, all the
rectangles have the same area.
We shall prove that P (n) is true for all positive integers n by induction.
This will prove the theorem.1
Base case. The result P (1) is obviously true since in any set containing
only one rectangle, all the rectangles in the set have the same area.
Inductive step. Suppose that, P (k) is true; in other words, in any set of
k rectangles, all the rectangles have the same area.
Let {R1 , R2 , . . . , Rk+1 } be an arbitrary set of k + 1 rectangles.
Since {R1 , R2 , . . . , Rk } is a set of k rectangles, all these have the same
area. Similarly, since {R2 , R3 , . . . , Rk+1 } is also a set of k rectangles, all
these have the same area.
Hence, taking the union of these two sets, all the rectangles in
{R1 , R2 , . . . , Rk+1 } have the same area.
Hence P (k + 1) is true and completes the inductive step.
Therefore, for all positive integers n, in any set of n rectangles, all the
rectangles have the same area, by induction.

Solution
The base case is uncontroversial. For any rectangle R, all rectangles
in the set {R} have the same area. The problem must lie in the in-
ductive step. Let’s examine one particular case of the inductive step,
say, when k = 4. The inductive hypothesis in this case is that every
set of four rectangles comprises rectangles with the same area. Now
let {R1 , R2 , R3 , R4 , R5 } be any set of five rectangles. By the induc-
tive hypothesis, both the sets of four rectangles, {R1 , R2 , R3 , R4 } and
{R2 , R3 , R4 , R5 }, contain rectangles that all have the same area; hence
all the rectangles in {R1 , R2 , R3 , R4 , R5 } also all have the same area.
This reasoning appears to be perfectly sound and indeed it is! The im-
plication considered here, P (4) ⇒ P (5), is a true proposition.
So what has gone wrong? Thinking about the truth of the result it-
self gives the clue. As we have indicated, the result is true for n = 1
(the base case) but clearly it is not true when n = 2. There are sets
1 To see why this is equivalent to the stated theorem, suppose we have established that

every finite set of rectangles contains rectangles which are all of the same area. The stated
theorem then follows by contradiction, as follows. Assume that not all rectangles have the
same area. Then there exist two rectangles, R and R′ , say, with different areas. Then any set
of rectangles containing both R and R′ contradicts the result that says all sets of rectangles
contain only rectangles with the same area.
312

containing two rectangles {R1 , R2 } that have different areas. In fact,

it is the implication P (1) ⇒ P (2) that is false. The inductive step
P (k) ⇒ P (k + 1) relies on there being a non-empty intersection between
the sets {R1 , R2 , . . . , Rk } and {R2 , R3 , . . . , Rk+1 } so that the ‘same
area’ property may be transferred to their union {R1 , R2 , R3 , . . . , Rk+1 }.
When k ≥ 2 this is, indeed, the case. However, when k = 1, removing
the first and last elements gives two disjoint sets, {R1 } and {R2 }, so
the ‘same area’ property does not transfer to their union. In summary,
we have: P (1) ̸⇒ P (2) ⇒ P (3) ⇒ P (4) ⇒ P (5) ⇒ · · · .
This example illustrates the importance of showing that the inductive
step P (k) ⇒ P (k + 1) is true for all k. In the present case, P (1) is true
and P (k) ⇒ P (k + 1) is true for all k ≥ 2, but this is not sufficient.

Exercises 8.1

1. Prove each of the following by induction.

(i) The sum of the first n odd positive integers is n2 :

1 + 3 + 5 · · · + (2n − 1) = n2 .
1
(ii) The sum of the squares of the first n positive integers is 6 n(n +
1)(2n + 1):
1
12 + 22 + 32 · · · + n2 = n(n + 1)(2n + 1).
6

(iii) For all positive integers n,

1 1 1 1 n
+ + + ··· + = .
1×2 2×3 3×4 n(n + 1) n+1

(iv) The sum of the cubes of the first n positive integers is 14 n2 (n + 1)2 :

1 2
13 + 23 + 33 · · · + n2 = n (n + 1)2 .
4

(v) For all positive integers n,

1 × 3 + 2 × 4 + 3 × 5 + . . . + n(n + 2) = 16 n(n + 1)(2n + 7).

(vi) For all n ∈ Z+ ,

1 × 6 + 2 × 7 + 3 × 8 + · · · + n(n + 5) = 31 n(n + 1)(n + 8).

 313

(vii) Generalising the two previous examples, let m be a (fixed) positive

integer. Then
1 × m + 2 × (m + 1) + 3 × (m + 2) + · · · + n(m + n − 1)
= 16 n(n + 1)(n + 3m − 2).

(viii) For all positive integers n,

1 2 3 n−1 1
+ + + ··· + =1− .
2! 3! 4! n! n!
2. Prove each of the following by induction.

(i) For all positive integers n, 2n > n.

(ii) For all positive integers n, the expression 9n + 7 is divisible by 8.
(iii) For all n ∈ Z+ , the expression 7n − 3n is divisible by 4.
(iv) For all n ∈ Z+ , the expression 11n − 4n is divisible by 7.
(v) Generalising the previous two examples, let a and b be two positive
integers such that a − b = m ∈ Z+ . Then, for all n ∈ Z+ , the
expression an − bn is divisible by m.
(vi) For all positive integers n, the expression 2n+2 + 32n+1 is divisible
by 7.
(vii) For all positive integers n, the expression 42n+1 + 3n+2 is divisible
by 13.

3. Prove each of the following results about products by mathematical

induction. In each case, try to find an alternative proof that does not
use mathematical induction.
(2n)!
(i) For all n ∈ Z+ , 1 × 3 × 5 × · · · × (2n − 1) = .
2n n!
2 4 6 2n √
(ii) For all n ∈ Z+ , × × × ··· × > 2n + 1.
1 3 5 2n − 1
4. (i) Prove that, for all n ∈ N and all real numbers x > −1,

(1 + x)n ≥ 1 + nx.

(ii) Prove that, for all n ∈ Z+ , if x1 , x2 , . . . , xn are positive real num-

bers, then

(1 + x1 )(1 + x2 ) . . . (1 + xn ) ≥ 1 + (x1 + x2 + · · · + xn ).

(iii) Prove that, for all n ∈ Z+ , if x1 , x2 , . . . , xn are real numbers in the

interval (0,1), (that is, 0 < xr < 1 for r = 1, 2, . . . , n), then

(1 − x1 )(1 − x2 ) . . . (1 − xn ) > 1 − (x1 + x2 + · · · + xn ).

314

5. (i) Prove by induction that, for all n ∈ Z+ , n3 − n is divisible by 6.

(ii) Give a direct proof of the result in part (i). Which of the two proofs
do you prefer?
(iii) Using the result of part (i), give a direct proof that n4 − n2 is
divisible by 12.
( )
cos θ − sin θ
6. Let R(θ) be the 2 × 2 matrix R(θ) = .
sin θ cos θ
The matrix represents an anti-clockwise rotation of the plane R2 about
the origin by an angle θ.
n
Prove that, for all positive integers n, R(nθ) = (R(θ)) .
Hint: you may assume the following trigonometric identities.
sin(A + B) = sin A cos B + cos A sin B
and cos(A + B) = cos A cos B − sin A sin B.

7. Prove that if a1 , a2 , . . . an are positive integers and p is a prime number

that is a factor of the product a1 a2 . . . an , then p is a factor of one of
the integers ar for some r = 1, 2, . . . , n.

8. The edges of a triangle are divided into n equal segments by inserting

n − 1 division points. Lines are drawn through each of these division
points parallel to each of the three edges of the triangle, thus forming a
set of small triangles as illustrated in the figure (for n = 4).
Prove that there are n2 small triangles.

9. Prove that, for all positive integers n,

( )
sin (n + 12 )θ
cos θ + cos 2θ + · · · + cos nθ = = 12 .
2 sin 12 θ

Hint: you may assume the following trigonometric identity

( ) ( )
A+B A−B
sin A − sin B = 2 cos sin .
2 2

10. Use mathematical induction to prove that, for all n ∈ Z+ , x + 1 is a

factor of x2n−1 + 1.
 315

11. (i) Prove that, for all n ∈ Z+ , a 2n × 2n grid with a single corner
square removed may be covered with L-shaped tiles like this one.

(ii) From part (i), prove that 22n − 1 is divisible by 3 for all n ∈ Z+ .
(iii) Prove the result in part (ii) by induction without reference to the
result in part (i).
(iv) Using a similar approach to part (i), prove that for all n ∈ Z+ , a
2n × 2n × 2n cube with a single small cube (with side length 1)
removed from one corner may be constructed using pieces like this
one.

(v) Deduce a ‘divisibility result’ about numbers of the form 23n − 1

from part (iv)?
Then give a standard proof by induction of this result.
12. Prove the following versions of De Morgan’s Laws for n sets.
(i) For all sets A1 , A2 , . . . , An ,
A1 ∪ A2 ∪ . . . ∪ An = Ā1 ∩ Ā2 ∩ . . . ∩ A¯n .
(ii) For all sets A1 , A2 , . . . , An ,
A1 ∩ A2 ∩ . . . ∩ An = Ā1 ∪ Ā2 ∪ . . . ∪ A¯n .
13. Let (G, ∗) be a group. Prove each of the following.
( )n
(i) Let g, h ∈ G. Then, for all n ∈ Z+ , h−1 gh = h−1 g n h.
−1
(ii) Let g1 , g2 , . . . , gn be n elements of G. Then (g1 g2 . . . gn ) =
gn−1 . . . g2−1 g1−1 .
(iii) Suppose that (G, ∗) is Abelian and let g, h ∈ G. Then, for all n ∈
Z+ , (gh)n = g n hn .
( ) ( )
1 1 1 n
14. (i) Let A = . Prove that, for all n ∈ Z+ , An = .
0 1 0 1
( ) ( )
1 0 1 0
(ii) Let A = . Prove that An = for all n ∈ Z+ .
−1 2 1 − 2n 2n
 
5 −2 −2
(iii) Let A = 2 1 −2. Prove that, for all n ∈ Z+ ,
2 −2 1
 
2 × 3n − 1 1 − 3n 1 − 3n
An =  3n − 1 1 1 − 3n  .
3 −1
n
1−3 n
1
316

15. This example generalises theorem 5.15.

Let A be an m × m matrix and let λ be an eigenvalue of A. Prove that,
for all n ∈ Z+ , λn is an eigenvalue of An .
Show also that the eigenspace of λ for the matrix A, EA (λ), is a subspace
of the eigenspace of λn for the matrix An , EAn (λn ).
16. Let f be a real-valued function such that lim f (x) = ℓ. For n ∈ Z+
x→a
n
define the function f n by f n (x) = (f (x)) for all x in the domain of f .
Prove by induction that, for all n ∈ Z+ , lim f n (x) = ℓn .
x→a
Note that you may assume, as background knowledge, the result of ex-
ercise 5.4.5.

8.3 Variations on proof by induction

In this section, we consider two variations to the method of proof by induction:

the first variation modifies the base case and the second modifies the inductive
step. The first modification is straightforward and just extends further the
notion that the base case will vary depending on the result to be proved. We
have already allowed two different base cases: n = 0 if the universe is N and
n = 1 if the universe is Z+ . We can readily extend this as follows.

Method of Proof by Mathematical Induction:

Varying the Base Case
Let P (n) be a propositional function with universe
Z≥m = {n ∈ Z : n ≥ m}.

Base case Prove that P (m) is true.

Inductive step Prove that, for all integers k ≥ m, P (k) ⇒ P (k + 1).
To do this:
assume P (k); this is the inductive hypothesis
then deduce P (k + 1).
Conclusion:
P (n) is true for all integers n ≥ m; that is, ∀n ∈ Z≥m • P (n) is true.
 317

Example 8.2
Since this modification is straightforward, we illustrate it with a single exam-
ple.

Theorem 8.4
For all integers n ≥ 5, n4 < 4n .

The following table compares the values of n4 and 4n for small positive integers
n. Although the initial values give no clear pattern, the table suggests that,
for n ≥ 5, the values of 4n increase more rapidly than do the values of n4 .

n n4 4n
1 1 4
2 16 16
3 81 64
4 256 256
5 625 1024
6 1296 4096

Thinking about the proof, the table takes care of the base case which is, of
course, n = 5. For the inductive step, we need to show that, for all k ≥ 5,
if k 4 < 4k , then (k + 1)4 < 4k+1 . So, let’s consider (k + 1)4 . Expanding the
expression, we have

(k + 1)4 = k 4 + 4k 3 + 6k 2 + 4k + 1.

Suppose we could show that the terms on the right-hand side were no greater
than k 4 + k 4 + k 4 + k 4 = 4 × k 4 . Then the inductive step would allow us to
reason: 4 × k 4 < 4 × 4k = 4k+1 . Now, considering the terms in the expression
k 4 + 4k 3 + 6k 2 + 4k + 1, we have:

4k 3 < k 4 since 4 < k

2
6k < k 4
since 6 < k 2 as k ≥ 5
4k + 1 < 4k + k = 5k < k 4 since 5 < k 3 as k ≥ 5.

From these ‘pieces’ we can deduce that k 4 + 4k 3 + 6k 2 + 4k + 1 < 4k 4 , which

then allows us to complete the inductive step. We can now marshal these
reasoning fragments into a coherent proof as follows.

Proof. The proof is by mathematical induction.

Base case. When n = 5, 54 = 625 and 45 = 1024, so n4 < 4n in this case.

Hence the result holds when n = 5.
318

Inductive step. Assume that, for all integers k ≥ 5, k 4 < 4k .

Now (k + 1)4 = k 4 + 4k 3 + 6k 2 + (4k + 1) and we consider the terms on the
right-hand side in turn. Since k ≥ 5, we have
4k 3 < k × k 3 = k 4
k 2 ≥ 25 so 6k 2 < 25k 2 ≤ k 2 × k 2 = k 4
k 3 ≥ 125 so 4k + 1 < 4k + k = 5k < 125k ≤ k 3 × k = k 4 .

Therefore (k + 1)4 = k 4 + 4k 3 + 6k 2 + (4k + 1)

< k4 + k4 + k4 + k4
= 4k 4
< 4 × 4k by the inductive hypothesis
k+1
= 4 .

This shows that (k + 1)4 < 4k+1 , which is the result for k + 1 and hence
completes the inductive step.

Therefore, for all integers n ≥ 5, we have n4 < 4n , by induction.

To motivate our second, more substantial, modification of the method of proof

by induction, we look again at the Prime Factorisation Theorem (theorem 4.3):
every integer greater than 1 can be expressed as a product of prime numbers.
Let P (n) be the propositional function:
P (n) : n can be expressed as a product of prime numbers.
Then the Prime Factorisation Theorem may be symbolised as ∀n•P (n) where
the universe is Z≥2 = {n ∈ Z : n ≥ 2}. Hence this is a prime (excuse the pun)
candidate for a proof by induction with base case n = 2. The base case itself
is clear: 2 is already expressed as a product of primes, albeit in a rather trivial
way.
Let’s consider the inductive step. The inductive hypothesis is: assume that
k ≥ 2 may be expressed as
k = p1 p2 . . . pm where p1 , p2 , . . . , pm are prime numbers.
Now consider k + 1; we need to show that k + 1 can also be expressed as a
product of primes. However, the factorisation of k does not give any infor-
mation about the factorisation of k + 1. For example, 15 has factorisation
3 × 5 but 16 has factorisation 2 × 2 × 2 × 2; or 17 is itself prime but 18 has
factorisation 2 × 3 × 3. However, there are two possibilities for k + 1: either it
is prime or it is composite. If it is prime, then there is nothing to prove; like
2, it is trivially expressed as a product of primes. If k + 1 is composite, then
it has a factorisation
k + 1 = a1 a2 where 2 ≤ a1 , a2 ≤ k.
 319

The inductive hypothesis only gives us information about k and not a1 and a2 .
In our previous, somewhat informal proof (page 134), we essentially applied
the same reasoning to a1 and a2 : each is either prime or it factorises. What
is needed is a ‘mathematical induction’ version of this. Suppose that we had
an inductive hypothesis that allowed us to assume P (a1 ) and P (a2 ) as well
as P (k). Then we may express each of a1 and a2 as a product of primes and
hence we can also express k + 1 = a1 a2 as a product of primes. An inductive
hypothesis that allows us to assume P (r) for all r ≤ k (rather than just P (k)
itself) is what is needed in this case.
Returning to our informal introduction to the Principle of Mathematical In-
duction expressed in terms of writing down integers, suppose we are given the
following two instructions.

1. Write down the number 1.

2. If you write down all of the integers 1, 2, . . . , k, then you must also write
down the integer k + 1.

The effect is the same as following the original two instructions. Firstly, to obey
the first instruction, we need to write down ‘1’. Then the second instruction
‘kicks in’: because we have now written down ‘1’, we have (trivially) written
down all the integers from 1 to 1; then need to write down ‘2’. But now we
have written down both ‘1’ and ‘2’, so to follow the second instruction, we also
need to write down ‘3’. Then we have written ‘1’, ‘2’, and ‘3’ so, to follow the
second instruction, we need to write ‘4’ and so on. In other words, to follow
both instructions, we again would need to write down all the positive integers.
This is an informal justification for the following variation of the method of
proof by induction.

Method of Proof by Mathematical Induction: Strong Form

Let P (n) be a propositional function with universe N.

Base case Prove that P (0) is true.
Inductive step Prove that, for all integers k ∈ N,
(P (0) ∧ P (1) ∧ . . . ∧ P (k)) ⇒ P (k + 1).
To do this:
assume P (r) for all r ≤ k (and r ∈ N);
this is the inductive hypothesis
then deduce P (k + 1).
Conclusion:
P (n) is true for all integers n ∈ N; that is, ∀n ∈ N • P (n) is true.
320

We have stated this for the case where the universe is N to match our original
statement of proof by induction but, of course, we may vary the base case
appropriately to cover cases where the universe is Z+ or Z≥m . The method
is called the ‘strong form’ of mathematical induction because it allows us to
make a stronger inductive hypothesis. In other words, we assume more than
in the standard version of mathematical induction: instead of assuming the
result for a single values k, we assume it for all values up to and including k.
However, despite this stronger inductive hypothesis, the strong form is actually
equivalent to the original method of proof by induction given in the previous
section in the sense that anything that can be proved using the strong form can
also be proved using the original version. Suppose that P (n) is a propositional
function with universe N. Let Q(n) be the propositional function

Q(n) : P (0) ∧ P (1) ∧ . . . ∧ P (n).

So Q(n) asserts that P (r) is true for all 0 ≤ r ≤ n. Then two things are
apparent:

• the propositions ∀n • P (n) and ∀n • Q(n) are logically equivalent: one is

true if and only if the other is true; and
• the steps in the strong form of mathematical induction applied to P (n)
are identical to the steps in the original form of mathematical induction
applied to Q(n).

This demonstrates that any statement that can be proved using the strong
form can actually be proved using the original form. Thus the strong form
is a ‘methodological convenience’: it makes writing proofs easier, but is no
stronger, logically, than the original form.

Examples 8.3
1. For our first example, we give an inductive proof of the Prime Factori-
sation Theorem.

Theorem 4.3 (Prime Factorisation Theorem).

Every integer greater than 1 can be expressed as a product of prime
numbers.

Proof. The proof is by mathematical induction, strong form.

Base case. First note that n = 2 is trivially expressed as a product of
primes since it is itself prime.
Inductive step. Let k ≥ 2, and suppose that every integer r such that
2 ≤ r ≤ k can be expressed as a product of prime numbers.
 321

Consider k+1. Either k+1 is prime or it is composite. If it is prime, then

there is nothing to do as it is expressed as a product of prime numbers.
If k + 1 is composite, then it has a factorisation

k + 1 = a1 a2 where 2 ≤ a1 , a2 ≤ k.

By the inductive hypothesis, each of a1 and a2 may be expressed as a

product of primes:

a1 = p1 p2 . . . ps and a2 = q1 q2 . . . qt

where p1 , p2 , . . . , ps and q1 , q2 , . . . , qt are all prime numbers. Therefore

k + 1 can be expressed as a product of prime numbers,

k + 1 = p1 p2 . . . ps q1 q2 . . . qt .

This completes the inductive step.

Therefore, every integer greater than 1 can be expressed as a product of

prime numbers, by induction.

2. Our second example is more geometric and concerns n-sided polygons.

For n = 3, 4, 5, 6, and so on, these are, of course, just triangles, rectan-
gles, pentagons, hexagons and so forth. Such a polygon is called convex
if, for any two points of the polygon, the line segment joining those
points is entirely contained within the polygon. This is illustrated in fig-
ure 8.1. Polygon A is convex: the line segment joining any two points of
the polygon is entirely contained in the polygon. This is illustrated for
two ‘arbitrary’ points in the figure. Polygon B is not convex: for the two
points shown, the line segment joining them passes outside the polygon
itself.

A B

FIGURE 8.1: Convex and non-convex polygons.

The following theorem was part of the ‘background knowledge’ that we

needed for our non-constructive existence proof in example 7.2.2.
322

Theorem 8.5
The sum of the interior angles in any n-sided convex polygon is (n−2)π.

We are measuring angles in radians here. Were we to measure angles in

degrees, the statement of the theorem would be ‘the sum of the interior
angles in any n-sided convex polygon is 180(n − 2) degrees’.
The proof will be by induction on the number of sides of the polygon.
The base case will therefore be n = 3 because a triangle is the polygon
with the fewest possible number of sides. In fact, it is implicit in the
statement of the theorem that n is an integer n ≥ 3. We will make the
strong inductive hypothesis that the result holds for all polygons with
r ≤ k sides. The idea behind the inductive step is very simple: consider
a (k + 1)-sided polygon and ‘cut it in two’ to produce two polygons each
with fewer sides; then apply the inductive hypothesis to the two smaller
polygons.

Proof. The proof is by mathematical induction, strong form.

Base case. The base case is n = 3 since every (convex) polygon must
have at least three sides. It is well known that the angle sum of any
triangle is π radians.2 Hence the base case is satisfied.
Inductive step. Suppose that k ≥ 3 and that for every r-sided convex
polygon where 3 ≤ r ≤ k, the angle sum is (r − 2)π.
Let P be an arbitrary (k +1)-sided polygon. Select any two non-adjacent
vertices of P and draw the line segment joining them. Cutting along the
line segment separates P into two convex polygons, Q and R, say. Of
the k + 1 sides belonging to P , suppose that the cut separates the sides
so that q of them belonging to Q and k + 1 − q of them belonging to
R. Since each of Q and R have an additional side (the cut itself), Q is
a (q + 1)-sided polygon and R is a (k + 2 − q)-sided polygon. This is
illustrated in figure 8.2 below.
Now, since there are at least two sides on each side of the cut, 2 ≤ q ≤
k − 1. Therefore 3 ≤ q + 1 ≤ k, so the inductive hypothesis applies to the
polygon Q. Also, multiplying 2 ≤ q ≤ k −1 by −1 gives 1−k ≤ −q ≤ −2
and then adding k + 2 gives 3 ≤ k + 2 − q ≤ k; hence the inductive
hypothesis also applies to the polygon R.
From the inductive hypothesis we may deduce

angle sum for Q = (q + 1 − 2)π = (q − 1)π

and angle sum for R = (k + 2 − q − 2)π = (k + q)π.
2 As we noted on page 167, this is a theorem of Euclidean geometry that is not true in

various non-Euclidean geometries. For the current proof, we will accept this result as part
of our background knowledge. For a proof, see exercise 8.2.10.
 323

q sides k + 1 – q sides

P Q R

k + 1 sides q + 1 sides k + 2 – q sides

FIGURE 8.2: Illustrating the inductive step.

Consider the angle sum for P . Apart from each interior angle where the
cut is made, each angle of P belongs either to Q or to R. For each of
the two angles split by the cut, the sum of the angles of the two ‘pieces’
in Q and R equals the original angle in P . Hence

angle sum for P = (angle sum for Q) + (angle sum for R)

= (q − 1)π + (k + q)π (from inductive hypothesis)
= (k − 1)π
= ((k + 1) − 2) π.

This completes the inductive step.

Therefore, the sum of the interior angles in any n-sided convex polygon
is (n − 2)π, by induction.

Recursively defined sequences

In section 5.4, we considered the convergence behaviour of sequences (an ). We

now consider a way of describing sequences, that is often used where sequences
are used to model real phenomena, and the role of mathematical induction
in proving results about sequences defined in this way. Many systems in the
real world that evolve over time can be described by measuring characteris-
tics of the system only at specific time points, for example, every second, day,
month, or year as appropriate. Such systems are called discrete dynamical
systems, and they may be modelled using a sequence (an ) where an repre-
sents the state of the system at the n th time point from the start. There are
many systems that may be modelled in this way; these include the size of a
population; the output of a factory or a whole economy; the temperature at a
particular location or of the whole globe; the atmospheric pressure, humidity,
wind speed at a particular place; and many others.
324

It is often the case that the state of the system at a particular time depends
only on its state at the time point or points immediately beforehand. For
example, a species whose growth is not constrained by environmental factors
may be such that its population at time n depends only on its population at
time n − 1. If an denotes the population at time n, then in this scenario, an
is a function of an−1 , an = f (an−1 ).
Another example is the famous Fibonacci sequence, which was introduced
to Western mathematics in 1202 by Leonardo of Pica, known as Fibonacci,
although this was known to Indian mathematics for centuries before this.
In attempting to understand the growth of a rabbit population, Fibonacci
proposed a model where the population at time n is the sum of the population
at times n − 1 and n − 2. In this case, if an denotes the rabbit population,
then Fibonacci’s model is expressed as an = an−1 + an−2 .
These are examples of a recursively defined sequence which is one where
each term in the sequence is defined in terms of the previous terms. We may
express this more formally by saying that the n th term is a function of (some
or all of) the terms a1 , a2 , . . . , an−1 ,

an = f (a1 , a2 , . . . , an−1 ) .

This expression is called the recursive relationship. The first term or the
first few terms will also need to be defined explicitly, and this will depend on
which terms appear in the recursive relationship.

Examples 8.4
1. If the recursive relationship defined an only in terms of an−1 , then it is
only the initial term a1 that also needs to be defined explicitly.
A sequence (an ) is defined by

an = 3an−1 + 2 for n ≥ 2,
a1 = 5.

The first few terms of the sequence are:

a1 = 5 (by definition)
a2 = 3a1 + 2 = 3 × 5 + 2 = 17
a3 = 3a2 + 2 = 3 × 17 + 2 = 53
a4 = 3a3 + 2 = 3 × 53 + 2 = 158
..
.

2. For the Fibonacci sequence, where the recursive relationship defines an

using both an−1 and an−2 , the first two terms are required explicitly.
 325

These are both taken as having the value 1, so the Fibonacci sequence
(an ) is defined by

an = an−1 + an−2 for n ≥ 3,

a1 = 1, a2 = 1.

The first few terms of the sequence are:

a1 = 1 (by definition)
a2 = 1 (by definition)
a3 = a2 + a1 = 1 + 1 = 2
a4 = a3 + a2 = 2 + 1 = 3
a5 = a4 + a3 = 3 + 2 = 5
a6 = a5 + a4 = 5 + 3 = 8
a7 = a6 + a5 = 8 + 5 = 13
..
.

3. In the previous two examples, the n th term is defined in terms of a fixed

number of previous terms — one or two in these examples. However, it
is possible to use a variable number of terms. For example, a sequence
(an ) is defined by

an = a1 + a2 + · · · a + an−1 for n ≥ 2,
a1 = 1.

In this case, the n th term is the sum of all of the previous terms, so the
first few terms of the sequence are:

a1 = 1 (by definition)
a2 = a1 = 1
a3 = a1 + a2 = 1 + 1 = 2
a4 = a1 + a2 + a3 = 1 + 1 + 2 = 4
a5 = a1 + a2 + a3 + a4 = 1 + 1 + 2 + 4 = 8
a6 = a1 + a2 + a3 + a4 + a5 = 1 + 1 + 2 + 4 + 8 = 16
..
.

In the previous examples, we have only considered the process of defining

sequences recursively. If we wish to prove properties of such sequences, then
a form of mathematical induction is an obvious tool. In particular, it is often
convenient to have an explicit formula for the n th term. For example, it would
require a lot of work to use the recursive definition to calculate the 1000 th
326

term of each of the sequences defined in examples 8.4. In some cases it might
be quite easy to conjecture a general formula; we suspect readers may ‘spot’ a
formula for an in example 3 above. In other cases, this might be considerably
more difficult; for the Fibonacci sequence, it would be almost impossible to
‘guess’ the following general formula
( √ )n ( √ )n
1 1+ 5 1 1− 5
an = √ −√ .
5 2 5 2
However a conjecture is obtained, mathematical induction will be an appro-
priate method of proof. For those cases where several initial terms are defined
explicitly, the result will need to be verified individually for these initial terms.
Thus there may be several ‘base cases’ that need verifying and the inductive
step will only apply where the recursive relationship is used. The general
method, based on the strong form of induction, is described as follows.

Method of Proof by Mathematical Induction:

Recursive Sequences
Let (an ) be a sequence that is defined recursively as follows.
a1 , . . . , am are defined explicitly
an = f (a1 , . . . , an−1 ) for n ≥ m + 1

Let P (an ) be a propositional function about the terms of the

sequence.
Base cases Prove that P (a1 ), . . . , P (am ) are true.
Inductive step Prove that, for all integers k ≥ m + 1,
(P (a1 ) ∧ . . . ∧ P (ak )) ⇒ P (ak+1 ).
To do this:
assume P (ar ) for all r ≤ k (and r ∈ Z+ );
this is the inductive hypothesis
then deduce P (ak+1 ).
Conclusion:
P (an ) is true for all integers n ∈ Z+ ; that is, ∀n ∈ Z+ • P (an ) is
true.

Examples 8.5
1. Let (an ) be the sequence in example 8.4.1 defined by
an = 3an−1 + 2 for n ≥ 2,
a1 = 5.
 327

Then an = 2 × 3n − 1 for all n ∈ Z+ .

Proof. The proof is by mathematical induction.

Base case. When n = 1, a1 = 5 (by definition) and 2 × 3n − 1 =

2 × 31 − 1 = 6 − 1 = 5. Hence the result is true for n = 1.

Inductive step. Let k ∈ Z+ . Suppose that, for every positive integer

r ≤ k, ak = 2 × 3k − 1.
Consider ak+1 . Since k + 1 ≥ 2, ak+1 is defined by the recursive rela-
tionship. So we have

ak+1 = 3ak + 2 (definition of ak+1 )

( )
= 3 2 × 3k − 1 + 2 (inductive hypothesis)
= 2 × 3 × 3k − 3 + 2
= 2 × 3k+1 − 1.

Hence ak+1 satisfies the given formula. This completes the inductive
step.

Therefore, an = 2 × 3n − 1 for all n ∈ Z+ , by induction.

Commentary
Alert readers will have noticed that, in this example, we have not ac-
tually used the strong form of the inductive hypothesis. The only place
where the inductive hypothesis is used is to replace ak with 2 × 3k − 1.
In this example, a proof using the ‘standard’ version of mathematical
induction would be perfectly valid. Of course, the reason for this is that
the recursive relationship is of the form an = f (an−1 ); in other words,
an is defined only in terms of its immediate predecessor an−1 .

2. Let (an ) be the sequence defined by

an = 3an−1 − 2an−2 for n ≥ 3,

a1 = 0, a1 = 2.

Then an = 2n−1 − 1 for all n ≥ 2.

Proof. The proof is by mathematical induction.

Base cases. Since an is defined explicitly for n = 1 and n = 2, there are

two base cases to consider.
When n = 1, a1 = 0 (by definition) and 2n−1 − 1 = 20 − 1 = 1 − 1 = 0.
When n = 2, a2 = 1 (by definition) and 2n−1 − 1 = 21 − 1 = 2 − 1 = 1.
328

Hence, the result is true for n = 1 and n = 2.

Inductive step. Let k ≥ 2. Suppose that, for every integer 2 ≤ r ≤ k,
ak = 2k−1 − 1.
Consider ak+1 . Since k + 1 ≥ 3, ak+1 is defined by the recursive rela-
tionship. So we have

ak+1 = 3ak − 2ak−1 (definition of ak+1 )

= 3(2 k−1
− 1) − 2(2 k−2
− 1) (inductive hypothesis)
= 3×2 k−1
−3−2×2 k−2
+2
= 3×2 k−1
−2 k−1
−1
= 2×2 k−1
−1
= 2 − 1.
k

Hence ak+1 satisfies the given formula. This completes the inductive
step.
Therefore, an = 2n−1 − 1 for all n ≥ 2, by induction.

Commentary
In contrast to the previous example, the strong form of the inductive
hypothesis is used in this proof. In the step labelled ‘inductive hypothe-
sis’ we have replaced ak−1 with 2k−1 − 1 and we have replaced ak−2 with
2k−2 − 1. In other words, we used the inductive hypothesis for r = k − 1
and r = k − 2. This means that a simple inductive hypothesis would not
have been sufficient in this case.

Exercises 8.2
1. Prove each of the following.
(i) For all integers n ≥ 4, n! > 2n .
(ii) For all integers n ≥ 7, n! > 3n .
(iii) For all integers n ≥ 5, n2 < 2n .
(iv) For all integers n ≥ 4, n3 < 3n .
(v) For all integers n ≥ 10, n3 < 2n .
2. Prove by induction that, for all integers n ≥ 2,
3 8 15 n2 − 1 n+1
× × × ··· × = .
4 9 16 n2 2n
 329

3. Using the strong form of mathematical induction, prove that every in-
teger n ≥ 6 can be written as n = 3a + 4b for some a, b ∈ N.
Proceed as follows.

• Prove the three base cases n = 6, n = 7, n = 8 directly.

• The inductive step applies for k ≥ 8; the strong inductive hypoth-
esis is that every integer r such that 6 ≤ r ≤ k can be expressed as
r = 3a + 4b for some a, b ∈ N.
• To complete the inductive step, write k + 1 = (k − 2) + 3 and use
the inductive hypothesis.

4. Prove each of the following about recursively defined sequences.

(i) Let (an ) be the sequence defined by

an = an−1 + 3 for n ≥ 2,
a1 = 3.

Then an = 3n for all n ∈ Z+ .

(ii) Let (an ) be the sequence defined by

an = 3 − an−1 for n ≥ 2,
a1 = 1.

3 + (−1)n
Then an = for all n ∈ Z+ .
2
(iii) Let (an ) be the sequence defined by
an = 5an−1 − 6an−2 for n ≥ 3,
a1 = 0, a2 = 6.

Then an = 2 × 3n − 3 × 2n for all n ∈ Z+ .

(iv) Let (an ) be the sequence defined by

an = a1 + a2 + · · · + an−1 for n ≥ 2,
a1 = 1.

Then an = 2n−2 for all n ≥ 2.

(v) Let (an ) be the sequence defined by

an = 2an−1 + an−2 − 2an−3 for n ≥ 4,
a1 = −3, a2 = 1, a3 = 3.

Then an = 2n + (−1)n − 4 for all n ∈ Z+ .

330

(vi) Let (an ) be the sequence defined by

an = 2an−1 + an−2 + an−3 for n ≥ 4,
a1 = 1, a2 = 2, a3 = 3.

Then an < 2n for all n ∈ Z+ .

5. Let (an ) be the Fibonacci sequence defined in example 8.4.2.

Prove each of the following.

(i) an < 2n for all n ∈ Z+ .

(ii) a1 + a2 + · · · + an = an+2 − 1 for all n ∈ Z+ .
(iii) a2n+2 − a2n+1 = an an+3 for all n ∈ Z+ .
(iv) a1 + a3 + · · · + a2n−1 = a2n for all n ∈ Z+ .
( √ )n ( √ )n
1 1+ 5 1 1− 5
(v) an = √ −√ for all n ∈ Z+ .
5 2 5 2

Hint: it will make the algebraic manipulation simpler if you first

establish that
( √ )2 √
1± 5 3± 5
= .
2 2

6. It is well known that positive integers can be written using binary nu-
merals. For example, 13 = 11012 because 13 = 23 + 22 + 20 . That every
n ∈ Z+ has a binary numeral representation is equivalent to the state-
ment: every n ∈ Z+ can be expressed as a sum of distinct non-negative
powers of 2.
Use the strong form of induction to prove this statement.

7. The definition of ‘convex’ for a polygon is given in example 8.3.2. The

definition applies to arbitrary subsets of the plane R2 or 3-dimensional
space R3 : a subset S of R2 or R3 is convex if, for all x, y ∈ S, the line
segment joining x and y is entirely contained in S. For convenience, we
regard the empty set ∅ and any singleton set {x} as being convex.
Prove that for all n ≥ 2, if A1 , A2 , . . . , An are convex subsets of R3 , then
their intersection A1 ∩ A2 ∩ . . . ∩ An is also a convex subset of R3 .

8. Let A = {1, 5, 9, 13, . . .} = {4n + 1 : n ∈ N} be the set of all positive

integers whose remainder on division by 4 is 1.

(i) Prove that A is closed under multiplication; that is, the product of
any two elements of A is always an element of A.
 331

Define an A-prime to be an element p ∈ A such that p > 1 and the

only factors of p which are elements of A are 1 and p.
For example, 9 is an A-prime, since the only divisors of 9 that are ele-
ments of A are 1 and 9.

(ii) Prove by induction on n that every element of A that is greater

than 1 can be written as a product of A-primes.
(iii) Show that A-prime factorisation of the elements of A is not neces-
sarily unique; in other words, it is possible for an element of A to
be written as a product of A-primes in at least two different ways.

9. By induction on |A|, prove that, for all finite sets A and B, if B ⊂ A,

then |B| < |A|.

10. Prove the base case in the proof of theorem 8.5: the angle sum in any
triangle is π.
The following diagram provides the idea behind the proof.

A C B

A B

11. In the proof of theorem 8.5, where is the requirement that the polygon
be convex used?
Is the theorem still true without the requirement that the polygon be
convex? If so, how might the proof given in example 8.3.2 be modified
to prove the more general theorem?

12. Let P (n) be a propositional function with universe Z+ . Suppose that

the following are true propositions.

(I) P (1);
(II) for all n ∈ Z+ , P (n) ⇒ P (2n);
(III) for all n ∈ Z+ , P (n + 1) ⇒ P (n).

Prove that P (n) is true, for all n ∈ Z+ .

Hints and Solutions to Selected
Exercises

Chapter 2

Exercises 2.1

1. (i) P Q ¬Q P ∨ ¬Q ¬(P ∨ ¬Q)

T T F T F
T F T T F
F T F F T
F F T T F
(v) P Q R P ⇒R R⇒Q (P ⇒ R) ∧ (R ⇒ Q)
T T T T T T
T T F F T F
T F T T F F
T F F F T F
F T T T T T
F T F T T T
F F T T F F
F F F T T T

2. (i) If the wind blows then, the sun doesn’t shine or the rain falls (or
both).
(ii) The wind blows and the rain falls if and only if the sun doesn’t
shine.
(v) If the rain doesn’t fall or the temperature rises (but not both), then
the sun shines and the wind doesn’t blow.

3. (ii) (S ∧ ¬W ) ⇒ T
(iv) (S ∧ T ) ∨ (W ∧ R)

333
334 Understanding Mathematical Proof

4. (iv) P R P ∧Q P ∨ Q ¬(P ∨ Q) (P ∧ Q) ∧ ¬(P ∨ Q)

T T T T F F
T F F T F F
F T F T F F
F F F F T F

Since all its truth values are F, (P ∧Q)∧¬(P ∨Q) is a contradiction.

(v) P Q R P ⇒Q Q⇒R (P ⇒ Q) ∨ (Q ⇒ R)
T T T T T T
T T F T F T
T F T F T T
T F F F T T
F T T T T T
F T F T F T
F F T T T T
F F F T T T

Since all its truth values are T, (P ⇒ Q) ∨ (Q ⇒ R) is a tautology.

5. (i) (a), (b), (d), and (f) are the substitution instances.
For example, the substitution that gives the proposition in (f) is
P −→ (R ⇒ S) and Q −→ S.

(iii) (d) is a substitution instance of question 4 (v) and so is a tautology.

Exercises 2.2

1. (iii) P Q ¬Q P ⇒ ¬Q P ∧Q ¬(P ∧ Q)
T T F F T F
T F T T F T
F T F T F T
F F T T F T
Since the truth values in the fourth and sixth columns are equal,
P ⇒ ¬Q ≡ ¬(P ∧ Q).
 Hints and Solutions to Selected Exercises 335

(iv) A B
P Q R P ∧Q (P ∧ Q) ∨ R P ∨R Q∨R A∧B
T T T T T T T T
T T F T T T T T
T F T F T T T T
T F F F F T F F
F T T F T T T T
F T F F F F T F
F F T F T T T T
F F F F F F F F

Since the truth values in the fifth and eighth columns are equal,
(P ∧ Q) ∨ R ≡ (P ∨ R) ∧ (Q ∨ R).

(vi) P Q R Q⇒R (P ⇒ (Q ⇒ R) P ∧ Q (P ∧ Q) ⇒ R
T T T T T T T
T T F F F T F
T F T T T F T
T F F T T F T
F T T T T F T
F T F F T F T
F F T T T F T
F F F T T F T

Since the truth values in the fifth and seventh columns are equal,
P ⇒ (Q ⇒ R) ≡ (P ∧ Q) ⇒ R.

2. (ii) P Q P ⇔Q (P ⇔ Q) ∧ Q
T T T T
T F F F
F T F F
F F T F

The only row where (P ⇔ Q) ∧ Q is true is row 1 and P is also

true in row 1.

Hence (P ⇔ Q) ∧ Q  P .
336 Understanding Mathematical Proof

(iv) P Q R Q∨R P ⇒ (Q ∨ R) P ∧ R P ∨R
T T T T T T T
T T F T T F T
T F T T T T T
T F F F F F T
F T T T T F T
F T F T T F F
F F T T T F T
F F F F T F F
The propositions P ⇒ (Q ∨ R) and P ∧ R are both true only in
rows 1 and 3. In each of these rows, P ∨ R is also true.
Therefore {P ⇒ (Q ∨ R), P ∧ R}  (P ∨ R).
(vii) P Q R ¬P P ⇒R Q⇒R ¬P ⇒ Q ¬P ∨ R
T T T F T T T T
T T F F F F T F
T F T F T T T T
T F F F F T T F
F T T T T T T T
F T F T T F T T
F F T T T T F T
F F F T T T F T
The propositions P ⇒ R, Q ⇒ R and ¬P ⇒ Q are all true only in
rows 1, 3, and 5. In each of these rows, ¬P ∨ R is also true.
Therefore {P ⇒ R, Q ⇒ R, ¬P ⇒ Q}  (¬P ∨ R).

3. (ii) ¬(P ∨ Q) ∨ (¬P ∧ Q)

≡ (¬P ∧ ¬Q) ∨ (¬P ∧ Q) De Morgan’s law
≡ ¬P ∧ (¬Q ∨ Q) Distributive law
≡ ¬P ∧ true Complement law
≡ ¬P Identity law

(iv) P ∨ (Q ∧ (P ∨ ¬Q))
≡ P ∨ ((Q ∧ P ) ∨ (Q ∧ ¬Q)) Distributive law
≡ P ∨ ((Q ∧ P ) ∨ false) Complement law
≡ P ∨ (Q ∧ P ) Identity law
≡ P ∨ (P ∧ Q) Commutative law
≡ P Absorption law

(vi) (P ∧ Q) ∨ (¬P ∨ R)
 Hints and Solutions to Selected Exercises 337

≡ ((P ∧ Q) ∨ ¬P ) ∨ R) Associative law

≡ (¬P ∨ (P ∧ Q)) ∨ R Commutative law
≡ ((¬P ∨ P ) ∧ (¬P ∨ Q)) ∨ R Distributive law
≡ (true ∧ (¬P ∨ Q)) ∨ R Complement law
≡ (¬P ∨ Q) ∨ R Identity law
(viii) [(P ∧ Q) ∨ (R ∧ S)]
≡ ((P ∧ Q) ∨ R) ∧ ((P ∧ Q) ∨ S) Distributive law
≡ (R ∨ (P ∧ Q)) ∧ (S ∨ (P ∧ Q)) Commutative law
≡ (R ∨ P ) ∧ (R ∨ Q) ∧ (S ∨ P ) ∧ (S ∨ Q) Distributive law
≡ (P ∨ R) ∧ (Q ∨ R) ∧ (P ∨ S) ∧ (Q ∨ S) Commutative law
≡ (P ∨ R) ∧ (P ∨ S) ∧ (Q ∨ R) ∧ (Q ∨ S) Commutative law
4. (ii) P ⇒ (P ∧ Q)
≡ ¬P ∨ (P ∧ Q) Material Implication law
≡ (¬P ∨ P ) ∧ (¬P ∨ Q) Distributive law
≡ true ∧ (¬P ∨ Q) Complement law
≡ ¬P ∨ Q Identity law
≡ P ⇒Q Material Implication law
(iv) (P ∨ Q) ⇒ R
≡ ¬(P ∨ Q) ∨ R Material Implication law
≡ (¬P ∧ ¬Q) ∨ R De Morgan’s law
≡ R ∨ (¬P ∧ ¬Q) Commutative law
≡ (R ∨ ¬P ) ∧ (R ∨ ¬Q) Distributive law
≡ (¬P ∨ R) ∧ (¬Q ∨ R) Commutative law
≡ (P ⇒ R) ∧ (Q ⇒ R) Material Implication law

Exercises 2.3
1. (ii) Universe: Footballers
Predicates: O : . . . is overpaid
T : . . . is talented.
Then symbolise as: ∃x • O(x) ∧ ¬T (x).

(iv) Universe: People (a specific group of people)

Predicates: F : . . . shouted ‘Fire’ !
P : . . . panicked.
Then symbolise as: ∃x • F (x) ∧ ∀x • P (x).
338 Understanding Mathematical Proof

(vii) We interpret this as: no-one likes all people who are rude.
Universe: People
Predicates: L : . . . likes . . .
R : . . . is rude.
Then symbolise as: ¬∃x ∀y • R(y) ⇒ L(x, y).

(xi) Universe: People who went to the races

Predicates: C : . . . was cold
P : . . . lost money.
Then symbolise as: ¬∃x • C(x) ∧ ∃x • L(x).

2. (ii) If Pete is clever and honest, then he’s likeable.

(iv) Some clever people are dishonest.
(vi) Not everyone is likeable and honest.
(viii) No one is likeable and dishonest.

3. (iii) ∃x • T (x, Statistics) ∧ ¬P (x, Statistics)

(v) ∃x ∀y • T (x, y) ⇒ P (x, y)
(vi) P (Gemma, Statistics) ⇒ (∀x • T (x, Statistics) ⇒ P (x, Statistics))

4. (iii) (a) Poppy enjoys every course she takes.

(b) ∃y • T (Poppy, y) ∧ ¬E(Poppy, y)
(c) Poppy take some courses that she doesn’t enjoy.

(v) (a) Everyone passes some course.

(b) ∃x ∀y • ¬P (x, y)
(c) Someone fails every course.

(vii) (a) Someone passes every course they take.

(b) ∀x ∃y • T (x, y) ∧ ¬P (x, y)
(c) Everyone fails some course that they take.

Exercises 2.4

1. (ii) 1. P ⇒ ¬Q premise
2. Q∨R premise
3. ¬¬Q ∨ R 2. Equivalence: involution law
4. ¬Q ⇒ R 3. Equivalence: material implication law
5. P ⇒R 1,5. Hypothetical syllogism
 Hints and Solutions to Selected Exercises 339

(iv) 1. (P ∧ Q) ⇒ (S ∧ T ) premise
2. Q∧P premise
3. P ∧Q 2. Equivalence: commutative law
4. S∧T 1,3. Modus ponens
5. S 4. Simplification

(vi) 1. Q premise
2. ¬S premise
3. (P ∧ R) ⇒ S premise
4. ¬(P ∧ R) 2,3. Modus tollens
5. ¬P ∨ ¬R 4. Equivalence: commutative law
6. Q∨P 1. Addition
7. P ∨Q 6. Equivalence: De Morgan’s law
8. Q ∨ ¬R 5,7. Resolution

(viii) 1. P ∨Q premise
2. R ⇒ ¬Q premise
3. ¬P premise
4. (¬R ∧ Q) ⇒ S premise
5. Q 1,3. Disjunctive syllogism
6. ¬(¬Q) 5. Equivalence: involution law
7. ¬R 2,6. Modus tollens
8. ¬R ∧ Q 5,7. Conjunction
9. S 4,8. Modus ponens

(x) 1. (P ⇒ Q) ∧ (R ⇒ Q) premise
2. S ⇒ (P ∨ R) premise
3. S premise
4. P ∨R 2,3. Modus ponens
5. P ⇒Q 1. Simplification
6. (R ⇒ Q) ∧ (P ⇒ Q) 1. Equivalence: commutative law
7. R⇒Q 6. Simplification
8. Q∨Q 4,5,7. Constructive dilemma
9. Q 8. Equivalence: idempotent law

2. (i) 1. P ⇒ Q premise
2. Q ⇒ P premise
3. (P ⇒ Q) ∧ (Q ⇒ P ) 1,2. Conjunction
4. P ⇔ Q 3. Equivalence: biconditional law
340 Understanding Mathematical Proof

(iii) 1. P ∨Q premise
2. P ⇒R premise
3. Q⇒R premise
4. R∨R 1,2,3. Constructive dilemma
5. R 3. Equivalence: idempotent law

3. (iii) Let B: the battery is flat

P : the car is out of petrol
S: the car won’t start
L: I’ll be late for work.
The argument has premises: (B ∨ P ) ⇒ (S ∧ L) and P ∨ B.
The conclusion is L.
1. (B ∨ P ) ⇒ (S ∧ L) premise
2. P ∨B premise
3. B∨P 2. Equivalence: commutative law
4. S∧L 1,3. Modus ponens
5. L∧S 4. Equivalence: commutative law
6. L 5. Simplification

(vi) Let V : Peter is brave

B: Peter is brainy
A: Peter is bald.
The argument has premises: (V ∨ B) ∧ (B ∨ A) and ¬B.
The conclusion is V ∧ A.
1. (V ∨ B) ∧ (B ∨ A) premise
2. ¬B premise
3. V ∨B 1. Simplification
4. B∨V 3. Equivalence: commutative law
5. V 2,4. Disjunctive syllogism
6. (B ∨ A) ∧ (V ∨ B) 1. Equivalence: commutative law
7. B∨A 6. Simplification
8. A 2,7. Disjunctive syllogism
9. V ∧A 5,8. Conjunction

(ix) Let G: ghosts are a reality

S: there are spirits roaming the Earth
F : we fear the dark
I: we have an imagination.
The argument has premises: (G ⇒ S) ∧ (¬G ⇒ ¬F ), F ∨ ¬I and
I ∧ G.
The conclusion is S.
 Hints and Solutions to Selected Exercises 341

1. (G ⇒ S) ∧ (¬G ⇒ ¬F ) premise
2. F ∨ ¬I premise
3. I ∧G premise
4. I 3. Simplification
5. ¬(¬I) 4. Equivalence: involution law
6. ¬I ∨ F 2. Equivalence: commutative law
7. F 5,6. Disjunctive syllogism
8. (¬G ⇒ ¬F ) ∧ (G ⇒ S) 1. Equivalence: commutative law
9. ¬G ⇒ ¬F 8. Simplification
10. ¬(¬F ) 7. Equivalence: involution law
11. ¬(¬G) 9,10. modus tollens
12. G 7. Equivalence: involution law
13. G⇒S 1. Simplification
14. S 12,13. Modus ponens

4. (ii) Let P: the project was a success

I: Sally invested her inheritance
S: Sally is sensible
B: Sally is broke.
The argument has premises: ¬P ∨ I, S ⇒ I, P and (¬S ∧ ¬I) ⇒ B.
The conclusion is B.
Suppose that: P is true, I is true, S is true, and B is false.
Then the premises are all true and the conclusion is false. Therefore
the argument is not valid.
(iv) With the notation in question 3 (ix) above, the argument has
premises: (G ⇒ S) ∨ (¬G ⇒ ¬F ), F ∨ ¬I and I ∨ G.
The conclusion is S.
Suppose that G is true, S is false, F is true, and I is true.
Then the premises are all true and the conclusion is false. Therefore
the argument is not valid.

5. (iii) Define the universe to be ‘numbers’ and symbolise the predicates

as follows.
Let E: . . . is even
R: . . . is rational
T: . . . is divisible by two
F: . . . is divisible by four.
Then the argument has premises: ∀x • E(x) ⇒ (R(x) ∧ T (x)) and
∃x • E(x) ∧ F (x). The conclusion is ∃x • T (x) ∧ F (x).
342 Understanding Mathematical Proof

1. ∀x • E(x) ⇒ (R(x) ∧ T (x)) premise

2. ∃x • E(x) ∧ F (x) premise
3. E(a) ⇒ (R(a) ∧ T (a)) 1. ∀-elimination
4. E(a) ∧ F (a) 2. ∃-elimination
5. E(a) 4. Simplification
6. R(a) ∧ T (a) 3, 5. Modus ponens
7. T (a) ∧ R(a) 6. Equivalence: commutative law
8. T (a) 7. Simplification
9. F (a) ∧ E(a) 4. Equivalence: commutative law
10. F (a) 9. Simplification
11 T (a) ∧ F (a) 8, 10. Conjunction
12. ∃x • T (x) ∧ F (x) 11. ∃-introduction
(viii) Define the universe to be ‘functions’ and symbolise the predicates
as follows.
Let P : . . . is a polynomial
D: . . . is differentiable
C: . . . is continuous.
Then the argument has premises: ¬∃x • P (x) ∧ ¬D(x) and ∀x •
D(x) ⇒ C(x). The conclusion is ∀x • P (x) ⇒ C(x).
1. ¬∃x • P (x) ∧ ¬D(x) premise
2. ∀x • D(x) ⇒ C(x) premise
3. ∀x • ¬(P (x) ∧ ¬D(x)) 1. Equivalence: negating
quantified propositions
4. D(a) ⇒ C(a) 2. ∀-elimination
5. ¬(P (a) ∧ ¬D(a)) 3. ∀-elimination
6. ¬P (a) ∨ ¬(¬D(a)) 5. Equivalence: De Morgan’s law
7. ¬P (a) ∨ D(a) 6. Equivalence: involution law
8. P (a) ⇒ D(a) 7. Equivalence: material implication
9. P (a) ⇒ C(a) 4, 8. Hypothetical syllogism
10. ∀x • P (x) ⇒ C(x) 9. ∀-introduction

Chapter 3
Exercises 3.1

1. (ii) {3, 6, 9, 12, . . .}

{ }
(iv) 13 , −2
 Hints and Solutions to Selected Exercises 343

(vi) {−2}
{ }
(viii) 12 , 1, 32 , 2, 52 , 3, . . .

2. (ii) {m : m = 2n for some n ∈ Z and 1 ≤ n ≤ 50} or, more simply,

{2n : n ∈ Z and 1 ≤ n ≤ 50}
(vi) {m : m = 5n + 2 for some n ∈ N} or, more simply, {5n + 2 : n ∈ N}
(ix) {n ∈ Z+ : n = 2m for some m ∈ Z}

3. (iii) True (iv) False (vi) False (vii) True.

4. (ii) |{x : x is an integer and 2/3 < x < 17/3}| = |{1, 2, 3, 4, 5}| = 5
(v) |{x ∈ R : x2 ≤ 2}| = ∞
(vii) |{2, 4, {{6, 8}, 10}}| = 3
(ix) |{1, {1}, {{1}}, {{{1}}}}| = 4

5. (ii) x ∈ A (iv) x ∈ A and x ⊆ A (vi) neither.

6. (ii) {{a, b}, {a, c}, {a, d}, {b, c}, {b, d}, {c, d}}
(iv) {{b}, {a, b}, {b, c}, {b, d}, {a, b, c}, {a, b, d}, {b, c, d}, {a, b, c, d}}

7. (ii) Both (iv) B ⊆ A (vi) neither.

Exercises 3.2

A B U A B U

1. (iii) (v)

C C

2. (i) {2} (iii) {4, 6, 8, 10}

(v) {3, 5, 7} (vii) {3, 5, 7}

3. (iii) None.
(vi) X ⊆ Y
(ix) Y ⊆ X and X ∩ Y = ∅ (since Y = ∅).

4. (i) Proof. Let A, B, and C be sets such that C ⊆ A and C ⊆ B.

Let x ∈ C. Then x ∈ A (since C ⊆ A) and x ∈ B (since C ⊆ B).
Therefore x ∈ A ∩ B.
We have shown that x ∈ C ⇒ x ∈ A ∩ B, which means that
C ⊆ A ∩ B.
344 Understanding Mathematical Proof

5. (i) (a) A ∗ B = {2, 6, 10, 12, 16}

(c) A ∗ (B ∩ C) = {2, 6, 8, 10, 16}
(e) (A ∗ B) ∩ (A ∗ C) = {2, 6, 10, 16}
(ii) (c) (A ∗ B) ∩ (A ∗ C)
A B A B A B

C C C

A* B A*C ( A * B) Ç ( A * C )

6. (iii) {∅, {{1}}, {{2, 3}}, {{1}, {2, 3}}}

(iv) {∅, {{1, 2, 3}}}

7. P(A) = {∅, {a}, {b}, {a, b}}.

P(B) = {∅, {b}, {c}, {d}, {b, c}, {b, d}, {c, d}, {b, c, d}}.
P(A ∪ B) = P({a, b, c, d})
= {∅, {a}, {b}, {c}, {d}, {a, b}, {a, c}, {a, d}, {b, c}, {b, d},
{c, d}, {a, b, c}, {a, b, d}, {a, c, d}, {b, c, d}, {a, b, c, d}}.
P(A)∪P(B) = {∅, {a}, {b}, {c}, {d}, {a, b}, {b, c}, {b, d}, {c, d}, {b, c, d}}.
P(A) ∪ P(B) ⊆ P(A ∪ B).
This relationship will always hold.
Reason: an element X ∈ P(A) ∪ P(B) is either a subset of A or a subset
of B. In either case, X ⊆ A ∪ B, so X ∈ P(A ∪ B). Hence every element
of P(A) ∪ P(B) is an element of P(A ∪ B), so P(A) ∪ P(B) ⊆ P(A ∪ B).

Exercises 3.3

1. (i) A × (X ∩ Y ) = {(1, c), (2, c), (3, c)}

(iii) (A × X) ∩ (B × Y ) = {(2, c), (3, c)}
(v) (A − B) × X = {(1, a), (1, b), (1, c)}
(vii) (A ∪ B) × (X ∩ Y ) = {(1, c), (2, c), (3, c), (4, c)}

2. (iii) Not every subset of X × Y is of the form A × B for some A ⊆ X

and B ⊆ Y .
Let X = {a, b, c} and Y = {1, 2, 3, 4}. The set S = {(a, 4), (b, 1)} is
a subset of X × Y but it is not of the form A × B (for any sets A
and B).
 Hints and Solutions to Selected Exercises 345

(iv) Suppose sets A, B, X, and Y are such that A × B ⊆ X × Y .

Provided both A and B are non-empty, it does follow that A ⊆ X
and B ⊆ Y .
However, if A = ∅, then for any set B, we have ∅×B = ∅ ⊆ X ×Y ,
so it does not follow that B ⊆ Y .
Similarly, if B = ∅, then it does not follow that A ⊆ X.

3. (i) (A ∩ B) × (X ∩ Y ) = {2, 4} × {y} = {(2, y), (4, y)} and

(A × X) ∩ (B × Y )
= {(1, x), (2, x), (3, x), (4, x), (1, y), (2, y), (3, y), (4, y)}
∩{(2, y), (4, y)(6, y), (2, z), (4, z), (6, z)}
= {(2, y), (4, y)},
so (A ∩ B) × (X ∩ Y ) = (A × X) ∩ (B × Y ).
(ii) The following diagrams illustrate (A ∩ B) × (X ∩ Y ) = (A × X) ∩
(B × Y ).
( A ´ X ) Ç (B ´ Y )
X ( A Ç B) ´ ( X Ç Y ) X A´ X

X ÇY

Y Y B ´Y

A AÇ B A
B B

Exercises 3.4
(1) 2× 1
1 4
11 = − 11 .
2
1. (ii) g = ( )2 =
2 1
−3 −4
2
(iv) (f ◦ f )(−2) = f (f (−2)) = f (| − 3|) = f (3) = 2.
(vi) (h ◦ h)(x) = h(h(x)) = h(2x + 1) = 2(2x + 1) + 1 = 4x + 3.
( )    
2x  2x   2x − x2 + 3 
(viii) (f ◦ g)(x) = f (g(x)) = f 2 
= 2  
− 1 =  .
x −3 x −3 x2 − 3 
( )
1
2. (i) (f ◦ g)(x) = f (g(x)) = f 2
x +1
2 2 + (x2 + 1) x2 + 3
= + 1 = = .
x2 + 1 x2 + 1 x2 + 1
346 Understanding Mathematical Proof
(√ ) 1 1
(iii) (g ◦ h)(x) = g(h(x)) = g x2 + 1 = (√ )2 = 2 .
2
x +1 +1 x +2

x2 + 3
(v) From part (i), (f ◦ g)(x) = .
x2 + 1
Hence ((f ◦ g) ◦ h)(x) = (f ◦ g)(h(x))
(√ )
= (f ◦ g) x2 + 1
(√ )2
x2 + 1 + 3
= (√ )2
x2 + 1 + 1
x2 + 4
= .
x2 + 2
4. (i) im f = {0, 1, 2, 3, 4}.
(ii) im f = {n2 : n ∈ Z} = {0, 1, 4, 9, 16, 25, 36, 49, . . .}.
(iii) im f = {capital cities of the world}.
(iv) im f = {∅, {a}}.
{ }
5. (iii) By definition, im f = {f (x) : x ∈ R} = 1/(x2 + 2) : x ∈ R .
1 1
For all x ∈ R, x2 + 2 ≥ 2, so 2 ≤ 12 . Also 2 > 0.
( 1] { x + 2 } x +2
Therefore im f ⊆ 0, 2 = x ∈ R : 0 < x ≤ 21 .
Conversely, if 0 < x ≤ 12 , then x1 ≥ 2, so x1 − 2 ≥ 0, in which case
√
x − 2 ∈ R.
1

(√ ) 1 1 1
Now f 1
− 2 = (√ )2 = (1 ) = = x,
x −2 +2
x 1/x
1
x − 2 + 2
so x ∈ im f.
{ }
Hence x ∈ R : 0 < x ≤ 12 ⊆ im f.
( ] { }
Therefore im f = 0, 12 = x ∈ R : 0 < x ≤ 21 .
{√ }
(v) By definition, im f = {f (x) : x ∈ R} = x2 + 1 : x ∈ R .
√
For all x ∈ R, x2 + 1 ≥ 1, so x2 + 1 ≥ 1; hence

im f ⊆ {x ∈ R : x ≥ 1}.
√
Conversely, if x ≥ 1, then x2 − 1 ≥ 0, so x2 − 1 ∈ R and
(√ ) √(√ )2 √ √
f x2 − 1 = x2 − 1 + 1 = (x2 − 1) + 1 = x2 = x,

since x ≥ 0. Thus x ∈ im f. Hence {x ∈ R : x ≥ 1} ⊆ im f).

Therefore im f = {x ∈ R : x ≥ 1}.
 Hints and Solutions to Selected Exercises 347

Exercises 3.5

1. (i) (a) For all x, y ∈ R,

f (x) = f (y) ⇒ 1
2x − 7 = 12 y − 7 ⇒ 1
2x = 21 y ⇒ x = y,

so f is injective.
(b) Let y ∈ R. Let x = 2y + 14 ∈ R. Then

f (x) = f (2y + 14) = 12 (2y + 14) − 7 = y,

so f is surjective.
(iii) (a) f (0) = |0−1| = |−1| = 1 and f (2) = |2−1| = 1, so f (0) = f (2);
hence f is not injective.
(b) Let y ∈ R≥0 . Let x = y + 1. Then x ∈ R≥0 and

f (x) = f (y + 1) = |(y + 1) − 1| = |y| = y.

Hence f is surjective.
(vi) (a) f is not injective. For example, f (2) = 1 = f (3) but 2 ̸= 3.
(b) Let m ∈ Z. Then n = 2m ∈ Z is even so
2m
f (n) = f (2m) = = m.
2
Therefore f is surjective.
(ix) (a) f ({1, 2, 3}) = {1, 2, 3} ∩ {1, 2} = {1, 2} and
f ({1, 2, 4}) = {1, 2, 4} ∩ {1, 2} = {1, 2}.
Therefore f ({1, 2, 3}) = f ({1, 2, 4}) but {1, 2, 3} =
̸ {1, 2, 4}, so
f is not injective.
(b) For all X ∈ A, we have f (X) = X ∩ {1, 2} ⊆ {1, 2}.
Hence there is no X ∈ A such that f (X) = {2, 3}, for example.
Therefore f is not surjective.
2. (i) Surjective but not injective.
(iii) Neither injective nor surjective.
(v) Injective and surjective.
3. (ii) (a) For all x, y ∈ R − {−1},
3x 3y
f (x) = f (y) ⇒ =
x+1 y+1
⇒ 3xy + 3x = 3xy + 3y
⇒ 3x = 3y
⇒ x = y.
Therefore f is injective.
348 Understanding Mathematical Proof

(b) Let y ∈ R − {3}.

y
Define x = − . Then x ∈ R − {−1} and
y−3
( )
y 3 × −y/(y − 3)
f (x) = f − =
y−3 −y/(y − 3) + 1
−3y −3y
= = = y.
−y + (y − 3) −3

Therefore f is surjective.
(c) The inverse function is

y
f −1 : R − {3} → R − {−1}, f −1 (y) = − .
y−3

(vi) (a) For all (x, y), (a, b) ∈ R2 ,

f (x, y) = f (a, b) ⇒ (2x − 1, 5y + 3) = (2a − 1, 5b + 3)

⇒ 2x − 1 = 2a − 1 and 5y + 3 = 5b + 3
⇒ 2x = 2a and 5y = 5b
⇒ x = a and y = b
⇒ (x, y) = (a, b).

Therefore f is injective.
(b) Let (a, b) ∈ R2 .
Define x = 21 (a + 1) and y = 15 (b − 3). Then (x, y) ∈ R2 and
( )
f (x, y) = f 12 (a + 1), 15 (b − 3)
( )
= 2 × 21 (a + 1) − 1, 5 × 15 (b − 3) + 3
= ((a + 1) − 1, (b − 3) + 3)
= (a, b).

Therefore f is surjective.
(c) The
( 1 inverse function
) is f −1 : R2 → R2 , f −1 (a, b) =
2 (a + 1), 5 (b − 3) .
1

4. (i) Let B ′ = im f, the image of f .

(ii) For each b ∈ im f, choose a single element a ∈ A such that f (a) = b.

Then define A′ to be the set of all of these chosen elements a.
 Hints and Solutions to Selected Exercises 349

Chapter 4

Exercises 4.1

1. (ii) Proof. Suppose that n is prime. There are two cases: n = 2 or n is

odd.
If n = 2, then n2 + 5 = 9, which is not prime.
If n is odd, then n2 is also odd — see example 1.3. Hence n2 + 5 is
even and so has a factor of 2. Thus n2 + 5 is not prime in this case.
In both cases n2 + 5 is not prime.
2. (i) Proof. Any three consecutive positive integers can be expressed as
n, n + 1, n + 2 for some n ∈ Z+ .
The sum is n + (n + 1) + (n + 2) = 3n + 3 = 3(n + 1), which is
clearly divisible by 3.
3. (i) Proof. Any three consecutive positive integers can be expressed as
n, n + 1, n + 2 for some n ∈ Z+ .
Consider their product n(n + 1)(n + 2). At least one of the factors
is even (divisible by 2). Another factor is divisible by 3 as this is
true of any three consecutive integers.
Therefore the product has a factor of 6.
4. (i) Proof. Any (two) consecutive positive integers can be expressed as
n, n + 1 for some n ∈ Z+ .
Their sum of their squares is n2 + (n + 1)2 = 2n2 + 2n + 1 =
2(n2 + n) + 1 where n2 + n ∈ Z+ .
Therefore n2 + (n + 1)2 is odd.

Exercises 4.2

1. (ii) (a) The proof assumes that n is an arbitrary prime number.

The proof immediately singles out a special case (n = 2) for
separate consideration, but the initial assumption is just that
n is (any) prime.
(b) Some of the background knowledge assumed:
• 9 is not prime;
• the square of an odd integer is also odd;
• any even integer (greater than 2) has a factor of 2 and
hence is not prime.
350 Understanding Mathematical Proof

(c) The initial statement is: n is prime.

The final concluding statement is: n2 +5 is odd (in both cases).
4. (i) (a) The proof implicitly assumes that a and b are arbitrary con-
secutive positive integers. It then makes the deduction that, if
n is the smaller of the two, then the larger of the two is n + 1.
This initial assumption and deduction are wrapped together as
the opening assumption that two arbitrary consecutive positive
integers can be expressed as n and n + 1.
(b) Some of the background knowledge assumed:
• some algebraic manipulation including (n+1)2 = n2 +2n+1
• some elementary factorisation: 2n2 + 2n = 2(n2 + n)
• n2 + n is a positive integer when n is a positive integer
• the oddness property of positive integers.
(c) The initial statement is: n and n + 1 are consecutive positive
integers.
The final concluding statement is: n2 + (n + 1)2 is odd.

Exercises 4.3

1. (iii) Proof. Let x and y be real numbers. There are two cases to consider:
|x| ≥ |y| and |x| < |y|.
From the triangle inequality for the modulus function, we have
|x| = |(x − y) + y| ≤ |x − y| + |y|.
Hence |x| − |y| ≤ |x − y|. (1)
Swapping the roles of x and y in (1) gives
|y| − |x| ≤ |y − x| = |x − y|. (2)
We can now consider the two cases.
Case A: |x| ≥ |y|.
In this case, ||x| − |y|| = |x| − |y| ≤ |x − y| from (1).
Case B: |x| < |y|.
In this case, ||x| − |y|| = − (|x| − |y|) = |y| − |x| ≤ |x − y| from (2).
Hence, in both cases, ||x| − |y|| ≤ |x − y|.
2. (iii) Proof. Let ε > 0.
{ ε}
Let δ = min 1, .
14
Then 0 < |x + 2| < δ
ε
⇒ |x + 2| < 1 and |x + 2| <
14
 Hints and Solutions to Selected Exercises 351

⇒ |x3 + x2 + 4| = |(x + 2)(x2 − x + 2)|

= |x + 2||(x + 2)2 − 5x − 2|
= |x + 2||(x + 2)2 − 5(x + 2) + 8|
( )
≤ |x + 2| |x + 2|2 + 5|x + 2| + 8
< 14|x + 2| (since |x + 2| < 13)
ε ε
< 14 × =ε (since |x + 2| < ).
14 14

4. (i) Proof. We need to show that d satisfies the conditions (M1), (M2),
and (M3) of definition 4.1.
(M1) Let x, y ∈ X. By definition, d(x, y) ≥ 0 and d(x, y) = 0 ⇔
x = y.
Hence d satisfies (M1).

(M2) Let x, y ∈ X.
If x = y, then d(x, y) = 0 = d(y, x).
If x ̸= y, then d(x, y) = 1 = d(y, x).
Hence d satisfies (M2).
(M3) Let x, y, z ∈ X. We consider five cases.
Case A: x = y = z.
Then d(x, z) = 0 = 0 + 0 = d(x, y) + d(y, z).
Case B: x = y ̸= z.
Then d(x, z) = 1 = 0 + 1 = d(x, y) + d(y, z).
Case C: x ̸= y = z.
Then d(x, z) = 1 = 1 + 0 = d(x, y) + d(y, z).
Case D: x ̸= y, y ̸= z but x = z.
Then d(x, z) = 0 < 1 + 1 = d(x, y) + d(y, z).
Case E: x ̸= y, y ̸= z and x ̸= z.
Then d(x, z) = 1 < 1 + 1 = d(x, y) + d(y, z).
In each case d(x, y) + d(y, z) ≥ d(x, z), so d satisfies (M3).
Since d satisfies the three conditions (M1), (M2), and (M3), it is a
metric on X.
{
{a} if 0 ≤ r < 1
(ii) B(a, r) =
X if r ≥ 1.

(iii) Theorem 4.8 (ii) is true for the discrete metric.

352 Understanding Mathematical Proof

Exercises 4.4

1. Here is an informal deduction in the style of formal deductions in chap-

ter 2. We show that, if we assume Γ (P ⇒ Q), then Q may be deduced
from Γ and P .

1. Γ (P ⇒ Q) assumption
2. Γ assumption
3. P assumption
4. P ⇒Q from 1 and 2.
5. Q from 3, 4: Modus ponens in propositional logic
6. Γ, P Q 2, 3, 5: Q has been deduced from Γ, P

2. Here is an informal deduction similar to that in question 1 above. We

show that, if we assume both Γ ¬Q and Γ, P Q, then ¬P may be
deduced from Γ.

1. Γ ¬Q assumption
2. Γ, P Q assumption
3. Γ assumption
4. Γ (P ⇒ Q) from 2 by Conditional rule
5. P ⇒Q from 3 and 4.
6. ¬Q ⇒ ¬P from 5 by Equivalence rule
7. ¬Q from 1 and 3.
8. ¬P from 6, 7: Modus ponens in propositional logic
9. Γ ¬P 3, 8: ¬P has been deduced from Γ

3. Formal structure ‘Real’ proof

Γ We need to prove P and Q.
.. ..
. (sequence of steps 1) . (sequence of steps 1)
P P
.. ..
. (sequence of steps 2) . (sequence of steps 2)
Q Q
P ∧ Q (Proof by conjunction rule) Hence P ∧ Q.
 Hints and Solutions to Selected Exercises 353

Chapter 5

Exercises 5.1

1. (ii) Proof. Let x and y be non-negative real numbers such that x2 ≥ y 2 .

Then x2 − y 2 ≥ 0.
If x = y = 0, then x ≥ y.
Otherwise at least one of x and y is non-zero so x + y > 0.
x2 − y 2
Hence x2 − y 2 ≥ 0 ⇒ ≥0 (since x + y > 0)
x+y
(x + y)(x − y)
⇒ ≥0
x+y
⇒ x−y ≥0
⇒ x ≥ y.
In both cases x ≥ y.
(iv) Proof. Let x, y ∈ R be such that x ̸= 0 and y ̸= 0.
( )2
Then x + 12 y ≥ 0 and y 2 > 0.
( )2
Hence x2 + xy + y 2 = x + 21 y + 34 y 2 > 0.
2. (i) Proof. Let n ∈ Z+ .
Then n3 − n = (n − 1)n(n + 1) is the product of three consecutive,
non-negative integers. One of the integers n − 1, n, n + 1 is divisible
by 3.
Therefore the product (n − 1)n(n + 1) = n3 − n is also divisible by
3.
3. Proof. Let a ≥ 1.
(√ √ √ √ )2
Then a+ a+ a− a
√ √ √ √ √
= (a + a) + 2 (a + a)(a − a) + (a − a)
√
= 2a + 2 a2 − a
√ √
≤ 2a + 2a since a2 − a ≤ a2 = a
= 4a.
√ √ √ √ √
Therefore, taking square roots, a + a + a − a ≤ 2 a.
5. (i) Proof. The only integers in the range

1, 2, . . . , p, p + 1, p + 2, . . . , 2p, 2p + 2, . . . , 3p, . . . p2
354 Understanding Mathematical Proof

which are divisible by p are p, 2p, 3p, . . . , p2 . There are p such inte-
gers.
Hence ϕ(p2 ) = p2 − p = p(p − 1).
αk
(iv) Proof. Let n = pα 1 α2
1 p2 . . . pk be the prime factorisation of n.
αk
Any pair of the terms p1 , pα
α1
2 , . . . , pk are coprime. By the gen-
2

eralisation of the result of part (iii) to a product with k terms, we

have
αk
ϕ(n) = ϕ (pα α2
1 ) ϕ (p2 ) . . . ϕ (pk ) .
1

Strictly, we need a proof by mathematical induction to justify this

step; see chapter 8.
Now, using the result of part (ii), we have
( )
α1 −1 1
ϕ (pα
1
1
) = p α1
1 − p1 = pα1
1 1 − ,
p1
..
. ( )
αk −1 1
ϕ (pk ) = pk − pk
αk αk
= pk 1 −
αk
.
pk

Piecing these results together gives

αk
ϕ(n) = ϕ (pα α2
1 ) ϕ (p2 ) . . . ϕ (pk )
1

( ) ( ) ( )
1 1 1
= p1 1 −
α1
p2 1 −
α1
. . . pk 1 −
αk
p1 p2 pk
( )( ) ( )
1 1 1
= pα
1
1 α1
p2 . . . p αk
k (1 − (1 − . . . 1 −
p1 p2 pk
( )( ) ( )
1 1 1
= n 1− 1− ... 1 − .
p1 p2 pk

6. (ii) Proof. Let p be prime.

Consider the sequence of factorials:

1!, 2!, 3!, . . . , p!, (p + 1)!, (p + 2)!, . . . , (2p)!, (2p + 1)!, . . . .

As noted in the proof of theorem 5.3, the first term that p divides
is p!.
In a similar way, p2 does not divide any of the terms (p + 1)!, (p +
2)!, . . . , (2p − 1)!. However, p2 does divide

(2p)! = 1 × 2 . . . × p × (p + 1) × . . . × (2p).

Therefore (2p)! is the smallest factorial that p2 divides, so S(p2 ) =

2p.
 Hints and Solutions to Selected Exercises 355

(iii) k 1 2 3 4 5 6 7 8 9 10
k
2 2 4 8 16 32 64 128 256 512 1024
S(2k ) 2 4 4 6 8 8 8 10 10 12

Exercises 5.2

4. Proof. Let A and B be sets.

There are two parts to the proof. We need to show both P(A ∩ B) ⊆
P(A) ∩ P(B) and P(A) ∩ P(B) ⊆ P(A ∩ B).
Let C ∈ P(A ∩ B). Then C ⊆ A ∩ B. Hence C ⊆ A and C ⊆ B, which
means C ∈ P(A) and C ∈ P(B). Therefore C ∈ P(A) ∩ P(B).
We have shown that C ∈ P(A ∩ B) ⇒ C ∈ P(A) ∩ P(B). Therefore
P(A ∩ B) ⊆ P(A) ∩ P(B).
Now let C ∈ P(A) ∩ P(B). Then C ∈ P(A) and C ∈ P(B) which means
C ⊆ A and C ⊆ B. Hence C ⊆ A ∩ B — see exercise 3.2.4 (i). Therefore
C ∈ P(A ∩ B).
We have shown that C ∈ P(A) ∩ P(B) ⇒ C ∈ P(A ∩ B). Therefore
P(A) ∩ P(B) ⊆ P(A ∩ B).
We have now shown both P(A ∩ B) ⊆ P(A) ∩ P(B) and P(A) ∩ P(B) ⊆
P(A ∩ B). Hence P(A ∩ B) ⊆ P(A) ∩ P(B).

6. (ii) Proof. We need to show both (A∩B)×(X ∩Y ) ⊆ (A×X)∩(B ×Y )

and (A × X) ∩ (B × Y ) ⊆ (A ∩ B) × (X ∩ Y ).
(⊆) Let (x, y) ∈ (A ∩ B) × (X ∩ Y ). Then x ∈ A ∩ B and y ∈ X ∩ Y .
Therefore x ∈ A and y ∈ X, so (x, y) ∈ A × X. Also x ∈ B
and y ∈ Y so (x, y) ∈ B × Y .
Hence (x, y) ∈ (A × X) ∩ (B × Y ).
We have shown that (x, y) ∈ (A ∩ B) × (X ∩ Y ) ⇒ (x, y) ∈
(A×X)∩(B ×Y ). Hence (A∩B)×(X ∩Y ) ⊆ (A×X)∩(B ×Y ).
(⊇) Let (x, y) ∈ (A × X) ∩ (B × Y ). Then (x, y) ∈ A × X, so x ∈ A
and y ∈ X. Also (x, y) ∈ B × Y , so x ∈ B and y ∈ Y .
Hence x ∈ A ∩ B and y ∈ X ∩ Y , so (x, y) ∈ (A ∩ B) × (X ∩ Y ).
We have shown that (x, y) ∈ ⇒ (A × X) ∩ (B × Y )(x, y) ∈
(A∩B)×(X ∩Y ). Hence (A×X)∩(B ×Y ) ⊆ (A∩B)×(X ∩Y ).
Since (A ∩ B) × (X ∩ Y ) ⊆ (A × X) ∩ (B × Y ) and (A × X) ∩
(B × Y ) ⊆ (A ∩ B) × (X ∩ Y ), it follows that (A ∩ B) × (X ∩ Y ) =
(A × X) ∩ (B × Y ).

7. (iii) Proof. For all sets A, B, C, we have

(A ∩ B) ∗ (A ∩ C)
356 Understanding Mathematical Proof

= ((A ∩ B) − (A ∩ C)) ∪ ((A ∩ C) − (A ∩ B))

(definition of ∗)
( ) ( )
= (A ∩ B) ∩ A ∩ C ∪ (A ∩ C) ∩ A ∩ B
(definition of set difference)
( ) ( )
= (A ∩ B) ∩ (Ā ∪ C̄) ∪ (A ∩ C) ∩ (Ā ∪ B̄)
(De Morgan’s law)
= (A ∩ B ∩ Ā) ∪ (A ∩ B ∩ C̄) ∪ (A ∩ C ∩ Ā) ∪ (A ∩ C ∩ B̄)
(Distributive law)
= ∅ ∪ (A ∩ B ∩ C̄) ∪ ∅ ∪ (A ∩ C ∩ B̄)
(Complement and commutative laws)
= (A ∩ B ∩ C̄) ∪ (A ∩ C ∩ B̄) (Identity law)
( )
= A ∩ (B ∩ C̄) ∪ (C ∩ B̄) (Distributive law)
= A ∩ ((B − C) ∪ (C − B)) (definition of set difference)
= A ∩ (B ∗ C) (definition of ∗)

8. (ii) Proof. For all x, y ∈ R − {−1},

3x 3y
f (x) = f (y) ⇒ =
x+1 y+1
⇒ 3xy + 3x = 3xy + 3y
⇒ 3x = 3y
⇒ x = y.

Therefore f is injective.
Now let y ∈ R − {3}.
y
Define x = − . Then x ∈ R − {−1} and
y−3
( )
y 3 × −y/(y − 3) −3y −3y
f (x) = f − = = = = y.
y−3 −y/(y − 3) + 1 −y + (y − 3) −3
Therefore f is surjective.
Hence f is a bijection.
(v) Proof. For all (x, y), (a, b) ∈ R2 ,

f (x, y) = f (a, b) ⇒ (2x − 1, 5y + 3) = (2a − 1, 5b + 3)

⇒ 2x − 1 = 2a − 1 and 5y + 3 = 5b + 3
⇒ 2x = 2a and 5y = 5b
⇒ x = a and y = b
⇒ (x, y) = (a, b).
 Hints and Solutions to Selected Exercises 357

Therefore f is injective.
Let (a, b) ∈ R2 .
Define x = 21 (a + 1) and y = 15 (b − 3). Then (x, y) ∈ R2 and
( )
f (x, y) = f 21 (a + 1), 15 (b − 3)
( )
= 2 × 12 (a + 1) − 1, 5 × 15 (b − 3) + 3
= ((a + 1) − 1, (b − 3) + 3)
= (a, b).

Therefore f is surjective.
Hence f is a bijection.
9. (iii) Proof. Let x ∈ C. Then f (x) ∈ f (C) by the definition of f (C).
Hence x ∈ f −1 (f (C)) by the definition of f −1 (D) with D = f (C).
We have shown that x ∈ C ⇒ x ∈ f −1 (f (C)). Therefore C ⊆
f −1 (f (C)).
(iv) Proof. Suppose that f is injective.
By part (iii), we only need to prove that f −1 (f (C)) ⊆ C.
Let x ∈ f −1 (f (C)). Then f (x) ∈ f (C) by the definition of f −1 (D)
with D = f (C).
By the definition of f (C), this means that f (x) = f (x′ ) for some
x′ ∈ C. But f is injective, so x = x′ , which means that x ∈ C.
We have shown that x ∈ f −1 (f (C)) ⇒ x ∈ C. Therefore
f −1 (f (C)) ⊆ C.
10. (ii) The required functions are defined by g(x) = 12 (f (x) + f (−x)) and
h(x) = 12 (f (x) − f (−x)).

Exercises 5.3

3. Proof. Let H be a subset of G that satisfies the three given conditions.

Condition (ii) says that ∗ is a binary operation on H, so we need to
verify the three group axioms.
(G1) The operation ∗ is associative on H because it is associative on G.
(G2) Since H is non-empty, we may choose an x ∈ H. By condition (iii)
its inverse is also in H, x−1 ∈ H.
Therefore, by condition (ii), x ∗ x−1 = e ∈ H. Hence H contains
the identity element for ∗.
(G3) Condition (iii) guarantees that every element of H has an inverse
in H.
358 Understanding Mathematical Proof

Therefore (H, ∗) is a group and hence is a subgroup of (G, ∗).

5. Proof. Suppose that S is a set with binary operation ∗ satisfying (A1)
and (A2).
Let x, y ∈ S.
By (A1), e ∈ S, so we may apply (A2) with elements e, x, y, which gives
x ∗ y = (e ∗ x) ∗ y = e ∗ (y ∗ x) = y ∗ x.
Therefore ∗ is commutative.
Now let x, y, z ∈ S. Then
(x ∗ y) ∗ z = x ∗ (z ∗ y) by (A2)
= x ∗ (y ∗ z) by commutativity.
Therefore ∗ is associative.
7. Proof. Let S and T be two subspaces of a vector space U . To prove
S + T is a subspace of U , we will use the Subspace Test, theorem 5.12.
First note that 0 ∈ S and 0 ∈ T , so 0 = 0 + 0 ∈ S + T . Therefore S + T
is non-empty.
Let v, w ∈ S + T . Then there exist vectors x1 , x2 ∈ S and y1 , y2 ∈ T
such that v = x1 + y1 and w = x2 + y2 .
Hence v + w = (x1 + y1 ) + (x2 + y2 )
= (x1 + x2 ) + (y1 + y2 ) (by vector space axioms
(A1) and (A2)).
Now x1 +x2 ∈ S since S is closed under addition of vectors and y1 +y2 ∈
T since T is also closed under addition of vectors. Therefore (x1 + x2 ) +
(y1 + y2 ) ∈ S + T . We have shown that S + T is closed under addition.
Let v = x1 + y1 ∈ S + T be as above and let λ ∈ R. Then
λv = λ(x1 + y1 ) = λx1 + λy1 (by vector space axiom (M2)).
Now λx1 ∈ S since S is closed under scalar multiplication and λy1 ∈ T
since T is also closed under scalar multiplication. Therefore λx1 + λy1 ∈
S + T . We have shown that S + T is closed under scalar multiplication.
Hence S + T satisfies the conditions of theorem 5.12 and is therefore a
subspace of U .
9. (i) Proof. Let λ be an eigenvalue of an n × n matrix A.
By definition, there exists a non-zero vector x such that Ax = λx.
Then (A + In )x = Ax + In x
= λx + x
= (λ + 1)x.
Therefore λ + 1 is an eigenvalue of A + In (with eigenvector x).
 Hints and Solutions to Selected Exercises 359

(ii) Proof. In part (i), we proved that x ∈ EA (λ) ⇒ x ∈ EA+In (λ+1).

This shows EA (λ) ⊆ EA+In (λ + 1) and we also need to prove the
subset relation the other way around.
Now x ∈ EA+In (λ + 1) ⇒ (A + In )x = (λ + 1)x
⇒ Ax + In x = λx + x
⇒ Ax + x = λx + x
⇒ Ax = λx
⇒ x ∈ EA (λ).
Therefore EA+In (λ + 1) ⊆ EA (λ).
Since we have established the subset relation both ways round, it
follows that EA (λ) = EA+In (λ + 1).

10. Use the Subspace Test, theorem 5.12.

11. For each part, use the Subspace Test, theorem 5.12.

Exercises 5.4
⌊ ⌋
19
1. (ii) Proof. Given ε > 0, let N = .
16ε
Then
19
n>N ⇒ n>
16ε
 3   
 n + 3n 1   4(n3 + 3n) − (4n3 + 7n2 ) 
⇒  3 − =  
4n + 7n2 4  4(4n3 + 7n2 ) 
 
 −7n2 + 12n 
=  
16n3 + 28n2 
7n2 + 12n
≤
16n3 + 28n2
7n2 + 12n2 19
≤ 3
= < ε.
16n 16n

n3 + 3n 1
Hence lim = as claimed.
n→∞ 4n3 + 7n2 4
2. Proof. Let (an ) be a convergent sequence with lim an = ℓ and let
n→∞
λ ∈ R.
If λ = 0, then (λan ) is the constant sequence with values 0. Hence (λan )
converges to limit 0 = λℓ.
360 Understanding Mathematical Proof

So suppose λ ̸= 0.
Let ε > 0.
ε
Then > 0 and, since lim an = ℓ, there exists N ∈ Z+ such that
|λ| n→∞

ε
n > N ⇒ |an − ℓ| < .
|λ|

Therefore, for n > N ,

ε
|λan − λℓ| = |λ||an − ℓ| < |λ| × = ε.
|λ|

Hence lim λan = λℓ.

n→∞
{ ε}
3. (iv) Proof. Let ε > 0. Let δ = min 1, . Then
14
ε
0 < |x − 1| < δ ⇒ |x − 1| < 1 and |x − 1| <
14
⇒ |3x2 + 5x − 8| = |3x + 8||x − 1|
= |3(x − 1) + 11||x − 1|
≤ (3|x − 1| + 11) |x − 1| (triangle
inequality)
< 14|x − 1| since |x − 1| < 1
ε
< ε since |x − 1| < .
14
Therefore lim 3x2 + 5x = 8.
x→1
{ √ }
(vi) Proof. Let ε > 0. Let δ = min 1, (1 + 2)ε .
First note that
√
|x − 2| < 1 ⇒ −1 < x − 2 < 1 ⇒ 1 < x < 3 ⇒ x > 1. (∗)
Then
√
0 < |x − 2| < δ ⇒ |x − 2| < 1 and |x − 2| < (1 + 2)ε
 √ 
√ √   ( x − √2)(√x + √2) 
 x − 2 =   
⇒ √ √ 
 x+ 2 
|x − 2|
= √ √
x+ 2
|x − 2| √
< √ since x > 1 by (∗)
1+ 2 √
< ε since |x − 1| < (1 + 2)ε.
√ √
Therefore lim x = 2.
x→2
 Hints and Solutions to Selected Exercises 361
ε
4. Proof. Let ε > 0. Then > 0.
2
Therefore there exist δ1 > 0, δ2 > 0 such that
ε
0 < |x − a| < δ1 ⇒ |f (x) − ℓ| <
2
ε
and 0 < |x − a| < δ2 ⇒ |g(x) − m| < .
2

Now let δ = min {δ1 , δ2 }. Then

0 < |x − a| < δ ⇒ 0 < |x − a| < δ1 and 0 < |x − a| < δ2

ε ε
⇒ |f (x) − ℓ| < and |g(x) − m| <
2 2
⇒ |f (x) + g(x) − (ℓ + m)| = |(f (x) − ℓ) + (g(x) − m)|
≤ |f (x) − ℓ| + |g(x) − m|
ε ε
< + = ε.
2 2

Therefore lim (f (x) + g(x)) = ℓ + m.

x→a

Chapter 6

Exercises 6.1

1. (ii) Proof. Let k be an odd integer. Then k = 2a + 1 for some a ∈ Z.

We prove the contrapositive: if n is odd, then kn + (k + 1) is odd.
So suppose that n is odd. Then n = 2m + 1 for some m ∈ Z. Now

kn + (k + 1) = (2a + 1)(2m + 1) + (2a + 1) + 1

= 4am + 2a + 2m + 1 + 2a + 2
= 2(2am + 2a + m + 1) + 1
= 2M + 1 where M = 2am + 2a + m + 1 ∈ Z.

Therefore kn + (k + 1) is odd, which completes the proof of the

contrapositive.
Hence, if kn + (k + 1) is even (and k is odd), then n is even.
362 Understanding Mathematical Proof

(iii) Proof. We prove the contrapositive: if n is not divisible by 3, then

n2 is not divisible by 3.
So suppose that n is an integer that is not divisible by 3.
There are two possibilities: n = 3m + 1 for some m ∈ Z or n =
3m + 2 for some m ∈ Z.
If n = 3m + 1, then

n2 = (3m + 1)2 = 9m2 + 6m + 1 = 3(3m2 + 2m) + 1

where 3m2 + 2m ∈ Z. Hence n2 is not divisible by 3.

If n = 3m + 2, then

n2 = (3m + 2)2 = 9m2 + 12m + 4 = 3(3m2 + 4m + 1) + 1

where 3m2 + 4m + 1 ∈ Z. Hence n2 is not divisible by 3.

In both cases n2 is not divisible by 3. Therefore, if n2 is divisible
by 3, then n itself is divisible by 3.
2. (ii) Proof. Let k be a positive integer.
We prove the contrapositive, so suppose ‘m ≤ k or n ≤ k’ is false.
Then m > k and n > k (by De Morgan’s Law, page 39). Therefore
mn > k 2 so, in particular, mn ̸= k 2 .
This completes the proof of the contrapositive.
5. Proof. We prove the contrapositive, so let f : A → B and g : B → C be
functions where g is not surjective. Then there exists c ∈ C such that
c ̸= g(b) for all b ∈ B.
Hence c ̸= g(f (a)) for all a ∈ A, so the composite function g ◦ f is not
surjective.
This completes the proof of the contrapositive. Hence if g ◦f is surjective
then so, too, is g.
7. (i) Proof. We prove the contrapositive, so suppose that the statement
‘A ⊆ X or B ⊆ Y ’ is false. Then, by De Morgan’s law, A ̸⊆ X and
B ̸⊆ Y .
This means that there exists a ∈ A such that a ̸∈ X and there exists
b ∈ B such that b ̸∈ Y . Therefore (a, b) ∈ A×B, but (a, b) ̸∈ X ×Y .
Hence A × B ̸⊆ X × Y .
This completes the proof of the contrapositive. Hence if A × B ⊆
X × Y , then A ⊆ X or B ⊆ Y .

Exercises 6.2

2. (ii) Proof. Let a, m, and n be positive integers.

 Hints and Solutions to Selected Exercises 363

(⇒) m|n ⇒ n = km for some k ∈ Z+

⇒ an = k(am) (where k ∈ Z+ )
⇒ am|an.

(⇐) am|an ⇒ an = k(am) for some k ∈ Z+

⇒ n = km since a ̸= 0
⇒ m|n.

4. (i) Proof. Suppose that f : A → B is injective and let C1 and C2 be

subsets of A. We need to prove that f (C1 ∩ C2 ) = f (C1 ) ∩ f (C2 ),
so we will show that each set is a subset of the other.
Firstly,
y ∈ f (C1 ∩ C2 )
⇒ y = f (x) for some x ∈ C1 ∩ C2
⇒ y ∈ f (C1 ) (since x ∈ C1 ) and y ∈ f (C2 ) (since x ∈ C2 )
⇒ y ∈ f (C1 ) ∩ f (C2 ).

Hence f (C1 ∩ C2 ) ⊆ f (C1 ) ∩ f (C2 ).

Secondly,
y ∈ f (C1 ) ∩ f (C2 )
⇒ y ∈ f (C1 ) and y ∈ f (C2 )
⇒ y = f (x1 ) for some x1 ∈ C1 and
y = f (x2 ) for some x2 ∈ C2
⇒ f (x1 ) = f (x2 )
⇒ x1 = x2 since f is injective
⇒ y = f (x1 ) where x1 ∈ C1 ∩ C2
⇒ y ∈ f (C1 ∩ C2 ).

Hence f (C1 ) ∩ f (C2 ) ⊆ f (C1 ∩ C2 ).

We have proved that each set is a subset of the other, so f (C1 ∩
C2 ) = f (C1 ) ∩ f (C2 ).

Conversely, suppose that f (C1 ∩C2 ) = f (C1 )∩f (C2 ) for all subsets
C1 and C2 of A. We need to show that f is injective.
Let x1 , x2 ∈ A. Then
364 Understanding Mathematical Proof

f (x1 ) = f (x2 )
⇒ f (x1 ) ∈ f ({x1 }) and f (x1 ) ∈ f ({x2 })
⇒ f (x1 ) ∈ f ({x1 }) ∩ f ({x2 })
⇒ f (x1 ) ∈ f ({x1 } ∩ f {x2 })
since f ({x1 }) ∩ f ({x2 }) = f ({x1 } ∩ f {x2 })
⇒ {x1 } ∩ f {x2 } ̸= ∅
⇒ x1 = x2 .
Hence f is injective.
5. (ii) Proof.
(⇒) Suppose that f : A → B is an increasing function and α ∈ R+ .
Let x, y ∈ A be such that x < y. Then
f (x) ≤ f (y) (since f is increasing)
⇒ αf (x) ≤ αf (y) (since α > 0).
Therefore αf is increasing.
(⇐) Suppose that αf : A → B is an increasing function where
α ∈ R+ .
Let x, y ∈ A be such that x < y. Then
αf (x) ≤ αf (y) (since αf is increasing)
⇒ f (x) ≤ f (y) (since α > 0).
Therefore f is increasing.

Note that there is a slicker proof of the converse, as follows.

(⇐) Suppose that αf : A → B is an increasing function where
α ∈ R+ .
Then 1/α ∈ R+ , so by the first part, f = (1/α)(αf ) is also
increasing.
6. Proof. Firstly, suppose that n = p4 where p is prime. Then n has exactly
5 factors, namely 1, p, p2 , p3 , and p4 .
Conversely, suppose that n is a positive integer that has exactly 5 fac-
tors. By the Prime Factorisation theorem 4.3, n may be expressed as a
product of prime numbers
n = pα 1 α2 αm
1 p2 . . . pm

where p1 , p2 , . . . , pm are prime and each αr is a positive integer. Fur-

thermore, this expression is unique except for the ordering of the prime
numbers — see theorem 7.16.
Any factor of n must therefore be of the form
pβ1 1 pβ2 2 . . . pβmm
 Hints and Solutions to Selected Exercises 365

where, for r = 1, . . . , m, 0 ≤ βr ≤ αr . Since any set of choices of the

βr gives a factor of n, there are (α1 + 1)(α2 + 1) . . . (αm + 1) factors
altogether.
Since n has exactly five factors, (α1 + 1)(α2 + 1) . . . (αm + 1) = 5. But
5 is a prime number, so we must have m = 1 and α1 = 4. Therefore
n = p41 (where p1 is prime), as required.

Exercises 6.3
1. (ii) Proof. If α and β are the roots of x2 + ax + b = 0, then
x2 + ax + b = (x − α)(x − β) = x2 − (α + β)x + αβ.
If α and β are odd integers, then their sum α + β = a is even and
their product αβ = b is odd.
(iii) Proof. Firstly,
−1 ≤ x ≤ 4 ⇒ −4 ≤ x − 3 ≤ 1
⇒ 0 ≤ (x − 3)2 ≤ 16
⇒ 0 ≤ x2 − 6x + 9 ≤ 16
⇒ −9 ≤ x2 − 6x ≤ 7.

Following a similar approach,

−1 ≤ x ≤ 4 ⇒ 2≤x+3≤7
⇒ 4 ≤ (x + 3)2 ≤ 49
⇒ 4 ≤ x2 + 6x + 9 ≤ 49
⇒ −5 ≤ x2 + 6x ≤ 40.

2. (iv) Proof. Let (a, x) ∈ (A ∩ B) × (X ∩ Y ).

Then a ∈ A ∩ B and x ∈ X ∩ Y .
Hence a ∈ A and x ∈ X, so (a, x) ∈ A × X. Also a ∈ B and x ∈ Y ,
so (a, x) ∈ B × Y . Therefore (a, x) ∈ (A × X) ∩ (B × Y ).
Hence (A ∩ B) × (X ∩ Y ) ⊆ (A × X) ∩ (B × Y ).
Now let (a, x) ∈ (A × X) ∩ (B × Y ).
Then (a, x) ∈ A × X, so a ∈ A and x ∈ X. Also (a, x) ∈ B × Y ,
so a ∈ B and x ∈ Y . Hence a ∈ A ∩ B and x ∈ X ∩ Y . Therefore
(a, x) ∈ (A ∩ B) × (X ∩ Y ).
Hence (A × X) ∩ (B × Y ) ⊆ (A ∩ B) × (X ∩ Y ).
Since we have proved that each set is a subset of the other, we have
(A ∩ B) × (X ∩ Y ) = (A × X) ∩ (B × Y ).
366 Understanding Mathematical Proof

3. (ii) Proof. Let x ∈ (A ∪ C) ∗ (B ∪ C). Then x ∈ (A ∪ C) − (B ∪ C) or

x ∈ (B ∪ C) − (A ∪ C).
In the first case,
x ∈ A ∪ C and x ̸∈ B ∪ C
⇒ (x ∈ A or x ∈ C) and (x ̸∈ B and x ̸∈ C)
(since x ∈ C and x ̸∈ C is impossible)
⇒ x ∈ A and x ̸∈ B and x ̸∈ C
⇒ x ∈ A − B and x ̸∈ C
⇒ x ∈ A ∗ B and x ̸∈ C (since A − B ⊆ A ∗ B)
⇒ x ∈ (A ∗ B) − C.
In the second case, interchanging A and B in the previous argument
shows that x ∈ (B ∗ A) − C = (A ∗ B) − C.
Therefore x ∈ (A ∗ B) − C in both cases, so that (A ∪ C) ∗ (B ∪ C ⊆
(A ∗ B) − C.
Conversely, let x ∈ (A ∗ B) − C. Then x ∈ A ∗ B and x ̸∈ C.
Now x ∈ A ∗ B means that x ∈ A − B or x ∈ B − A.
In the first case, we have x ∈ A, x ̸∈ B, and x ̸∈ C. Hence x ∈ A
and x ̸∈ B ∪ C. Since A ⊆ (A ∪ C), this means that x ∈ A ∪ C and
x ̸∈ B ∪ C so x ∈ (A ∪ C) − (B ∪ C).
In the second case, reversing the roles of A and B in the previous
argument shows that x ∈ (B ∪ C) − (A ∪ C).
Putting the two cases together we have
x ∈ ((A ∪ C) − (B ∪ C)) ∪ ((B ∪ C) − (A ∪ C))
= (A ∪ C) ∗ (B ∪ C).
Hence (A ∗ B) − C ⊆ (A ∪ C) ∗ (B ∪ C).
As we have shown that each set is a subset of the other, we have
(A ∪ C) ∗ (B ∪ C) = (A ∗ B) − C.
4. (iii) To prove X ⊆ Y : let x ∈ X. Clearly x > 0. To show that x ≤ 12 ,
use the approach given in the proof in example 5.1.
To prove Y ⊆ X: let 0 < c ≤ 12 . Then use the quadratic formula to
show that the equation x/(x2 + 1) = c has a solution for x ≥ 1.

Exercises 6.4
1. (i) Proof. Let x be rational and y be irrational.
Suppose that x + y is rational. Then y = (x + y) − x is the dif-
ference of two rational numbers, which is therefore rational. This
contradicts the fact that y is irrational.
Therefore x + y is irrational.
 Hints and Solutions to Selected Exercises 367

3. Firstly, suppose that ax2 + bx + c = 0 has a rational root (where a, b,

and c are odd integers).
Deduce that both roots must be rational — question 1 may be of use
here.
Then show that ax2 +bx+c = (px−q)(rx−s) for some integers p, q, r, s.
Deduce that p, q, r, and s are all odd.
The contradiction comes from the fact that the coefficient of x must
then be even.

5. (i) Proof. Let m and n be integers where n ̸= 0.

√ √ p
Suppose that m+ 2n is rational; then m+ 2n = where p, q ∈ Z.
q
√ p p − qm √ p − qm
Hence 2n = − m = , so 2 = ∈ Q. This is a
q √ q qn
contradiction since 2 is irrational.
√
Therefore m + 2n is irrational.

6. Proof. Suppose that there is a rational number r such that 2r = 3.

Now r may be expressed as r = p/q where p, q ∈ Z and q ̸= 0. Hence
2p/q = 3. Raising both sides to the power q gives 2p = 3q . Now 2p is
even and 3q is odd, so 2p ̸= 3q .
Hence there is no rational number r such that 2r = 3.

7. Suppose that x is the smallest real number such that x > 21 .

x 1
Show that y = + is smaller than x but still greater than 12 .
2 4
m−1
11. Proof. Suppose ∈ A (where m ∈ Z+ ) is the largest element of A.
m
m (m + 1) − 1
Note that = ∈ A and
m+1 m+1

m m−1 m2 − (m + 1)(m − 1)
− =
m+1 m m(m + 1)
m2 − (m2 − 1) 1
= = > 0.
m(m + 1) m(m + 1)

m m−1
Hence > .
m+1 m
m m−1
In other words, is an element of A that is greater than ,
m+1 m
which is therefore not the largest element of A.
Hence A has no largest element.
368 Understanding Mathematical Proof

Exercises 6.5

1. (i) The identity element of H is the element eH that satisfies h ◦ eH =

h = eh ◦ eH for all h ∈ H.
Show that θ(eG ) satisfies this property for any element h that is in
the image of θ; h = θ(g) for some g ∈ G.
(iii) Proof. Let (G, ∗) and (H, ◦) be groups and let θ : G → H be a
morphism.
To prove that ker θ is a subgroup of G, we use the Subgroup Test,
theorem 5.16 given in exercise 5.3.3.
Firstly, since θ(e) = e, by part (i), it follows that e ∈ ker θ. (Note
that we have dropped the subscripts on the identity elements to
keep the notation simple.) Hence ker θ ̸= ∅.
Let x, y ∈ ker θ. Then, by definition of the kernel, θ(x) = e and
θ(y) = e. Therefore θ(x ∗ y) = θ(x) ◦ θ(y) = e ◦ e = e, so xy ∈ ker θ.
−1
By part (ii), θ(x−1 ) = (θ(x)) = e−1 = e, so x−1 ∈ ker θ.
We have shown that ker θ satisfies the three conditions of the Sub-
group Test. It therefore follows that ker θ is a subgroup of G.

2. Proof.
(⇒) Suppose that G is Abelian.
Let g, h ∈ G. Then

(gh)−1 = h−1 g −1 (theorem 5.9)

−1 −1
= g h (G is Abelian.)

(⇐) Now suppose that (gh)−1 = g −1 h−1 for all elements g, h ∈ G.

In particular, since this equation is satisfied all elements of G, it is
( )−1 ( −1 )−1 ( −1 )−1
satisfied by g −1 and h−1 : g −1 h−1 = g h .
Hence we have
( )−1 ( −1 )−1
gh = g −1 h
( −1 −1 )−1
= g h
( )−1 ( −1 )−1
= h−1 g (theorem 5.9)
= hg.

Therefore G is Abelian.

5. Let X be a non-empty subset of R. A real number m satisfying:

(i) m is a lower bound for X: m ≤ x for all x ∈ X and
 Hints and Solutions to Selected Exercises 369

(ii) if b > m, then b is not a lower bound for X; if b > m then, there
exists x ∈ X such that x < b
is called a infimum or greatest lower bound for X; it is denoted
m = inf X.
x−1
6. (i) Proof. First note that, for all x ∈ R+ , x − 1 < x, so < 1.
x
Hence 1 is an upper bound for X.
1
Now let a < 1. Then > 0.
1−a
Since there is no smallest positive real number (compare with ex-
a
ercise 6.4.7), there exists x ∈ R+ such that x < . Hence
1−a
a
x< ⇒ x(1 − a) < 1 (since a − 1 > 0)
1−a
⇒ x − ax < 1
⇒ x − 1 < ax
x−1
⇒ <a (since x > 0).
x
In other words, there exists an element (x − 1)/x ∈ X that is less
than a. Hence a is not an upper bound for X.
Therefore sup X = 1.
(v) Proof. Let x ∈ X. Then x = m/n for some m, n ∈ Z+ such that
m < 2n. Hence x < 2. Therefore 2 is an upper bound for X
1
Now let a < 2. Then 2−a > 1. Choose n ∈ Z+ such that n > .
2−a
Then
1 1 1 2n − 1
n> ⇒ 2−a> ⇒ a<2− =
2−a n n n
2n − 1
where ∈ X.
n
Therefore a is not an upper bound for X.
Hence sup X = 2.

Chapter 7

Exercises 7.1

1. (iii) 73 = 32 + 82 and 74 = 52 + 72 .
370 Understanding Mathematical Proof
√ √ √
(vi) Both 1 + 2 and (1 + 2)2 = 3 + 2 2 are irrational.
(ix) 1729 = 13 + 123 = 93 + 103 .
(xi) Proof. Let n be an odd positive integer.
Then n = 2k + 1 for some k ∈ N. Hence n2 = (2k + 1)2 = 4k 2 +
4k + 1 = 4k(k + 1) + 1.
Now k(k + 1) is even — see theorem 1.2. Hence k(k + 1) = 2m for
some m ∈ N.
Therefore n2 = 4k(k + 1) + 1 = 8m + 1 for some m ∈ N.
2. Try to construct examples where A, B, and C are small finite sets (at
most three elements) and the functions f and g are defined by specifying
explicitly the image of each element.
( )
−1 0
3. (ii) For example, A = .
0 1
 
( ) 1 1
1 0 −1
(iv) For example, A = and B = 1 0.
2 −2 −1
0 1
Then AB = I2 but BA ̸= I3 .
5. The group S(△), defined in example 5.4.4, is a suitable example to
establish parts (i), (ii), and (vi).

Exercises 7.2

2. (i) Proof. Let {a1 , a2 , . . . , an } be a set of non-zero integers such that

∑n
ak < n and suppose that none of the ai are negative.
k=1
∑
n
Then each a1 ≥ 1 so ak ≥ 1 + 1 + · · · + 1(n terms) = n.
k=1
This is a contradiction. Hence at least one of the ai is negative.
4. Proof. Suppose that x ∈ R − Q is such that, in its decimal expansion,
each digit 0, 1, 2, . . . , 9 occurs a finite number of times.
Then the decimal expansion of x terminates. Suppose the decimal ex-
pansion terminates after N decimal places, so that x = x0·x1 x2 . . . xN
where x0 ∈ Z and x1 , . . . , xN ∈ {0, 1, . . . , 9}. Then x × 10N is an integer,
so x is rational.
6. Consider the prime numbers less than 30. Place each integer in the set
{2, 3, . . . , 30} into a pigeonhole according to its smallest prime factor.
Then use the Pigeonhole Principle to show that there are two integers
with the same smallest prime factor.
 Hints and Solutions to Selected Exercises 371

7. (ii) Proof. Suppose that n2 + 1 points are placed in a square of side n.

Divide the square into n2 smaller squares, each of side length 1 as
shown.

Since there are n2 + 1 points placed in the (large) square and there
are only n2 small squares, there must be a small square that
√ con-
tains two points. These two points are no more than 2 units
apart.
10. (ii) If a pair of dice is rolled, there are 11 possible scores: 2, 3, . . . , 12.
Since 45 > 4 × 11, if a pair of dice is rolled 45 times, there must be
one score that occurs at least 5 times by the Generalised Pigeonhole
Principle.
12. (iii) Proof. Let f and g be two functions, each continuous on [a, b], with
the property that f (a) < g(a) and f (b) > g(b).
Let h be the function h = f − g defined on [a, b]; in other words
h(x) = f (x) − g(x) for all x ∈ [a, b].
Then h is continuous on [a, b], h(a) = f (a) − g(a) < 0 and h(b) =
f (b) − g(b) > 0. Therefore, by the Intermediate Value Theorem,
there exists c ∈ (a, b) such that h(c) = 0. Hence f (c) = g(c).

Exercises 7.3

2. (i) For example, take a = 6, b = 5, c = 4, d = 1.

(iii) f (17) = 172 is not prime.
(v) The smallest counter-example is n = 10, where 310 = 59049 >
40000 = 4 × 104 .
4. (i) This is true. If a = p/q and b = r/s where p, q, r, s ∈ Z and q ̸= 0,
s ̸= 0, then ab = (pr)/(qs) where pr, qs ∈ Z and qs ̸= 0.
√
(ii) This is false. A counter-example is a = b = 2.
√
(v) This is false. A counter-example is α = β = 2, m = 1, and
n = −1.
6. (ii) This is false: 8n + 1 factorises as (2n + 1)(4n − 2n + 1).
372 Understanding Mathematical Proof

7. (i) This is false. For example, let f and g be functions R → R defined

by f (x) = 2x and g(x) = −x. Then f + g is increasing but g is not
increasing.
Note that the implication ‘f and g are increasing ⇒ f + g is in-
creasing’ is true and has a simple direct proof.

9. (i) Let G = {e, r, v, h} be the Klein 4-group defined in example 5.4.3

and let H = {e, r} and K = {e, v}. Then H and K are subgroups
of G but H ∪ K is not a subgroup of G (because r, v ∈ H ∪ K but
r ∗ v = h ̸∈ H ∪ K).
(ii) The group (S△), defined in example 5.4.4, is a suitable example.

10. (i) This is true and has a simple direct proof.

(ii) This is false. For example, let A = {1}, B = {2}, and C = {1, 3}.
Then A ̸⊆ B and B ̸⊆ C but A ⊆ C.
(v) This is false. Let A = {1, 2}, B = {a, b}, and C = {(1, a), (2, b)}.
Clearly C ⊆ A × B = {{(1, a), (1, b), (2, a)(2, b)}.
We need to show that C is not the Cartesian product of any two
sets X and Y , which we do by contradiction.
So suppose that C = X ×Y for some sets X and Y . Since (1, a) ∈ C,
we have 1 ∈ X; similarly, since (2, b) ∈ C, we have b ∈ Y . Then
(1, b) ∈ X × Y = C, which is a contradiction. Therefore C is not
the Cartesian product of any sets X and Y .

Exercises 7.4
√ √
1. (ii) Proof. Assuming the existence of a as a real number, then x = a
is a solution of x2 = a.
Suppose that x and y are both positive real numbers such that
x2 = a and y 2 = a. Then

x2 = y 2 ⇒ x2 − y 2 = 0 ⇒ (x + y)(x − y) = 0.

Now x + y > 0, so x − y = 0 so x = y.
Hence x2 = a has a unique positive solution.
ds − bt at − cs
2. Proof. First let x = and y = .
ad − bc ad − bc
a(ds − bt) + b(at − cs) ads − bcs
Then ax + by = = =s
ad − bc ad − bc
c(ds − bt) + d(at − cs) −bct + adt
and cx + dy = = = t.
ad − bc ad − bc
 Hints and Solutions to Selected Exercises 373
( )
ds − bt at − cs
Hence (x, y) = , is a solution to the simultaneous
ad − bc ad − bc
equations.
For the uniqueness part, suppose that (x, y) and (x′ , y ′ ) are solutions to
the simultaneous equations. In other words,
ax + by = s (1) ax′ + by ′ = s (3)
and
cx + dy = t (2) cx′ + dy ′ = t (4).

Subtracting equation (3) from equation (1) and subtracting equation (4)
from equation (2) gives:

a(x − x′ ) + b(y − y ′ ) = 0 (5)

⇒ (ab − cd)(y − y ′ ) = 0.
c(x − x′ ) + d(y − y ′ ) = 0 (6)

Since ad − bc ̸= 0, it follows that y − y ′ = 0, so y = y ′ .

Then equations (5) and (6) simplify to a(x − x′ ) = 0 and c(x − x′ ) = 0
respectively. Now a and c cannot both be zero (as ad − bc ̸= 0), so at
least one of these equations now implies that x − x′ = 0 so x = x′ .
Hence (x, y) = (x′ , y ′ ), so the simultaneous equations have a unique
solution.

4. Proof. Note that if p = 3, then p2 + 2 = 11, which is also prime. Hence

there exists a prime p for which p2 + 2 is also prime.
Now suppose that p is prime and p ̸= 3. Then p is not divisible by 3, so
p = 3n + 1 or p = 3n + 2 for some n ∈ N.
If p = 3n+1, then p2 +2 = (3n+1)2 +2 = 9n2 +6n+3 = 3(3n2 +2n+1).
If p = 3n+2, then p2 +2 = (3n+2)2 +2 = 9n2 +12n+6 = 3(3n2 +4n+2).
In either case, p2 + 2 is divisible by 3 and therefore is not prime.
Hence p = 3 is the only prime for which p2 + 2 is also prime.

7. (ii) Suppose that gx = h and gy = h in G. Then gx = gy and ‘multiply’

this equation on the left by g −1 .

9. (i) Proof. Clearly x = 0 is such that |x| < ε for all ε > 0, ε ∈ R.
|x|
Suppose x ̸= 0. Then |x| > 0. Let ε = . Then ε > 0 and |x| > ε.
2
Hence x = 0 is the only real number that satisfies |x| < ε for all
ε ∈ R+ .
(ii) Proof. Let (an ) be a convergent sequence and suppose that
lim an = ℓ and lim an = ℓ′ .
n→∞ n→∞
Let ε > 0, ε ∈ R.
Then ε/2 > 0, so as lim an = ℓ, there exists N1 ∈ Z+ such that
n→∞
374 Understanding Mathematical Proof

n > N1 ⇒ |an − ℓ| < ε/2. Similarly, since lim an = ℓ′ , there exists

n→∞
N2 ∈ Z+ such that n > N1 ⇒ |an − ℓ′ | < ε/2.
For n > max{N1 , N2 } we have
ε ε
|ℓ′ − ℓ| = |(an − ℓ) − (an − ℓ′ )| ≤ |an − ℓ| + |an − ℓ′ | < + = ε.
2 2
Hence |ℓ − ℓ′ | < ε for all ε ∈ R+ . Therefore, by part (i), ℓ − ℓ′ = 0
so ℓ = ℓ′ .

Chapter 8

Exercises 8.1

1. (vi) Proof. The proof is by mathematical induction.

Base case. When n = 1, LHS = 1×6 = 6 and RHS = 13 ×1×2×9 =
6. Hence the result holds when n = 1.
Inductive step. Assume that 1 × 6 + 2 × 7 + 3 × 8 + · · · + k(k + 5) =
1
3 k(k + 1)(k + 8); this is the inductive hypothesis.
The next term on the left-hand side is (k + 1)(k + 6). Adding this
to both sides gives
1 × 6 + 2 × 7 + 3 × 8 + · · · + k(k + 5) + (k + 1)(k + 6)
= 13 k(k + 1)(k + 8) + (k + 1)(k + 6)
= 13 (k + 1) (k(k + 8) + 3(k + 6))
= 13 (k + 1)(k 2 + 11k + 18)
= 13 (k + 1)(k + 2)(k + 9).

This is the result for n = k + 1 and hence completes the inductive

step.
Therefore, for all n ∈ Z+ , 1 × 6 + 2 × 7 + 3 × 8 + · · · + n(n + 5) =
1
3 n(n + 1)(n + 8), by induction.
(viii) Proof. The proof is by mathematical induction.
Base case. When n = 1, LHS = 1/2! = 12 and RHS = 1−1/2! = 12 .
Hence the result holds when n = 1.
1 2 3 k−1 1
Inductive step. Assume that + + + ··· + = 1− ;
2! 3! 4! k! k!
this is the inductive hypothesis.
 Hints and Solutions to Selected Exercises 375
k
The next term on the left -hand side is . Adding this to
(k + 1)!
both sides gives

1 2 k−1 k 1 k
+ + ··· + + = 1− +
2! 3! k! (k + 1)! k! (k + 1)!
k+1 k
= 1− +
(k + 1)! (k + 1)!
1
= 1−
(k + 1)!

This is the result for n = k + 1 and hence completes the inductive

step.
1 2 3 n−1 1
Therefore, for all n ∈ Z+ , + + + ··· + = 1 − , by
2! 3! 4! n! n!
induction.

2. (iv) Proof. The proof is by mathematical induction.

Base case. When n = 1, 11n − 4n = 11 − 4 = 7, which is clearly
divisible by 7. Hence the result holds when n = 1.
Inductive step. Assume that 11k − 4k is divisible by 7; in other
words, 11k − 4k = 7a for some integer a. Hence 11k = 7a + 4k . Now

11k+1 − 4k+1 = 11 × 11k − 4 × 4k

= 11(7a + 4k ) − 4 × 4k (by inductive hypothesis)
= 7 × 11a + 7 × 4k
= 7(11a + 4k ) where 11a + 4k ∈ Z.

Therefore 11k+1 − 4k+1 is divisible by 7. This completes the induc-

tive step.
Therefore, for all n ∈ N, the expression 11n − 4n is divisible by 7,
by induction.

3. (i) Proof. The proof is by mathematical induction.

2!
Base case. When n = 1, LHS = 1 and RHS = = 1. Hence
21 × 1!
the result holds when n = 1.
(2k)!
Inductive step. Assume that 1 × 3 × 5 × · · · × (2k − 1) = k ; this
2 k!
is the inductive hypothesis.
The next term on the left -hand side is 2(k + 1) − 1 = 2k + 1.
376 Understanding Mathematical Proof

Multiplying both sides by 2k + 1 gives

1 × 3 × 5 × · · · × (2k − 1) × (2k + 1)
(2k)!
= × (2k + 1)
2k k!
(2k)! × (2k + 1) × (2k + 2)
=
2k k! × (2k + 2)
(2k + 2)!
= k
2 k! × 2(k + 1)
(2(k + 1))!
= k+1 .
2 (k + 1)!

This is the result for n = k + 1 and hence completes the inductive

step.
(2n)!
Therefore, for all n ∈ Z+ , 1 × 3 × 5 × · · · × (2n − 1) = n , by
2 n!
induction.
The following is an alternative, and simpler, direct proof.
Proof. Let n ∈ Z+ . Then

1 × 3 × 5 × · · · × (2n − 1)
1 × 2 × 3 × 4 × 5 × · · · × (2n − 1) × (2n)
=
2 × 4 × 6 × · · · × (2n)
(2n)!
= n
2 (1 × 2 × 3 × · · · × n)
(2n)!
= n .
2 n!

5. (i) Proof. The proof is by mathematical induction.

Base case. When n = 1, n3 − n = 0, which is divisible by 6. Hence
the result holds when n = 1.
Inductive step. Assume that k 3 −k is divisible by 6; then k 3 −k = 6a
for some integer a. Now

(k + 1)3 − (k + 1) = (k 3 + 3k 2 + 3k + 1) − (k + 1)
= k 3 − k + 3(k 2 + k)
= 6a + 3k(k + 1) (by inductive assumption).

Now k(k + 1) is the product of an even and an odd integer, so is

even — see theorem 1.2. Therefore 3k(k + 1) is divisible by 6.
 Hints and Solutions to Selected Exercises 377

Hence is (k + 1)3 − (k + 1) = 6a + k(k + 1) is divisible by 6. This

completes the inductive step.
Therefore, for all n ∈ Z+ , n3 − n is divisible by 6, by induction.
(ii) Proof. Let n ∈ Z+ . Then n3 − n = n(n2 − 1) = (n − 1)n(n + 1).
Therefore n3 − n is the product of three consecutive integers, one
of which is divisible by 3 and (at least) one of which is even.
Therefore n3 − n is divisible by 6.

8. For the inductive step, note that adding the (k + 1) st row at the bottom
of the large triangle adds 2k + 1 additional small triangles. Then use the
inductive hypothesis and (k + 1)2 = k 2 + (2k + 1).

11. (i) Proof. The proof is by mathematical induction.

Base case. When n = 1, the grid comprises a single L-shaped tile
so the result holds (trivially) when n = 1.
Inductive step. Assume that a 2k × 2k grid with a single corner
square removed may be covered with L-shaped tiles.
Consider a 2k+1 × 2k+1 grid with a single corner square removed.
Split the grid into four, as shown in the diagram below, and place
a single L-shaped tile in the middle as indicated.

The remaining areas in each of the four quarters is a 2k × 2k grid

with a single corner square removed. By the inductive hypothesis,
each of these quarters may be covered with L-shaped tiles. There-
fore the 2k+1 × 2k+1 grid with a single corner square removed may
also be covered with L-shaped tiles. This completes the inductive
step.
Therefore, for all n ∈ Z+ , a 2n × 2n grid with a single corner square
removed may be covered with L-shaped tiles, by induction.
(ii) The result follows by comparing areas. The area of the 2n × 2n grid
with a single corner square removed is 22n − 1. Each tile has area
3. Hence, if the grid may be covered with tiles of area 3, it follows
that 3 divides 22n − 1.
378 Understanding Mathematical Proof

(iii) Proof. The proof is by mathematical induction.

Base case. When n = 1, 22n − 1 = 22 − 1 = 3, which is divisible by
3. Hence the result holds when n = 1.
Inductive step. Assume that 22k − 1 is divisible by 3; then 22k − 1 =
3a for some integer a. Hence 22k = 3a + 1. Now

22(k+1) − 1 = 22 × 22k − 1
= 4(3a + 1) − 1 (by inductive assumption)
= 3 × 4a + 3
= 3(4a + 1) where 4a + 1 ∈ Z.

Hence 22(k+1) − 1 is divisible by 3. This completes the inductive

step.
Therefore, for all n ∈ Z+ , 22n −1 is divisible by 3, by induction.

13. (i) Proof. The proof is by mathematical induction.

Base case. When n = 1, LHS = h−1 gh = RHS. Hence the result
holds when n = 1.
( )k
Inductive step. Assume that h−1 gh = h−1 g k h. Then (assuming
the associative law throughout so that terms may be regrouped as
appropriate), we have
( )k+1 ( )( )k
h−1 gh = h−1 gh h−1 gh
( )( )
= h−1 gh h−1 g k h (by inductive assumption)
( )
= h−1 g hh−1 g k h
= h−1 gg k h (since hh−1 = e)
= h−1 g k+1 h.

This completes the inductive step.

( )n
Therefore, for all n ∈ Z+ and for all g, h ∈ G, h−1 gh = h−1 g n h,
by induction.

14. (ii) Proof. The proof is by mathematical induction.

( )
1 0
Base case. When n = 1, LHS = A = and
−1 2
( ) ( )
1 0 1 0
RHS = = .
1 − 21 21 −1 2
Hence the result holds when n = 1.
( )
1 0
Inductive step. Assume that Ak = .
1 − 2k 2k
 Hints and Solutions to Selected Exercises 379

Then

Ak+1 = AAk
( )( )
1 0 1 0
= (by inductive assumption)
−1 2 1 − 2k 2k
( )
1 0
=
−1 + 2(1 − 2k ) 2 × 2k
( )
1 0
= .
1 − 2k+1 2k+1

This completes the inductive step.

( )
1 0
Therefore, for all n ∈ Z , A =
+ n
, by induction.
1 − 2n 2n

Exercises 8.2

1. (ii) Proof. The proof is by mathematical induction.

Base case. For n = 7 we have 7! = 5040 > 2187 = 37 , so the result
holds for n = 7.
Inductive step. Assume that k! > 3k where k ≥ 7. Then

(k + 1)! = (k + 1) × k! > (k + 1) × 3k > 3k+1

since k +1 > 3. This is the result for n = k + 1 and hence completes

the inductive step.
Therefore, for all integers n ≥ 7, n! > 3n , by induction.

2. Proof. The proof is by mathematical induction.

3
Base case. When n = 2, LHS = 4 = RHS. Hence the result holds
when n = 2.

Inductive step. Assume that

3 8 15 k2 − 1 k+1
× × × ··· × = ,
4 9 16 k2 2k
where k ≥ 2; this is the inductive hypothesis.
(k + 1)2 − 1
The next term on the left-hand side is . Multiplying both
(k + 1)2
380 Understanding Mathematical Proof

sides by this gives

3 8 k 2 − 1 (k + 1)2 − 1 k + 1 (k + 1)2 − 1
× × ··· × × = ×
4 9 k2 (k + 1)2 2k (k + 1)2
2
k+1 k + 2k
= ×
2k (k + 1)2
k + 1 k(k + 2)
= ×
2k (k + 1)2
k+2
= .
2(k + 1)

This is the result for n = k + 1 and hence completes the inductive step.

n2 − 1 n+1
Therefore, for all n ∈ Z+ , 3
4 × 8
9 × 15
16 × ··· × = , by
n2 2n
induction.

4. (iii) Proof. The proof is by mathematical induction.

Base cases. Since an is defined explicitly for n = 1 and n = 2, there
are two base cases to consider.
When n = 1, a1 = 0 (by definition) and 2 × 3n − 3 × 2n = 2 × 3 −
3 × 2 = 0.
When n = 2, a2 = 6 (by definition) and 2 × 3n − 3 × 2n = 2 × 9 −
3 × 4 = 6.
Hence the result is true for n = 1 and n = 2.
Inductive step. Let k ≥ 2. Suppose that, for every integer 2 ≤ r ≤ k,
ar = 2 × 3r − 3 × 2r .
Consider ak+1 . Since k + 1 ≥ 3, ak+1 is defined by the recursive
relationship. So we have

ak+1 = 5ak − 6ak−1 (definition of ak+1 )

= 5(2 × 3 − 3 × 2 ) − 6(2 × 3k−1 − 3 × 2k−1 )
k k

(inductive assumption)
= 10 × 3k − 15 × 2k − 4 × 2k + 9 × 2k
= 6 × 3k − 6 × 2k
= 2 × 3k+1 − 3 × 2k+1 .

Hence ak+1 satisfies the given formula. This completes the inductive
step.
Therefore, an = 2 × 3n − 3 × 2n for all n ≥ 2, by induction.

5. (ii) Proof. The proof is by mathematical induction.

 Hints and Solutions to Selected Exercises 381

Base cases. When n = 1, LHS = a1 = 1 and RHS = a3 − 1 =

2 − 1 = 1.
When n = 2, LHS = a1 + a2 = 1 + 1 = 2 and RHS = a4 − 1 =
3 − 1 = 2.
Hence the result is true for n = 1 and n = 2.
Inductive step. Let k ≥ 2 and suppose that a1 + a2 + · · · + ak =
ak+2 − 1.
Adding ak+1 to both sides gives

a1 + a2 + · · · + ak + ak+1 = ak+2 − 1 + ak+1

= ak+3 − 1.

since ak+2 + ak+1 = ak+3 from the Fibonacci recurrence relation.

This completes the inductive step.
Therefore, a1 + a2 + · · · + an = an+2 − 1, by induction.
(iii) This example may serve as a caution to assuming that proof by
induction is always the best approach. Although a proof by induc-
tion is possible, it is complicated. The following is a much simpler
direct proof.
Proof. Let n ∈ Z+ . Then the Fibonacci recurrence relation gives
both an+3 = an+2 + an+1 and an+2 = an+1 + an . Using these we
have:
a2n+2 − a2n+1 = (an+2 + an+1 )(an+2 − an+1 )
= an+3 (an+1 + an − an+1 )
= an+3 an .

6. Proof. The proof is by mathematical induction.

Base case. When n = 1, 1 = 20 can be expressed as a ‘sum’ of distinct
non-negative powers of 2. Hence the result is true for n = 1.
Inductive step. Suppose that, for all integers 1 ≤ r ≤ k, r can be
expressed as a sum of distinct non-negative powers of 2.
Consider k + 1. We consider two cases: k + 1 is odd or k + 1 is even.
If k + 1 is odd, then k is even. By the inductive hypothesis, k can
be expressed as a sum of distinct non-negative powers of 2. Since k
is even, this expression does not include 20 = 1. Hence adding 20 to
the expression for k gives an expression for k + 1 as a sum of distinct
non-negative powers of 2.
If k + 1 is even, then k + 1 = 2r where 1 ≤ r ≤ k. By the inductive
hypothesis, r can be expressed as a sum of distinct non-negative powers
382 Understanding Mathematical Proof

of 2. Multiplying this expression for r by 2 gives an expression for k + 1

as a sum of distinct non-negative powers of 2.
In either case, k + 1 has an expression as a sum of distinct non-negative
powers of 2. This completes the inductive step.

Therefore every n ∈ Z+ can be expressed as a sum of distinct non-

negative powers of 2, by induction.

8. (i) Proof. Let a, b ∈ A. Then a = 4m + 1 and b = 4n + 1 for some

integers m and n. Hence ab = (4m + 1)(4n + 1) = 16mn + 4m +
4n + 1 = 4(4mn + m + n) + 1 where 4mn + m + n ∈ Z. Therefore
ab ∈ A.
(ii) Proof. The proof is by mathematical induction.
Base case. When n = 1, 5 = 4 × 1 + 1 is A-prime and so is a
‘product’ of A-primes. Hence the result is true for n = 1.
Inductive step. Suppose that, for all integers 1 ≤ r ≤ k, 4r + 1 can
be expressed as a product of A-primes.
Consider a = 4(k + 1) + 1.
If a is A-prime, then it is trivially expressed as a ‘product’ of A-
primes.
Otherwise a has a factorisation a = a1 a2 where both a1 and a2 are
elements of A. By the inductive hypothesis, a1 and a2 can each be
written as a product of A-primes

a1 = p1 p2 . . . ps and a2 = q1 q2 . . . qt

where p1 , p2 , . . . , ps and q1 , q2 , . . . , qt are all A-primes. Therefore a

can be expressed as a product of A-primes

a = a1 a2 = p1 p2 . . . ps q1 q2 . . . qt .

This completes the inductive step.

Therefore every element of A that is greater than 1 can be written
as a product of A-primes, by induction.
(iii) 693 ∈ A has two distinct A-prime factorisations, 693 = 9 × 77 and
693 = 21 × 33.

9. Proof. The proof is by mathematical induction on |A|.

Base case. When n = 0, |A| = 0 implies that A = ∅, which has no

proper subsets. Hence the result is (trivially) true for n = 0.3
3 Readers who don’t like this argument may wish to start the induction at n = 1. If

|A| = 1 and B ⊂ A it follows that B = ∅, so |B| = 0 < 1 = |A| and the result holds for
n = 1.
 Hints and Solutions to Selected Exercises 383

Inductive step. Suppose that, for all sets A with |A| = k, if B ⊂ A, then
|B| < |A|.
Let A be a set with cardinality |A| = k + 1 and let B ⊂ A.
If B = ∅, then |B| = 0 < k + 1 = |A|.
Otherwise choose an element b ∈ B. Then B ′ = B−{b} is a proper subset
of A′ = A − {b}, B ′ ⊂ A′ . But |A′ | = |A| − 1 = k so, by the inductive
hypothesis, |B ′ | < |A′ |. Therefore |B| = |B ′ | + 1 < |A′ | + 1 = |A|. This
completes the inductive step.
Therefore, for all finite sets A and B, if B ⊂ A, then |B| < |A|, by
induction.
Bibliography

[1] A. Cupillari. The Nuts and Bolts of Proof, 4 th edition. Elsevier Academic
Press, 2013.
[2] P.J. Eccles. An Introduction to Mathematical Reasoning: Numbers, Sets
and Functions. Cambridge University Press, 1997.
[3] J. Fauvell and J. Gray. The History of Mathematics: A Reader. Palgrave
Macmillan, 1987.
[4] J. Franklin and A. Daoud. Proof in Mathematics: An Introduction. Key
Books, 2010.
[5] R. Garnier and J. Taylor. 100% Mathematical Proof. John Wiley & Sons,
1996.
[6] R. Garnier and J. Taylor. Discrete Mathematics: Proofs, Structures and
Applications, 3 rd edition. Taylor and Francis, 2010.
[7] G.H. Hammack. Book of Proof, 2 nd edition. Creative Commons, 2013.
[8] G.H. Hardy. Mathematical proof. Mind, pages 1–25, 1929.
[9] G.H. Hardy. A Mathematician’s Apology (reprinted edition). Cambridge
University Press, 2012.
[10] S.-J. Shin. The Logical Status of Diagrams. Cambridge University Press,
1994.
[11] A.B. Slomson. Mathematical proof and its role in the classroom. Math-
ematics Teaching, 155:10–13, 1996.
[12] D. Solow. How to Read and Do Proofs: An Introduction to Mathematical
Thought Processes, 5 th edition. John Wiley & Sons, 2010.
[13] D. Tall. Thinking through three worlds of mathematics. Proceedings
of the 28th Conference of the International Group for the Psychology of
Mathematics Education, Bergen, Norway, pages 281–288, 2004.
[14] D.J. Vellerman. How to Prove It: A Structures Approach, 2 nd edition.
Cambridge University Press, 2006.

385
Index

Abel, Niels, 202 Biconditional connective, 23

Abelian group, 202, 208, 211* 294 Biconditional law, 41* 246
Absorption law Bijection, see bijective function
propositions, 32 Bijective function, 121, 198
sets, 9Q Binary operation, 201^ 202, 204
Adding assumptions deduction rule, associative property, 201
170 Cayley table, 204
Addition deduction rule, 68 closure property, 201
Argument c o m m u t a t i v e property, 202
predicatel logic, Ih identity element, 201
propositional logic, 2 1 inverse of an element, 201
Aristotle, 162 Binary word, 153, 165
Arithmetic, Fundamental Theorem, Binomial coefficient, 190, 284
132*299 Bolyai, J a n o s , 163, 167
Associative law Bounded
propositions, 39 function, 295
sets, 9Q set, 264
Associative property
addition, 138 Cancellation property
binary operation, 201 in a group, 206
m a t r i x multiplication, 298 real n u m b e r s , 206
Axiom of Induction, 296, 304 Cantor, Georg, 281
Axiom system, 165, 166, 168-171 Cardinality, 86, 189, 197, 198, 2 8 1
deduction rule(s), 169-171 Cartesian p r o d u c t , 103, 105, 199, 256
interpretation, 166 Cayley table, 204
model, 166 Cayley, A r t h u r , 204
proof, 168 Closed interval, 90
theorem, 168 Closure
Axiom(s), 4 , 1 6 1 * 163-165, 304 binary operation, 201
consistent, 164 subspace of a vector space, 218
induction, 304 Codomain, of a function, 108
parallel, 163 C o m m u t a t i v e law
P e a n o , 304 propositions, 3 9
Axiomatic m e t h o d , 161 sets, M
Commutative property
Background knowledge, 139, 141 addition, 138
Ball, in a metric space, 157 binary operation, 202

387
388

Complement law Coprime integers, 182

propositions, 32 Corollary, 5 , 1 9 8
sets, 2 2 Counting a r g u m e n t s , 280
Complement, of a set, Qh^ 252 Cyclic group, 276
Completeness
propositional logic, 69 De Morgan's law
Completeness axiom propositions, 37, 39, 241
for real n u m b e r s , 296 sets, 22^ 315
Complex n u m b e r , 2Q Deduction
Composite integer, 134, 242 in propositional logic, 69
Composition Deduction rule(s)
of functions, 113, 144, 202^ 244 adding assumptions, 170
of symmetries, 203 axiom systems, 169-171
C o m p o u n d proposition, IE biconditional elimination, 2 9
t r u t h table, 2 1 biconditional introduction, 29
Conclusion, 4 case analysis, 29^ 171
of a deduction, 6 2 conditional, 169
of an argument, 2 1 conjunction, 169
Conditional connective, 20. equivalence, 170
contrapositive, 4 1 , 117, 240 m o d u s ponens, 169
converse, 2 3 m o d u s tollens, 174
t r u t h table, 2 1 predicate logic, 75
Conditional deduction rule, 169 proof by cases, 170
Conjecture, 5+ 180 propositional logic, 62> 167, 168,
Goldbach, h^ IB. 121
Conjunction, 1&, 2 1 Deductive m e t h o d , 138, 141
t r u t h table, IE Deductive reasoning, E^. 138
Conjunction deduction rule, 68, 169, Defmition(s), 4 ^ 1 6 2
252 D e t e r m i n a n t , 297
Conjuncts, IE Difference, of sets, 2 5
Connective, 18-20, 2 3 Direct proof, 141, 142, 1 4 4 , 1 2 5
Constructive dilemma, QE Disc, 15Z
Constructive proof, 270 Discrete dynamical systems, 323
Contradiction, 32 Discrete metric, 160, 161
proof by, 257 Disjoint sets, 9 5 , 252
Contradiction law, 4J^ 257 Disjunction, 19, 20, 94
Contrapositive, 4J^ 117, 240 exclusive, 22
proof using, 240 inclusive, 19
Contrapositive law, 4H, 240 t r u t h table, 1 2 , 2Q
Convergent sequence(s), 222 Disjunctive syllogism, QE
sum of, 231 Disjuncts, 1 2
Converse, 23 Distributive law
Convex propositions, 3 2
polygon, 321 sets, 22
subset ofM 2 or R 3 , 330 Distributive property
 389

integers, 139 floor, 230

Divides, 182 formal definition, 112
Division Algorithm, 282, 3 0 1 function machine, 108
Domain, of a function, 108 horizontal line test, 121
image, 108
Eigenspace, 221, 316 increasing, 251, 293
Eigenvalue, 220, 316 informal definition, 107
Eigenvector, 220 injective, 117, 144, 194, 244, 275,
Element, of a set, 83 290
E m p t y set, 85 integer p a r t , 230
E n u m e r a t i o n theory, 280 inverse, 121
Equality, of sets, 8 5 , 8 ^ 253 limit, 232
Equivalence deduction rule, 170 modulus, 146
Error correcting codes, 165 odd, 200
Euclid, 1 6 1 , 1£2^ 258 one-one, see injective
Euclidean geometry, 163, 322 onto, see surjective
Euclidean metric, 152 propositional, 46
Euler's phi function, 182 square, 112
Euler, Leonhard, 92^ 287, 292 surjective, 117, 144, 194, 197,
Existence proof, 270 244, 275, 290
constructive, 270 F u n d a m e n t a l T h e o r e m of A r i t h m e t i c ,
infinitely m a n y primes, 258 132, 299
irrational n u m b e r s , 271
non-constructive, 276 Godel, K u r t , 1QA
using counting a r g u m e n t s , 280 Galois, Evariste, 200
Existential quantifier, I S Gauss, Carl Friedrich, 163
Generalised pigeonhole principle, 285
Factor, 182 Generator, in a group, 276
Fermat n u m b e r s , 287 Geometric progression, 242, 310
F e r m a t ' s Last T h e o r e m , h Goldbach Conjecture, ^ lfi
Fermat, Pierre de, 5 Greatest common divisor, 182
Fibonacci sequence, 324 Greatest lower b o u n d , 264, 268, 3£&
Finite set, SA G r o u p ( s ) , 201^ 202, 206, 207, 262, 276,
Floor function, 230 294,298
Function, 1 0 7 , 1 0 8 , 1 1 2 , 1 1 3 , 117, 121, Abelian, 202, 2Q&, 2 1 1 , 276, 294
194, 197, 198, 200, 203, 213, cancellation property, 206
232^ 244, 2 5 1 , 275, 290, 293, cyclic, 276
295 generator, 276
bijective, 1 2 1 , 1 9 8 Klein 4-group, 205, 2Q$>
b o u n d e d , 295 morphism, 2£2
codomain, 108 Shoes and Socks theorem, 207
composition, 113, 144, 203, 244 subgroup, 225^ 276, 294
decreasing, 251, 293 uniqueness of inverses, 298
domain, 108
even, 200 Highest common factor, 182
390

Horizontal line test, 121 Involution law

Hypotheses, 4 propositions, 3 9
Hypothetical syllogism, £ 8 sets, 99
Irrational n u m b e r ( s ) , 271
I d e m p o t e n t law denseness in t h e reals, 273
propositions, 39 existence, 271
sets, 99 Isomorphic, model, 167
Identity element, for a binary opera­
tion, 2 0 1 Kernel, 226, 263
Identity law of a group morphism, 263
propositions, 39 of a linear transformation, 226
sets, 9 9 Klein 4-group, 205, 208
Identity m a t r i x , 225 Klein, Felix, 205
Image
of a function, 108 Law
of a linear transformation, 226 logical equivalences, 39
of an element, 108 set theory, 9 9
Inconsistent premises, 2 4 Least u p p e r b o u n d , 264
Indirect existence proof, 276 L e m m a , 5 , 193
Inductive reasoning, 2 , 138 Limit
Infimum, 264, 268, 368 of a function, 232
Infinite cardinality, 86 of a sequence, 222
Infinite set, 84 Linear transformation, 226, 262
Injection, see injective function image, 226
Injective function, 1 1 2 , 144, 194, 2 4 4 , kernel, 226
275, 290 Lobachevsky, Nikolai, 163, 167
horizontal line test, 121 Logical connective, 18-20, 2 3
Integer p a r t function, 230 biconditional, 2 3
Integer(s), 90, 138, I M i 2 0 2 conditional, 20
as a group, 202 conjunction, 1 8
associative property, 138 disjunction, I S
c o m m u t a t i v e property, 138 exclusive disjunction, 29
distributive property, 139 inclusive disjunction, 19
Intermediate value theorem, 286 Logical equivalence, 35, 3 9 - 4 1 , 44, 6 3 ,
Interpretation, of an axiom system, 240, 241, 246, 257
166 absorption laws, 39
Intersection associative laws, 3 9
of sets, 9 3 , 94 biconditional law, 4 1 , 246
of subgroups, 225 c o m m u t a t i v e laws, 39
of subspaces, 219 complement laws, 3 9
Interval, 9 0 , 263 contradiction law, 41> 257
Inverse contrapositive law, 40^ 240
of a m a t r i x , 297 De Morgan's laws, 39, 241
of an element in a group, 201 distributive laws, 3 9
Inverse function, 121 idempotent laws, 39
 391

identity laws, 39 P e a n o axioms, 296

involution law, 39 Necessary condition, 2 2 , 2 2 , 247
laws, 3 9 - 4 1 , 240, 246, 257 Negation, 12
material implication, 44, S3 of a quantified proposition, 5 2 , 6 3
Logical implication, 4 2 , 6 6 t r u t h table, IE.
Lower b o u n d , 264 Non-constructive existence proof, 276
Lucas, E d o u a r d , 184 Non-Euclidean geometry, 164, 167
Null set, E5.
M a n h a t t a n metric, 153, 155 Null space, of a m a t r i x , 226
Mapping, 1 0 7 , 1 1 2
Material implication, 44, 63 One-one function, see injective func­
Matrix, 212, 219-221, 225, 226, 2ZU tion
2 9 4 297,216 One-place predicate, 55.
symmetric, 294 O n t o function, see surjective function
transpose, 294 O p e n interval, £0.
d e t e r m i n a n t , 297 Ordered n-tuple, 105
eigenspace, 221, 316 Ordered pair, 103
eigenvalue, 220, 316
eigenvector, 220 Parallel axiom, 163, 167
identity m a t r i x , 225 Peano axioms, 296, 304
inverse, 297 Peano, Giuseppe, 304
multiplication, 271, 298 Pigeon hole principle, 282
null space, 226 Pigeonhole principle
Mersenne primes, 277, 288 generalised, 285
M e t h o d , of direct proof, 144 Polygon, 277, 321
Metric, 152, 155 convex, 284
discrete, 160, 161 Power set, 99, 1 8 9 , 1 2 2
Euclidean, 152 Predicate, 45^ 85
M a n h a t t a n , 153, 155 one-place, 45^ 55
on binary words, 153, 165 relational, 56
sphere, 153 two-place, 56
Metric space, 152, 155 Predicate logic
discrete, 161 a r g u m e n t , 25.
triangle inequality, 155 deduction rules, 15.
Model, of an axiom system, 166 Premise(s)
isomorphic, 167 inconsistent, 74
Modulo arithmetic, 275 of a deduction, 69
Modulus function, 146 of an argument, 7 1
triangle inequality, 146, 231, 234 P r i m e factorisation theorem, 132,
M o d u s ponens, 68, 169 162, 318, 32Q
M o d u s tollens, 68, 174 P r i m e n u m b e r ( s ) , 4^ 133, 162, 242,
Morphism, 262 270
Mersenne primes, 277
N a t u r a l language, 21 Primitive t e r m , 162
N a t u r a l n u m b e r ( s ) , £9 Principle
392

of M a t h e m a t i c a l Induction, 305 existential, 49

universal generalisation, 137, 141, introduction rules, 2 5
145^241 negation, 52^ £ 3
Proof universal, 49
by construction, 270
by contradiction, 257 Rational n u m b e r ( s ) , 90, 203
by M a t h e m a t i c a l Induction, 305 Real n u m b e r ( s ) , 90^ 203
change of base case, 316 cancellation property, 206
Strong form, 319 completeness axiom, 296
deductive m e t h o d , 138 Reasoning, ^ 129, 138
direct, 1 4 1 , 1 4 2 , 1 4 4 deductive, 8
in an axiom system, 168 inductive, 2
two phase, 196 Recursively defined sequences, 324
using contrapositive, 240 Relational predicate, 5fi
using counting a r g u m e n t s , 280 Resolution deduction rule, 68
working backwards, 178
Proof by cases deduction rule, 170 Scalar multiplication, 211
Proof by M a t h e m a t i c a l Induction Sequence (s), 222+324
Strong form, 326 chaotic, 227
P r o p e r subgroup, 225, 276 convergent, 227
P r o p e r subset, R2 defined recursively, 324
Proposition, 16» 18» 32+ 33, 35, 42, M diagram, 222
c o m p o u n d , 18 Fibonacci, 324
contradiction, 32 periodic, 222
deducible from, £9 Set theory laws, 99
logical equivalent, 3 5 De Morgan's law, 315
logical implication, 42 Set(s), 83-88, 93-95, 99, 102, 103^
simple, 18 105, 189, 192, 197-199, 223^
substitution instance, 33 252, 253, 255, 256, 263, 264,
tautology, 3 2 281
Propositional function, 4£ bounded, 2 M
quantified, 48 b o u n d e d above, 263
two variable, 56 b o u n d e d below, 264
Propositional logic cardinality, 86, 189,197, 198, 2 8 1
argument, 71 Cartesian p r o d u c t , 103, 105, 199,
completeness, 69 256
deduction, 69 complement, 95, 252
deduction rules, 67, 1£2+ 168. 171 defined using predicates, 85
soundness, £9 defining sets, 8 1
P u n c t u r e d neighbourhood, 233 difference, 95.
disjoint, 95, 252
Quantified propositional function, 48 e m p t y set, 85
negation, fi3 equality, 8 5 , 8&+ 253
Quantifier, 48. finite, £4
elimination rules, Z5. infinite, 8 1
 393

intersection, 9 3 , 94 Tautology, 32
power set, 99, 1 8 9 , 1 0 2 T h e o r e m , 5 , 166-168
solution set, 8 5 denseness of rational n u m b e r s ,
subset, 8 1 , 2 2 3 273
superset, 8 1 Euclid's theorem on primes, 258
symmetric difference, 102, 199, existence of irrationals, 281
255, 256 in an axiom system, 167, 168
union, M , 95 Intermediate value, 286
universe, 86 prime factorisation, 132, 162,
Shoes and socks theorem 318, 32Q
for groups, 207 Shoes and Socks, 207
Simple proposition, 1 8 Totient function, see Euler's phi func­
Simplification deduction rule, OR tion
S m a r a n d a c h e function, 184 Transitive property of subset, 193
Smarandache, Florentin, 184 Transpose, of a m a t r i x , 294
Soundness, of propositional logic, 6£ Triangle inequality
Sphere, 157 metric space, 155
Sphere metric, 153 m o d u l u s function, 146, 231, 234
Statement, 3 T r u t h table, 1 7 - 2 1 , 23-26
Subgroup, 225, 276, 294 biconditional, 2 3
intersection, 225 conditional, 2 1
Subset, 87, 223 conjunction, 18
counting theorem, 280 disjunction, 1£), 2Q
transitive property, 193 negation, IS.
Subspace(s) of a vector space, 215, T r u t h value, lfi
260 Two-place predicate, 56
direct sum, 251
intersection, 219 Undefined t e r m , 162
sum, 225, 261 Union, of sets, 94, 95
test for, 218, 2m Universal generalisation
Substitution instance, 3 2 principle, 137, 141^ 145, 241
Sufficient condition, 2 2 , 2 3 , 247 Universal quantifier, 49
Superset, 87 Universal set, 86
S u p r e m u m , 264^ 300 Universe of discourse, 54
Surjection, see surjective function Upper b o u n d , 263
Surjective function, 117, 144, 194,
197, 244, 275, 290 Valid argument
horizontal line test, 121 in predicate logic, 1 5
Symmetric difference, of sets, 102, in propositional logic, 7 1
255^256 Vector space(s), 2 1 1 , 215, 218, 219,
Symmetric m a t r i x , 294 225, 251, 260, 261, 2 9 1
Symmetries, 203 subspace, 215, 260, 294
composition, 203 direct sum, 251
of a rectangle, 203 intersection, 219
of an equilateral triangle, 205 sum, 2 2 5 , 261
394

test for, 218, 2fiQ

Vector(s), 2 1 1
i n R 2 , 209
addition, 210
m a g n i t u d e , 209
scalar multiplication, 210
Venn, J o h n , 92
Venn-Euler diagram, 92-95, 2 M
complement, 95
disjoint sets, 9 5
intersection, M
set difference, 95
subset, 9 3
union, QA
 Mathematics

UNDERSTANDING MATHEMATICAL PROOF

The notion of proof is central to mathematics yet it is one of the most difficult
aspects of the subject to master. Understanding Mathematical Proof
describes the nature of mathematical proof, explores the various techniques
that mathematicians adopt to prove their results, and offers advice and
strategies for constructing proofs. It will improve your ability to understand
proofs and construct correct proofs of your own. UNDERSTANDING
The first chapter of the text introduces the kind of reasoning that mathemati-
cians use when writing their proofs and gives some example proofs to set
the scene. The book then describes basic logic to enable an understand-
ing of the structure of both individual mathematical statements and whole
MATHEMATICAL
mathematical proofs. It also explains the notions of sets and functions and
dissects several proofs with a view to exposing some of the underlying fea-
tures common to most mathematical proofs. The remainder of the book
PROOF
delves further into different types of proof, including direct proof, proof using
contrapositive, proof by contradiction, and mathematical induction. The au-
thors also discuss existence and uniqueness proofs and the role of counter
examples.

Features
• Presents a rigorous yet accessible introduction to one of the defining
concepts of mathematics
• Explores the structure of mathematical proof
• Classifies different types of proof
• Develops all the standard techniques for proving mathematical
theorems
• Provides hints and guidance on finding the appropriate proof
• Includes many elementary examples and, in addition, develops
proofs in linear algebra, group theory, and real analysis that will be of
Taylor • Garnier John Taylor
particular interest to undergraduate mathematicians
Rowan Garnier

K15018

K15018_Cover.indd 1 2/4/14 10:37 AM