MA102

Mathematical Proof and Analysis

and

MA103
Introduction to Abstract Mathematics

Michaelmas Term

Lecture Notes

These lecture notes were compiled by Peter Allen in 2022. However much of the content
comes from Martin Anthony, Graham Brightwell, Michele Harvey, Jan van den Heuvel and Amol
Sasane.
 Contents

1 Introduction 6
1.1 What is this course about? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.1 How to get the most out of this course (and all the other maths courses) 9
1.1.2 Topics covered (MA102, first half of MA103) . . . . . . . . . . . . . . . . . 11
1.2 Moodle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3 Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4 Activities and sample exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Mathematical statements, proof, and logic 13

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Mathematical statements and proof . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 Examples of Mathematical Statements . . . . . . . . . . . . . . . . . . . . 14
2.2.2 Introduction to proving statements . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Some basic logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 Negation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.2 Conjunction and disjunction . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3.3 If-then statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.4 If and only if statements; logical equivalence . . . . . . . . . . . . . . . . . 23
2.4 Implications and associated statements . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4.1 Converse statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4.2 Contrapositive statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4.3 Converse of the contrapositive . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5 What is a proof? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.6 How to prove it . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.7 What is not a proof? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.8 Sample exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.9 Comments on selected activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.10 Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

3 Sets and quantifiers 33

3.1 Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1.2 A note on notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.1.3 Set equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.1.4 Subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2
 3

3.1.5 Unions and intersections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.1.6 Arbitrary unions and intersections . . . . . . . . . . . . . . . . . . . . . . . 37
3.1.7 Universal sets and complements . . . . . . . . . . . . . . . . . . . . . . . . 37
3.1.8 Sets and logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.1.9 Cartesian products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.1.10 Power sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2 Quantifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2.1 Quantifiers and arbitrary unions and intersections; empty sets . . . . . . 40
3.3 Proof by contradiction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4 Some terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.5 General advice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.5.2 Definition chasing and cases . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.5.3 How to write mathematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.5.4 How to do mathematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.5.5 How to become better in mathematics . . . . . . . . . . . . . . . . . . . . . 49
3.6 Non-examinable: set theory—take 2 . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.7 Sample exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.8 Comments on selected activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.9 Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4 Structures, natural numbers and proof by induction 53

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2 Mathematical structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2.1 Greatest and least elements . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.3 The principle of induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.3.1 Proof by induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.3.2 An example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.3.3 Induction: why be careful? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3.4 Variants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.4 Summation formulae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.5 Recursively defined sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.6 Sample exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.7 Comments on selected activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.8 Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

5 Functions and counting 68

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.2 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.2.1 Basic definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.2.2 Function equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.2.3 Composition of functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.3 Bijections, surjections and injections . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.4 Inverse functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.4.1 Definition, and existence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.4.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.5 Functions on sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.6 Counting as a bijection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.7 The pigeonhole principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.7.1 The principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
 4

5.7.2 What will be on the exam? . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

5.7.3 Some applications of the Pigeonhole Principle . . . . . . . . . . . . . . . . 81
5.8 A generalised form of PP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.9 Infinite sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.10 Sample exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.11 Comments on selected activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.12 Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

6 Equivalence relations and the rational numbers 87

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.2 Equivalence relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.2.1 Relations in general . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.2.2 The special properties of equivalence relations . . . . . . . . . . . . . . . . 88
6.3 Equivalence classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.3.1 What’s the point? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.4 Rational numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.4.1 An important equivalence relation . . . . . . . . . . . . . . . . . . . . . . . 92
6.4.2 Rational numbers as equivalence classes . . . . . . . . . . . . . . . . . . . . 93
6.4.3 Non-examinable: Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
6.5 Sample exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.6 Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

7 Real and complex numbers 97

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
7.2 Rational numbers and real numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
7.2.1 Non-examinable: what are the real numbers exactly? . . . . . . . . . . . . 98
7.2.2 Real numbers: a ‘sketchy’ introduction . . . . . . . . . . . . . . . . . . . . 99
7.2.3 Rationality and repeating patterns . . . . . . . . . . . . . . . . . . . . . . . 101
7.2.4 Irrational numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.2.5 ‘Density’ of the rational numbers . . . . . . . . . . . . . . . . . . . . . . . . 104
7.3 Complex numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.3.2 Complex numbers: a formal approach . . . . . . . . . . . . . . . . . . . . . 105
7.3.3 Complex numbers: a more usual approach . . . . . . . . . . . . . . . . . . 105
7.3.4 Roots of polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.3.5 The complex plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
7.3.6 Polar form of z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.3.7 Exponential form of z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
7.4 Sample exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.5 Comments on selected activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.6 Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

8 Analysis: the real numbers 115

8.1 What is analysis? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
8.1.1 Analysis and calculus: the history . . . . . . . . . . . . . . . . . . . . . . . 115
8.1.2 Greek letters, conventions and inexact calculation . . . . . . . . . . . . . . 117
8.2 The real numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
8.2.1 Bounded sets; least upper bound . . . . . . . . . . . . . . . . . . . . . . . . 119
8.2.2 The least upper bound property . . . . . . . . . . . . . . . . . . . . . . . . 121
8.2.3 The Archimedean property of the real numbers . . . . . . . . . . . . . . . 123
8.2.4 Intervals and absolute values . . . . . . . . . . . . . . . . . . . . . . . . . . 125
 5

8.3 Sample exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

8.4 Comments on selected activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
8.5 Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

9 Analysis: Sequences and limits 132

9.1 Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
9.2 Limit of a convergent sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
9.2.1 Proving convergence of a sequence . . . . . . . . . . . . . . . . . . . . . . . 137
9.2.2 Writing analysis proofs: formal versus informal . . . . . . . . . . . . . . . 144
9.3 Bernoulli’s inequality and the sequence (xn )n∈N . . . . . . . . . . . . . . . . . . . . 147
9.4 Bounded and monotone sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
9.4.1 Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
9.5 Algebra of limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
9.6 The Sandwich Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
9.7 Subsequences and the Bolzano-Weierstrass theorem . . . . . . . . . . . . . . . . . 162
9.7.1 Non-examinable: another proof of Bolzano-Weierstrass . . . . . . . . . . . 165
9.8 Sample exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
9.9 Comments on selected activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
9.10 Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

10 Analysis: Continuity 174

10.1 Definition of continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
10.2 Continuous functions preserve convergent sequences . . . . . . . . . . . . . . . . . 178
10.3 Restrictions and compositions of functions . . . . . . . . . . . . . . . . . . . . . . . 181
10.4 Intermediate Value theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
10.5 Extreme Value theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
10.6 Sample exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
10.7 Comments on selected activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
10.8 Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
 1
Introduction

This chapter is intended to tell you what ‘abstract mathematics’ and ‘proof’ mean, and why
you should care about studying them.

1.1 What is this course about?

There are two main concepts in this course, and they are the two main concepts you will learn,
use, and re-use throughout your degree. After you have finished your degree, you might never
again use some of the mathematics you learn: but the ways of thinking which you will be shown,
will practice, and will steadily improve through your time here will stay with you. These ways of
thinking are what in the end prepare you for your future career. These concepts are abstraction
and proof.
You probably saw before at least some idea of what a mathematical proof is (but it is fine if
you did not—we will cover it!), and you probably do not know what ‘abstraction’ should be
(which is also fine). So I will begin by giving an example of abstraction which you met long ago.
Choose a number, multiply it by itself, then add your chosen number four times, and finally
add four. For example:

1 × 1 + 1 + 1 + 1 + 1 + 4 = 9 = 3 × 3 = (1 + 2) × (1 + 2)
2 × 2 + 2 + 2 + 2 + 2 + 4 = 16 = 4 × 4 = (2 + 2) × (2 + 2)
3 × 3 + 3 + 3 + 3 + 3 + 4 = 25 = 5 × 5 = (3 + 2) × (3 + 2) and so on . . .

These are concrete examples. You probably see that there is a pattern to the answers we get.
We can write it more generally:

x × x + 4 × x + 4 = (x + 2) × (x + 2) .

This is a mathematical statement. It’s something which is either true or false (depending on
what x is). It means the same as the following English:
Choose a number, multiply it by itself, then add your chosen number four times, and finally
add four. You will get the same answer as if you add two to your chosen number to get a new
number, then multiply the new number by itself.
Writing x in an equation, rather than ‘your chosen number’ in an English phrase, is an
example of a (simple) abstraction. Here the purpose is to simplify the presentation. There is
no need to write equations with xs in them; you could do it all in words—and indeed long ago
that is what people did. Of course, it’s hard to get anything done like that. If you show the

6
CHAPTER 1. INTRODUCTION 7

equation to a small child, it won’t mean anything to them, while they can read and understand
the sentence. But once you understand what the symbols in the equation mean, then it’s much
quicker and easier to read or write.
Now we come to proof. Is the statement above (however it’s written) true for some other
values of x than the three we checked by calculation? And if so, why? The purpose of a proof is
not just to be certain that a statement is true. It also explains why a statement is true. As you
probably know, the statement we wrote is true for all integers. Here is a proof.
Proof.

(x + 2) × (x + 2)
=(x + 2) × x + (x + 2) × 2 (multiplication distributes over addition)
=x × x + 2 × x + (x + 2) × 2 (multiplication distributes over addition)
=x × x + 2 × x + x × 2 + 2 × 2 (multiplication distributes over addition)
=x × x + 2 × x + x × 2 + 4 (2 × 2 = 4)
=x × x + 2 × x + 2 × x + 4 (multiplication is commutative)
=x × x + (2 + 2) × x + 4 (multiplication distributes over addition)
=x × x + 4 × x + 4 (2 + 2 = 4)

We can see that each line is equal to the previous one, for any integer x, because of the reason
given on the right. Most of the reasons are axioms—statements which we are assuming to be
true—and a couple are little calculations which you should check. So in particular the first and
last lines are equal for any integer x, in other words the statement

(x + 2) × (x + 2) = x × x + 4 × x + 4

is true for any integer x. That’s what we wanted to prove.

Of course, you will never want to write down a proof in this kind of detail. You would much
rather write at most a couple of lines of algebra expanding out the brackets, just as you would
have done in school, or simply write ‘it is obvious that (x + 2) × (x + 2) = x × x + 4 × x + 4’. This
is fine (unless a question explicitly asks you to prove something from axioms). You just need to
be aware that when you write ‘it is obvious...’ that you are promising that if someone really
wants to see the details, you would be able to write out the details as above.
Let’s go back to abstraction. These axioms we wrote down above (multiplication distributes
over addition, multiplication is commutative) are statements which you presumably agree are
true for the integers. Of course, they are also true for other numbers—they are true for real
numbers, or complex numbers. That means that the proof we wrote down works equally well for
real numbers, or complex numbers. So you know, for instance, that

(4.5 + 6i + 2) × (4.5 + 6i + 2) = (4.5 + 6i) × (4.5 + 6i) + 4 × (4.5 + 6i) + 4

is a true statement. This is a second reason abstraction is important: it is a time- and memory-
saving device. You can prove something once—or remember one fact—in an abstract setting
and use it in many different concrete examples.
Later on, you will see examples of mathematical structures which are not just numbers. For
some of these structures, the two axioms we mentioned above will be true, and (if you can find a
reasonable way of saying what ‘2’ and ‘4’ are!) the above proof still works. For other structures,
one or both of these axioms might not be true, so the proof will not work. That doesn’t mean
the statement is automatically false, but at least you should be suspicious.
CHAPTER 1. INTRODUCTION 8

Actually, you probably already know an example (or, at least, by the time you come to
revision you will know it). We can look at 2-by-2 matrices. Here, it’s reasonable to say that ‘2’
2 0 4 0
should mean the matrix ( ), and ‘4’ should be ( ). Assuming you know how to add and
0 2 0 4
multiply 2-by-2 matrices, you can make sense of the statement ‘(x + 2) × (x + 2) = x × x + 4 × x + 4’
now when x is a 2-by-2 matrix. Does the proof we gave still work, and is the statement true?
Well, multiplication of matrices does still distribute over addition, and the two small cal-
culations do still work. But matrix multiplication is not commutative; you can find pairs of
matrices where the order you multiply them makes a difference to the answer. So the proof does
not work.
2 0
But it happens (luckily!) to be the case that multiplication of any 2-by-2 matrix by ( ) does
0 2
commute (think about why!) and since the only place we used commutativity of multiplication
in our proof above was to say 2 × x = x × 2, we can make our proof work by changing the
2 0
reason ‘multiplication is commutative’ to ‘multiplication by ( ) is commutative’. Phew! The
0 2
statement is still true for 2-by-2 matrices, and we can prove it.
However, you should be a bit careful with matrices. Is it true that
2 2 2
⎛ 0 1 0 0 ⎞ 0 1 0 1 0 0 0 0
( )+( ) =( ) +2⋅( )⋅( )+( ) ?
⎝ 0 0 1 0 ⎠ 0 0 0 0 1 0 1 0

This looks like the same ‘expanding out the brackets’ that we just did, but (if you try to
mimic the proof above) you’ll see that there is a step where you would like to say that
0 1 0 0 0 0 0 1
( )⋅( ) = ( )⋅( ), i.e. that these two matrices commute. They don’t, and
0 0 1 0 1 0 0 0
this is where the calculation goes wrong.
Next term, we’ll give axiomatic definitions of a group and a vector space, and start proving
theorems about abstract groups (and vector spaces). Here ‘abstract’ means we don’t assume
anything about the group except the axioms. This will seem painful and useless at first: you’ll
(by then) know a few concrete examples of groups and of vector spaces. It will usually be easier
to see how to prove the theorems for the concrete examples. Usually you will have some idea
already why the theorems should be true in the examples, while you won’t have much intuition
for how abstract groups behave. The natural response will be that you don’t want to study
abstract groups, you want to work with the concrete examples you know. But this is the wrong
reaction. The reason is that you will then only learn about the concrete examples you already
know, and you will suffer as soon as in future courses you see new examples of groups and are
expected to immediately know a bunch of facts about them (and also in the exam, where we
will likely test your ability to work with a new example of an abstract structure).

Finally, let’s return to proof. Why should you care that you can mathematically prove a
statement, when it’s obviously true (like the one above) or when you can check lots of cases and
become convinced?
First, we will not generally be interested in proving obvious statements, we will rather be
trying to prove statements which aren’t obvious. We will discuss later what exactly that word
‘obvious’ means, and we will see lots of examples of statements where you don’t immediately see
whether it is true or false, or how to decide which.
Second, what do you learn from checking cases? If you are trying to find out whether a
claim is true or false, it’s a good idea to start checking cases. That might give you an idea why
the claim is true, or find out if it is complete nonsense. But what if the statement is true for
CHAPTER 1. INTRODUCTION 9

most cases, but there are some special cases where it goes wrong? You most likely won’t find
them. Similarly, if you’re writing a computer program (a likely part of your future career!) and
your program works most of the time, but you don’t consider some special cases (‘edge cases’ is
the jargon), you might end up writing a program which causes a disaster—not at the level of
say crashing a plane, because such programs are checked in detail, but you could easily find
your automated trading program has lost your bank a lot of money and you your job. To avoid
that, you need to learn how to keep an overview of a complicated problem: which parts have I
checked, and what is still left that could go wrong? Learning to write formal proofs is a good
way to train.

In this course, we need to work with precise definitions and statements, and you will need to
know these. Not only will you need to know these, but you will have to understand them, and
be able (through the use of them) to demonstrate that you understand them. Simply learning
the definitions without understanding what they mean is not going to be adequate. I hope that
these words of warning don’t discourage you, but I think it’s important to make it clear that
this is a subject at a higher conceptual level than most of the mathematics you are likely to have
studied before. This does not mean it is incredibly hard and you will struggle. It is not incredibly
hard, and you are quite capable of doing well in this course (or you would not be here). It does
mean, though, that if you are used to getting through school courses by memorising material
without understanding it, then now is the time to change that (and, by the way, no-one will
hire you for your memorisation ability—a computer does that better!).
One of the standard problems students have in this course is around what it means to
‘know precise definitions’. We will be using English language—not, for the most part, logical
symbols—to define various concepts. If you know the string of words as it appears here by heart,
then, yes, you know the precise definition. But most likely I will not be completely consistent,
and certainly your textbooks and other courses will use slightly different strings of words for the
same concept. What will be changed will turn out to be things that do not alter the meaning of
the concept—you’re completely used to the idea that there are words one can change without
changing the intent of a sentence in English. Mathematical English is, however, a bit more picky
than the usual spoken English; there are some words which you cannot change, and in particular
the order of words is often important. I’ll highlight this when it gets relevant in the course. For
now, if you’re not sure whether two sentences mean the same thing, that tells you you don’t
understand either of them and you need to think a bit more and look at the examples.

1.1.1 How to get the most out of this course (and all the other
maths courses)
There are two theories about mathematical ability (and intelligence in general). One theory says
that you have what you are born with. The other says that (just like strength or stamina) it’s
something you develop by practice. Various studies have shown that broadly similar number
of students believe each theory, but the ones who believe ability is something you develop are
consistently the ones who do better—and almost all academic mathematicians believe ability is
something you learn and train.
Some people are faster than others, but speed is in the end not all that useful: no matter
how fast you are, if you switch off and coast for a while, you will have trouble catching up with
people who pay attention and work on understanding their courses. In particular—and this
is different to school maths—we will always assume you understood the previous lectures and
courses, and we will use things from those previous lectures and courses all the time. If you do
understand the previous material—even if you are not so fast—you’ll understand a good deal of
the current lecture (maybe all of it, maybe not quite) in real time, and you won’t need to spend
CHAPTER 1. INTRODUCTION 10

much time after the lecture going over the material. If you don’t really understand the previous
material, you won’t have a chance to understand large parts of the lecture and you’ll have to do
even more work afterwards to catch up.
In this course, all the theory will be introduced in the lectures, together with some examples.
There will be extra examples sessions (which don’t exist in most courses—don’t expect them!),
which you do not have to attend but which may well be useful.
In the lectures and examples sessions, you should be trying to understand what is going
on. Don’t waste time copying things down which will appear on Moodle (which is essentially
everything). Certainly don’t waste time with a newspaper or games on your phone, which annoys
me and distracts your classmates. No-one is taking attendance in the lectures; if you’re not going
to pay attention, go to a café instead. If you do not understand something I said or wrote, then
probably either I didn’t explain it properly or I made a mistake, so you should ask questions
(louder, or put your hand up, if I don’t hear). When I ask a question, I really want an answer.
Probably you need to think about the question to answer it, so I will wait until someone does
answer.
For many of you there will be some point in the lectures where you do not immediately
understand what I say, and when you ask a question the answer is still not very useful. You
should keep asking me to explain better in such a situation. It is possible that I will eventually
say that I want to move on and you should think about it after the lecture. That doesn’t mean
I think you are stupid, it generally means I am failing to understand what exactly you don’t
understand, or maybe I do understand but cannot think of a good way to explain it on the spot.
In either case, if you try to formulate clearly exactly what it is that you do not like (which will
take time, which is why you should do it after the lecture), you will probably find that doing so
also helps you figure out what is going on; once you understand something deeply in this way
you will not forget it. But it would still be useful to tell me about it after the lecture, so that I
can improve the lectures for next year (and if possible give some more explanation on Moodle
directly).
There will also be problems set every week, some online (for which you’ll see results immedi-
ately) and some which you will solve and hand in to your class teacher, who will mark them
and discuss in the next class. The class work will be marked, and in addition it will contribute
10% to your eventual course grade.
The purpose of the problems is for you to practice and check you really know what is going
on. If you get stuck, hand in half a solution with ‘I don’t know what to do next’ and your class
teacher will tell you (either written on the work, or maybe many people were stuck in the same
place and the class teacher will go over it in class; usually then there will be a short comment
like ‘Will discuss in class’). Then you learn something. If you don’t hand anything in, or you
only hand in the problems you could solve, you don’t learn anything. The written comments on
your work, and the explanations in class, are the most important piece of feedback you get—but
you only get it if you show us something on which we can give feedback. On that note—please
do not copy work from someone else (or from last year’s solutions). Doing this is a waste of
your time and ours, and it is plagiarism which can potentially land you in serious trouble. Your
mark for each week will reflect how well your class teacher feels you did on the exercises.
The contribution to the course grade is different. Each week, your work will be either judged
as acceptable or not—this is a binary system, you don’t get extra points for amazing work—and
to get all of the 10% course grade, you need sufficiently many acceptable pieces of work, handed
in on time, over the term.
There are two ways in which a piece of work can be judged acceptable. One is if you have a
‘Satisfactory’ or better grade. The other is if you do not have such a grade, but you have made a
serious attempt at all the questions. This is defined to mean: you have written down all the
definitions (often there will be only one) relevant to that question.
CHAPTER 1. INTRODUCTION 11

The intention of this contribution to the course grade is to reward students who keep up
with the course and make some effort to learn actively. If you know how to do most exercises,
you’re guaranteed to get the ‘acceptable’. If you don’t, then the first thing you should do is to
write down the definitions, not just because this will guarantee you your ‘acceptable’, but also
(and mainly) because the most common reason why students cannot do exercises is that they
do not know what the exercise is actually asking—writing down the definitions will often give
you an idea of how to get on with the solution.
The only way to fail to get the 10% for coursework is for you to decide that it is not worth
your time to make any serious attempt at the classwork. In recent years we noticed students
increasingly doing this, usually then telling us that they are ‘a bit behind and need to catch up’,
or that they will ‘do all the questions in revision’. Usually, these students failed their exams; we
hope that if you are thinking of studying ‘school style’, even if you believe that you personally
will be able to make it work, you will at least recognise that throwing away 10%—an entire class
grade—is a bad move.
Finally, there are office hours and the Maths Support Centre. If you don’t understand
something, you should first try to figure it out for yourself—if you manage, then you won’t
forget it (and you should be happy with yourself). But if you get stuck, then you should not
wait and hope that it magically gets clear. It probably will not, and you will suffer because you
don’t understand something I am assuming you do understand in my lectures. So go to office
hours or the Support Centre and ask questions. You have already paid for those office hours;
use them. You can also try talking with your friends on the course and seeing if you can figure
out what’s going on—group work can be fun and productive.

1.1.2 Topics covered (MA102, first half of MA103)

Descriptions of topics to be covered appear in the relevant chapters. However, it is useful to
give a brief overview at this stage. These notes are for the ten weeks of MA102, which is the
first half of MA103.
We are concerned primarily with proof and logic. We will first investigate how precise
mathematical statements can be formulated, and here we will use the language and symbols of
mathematical logic. We will then study how one can prove or disprove mathematical statements,
and introduce some important basic structures and concepts. This will occupy the first (roughly)
five weeks, at something like one week per topic. In each new topic, we will begin from scratch,
and the way you need to think about each topic will be different.
After this, we will spend the next five weeks concentrating on Analysis. This is one of the
major branches of abstract mathematics. While these five weeks are split into three topics, the
way you need to think about all three is very much the same.
Most of the material in these notes is intended to help you prepare for the rest of this course;
all of it is intended to prepare you for the second-year and later mathematics courses. All of it
is examinable, with the exception of sections which are clearly marked ‘non-examinable’. Just
to be clear—some of the non-examinable material will be useful for understanding the course
(and I’ll probably talk about it in lectures), some is background which you will not need to
understand the course (but which you might find interesting, and which I will probably not talk
about in lectures). The way I choose what material is examinable and what is not, is I try to
come up with a good exam question; if I can’t, then I’ll mark it as non-examinable. That means,
anything in the course marked as examinable is material which I know how to test in an exam.
CHAPTER 1. INTRODUCTION 12

1.2 Moodle
All information and materials for this course are on Moodle:
http://moodle.lse.ac.uk/course/view.php?id=1989
On the course Moodle page, you will find assignments, solutions, lecture notes, and so on.

1.3 Reading
These notes are intended to be a comprehensive treatment. That means, I think you should not
need to buy or borrow any textbooks for this course.
However, you might disagree. If you don’t like my writing style, or you want to understand
a particular topic better, try looking at a textbook. If you want more exercises, and you are
actually going to do the exercises, look at a textbook. If you want more exercises in order to
read the solutions, you’re wasting your time!
There are many books that would be useful for this subject, since abstract mathematics is a
component of all university-level mathematics degree programmes I know of.
For the first half of the course (the part covered by these notes), the following two books are
recommended, and most chapters of the notes will start with a reference to the corresponding
chapters in these two books.

• Biggs, Norman L., Discrete Mathematics, Second edition. (Oxford University Press, 2002).
[ISBN 0198507178].

• Eccles, P.J., An Introduction to Mathematical Reasoning: numbers, sets and functions. (Cam-
bridge University Press, 1997). [ISBN 0521597188].

There is one topic that neither of these covers, which is the topic of Complex Numbers.
However, this is a topic that is well-covered in a number of other textbooks—look around.

1.4 Activities and sample exercises

Throughout the chapters of these notes, you’ll find ‘activities’. These are things for you to do or
think about as you read, just to reaffirm that you’ve understood the material.
At the end of each chapter of these notes you will find some sample exercises together with
solutions. These are not the exercises that will be assigned for classes, but are additional to
those. They are a very useful resource. You should try them once you think you have mastered
a particular chapter. Really try them: don’t just simply read the solutions provided. Make a
serious attempt before consulting the solutions. Note that the solutions are often just sketch
solutions, to indicate to you how to answer the questions.
 Mathematical statements, proof, and logic
2
In this chapter we go over the basics which one needs in order to start doing abstract mathematics
and proof, namely statements and logic. This will go by fairly quickly—there is nothing hard
here. Even if some things look funny the first time you see them, expect that as you see them
repeatedly through the course, you will get used to them.
The material in this chapter is also covered in:

• Biggs, N.L. Discrete Mathematics. Chapters 1–3.

• Eccles, P.J. An Introduction to Mathematical Reasoning. Chapters 1–4 and 6.

2.1 Introduction
In this course, we want to make precise mathematical statements and establish whether they
are true or not—we want to prove things. But for that, we have to first understand what a
proof is. We will look at fairly simple types of mathematical statement, in order to emphasise
techniques of proof. Some of these statements are going to be interesting, others are not so
interesting—bear in mind that what you are doing in this part of the course is learning the rules
of the game: the play (and more of the fun) comes later.
In later chapters (such as those on numbers, analysis and algebra) we will use these proof
techniques extensively. You might think that some of the things we prove in this chapter are
very obvious and hardly merit proving, but proving even ‘obvious’ statements can be quite
tricky sometimes, and it is good preparation for proving more complicated things later.

2.2 Mathematical statements and proof

To introduce the topics of mathematical statement and proof, we start by giving some explicit
examples. Later in the chapter we give some general theory and principles. Our discussion of
the general theory is limited because this is not a course in logic. We need enough logic to
understand what mathematical statements mean and how we might prove or disprove them. We
don’t need to start talking about things like which statements are provable and which statements
are true (and whether those are the same or not). There are interesting mathematical things to
say there (and interesting philosophical things), but you don’t need to know them in order to
do mathematics.

13
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 14

2.2.1 Examples of Mathematical Statements

Consider the following statements (in which you should recall that the natural numbers are the
positive integers):

(a) 20 is divisible by 4.

(b) 21 is not divisible by 7.

(c) 21 is divisible by 4.

(d) 21 is divisible by 3 or 5.

(e) 50 is divisible by 2 and 5.

(f) n2 is even.

(g) For every natural number n, the number n2 + n is even.

(h) There is a natural number n such that 2n = 2n .

(i) If n is even, then n2 is even.

(j) For all odd numbers n, the number n2 is odd.

(k) For natural numbers n, the number n2 is even if and only if n is even.
√
(l) There are no natural numbers m and n such that 2 = m/n.

These are all mathematical statements, of different sorts (all of which will be discussed in
more detail in the remainder of this chapter).
Statements (a) to (e) are straightforward propositions about certain numbers, and these are
either true or false. Statements (d) and (e) are examples of compound statements. Statement (d)
is true precisely when either one (or both) of the statements ‘21 is divisible by 3’ and ‘21 is
divisible by 5’ is true. Statement (e) is true precisely when both of the statements ‘50 is divisible
by 2’ and ‘50 is divisible by 5’ are true.
Statement (f) is different, because the number n is not specified and whether the statement
is true or false will depend on the value of the so-called free variable n. Such a statement is
known as a predicate.
Statement (g) makes an assertion about all natural numbers and is an example of a universal
statement.
Statement (h) asserts the existence of a particular number and is an example of an existential
statement.
Statement (i) can be considered as an assertion about all even numbers, and so it is a
universal statement, where the ‘universe’ is all even numbers. But it can also be considered as
an implication, asserting that if n happens to be even, then n2 is even.
Statement (j) is a universal statement about all odd numbers. It can also be thought of (or
rephrased) as an implication, for it says precisely the same as ‘if n is odd, then n2 is odd’.
Statement (k) is an ‘if and only if’ statement: what it says is that n2 is even, for a natural
number n, precisely when n is even. But this means two things: namely that n2 is even if n is
even, and n is even if n2 is even. Equivalently, it means that n2 is even if n is even and that n2
is odd if n is odd. So statement (k) will be true precisely if (i) is true for all natural numbers,
and (j) is true.
Statement (l) asserts the non-existence of a certain pair of numbers (m, n). Another way
of thinking about this statement is that it says that for all choices of (m, n), it is not the case
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 15
√
that m/n = 2. (This is an example of the general rule that a non-existence statement can be
thought of as a universal statement, something to be discussed later in more detail.)

It’s probably worth giving some examples of things that are not proper mathematical
statements.
‘6 is a nice number’ is not a mathematical statement. This is because ‘nice number’ has
no mathematical meaning. However, if, beforehand, we had defined ‘nice number’ in some way,
then this would not be a problem. For example, suppose we said:
Let us say that a number is nice if it is the sum of all the positive numbers that
divide it and are less than it.
Then ‘6 is a nice number’ would be a proper mathematical statement, and it would be true,
because 6 has positive divisors 1, 2, 3, 6 and 6 = 1 + 2 + 3. But without defining what ‘nice’ means,
it’s not a mathematical statement. Definitions are important1 .
‘n2 + n’ is not a mathematical statement, because it does not say anything about n2 + n.
It is not a mathematical statement in the same way that ‘Boris Johnson’ is not a sentence: it
makes no assertion about what Boris Johnson did or did not do to get thrown out. However,
‘n2 + n > 0’ is an example of a predicate with free variable n and, for a particular value of n, this
is a mathematical statement. Likewise, ‘for all natural numbers n, n2 + n > 0’ is a mathematical
statement.
Finally, anything which does not make sense as an English sentence is not a mathematical
statement. We will use lots of symbols—some you know, like =, some you don’t yet, like ∀—which
all mean some English word or words. It’s easy to write something with symbols that, when you
read it out, doesn’t make sense. If when you read your work out, you are saying something like
‘five is true’ or ‘for every integer n we have n = 2’, something is wrong. Figure out what you
meant to write, then write that.

2.2.2 Introduction to proving statements

We’ve seen, above, various types of mathematical statement, and such statements are either
true or false. But how would we establish the truth or falsity of these?
We can, even at this early stage, prove (by which we mean establish the truth of) or disprove
(by which we mean establish the falsity of) most of the statements given above. Before we do
this, we need to be sure that we really know precisely what all the statements mean. We already
said what we mean by the ‘natural numbers’, and I assume you know what the algebra means
(i.e. that n2 means n multiplied by n, and so on). We haven’t formally defined ‘divisible’, though,
and you might not have seen this in school. So we need to do that:
Let us say that a natural number n is divisible by a natural number d if we can write n = d ⋅ k
for some natural number k. We say that a natural number is even if it is divisible by 2, and odd
if it is not.
Note that saying n is divisible by d is the same thing as saying that if we try to divide
n by d we get no remainder. This definition is probably what you thought ‘divisible’ meant
when you read the statements in the previous section—now you know you were right, and
you know everyone else will (by definition!) agree with you. For the rest of your degree, we’ll
assume you know what ‘divisible’ means, and the meaning will not be changed. We might say
‘divisible means when we try to divide we get no remainder’, or some other phrase which has
the same mathematical meaning: the precise words aren’t important. What is important is that
the mathematical meaning is now fixed.
1
Usually we say that a natural number which is equal to the sum of all smaller positive numbers which divide
it is perfect. The reason for using ‘nice’ in the text is because that term is not commonly defined!
 CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 16

Now that we’re all clear on exactly what the statements mean, let’s see which ones are true
and prove them.

(a) 20 is divisible by 4.
This statement is true. Since 20 = 5 × 4, we see that (by the definition) 20 is divisible by 4.
And that’s a proof! It’s utterly convincing, watertight, and not open to debate. Nobody can
argue with it, not even a sociologist! Isn’t this fun? Well, maybe it’s not that impressive in
such a simple situation, but we will certainly prove more impressive results later.

(b) 21 is not divisible by 7.

This is false. It’s false because 21 is divisible by 7, because 21 = 3 × 7.

(c) 21 is divisible by 4.
This is false, as can be established in a number of ways. First, we note that if the natural
number m satisfies m ≤ 5, then m × 4 will be no more than 20. And if m ≥ 6 then m × 4
will be at least 24. Well, any natural number m is either at most 5 or at least 6 so, for all
possible m, we do not have m × 4 = 21 and hence there is no natural number m for which
m × 4 = 21. In other words, 21 is not divisible by 4. Another argument (which is perhaps
more straightforward, but which relies on properties of rational numbers rather than just
simple properties of natural numbers) is to note that 21/4 = 5.25, and this is not a natural
number, so 21 is not divisible by 4. (This second approach is the same as showing that 21
has remainder 1, not 0, when we divide by 4.)
Most of you are probably completely happy with these proofs. Maybe one or two of you
would like to know things like: why is there no natural number between 5 and 6? Do we need
to prove it? We’ll get to that next term; for now, don’t worry about it.

(d) 21 is divisible by 3 or 5.
As we noted above, this is a compound statement. It is true precisely if one (or both) of the
following statements is true:

(i) 21 is divisible by 3
(ii) 21 is divisible by 5.

Statement (i) is true, because 21 = 7 × 3. Statement (ii) is false. Because at least one of these
two statements is true, statement (d) is true.

(e) 50 is divisible by 2 and 5.

This is true. Again, this is a compound statement and it is true precisely if both of the
following statements are true:

(i) 50 is divisible by 2
(ii) 50 is divisible by 5.

Statements (i) and (ii) are indeed true because 50 = 25 × 2 and 50 = 10 × 5. So statement (e)
is true.

(f) n2 is even
As mentioned above, whether this is true or false depends on the value of n. For example,
if n = 2 then n2 = 4 is even, but if n = 3 then n2 = 9 is odd. So, unlike the other statements
(which are propositions), this is a predicate P (n). The predicate will become a proposition
 CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 17

when we assign a particular value to n to it, and the truth or falsity of the proposition can
then be established. You probably implicitly assume that n has to be a natural number,
but there isn’t actually anything in the statement to tell you that—maybe n is a matrix, in
which case it’s not even clear what ‘even’ should mean for a matrix (we only defined ‘even’
for natural numbers). If we assume n is a natural number, then (i) and (j) cover all the
possibilities.

(g) For every natural number n, the number n2 + n is even.

Here’s our first non-immediate, non-trivial, proof. How on earth can we prove this, if it is
true, or disprove it, if it is false? Suppose it was false. How would you convince someone of
that? Well, the statement says that for every natural number n, n2 + n is even. So if you
managed (somehow!) to find a particular N for which N 2 + N happened to be odd, you could
prove the statement false by simply observing that ‘When n = N , it is not the case that
n2 + n is even.’ And that would be the end of it. So, in other words, if a universal statement
about natural numbers is false, you can prove it is false by showing that its conclusion is
false for some particular value of n. But suppose the statement is true. How could you prove
it. Well, you could prove it for n = 1, then n = 2, then n = 3, and so on, but at some point
you would expire and there would still be numbers n that you hadn’t yet proved it for. And
that simply wouldn’t do, because if you proved it true for the first 9999 numbers, it might
be false when n = 10000. So what you need is a more sophisticated, general argument that
shows the statement is true for any arbitrary n.
Now, it turns out that this statement is true. So we need a nice general argument to establish
this. Well, here’s one approach. We can note that n2 + n = n(n + 1). The numbers n and n + 1
are consecutive natural numbers. So one of them is odd and one of them is even. When you
multiply any odd number and any even number together, you get an even number, so n2 + n
is even. Are you convinced? Maybe not? We really should be more explicit. Suppose n is
even. What that means is that, for some integer k, n = 2k. Then n + 1 = 2k + 1 and hence

n(n + 1) = 2k(2k + 1) = 2 (k(2k + 1)) .

Because k(2k + 1) is an integer, this shows that n2 + n = n(n + 1) is divisible by 2; that is, it
is even. We supposed here that n was even. But it might be odd, in which case we would
have n = 2k + 1 for some integer k. Then

n(n + 1) = (2k + 1)(2k + 2) = 2 ((2k + 1)(k + 1)) ,

which is, again, even, because (2k + 1)(k + 1) is an integer.

Right, we’re really proving things now. This is a very general statement, asserting something
about all natural numbers, and we have managed to prove it. I find that quite satisfying,
don’t you?

(h) There is a natural number n such that 2n = 2n .

This is an existential statement, asserting that there exists n with 2n = 2n . Before diving
in, let’s pause for a moment and think about how we might deal with such statements. If
an existential statement like this is true we would need only to show that its conclusion
(which in this case is 2n = 2n ) holds for some particular n. That is, we need only find an n
that works. If the statement is false, we have a lot more work to do in order to prove that
it is false. For, to show that it is false, we would need to show that, for no value of n does
the conclusion holds. Equivalently, for every n, the conclusion fails. So we’d need to prove a
universal statement and, as we saw in the previous example, that would require us to come
up with a suitably general argument.
 CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 18

In fact, this statement is true. This is because when n = 1 we have 2n = 2 = 21 = 2n ; we’re

done.
We could also use n = 2 to prove this statement is true: we have 2n = 2 ⋅ 2 = 4 = 22 = 2n . But
to prove an existential statement to be true, it’s enough to find one example; once we saw
n = 1 is such an example, we don’t need to care that n = 2 is also an example.

(i) If n is even, then n2 is even

This is true. The most straightforward way to prove this is to assume that n is some (that is,
any) even number and then show that n2 is even. So suppose n is even. Then n = 2k for some
integer k (by definition) and hence n2 = (2k)2 = 4k 2 . This is even because it is 2(2k 2 ) and
2k 2 is an integer.

(j) For all odd numbers n, n2 is odd.

This is true. The most straightforward way to prove this is to assume that n is any odd
number and then show that n2 is also odd. So suppose n is odd. Then n = 2k + 1 for some
integer k and hence n2 = (2k + 1)2 = 4k 2 + 4k + 1. To establish that this is odd, we need to show
that it can be written in the form 2K +1 for some integer K. Well, 4k 2 +4k +1 = 2(2k 2 +2k)+1.
This is indeed of the form 2K + 1, where K is the integer 2k 2 + 2k. Hence n2 is odd.
Another way to prove this result is to prove that if n2 is even then n must be even. We won’t
do that right now, because to do it properly requires a result we meet later concerning the
factorisation of numbers into prime numbers. But think about the strategy for a moment.
Suppose we were able to prove the following statement, which we’ll call Q:
Q: if n2 is even then n is even.
Why would that establish what we want (namely that if n is odd then n2 is odd). Well, one
way is to observe that Q is what’s called the contrapositive of statement (j) that we’re trying
to prove, and the contrapositive is logically equivalent to the initial statement. (This is a bit of
formal logic, and we will discuss this more later). But there’s another way of thinking about
it, which is perhaps easier to understand at this stage. Suppose we have proved statement Q
and suppose that n is odd. Then it must be the case that n2 is odd. For, if n2 was not odd, it
would be even and then Q would tell us that this means n is even. But we have assumed n is
odd. It cannot be both even and odd, so we have reached a contradiction. By assuming that
the opposite conclusion holds (n2 even) we have shown that something impossible happens.
This type of argument is known as a proof by contradiction and it is often very powerful. We
will see more about this later.

(k) For natural numbers n, n2 is even if and only if n is even.

This is true. What we have shown in proving (i) and (j) is that if n is even then n2 is even,
and if n is odd then n2 is odd. The first, (statement (i)) establishes that if n is even, then
n2 is even. The second of these (statement (j)) establishes that n2 is even only if n is even.
This is because it shows that n2 is odd if n is odd, from which it follows that if n2 is even, n
must not have been odd, and therefore must have been even. ‘If and only if’ statements if
this type are very important. As we see here, the proof of such statements breaks down into
the proof of two ‘If-then’ statements.
√
(l) There are no natural numbers m and n such that 2 = m/n.
This is, in fact, true, though we defer the proof for now, until we know more about factorisation
of numbers into prime numbers. We merely comment that the easiest way to prove the
statement is to use a proof by contradiction.
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 19

These examples hopefully demonstrate that there are a wide range of statements and proof
techniques, and in the rest of this chapter we will explore these further.
Right now, one thing I hope comes out very clearly from these examples is that to prove a
mathematical statement, you need to know precisely what it means. Well, that sounds obvious,
but you can see how detailed we had to be about the meanings (that is, the definitions) of the
terms ‘divisible’, ‘even’ and ‘odd’.
Something you can also notice is that we like to come up with special names to distinguish
things even when it’s ‘unnecessary’. For example, we talked about ‘propositions’ and ‘predicates’
as being different types of statement; why bother with these two funny words? Right now,
this no doubt feels like me inventing more words that you have to learn for no good reason.
If I write down one of the statements above, you’ll be able to see immediately whether it is a
simple true-or-false statement (a proposition) or whether there is some free variable in it and
its truth could depend on the value of the free variable (a predicate). However later, when we
are dealing with more complicated statements and have to explain something difficult, it will
be useful for me to be able to say ‘consider the proposition ...’ and ‘we have the predicate ...’
and expect that these words have made your life easier—you know already that what is coming
should be respectively a true-or-false statement, and have a free variable (or two) in it. Quite a
lot of mathematical vocabulary and notation is there ‘to help the reader’. It will always look
unnecessary when it’s introduced, because that will always be in a simple situation where what
is intended is obvious. We will never test you on it (there will not be an exam question asking
which of the following statements are predicates), but knowing it will help you understand and
write mathematics better.

2.3 Some basic logic

Mathematical statements can be true or false. Let’s denote ‘true’ by T and ‘false’ by F. Given a
statement, or a number of statements, it is possible to form other statements. This was indicated
in some of the examples above (such as the compound statements). A technique known as the
use of ‘truth tables’ enables us to define ‘logical operations’ on statements, and to determine
when such statements are true. This is all a bit vague, so let’s get down to some concrete
examples.

2.3.1 Negation
The simplest way to take a statement and form another statement is to negate the statement.
The negation of a statement P is the statement ¬P (sometimes just denoted ‘not P ’), which is
defined to be true exactly when P is false. This can be described in the very simple truth table,
Table 2.1:
P ¬P
T F
F T

Table 2.1: The truth table for ‘negation’ or ‘not’

What does the table signify? Quite simply, it tells us that if P is true then ¬P is false and if
P is false then ¬P is true.
Example 2.1. If P is ‘20 is divisible by 3’ then ¬P is ‘20 is not divisible by 3’. Here, P is false
and ¬P is true.
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 20

It has, I hope, been indicated in the examples earlier in this chapter, that to disprove a
universal statement about natural numbers amounts to proving an existential statement. That
is, if we want to disprove a statement of the form ‘for all natural numbers n, property p(n) holds’
(where p(n) is some predicate, such as ‘n2 is even’) we need only produce some N for which
p(N ) fails. Such an N is called a counterexample. Equally, to disprove an existential statement
of the form ‘there is some n such that property p(n) holds’, one would have to show that for
every n, p(n) fails. That is, to disprove an existential statement amounts to proving a universal
one. But, now that we have the notion of the negation of a statement we can phrase this a little
more formally. Proving that a statement P is false is equivalent to proving that the negation
¬P is true. In the language of logic, therefore, we have the following:

• The negation of a universal statement is an existential statement.

• The negation of an existential statement is a universal statement.

More precisely,

• The negation of the universal statement ‘for all n, property p(n) holds’ is the existential
statement ‘there is n such that property p(n) does not hold’.

• The negation of the existential statement ‘there is n such that property p(n) holds’ is the
universal statement ‘for all n, property p(n) does not hold’.

We could be a little more formal about this, by defining the negation of a predicate p(n) (which,
recall, only has a definitive true or false value once n is specified) to be the predicate ¬p(n)
which is true (for any particular n) precisely when p(n) is false. Then we might say that

• The negation of the universal statement ‘for all n, the statement p(n) is true’ is the existential
statement ‘there is n such that ¬p(n) is true’.

• The negation of the existential statement ‘there is n such that p(n) is true’ is the universal
statement ‘for all n, the statement ¬p(n) is true’.

Now, let’s not get confused here. None of this is really difficult or new. We meet such logic
in everyday life. If I say ‘It rains every day in London’ then either this statement is true or
it is false. If it is false, it is because on (at least) one day it does not rain. The negation (or
disproof) of the statement ‘On every day, it rains in London’ is simply ‘There is a day on which
it does not rain in London’. The former is a universal statement (‘On every day, . . . ’) and the
latter is an existential statement (‘there is a day . . . ’). Or, consider the statement ‘There is a
student who enjoys reading these lecture notes’. This is an existential statement (‘There is . . . ’).
This is false if ‘No student enjoys reading these lecture notes’. Another way of phrasing this
last statement is ‘Every student reading these lecture notes does not enjoy it’. This is a more
awkward expression, but it emphasises that the negation of the initial, existential statement, is
a universal one (‘Every student . . . ’).
The former is an existential statement (‘there is something I will write that . . . ’) and the
latter is a universal statement (‘everything I write will . . . ). This second example is a little more
complicated, but it serves to illustrate the point that much of logic is simple common sense.
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 21

2.3.2 Conjunction and disjunction

There are two very basic ways of combining propositions: through the use of ‘and’ (known as
conjunction) and the use of ‘or’ (known as disjunction).
Suppose that P and Q are two mathematical statements. Then ‘P and Q’, also denoted
P ∧ Q, and called the conjunction of P and Q, is the statement that is true precisely when both
P and Q are true. For example, statement (e) above, which is
‘50 is divisible by 2 and 5’
is the conjunction of the two statements

• 50 is divisible by 2

• 50 is divisible by 5.

Statement (e) is true because both of these two statements are true.
Table 2.2 gives the truth table for the conjunction P and Q:

P Q P ∧Q
T T T
T F F
F T F
F F F

Table 2.2: The truth table for ‘and’

What Table 2.2 says is simply that P ∧ Q is true precisely when both P and Q are true (and
in no other circumstances).
Suppose that P and Q are two mathematical statements. Then ‘P or Q’, also denoted P ∨ Q,
and called the disjunction of P and Q, is the statement that is true precisely when P , or Q, or
both, are true. For example, statement (d) above, which is
‘21 is divisible by 3 or 5’
is the disjunction of the two statements

• 21 is divisible by 3

• 21 is divisible by 5.

Statement (d) is true because at least one (namely the first) of these two statements is true.
Note one important thing about the mathematical interpretation of the word ‘or’. It is always
used in the ‘inclusive-or’ sense. So P ∨ Q is true in the case when P is true, or Q is true, or
both. In some ways, this use of the word ‘or’ contrasts with its use in normal everyday language,
where it is often used to specify a choice between mutually exclusive alternatives. (For example
‘You’re either with us or against us’.) But if I say ‘Tomorrow I will wear brown trousers or I will
wear a yellow shirt’ then, in the mathematical way in which the word ‘or’ is used, the statement
would be true if I wore brown trousers and any shirt, any trousers and a yellow shirt, and also if
I wore brown trousers and a yellow shirt. You might have your doubts about my dress sense in
this last case, but, logically, it makes my statement true.
Table 2.3 gives the truth table for the disjunction P and Q:
What Table 2.3 says is simply that P ∨ Q is true precisely when at least one of P and Q is
true.
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 22

P Q P ∨Q
T T T
T F T
F T T
F F F

Table 2.3: The truth table for ‘or’

2.3.3 If-then statements

It is very important to understand the formal meaning of the word ‘if’ in mathematics. The
word is often used rather sloppily in everyday life, but has a very precise mathematical meaning.
Let me give you an example. Suppose I tell you ‘If it rains, then I wear a raincoat’, and suppose
that this is a true statement. Well, then, suppose it rains. You can certainly conclude I will wear
a raincoat. But what if it does not rain? Well, you can’t conclude anything. My statement only
tells you about what happens if it rains. If it does not, then I might, or I might not, wear a
raincoat: and whether I do or not does not affect the truth of the statement I made. You have
to be clear about this: an ‘if-then’ statement only tells you about what follows if something
particular happens.
More formally, suppose P and Q are mathematical statements (each of which can therefore
be either true or false). Then we can form the statement denoted P Ô⇒ Q (‘P implies Q’ or,
equivalently, ‘if P , then Q’), which has as its truth table Table 2.3.3. (This type of statement is
known as an if-then statement or an implication.)

P Q P Ô⇒ Q
T T T
T F F
F T T
F F T
Table 2.4: The truth table for ‘P Ô⇒ Q’

Note that the statement P Ô⇒ Q is false only when P is true but Q is false. To go back to
the previous example, the statement ‘If it rains, I wear a raincoat’ is false precisely if it does
rain but I do not wear a raincoat.
Warning 2.2. Many students focus on the ‘if the premise is true’ first two lines of the truth
table above, and forget the last two lines. We will need to use all four lines regularly, so do not
do this. Yes, the mathematical Ô⇒ is a bit different to the usual English ‘implies’, but this is
something you simply need to get used to. For the next few months, every time you use Ô⇒ ,
think for a few seconds about whether you have really written what you wanted to write.
The statement P Ô⇒ Q can also be written as Q ⇐Ô P . There are different ways of
describing P Ô⇒ Q, such as:

• if P then Q

• P implies Q

• P is sufficient for Q

• Q if P
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 23

• P only if Q

• Q whenever P

• Q is necessary for P .
All these mean the same thing. The first two are the ones I will use most frequently.

2.3.4 If and only if statements; logical equivalence

If P Ô⇒ Q and Q Ô⇒ P then this means that Q will be true precisely when P is. That is
Q is true if and only if P is. We use the single piece of notation P ⇐⇒ Q instead of the two
separate P Ô⇒ Q and P ⇐Ô Q. There are several phrases for describing what P ⇐⇒ Q
means, such as:
• P if and only if Q (sometimes abbreviated to ‘P iff Q’)

• P is equivalent to Q

• P is necessary and sufficient for Q

• Q is necessary and sufficient for P .

The truth table is shown in Table 2.5, where we have also indicated the truth or falsity
of P Ô⇒ Q and Q Ô⇒ P to emphasise that P ⇐⇒ Q is the same as the conjunction
(P Ô⇒ Q) ∧ (Q Ô⇒ P ).

P Q P Ô⇒ Q Q Ô⇒ P P ⇐⇒ Q
T T T T T
T F F T F
F T T F F
F F T T T

Table 2.5: The truth table for ‘P ⇐⇒ Q’

What the table shows is that P ⇐⇒ Q is true precisely when P and Q are either both true
or both false.
Activity 2.1. Look carefully at the truth table and understand why the values for P ⇐⇒ Q
are as they are. In particular, try to explain in words why the truth table is the way it is.
So far in mathematics, most statements you have seen are ‘if and only if’ statements. In
particular when you rearrange equations, you’re (usually!) saying ‘these two things are equal if
and only if those two things are equal’. In fact, most of the times that you have seen a ‘genuine’
Ô⇒ (I mean, one where it would not be true to write ⇐⇒ ) it’s been as a warning that
something nasty might be around the corner: it’s true that if a = b then a2 = b2 , but it’s not
always true that if a2 = b2 then a = b, so be careful.
That is not how things will be for most of the mathematics you will study, and you will get
used to ‘implies’ being the normal thing. That shouldn’t be surprising. There are usually several
different possible causes for the same effect, so any one of these causes will imply the effect. If
you stay inside, you won’t get sunburnt; if you use sunscreen, you won’t get sunburnt; if you
wear a spacesuit, you won’t get sunburnt. The converse is generally going to be false—it is not
true that if you don’t get sunburnt, then the reason is that you used sunscreen, and stayed
inside, and wore a spacesuit.
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 24

Another piece of vocabulary we will sometimes use, when we are told A ⇐⇒ B, is that
A and B are logically equivalent. Spelling it out, we say A and B are logically equivalent if
either they are both true, or they are both false. Generally, we will say things like ‘P is true if
and only if Q is true’ when we need to look at what the statements P and Q actually are—as
mathematical statements, maybe talking about integers—in order to see why the ⇐⇒ is the
case. We will say that A and B are ‘logically equivalent’ if we do not need to understand the
mathematical meaning of the statements at all, we only need to look at the logic. This is ‘to
help the reader’.
Example 2.3. The statements ¬(P ∨ Q) and ¬P ∧ ¬Q are logically equivalent.
To see that this is true, we can draw out the truth tables:
P Q P ∨ Q ¬(P ∨ Q) ¬P ¬Q ¬P ∧ ¬Q
T T T F F F F
T F T F F T F
F T T F T F F
F F F T T T T

Table 2.6: The truth tables for ¬(P ∨ Q) and ¬P ∧ ¬Q

We can see that the two bold lines are the same—these two statements are logically equivalent.
So I might say ‘We know that the flobble is not either pretty or quick. It is logically equivalent
to say that the flobble is not pretty, and the flobble is not quick.’—and I presumably want to go
on for a few more lines of argument to tell you something interesting about the flobble. However
what I’ve signalled here is that you do not need to know what a flobble is, nor what it should
mean for one to be pretty or quick, in order to be happy with this particular line of argument.
If on the other hand, I say ‘a graph is bipartite if and only if it contains no odd cycle’ then
I’m signalling that in order to be happy that this statement is true (it is) you will need to look
up definitions of all the funny words in the sentence (don’t do that now!) and do some ‘real
maths’ not ‘just logic’.
Activity 2.2. Show that the statements ¬(P ∧ Q) and ¬P ∨ ¬Q are logically equivalent.

2.4 Implications and associated statements

Given an implication P Ô⇒ Q, there are three more ‘associated’ statements we can make
by swapping P and Q for ¬P and ¬Q, by reversing the implication, or both. One of these is
important because it is logically equivalent to P Ô⇒ Q (and this turns out to be very useful)
and the other two are important because they are not logically equivalent to P Ô⇒ Q (and
this is a standard way to make mistakes).

2.4.1 Converse statements

The implication Q Ô⇒ P is the converse of P Ô⇒ Q. Generally, there is no reason why the
converse should be true just because the implication is. For example, consider the statement
‘If it is Tuesday, then I buy the Guardian newspaper’. The converse is ‘If I buy the Guardian
newspaper, then it is Tuesday’. Well, I might buy that newspaper on other days too, in which
case the implication can be true but the converse false.
We’ve seen, in fact, that if both P Ô⇒ Q and Q Ô⇒ P then we have a special notation,
P ⇐⇒ Q, for this situation. Generally, then, the truth or falsity of the converse Q Ô⇒ P has
to be determined separately from that of the implication P Ô⇒ Q.
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 25

Activity 2.3. What is the converse of the statement ‘if the natural number n divides 4 then n
divides 12’ ? Is the converse true? Is the original statement true?

2.4.2 Contrapositive statements

The contrapositive of an implication P Ô⇒ Q is the statement ¬Q Ô⇒ ¬P . The contrapositive
is logically equivalent to the implication, as Table 2.7 shows. (The columns highlighted in bold
are identical.)

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

Table 2.7: The truth tables for P Ô⇒ Q and ¬Q Ô⇒ ¬P .

If you think about it, the equivalence of the implication and its contrapositive makes sense.
For, ¬Q Ô⇒ ¬P says that if Q is false, P is false also. So, it tells us that we cannot have Q
false and P true, which is precisely the same information as is given by P Ô⇒ Q.
So what’s the point of this? Well, sometimes you might want to prove P Ô⇒ Q and it will,
in fact, be easier to prove instead the equivalent (contrapositive) statement ¬Q Ô⇒ ¬P . You
will see many examples of this through your degree.

2.4.3 Converse of the contrapositive

Finally, ¬P Ô⇒ ¬Q is the converse of the contrapositive of P Ô⇒ Q. As we’ve seen, this is
logically equivalent to the converse, so not logically equivalent to P Ô⇒ Q, but all the ‘not’s
floating around can make this hard to see, especially if P and Q are complicated statements
with ‘not’s in themselves.
Warning 2.4. It is very easy to get tricked into believing that just because a statement is true,
so is its converse (or the contrapositive of its converse). If you wear sunscreen, you will not get
sunburnt. If you tell someone ‘you are not wearing sunscreen, so you will get sunburnt’ you
might be right; on the other hand if it’s midnight, you will probably get laughed at.

Mistake 1 (The theorem doesn’t apply, so its conclusion is false). I’ve just finished (summer
2019) marking MA103 exams in which a large number of students wrote ‘the conditions of
Theorem A are not met, so the conclusion is false’. That is exactly the same error as the midnight
sunscreen advocate: Theorem A is an ‘if P then Q’ statement, and it can perfectly well be that
P is false but (for some other reason) Q is still true. So these answers received zero marks, and
this paragraph has been added.
Summer 2020: The same mistake again. I’ll keep adding to this each year many students
lose marks for this class of error in the exam.
Summer 2021: Well, there were less of these mistakes, but it made the difference between
passing and failing for quite a few.
Summer 2022: We didn’t have a question of this form this year.
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 26

2.5 What is a proof ?

You should probably have some idea of what a proof is by now: you start with some statements
you’re assuming to be true (usually called axioms), from these statements you deduce others
(using the rules of logic) and eventually you get to the statement you wanted to prove. If you
are being very formal, you should write down every single step.

If you write down every single step, you’re in a great position if someone wants to argue with
your proof. If someone doesn’t agree with your conclusion—the statement you’re proving—it’s
their problem to find a mistake in your proof. That means they have to point at some statement
in your proof and say that they do not believe it. Now there are two sorts of statements in your
proof: ones which follow logically from earlier statements, and your axioms. If the doubter says
they don’t believe something which follows logically from earlier statements, then they have
to point at one of these earlier statements and say they don’t like that one either (or they tell
you they don’t believe in logic, in which case you can safely stop listening). Eventually they
will either be convinced you were right all along, or they will get back to one of your axioms
and say they disagree with that. Now, if you have some strange non-standard axiom, then there
might even be a good reason to argue. But if you stick to standard axioms, like ‘addition of
natural numbers is commutative’, then no-one is going to argue—which means you will convince
everyone that what you claim is true. This is the gold standard of proof.
The problem with writing down every single step is that it takes a very long time to actually
get anywhere. Look back to the proof on page 7—it takes eight lines to do a piece of algebra
which you would normally write out in one line, and even that proof skips the steps of proving
from axioms that 2 × 2 = 2 + 2 = 4 (which we’ll see how to do next term). You don’t want to
spend the next three years taking pages and pages to write out simple algebra, so we need to
agree on a way to write proofs which is shorter. There are two ways to do this, and we will use
both.

The first way is that, as we go through the course (and the degree) we will make for ourselves
a library of true statements—ones which we already proved—and we will not repeat the proofs
every time we want to use them. So, for example, we already proved that for every natural
number n, the number n2 + n is even (We didn’t really write out every single step—if you don’t
like that, try doing it yourself). Next time we want to know that n2 + n is even for some natural
number n, we won’t need to prove it, we can just say ‘proved in MA103’. There’s nothing much
anyone can object to here—it’s clear that we could have written out a gold standard proof just
by copying-and-pasting in the proof from MA103.
The second way we will save time is by not writing out every single step. When you need to
do a piece of algebra, do it just as you did in school, and we will assume you do know how to
justify all the steps by going back to the axioms (or at least that you know where to look in
order to find out how). We will also sometimes save steps by saying that something is ‘obvious’,
or ‘clear’. When you (or I) write ‘obvious’ or ‘clear’ in a proof, it is there to tell the reader that
there are some steps missing, that you (or I) know what those steps are, and that the reader
should have no trouble figuring out what the missing steps are. What this also means is: if you
cannot explain why a statement is true, then you cannot write that it is ‘obvious’
in a proof. You will need to make a judgement of how many steps it is OK to skip.

You will quickly get used to what is and what is not acceptable as a proof—assuming you
do the weekly exercises—because your class teacher will correct you. What you should keep in
mind is that whatever you write as a proof should be something which you could expand out to
a gold standard proof if you were forced to, either from memory or because you know where to
look for the missing pieces and previously proved statements.
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 27

As we go on, those ‘missing pieces and previously proved statements’ will get pretty long:
there will be proofs you write later this year in a page or two which might take a hundred or
more pages to write out in ‘gold standard’ style. For an example (which you shouldn’t expect to
understand when you read this the first time; but it will make sense when you’re revising) think
about how to prove that a piece of simple algebra with the rational numbers makes sense, in
terms of the axioms for the natural numbers. We prove in this course that you can do it (which
is enough—if I know something is possible, I don’t have to actually do it to check it works)—but
try actually doing it!

2.6 How to prove it

As you will soon see, it is not easy to find proofs. Sometimes you will be asked to prove a
statement where there is an ‘obvious’ way to proceed—as soon as you understand the statement,
you have an idea what to try—but mainly you will not see what to do at first. For some (most!)
true statements, no-one has ever figured out a proof; you shouldn’t feel bad that you do not
find it easy!
However, there are some strategies which you can use to help. The thing to keep in mind is
that
a proof is a sequence of implications, but that is not normally the order in which you think of it.
What that means is that you may not see how to get started—or maybe you know what
the first thing to do is, but not what comes next in the proof—but you perhaps can see that
if you could prove some statement S, then that would imply what you want to prove. If S is
‘easier’ somehow than the conclusion you want to get to, then that’s progress. Sometimes it can
be easier to start at the ‘end’ of the proof and ‘work backwards’. You have to be a bit careful
doing this—see Mistake 4 below—but it is still a good strategy.
My suggestion, if you think you would like to ‘work backwards’ to solve a problem, is to
write the conclusion at the bottom of the sheet and literally work backwards, writing up the
page, occasionally adding stuff at the top, and try to meet in the middle. You’ll probably have
a big gap in the middle when you’re done, but that is fine (it’s certainly better than running
out of room). If you really don’t like it, recopy the proof on a fresh sheet of paper.
However, I can’t do that in printed notes to give an example, so I will use different colours.
I’m first going to simply write out a proof, then explain what the colours mean and how I got
to it.
a2 +b2
Example 2.5. Prove that for all real numbers a and b we have ab ≤ 2 .

2 2
Proof. For all real a, b we have ab ≤ a 2+b .
2 2
Let p and q be real numbers. We have pq ≤ p +q2 .
We have 2pq ≤ p2 + q 2 .
We have p2 − 2pq + q 2 ≥ 0.
We have (p − q)2 ≥ 0.
Since p and q are real numbers, p − q is a real number. Since the square of any real number is
non-negative, we have (p − q)2 ≥ 0.
Expanding the brackets, we have p2 − 2pq + q 2 ≥ 0.
2 2
Rearranging, we get pq ≤ p +q 2 .
2 2
Since we proved pq ≤ p +q 2 for an arbitrary pair p and q of real numbers, we can conclude that
2 2
for all real a, b we have ab ≤ a 2+b .
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 28

What is going on here? The black text on the left is the proof we wanted. I’ve written it
out in a bit more detail than you would maybe feel necessary, in order to mention a couple of
important points. The red text on the right is the ‘current aim’—this is what we want to prove,
we have not yet proved it! The first line is simply repeating the text of the example. Let me
repeat what this aim is, in English. It is:
Pick any two real numbers. Then their product is at most half the sum of their squares.
Next, we pick a couple of real numbers p and q. We don’t assume anything about them apart
from that they are real numbers—that’s what the word ‘arbitrary’ means. We want to check
that for this particular pair of real numbers, we have the inequality we want—so the current
aim (the red text on the right) gets simpler. This is a standard approach to proving ‘for all’
statements; again, we’ll say more about this later.
At this point, I don’t see how to proceed ‘forwards’ in the proof; it’s not obvious what
the next black line should be, because the ‘aim’ inequality is complicated. So I try to ‘work
backwards’ and rearrange the ‘aim’ to something easier. That’s the next few red lines: get rid
of fractions, collect all the terms on one side, try to factorise—these are all things you can try.
If one doesn’t turn out to help, no problem, try another! In this example, we get to the nice
simple aim (p − q)2 ≥ 0.
Now I have reached an aim which I know how to prove true, so I write it down (that’s the
next black line). Finally, I can write out the rest of the proof, by writing out the red lines in
reverse; if you were trying this following the suggestion to work literally backwards from the
bottom of the paper, you’d already have written these lines from the bottom of the paper, and
this is where you would stop.
Finally—check that this proof makes sense! Does each black line really follow from the
previous ones?

I would be perfectly happy with a proof like:

a2 +b2
Proof. Let a, b be any real numbers, then (a − b)2 ≥ 0, rearranging we get ab ≤ 2 so we are
done.
Reading this proof, there is a ‘magic step’: for some reason we write (a − b)2 ≥ 0 and it is
completely unclear how we thought of writing that. We can check it works, but we don’t get
any idea from this of how to find such a proof. You know how—and more or less always, if you
read a proof and there is a ‘magic step’, there is some kind of reason, some thought process
which hasn’t been written down. If you try to follow this course by just reading all the solutions
rather than actually trying to do the exercises, then what you will not learn is how to find these
‘magic steps’. Since that will be tested in the exam, you will then suffer.

It’s important to be a bit careful about what is going on with the ‘for all’, because many
students get confused here. Read this now, but come back and re-read it once you get to the
end of the next chapter and we have formally discussed quantifiers.
When we write ‘for all a, b ...’ the a and b are placeholders (we say bound variables, as
opposed to the free variables that appear in predicates) that we introduce just in order to write
the inequality conveniently. If you change these two letters for any others, it doesn’t change
the meaning of the sentence, or indeed if you write it in English without any algebra at all (as
in the box above). It doesn’t make sense to talk about ‘what a is’ on the first line; a is just a
placeholder. This is why I used different letters p and q on the second line: here we declare that
for the rest of the proof, we are going to work with a particular pair of real numbers p and q,
and they won’t change from line to line. I won’t normally bother with this (because normally
we are too lazy to use new letters) but you should be aware that this is a little bit naughty.
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 29

Finally, we wrapped up the proof by stressing that what ‘for all’ means is a promise: ‘pick
any pair of real numbers, check the inequality for that particular pair, and you will find that it
is a true inequality.’

Warning 2.6. Is the following logic valid?

Since we picked a pair of real numbers a, b, actually we have (a − b)2 > 0, so we could say that
2 2
for all real a, b we have ab < a 2+b .
2 2
The answer is no. It is not true that for all real numbers a and b we have ab < a 2+b . For
example, it is not true for the real numbers 1 and 1 (as you can check). When we say ‘for all
a, b..’ we do include the possibility that a and b are in fact the same.

One final point to note is that this use of red text on the right in a proof is not standard ;
don’t expect to see it elsewhere. This is just my best attempt to show you how we get to a proof.
I’ll do this in several proofs later in the notes: it will always be the case that if you completely
ignore the red lines, what you have is a complete proof. If you are ‘working backwards’, you can
avoid having to write red lines by literally working back from the bottom of the page; if you
want to copy my red lines style, feel free, but think of the red lines as being part of your rough
work that should be crossed out once you figured out and wrote down the complete proof.

2.7 What is not a proof ?

There are several common mistakes made by students when they are asked to prove something.
I’ve mentioned one already, and more will appear later in the notes. But here are the ‘three
classics’ which I would like you not to repeat.
Mistake 2 (The goose’s mistake, ‘proof by example’). In January, a goose hatches from an
egg. Every day, the farmer feeds it. Towards the middle of December, the goose is sure that it
will be fed every day forever...
Whenever you are supposed to prove ‘for all...’ statements, you need to do all the cases not
one or two; whenever you want a counterexample to ‘there exists...’ statements, than means you
have to show all the possibilities fail, not just that the most obvious one fails. This probably
sounds obvious written out like this, but nevertheless probably about half of you will make the
goose’s mistake at some point.
Mistake 3 (The ends justify the means). You are in a park and buy an ice-cream; a small
child snatches it away from you. In the end, you will get your ice-cream back—explain how.
That means: write a story. The first and last lines are given: ‘You are in a park and buy an
ice-cream; a small child snatches it away from you’ and ‘You get your ice-cream back’. What’s
in the middle is important. Maybe it’s ‘You have a long discussion of comparative morality with
the child. It realises the error of its ways’.
You’re used to ‘doing maths’ meaning making a calculation, and the point of a calculation
is to ‘get the right answer’. Now, of course, it can happen that you make two mistakes in a
calculation which happen to cancel out and you get the right answer even though you made
mistakes—but you have to be really lucky for that to happen. Normally, if you make mistakes
you get the wrong answer. So you’re used to thinking (maybe subconsciously) that if the last
line is right, then everything else was probably also good.
We’re not doing calculations in this course, though, we’re doing proofs. When you write a
proof, you usually know the first and last lines before anything else: the first line is what you’re
assuming, and the last line is what you want to prove. What is important is actually what’s in
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 30

the middle which explains why the last line is true. If (when) you get a proof back from your
class teacher marked as wrong even though ‘the answer is right’, before complaining, think: does
it make a difference to the story if the middle line is instead ‘You pull out your gun and shoot
the child’ ?

Mistake 4 (Backwards thinking). Working in reverse to obtain a proof but then not writing
the proof out forwards.

For example, consider trying to prove the following trigonometric identity: for all real numbers
x, we have
(cos x)2 − sin x = 1 − (sin x)2 + sin x . (2.1)
If you just work in reverse, your proof might be:
Proof. Fix a real number x.

We want (cos x)2 − sin x = 1 − (sin x)2 + sin x

so − sin x = 1 − (sin x)2 + sin x − (cos x)2 subtracting (cos x)2
2
so (sin x)2 = (1 − (sin x)2 + sin x − (cos x)2 ) squaring both sides
so 0 = (1 − 1 + sin x)2 − (sin x)2 = 0 subtracting (sin x)2 ,

where to get to the last line we used the identity (sin x)2 + (cos x)2 = 1, which holds for all real
numbers x by Pythagoras’ Theorem. The last line is true, so we are done.
Note that normally you wouldn’t write justifications for each line of simple algebra—it’s
obvious enough how we got from each line to the next—but I wanted to do this here for extra
clarity.
This looks a lot like what we did in the last section to prove Example 2.5; it’s a lot like the
red rearranging-the-inequality lines there. We just didn’t bother to write the remaining black
lines out. What’s the problem?
What the above proof shows is that if the identity we want to prove, (2.1), holds, then 0 = 0,
which is a true statement. But that is the converse of the statement we want to prove, if 0 = 0
then (2.1) holds. (Which is the same as just saying that (2.1) holds: 0 = 0 is True.) We already
know that the converse being true doesn’t tell us if the original statement is true. If we want to
prove the original statement, we need to end with the statement we want to prove, not start
with it.
That might seem picky—let’s see what happens if we try to write it out in the ‘right order’.
Proof, take 2. Fix a real number x. We have

0 = (1 − 1 + sin x)2 − (sin x)2

2
so (sin x)2 = (1 − 1 + sin x) adding (sin x)2
2
so (sin x)2 = (1 − (sin x)2 + sin x − (cos x)2 ) since 1 = (sin x)2 + (cos x)2
so − sin x = 1 − (sin x)2 + sin x − (cos x)2 taking square roots
so (cos x)2 − sin x = 1 − (sin x)2 + sin x adding (cos x)2

which is what we wanted to prove.

CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 31

Looks better—but wait! In the last section, I told you to check the proof. The first two ‘so’s
are fine, but the third ‘so’, ‘taking square roots’, boils down to ‘If a2 = b2 then −a = b’—and
that’s not true; it could equally well be that a = b. There is a problem with the proof here—and
the reason is that we are trying to prove a false statement! In fact,
(cos π2 )2 − sin π2 = 02 − 1 = −1 but 1 − (sin π2 )2 + sin π2 = 1 − 12 + 1 = 1 .
so the ‘identity’ simply isn’t true.
What you should learn from this example is that it is not being picky to insist on writing
arguments (especially calculations with algebra) properly so that the statement to be proved
comes at the end not the beginning. It is very easy to do some operation to both sides which is
not reversible—in this example, squaring—without noticing and ‘prove’ a false statement. If
you write a proof properly, i.e. forwards, then you are more likely to notice a potential problem.

2.8 Sample exercises

Exercise 2.1. Is the following statement about natural numbers n true or false? Justify your
answer by giving a proof or a counterexample:
If n is divisible by 6 then n is divisible by 3.
What are the converse and contrapositive of this statement? Is the converse true? Is the contra-
positive true?
Exercise 2.2. Is the following statement about natural numbers n true or false? Justify your
answer by giving a proof or a counterexample:
If n is divisible by 2 then n is divisible by 4.
What are the converse and contrapositive of this statement? Is the converse true? Is the contra-
positive true?
Exercise 2.3. Prove that ¬(P ∧ Q) and ¬P ∨ ¬Q are logically equivalent.
Exercise 2.4. Prove that the negation of P ∨ Q is ¬P ∧ ¬Q.
Exercise 2.5. Prove by contradiction that there is no largest natural number.

2.9 Comments on selected activities

Comment on Activity 2.2. We can do this by constructing a truth table. Consider Table 2.8.
This proves that ¬(P ∧ Q) and ¬P ∨ ¬Q are equivalent.

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

Table 2.8: The truth tables for ¬(P ∧ Q) and ¬P ∨ ¬Q

Comment on Activity 2.3. The converse is ‘if n divides 12 then n divides 4’. This is false. For
instance, n = 12 is a counterexample. This is because 12 divides 12, but it does not divide 4.
The original statement is true, however. For, if n divides 4, then for some m ∈ Q, 4 = nm and
hence 12 = 3 × 4 = 3nm = n(3m), which shows that n divides 12.
CHAPTER 2. MATHEMATICAL STATEMENTS, PROOF, AND LOGIC 32

2.10 Solutions to exercises

Solution to Exercise 2.1. The statement is true. For, suppose n is divisible by 6. Then for some
m ∈ N, n = 6m, so n = 3(2m) and since 2m ∈ N, this proves that n is divisible by 3.
The converse is ‘If n is divisible by 3 then n is divisible by 6’. This is false. For example,
n = 3 is a counterexample: it is divisible by 3, but not by 6.
The contrapositive is ‘If n is not divisible by 3 then n is not divisible by 6’. This is true,
because it is logically equivalent to the initial statement, which we have proved to be true.
Solution to Exercise 2.2. The statement is false. For example, n = 2 is a counterexample: it is
divisible by 2, but not by 4.
The converse is ‘If n is divisible by 4 then n is divisible by 2’. This is true. For, suppose n is
divisible by 4. Then for some m ∈ N, n = 4m, so n = 2(2m) and since 2m ∈ N, this proves that n
is divisible by 2.
The contrapositive is ‘If n is not divisible by 4 then n is not divisible by 2’. This is false,
because it is logically equivalent to the initial statement, which we have proved to be false.
Alternatively, you can see that it’s false because 2 is a counterexample: it is not divisible by 4,
but it is divisible by 2.
Solution to Exercise 2.3. This can be established by using the truth table constructed in
Activity 2.2. See the solution above.
Solution to Exercise 2.4. This is established by Table 2.6. That table shows that ¬(P ∨ Q)
is logically equivalent to ¬P ∧ ¬Q. This is the same as saying that the negation of P ∨ Q is
¬P ∧ ¬Q.
Solution to Exercise 2.5. Let’s prove by contradiction that there is no largest natural number.
So suppose there is a largest natural number. Let us call it N . (What we want to do now is
somehow show that a conclusion, or something we know for sure must be false, follows.) Well,
consider the number N + 1. This is a natural number. But since N is the largest natural number,
we must have N + 1 ≤ N , which means that 1 ≤ 0, and that’s nonsense. So it follows that we
must have been wrong in supposing there is a largest natural number. (That’s the only place in
this argument where we could have gone wrong.) So there is no largest natural number. We
could have argued the contradiction slightly differently. Instead of using the fact that N + 1 ≤ N
to obtain the absurd statement that 1 ≤ 0, we could have argued as follows: N + 1 is a natural
number. But N + 1 > N and this contradicts the fact that N is the largest natural number.
 Sets and quantifiers
3
In this chapter, we discuss a fundamental concept in mathematics: sets. We need sets in order
to talk about quantification, which means talking about a statement being true ‘for all x’, or ‘for
some x’. It doesn’t really make sense to say that the statement (x + 2)2 = x2 + 2x + 4 is ‘true for
all x’—it’s not even clear what the statement should mean if x is a banana—but this statement
is true for all x in the set of real numbers. That’s quantification.
As with the previous chapter, there is nothing here that is difficult. However, sets and
quantification are not intuitive, and unless you pay attention, you will fall into a whole collection
of traps.
Actually, let me clarify that a bit. Sets and quantification are not intuitive yet. Once you
get to the point where you automatically avoid all the traps in this chapter without having to
think about it, you’re most of the way to the ‘thinking like a mathematician’ which is what
your future employer is looking for. It will happen, ideally before you sit the exam.

3.1 Sets
You have probably already met some basic ideas about sets and there is not too much more to
add at this stage, but they are such an important idea in abstract mathematics that they are
worth discussing here.
If you look around on the Internet, you might run into some things talking about ‘set theory’
and saying that this is all very subtle, and ‘unproveable’ and such things. This is not what we
are going to do. We are going to take a very simple view of sets (sometimes called naı̈ve set
theory). We are not going to go looking for trouble, and we will not find it, so don’t worry. If
you are curious about what trouble you might find if you insist on looking for it, see Section 3.6.

3.1.1 Basics
Loosely speaking, a set may be thought of as a collection of objects. A set is usually described
by listing or describing its members, or elements, inside curly brackets. For example, when we
write A = {1, 2, 3}, we mean that the objects belonging to the set A are the numbers 1, 2, 3 (or,
equivalently, the set A consists of the numbers 1, 2 and 3). Equally (and this is what we mean
by ‘describing’ its members), this set could have been written as

A = {n ∣ n is a whole number and 1 ≤ n ≤ 3}.

Here, the symbol ∣ stands for ‘such that’. Often, the symbol ‘∶’ is used instead, so that we might
write
A = {n ∶ n is a whole number and 1 ≤ n ≤ 3}.

33
CHAPTER 3. SETS AND QUANTIFIERS 34

When x is an object in a set A, we write x ∈ A and say ‘x belongs to A’, or ‘x is in A’, or ‘x

is a member of A’. If x is not in A we write x ∈/ A.
As another example, the set
B = {x ∈ N ∣ x is even}
has as its members the set of positive even integers. Here we are specifying the set by describing
the defining property of its members.
One point which is important is that it doesn’t make sense to say that an object is in a
set twice. It’s either in or not, and this is the end. We’ll avoid writing obvious repetitions, like
S = {1, 2, 3, 1}. That is a set, and it is the same as the set {1, 2, 3}; whichever way I write it, it
contains 1, 2 and 3 and nothing else. But sometimes it will be painful to write a description
avoiding repetition.
Sometimes it is useful to give a constructional description of a set. For example,
C = {n2 ∣ n ∈ N} is the set of natural numbers known as the ‘perfect squares’.
We could also write D = {z 2 ∣ z ∈ Z}, where Z is the set of all (not just positive) integers. The
difference between C and D is simple: D contains 0 and C does not. That’s the only difference.
By definition (−3)2 = 9 is in D, but it is also in C, because 32 = 9 is by definition in C. It doesn’t
matter that our definition of D repeats some elements (like 9 = (−3)2 = 32 ).
The set which has no members is called the empty set and is denoted by ∅. The empty
set may seem like a strange concept, but it is useful to define. Think about lengths—‘zero
centimetres’ is a funny length, but if we didn’t want to use it, we would have trouble with the
question ‘How much longer is a metre than 100 centimetres?’.
The short version of all of this is: when you want to define a set, you write the complete
description within the curly braces, in any of the ways above.
Warning 3.1. You do need to be careful about writing down sets. When we write
B = {n ∈ N ∣ n is even}, the letter n is a bound variable: a placeholder. It doesn’t mean
anything to ask ‘what is n in B’. It makes no difference if you change the letter n for another
letter.
Rather often students intend to write down a set, but put some of the description outside
the curly braces. Things like writing ‘B = {n ∈ N} n is even’. This particular example doesn’t
make sense—in English, it reads ‘B is the set of all natural numbers n is even’ which looks like
someone forgot a full-stop. Sometimes though, it does make sense but isn’t what was intended.
Either way, this is guaranteed to lose marks.
You may find this picky: bear in mind that in your future job you’ll probably ask computers
for answers rather often. If you cannot formulate the question accurately, the computer will
either complain or happily give you an answer to a question you didn’t mean to ask. This is no
different.
CHAPTER 3. SETS AND QUANTIFIERS 35

3.1.2 A note on notation

You should notice that the ‘is a member of’ symbol ∈ is written by drawing a semicircle on its
side, lifting the pencil and putting a bar from the middle. There is another symbol ε which
looks rather similar; this symbol is drawn in one stroke. This second symbol is the Greek letter
epsilon.
Many students in recent years confuse these two symbols. I don’t know why—maybe your
teacher at school used ε for ‘is a member of’. What I do know is that you must stop doing
this. Later this term, you will be using the Greek letter ε a great deal—in Analysis—and you
will at the same time be working with sets. If you draw ∈ and ε in the same way, you will end
up writing ‘for all εε(0, 1)’ and having to remember that one of these ε symbols is supposed to
be a real number and the other means ‘is a member of’. It’s rather easier to see what’s going on
if you write ε ∈ (0, 1).
Getting into the bad habit of writing ε when you mean ∈ will make your life difficult, especially
when you do it in the exam and lose marks unnecessarily.
Similarly, the brackets that go around sets are { and }. They are not ( and ). Nor [ and
]. Not even ⟨ and ⟩. Those other kinds of brackets all have different meanings in mathematics
(we’ll see all but ⟨ and ⟩ in this course).
In general, the more mathematics you do, the more symbols you will encounter, and the
more your life will become difficult if you cannot write them distinctly or if you misuse them.
You’ve no doubt already noticed at school that × and x are distinct symbols and that if you
write the latter with two diagonal lines, then you probably at some time tried to cancel an x
with a multiplication and got the wrong answer. Similarly, 2, z and Z are sometimes written
indistinguishably, with similar consequences. If we can’t tell whether your exam answer is correct
because we cannot distinguish the symbols you use, then we cannot give you the marks; that
would be a particularly silly way to not get the First you want.

3.1.3 Set equality

We’ve already written = between two sets above, but let’s be completely clear what it means. So
far, we saw = only to talk about when two numbers are equal—that’s something you’re so used
to that you don’t think about what it means (which is fine). But we need to define set equality.
Suppose A and B are two sets. We can write A = B when

for all x we have x ∈ A ⇐⇒ x ∈ B .

Let’s see why {1, 2, 3} = {1, 2, 3, 1} according to this definition. We have to check a certain
predicate (namely x ∈ {1, 2, 3} ⇐⇒ x ∈ {1, 2, 3, 1} ) is true for every x. Well, for x = 1 it’s true, 1
is in both sets. For x = 2 it is true, 2 is in both sets. For x = 3 it is true, 3 is in both sets. For
x = 4 it is true, 4 is in neither set. For x = banana it is true, banana is in neither set. And so on...
for any x except the ones we already checked, the predicate is true because x is in neither set.

3.1.4 Subsets
We say that the set S is a subset of the set T , and we write S ⊆ T , if every member of S is a
member of T . For example, {1, 2, 5} ⊆ {1, 2, 4, 5, 6, 40}. (Be aware that some texts use ⊂ where
we use ⊆.) What this means is that we have

for all x we have x ∈ A Ô⇒ x ∈ B .

CHAPTER 3. SETS AND QUANTIFIERS 36

A rather obvious, but sometimes useful, observation is that, given two sets A and B, A = B
if and only if A ⊆ B and B ⊆ A. So to prove two sets are equal, we can prove that each of these
two ‘containments’ holds. That might seem clumsy, but it is, in many cases, the best approach.
For any set A, the empty set, ∅, is a subset of A. You might think this is strange, because
what it means is that ‘every member of ∅ is also a member of A’. But ∅ has no members—how
can that be true? Let’s go back to the logic: ‘every member of ∅ is also a member of A’ means
‘for each x, if x in ∅ then x ∈ A’. Check the truth table of if—then ( Ô⇒ ). The only way some
x can be a counterexample to this statement is if x is in ∅ and not in A. But there is no x such
that x ∈ ∅, by definition—so we proved ∅ ⊆ A.

It’s very easy to get confused about what sets are equal, what are members and what are
subsets of a set. I’m about to give an example, which right now will look like a deliberate attempt
to trick you. But things like this will show up later, not as a trick, and you need to get it right.
Warning 3.2. Consider the set S = {0, 1, {0, 1}, {2}}. What are its members and subsets?
Well, 0 is a member. And so is 1, and so is {0, 1}, and so is {2}. But 2 is not a member
of S. Furthermore, {0, 1} is a subset of S (because 0 and 1 are both members of S) and so is
{{0, 1}}. These are two different sets—{0, 1} ≠ {{0, 1}}. And there are some other subsets of
S too—try to write them all out; you should get 16 in total.

If you don’t like the statements above, maybe think of it this way. Any (mathematical)
object can go in a set, so the number 1 can go in, or a function can go in, or even another set.
This is just the same thing as saying that you can put a (normal) object in a parcel, so an apple
can go in a parcel, or an orange can go in a parcel, or a parcel full of sweets can go in another
parcel, and so on. If you think a parcel containing a parcel full of sweets is the same as a parcel
full of sweets (or it’s the same as just having a lot of sweets), think back to childhood games of
Pass-the-Parcel. Just like that game, it really matters how many of the { and } set brackets
there are, and what exactly they go round.

3.1.5 Unions and intersections

Given two sets A and B, the union A ∪ B is the set whose members belong to A or B (or both
A and B): that is,
A ∪ B = {x ∣ x ∈ A or x ∈ B} .
Equivalently, writing out the definition of set equality:

for all x we have x ∈ A ∪ B ⇐⇒ (x ∈ A) ∨ (x ∈ B) .

Example 3.3. If A = {1, 2, 3, 5} and B = {2, 4, 5, 7}, then A ∪ B = {1, 2, 3, 4, 5, 7}.

Similarly, we define the intersection A ∩ B to be the set whose members belong to both A
and B:
A ∩ B = {x ∣ x ∈ A and x ∈ B} .
In other words,
for all x we have x ∈ A ∩ B ⇐⇒ (x ∈ A) ∧ (x ∈ B) .
CHAPTER 3. SETS AND QUANTIFIERS 37

3.1.6 Arbitrary unions and intersections

Often we will want to take the union of a lot of sets, for example A1 ∪ A2 ∪ A3 ∪ A4 ∪ A5 . This is
a pain to write out in this way, and if we wanted to take the union of infinitely many sets, we
wouldn’t be able to do it at all. So we define a notation which lets us write such a thing easily.
Suppose that I is a set, which we will call the index set, and that for each i ∈ I we have some
set Ai (so in the example above, I = {1, 2, 3, 4, 5}). Then we define the arbitrary union
⋃ Ai = {x ∣ x ∈ Ai for at least one i ∈ I} .
i∈I
The phrase ‘index set’ is supposed to help the reader: it is telling you ‘this set is here so that we
can put it under a ⋃’. It doesn’t mean that I is in any way special.
Similarly, we define the arbitrary intersection
⋂ Ai = {x ∣ x ∈ Ai for all i ∈ I} .
i∈I

You should check for yourself that

⋃ Ai
i∈{1,2,3,4,5}

really defines the same set as A1 ∪ A2 ∪ A3 ∪ A4 ∪ A5 , and similarly with the arbitrary intersection.
What do these definitions mean if I = ∅? It’s not very obvious, and we need to talk about
universal sets to understand it. We’ll get back to this later; for now, just think of ⋃ as a
convenient way to avoid writing a long string of ∪s.

3.1.7 Universal sets and complements

We’ve been a little informal about what the possible ‘objects’ in a set might be. In fact, we
haven’t been very clear about what exactly is and is not a set—this is a genuine difficulty. See
Section 3.6 for a brief discussion of this. In this course, we will take the (not very rigorous!)
point of view that anything we claim is a set, really is. In order for this to make some kind of
sense, we will always work with respect to some ‘universal set’ E. For example, if we are thinking
about sets of natural numbers, the universal set (the possible candidates for membership of the
sets we might want to consider) is the set N of all natural numbers.
This might seem like an unnecessary complication, but it is essential. Suppose I tell you
that the set A is the set of all even natural numbers. What are the objects that do not belong
to A? Well, in the context of natural numbers, it is all odd natural numbers. The context is
important (and it is this that is encapsulated in the universal set). Without that context (or
universal set), then there are many other objects that we could say do not belong to A, such as
negative integers, apples, bananas and elephants. (I could go on, but I hope you get the point!)
Given a universal set E and a subset A of E, the complement of A (sometimes called the
complement of A in E) is denoted by E ∖ A and is
E ∖ A = {x ∈ E ∣ x ∈/ A}.
If the universal set is clear, the complement of A is sometimes denoted by Ā or Ac (with textbooks
differing in their notation).
Suppose A is any subset of E. Because each member of E is either a member of A, or is not
a member of A, it follows that
A ∪ (E ∖ A) = E.
You should never worry ‘what is the universal set’ in this course. If you need to know it
(which is rare), it will be clearly stated what it is. If you don’t need to know it, you also don’t
need to worry about it.
CHAPTER 3. SETS AND QUANTIFIERS 38

3.1.8 Sets and logic

There are a great many comparisons and analogies between set theory and logic. Using the
shorthand notation for complements, one of the ‘De Morgan’ laws of complementation is that

A ∩ B = Ā ∪ B̄.

This looks a little like the fact (see Activity 2.2) that ¬(P ∧ Q) is equivalent to ¬P ∨ ¬Q. And
this is more than a coincidence. The negation operation, the conjunction operation, and the
disjunction operation on statements behave entirely in the same way as the complementation,
intersection, and union operations (in turn) on sets. In fact, when you start to prove things
about sets, you often end up giving arguments that are based in logic.
For example, how would we prove that A ∩ B = Ā ∪ B̄? We could argue as follows:

x∈A∩B ⇐⇒ x ∈/ A ∩ B
⇐⇒ ¬(x ∈ A ∩ B)
⇐⇒ ¬((x ∈ A) ∧ (x ∈ B))
⇐⇒ ¬(x ∈ A) ∨ ¬(x ∈ B)
⇐⇒ (x ∈ Ā) ∨ (x ∈ B̄)
⇐⇒ x ∈ Ā ∪ B̄.

What the result says is, in fact, easy to understand: if x is not in both A and B, then that’s
precisely because it fails to be in (at least) one of them.
For two sets A and B (subsets of a universal set E), the complement of B in A, denoted by
A ∖ B, is the set of objects that belong to A but not to B. That is,

A ∖ B = {x ∈ A ∣ x ∈/ B}.

Activity 3.1. Prove that A ∖ B = A ∩ (E ∖ B).

3.1.9 Cartesian products

For sets A and B, the Cartesian product A × B is the set of all ordered pairs (a, b), where a ∈ A
and b ∈ B. For example, if A = B = R then A × B = R × R is the set of all ordered pairs of real
numbers (‘the set of points in the plane’), usually denoted by R2 .
We can similarly define products of many sets. You’ve already seen this, for example

R × R × R = R3 = {(a, b, c) ∶ a, b, c, ∈ R} ,

which you’ve probably seen before as the set of points in space.

3.1.10 Power sets

For a set A, the set of all subsets of A, denoted P(A), is called the power set of A. Note that
the power set is a set of sets. For example, if A = {1, 2, 3}, then

P(A) = {∅, {1}, {2}, {3}, {1, 2}, {1, 3}, {2, 3}, {1, 2, 3}}.

Activity 3.2. Write down the power set of the set A = {1, 2, 3, 4}.
Activity 3.3. Suppose that A has n members, where n ∈ N. How many members does P(A)
have?
CHAPTER 3. SETS AND QUANTIFIERS 39

3.2 Quantifiers
We have already met the ideas of universal and existential statements involving natural numbers.
More generally, given any set E, a universal statement on E is one of the form ‘for all x ∈ E,
P (x)’. This statement is true if P (x) is true for all x in E, and it is false if there is some x in
E (known as a counterexample) such that P (x) is false. We have a special symbol that is used
in universal statements: the symbol ‘∀’ means ‘for all’. So the typical universal statement can
be written as
∀x ∈ E, P (x).
(The comma is not necessary, but I think it looks better.) An existential statement on E is one
of the form ‘there is x ∈ E such that P (x)’, which is true if there is some x ∈ E for which P (x)
is true, and is false if for every x ∈ E, P (x) is false. Again, we have a useful symbol, ‘∃’, meaning
‘there exists’. So the typical existential statement can be written as
∃x ∈ E, P (x).
Here, we have omitted the phrase ‘such that’, but this is often included if the statement reads
better with it. For instance, we could write
∃n ∈ N, n2 − 2n + 1 = 0,
but it would probably be easier to read
∃n ∈ N such that n2 − 2n + 1 = 0.
Often ‘such that’ is abbreviated to ‘s.t.’. (By the way, this statement is true because n = 1
satisfies n2 − 2n + 1 = 0.)
We have seen that the negation of a universal statement is an existential statement and
vice versa. In symbols, ¬(∀x ∈ E, P (x)) is logically equivalent to ∃x ∈ E, ¬P (x); and ¬(∃x ∈
E, P (x)) is logically equivalent to ∀x ∈ E, ¬P (x).
With these observations, we can now form the negations of more complex statements. Consider
the statement
∀n ∈ N, ∃m ∈ N, m > n.
Activity 3.4. What does the statement ∀n ∈ N, ∃m ∈ N, m > n mean? Is it true?
What would the negation of the statement be? Let’s take it gently. First, notice that the
statement is
∀n ∈ N, (∃m ∈ N, m > n).
The parentheses here do not change the meaning. According to the rules for negation of universal
statements, the negation of this is
∃n ∈ N, ¬(∃m ∈ N, m > n).
But what is ¬(∃m ∈ N, m > n)? According to the rules for negating existential statements, this
is equivalent to ∀m ∈ N, ¬(m > n). What is ¬(m > n)? Well, it’s just m ≤ n. So what we see is
that the negation of the initial statement is
∃n ∈ N, ∀m ∈ N, m ≤ n.
We can put this argument more succinctly, as follows:
¬ (∀n ∈ N(∃m ∈ N, m > n)) ⇐⇒ ∃n ∈ N, ¬(∃m ∈ N, m > n)
⇐⇒ ∃n ∈ N, ∀m ∈ N, ¬(m > n)
⇐⇒ ∃n ∈ N, ∀m ∈ N, m ≤ n.
CHAPTER 3. SETS AND QUANTIFIERS 40

Warning 3.4. This argument is succinct, but it is also hard to read, at least for me. Just to
understand what each line means requires some thought, and then some more thought to see
that it actually is equivalent to the previous line. It’s also fragile in the sense that making some
tiny change could break it.
In particular, the order of quantifiers is important. Change them, and you probably change
the meaning. If you change the order of the quantifiers in Activity 3.4, is what you get a true
statement? Try writing out what it means in English.

You want to prove an existential statement. That means you need to find one example.
There is a person who has run under 10 seconds for the 100m, because Usain Bolt did it.
You want to prove a universal statement. That means you need to check every single
possibility.
There is no person over 10 metres tall, because (you went round the world and measured the
heights of all 8 000 000 000 people).
You want to disprove an existential statement. That means you need to check every single
possibility doesn’t work —in other words, prove a universal statement.
You want to disprove a universal statement. That means you need to find one example where
it goes wrong—which is the same as proving an existential statement, and we normally call the
bad example a counterexample.

How do proofs actually look that do these things? I can’t help you much with proving an
existential statement (yet). Sit down and think about what the object you need to find is, and
hopefully at some point you can write down ‘Usain Bolt’ or ‘Lamont Marcell Jacobs’ or some
other example.
But there is a standard first thing to try if you are supposed to prove a universal statement.
If the statement is ‘for all z ∈ R, P (z)’ then the proof will often start ‘Pick z ∈ R’ or ‘Given
z ∈ R’. Then the aim is to prove P (z) for this one particular z.
We will get to using existential and universal statements later. You are told some universal
statement is true—what can you do with that information? It’s best to think of that as a
completely different thing to the process of proving existential and universal statements; again,
we’ll get to that later.

3.2.1 Quantifiers and arbitrary unions and intersections; empty sets

Another way of defining arbitrary union is

⋃ Ai = {x ∣ ∃i ∈ I, x ∈ Ai } ,
i∈I

and the arbitrary intersection is

⋂ Ai = {x ∣ ∀i ∈ I, x ∈ Ai } .
i∈I

Check that you see these definitions agree with the ones we gave earlier!
Now, what exactly do we do if I is an empty set? Well, for union it is intuitively clear: the
union of no sets had better be an empty set. That’s what the definition above says. If I is
empty, there is no i ∈ I, so whatever the condition after ‘∃i ∈ I’ is is irrelevant. The statement
‘∃X ∈ ∅, P (x)’ is False whatever P (x) is. This looks obvious written like this, but if P (x) is a
statement that looks ‘obviously true’ you will be tempted to say that ‘∃X ∈ ∅, P (x)’ should be
True, and then you will run into trouble.
CHAPTER 3. SETS AND QUANTIFIERS 41

For the arbitrary intersection, it is not so clear what the right answer should be—and in fact
we will avoid using this notation—but what the answer should be is that

⋂ Ai = E
i∈∅

where E is the universal set we’re working in. Why? Well, because ‘∀x ∈ ∅, P (x)’ is True
whatever P (x) is, so by definition every x we are considering is in the arbitrary intersection of
no sets. This might sound strange, and for sets it is a bit funny. But it is important in logic:
and again, if P (x) is some statement that looks ‘obviously false’ then you will be tempted to
say that ‘∀x ∈ ∅, P (x)’ should be False and get into trouble.

Quantifiers and the empty set can be a bit confusing.

Warning 3.5. Suppose ∀x ∈ X, P (x) is true. Is ∃x ∈ X, P (x) true?
You probably automatically say: yes, of course it is true! Pick an x in X, then we know P (x)
is true (because it is true for all x in X) and this is the example that shows ∃x ∈ X, P (x) is true.
But the answer is No!
How can that be? Well, if X is the empty set, then we cannot ‘pick an x in X’. There is
nothing to pick! The argument we gave works for any set X which is not empty, but it does not
work when X is the empty set.
We have ∀x ∈ ∅, P (x) is True whatever P (x) is (even if it is some ‘ridiculous’ statement like
‘x is a ten metre tall person’). This is because there is nothing we need to check in order to
prove it, or if you prefer, if we try to disprove it there is nothing in ∅ to be a counterexample.
We have ∃x ∈ ∅, P (x) is False whatever P (x) is (even if it is ‘x can run a 10 second 100m’),
because there doesn’t exist anything in ∅, so we never even get to the point of asking if it
satisfies P (x).
The name for this funny behaviour is we say a statement is vacuously true when it looks like
∀x ∈ ∅, P (x). We might say a statement is vacuously false if it looks like ∃x ∈ ∅, P (x), though
in practice that does not show up so often.
Even though this looks like nonsense, it turns out to be useful to allow it. Let X be the set
of monsters. I can prove that everything in X lives under a child’s bed. And everything in X is
purple. And everything in X is at least three metres tall. But child beds are less than 2 metres
long, so everything that lives under a child’s bed is less than 2 metres tall. So it turns out that
X is actually an empty set, because everything in X is simultaneously at least 3 metres tall and
less than 2 metres tall; there are no monsters.
If we said ‘you can’t quantify over empty sets’ then I would need to write something much
more complicated than ‘everything in X lives under a child’s bed’. I’d need to write ‘either X
is empty, or everything in X lives under a child’s bed’. I don’t want to keep having to write
something like that, and so (and only for that reason!) we allow it.
CHAPTER 3. SETS AND QUANTIFIERS 42

3.3 Proof by contradiction

We’ve seen a small example of proof by contradiction earlier in the chapter. Suppose you want
to prove a statement P . One way to do this is by contradiction. What this means is that you
suppose P is false, and you show that, somehow, this leads to a conclusion that you know,
definitely, to be false.
Here’s an example.

Example 3.6. There are no integers m, n such that 6m + 8n = 1099.

To prove this by contradiction, we can argue as follows:

Proof. Suppose that integers m, n do exist such that 6m + 8n = 1099. Then since 6 is even, 6n is
also even; and, since 8 is even, 8n is even. Hence 6m + 8n, as a sum of two even numbers, is
even. But this means 1099 = 6m + 8n is an even number. But, in fact, 1099 is odd, so we have a
contradiction. It follows that m, n of the type required do not exist.
This sort of argument can be a bit perplexing when you first meet it. What’s going on in the
example just given? Well, what we show is that if such m, n exist, then something impossible
happens: namely the number 1099 is both even and odd. Well, this can’t be. If supposing
something leads to a conclusion you know to be false, then the initial supposition must be false.
So the conclusion is that such integers m, n do not exist.
Probably the most famous proof by contradiction is Euclid’s proof that there are infinitely
many prime numbers1 . A prime number is a natural number greater than 1 which is only divisible
by 1 and itself. Such numbers have been historically of huge importance in mathematics, and
they are also very useful in a number of important applications, such as information security.
The first few prime numbers are 2, 3, 5, 7, 11, . . . . A natural question is: does this list go on
forever, or is there a largest prime number? In fact, the list goes on forever: there are infinitely
many prime numbers. We’ll mention this result again later. A full, detailed, understanding of
the proof requires some results we’ll meet later, but you should be able to get the flavour of it
at this stage. So here it is, a very famous result:

There are infinitely many prime numbers.

Proof. (Informally written for the sake of exposition) Suppose not. That is, suppose there are
only a finite number of primes. Then there’s a largest one. Let’s call it M . Now consider the
number
X = (2 × 3 × 5 × 7 × 11 × ⋯ × M ) + 1,
which is the product of all the prime numbers (2 up to M ), with 1 added. Notice that X > M ,
so X is not a prime (because M is the largest prime). If the number X is not prime, that means
that it has a divisor p that is a prime number and which satisfies 1 < p < X. [This is the key
observation: we haven’t really proved this yet, but we will later.] But p must therefore be one
of the numbers 2, 3, 5, . . . , M . However, X is not divisible by any of these numbers, because it
has remainder 1 when divided by any of them. So we have reached a contradiction: on the one
hand, X must be divisible by one of these primes, and on the other, it is not. So the initial
supposition that there were not infinitely many primes simply must be wrong. We conclude
there are infinitely many primes.
1
Historians of mathematics will probably tell you that Euclid’s proof is not a proof by contradiction. Which
is true, but I want to show you a proof by contradiction, so I am going to write down something which is not
actually what Euclid wrote (but it’s similar) and call it ‘Euclid’s proof’. What Euclid actually proved is ‘given
any finite list of prime numbers, there is a prime number not on the list’ and his proof does not use contradiction.
CHAPTER 3. SETS AND QUANTIFIERS 43

This proof has been written in a fairly informal and leisurely way to help explain what’s
happening. It could be written more succinctly and a bit more formally:
Proof. Suppose the set of prime numbers is not infinite. Then there are t prime numbers, for
some integer t. In other words, the set of prime numbers is {p1 , . . . , pt }. Consider the integer
N = (p1 × p2 × ⋅ ⋅ ⋅ × pt ) + 1. Now N is bigger than any of p1 , . . . , pt , so (by our assumption that
p1 , . . . , pt are all the prime numbers) it cannot be prime. And by construction N is not divisible
by any of p1 , . . . , pt (if we divide by any of them we have a remainder of 1). And since 2 and 3
are prime, certainly N is at least 7, in particular it is bigger than 1. But any integer bigger than
1 is either prime or it is divisible by a prime number, which is a contradiction.
This proof is still missing a few things—which you can see a bit more clearly because it’s
written formally. Why does the first sentence imply the second? Well, we didn’t formally define
the word ‘infinite’ yet. When we do, you’ll see that the second sentence is just writing out the
definition of ‘not infinite’, also known as ‘finite’. And we still didn’t prove the final sentence—but
hopefully it is a bit more clear what exactly we do need to prove. It’s worth thinking about this
a little bit now—what exactly is missing? We defined a prime number to be an integer greater
than 1 which is only divisible by 1 and itself. So we need to know what to do if we are given an
integer bigger than 1 which is not prime.
The other point which we should be careful about is the following. Suppose that we take the
first t prime numbers, multiply them together and add one. What we just proved is that either
we will get a new prime number or what we get will be divisible by a prime number which isn’t
one of the first t primes. We don’t have any idea which of these two things will happen. If you
try this for the first few values of t, you see
2+1=3
2×3+1=7
2 × 3 × 5 + 1 = 31
2 × 3 × 5 × 7 + 1 = 211
2 × 3 × 5 × 7 × 11 + 1 = 2311
which are all prime. It’s tempting to think this pattern will continue, but in fact
2 × 3 × 5 × 7 × 11 × 13 + 1 = 30031 = 59 × 509
is not prime.

3.4 Some terminology

At this point, it’s worth introducing some important terminology. When, in Mathematics, we
prove a true statement, we often say we are proving a Theorem, or a Proposition. (Usually the
word ‘Proposition’ is used if the statement does not seem quite so significant as to merit the
description ‘Theorem’.) A theorem that is a preliminary result leading up to a Theorem is often
called a Lemma, and a minor theorem that is a fairly direct consequence of, or special case of, a
theorem is called a Corollary, if it is not significant enough itself to merit the title Theorem.
For your purposes, it is important just to know that these words all mean true mathematical
statements.
You should realise that these terms are used subjectively. Some authors call Euclid’s result that
there are infinitely many prime numbers a Theorem, others call it a Proposition. Mathematically,
it doesn’t make a difference; the different words are just to help the reader—to give you an idea
of how hard the proof might be, or whether you should be interested in the statement for its
own sake or for what you can do with it.
CHAPTER 3. SETS AND QUANTIFIERS 44

3.5 General advice

3.5.1 Introduction
Proving things is difficult. Yes, I already said this, but it bears repetition. So far, everything
you did in mathematics was relatively easy—maybe it didn’t feel that way at the time, but
you probably always had the feeling that whatever problem your teachers set, you could do it.
Maybe you didn’t get it right first time, maybe you needed a bit more time than the teacher
gave you, but would you ever be so stuck that you would spend a week trying, checking your
answer, trying again... and never getting it right? Of course not.
Professional mathematicians are used to failure. If I tell a colleague I’ve got nowhere on some
problem for a week, they will probably wonder why I’m even bothering to tell them that—that’s
normal; in fact months or even years is normal. If I actually managed to solve a problem in less
than a week, I’d be likely to go around boasting about it!

In this course, we’ll try to give you a mix of problems. Some will not be harder than the
ones you did at school—these are there to check you understand the concept we just introduced.
A few will be either very hard (so that while I might know how to solve them, you probably will
not be able to) or even unsolved problems. These are there so that you get some experience with
trying something genuinely difficult and seeing how far you can get. Most will be somewhere in
the middle: harder than anything you did in school, but you can solve some or most of them,
in more or less time. These are the kinds of questions that will appear on the exam—but by
then you will have more experience and things you find difficult now will not be so bad any
more—and so training yourself to solve questions at this level will be needed to pass the course.
Inevitably, when you read a proof, in the textbooks or in these notes, you will ask ‘How did
the writer know to do that?’ (‘magic steps’) and you will often find you asking yourself ‘How
can I even begin to prove this?’. This is perfectly normal.
Look back to the two-line proof of Example 2.5. That proof has a ‘magic step’, but you know
how the writer thought of it. We will meet more proofs with magic steps in this course (and I
will generally try to explain why they are not really magic) and in future courses (where you
might be expected to figure things out for yourself a bit more), and there will always be some
reason why the step is not as magical as it seems.
We’ll discuss more strategies, more things to try, more tools to use, as we go on in the course.
At the same time, we’ll look at more difficult problems and more complicated concepts. You
may well feel the whole time that you are only barely coping with the course, and everything is
almost too hard. That’s what we are aiming for, more or less: to push your problem solving
ability to improve as fast as possible. Every so often, look back at the problems from the first
few weeks that you struggled with so that you can see how much you have moved on.
For now, the main thing to remember is: if you don’t try, you will never succeed. Try
something. You don’t have to justify to anyone why you should start with this particular
calculation, or why that theorem might help you. No-one will see your rough work. When you
fail, think about why—what is missing? What else could you try? Eventually you will get there.
This is a bit like integration—there are several methods, different substitutions and so forth; try
one until you get there. It’s more open ended in that there will be many more things to try.
One thing is vital: before you try to prove anything, you need to understand what it is that
you want to prove. That no doubt sounds totally obvious—but every year, I read lots of work
from students who obviously do not know what all the words in the question mean. If you do
not know what a word means, you have no chance of writing a correct solution! Look up the
definition. Then use the definition—there has to be a reason why that word is there!
CHAPTER 3. SETS AND QUANTIFIERS 45

This is particularly the case when a word has a meaning in mathematics and a meaning in
normal English, and these meanings are not the same. We saw that already with ‘implies’, and
there will be many more examples. You don’t get to choose, you have to use the mathematical
definition.
In general, you should expect that it takes time to read and understand even a rather short
mathematical statement. Take the time, look up any words you don’t know or are unsure about,
check that you know the meanings of all the symbols, and put all the pieces together. As a quick
example, what does A = B mean? Well, that depends what A and B are. Are they numbers?
vectors? sets? functions? In each of those cases, the symbol = means something different.

3.5.2 Definition chasing and cases

There is a ‘how to prove it’ strategy that comes together with ‘understand the statement’.
Namely, try replacing words (or symbols) with a defined meaning with the definition written
out. Often this can suggest a way to proceed.

Example 3.7. Prove that {x ∈ R ∣ x2 + 4 ≥ 8} ∩ N = {y ∈ Z ∣ y ≥ 2}.

The first thing to notice with this statement is that the = is a set equality. That has a
definition, so we might get somewhere by writing it in. Since this is the aim—this is what we
want to prove—we’re going to be changing our aim, i.e. working backwards, to start with. Let’s
first give the proof, then explain it a little bit.

Proof. {x ∈ R ∣ x2 + 4 ≥ 8} ∩ N = {y ∈ Z ∣ y ≥ 2}
∀z , (z ∈ {x ∈ R ∣ x2 + 4 ≥ 8} ∩ N) ⇐⇒ (z ∈ {y ∈ Z ∣ y ≥ 2})
∀z , (z 2 + 4 ≥ 8 and z is a positive integer) ⇐⇒ (z ≥ 2 is an integer)
Fix z. (z 2 + 4 ≥ 8 and z is a positive integer) ⇐⇒ (z ≥ 2 is an integer)

If z is not an integer, or z is not positive, then obviously both sides of

(z 2 + 4 ≥ 8 and z is a positive integer) ⇐⇒ (z ≥ 2 is an integer) are false, so the ⇐⇒ evaluates
to True.
If z = 1, then we can check both sides of the ⇐⇒ are false, so the ⇐⇒ evaluates to True.
If z ≥ 2 is an integer, then the right hand side of the ⇐⇒ is obviously true. For the left
hand side, we need to check that since z ≥ 2 we have z 2 ≥ 4, and so z 2 + 4 ≥ 8. That means that
in this case both sides of the ⇐⇒ are true, so the ⇐⇒ evaluates to True.
Since any z is either not an integer, or is not positive, or it is 1, or it is an integer at least 2,
we considered all the possibilities for z, and in each case we checked that
(z 2 + 4 ≥ 8 and z ∈ N) ⇐⇒ (z ≥ 2 is an integer) is true. So we can conclude

∀z , (z 2 + 4 ≥ 8 and z ∈ N) ⇐⇒ (z ≥ 2 is an integer)

By definition (of and and of the sets written out below), that is the same thing as

∀z , (z ∈ {x ∈ R ∣ x2 + 4 ≥ 8} ∩ N) ⇐⇒ (z ∈ {y ∈ Z ∣ y ≥ 2}) ,

and by definition, that is the same thing as

{x ∈ R ∣ x2 + 4 ≥ 8} ∩ N = {y ∈ Z ∣ y ≥ 2}

so we are done.
CHAPTER 3. SETS AND QUANTIFIERS 46

Again, the black lines are a complete proof. But we didn’t know how to get started without
thinking a bit first about what it is we actually wanted to prove. The first red line is just
repeating what we want to prove. The second red line is writing out the definition of set equality
in this particular example. That’s what I mean by ‘know and use’ the definition—it’s never
going to help much to simply copy the definition from your notes; what you need to do is to
write the definition as it applies to the thing you’re working with.
The third red line is, again, simply copying out the definitions as they apply in this example.
On the right, we’re simply filling in what it means for z to be in the set {y ∈ Z ∣ y ≥ 2}. On
the left, we’re filling in what it means for z to be in the conjunction of the two sets: namely
(by definition) it is a real number such that z 2 + 4 ≥ 8, and also it is a positive integer. Since all
positive integers are real, I didn’t bother to write the ‘is a real number’ bit. So far, our ‘current
aim’ has been getting longer each line, which looks like negative progress—but it is also getting
more concrete; we replaced abstract notation with things that you are familiar with. Generally
that means it will be easier to handle.
At this point, we can see a standard strategy to try. We’re supposed to prove a ‘for all’
statement, so let’s pick a particular z and try to prove it for that particular z. This gives us the
first black line of our proof, and (for the first time) the current aim actually gets shorter. What
we now have to prove is something simple. Saying that z is an integer at least 2 is supposed to
be the same thing as saying that z is a positive integer such that z 2 + 4 ≥ 8.
There are a few ways to proceed at this point, but the one I chose is to illustrate another
standard technique, ‘proof by cases’. At this moment, we said nothing about what z is. The ⇐⇒
statement we are trying to prove could be true for any of several different reasons, depending on
what z is. We simply list a bunch of reasons, called dividing into cases, and then check that
every z is covered by one of these reasons.
Once we checked that our cases are exhaustive—that is, any possible z falls into at least one
of them—then what we have proved is that for any z we have
(z 2 + 4 ≥ 8 and z is a positive integer) ⇐⇒ (z ≥ 2 is an integer). So we can write that down as
the next black line; and then we finish the proof off by recopying the red lines from earlier in
the reverse order, and checking that they really make sense written out forwards.

It’s important that you are happy with this proof. If not, you should talk to me or your class
teacher for a better explanation.

You may well feel that we’ve done a lot of unnecessary formalism to prove something ‘obvious’,
if you’re already happy with what the set notation means. This is a trap—it’s important to
be able to figure out how to write a formal proof from the definitions now, even though you
know how to write an intuitive and convincing explanation of why these two sets are the same
without bothering. This is because later (very soon) you’ll be dealing with statements which are
not so obvious, and you will not be able to rely on your intuition; then you need to be able to
get started with a formal proof.
In particular, you may well feel that the definition chasing we did—replacing the set equality
with its definition, and replacing the set membership and conjunction with their definitions—was
just some formal nonsense that you did not need in order to see why the statement is true. Later
in the course, the statements you can attack by definition chasing will not be obvious, but it
will still sometimes be the case that the only thing you need to do to get a solution is to replace
notation or terms with their definitions till you get to something obvious.

Finally, we saw a ‘proof by cases’ of a ‘for all’ statement. There will be lots more of these to
come. You haven’t seen anything like this before because simple algebra statements, when they’re
true, are true for exactly one reason—the calculation you do to prove them. More complicated
statements generally have multiple possible reasons for being true, as we saw here. If you’re
CHAPTER 3. SETS AND QUANTIFIERS 47

not happy with the logic, think of it the following way. Whatever z is, we need to provide a
reason why the predicate we’re looking at (the ⇐⇒ statement) is True for that particular z. If
there were only say 5 possible values of z, we could just do that by writing out each of the five
corresponding statements and checking them. Since there are infinitely many possible values of
z, we can’t do that.
But we can tell a Checker how they should go about checking any particular z they want.
You can imagine a dialogue with the Checker. The line ‘fix z’ means, we tell the Checker to
decide on a particular z that they should check; maybe it’s 5, or 0, or π, or banana. Then the ‘if
z is not an integer’ line means: we tell the Checker to first ask themselves whether their favourite
z is an integer; if it’s not, we explain to them why the ⇐⇒ is true (your favourite z is not an
integer, so both the left and right side of the ⇐⇒ come out to False, so the ⇐⇒ comes out
True). Then the next line tells the Checker what to do if z = 1, and so on. Finally we make sure
our cases are exhaustive—that means we are now confident that whatever z the Checker asks us
for help with checking, we have written down a reason for the Checker why the ⇐⇒ comes out
True.
In particular, ‘fix z’ does not mean that z is somehow ‘all the possibilities at once’.
How should you know to think about proving something by cases? This is simple to say
(but not always easy to do). If you can’t find one argument that works for every z, then find an
argument that works for some z, write it down, figure out which zs exactly it works for, and
then think about how to handle the other zs. Keep going until you find you’ve dealt with every
possible z.

3.5.3 How to write mathematics

You should write mathematics in English. You shouldn’t think that writing mathematics is
just using formulae. A good way to see if your writing makes sense is by reading it aloud (where
you should only read what you really have written, not adding extra words). If it sounds like
nonsense, a sequence of loose statements with no obvious relations, then you need to write it
again.
Don’t use more symbols than necessary.
Since many people seem to think that mathematics involves writing formulae, they often use
symbols to replace normal English words. An eternal favourite is the double arrow “Ô⇒” to
indicate that one thing follows from the other. As in :

x2 = 1 Ô⇒ x = 1 or x = −1.

This is not only pure laziness, since it’s just as easy to write :

x2 = 1, hence x = 1 or x = −1.

But it is even probably not what was meant! The implication arrow “Ô⇒” has a logical
meaning “if . . . , then . . . ”. So if you write “x2 = 1 Ô⇒ x = 1 or x = −1”, then that really means
“if x2 = 1, then x = 1 or x = −1”. And hence this gives no real information about what x is. On
the other hand, writing
I know x2 = 1, hence x = 1 or x = −1,
means that now we know x = 1 or x = −1 and can use that knowledge in what follows.
Some other unnecessary symbols that are sometimes used are “∴” and “ ∵ ”. They mean
something like “therefore/hence” and “since/because”. It is best not to use them, but to write
the word instead. It makes things so much easier to read.
CHAPTER 3. SETS AND QUANTIFIERS 48

Provide all information required.

A good habit is to start by writing what information is given and what question needs to be
answered. For instance, suppose you are asked to prove the following :

Problem 3.8. For any natural numbers a, b, c with c ≥ 2, there is a natural number n such that
an2 + bn + c is not a prime.

A good start to an answer would be :

Given : natural numbers a, b, c, with c ≥ 2.
To prove : there is a natural number n such that an2 + bn + c is not a prime.
At this point you (and any future reader) has all the information required, and you can start
thinking what really needs to be done.

3.5.4 How to do mathematics

In a few words : by trying and by doing it yourself !!
Try hard
The kind of questions you will be dealing with in this subject often have no obvious answers.
There is no standard method to come to an answer. That means that you have to find out what
to do yourself. And the only way of doing that is by trial and error.
So once you know what you are asked to do (plus all the information you were given), the
next thing is to take a piece of paper and start writing down some possible next steps. Some of
them may look promising, so have a better look at those and see if they will help you. Hopefully,
after some (or a lot) of trying, you see how to answer the question. Then you can go back
to writing down the answer. This rough working is a vital part of the process of answering a
question (and, in an examination, you should make sure your working is shown). Once you have
completed this part of the process, you will then be in a position to write the final answer in a
concise form indicating the flow of the reasoning and the arguments used.
Keep trying
You must get used to the situation that not every question can be answered immediately.
Sometimes you immediately see what to do and how to do it. But other times you will realise
that after a long time you haven’t got any further.
Don’t get frustrated when that happens. Put the problem aside, and try to do another
question (or do something else). Look back at the question later or another day, and see if it
makes more sense then. Often the answer will come to you as some kind of “ah-ha” flash. But
you can’t force these flashes. Spending more time improves the chances they happen, though.
Don’t get the idea that you are looking for ‘the right answer’. That might seem funny—
in every mathematics class you ever took so far, you were probably told that the point of
mathematics is ‘to find the right answer’. This is not true. We would like to know which
statements are true and which are false—but usually there are lots of different correct ways to
prove a statement is true. They are all ‘right answers’. So don’t be surprised if your answer to a
problem is not the same as the model solution but it is marked as correct—that just means you
found a different way to solve the problem, which is fine.
If you need a long time to answer certain questions, you can consider yourself in good
company. For the problem known as “Fermat’s Last Theorem”, the time between when the
problem was first formulated and when the answer was found was about 250 years.
Finally, you should not be unhappy if you find some problems you can’t solve at all. What
about the following: Suppose I take the first t primes, multiply them together and add one
(remember we saw this when we proved that there are infinitely many primes). We know the
CHAPTER 3. SETS AND QUANTIFIERS 49

result is sometimes prime and sometimes not, depending on t (we saw examples of both). Are
there infinitely many values of t such that we get a prime number? No-one knows the answer;
that problem has been open for over 2 300 years.
Do it yourself
Here is one (of many possible) solutions to Problem 3.8:
Given : natural numbers a, b, c, with c ≥ 2.
To prove : there is a natural number n such that an2 + bn + c is not a prime.
By definition, a natural number p is prime if p ≥ 2 and the only divisors of p are 1 and p itself.
Hence to prove : there is a natural number n for which an2 + bn + c is smaller than 2 or it has
divisors other than 1 or itself.
Let’s take n = c. Then we have an2 + bn + c = ac2 + bc + c.
But we can write ac2 + bc + c = c (ac + b + 1), which shows that ac2 + bc + c has c and ac + b + 1 as
divisors.
Moreover, it’s easy to see that neither c nor ac + b + 1 can be equal to 1 or to ac2 + bc + c.
We’ve found a value of n for which an2 + bn + c has divisors other than 1 or itself.
The crucial step in the answer above is the one in which I choose to take n = c. Why did I
choose that? Because it works. How did I get the idea to take n = c? Ah, that’s far less obvious.
Probably some rough paper and lots of trying was involved. In the final answer, no information
about how this clever idea was found needs to be given.
You probably have no problems following the reasoning given above, and hence you may
think that you understand this problem. But being able to follow the answer, and being able to
find the answer yourself are two completely different matters. And it is the second skill you
are suppose to acquire in this course. (And hence the skill that will be tested in the examination.)
Once you have learnt how to approach questions such as the above and come up with the clever
trick yourself, you have some hope of being able to answer other questions of a similar type.
But if you only study answers, you will probably never be able to find new arguments for
yourself. And hence when you are given a question you’ve never seen before, how can you trust
yourself that you have the ability to see the “trick” that that particular question requires ?
For many, abstract mathematics seems full of clever “tricks”. But these tricks have always
been found by people working very hard to get such a clever idea, not by people just studying
other problems and the tricks found by other people.

3.5.5 How to become better in mathematics

One thing you might consider is doing more questions. The books are a good source of exercises.
Trying some of these will give you extra practice.
But if you want to go beyond just being able to do what somebody else has written down,
you must try to explore the material even further. Try to understand the reason for things that
are maybe not explicitly asked.
As an illustration of thinking that way, look again at the formulation of the example we
looked at before :
For any natural numbers a, b, c with c ≥ 2, there is a natural number n such that
an + bn + c is not a prime.
2

Why is it so important that c ≥ 2 ? If you look at the proof in the previous section, you see
that that proof goes wrong if c = 1. (Since we want to use that c is a divisor different from 1.)
Does that mean the statement is wrong if c = 1 ? (No, but a different proof is required.)
And what happens if we allow one or more of a, b, c to be zero or negative?
And what about more complicated expression such as an3 + bn2 + cn + d for some numbers
a, b, c, d with d ≥ 2 ? Could it be possible that there is an expression like this for which all n give
CHAPTER 3. SETS AND QUANTIFIERS 50

prime numbers ? If you found the answer to the original question yourself, then you probably
immediately see that the answer has to be “no”, since similar arguments as before work. But if
you didn’t try the original question yourself, and just studied the ready-made answer, you’ll be
less well equipped to answer more general or slightly altered versions.
Once you start thinking like this, you are developing the skills required to be good in
mathematics. Trying to see beyond what is asked, asking yourself new questions and seeing
which you can answer, is the best way to train yourself to become a mathematician.

We’ve now reached the point in the course where you have all the basic tools you need to
start looking at problems. There will be more concepts to introduce in the next chapters, but
we will stop with introducing a new concept every page, and start spending much more time
finding out what we can do.

3.6 Non-examinable: set theory—take 2

What is a set, exactly? It’s supposed to be a mathematical object, which contains other
mathematical objects. That sounds like a definition—why not just say that anything goes; put
a bunch of objects in a bag and you have a set, which you can name (and in turn you can put it
in further sets.

One of the properties we would rather like to have sets to have is that we can write things
like
{n ∈ N ∶ n is even}
and say that this too is a set. More generally, if we have some statement P (s) (whose truth
depends on s) and a set S, we would like to say that {s ∈ S ∶ P (s) is true} is a set. We’ll see
that this kind of statement shows up continually throughout your degree programme.

Now, so far this looks fine—if ‘anything goes’ then certainly this is OK. But if ‘anything
goes’, we can also ask about the set of all mathematical objects—this would also be a set, let’s
call it U for ‘universe’. And we can write our favourite statement P (s), for example P (s) could
be the statement ‘s is not a member of s’. In that case we get a set

X = {s ∈ U ∶ P (s) is true} .

Now, you might notice this statement P (s) is a bit funny—how can a set possibly be a member
of itself? Well, actually if U is a set, then U is a mathematical object so U has to contain itself.
That might already raise a warning sign that strange things are going to happen, but it’s not
actually a logical contradiction; it’s just a bit funny.
But what about this set X? Well, by definition X contains everything which is not a member
of itself (and nothing else). So it certainly contains anything which isn’t a set (because something
which isn’t a set doesn’t contain anything at all, let alone itself). And it certainly contains
a lot of sets, like ∅ and {1, 2, 53}. OK, does X contain X? Well, if not, then by definition it
should. So X must contain X. But then by definition, X cannot contain X. That’s a logical
contradiction, pointed out by Bertrand Russell.
That’s really nothing more than a mathematical version of the ‘Barber of Seville’, who shaves
everyone in Seville that doesn’t shave themself. Who shaves the Barber?

What this logical contradiction tells us is that ‘anything goes’ is not OK. Some things are
not sets. We need to give some rules which allow you to construct new sets from old sets; some
axioms of set theory. This is what most mathematicians do (when we think about such things
CHAPTER 3. SETS AND QUANTIFIERS 51

at all!), and usually we use some axioms called ZFC (Zermelo-Fraenkel with Choice). These
axioms don’t, for instance, allow you to construct a ‘set of everything’; in fact, they don’t allow
any set to contain itself (because you have to construct new sets from old sets you already have).
These rules don’t—as far as we know—lead to logical contradictions like Russell’s. If you are
worried about trying to explain everything in mathematics, then a good place to start is with
ZFC set theory.
However, ZFC set theory is hard work; you spend a lot of time and energy proving things
which look ‘obvious’. We had to make a choice: do we spend all year building up the basics of
mathematics from set theory, so that you have one (hopefully) consistent foundation for the rest
of your degree? Or do we want to actually do some mathematics? We chose to do the latter,
which means that in this course we are going to assume some things are true without proving
them. In particular, we are going to assume statements like that there is such a thing as the set
of natural numbers N, that it makes sense to talk about sets of pairs such as {(a, b) ∶ a, b ∈ N},
and so on. All these are things which one can prove from the ZFC axioms, but we will not do so.
If you dislike this, you should go study ZFC set theory (in the summer, when you have
time!). However don’t expect it to be particularly easy, and don’t expect it to be an ‘answer to
everything’. You’ll still need to assume that ZFC set theory itself makes sense; there is no proof
that it makes sense.

3.7 Sample exercises

Exercise 3.1. Prove that for all real numbers a, b, c, ab + ac + bc ≤ a2 + b2 + c2 .
Exercise 3.2. Prove that there is no smallest positive real number.
Exercise 3.3. Suppose A and B are subsets of a universal set E. Prove that
(E × E) ∖ (A × B) = ((E ∖ A) × E) ∪ (E × (E ∖ B)).
Exercise 3.4. Suppose that P (x, y) is a predicate involving two free variables x, y from a set
E. (So, for given x and y, P (x, y) is either true or false.) Find the negation of the statement
∃x ∈ E, ∀y ∈ E, P (x, y) .

3.8 Comments on selected activities

Comment on Activity 3.1. We have
x∈A∖B ⇐⇒ (x ∈ A) ∧ (x ∈/ B)
⇐⇒ (x ∈ A) ∧ (x ∈ E ∖ B)
⇐⇒ x ∈ A ∩ (E ∖ B).
Comment on Activity 3.2. P(A) is the set consisting of the following sets:
∅, {1}, {2}, {3}, {4}, {1, 2}, {1, 3}, {1, 4}, {2, 3}, {2, 4}, {3, 4},
{1, 2, 3}, {2, 3, 4}, {1, 3, 4}, {1, 2, 4}, {1, 2, 3, 4}.
Comment on Activity 3.3. The members of P(A) are all the subsets of A. A subset S is
determined by which of the n members of A it contains. For each member x of A, either x ∈ S or
x ∈/ S. There are therefore two possibilities, for each x ∈ A. It follows that the number of subsets
is 2 × 2 × ⋯ × 2 (where there are n factors, one for each element of A). Therefore P(A) has 2n
members.
Comment on Activity 3.4. The statement means that if we take any natural number n there will
be some natural number m greater than n. Well, this is true. For example, m = n + 1 will do.
CHAPTER 3. SETS AND QUANTIFIERS 52

3.9 Solutions to exercises

Solution to Exercise 3.1. We work backwards, since it is not immediately obvious how to begin.
We note that what we’re trying to prove is equivalent to
a2 + b2 + c2 − ab − ac − bc ≥ 0.
This is equivalent to
2a2 + 2b2 + 2c2 − 2ab − 2ac − 2bc ≥ 0,
which is the same as
(a2 − 2ab + b2 ) + (b2 − 2bc + c2 ) + (a2 − 2ac + c2 ) ≥ 0.
You can perhaps now see how this is going to work, for (a2 − 2ab + b2 ) = (a − b)2 and so on.
Therefore the given inequality is equivalent to
(a − b)2 + (b − c)2 + (a − c)2 ≥ 0.
We know this to be true because squares are always non-negative. If we wanted to write this
proof ‘forwards’ we might argue as follows. For any a, b, c, (a − b)2 ≥ 0, (b − c)2 ≥ 0 and (a − c)2 ≥ 0,
so
(a − b)2 + (b − c)2 + (a − c)2 ≥ 0
and hence
2a2 + 2b2 + 2c2 − 2ab − 2ac − 2bc ≥ 0,
from which we obtain
a2 + b2 + c2 ≥ ab + ac + bc,
as required.
Solution to Exercise 3.2. We use a proof by contradiction. Suppose that there is a smallest
positive real number and let’s call this r. Then r/2 is also a real number and r/2 > 0 because
r > 0. But r/2 < r, contradicting the fact that r is the smallest positive real number. (Or, we
could argue: because r/2 is a positive real number and r is the smallest such number, then we
must have r/2 ≥ r, from which it follows that 1 ≥ 2, a contradiction.)
Solution to Exercise 3.3. We need to prove that
(E × E) ∖ (A × B) = ((E ∖ A) × E) ∪ (E × (E ∖ B)).
Now,
(x, y) ∈ (E × E) ∖ (A × B) ⇐⇒ ¬((x, y) ∈ A × B)
⇐⇒ ¬((x ∈ A) ∧ (y ∈ B))
⇐⇒ ¬(x ∈ A) ∨ ¬(y ∈ B)
⇐⇒ (x ∈ E ∖ A) ∨ (y ∈ E ∖ B)
⇐⇒ ((x, y) ∈ (E ∖ A) × E) ∨ ((x, y) ∈ E × (E ∖ B))
⇐⇒ (x, y) ∈ ((E ∖ A) × E) ∪ (E × (E ∖ B)).
Solution to Exercise 3.4. We deal first with the existential quantifier at the beginning of the
statement. So, the negation of the statement is
∀x ∈ E, ¬(∀y ∈ E, P (x, y))
which is the same as
∀x ∈ E, ∃y ∈ E, ¬P (x, y).
 Structures, natural numbers and proof by induction
4
The material in this chapter is also covered in:

• Biggs, N. L. Discrete Mathematics. Chapter 4.

• Eccles, P.J. An Introduction to Mathematical Reasoning. Chapters 1–4 and 6.

4.1 Introduction
In this chapter we will discuss what is meant by a ‘mathematical structure’, and explore some
of the properties of one of the most important mathematical structures: the natural numbers.
These will not be new to you, but they shall be explained a little more formally. The chapter
also studies a very powerful proof method, known as proof by induction. This enables us to
prove many universal statements about natural numbers that would be extremely difficult to
prove by other means.

4.2 Mathematical structures

A mathematical structure is a precisely specified object which one can study. We already saw,
informally, several examples in the course:
(1) The natural numbers N = {1, 2, 3, . . . } which come with the operations + and ×, and the
relation <.
(2) The integers Z which come with the operations + and ×, and the relation <.
(3) The rational numbers Q (intuitively, the fractions: numbers which you can write as ab where
a and b are integers and b is not zero), which again come with the operations + and ×, and
the < relation.
(4) The real numbers R (intuitively: points on the number line) which again come with the
operations + and ×, and the < relation.
(5) The complex √ numbers C which are numbers of the form a + bi, where i is a special symbol
representing −1. Again you can add and multiply these, but you don’t know any < on
these (and in fact, there isn’t any sensible choice).
All these examples are structures where you can do arithmetic as you’re used to it. Here are
another couple of examples. Don’t worry if you haven’t seen these before. We won’t try to study
them just yet; they will appear in Lent Term in MA103.

53
CHAPTER 4. STRUCTURES, NATURAL NUMBERS AND PROOF BY INDUCTION 54

(6) The ‘clock numbers’ Z24 , which are the integers {0, 1, 2, . . . , 23} on a 24-hour clock, where
you add and multiply as you would on a clock; if you get 24 you replace it with 0, if you get
25 you replace it with 1, and so on.

a b
(7) The 2 × 2 matrices ( ) where a, b, c, d are real numbers. Here too we can define addition
c d
and multiplication:

a b a′ b ′ a + a′ b + b ′
( ) + ( ′ ′) = ( ) and
c d c d c + c′ d + d ′
a b a′ b ′ aa′ + bc′ ab′ + bd′
( ) × ( ′ ′) = ( ′ ).
c d c d ca + dc′ cb′ + dd′

These still look like structures where you can ‘do arithmetic as you’re used to it’. But you
have to be a little careful now. In Z24 we have 4 × 5 = 20 = 4 × 11. So what should we say 20/4
is? You’re used to the idea that ‘division by zero’ doesn’t make sense, but in Z24 ‘division by
four’ also doesn’t make sense. When you work with 2 × 2 matrices, then multiplication turns out
not to be commutative:
0 1 1 0 0 −1 1 0 0 1 0 1
( )( )=( ) but ( )( )=( ).
−1 0 0 −1 −1 0 0 −1 −1 0 1 0

Here is a rather different example.

(8) The set of social networks, where a social network consists of a (finite) collection of people
and a relation ‘friends’ between pairs of people.

Think of taking a snapshot of the Facebook network at some moment: there are something like
1 000 000 000 people in the network, and if I look at any particular pair I will find they are either
friends or they are not. That’s a social network (by the definition we gave); if we let some time
pass, some people join or leave, some pairs of people friend or de-friend each other, we get a
different social network.
It’s not clear what + or × should mean here—how can we multiply social networks? But
I probably don’t have to convince you that there are interesting things to study here; and in
fact the (results of the) mathematical study of networks (‘Graph Theory’) turns out to be very
important in today’s technology. We’re not going to go further into this in MA103; the point of
giving this example is to show you that we can be interested as mathematicians in things which
don’t involve arithmetic.
More or less, any time you find a precise, unambiguous definition of something, then you have
a mathematical structure which you can start studying. Mathematics is a much broader subject
than the arithmetic you saw in school. A lot of mathematics is not about numbers. Of
course, not everything interesting is mathematics—you (maybe) find politics interesting, but
you will not be able to come up with a definition of ‘left-wing’ or ‘economically good’ which is
generally agreed on, let alone one which is precise and unambiguous. We’ll have to leave politics
to the political scientists. The flip side of this is: it’s (more or less) true that all mathematicians
agree that all of mathematics is correct, which keeps fights to a minimum. That’s certainly not
true for political scientists, who (sometimes) write books whose messages boil down to ‘My idea
is right’, ‘You’re wrong’, ‘Am not!’, ‘Wrongy wrongy wrong!’... and so on.
If you’re thinking carefully, you might notice that the structures we mentioned above aren’t
really very clearly defined. What are ‘the points on the number line’ ? In fact, what are ‘the
natural numbers’ ? We probably all feel we know what is meant by a positive integer, how to
CHAPTER 4. STRUCTURES, NATURAL NUMBERS AND PROOF BY INDUCTION 55

add and multiply them, and that all of us will get the same answers if we try it. But that’s not
good enough. It would be very embarrassing if it turned out that some of us made different
assumptions to others about the natural numbers, and we started arguing about what statements
are true.
The way we solve this in mathematics is to be very careful with assumptions. We will write
down a rather short list of assumptions, called axioms. And then we will prove that all the other
properties of the natural numbers which you are used to follow from those few axioms. This is
called the axiomatic approach to the natural numbers. We’ll develop it in MA103, in Lent Term,
as a warm-up to the axiomatic approach to groups and to abstract vector spaces—these are
structures which you quite possibly have not yet met, and about which you have no intuition.
The only way you can hope to prove anything about groups or abstract vector spaces is to get
good at working with axioms.
But for now, we will stick to structures that you are familiar with, like the natural numbers.
And we will not worry too much about justifying properties carefully from ‘axioms’, instead we
will get on with some mathematics.

4.2.1 Greatest and least elements

Let S be a subset of N. We say ` is a least element or least member of S if ` ∈ S and for all s ∈ S
we have ` ≤ s. Similarly, we say g is a greatest element or greatest member of S if g ∈ S and for
all s ∈ S we have g ≥ s.
It’s obvious that some subsets of N do not have a greatest element—for example N itself
doesn’t have a greatest element, nor does the set of even natural numbers, nor the set of primes
(this is what Euclid proved). And by definition the empty set ∅ doesn’t have either a least or a
greatest element: it doesn’t have any elements at all.
But every non-empty subset of N has a least element. This is called the Well-Ordering
Principle (or sometimes the Least Element Principle). It’s a rather special property of the natural
numbers, which doesn’t hold for many other structures, such as the real numbers.
CHAPTER 4. STRUCTURES, NATURAL NUMBERS AND PROOF BY INDUCTION 56

4.3 The principle of induction

4.3.1 Proof by induction
One particularly useful theorem that follows from the axioms of the natural numbers is the
following one, known as the Principle of Induction. Officially, this is a theorem—a statement
which we don’t assume is true, but which we prove. However, the proof will come next term; for
now, I will try to explain why it is plausible and then we will assume it is true.
The Principle of Induction: Suppose P (n) is a statement involving natural numbers n.
Suppose furthermore that the following two statements are true.
(i) P (1) is true; (we call this the ‘Base case’)
(ii) For all k ∈ N , P (k) Ô⇒ P (k + 1). (we call this the ‘Induction step’)
Then P (n) is true for all natural numbers n.
We aren’t going to prove this right now, but let’s give an intuition for why it is true.
We know that P (1) is true. We know that ∀k ∈ N , P (k) Ô⇒ P (k + 1). The second of these
is a bit complicated—it is saying that we have an infinite list of true statements:

P (1) Ô⇒ P (2) and P (2) Ô⇒ P (3) and P (3) Ô⇒ P (4) and so on .

Well, we can prove P (2) from this. We know P (1) is true, and we know P (1) Ô⇒ P (2) is
true. Look at the truth table for Ô⇒ ; the only way that that can happen is that P (2) is true.
Now we can prove P (3). We know (now!) that P (2) is true, and we know P (2) Ô⇒ P (3).
Again, the only way that can happen is that P (3) is true.
And so on...

That looks fairly convincing; and I said (truthfully) that this does prove P (2) and P (3).
Why is this not in fact a proof that the Principle of Induction is true? Well, in mathematics we
insist that a proof is always a finite argument: it has to be something you can get to the end of
and check, not an infinite sequence of statements, nor something finishing with a vague ‘carry
on like that’.
You will probably feel that this particular ‘and so on’ is clear enough that you would be
happy to accept this argument as valid, even though it doesn’t quite fit the definition of a
mathematical proof. This is a reasonable thing to say; though in Lent Term we will write down
some axioms for the natural numbers and use them to prove the Principle of Induction.
What is not fine, though, is saying the same thing about other similar arguments.
If you start allowing ‘and so on’ statements into proofs, it is easy to run into trouble. You
might miss something that works fine for the first ten steps but then goes wrong, because you
didn’t notice it before writing ‘and so on’. Your readers might guess a different pattern than you
intended for ‘and so on’—if I give you the sequence 1, 1, 2, 6, ... what is actually the next term?
there are a few integers you could reasonably argue for. Your readers might not be able to guess
at all what you meant ‘and so on’ to mean, because you are proving something complicated.
What certainly is the case is that your mathematics will no longer be something that anyone
can check and agree on; different readers might disagree on whether your proof is valid.
So we do not allow ‘and so on’ in proofs. If you can formulate clearly enough what you
intend ‘and so on’ to mean, then what you will find you have written is a proof by induction.
CHAPTER 4. STRUCTURES, NATURAL NUMBERS AND PROOF BY INDUCTION 57

4.3.2 An example
Here’s an example of how we might prove by induction a result we proved directly earlier, in
the previous chapter.

Example 4.1. Prove that

∀n ∈ N , n2 + n is even .

Suppose you looked at this statement for a bit, and didn’t notice the ‘trick’ we used earlier.
You would probably see what n2 + n is for a few integers first, to get some idea. 12 + 1 = 2 is even.
22 + 2 = 6 is even. 32 + 3 = 12 is even. Then you might think, the difference between consecutive
squares is always odd, and obviously the difference between consecutive integers is always odd,
so the difference between consecutive values of n2 + n is always even—if I know that n2 + n is
even, that tells me (n + 1)2 + (n + 1) is even. As soon as you start thinking that it would be
useful to know an earlier case to prove a later case, that generally means you want to write a
proof by induction. Here it is.
Proof. Let Q(n) be the statement ‘n2 + n is even’.
The base case is n = 1. The statement Q(1) is ‘12 + 1 is even’. That is true, because 12 + 1 = 2.
Fix a natural number k.
As an induction hypothesis, suppose Q(k) is true, i.e. k 2 + k is even.
We have (k + 1)2 + (k + 1) = k 2 + 2k + 1 + k + 1 = (k 2 + k) + 2(k + 1). Since k 2 + k is an even
number by the induction hypothesis, and 2(k + 1) is obviously even, we see that (k + 1)2 + (k + 1)
is even, which is Q(k + 1).
So for this k, we proved Q(k) Ô⇒ Q(k + 1), and since k ∈ N is arbitrary, we proved
∀k ∈ N , Q(k) Ô⇒ Q(k + 1), the induction step.
By the Principle of Induction, we conclude that Q(n) is true for all n ∈ N.
The reason why I used the letter Q rather than P is just to remind you that it’s not important
which particular letter we use.
Let’s recap the logic here quickly. The Principle of Induction says: if you know the base case
Q(1) and the induction step ∀k ∈ N , Q(k) Ô⇒ Q(k + 1) are true statements, then you also
know that ∀n ∈ N , Q(n) (which is our goal) is a true statement. So a proof by induction will
always mean proving the base case Q(1) (which is usually, as here, a simple calculation), and
then proving the induction step, and then saying ‘so we are done by induction’.
The induction step is a complicated statement: it is a universal statement, and the thing
inside the ‘for all’ that we want to show is itself an implication. Nevertheless, there is a standard
thing to try for both of these. Since we want to prove ∀k ∈ N , Q(k) Ô⇒ Q(k + 1), the proof of
the induction step will start ‘fix k ∈ N’ or ‘given k ∈ N’ (these mean the same) and then we just
have to prove the implication Q(k) Ô⇒ Q(k + 1) for this particular k, about which we are not
going to assume any more (it is ‘arbitrary’).
There is also a standard first thing to try when we want to prove an implication: assume
the premise Q(k). We give it the name induction hypothesis to help the reader; to remind them
that this is a standard part of the induction proof. We then just need to prove Q(k + 1) holds,
and somewhere along the way we presumably will use the statement Q(k) we assumed. I can’t
help you any more with this bit—this is usually the hard part of an induction proof, where you
need to figure out how in fact you want to prove your implication.
CHAPTER 4. STRUCTURES, NATURAL NUMBERS AND PROOF BY INDUCTION 58

4.3.3 Induction: why be careful?

At least for now, I’m going to insist that when you write a proof by induction, you really need
to write it out formally as in the examples in this chapter. I want to see the words ‘base case’
appearing with a proof of the base case, I want to see the words ‘induction step’ appearing
with a proof of the induction step, and then I want to see a final line like ‘so by the principle
of induction, ... ’. You can afford to give a little less detail than the example above—see the
examples below—but those features need to be present.
This is not (just) because I am picky; it is because induction is an easy thing to mess up and
‘prove’ something which isn’t true. Furthermore, later on you may well write a long complicated
proof that uses induction in two or three different places, and writing it out formally like this
gives you some structure and lets you see clearly where you are using induction and when you
are done.
You may get worried about why induction works—it can get confusing, when you have some
complicated statement which you are trying to prove, and especially if you are using some
variants of induction (see below). Keep in mind that while the predicate P (n) you’re working
with may be complicated, the logic of induction is just the simple logic above.
You may alternatively begin to feel that induction is obvious and it’s not clear why you
need all the careful formalities; the examples we will see next mainly look like ‘calculate the
first case, then just keep doing the same calculation over and over again’. Why can’t we simply
write in a proof ‘and now keep doing this calculation forever’ ? The answer is that it is easy to
write down something which looks convincing, where the ‘calculation you do forever’ works for
the first one or two times, but then it stops working because you missed some difficulty which
doesn’t show up in the first one or two cases. Induction is nothing more than ‘and now keep
doing this calculation forever’, except that writing out the formalities forces you to say in detail
exactly what calculation you will do and check it really works.
Finally, you need to avoid getting confused with what you have proved, when. When you
are proving the induction step, you have not proved the P (k) induction hypothesis. You’ve
simply assumed it’s true—you don’t know it. When you finish proving the implication with ‘so
P (k + 1) is true’ you have not proved P (k + 1), what you have proved is that if your induction
hypothesis P (k) is true, then so is P (k + 1).
There is a story which goes like this:
Johnny: I’ve made a diamond machine! If you give me some wood, I’ll turn it into diamonds!
Frank : Sounds good.
Johnny: I am going to sell half the diamonds and buy some gold!
Frank : Sounds good.
Johnny: I’ll pay a goldsmith to show me how to make a ring!
Frank : Sounds good.
Johnny: I’m going to make a diamond wedding ring!
Frank : Sounds good.
Johnny: Frankie baby, we’re getting married!
Frank : Johnny... wait a minute... I’m not sure I’m ready for that...
What has happened here is that Johnny has forgotten the base of his induction. If Frank
gives Johnny wood, he can make diamonds. If Johnny sells diamonds, he can buy gold, and so
on. These logical implications are all fine. But they do not prove a wedding is in the waiting.
Frank doesn’t want to get married, so he probably will not be giving Johnny wood.
CHAPTER 4. STRUCTURES, NATURAL NUMBERS AND PROOF BY INDUCTION 59

4.3.4 Variants
Sometimes you want to prove that a statement is true not for all positive integers (natural
numbers) but perhaps for all non-negative integers, or all integers at least 8, or something similar.
Something like induction still works, commonly called ‘induction with base case N ’. Here N is
some particular integer, which is the smallest case you want to prove (such as 0, or 8).
The Principle of Induction with base case N : Suppose P (n) is a statement involving
integers n ≥ N . Suppose furthermore that the following two statements are true.
(i) P (N ) is true; (the ‘Base case’)
(ii) For all integers k ≥ N , P (k) Ô⇒ P (k + 1). (the ‘Induction step’)
Then P (n) is true for all integers n ≥ N .
Note that the Principle of Induction is the same thing as the Principle of Induction with
base case 1. The more general statement above can be proved using the (original) Principle of
Induction: this is an exercise.
Example 4.2. Prove that
∀n ≥ 4 , n2 ≤ 2n .
Let’s notice that we can’t prove this by the usual induction. The ‘base case’ n = 1 is true, so
is the n = 2 case, but the n = 3 case is false; 32 is bigger than 23 . That means that we would get
stuck proving the induction step if we tried. Try to figure out the proof for yourself!
Activity 4.1. Prove that ∀n ≥ 4 , n2 ≤ 2n .
Another variant of the Induction Principle is the following, known as the Strong Induction
Principle:
The Strong Induction Principle: Suppose P (n) is a statement involving natural numbers
n. Suppose furthermore that the following statement is true.

∀k ∈ N , (∀t ∈ N such that t < k , P (t)) Ô⇒ P (k) .

Then P (n) is true for all natural numbers n.
It’s worth taking several minutes to think about this statement. What on Earth does it
mean? I don’t think the string of symbols in the middle is easy to understand, and I suspect
you are less happy with it than I am. Nevertheless, the ability to make sense of a statement like
this is an important skill you need to learn.
To begin understanding it, remember that the ∀k ∈ N is another way of saying ‘all of the
following infinite list of statements are true’, where the statements in question are

(∀t ∈ N such that t < 1 , P (t)) Ô⇒ P (1)

(∀t ∈ N such that t < 2 , P (t)) Ô⇒ P (2)
(∀t ∈ N such that t < 3 , P (t)) Ô⇒ P (3)
(∀t ∈ N such that t < 4 , P (t)) Ô⇒ P (4)
(∀t ∈ N such that t < 5 , P (t)) Ô⇒ P (5) and so on.

Does that help? Well, yes, a bit. We can recognise that over on the right, the conclusions of
these implications are P (1), P (2), and so on—we’re hoping to find that all those statements are
true. So presumably we expect to find all the premises are true, for some reason. The premises
are still complicated, so let’s replace the quantifier by writing out the lists of statements explicitly.
CHAPTER 4. STRUCTURES, NATURAL NUMBERS AND PROOF BY INDUCTION 60

These are all finite lists of statements. In fact, on the first line we are quantifying over an empty
set—there is no natural number less than 1—so for that line we need to check the definition to
remember that ‘for all’ things in an empty set is vacuously true. What we get is

true Ô⇒ P (1)
P (1) Ô⇒ P (2)
P (1) ∧ P (2) Ô⇒ P (3)
P (1) ∧ P (2) ∧ P (3) Ô⇒ P (4)
P (1) ∧ P (2) ∧ P (3) ∧ P (4) Ô⇒ P (5) and so on.

At this point, you can start to believe that this Strong Induction Principle makes sense. The
first line above is true; that tells us (check the truth table for Ô⇒ ) that P (1) is true.
But if P (1) is true, the second line tells us P (2) is true.
And then the third line says, since P (1) and P (2) are true, that P (3) is true. And so on.
This ‘and so on’ is of course not a proof of the Strong Induction Principle. But we can prove
it using the Principle of Induction.

Activity 4.2. Try to understand why the Strong Induction Principle follows from the Principle
of Induction. Hint: consider Q(n), the statement ‘∀s ≤ n, P (s) is true’.
This is difficult, so you may want to omit this activity at first.

Here is a reformulation, less ‘mathematically precise’ but maybe more useful, of the Strong
Induction Principle. Remember ‘asssuming P (1) we prove P (2) is the same thing as ‘we prove
P (1) Ô⇒ P (2)’, and so on.
The Strong Induction Principle: Suppose P (n) is a statement about natural numbers n.
Suppose furthermore that you can prove true Ô⇒ P (1) (i.e. you can prove P (1) ).
And, if you assume P (1), you can prove P (2).
In fact for every k ∈ N, if you assume P (1), P (2), . . . , P (k − 1) are true, you can prove P (k).
Then P (n) is true for all natural numbers n.
It’s immediately worth pointing out that just because you assume P (1), P (2), ..., P (k − 1)
when you want to prove P (k) doesn’t mean you have to use all of them in your proof. It just
means you can if you want to.
A standard question at this point is ‘what is the base case in strong induction? is it P (1)?’
The answer to this is a bit complicated—it can be yes, it can be no, it depends. This will be
easier to understand once you saw a few examples!
What is probably very unclear at this point is when or why you might want to use this
complicated-looking Strong Induction. The answer, below, is simple enough, but probably it is
not easy to understand until after you’ve read the next couple of sections.
Just as induction is what you should think of using when you try to prove a predicate P (n)
and think ‘it would really help me if I knew the last case P (n − 1) was true’, strong induction is
what you should think of using when you try to prove P (n) and think ‘it would really help me
if I knew one or several smaller cases were true’.
CHAPTER 4. STRUCTURES, NATURAL NUMBERS AND PROOF BY INDUCTION 61

4.4 Summation formulae

Suppose a1 , a2 , a3 , . . . is a sequence (an infinite, ordered, list) of real numbers. Then the sum
n
∑r=1 ar is the sum of the first n numbers in the sequence. It is useful to define these sums
‘recursively’ or ‘inductively’, as follows:
1 n+1 n
∑ ar = a1 and for n ∈ N, ∑ ar = (∑ ar ) + an+1 .
r=1 r=1 r=1

With this observation, we can use proof by induction to prove many results about the values
and properties of such sums. Here is a simple, classical, example.

Example 4.3. For all n ∈ N, ∑nr=1 r = 12 n(n + 1). This is simply the statement that the sum of
the first n natural numbers is 12 n(n + 1).

Proof. We prove the result by induction. Let P (n) be the statement that ∑nr=1 r = 12 n(n + 1).
Then P (1) states that 1 = 12 × 1 × 2, which is true; that is the base case.
Given k ∈ N, suppose (the induction hypothesis) ∑kr=1 r = 12 k(k + 1) is true.
Consider ∑k+1
r=1 r. We have

k+1 k
∑ r = ∑ r + (k + 1)
r=1 r=1
= 2 k(k + 1) + (k + 1) by
1
the induction hypothesis
= 2 (k + k + 2k + 2)
1 2

= 2 (k + 3k + 2)
1 2

= 2 (k + 1)(k + 2)
1

= 2 (k + 1)((k + 1) + 1).
1

Checking the first and last lines, what we have proved (assuming the induction hypothesis) is
P (k + 1), i.e. we proved P (k) Ô⇒ P (k + 1). We did this for an arbitrary k, so we proved the
induction step. By the Principle of Induction, P (n) is true for all natural numbers n.
Note how the the induction hypothesis was used. In the induction step, you always prove
P (k + 1) to be true assuming P (k) is. Unless you do so, it isn’t really a proof by induction.
If you write a ‘proof by induction’ and notice that you never use the induction hypothesis in
the induction step, then what you have is a fake induction. Cross out all the lines talking about
induction, and check that what is left is still a proof. Unless you’re answering a question that
explicitly says ‘prove by induction...’, we’re happier to get a direct proof than a fake induction
(even though both are proofs!).

Activity 4.3. Prove by induction that the sum of the first n terms of an arithmetic progression
with first term a and common difference d (that is, the sequence a, a + d, a + 2d, a + 3d,. . . ) is
2 n(2a + (n − 1)d).
1
CHAPTER 4. STRUCTURES, NATURAL NUMBERS AND PROOF BY INDUCTION 62

4.5 Recursively defined sequences

Sequences of numbers are often defined ‘recursively’ or ‘inductively’.
Example 4.4. The sequence xn is given by x1 = 9, and x2 = 13, and, for n ≥ 3, by xn =
3xn−1 − 2xn−2 . Prove that for all n ∈ N we have xn = 5 + 2n+1 .
This will be our first proof using Strong Induction—I’ll explain it after.
Proof. First, we can check that the formula works for n = 1 and n = 2, which we will call base
cases.
We have 9 = x1 = 5 + 22 . And we have 13 = x2 = 5 + 23 .
Now suppose k ≥ 3. Aim: if our formula holds for all t < k then xk = 5 + 2k+1 .

Assume, as the strong induction hypothesis, that xt = 5 + 2t+1 is true for each integer t < k.
In particular, the induction hypothesis means we assume xk−2 = 5 + 2k−1 , and xk−1 = 5 + 2k .
Now by definition (since k ≥ 3) we have
xk = 3xk−1 − 2xk−2
= 3(5 + 2k ) − 2(5 + 2k−1 )
= 15 + 6 × 2k−1 − 10 − 2 × 2k−1
= 5 + 4 × 2k−1
= 5 + 2k+1
which is the statement we wanted to show, so we proved the induction step.
By strong induction, we conclude the formula holds for all natural numbers n.
Let’s notice that we could replace ‘the statement we want to show’ with a defined predicate
so that it looks more like Strong Induction. Then we would have written:
Proof. For each n ∈ N, let S(n) be the statement ‘xn = 5 + 2n+1 ’.
First, we can check that S(1) and S(2) hold, which we will call base cases.
We have 9 = x1 = 5 + 22 . And we have 13 = x2 = 5 + 23 .
Now suppose k ≥ 3. We want to prove (∀t < k , S(t)) Ô⇒ S(k).

Assume, as the strong induction hypothesis, that S(t) is true for each integer t < k.
In particular, the induction hypothesis means we assume S(k − 2) and S(k − 1), i.e. that
xk−2 = 5 + 2k−1 , and xk−1 = 5 + 2k .
Now by definition (since k ≥ 3) we have
xk = 3xk−1 − 2xk−2
= 3(5 + 2k ) − 2(5 + 2k−1 )
= 15 + 6 × 2k−1 − 10 − 2 × 2k−1
= 5 + 4 × 2k−1
= 5 + 2k+1
which is S(k), so we proved the induction step.
By strong induction, we conclude S(n) holds for all natural numbers n.
The second version looks ‘more formal’, but both are equally good. As long as you can write
clearly, you don’t need to write some predicate P (n) (or S(n), or whatever other letter) in an
induction proof. But if you do write some P (n), you need to define it. ‘It’s obvious from the
question what that should be’ isn’t acceptable.
CHAPTER 4. STRUCTURES, NATURAL NUMBERS AND PROOF BY INDUCTION 63

Warning 4.5. When you write a predicate, by definition that is a true-or-false statement. If you
write ‘Let P (k) = 5 + 2k+1 ’ then you will definitely lose marks, because on the next few lines you
will write things like ‘7 is True’ which are nonsense. In general, any time you write P (n) = ..
where P (n) is supposed to be a predicate, the = sign has instantly lost you marks. We never
write that a logical statement is = something. (You can use the symbol ⇐⇒ , but usually it is
better to use words.)
Let’s check that we really are using Strong Induction correctly here. When we say ‘by Strong
Induction’ we’re claiming that we proved each of the implications that we have to prove.
We proved true Ô⇒ S(1) by proving S(1) directly. And we proved S(1) Ô⇒ S(2) by
proving S(2) directly.
And for each k ≥ 3, we proved S(1) ∧ S(2) ∧ ⋅ ⋅ ⋅ ∧ S(k − 1) Ô⇒ S(k) in the ‘induction step’.
So, yes, we did prove all the statements we were supposed to.
This also explains why we called S(1) and S(2) ‘base cases’ and the rest ‘the induction
step’. We did something special and different to prove those first two cases, which didn’t use
any induction hypothesis. To help the reader (who might expect us to have used S(1) to prove
S(2) !) we call it a base case; that’s just telling the reader ‘this case will be special’. And for
the rest of the cases, we used one argument that deals with all of them (and it does assume
some smaller cases are true) so we call it ‘the induction step’.
We’ll see later examples of strong induction arguments with one base case, or two, or three—or
even sometimes with no base case at all. The base cases are just the cases you find you need to
handle separately because the ‘main argument’ doesn’t work for them. In the example above,
in the ‘main argument’ we used the recursion formula xk = xk−1 + xk−2 . We were only told that
that formula makes sense if k ≥ 3, so the ‘main argument’ can’t handle k = 1 or k = 2. You can
find out what base cases you need by reading over the induction step, once you figure it out,
and checking whether it really works for all values of k (if so, no base cases) or if there are a
few small values of k for which it doesn’t work (these are the base cases, make sure you write a
proof separately for each of them).

4.6 Sample exercises

Exercise 4.1. Prove by induction that, for all n ∈ N, 2n ≥ n + 1.
Exercise 4.2. Prove by induction that the sum a + ar + ar2 + ⋯ + arn−1 of the first n terms of a
geometric progression with first term a and common ratio r ≠ 1 is a(1 − rn )/(1 − r).
Exercise 4.3. Prove by induction that for all n ∈ N,
n
1
∑ r2 = n(n + 1)(2n + 1).
r=1 6
n
1 n
Exercise 4.4. Prove by induction that ∑ = .
i=1 i(i + 1) n+1
Exercise 4.5. Suppose the sequence xn is given by x1 = 7, x2 = 23 and, for n ≥ 3, xn =
5xn−1 − 6xn−2 . Prove by induction that, for all n ∈ N, xn = 3n+1 − 2n .
Exercise 4.6. Prove by induction that, for all n ∈ N, 2n+2 + 32n+1 is divisible by 7.
Exercise 4.7. For a sequence of numbers x1 , x2 , x3 , . . . , and for n ∈ N, the number ∏nr=1 xr is
the product of the first r numbers of the sequence. It can be defined inductively as follows:
1 k+1 k
∏ xr = x1 , and for k ≥ 1, ∏ xr = (∏ xr ) xk+1 .
r=1 r=1 r=1
 CHAPTER 4. STRUCTURES, NATURAL NUMBERS AND PROOF BY INDUCTION 64

Suppose that x ≠ 1. Prove that

1 − x2
n n
2r−1
∏(1 + x )= .
r=1 1−x

4.7 Comments on selected activities

Comment on Activity 4.1. When n = 4, n2 = 16 and 2n = 24 = 16, so in this base case, the
statement is true. Suppose we make the inductive hypothesis that for some k ≥ 4, k 2 ≤ 2k . We
want to show
(k + 1)2 ≤ 2k+1 .
We have
(k + 1)2 = k 2 + 2k + 1 ≤ 2k + 2k + 1
(by the inductive hypothesis). So we’ll be done if we can show that 2k + 1 ≤ 2k . This will follow
from 2k + 1 ≤ k 2 and the assumed fact that k 2 ≤ 2k . Now,

2k + 1 ≤ k 2 ⇐⇒ k 2 − 2k − 1 ≥ 0 ⇐⇒ (k − 1)2 ≥ 2,

which is true for k ≥ 4. So, finally,

(k + 1)2 ≤ 2k + 2k + 1 ≤ 2k + k 2 ≤ 2k + 2k = 2k+1 .

as required. So the result is true for all n ≥ 4.

Comment on Activity 4.2. Let Q(n) be the statement ‘∀s ≤ n, P (s) is true’. Then Q(1) is true
if and only if P (1) is true. The statement Q(k) Ô⇒ Q(k + 1) is the same as

(P (s) true ∀s ≤ k) Ô⇒ (P (s) true ∀s ≤ k + 1).

But if P (s) is true for all s ≤ k then its truth for all s ≤ k + 1 follows just from its truth when
s = k + 1. That is, Q(k) Ô⇒ Q(k + 1) is the same as (P (s) true ∀s ≤ k) Ô⇒ P (k + 1). The
(standard) Induction Principle applied to the statement Q(n) tells us that: Q(n) is true for all
n ∈ N if the following two statements are true:

(i) Q(1) is true;

(ii) For all k ∈ N, Q(k) Ô⇒ Q(k + 1).

What we’ve established is that (i) and (ii) can be rewritten as:

(i) P (1) is true;

(ii) For all k ∈ N, (P (s) true ∀s ≤ k) Ô⇒ P (k + 1).

We deduce that: P (n) is true for all n ∈ N if the following two statements are true:

(i) P (1) is true;

(ii) For all k ∈ N, (P (s) true ∀s ≤ k) Ô⇒ P (k + 1).

This is exactly the Strong Induction Principle. So the Strong Induction Principle follows from
the standard one and is, therefore, not really ‘stronger’.
CHAPTER 4. STRUCTURES, NATURAL NUMBERS AND PROOF BY INDUCTION 65

Comment on Activity 4.3. Let P (n) be the statement that the sum of the first n terms is
(n/2)(2a + (n − 1)d). The base case is straightforward. The first term is a, and the formula
(n/2)(2a + (n − 1)d) gives a when n = 1. Suppose that P (k) holds, so the sum of the first k
terms is (k/2)(2a + (k − 1)d). Now, the (k + 1)st term is a + kd, so the sum of the first k + 1
terms is therefore
k k(k − 1)
a + kd + (2a + (k − 1)d) = a + kd + ak + d
2 2
k(k + 1)
= (k + 1)a + d
2
(k + 1)
= (2a + kd)
2
(k + 1)
= (2a + ((k + 1) − 1)d),
2
so P (k + 1) is true. The result follows for all n by induction.

4.8 Solutions to exercises

Solution to Exercise 4.1. Let P (n) be the statement ‘2n ≥ n + 1’. When n = 1, 2n = 2 and n + 1 = 2,
so P (1) is true. Suppose P (k) is true for some k ∈ N. Then 2k ≥ k + 1. It follows that

2k+1 = 2.2k ≥ 2(k + 1) = 2k + 2 ≥ k + 2 = (k + 1) + 1,

so P (k + 1) is also true. Hence, by induction, for all n ∈ N, 2n ≥ n + 1.

Solution to Exercise 4.2. Let P (n) be the statement that the sum of the first n terms is
a(1 − rn )/(1 − r). P (1) states that the first term is a(1 − r1 )/(1 − r) = a, which is true. Suppose
P (k) is true. Then the sum of the first k + 1 terms is the sum of the first k plus the (k + 1)st
term, which is ark , so this sum is

a(1 − rk ) a(1 − rk ) + (1 − r)ark

+ ark =
1−r 1−r
a − ark + ark − ark+1
=
1−r
a(1 − r )
k+1
= ,
1−r
which shows that P (k + 1) is true. Hence, for all n ∈ N, P (n) is true, by induction.
Solution to Exercise 4.3. Let P (n) be the statement that
n
1
∑ r2 = n(n + 1)(2n + 1).
r=1 6

Then P (1) states that 1 = 1(2)(3)/6, which is true. Suppose P (k) is true for k ∈ N. Then
k
1
∑ r2 = k(k + 1)(2k + 1)
r=1 6

and P (k + 1) is the statement that

k+1
1 1
∑ r2 = (k + 1)(k + 2)(2(k + 1) + 1) = (k + 1)(k + 2)(2k + 3).
r=1 6 6
CHAPTER 4. STRUCTURES, NATURAL NUMBERS AND PROOF BY INDUCTION 66

We have
k+1 k
∑ r2 = (k + 1)2 + ∑ r2
r=1 r=1
1
= (k + 1)2 + k(k + 1)(2k + 1) (by the induction hypothesis)
6
1
= (k + 1) [6(k + 1) + k(2k + 1)]
6
1
= (k + 1) (2k 2 + 7k + 6)
6
1
= (k + 1)(k + 2)(2k + 3),
6
so P (k + 1) is true. By induction, P (n) is true for all n ∈ N.
n
1 n
Solution to Exercise 4.4. Let P (n) be the statement that ∑ = . Then P (1) states
i=1 i(i + 1) n+1
1 1
that = , which is true. Suppose P (k) is true for k ∈ N. Then
1×2 1+1
k
1 k
∑ =
i=1 i(i + 1) k+1

and P (k + 1) is the statement that

k+1
1 k+1
∑ = .
i=1 i(i + 1) k + 2

Now,
k+1 k
1 1 1
∑ = +∑
i=1 i(i + 1) (k + 1)(k + 2) i=1 i(i + 1)
1 k
= + (by the induction hypothesis)
(k + 1)(k + 2) k + 1
1 + k(k + 2)
=
(k + 1)(k + 2)
k 2 + 2k + 1
=
(k + 1)(k + 2)
(k + 1)2
=
(k + 1)(k + 2)
k+1
= ,
k+2
so P (k + 1) is true. By induction, P (n) is true for all n ∈ N.
Solution to Exercise 4.5. Let P (n) be the statement that xn = 3n+1 − 2n . We use the Strong
Induction Principle to prove P (n) is true for all n ∈ N. The base cases are n = 1 and n = 2.
When n = 1, x1 = 7 and 3n+1 − 2n = 9 − 2 = 7. When n = 2, x2 = 23 and 3n+1 − 2n = 27 − 4 = 23, so
these are true. Suppose that k ≥ 2 and that for all s ≤ k, P (s) is true. In particular, P (k) and
CHAPTER 4. STRUCTURES, NATURAL NUMBERS AND PROOF BY INDUCTION 67

P (k − 1) are true and so

xk+1 = 5xk − 6xk−1

= 5(3k+1 − 2k ) − 6(3k − 2k−1 )
= 5(3k+1 ) − 5(2k ) − 6(3k ) + 6(2k−1 )
= 15(3k ) − 6(3k ) − 10(2k−1 ) + 6(2k−1 )
= 9(3k ) − 4(2k−1 )
= 3k+2 − 2k+1
= 3(k+1)+1 − 2k+1 ,

so P (k + 1) is true. Therefore, P (n) is true for all n ∈ N.

Solution to Exercise 4.6. Let P (n) be the statement that 2n+2 + 32n+1 is divisible by 7. When
n = 1, 2n+2 + 32n+1 = 8 + 27 = 35 and this is a multiple of 7 because 35 = 5 × 7. Suppose P (k) is
true, which means that for some m ∈ N, 2k+2 + 32k+1 = 7m. Now, when we take n = k + 1,

2n+2 + 32n+1 = 2k+3 + 32k+3

= 2(2k+2 ) + 9(32k+1 )
= 2(2k+2 + 32k+1 ) + 7(32k+1 )
= 14m + 7(32k+1 )
= 7 (2m + 32k+1 ) ,

which is a multiple of 7. So the statement is true for P (k + 1). This proves P (k) Ô⇒ P (k + 1),
the induction step, and hence, by induction, for all n ∈ N.
Solution to Exercise 4.7. Let P (n) be the statement

1 − x2
n n
r−1
∏(1 + x2 )= .
r=1 1−x

When n = 1, the left hand side is 1 + x2 = 1 + x and the right hand side is (1 − x2 )/(1 − x) = 1 + x,
0

so P (1) is true. Suppose P (k) is true, so that

1 − x2
k k
2r−1
∏(1 + x )= .
r=1 1−x

Then
k+1 k
r−1 r−1
∏(1 + x2 ) = (1 + x2 ) × ∏(1 + x2 )
(k+1)−1

r=1 r=1
1 − x2
k
k
= (1 + x2 ) (by the induction hypothesis)
1−x
1 − (x2 )2
k

= (where we’ve used (1 + y)(1 − y) = 1 − y 2 )

1−x
1 − x2 ×2
k

=
1−x
1 − x2
k+1

= ,
1−x
which shows that P (k + 1) is true. So P (n) is true for all n ∈ N, by induction.
 Functions and counting
5
The material in this chapter is also covered in:

• Biggs, N. L. Discrete Mathematics. Chapters 5 and 6.

• Eccles, P.J. An Introduction to Mathematical Reasoning. Chapter 10, Sections 10.1 and 10.2,
and Chapter 11.

5.1 Introduction
In this chapter we look at the theory of functions, and we see how the idea of the ‘size’ of a set
can be formalised.

5.2 Functions
5.2.1 Basic definitions
You have worked extensively with functions in your previous mathematical study. Chiefly, you
will have worked with functions from the real numbers to the real numbers, these being the
primary objects of interest in calculus.
You are probably used to writing a function down by writing a formula, something like
‘f (x) = x2 + sin x’. This is not the approach we are going to take, because it’s too restrictive.
For a very simple example, take the function g(x) which is defined as follows:
⎧
⎪ 0 if x ≤ 11850 ,
⎪
⎪
⎪1
g(x) = ⎨ 5 (x − 11850) if 11850 < x ≤ 46350 and
⎪
⎪
⎪
⎪
⎩ 5 (x − 46350) + 6900
2
if x > 46350 .

This is a perfectly good function, but finding a single formula for it is a bit tricky. Furthermore,
once you find it you’ll notice that the formula is much less helpful than the definition above.
This function was actually an important function (at least in the UK): it’s the (in 2018) income
tax you pay on income £x. I won’t bother trying to update that for 2022, because Liz Truss
will probably have changed her mind at least twice between me writing it and you reading it.

Activity 5.1. Find a single formula which gives the function g(x) above.

68
CHAPTER 5. FUNCTIONS AND COUNTING 69

So we do not want to think of ‘function’ as meaning ‘defined by a formula’. In fact, we don’t

want to think about how to go from the input x to the output f (x) at all—we will think of a
function as a ‘black box’ which takes in a number and spits out a number; the only rule is that
we insist that it always spits out the same number.
Actually, even that is too restrictive; we don’t want to insist that the input or output is a
number. Maybe we would like the input or output to be ‘Yes’, or ‘No’, or a colour, or a social
network... we need a definition which allows any of these possibilities. The only thing we want
to stick to is: if we give the function the same input twice, we should get the same output each
time. Here is the definition which formalises this.

Definition 5.1. Suppose that X and Y are sets. Then a function (also known as a mapping)
from X to Y is a rule that associates a unique member of Y to each member of X. We write
f ∶ X → Y . The set X is called the domain of f and Y is called the codomain.

The element of Y that is assigned to x ∈ X is denoted by f (x) and is called the image of x.
We can write x ↦ f (x) to indicate that x maps to f (x).
There are lots of examples of functions you already know, such as sin x, or g(x) defined
above.
If you have a social network, then that social network contains a number of friendships (i.e.
pairs of people who are friends); that defines a function from social networks to the integers,
which given a social network returns the total number of friendships.
If you have a road map of some country, then there may or there may not be a way to drive
through all the villages without ever having to return to a village you already visited. That
defines a function from road maps to {Yes, No}.
You can also generate your own personal function as follows. Throw a die 1 000 000 times, and
write down the numbers in order that you get—that defines you a function from {1, . . . , 1 000 000}
to {1, . . . , 6}. (It’s extremely unlikely anyone ever wrote down your personal function before. Of
course, the next time you try this you are very likely to get a different function..!)
Some of these functions are easier to work with, or more interesting, than others. You know
sin x shows up a lot in real-world calculations (in engineering, for example), and you know how
to do algebra and calculus with it.
What about the road map function? If you’re a fraudster, you need to keep moving on, and
you probably care a lot about not going back to villages where you already conned people—but
how do you actually work out, for a given road map with maybe 50 000 villages, whether the
answer is ‘Yes’ or ‘No’ ? It’s an interesting function, but it’s very hard to work with.
Finally, what about one of these generated-by-dice functions? It’s not easy to describe—you
don’t want to read a list a million characters long—and it’s not clear what it should be useful
for. Often (but certainly not all the time), we are really only interested in functions which we
can describe in some useful way.
There are various ways of describing a function.
If X has only finitely many members, we can simply list the images of the members of X.
You’re used to seeing a function defined by giving a formula for the function. For instance,
f ∶ R → R given by f (x) = 2x is the function that maps each real number a to the real number
2a.
Sometimes a function can be defined recursively. For example, we might define f ∶ N → N by

f (1) = 1 and f (n) = 2 + 3f (n − 1), for n ≥ 2 .

We might also define a function by writing down some properties it has. For example, I could
say ‘let h ∶ R → R be the function such that h(0) = 1 and dh(x)
dx = h(x) holds for all x ∈ R.’ You
CHAPTER 5. FUNCTIONS AND COUNTING 70

probably recognise from school that h(x) = ex is the exponential function. Here, we really need
to be careful: am I actually defining a function? In this case, yes: there is exactly one function
that satisfies the properties I wrote down. But if I left out the condition h(0) = 1 then I would
be writing something not well-defined, i.e. something that looks like it’s defining a function but
in fact isn’t. The reason is there would be many possible valid answers, such as h(x) = 2022ex .
Finally, we define one very basic function. For any set X, the identity function 1 ∶ X → X is
given by 1(x) = x.

5.2.2 Function equality

What does it mean to say that two functions f and g are equal? Well, first, they must have the
same domain X and codomain Y . Then, for each x ∈ X, we must have f (x) = g(x). For example,
if R+ is the set of positive real numbers, then the function f ∶ R+ → R given by f (x) = x2 and
the function g ∶ R → R given by g(x) = x2 are not equal because their domains are different.
You might think it is picky to say that, for example, the function f ∶ R≥0 → R≥0 defined
by f (x) = x2 and the function g ∶ R≥0 → R defined by g(x) = x2 are different (The set R≥0 is
the non-negative real numbers). After all, what you can put into both functions is the same,
and what comes out is also the same—the only difference is that the codomains of f and g are
different. However, it turns out often to be important what the codomain is—for example, we’ll
see later that only one of f and g is a ‘bijection’.
To repeat from a previous chapter—we’ve just met another definition of the symbol =. When
we write f = g, and f and g are functions, we mean:
the domains of f and g are equal (as sets),
the codomains of f and g are equal (as sets), and
f (x) = g(x) is true for all x in the domain of f (which is the same as the domain of g).
What do we mean by f (x) = g(x)? Well, if the codomains of f and g are numbers, we mean
equality of numbers. If they are sets, we mean equality of sets. If they are functions (yes, the
output of a function could actually be a function..!) then it means equality as we just defined it
of functions; and there are more things that f (x) and g(x) could be.

5.2.3 Composition of functions

Suppose that X, Y, Z are sets and that f ∶ X → Y and g ∶ Y → Z. Then the composition g ○ f is
the function from X to Z given by

(g ○ f )(x) = g(f (x)) for x ∈ X .

If X and Z are distinct sets, there is only one way we can compose f and g.
If X = Z, then both f ○ g and g ○ f make sense—but they are generally not the same function:
the order is important.
In some textbooks you will see a different notation for function composition, leaving out
the ○. So you might see f g where I would write f ○ g. This notation f g can cause confusion
(which is why I won’t use it). For example, suppose X = Y = Z = R. Then you might be tempted
to think that gf denotes the product function x → g(x)f (x). But this would be wrong. The
notation g ○ f avoids this confusion.

Example 5.2. Suppose f ∶ N → N and g ∶ N → N are given by f (x) = x2 + 1 and g(x) = (x + 1)2 .
CHAPTER 5. FUNCTIONS AND COUNTING 71

2
Then (f ○ g)(x) = f (g(x)) = f ((x + 1)2 ) = ((x + 1)2 ) + 1 = (x + 1)4 + 1 ,
2
while (g ○ f )(x) = g(f (x)) = g(x2 + 1) = ((x2 + 1) + 1) = (x2 + 2)2 ,
and g(x)f (x) = (x + 1)2 (x2 + 1) .

All three are different.

5.3 Bijections, surjections and injections

There are three very important properties that a function might possess:

Definition 5.3 (Surjection). Suppose f is a function with domain X and codomain Y . Then f
is said to be a surjection (or ‘f is surjective’) if every y ∈ Y is the image of some x ∈ X; that is,
f is a surjection if and only if ∀y ∈ Y, ∃x ∈ X, s.t. f (x) = y.

Definition 5.4 (Injection). Suppose f is a function with domain X and codomain Y . Then f
is said to be an injection (or ‘f is injective’) if every y ∈ Y is the image of at most one x ∈ X. In
other words, the function is an injection if different elements of X have different images under
f . Thus, f is an injection if and only if

∀x, x′ ∈ X, x ≠ x′ Ô⇒ f (x) ≠ f (x′ )

or (equivalently, taking the contrapositive), if and only if

∀x, x′ ∈ X, f (x) = f (x′ ) Ô⇒ x = x′ .

This latter characterisation often provides the easiest way to verify that a function is an injection.

Definition 5.5 (Bijection). Suppose f is a function with domain X and codomain Y . Then
f is said to be a bijection (or ‘f is bijective) if it is both an injection and a surjection. So this
means two things: each y ∈ Y is the image of some x ∈ X, and each y ∈ Y is the image of no
more than one x ∈ X. Well, of course, this is equivalent to: each y ∈ Y is the image of precisely
one x ∈ X.

Example 5.6. f ∶ N → N given by f (x) = 2x is not a surjection, because there is no n ∈ N such

that f (n) = 1. (For, 2n = 1 has no solution where n ∈ N.) However, it is an injection. To prove
this, suppose that m, n ∈ N and f (m) = f (n). Then 2m = 2n, which implies m = n.

Activity 5.2. Prove that f ∶ R → R given by f (x) = 2x is a bijection.

We write (a, b) (which is called an open interval and we will meet again later) for the set
of real numbers x such that a < x < b. And we write ∣x∣ for the absolute value of x, defined by
∣x∣ = x if x ≥ 0 and ∣x∣ = −x if x < 0. Thus ∣x∣ is always non-negative.

Example 5.7. The function f ∶ R → (−1, 1) given by f (x) = x

1+∣x∣ is a bijection.

Proof. First, we prove f is injective. To do this, we prove that f (x) = f (y) implies x = y. So,
suppose f (x) = f (y). Then
x y
= .
1 + ∣x∣ 1 + ∣y∣
Rearranging, we want to solve x + x∣y∣ = y + y∣x∣.
Suppose x ≥ 0. If y < 0, then the left hand side of the above equation is non-negative and the
CHAPTER 5. FUNCTIONS AND COUNTING 72

right hand side is negative—this cannot be a solution. So y ≥ 0. But then ∣x∣ = x and ∣y∣ = y, and
we get x + xy = y + xy, which tells us x = y.
Suppose x < 0. If y ≥ 0, then the left hand side of the above equation is negative and the right
hand side is non-negative—this cannot be a solution. So y < 0. Then ∣x∣ = −x and ∣y∣ = −y, we
get x − xy = y − xy and again x = y.
Next, we show f is surjective. We need to prove that, for each y ∈ (−1, 1), there is some
x ∈ R such that 1+∣x∣
x
= y.
Suppose y ≥ 0. Then, to have 1+∣x∣ x
= y, we need x ≥ 0. So ∣x∣ = x and we need to solve 1+x
x
= y.
This has solution x = 1−y , which is well-defined and non-negative because we know 0 ≤ y < 1.
y

Suppose y < 0. Then we’ll need to have x < 0 and the equation to solve is 1−xx
= y, for a solution
x = 1+y ; this is well-defined and negative since 0 > y > −1.
y

Since we showed f is injective and surjective, by definition it is bijective.

CHAPTER 5. FUNCTIONS AND COUNTING 73

At first, you might well think that the above proof is difficult; it’s certainly not short, and
has a bunch of somewhat complicated formulae and cases to consider.
But actually, this proof is long, but not hard. We will see quite a few proofs which are long,
but not hard, in this course. This is one of the standard places where students are put off the
course (and maybe mathematics as a whole), because they feel that how difficult it will be to
find a proof has to be proportional to the length of the proof, and the proofs are getting rapidly
longer.
There will certainly be difficult proofs in the course. But proof difficulty doesn’t have much
to do with length. Let me explain why this proof is not hard.
To begin with, we’re supposed to prove a function is bijective. That means (definition chasing)
we need to prove it is injective and surjective (because that’s what ‘bijective’ means). Well, if
we want to prove two things, we should probably do them one after the other. So we do that
(and unsurprisingly, if we prove two things it will be twice as long).
Next, we look up the definition of ‘f is injective’ in order to prove it. We take the hint from
the lecture notes to use the contrapositive form: we should (definition chasing) try to prove
f (x) = f (y) Ô⇒ x = y is true for all x, y ∈ R. Well, this is a ‘for all’ statement, so we use the
standard first thing to try: fix x and y, and try to prove the statement for this particular x and
y. We write in the definition of f (x) and f (y) (definition chasing, again) and hope to get some
nice equation that we can hit with algebra and solve. What we get is x + x∣y∣ = y + y∣x∣.
This isn’t quite a nice equation, because of the ∣⋅∣ signs; we would be much happier if we could
get rid of them. How can we do that? Well, we can get rid of ∣x∣ by definition chasing (again!):
let’s think about the cases x ≥ 0 (which is when ∣x∣ = x) and x < 0 (so ∣x∣ = −x) separately. That
case distinction in the proof is not magic, it was copied straight from the definition of ∣x∣.
We still have a nasty ∣y∣ around. Let’s repeat the definition chasing: in each of our two cases
for x, let’s separately consider whether y ≥ 0 or y < 0 (so we have four cases in total).
At this point, we need to think for a couple of seconds to notice that two of our four cases
can’t really happen: if x ≥ 0 and y < 0 (or vice versa) then we don’t need to start doing algebra
because there can’t be a solution.
What would happen if you did just start doing algebra here? Well, you’d try to solve
x − xy = y + xy (plugging in ∣x∣ = x and ∣y∣ = −y) and so x = y + 2xy, so y = 1+2x
x
. Then you need
to notice that since x ≥ 0, the ‘solution’ you’ve just found gives us y ≥ 0, whereas we assumed
y < 0. So it’s not really a solution; it violates the assumption we made.
And then we do just do the algebra in the two remaining cases, and in both cases it is easy.
Now we move to ‘f is surjective’. Again, we definition-chase, and write out what that means.
Again, we need to deal with the ∣x∣ and again the right thing to do is to separate the two cases
(since we are given y and want to find x such that f (x) = y, it makes sense to consider the
two possible cases for y). And again, we can then do the algebra and double-check our solution
makes sense.
What you should notice is that although there are a lot of steps here (and this only got to ‘f
is injective’) all the steps are basic strategies: mostly definition chasing, plus a couple of times
we did some high-school algebra to solve equations. Because there are a lot of steps, you have
no chance of looking at the problem and seeing how the proof will go; it’s easy to get scared.
But if you simply try, you can write the proof down without ever having to pause for thought
for more than a minute (if you’re revising, you’re probably at the stage where it takes longer to
write the next line than to think what it should be). Whenever we had some not-so-nice concept
left over, we used definition chasing to replace it with something nicer (even when that means
considering cases, this is a winner: two nice things is better than one not-so-nice thing). Until we
finally got down to a problem you know how to do from high school ‘solve this nice equation’.
That is ‘long but not hard’; get used to it. Don’t get scared until standard strategies don’t help.
CHAPTER 5. FUNCTIONS AND COUNTING 74

5.4 Inverse functions

5.4.1 Definition, and existence
Suppose we are given a function f ∶ X → Y . Then g ∶ Y → X is an inverse function of f if
(g ○ f )(x) = x for all x ∈ X and (f ○ g)(y) = y for all y ∈ Y . An equivalent characterisation is
that y = f (x) ⇐⇒ x = g(y).
The following theorem tells us precisely when a function has an inverse. It also tells us that
if an inverse exists, then there is only one inverse. For this reason we can speak of the inverse
function, and give it a specific notation, namely f −1 .

Theorem 5.8. f ∶ X → Y has an inverse function if and only if f is a bijection. If an inverse

function exists, it is unique.

First, we prove:
f ∶ X → Y has an inverse ⇐⇒ f is bijective.
Proof. This is an ⇐⇒ theorem, so there are two things to prove: the ⇐ and the ⇒.
First, we show: f ∶ X → Y has an inverse ⇐ f is bijective.
Suppose f is a bijection. For each y ∈ Y there is exactly one x ∈ X with f (x) = y. Define
g ∶ Y → X by g(y) = x. Then this is an inverse of f . Check this!
Next, we show: f ∶ X → Y has an inverse ⇒ f is bijective.
Suppose f has an inverse function g. We know that for any y ∈ Y , f (g(y)) = (f ○ g)(y) = y,
so there is some x ∈ X (namely x = g(y)) such that f (x) = y. So f is surjective.
Now suppose f (x) = f (x′ ). Then g(f (x)) = g(f (x′ )). But g(f (x)) = (g ○ f )(x) = x and,
similarly, g(f (x′ )) = x′ . So: x = x′ and f is injective.
Now we prove that if an inverse function exists, it is unique.
Proof. Suppose that g and h are inverses of f . Then both have domain Y and codomain X, and
we just need to check that g(y) = h(y) for every y ∈ Y . Well, h ○ f is the identity function on X
and f ○ g is the identity function on Y . So, for any y ∈ Y we have

g(y) = (h ○ f )(g(y)) = ((h ○ f ) ○ g)(y) = (h ○ (f ○ g))(y) = h((f ○ g)(y)) = h(y) ,

so g = h.
Note that if f ∶ X → Y is a bijection, then its inverse function (which exists, by Theorem 5.8)
is also a bijection. The easiest way to see that is: if g is the inverse function of f , then also by
definition f is the inverse function of g. So g has an inverse function, so by Theorem 5.8 g is
bijective.
Again, you need to be a bit careful with the notation if your function is (for example) from
R to R. Do not confuse f −1 , the inverse function, with the function x → (f (x)) = 1/f (x).
−1

5.4.2 Examples
Example 5.9. The function f ∶ R → R is given by f (x) = 3x + 1. Find the inverse function.
To find a formula for f −1 , we use: y = f (x) ⇐⇒ x = f −1 (y). Now,

y = f (x) ⇐⇒ y = 3x + 1 ⇐⇒ x = 31 (y − 1),

so
f −1 (y) = 31 (y − 1).
CHAPTER 5. FUNCTIONS AND COUNTING 75

Recall Z denotes the set of all integers (positive, zero, and negative).

Example 5.10. The function f ∶ Z → N ∪ {0} is defined as follows:

2n if n ≥ 0
f (n) = {
−2n − 1 if n < 0.

Prove that f is a bijection and determine a formula for the inverse function f −1 .
First, we prove that f is injective: Suppose f (n) = f (m). Since 2n is even and −2n − 1 is
odd, either (i) n, m ≥ 0 or (ii) n, m < 0. (For otherwise, one of f (n), f (m) is odd and the other
even, and so they cannot be equal.)
In case (i), f (n) = f (m) means 2n = 2m, so n = m.
In case (ii), f (n) = f (m) means −2n − 1 = −2m − 1, so n = m. Therefore f is injective.
Next, we prove that f is surjective: We show that ∀m ∈ N ∪ {0}, ∃n ∈ Z such that f (n) = m.
Consider separately the case m even and the case m odd.
Suppose m is even. Then n = m/2 is a non-negative integer and f (n) is 2(m/2) = m).
If m odd, then n = −(m + 1)/2 is a negative integer and

(m + 1)
f (n) = f (−(m + 1)/2) = −2 (− ) − 1 = m.
2
The proof that f is surjective reveals to us what the inverse function is. We have

m/2 if m even
f −1 (m) = {
−(m + 1)/2 if m odd.

Finally, let’s give an important non-example.

√ 5.11. Let f ∶ R → R≥0 be defined by f (x) = x , and let g ∶ R≥0 → R be defined by

Example 2

g(x) = x.
√ function of f , and indeed (f ○ g)(x) = x for all
It’s tempting to think that g is the inverse
x ∈ R≥0 . But (g ○ f )(−1) = g(1) = 1, because x means the non-negative square root of x. If you
check Theorem 5.8 you’ll see that in fact f doesn’t have an inverse function: it is not a bijection.
√ example f (1) = 1 = f (−1). It’s a somewhat common mistake in basic algebra to assume
For
x2 = x; as we just saw it’s not true when x < 0. We saw essentially this error as Mistake 4 in
Section 2.7.
CHAPTER 5. FUNCTIONS AND COUNTING 76

5.5 Functions on sets

Suppose we have a function f ∶ X → Y . It is very common that, given some S ⊆ X, we want to
talk about the set {f (x) ∶ x ∈ S}. To make this easier, we define

f (S) = {f (x) ∶ x ∈ S} .

Note that f (∅) = ∅, and for any single x ∈ X we have f ({x}) = {f (x)}. It’s important to
remember that {f (x)} is not the same as f (x) (in the same way that an apple in a box is not
the same as an apple).
We also define, for any function f ∶ X → Y and any T ⊆ Y , the set

f −1 (T ) = {x ∈ X ∶ f (x) ∈ T } .

Again, it’s important to remember that for y ∈ Y , the set f −1 ({y}) is a set of elements in X,
and it always exists, in contrast to f −1 (y) which is a member of X and is only defined if f is an
invertible function.
If f is invertible, then for every y ∈ Y the set f −1 ({y}) contains exactly one element, namely
f −1 (y). However if f is not invertible, then by Theorem 5.8 either there will be some y ∈ Y such
that f −1 ({y}) = ∅ (i.e. f is not surjective) or there will be some y ∈ Y such that f −1 ({y}) has
two or more elements (i.e. f is not injective), or both.
Given a function f ∶ X → Y , the set f (X) is sometimes called the image of f . The image
f (X) of f is always a subset of the codomain Y (by definition!). It might be that f (X) = Y , or
it might not be—by definition, f (X) = Y if and only if f is surjective.

5.6 Counting as a bijection

What does it mean to say that a set has three objects? Well, it means that I can take an object
from the set, and call that ‘Object 1’, then I can take a different object from the set and call
that ‘Object 2’, and then I can take a different object from the set and call that ‘Object 3’, and
then I have named all the objects in the set. Obvious, I know, but this is the fundamental way
in which we can abstractly define what we mean by saying that a set has m members.
For m ∈ N, let Nm be the set {1, 2, . . . , m} consisting of the first m natural numbers. Then
we can make the following formal definition:

Definition 5.12. A set S has m members if there is a bijection from Nm to S.

So, the set has m members if to each number from 1 to m, we can assign a corresponding
member of the set S, and all members of S are accounted for in this process. This is like the
attachment of labels ‘Object 1’, etc, described above.
Note that an entirely equivalent definition is to say that S has m members if there is a
bijection from S to Nm . This is because if f ∶ Nm → S is a bijection, then the inverse function
f −1 ∶ S → Nm is a bijection also. In fact, because of this, we can simply say that S has m
members if there is a bijection ‘between’ Nm and S. (Eccles uses the definition that involves a
bijection from Nm to S and Biggs uses the definition that involves a bijection from S to Nm .)
For m ∈ N, if S has m members, we say that S has cardinality m (or size m). The cardinality
of S is denoted by ∣S∣, so we would usually simply write ∣S∣ = m for ‘S has cardinality m’.

Warning 5.13. If you are very alert, you might notice that there is a potential problem with our
definition of cardinality. We said something about ‘the cardinality of S’. That means we have
some idea that there should only be one number m such that ∣S∣ = m. Well, if I have a set of
CHAPTER 5. FUNCTIONS AND COUNTING 77

five fruit, you’ll probably happily agree with me that it has cardinality five and nothing else.
But is that kind of statement always true whatever S is a set of? What we’re worried about
here is whether cardinality is well-defined. We’ll shortly see that it is.
In general ‘well-defined’ means that whatever definition we just wrote down is not ‘cheating’
or ‘wrong’ in some way. What might be an example of a bad definition? Suppose I say ‘let t
be the number of cards in a deck’. I am claiming to define a number t here; there should be
only one answer to the question of what t is. But what deck of cards? A bridge deck (with 52
cards)? or a skat deck (with 32)? Or something else? This t is not well-defined, and it’s exactly
this kind of problem that the warning is getting into. Could it be that there is a set S such that
by our definition we have ∣S∣ = 32 and also ∣S∣ = 52?
It’s usually easiest to write down a definition and then try to argue that it makes sense; we
say we are showing the definition is well-defined. We’ll do that for cardinality shortly, but we
need some more theory first.

5.7 The pigeonhole principle

5.7.1 The principle
The ‘pigeonhole principle’ is something that you might find obvious, but it is very useful.
Informally, what it says is that says is that if you have n letters and you place them into
m pigeonholes in such a way that no pigeonhole contains more than one letter, then n ≤ m.
Equivalently, if n > m (so that you have more letters than pigeonholes), then some pigeonhole
will end up containing more than one letter. This is very intuitive. Obvious as it may be, however,
can you think about how you would actually prove it?
We can’t really hope to prove any vague statement until we make it more formal. So let’s
first do that.

Theorem 5.14 (Pigeonhole Principle (PP)). Suppose that A and B are sets with ∣A∣ = n and
∣B∣ = m, where m, n ∈ N. If there is an injection from A to B, then n ≤ m.

We’ve just formalised the first statement above: if we place (the function f ) letters (the
set A) into pigeonholes (the set B) such that no pigeonhole contains more than one letter (f
is injective) then A cannot be bigger than B. This is now a clear formal statement: we know
exactly what we need to prove.
But coming up with a proof is not easy. We’ll need to talk about injective functions (because
there is an injective function in the statement), but we will also need to use the definition
of cardinality, because that also shows up (we say ∣A∣ = n) and that talks about (completely
different!) bijective functions. And furthermore, we will probably need to talk about the members
of A and of B, which are two arbitrary sets—we don’t know what the members are. To get
around that (temporarily!) let’s try to prove the statement for a couple of specific sets.

Theorem 5.15 (Pigeonhole Principle (PP), special case). The following statement is true for
all n ∈ N: For all natural numbers m, if there is an injection from Nn to Nm , then n ≤ m.

This version doesn’t talk about cardinality; we know (by definition!) that ∣Nn ∣ = n and
∣Nm ∣ = m, and we know what the elements of these two sets are. This will make it easier to write
a formal proof. But it’s still not easy to see what to do next.
We know we need to deal with injective functions to prove this special case. So let’s prove a
statement about injective functions. For now, it is going to be unclear what this statement has
to do with the Pigeonhole Principle; I’ll try to explain where it comes from later.
CHAPTER 5. FUNCTIONS AND COUNTING 78

Lemma 5.16. Suppose that A and B are sets, each of which has at least two distinct elements.
Suppose that a is an element of A, and b is an element of B. If there is an injection f ∶ A → B,
then there is an injection g ∶ A ∖ {a} → B ∖ {b}.
Proof. Given A and B, and elements a and b, as in the lemma statement, we want to prove that
if there is an injection f ∶ A → B, then there is an injection g ∶ A ∖ {a} → B ∖ {b}.
So suppose that f ∶ A → B is an injection.
We want to use f to help us construct g. We consider two cases.
Case 1: f (x) ∈ B ∖ {b} for each x ∈ A ∖ {a}.
This case is easy. We define a function g ∶ A ∖ {a} → B ∖ {b} by g(x) = f (x) for each x. This
is well-defined because we assumed that for each x in A ∖ {a}, indeed f (x) is in B ∖ {b}. We
just need to check that g is indeed injective. Well, suppose g(x) = g(y). Then by definition
f (x) = f (y), and since f is injective x = y. So g is indeed injective.
Case 2: there is s ∈ A ∖ {a} such that f (s) = b.
This case is simply what we get when we say ‘we are not in Case 1’. It’s what is left over
after dealing with the easy case.
Let’s first check that there is only one s such that f (s) = b. Indeed, suppose that for some
x ∈ A we have f (x) = b. Then f (x) = f (s), and since f is injective, we conclude x = s.
This time, if we tried to define g as in Case 1, we would find g is not well-defined. g is
supposed to have codomain B ∖ {b}, but f (s) = b. But this is the only ‘problem’; we can hope
to define g(s) in some other way. The trick is: we define g ∶ A ∖ {a} → B ∖ {b} by
⎧
⎪
⎪f (x) if x ≠ s
g(x) = ⎨ .
⎪
⎩f (a)
⎪ if x = s

This is well-defined—that is, g(x) is in B ∖ {b} for each x in the domain—because we know
f (x) is always in B and we don’t use f (s) which is the only way of getting b.
What is not quite so clear is that g is injective. Let’s check. Suppose that g(x) = g(y) for
some x, y. We want to show x = y.
If neither x nor y is equal to s, then by definition we have g(x) = f (x) and g(y) = f (y), so
f (x) = f (y), so since f is injective we have x = y. We need to deal with the case that at
least one of x and y is equal to s; suppose without loss of generality it is x. Then we have
g(y) = g(x) = f (a). If y ≠ x then g(y) = f (y) = f (a), but then since f is injective we have
y = a—and this is impossible, since y ∈ A ∖ {a}. So in this case also y = x and we are done.
In either case, we were able to construct an injective g as desired, and the two cases are
exhaustive.
This proof is not all that easy to understand—because it is quite abstract—so here is a
concrete ‘story’ of the proof.
Suppose you have a set of hotel guests (A) who are booked into the set of single rooms (B)
in a hotel. The function f ∶ A → B says which guest is booked into each room: that it is injective
is telling you that each room has at most one guest booked in (some rooms might be empty,
but no room has two guests booked in to it).
Now one guest (called a) checks out, and there is a water leak in one room (room number b)
so that room becomes unusable. What does the hotel manager do? Well, if none of the remaining
guests (the set A ∖ {a}) is booked into the wet room, they don’t have to do anything. That’s
case 1.
If on the other hand there is a guest s booked into the wet room, then the manager can
solve the problem by changing s’s room to the one a has vacated. That’s case 2.
We’ll see this Lemma is what we need to prove Theorem 5.15 by induction. As a quick
remark, it’s maybe not clear why in the statement of the Lemma we say that A and B each
CHAPTER 5. FUNCTIONS AND COUNTING 79

have at least two distinct elements. The reason is that we do not want A ∖ {a} or B ∖ {b} to be
the empty set; it’s not clear what a function with domain or codomain the empty set should be.
We can now prove Theorem 5.15.
Proof of Theorem 5.15. We prove this by induction. The statement we want to prove is the
statement P (n): ‘for all m ∈ N, if there is an injection from Nn to Nm , then n ≤ m.’
The base case, n = 1, is true because for all m ∈ N we have 1 ≤ m.
Given a natural number k, we want to prove P (k) Ô⇒ P (k + 1).
Suppose for an induction hypothesis that P (k) is true. We want to prove P (k + 1). That is,
given m, we want to show that if there is an injection f ∶ Nk+1 → Nm , then k + 1 ≤ m.
So suppose there is an injection f ∶ Nk+1 → Nm . We want to show k + 1 ≤ m.
Since k ≥ 1, we have k + 1 ≥ 2.
If m = 1, then the codomain of f is {1}, so f (1) = f (2) = 1. But this is a contradiction to
our assumption that f is injective; this case cannot occur.
If m ≥ 2, then f is an injective function from Nk+1 to Nm , and both of these sets have at least
two elements (both contain 1 and 2). So we can apply Lemma 5.16, with A = Nk+1 and a = k + 1,
and B = Nm and b = m. The Lemma says that there is an injective function g ∶ Nk → Nm−1 .
And now our induction hypothesis P (k) tells us that k ≤ m − 1. Adding 1 to both sides, we
conclude k + 1 ≤ m, which is what we wanted.
This proves the induction step. By the Principle of Induction, we conclude that P (n) is true
for all n ∈ N.
Finally, let’s explain why the special case of the Pigeonhole Principle implies the general
case, Theorem 5.14.
Proof of Theorem 5.14. From the definition of cardinality, there are bijections g ∶ Nn → A and
h ∶ Nm → B. We also have an inverse bijection h−1 ∶ B → Nm by Theorem 5.8.
Suppose there is an injection f ∶ A → B. Consider the composite function h−1 ○f ○g ∶ Nn → Nm .
If we can prove that this is an injection, then from Theorem 5.15 it follows that n ≤ m.
So, let us prove injectivity. Suppose a, b ∈ Nn with a ≠ b. Since g is a bijection g(a), g(b) ∈ A
with g(a) ≠ g(b). Since f is an injection, there are f (g(a)), f (g(b)) ∈ B with f (g(a)) ≠ f (g(b)).
Since h−1 is a bijection, h−1 (f (g(a))) and h−1 (f (g(b))) belong to Nm , and h−1 (f (g(a))) ≠
h−1 (f (g(b))). This last inequality is what we need.
This was a long proof. Before we make a couple of comments on what you should learn from
it, let’s deduce one important conclusion.
Theorem 5.17. Suppose n, m are two natural numbers. If there is a bijection from Nn to Nm ,
then n = m.
Proof. Suppose f ∶ Nn → Nm is a bijection. Then f is an injection. So from Theorem PP, n ≤ m.
But by Theorem 5.8 there is an inverse function f −1 ∶ Nm → Nn and this is also a bijection.
In particular, f −1 is an injection from Nm to Nn , and hence m ≤ n.
Now we have both n ≤ m and m ≤ n, hence n = m.
What this theorem tells us is that our definition of cardinality is well-defined. Remember we
were worried that possibly there is some set S such that we can write ∣S∣ = m and ∣S∣ = n, and
m and n aren’t the same; then it wouldn’t make sense to say that either is ‘the size of S’. But if
both ∣S∣ = m and ∣S∣ = n, then there are by definition bijections f ∶ Nm → S and g ∶ Nn → S, and
so f −1 ○ g is a bijection from Nn → Nm . And now Theorem 5.17 says n = m.
The pigeonhole principle is remarkably useful (even in some very advanced areas of mathe-
matics). It has many applications. For most applications, it is the contrapositive form of the
principle that is used. This states:
CHAPTER 5. FUNCTIONS AND COUNTING 80

If m < n and f ∶ Nn → Nm is any function, then f is not an injection..

So, if m < n, and f is any function f ∶ Nn → Nm , then there are x, y ∈ Nn with x ≠ y such that
f (x) = f (y).
In other words, if you have more letters than pigeonholes, then you will have to put at least
two letters into some one pigeonhole.

5.7.2 What will be on the exam?

We’ve just seen our first ‘long’ proof which is examinable, the proof of the Pigeonhole Principle.
You might be tempted to make a tactical guess that I will not ask you to reproduce this proof
in the exam (which is correct, I won’t ask it) and hence skip it. And you might think that it is
too obvious to be interesting.
This would be an error. The proof is in the course for a reason: it’s the first proof you have
seen which uses ‘abstract information’ in a serious way, and I can and quite possibly will ask
questions on the exam which test your ability to do something similar, maybe in a simpler
scenario (the proof of PP is too long for an exam question, and would be too hard if you hadn’t
seen it before).
There are a few steps to the proof of the Pigeonhole Principle. The first one, after sorting
out how to write down the right formal statement (Theorem 5.14) is to notice that it’s enough
to prove Theorem 5.15. This isn’t ‘necessary’—you can write a proof of Theorem 5.14 without
deducing it from Theorem 5.15—but it does make a lot of statements simpler; it’s easier to
understand the way we wrote it.
After this, how does one think of the proof of Theorem 5.15? If you hadn’t seen the proof
before, most likely you would try to prove it directly, and at some point you’d get stuck. Then
you might notice that since it is a ‘for all natural numbers’ statement, a possibility would be to
try an induction proof. I think realistically you won’t find the induction argument unless you’re
looking for it here.
Once you think of trying induction on n, then it’s obvious that the base case is true—we
don’t need to think at all about the condition ‘if there is an injection from N1 to Nm ’ at all,
because 1 ≤ m is true for all natural numbers m.
So the difficulty is to prove the induction step. Now, it is not obvious how to do this—it is
certainly not the case that a Real Mathematician instantly sees how to do it. What we do is look
for something more we can say, ideally something that will let us use our induction hypothesis.
There is one more thing we can easily say. We are given that there is an injection from Nk+1
to Nm ; in particular k + 1 is at least 2, and we can immediately rule out the possibility m = 1.
Note we do not use the induction hypothesis to do this (even though we are in the middle of
the induction step). So what is left (in the induction step) is to deal with the case m ≥ 2.
Now, at this point Lemma 5.16 plus the induction assumption immediately deals with this
case and we are done. But this is ‘cheating’—the only reason you would care about Lemma 5.16
is in order to prove Theorem 5.15. It is certainly not the case that some historical mathematician
proved Lemma 5.16 and then noticed that they could use it to prove Theorem 5.15. What we
can see at this point is that our induction hypothesis P (k) will tell us k ≤ m − 1 (which is
basically what we want) provided we can somehow find an injective function from Nk to Nm−1 .
So our aim has to be to find such an injective function, and this has to come somehow from the
injection we know exists, namely f .
That means it’s natural to write down a statement ‘if m ≥ 2 and there is an injection
f ∶ Nk+1 → Nm then there is an injection g ∶ Nk → Nm−1 ’; it’s what we want to be true. And
this is more or less the same thing as Lemma 5.16. One could (and in previous years we did)
not bother to write a separate Lemma, but simply prove the statement in quotes above at this
CHAPTER 5. FUNCTIONS AND COUNTING 81

point. But students generally complained it was confusing, so now we separate the Lemma out
explicitly.
At last, we have one more question: how do we think of the proof of the Lemma? Well, the
first few lines are ‘automatic’; we’ve just written down the information we’re given in the lemma
statement, and then we have to prove an implication—so we go for the simplest route, namely
assume the premise and try to prove the conclusion from it.
Then we get to a case distinction. This case distinction looks a bit complicated at first,
but it follows the basic idea mentioned earlier in these notes: if you’re not sure how to prove
something, identify a special ‘easy’ case you can do, do it, then figure out how to do the rest.
The ‘easy case’ is Case 1; here f really immediately gives us the injection g we want, we just
need to write it down and check it.
The ‘hard’ case is Case 2. All we do to write it down is figure out what it means that ‘we are
not in Case 1’, but it turns out to give us a piece of abstract information; we get told something
about the function values of f , namely f (s) = b, which we did not know before, and which we
should try to use. And, finally, once we got this far it turns out not to be that hard!
This kind of understanding is what I want you to get from the proof of PP. For all the longer
proofs in these notes, I would like you to get an idea of why the proof works and what ideas you
are being shown that you can use elsewhere in your own proofs; this is why these proofs are
there. Sometimes, as here, I’ll break the proof into bitesize pieces and give more details of what
and why we are doing something, but not always. It is good for you to learn to break a long
complicated argument into pieces yourself—identify the key points, figure out which things are
‘automatic’ (i.e. the first thing you should try works) and which are ‘difficult’ (everything else,
especially the times where the second and third things you should try don’t work either). It’s
not quite as good as coming up with a long complicated proof of your own, but it’s a next best.

5.7.3 Some applications of the Pigeonhole Principle

We start with an easy example.

Theorem 5.18. In any group of 13 or more people, there are two persons whose birthday is in
the same month.

Proof. Consider the function that maps the people to their months of birth. Since 13 > 12, this
cannot be a bijection, so two people are born in the same month.
This next one is not hard, but perhaps not immediately obvious.

Theorem 5.19. In a room full of people, there will always be at least two people who have the
same number of friends in the room.

Proof. Let X be the set of people in the room and suppose ∣X∣ = n ≥ 2. Consider the function
f ∶ X → N ∪ {0} where f (x) is the number of friends x has in the room.
Let’s assume that a person can’t be a friend of themselves. (We could instead assume that a
person is always friendly with themselves: we simply need a convention one way or the other.)
Then f (X) = {f (x) ∶ x ∈ X} ⊆ {0, 1, . . . , n − 1}. But there can’t be x, y with f (x) = n − 1 and
f (y) = 0. Why? Well, such an x would be a friend of all the others, including y, which isn’t
possible since y has no friends in the room.
So either f (X) ⊆ {0, 1, . . . , n − 2} or f (X) ⊆ {1, . . . , n − 1}. In each case, since f (x) can take
at most n − 1 values, there must, by PP, be at least two x, y ∈ X with f (x) = f (y). And that’s
what we needed to prove.
CHAPTER 5. FUNCTIONS AND COUNTING 82

Here’s an interesting geometrical example. For two points (x1 , y1 ), (x2 , y2 ) in the plane, the
midpoint of (x1 , y1 ) and (x2 , y2 ) is the point

( 21 (x1 + x2 ), 12 (y1 + y2 ))

(the point on the middle of the line connecting (x1 , y1 ) to (x2 , y2 ) ).

Theorem 5.20. If we have a set A of five or more points in the plane with integer coordinates,
then there are two points in A whose midpoint has integer coordinates.
Proof. For two integers a, b, 12 (a + b) is an integer if and only if a + b is even, so if and only if
a, b are both even or are both odd.
So the midpoint of (x1 , y1 ), (x2 , y2 ) has both coordinates integer if and only if x1 , x2 are
both even or both odd, and also y1 , y2 are both even or both odd.
Let’s label each of the points (a, b) of A with one of “(even,even)”, “(even,odd)”, “(odd,even)”
or “(odd,odd)”.
Since ∣A∣ ≥ 5, there will be at least two points which receive the same label. Hence these two
points have the same parity (odd or even) for the first coordinate, and the same parity for the
second coordinate. This means the midpoint of these two points must be integer as well.
By the way, this result would not necessarily hold if we only had four points in the set.
Consider (0, 0), (1, 0), (1, 0) and (1, 1).
Here’s a very interesting number theory application (with a very sneaky proof). It uses the
notion of remainders on division by n, which we’ll cover properly in Lent Term: for now, all we
need is that, for every natural number m, the “remainder, r, upon division by n” is one of the
numbers 0, 1, . . . , n − 1, and that m − r is divisible by n.
Theorem 5.21. Let a1 , a2 , . . . , an be n integers (where n ≥ 2). Then there exists a non-empty
collection of these integers whose sum is divisible by n.
Proof. Consider the numbers s0 , s1 , . . . , sn given by

s0 = 0,

s 1 = a1 ,
s 2 = a1 + a2 ,
s 3 = a1 + a2 + a3 ,
etc., until
s n = a1 + a2 + ⋯ + an .
(It is not obvious, at all, why we should do this, but it will work!)
For each of these si , consider the remainder upon division by n. Since there are n + 1
numbers si , but only n possible remainders (0, 1, . . . , n − 1), two of the si will have the same
remainder upon division by n.
So suppose sk and s` have the same remainder, where k < `. Then s` − sk is divisible by n.
But since s` − sk = ak+1 + ak+2 + ⋯ + a` , this means that the sum ak+1 + ak+2 + ⋯ + a` is divisible
by n. Se we have proved the result.
In fact we proved something even stronger than what we set out to prove :
Let a1 , a2 , . . . , an be a list of n integers ( where n ≥ 2 ). Then there exists a non-empty collection
of consecutive numbers from this list ak+1 , ak+2 , . . . , a` whose sum is divisible by n.
The theorem isn’t true if we have fewer than n integers. For instance, if for any n ≥ 2 we
take the numbers a1 , . . . , an−1 all equal to 1, then it’s impossible to find a sum that adds up to
something divisible by n.
CHAPTER 5. FUNCTIONS AND COUNTING 83

5.8 A generalised form of PP

We state without proof the following more general version of the PP. Again, it’s rather obvious.
Isn’t it?
Theorem 5.22. Suppose f ∶ A → B and that ∣A∣ > k∣B∣ where k ∈ N. Then there is some element
of B that is the image of at least k + 1 elements of A.
I should maybe point out why the proof of this is not in the course. First, it is something
you can find or generate for yourself fairly easily if you want. More importantly, it won’t show
you any new ideas; you wouldn’t learn anything you didn’t already see earlier.
Last year, 241 students were registered for this course. I knew, before marking the exams,
that at least three of them would get the same exam mark.
Why? Well, apply the theorem, with A being the students, B being the set {0, 1, . . . , 100} of
all possible marks (which is of size 101) and f (x) the mark of student x. Since 241 > 2(101),
there’s some mark y such that at least 2 + 1 = 3 students will have y = f (x), which means they
get the same mark.

5.9 Infinite sets

We say that a set A is finite when there is some n ∈ N such that ∣A∣ = n. Otherwise, A is said to
be infinite.
For example, the set of natural numbers is infinite. You might think that’s obvious, but how
would you prove it? (Remember that the formal definition that a set A has cardinality n is that
there is a bijection between Nn and A.)
One way to show this is to use a proof by contradiction. Suppose (for a contradiction)
that N is finite, of cardinality n ∈ N, and that f ∶ Nn → N is a bijection. Consider the number
N = f (1) + f (2) + ⋯ + f (n). Since each f (i) is a natural number, for all i ∈ Nn , N is also a
natural number. But N > f (i) for all i ∈ Nn . So here is a natural number, N , that is not equal to
f (i) for any i ∈ Nn . But that contradicts the fact that f is a bijection, because if it’s a bijection
then it’s certainly a surjection and there should be some i ∈ Nn with f (i) = N .

5.10 Sample exercises

Exercise 5.1. Suppose that X, Y, Z are sets and that f ∶ X → Y and g ∶ Y → Z. Prove that if f
and g are injections, so is the composition g ○ f . Prove also that if f and g are surjections, then
so is the composition g ○ f .
Exercise 5.2. Let Z be the set of all integers and suppose that f ∶ Z → Z is given, for x ∈ Z, by

x + 1 if x is even
f (x) = {
−x + 3 if x is odd.

Determine whether f is injective. Determine also whether f is surjective.

Exercise 5.3. Suppose that X, Y, Z are sets, and we have functions f ∶ X → Y , g ∶ Y → Z, and
h ∶ Y → Z. Suppose that the compositions h ○ f and g ○ f are equal, and also that f is surjective.
Prove that g = h.
Exercise 5.4. Suppose that X, Y, Z are sets and that f ∶ X → Y and g ∶ Y → Z. Prove that if
the composition g ○ f is injective, then f is injective. Prove that if g ○ f is surjective, then g is
surjective.
CHAPTER 5. FUNCTIONS AND COUNTING 84

Exercise 5.5. Suppose that A and B are non-empty finite sets and that they are disjoint (i.e.
A ∩ B = ∅). Prove, using the formal definition of cardinality, that ∣A ∪ B∣ = ∣A∣ + ∣B∣.

Exercise 5.6. Suppose that X, Y are any two finite sets. By using the fact that

X ∪ Y = (X ∖ Y ) ∪ (Y ∖ X) ∪ (X ∩ Y ),

together with the result of Exercise 5.5, prove that

∣X ∪ Y ∣ = ∣X∣ + ∣Y ∣ − ∣X ∩ Y ∣.

Exercise 5.7. Suppose n ∈ N and that f ∶ N2n+1 → N2n+1 is a bijection. Prove that there is some
odd integer k ∈ N2n+1 such that f (k) is also odd. (State clearly any results you use.)

5.11 Comments on selected activities

Comment on Activity 5.1. To get started, observe that we can describe the function h(x) defined
by h(x) = 0 for x < 0 and h(x) = 2x for x ≥ 0 using the formula h(x) = x + ∣x∣, where ∣x∣ is (as is
√ value of x, i.e. the function ∣x∣ = x if x ≥ 0 and ∣x∣ = −x if x < 0. (We could
usual) the absolute
also write ∣x∣ = x2 ). It follows that

g(x) = 1
10
((x − 11850) + ∣x − 11850∣) + 10
1
((x − 46350) + ∣x − 46350∣) .

Would that formula be more or less useful to you than the description we gave to define it?
Comment on Activity 5.2. Given any y ∈ R, let x = y/2. Then f (x) = 2(y/2) = y. This shows
that f is surjective. Also, for x, y ∈ R,

f (x) = f (y) Ô⇒ 2x = 2y Ô⇒ x = y,

which shows that f is injective. Hence f is a bijection.

5.12 Solutions to exercises

Solution to Exercise 5.1. Suppose f and g are injective. Then, for x, y ∈ X,

(g ○ f )(x) = (g ○ f )(y) Ô⇒ g(f (x)) = g(f (y))

Ô⇒ f (x) = f (y) (because g is injective)
Ô⇒ x = y (because f is injective).

This shows that g ○ f is injective.

Suppose that f and g are surjective. Let z ∈ Z. Then, because g is surjective, there is some
y ∈ Y with g(y) = z. Because f is surjective, there is some x ∈ X with f (x) = y. Then

(g ○ f )(x) = g(f (x)) = g(y) = z,

so z is the image of some x ∈ X under the mapping gf . Since z was any element of Z, this shows
that g ○ f is surjective.
Solution to Exercise 5.2. Suppose one of x, y is even and the other odd. Without any loss of
generality, we may suppose x is even and y odd. (‘Without loss of generality’ signifies that there
is no need to consider also the case in which x is odd and y is even, because the argument we’d
use there would just be the same as the one we’re about to give, but with x and y interchanged.)
CHAPTER 5. FUNCTIONS AND COUNTING 85

So f (x) = x + 1 and f (y) = −y + 3. But we cannot then have f (x) = f (y) because x + 1 must be
an odd number and −y + 3 an even number. So if f (x) = f (y), then x, y are both odd or both
even. If x, y are both even, this means x + 1 = y + 1 and hence x = y. If they are both odd, this
means −x + 3 = −y + 3, which means x = y. So we see that f is injective.
Is f surjective? Let z ∈ Z. If z is odd, then z − 1 is even and so f (z − 1) = (z − 1) + 1 = z. If
z is even, then 3 − z is odd and so f (3 − z) = −(3 − z) + 3 = z. So for z ∈ Z there is x ∈ Z with
f (x) = z and hence f is surjective.
Solution to Exercise 5.3. Suppose f is surjective and that h ○ f = g ○ f . Let y ∈ Y . We show
g(y) = h(y). Since y is any element of Y in this argument, this will establish that g = h.
Because f is surjective, there is some x ∈ X with f (x) = y. Then, because h ○ f = g ○ f , we have
h(f (x)) = g(f (x)), which means that h(y) = g(y). So we’ve achieved what we needed.
Solution to Exercise 5.4. Suppose g ○ f is injective. To show that f is injective we need to show
that f (x) = f (y) Ô⇒ x = y. Well,

f (x) = f (y) Ô⇒ g(f (x)) = g(f (y))

by definition of a function. Now g(f (x)) = (g ○ f )(x), and similarly for y; this is what ○ means.
And
(g ○ f )(x) = (g ○ f )(y) Ô⇒ x = y ,
because g ○ f is injective. So we proved

f (x) = f (y) Ô⇒ x = y ,

i.e. f is injective.
Now suppose g ○ f is surjective. So for all z ∈ Z there is some x ∈ X with (g ○ f )(x) = z. So
g(f (x)) = z. Denoting f (x) by y, we therefore see that there is y ∈ Y with g(y) = z. Since z was
any element of Z, this shows that g is surjective.
Solution to Exercise 5.5. Suppose ∣A∣ = m and ∣B∣ = n. We need to show that ∣A ∪ B∣ = m + n
which means, according to the definition of cardinality, that we need to show there is a bijection
from Nm+n to A ∪ B. Because ∣A∣ = m, there is a bijection f ∶ Nm → A and because ∣B∣ = n, there
is a bijection g ∶ Nn → B. Let us define h ∶ Nm+n → A ∪ B as follows:

for 1 ≤ i ≤ m, h(i) = f (i) and for m + 1 ≤ i ≤ m + n, h(i) = g(i − m).

Then h is injective. We can argue this as follows: if 1 ≤ i, j ≤ m then

h(i) = h(j) Ô⇒ f (i) = f (j) Ô⇒ i = j,

because f is injective. If m + 1 ≤ i, j ≤ m + n then

h(i) = h(j) Ô⇒ g(i − m) = g(j − m) Ô⇒ i − m = j − m Ô⇒ i = j,

because g is injective. The only other possibility is that one of i, j is between 1 and m and the
other between m + 1 and m + n. In this case, the image under h of one of i, j belongs to A and
the image of the other to B and these cannot be equal because A ∩ B = ∅. So h is indeed an
injection. It is also a surjection. For, given a ∈ A, because f is a surjection, there is 1 ≤ i ≤ m
with f (i) = a. Then h(i) = a also. If b ∈ B then there is some 1 ≤ j ≤ n such that g(j) = b. But
then, this means that h(m + j) = g((m + j) − m) = b, so b is the image under h of some element
of Nm+n . So h is a bijection from Nm+n to A ∪ B and hence ∣A ∪ B∣ = m + n.
CHAPTER 5. FUNCTIONS AND COUNTING 86

Solution to Exercise 5.6. Note first that the two sets (X ∖ Y ) ∪ (Y ∖ X) and X ∩ Y are disjoint.
Therefore,
∣X ∪ Y ∣ = ∣(X ∖ Y ) ∪ (Y ∖ X)∣ + ∣X ∩ Y ∣.
Now, (X ∖ Y ) and (Y ∖ X) are disjoint, so

∣(X ∖ Y ) ∪ (Y ∖ X)∣ = ∣(X ∖ Y )∣ + ∣(Y ∖ X)∣

and therefore
∣X ∪ Y ∣ = ∣(X ∖ Y )∣ + ∣(Y ∖ X)∣ + ∣X ∩ Y ∣.
Now, the sets X ∖ Y and X ∩ Y are disjoint and their union is X, so

∣X∣ = ∣(X ∖ Y ) ∪ (X ∩ Y )∣ = ∣X ∖ Y ∣ + ∣X ∩ Y ∣.

A similar argument shows that

∣Y ∣ = ∣(Y ∖ X) ∪ (X ∩ Y )∣ = ∣Y ∖ X∣ + ∣X ∩ Y ∣.

These mean that

∣X ∖ Y ∣ = ∣X∣ − ∣X ∩ Y ∣ and ∣Y ∖ X∣ = ∣Y ∣ − ∣X ∩ Y ∣.
So we have

∣X ∪ Y ∣ = ∣(X ∖ Y )∣ + ∣(Y ∖ X)∣ + ∣X ∩ Y ∣

= (∣X∣ − ∣X ∩ Y ∣) + (∣Y ∣ − ∣X ∩ Y ∣) + ∣X ∩ Y ∣
= ∣X∣ + ∣Y ∣ − ∣X ∩ Y ∣.

Solution to Exercise 5.7. Let E be the set of even integers, and O the set of odd integers, in the
range {1, 2, . . . , 2n + 1}. Then ∣E∣ = n and ∣O∣ = n + 1. If f was such that f (k) was even for all
k ∈ O, then f ∗ ∶ O → E given by f ∗ (x) = f (x) would be an injection. But, by the pigeonhole
principle, since ∣O∣ > ∣E∣, such an injection cannot exist. Hence there is some odd k such that
f (k) is odd.
 Equivalence relations and the rational numbers
6
The material in this chapter is also covered in:

• Biggs, N. L. Discrete Mathematics. Chapter 7.

• Eccles, P.J. An Introduction to Mathematical Reasoning. Chapter 22.

6.1 Introduction
In this chapter of the notes we study the important idea of an equivalence relation, a concept
that is central in abstract mathematics. As an important example, we show how to formally
construct the rational numbers using the integers and a carefully chosen equivalence relation.
We will return to this in Lent Term.

6.2 Equivalence relations

6.2.1 Relations in general
The idea of a relation is quite a general one. For example, consider the set of natural numbers N
and let us say that two natural numbers m, n are related, denoted by m R n, if m + n is even.
So we have, for instance, 6 R 2 and 7 R 5, but that 6 and 3 are not related. This relation has
some special properties. For one thing, since 2n is even for all n ∈ N, n R n for all n ∈ N. (We say
such a relation is reflexive.) Also, if m R n, then m + n is even. But m + n = n + m and hence,
also, n R m. (We say such a relation is symmetric.) It is because m R n ⇐⇒ n R m that we can
simply say that ‘m and n are related’ rather than ‘m is related to n’ or ‘n is related to m’. The
relation R has other important properties that we will come back to later.
Formally, a relation R on a set X is a subset of the Cartesian product X × X (which, recall,
is the set of all ordered pairs of the form (x, y) where x, y ∈ X). You should just keep in mind
that x R y is a true-or-false statement; if you’re not told any more about the relation, there’s
not much more you can say—maybe for some x and y you are told x R y is true, but it doesn’t
tell you whether or not y R x is true, for example.
In some textbooks, the author insists on using the Cartesian product notation; so you might
see (6, 2) ∈ R where we write 6 R 2. The Cartesian product notation has the advantage of being
clear and unambiguous, but the (big!) disadvantage that you already know a lot of relations,
such as equality, greater than, and so on, and in fact you write them in the 6 R 2 style.
Example 6.1. Suppose R is the relation on R given by x R y ⇐⇒ x > y. Regarded as a subset
of R×R, this is the set {(x, y) ∣ x > y}. This relation does not possess the reflexive and symmetric

87
CHAPTER 6. EQUIVALENCE RELATIONS AND THE RATIONAL NUMBERS 88

properties we met in the example above. For no x ∈ R do we have x R x because x is not greater
than x. Furthermore, if x R y then x > y, and we cannot therefore also have y R x, for that would
imply the contradictory statement that y > x.

In many cases, we use special symbols for relations. For instance ‘=’ is a relation, as is >. It
is often convenient to use a symbol other than R: for instance, many textbooks use x ∼ y rather
than x R y as a symbol for ‘some relation’, particularly if the relation is an equivalence relation
(see below).

6.2.2 The special properties of equivalence relations

There are three special properties that a relation might have (two of which we saw in one of the
earlier examples):

Definition 6.2. Suppose that R is relation on a set X. Then

• [The reflexive property] R is said to be reflexive if, for all x ∈ X, x R x.

• [The symmetric property] R is said to be symmetric if, for all x, y ∈ X, x R y implies

y R x (equivalently, for all x, y ∈ X, x R y ⇐⇒ y R x).

• [The transitive property] R is said to be transitive if, for all x, y, z ∈ X, whenever x R y

and y R z, we also have x R z; that is, (x R y) ∧ (y R z) Ô⇒ xRz.

A relation that has all three of these properties is called an equivalence relation.

Definition 6.3. A relation is an equivalence relation if is reflexive, symmetric and transitive.

Example 6.4. We saw earlier that the relation on N given by

m R n ⇐⇒ m + n is even

is reflexive and symmetric. It is also transitive. To prove that, suppose x, y, z are three natural
numbers and that x R y and y R z. Then x + y is even and y + z is even. To show that x R z we
need to establish that x + z is even. Well,

x + z = (x + y) + (y + z) − 2y,

and all three terms on the right (x + y, y + z, and 2y) are even. Therefore, x + z is even and so
x R z.

Example 6.5. Let X be the set of n × n real matrices. Define a relation ∼ on X by:

M ∼ N ⇐⇒ ∃r, s ∈ N such that M r = N s .

Then ∼ is an equivalence relation.

Reflexivity and symmetry are easy to see: M 1 = M 1 and, if M r = N s , then N s = M r . Proving
transitivity requires more work. Suppose M ∼ N and N ∼ R. Then there are r, s, t, u ∈ N with
M r = N s and N t = Ru . Then

M rt = (M r )t = (N s )t = (N t )s = (Ru )s = Rus ,

so there are integers w = rt and x = us such that M w = Rx and hence M ∼ R.

CHAPTER 6. EQUIVALENCE RELATIONS AND THE RATIONAL NUMBERS 89

Example 6.6. Let S be a set of people in a given social network, and let F be the relation
‘friendship’, i.e. aF b if a and b are people in S who are friends in the social network. This
relation is symmetric (in real life, it might be that a says they are friends with b but b disagrees.
Social networks such as Facebook don’t allow this one-sided ‘friendship’). Let’s say that you are
automatically a friend of yourself, so the relation is reflexive.
Is the relation transitive? Well, that depends on the social network. You probably want to
say ‘No’, because (if you’re on Facebook) you surely have some friend not all of whose friends
you know. So for the example of S and F coming from Facebook, you know the relation F is
not transitive; you have a counterexample—and hence it’s also not an equivalence relation. But
it doesn’t have to be that way. If S is all the people in this lecture hall—well, we’re all friends
(I hope!) and so from the lecture example we do get a transitive relation, and hence (because we
checked all three properties) an equivalence relation.

6.3 Equivalence classes

Given an equivalence relation, it is natural to group together objects that are related to each
other. The resulting groupings are known as equivalence classes. In this section, we formally
define equivalence classes and discuss some of their properties.
Definition 6.7. Suppose R is an equivalence relation on a set X and, for x ∈ X, let [x]R be
the set of all y ∈ X such that y R x. So,

[x]R = {y ∈ X ∣ y R x}.

Often, we will want to talk about the set of all equivalence classes of R. This set is written
X/R, and referred to as the quotient set of X by R. So we have

X/R = {[x]R ∶ x ∈ X} .

Notice that each [x]R is a subset of X. If R is clear from the context—which it usually will
be; in general we will only be talking about one equivalence relation at any given time—we may
just write [x] for [x]R .
Example 6.8. Consider again R on N given by m R n ⇐⇒ m + n is even. Any even number is
related to any other even number; and any odd number to any odd number. So there are two
equivalence classes:
[1] = [3] = [5] = ⋯ = {n ∈ N ∣ n is odd} ,
[2] = [4] = [6] = ⋯ = {n ∈ N ∣ n is even} ,
and we have N/R = {[1], [2]}.
You should keep in mind that even though we use the word ‘equivalence class’, what an
equivalence class is, is simply a set: and you know how to handle sets. The name ‘equivalence
class’ is just to remind you that this particular set is a special set and (as we’ll shortly see)
they have some extra nice properties. Similarly, we might say that 3 is a representative of the
equivalence class [1] (in the above example). That means exactly the same as saying that 3 is a
member of the set [1]; we use the word ‘representative’ instead of ‘member’ to remind ourselves
that we are dealing with a special set.
Example 6.9. Given a function f ∶ X → Y , define a relation R on X by x R z ⇐⇒ f (x) = f (z).
Then R is an equivalence relation. If f is a surjection, the equivalence classes are the sets

{x ∈ X ∶ f (x) = y} = f −1 ({y}),
 CHAPTER 6. EQUIVALENCE RELATIONS AND THE RATIONAL NUMBERS 90

for y ∈ Y . Note that the place where we use that f is a surjection is that it implies each f −1 ({y})
is non-empty. If f is not a surjection, then the equivalence classes are the sets f −1 ({y}) for all
y ∈ Y such that there is an x ∈ X with y = f (x), in other words for each y ∈ f (X).

Activity 6.1. Check this!

The equivalence classes have a number of important properties. These are given in the
following result.

Theorem 6.10. Suppose R is an equivalence relation on a set X. Then

(i) For x, y ∈ X, [x] = [y] ⇐⇒ x R y

(ii) For x, y ∈ X, if x and y are not related by R, then [x] ∩ [y] = ∅.

Proof. (i) This is an if and only if statement, so we have two things to prove: namely that
[x] = [y] Ô⇒ x R y and that x R y Ô⇒ [x] = [y].
Suppose, then, that [x] = [y]. The relation R is reflexive, so we have x R x. This means that
x ∈ [x]. But if [x] = [y], then we must have x ∈ [y]. But that means (by definition of [y]) that
x R y.
Conversely, suppose that x R y. We now want to show that [x] = [y]. So let z ∈ [x]. (We
will show that z ∈ [y].) Then z R x. But, because x R y and R is transitive, it follows that z R y
and hence z ∈ [y]. This shows [x] ⊆ [y]. We now need to show that [y] ⊆ [x]. Suppose w ∈ [y].
Then w R y and, since x R y, we also have, since R is symmetric, y R x. So w R y and y R x. By
transitivity of R, w R x and hence w ∈ [x]. This shows that [y] ⊆ [x]. Because [x] ⊆ [y] and
[y] ⊆ [x], [x] = [y], as required.
(ii) Suppose x and y are not related. We prove by contradiction that [x] ∩ [y] = ∅. So suppose
[x] ∩ [y] ≠ ∅. Let z be any member of the intersection [x] ∩ [y]. (The fact that we’re assuming
the intersection is non-empty means there is such a z.) Then z ∈ [x], so z R x and z ∈ [y], so
z R y. Because R is symmetric, x R z. So: x R z and z R y and, therefore, by transitivity, x R y.
But this contradicts the fact that x, y are not related by R. So [x] ∩ [y] = ∅.
Theorem 6.10 shows that either two equivalence classes are equal, or they are disjoint.
Furthermore, because an equivalence relation is reflexive, any x ∈ X is in some equivalence class
(since it certainly belongs to [x] because x R x). So what we see is that the equivalence classes
form a partition of X: their union is the whole of X, and no two equivalence classes overlap.

Example 6.11. Consider again the equivalence relation R on N given by

m R n ⇐⇒ m + n is even.

We have seen that there are precisely two equivalence classes: the set of odd positive integers
and the set of even positive integers. Note that, as the theory predicted, these form a partition
of all of N (since every natural number is even or odd, but not both).
CHAPTER 6. EQUIVALENCE RELATIONS AND THE RATIONAL NUMBERS 91

6.3.1 What’s the point?

You’re used to saying that two numbers are equal (or indeed two sets, or two functions...).
Equality of integers (for example) is an equivalence relation, as you can easily check. It’s not a
very interesting equivalence relation, because its equivalence classes are all sets of size one.
In general, an equivalence relation is what we want to have when we get to the situation
‘these two things are kind of the same, but they’re not actually equal’. Let’s give a familiar
example.
If I give you a function f ∶ R → R which has an integral ∫ f (x) dx, what do I mean by that?
Well, I might say (for a specific example) ∫ 3x2 dx = x3 . But I could also write ∫ 3x2 dx = x3 + 10.
As you know from school, an indefinite integral is only defined up to a constant; you probably
were told to write ∫ 3x2 dx = x3 + C where C is a constant. As you know, if you want to work
7
out some definite integral, like ∫x=−4 3x2 dx, the constant C cancels and it doesn’t really matter
what you chose.
We’d like to say ‘the indefinite integral of f (x) is the function F (x) such that dx
d
F (x) = f (x)’.
But that is not well-defined. There are lots of possible choices for F (x). They are functions
which differ by a constant, so we can’t say ‘the’ function; ‘the’ implies there is exactly one, which
is not true.
In school, you dealt with this by always writing +C. That’s fine (i.e. it works), but it gets
annoying, especially if you have several integrals to do and you therefore need to write down
several different letters for the different constants of integration. Most of the time (not all..!) it
doesn’t really matter what the constant is. You’d like to say these different functions are ‘kind
of the same’.
Define an equivalence relation S on functions f ∶ R → R by f Sg if and only if

∃C ∈ R ∶ ∀x ∈ R , f (x) − g(x) = C .

In English, we say f Sg if the functions f and g differ by a constant.

Activity 6.2. Check that S is an equivalence relation.

What are the equivalence classes of S? Well, they are sets of functions. If f is any particular
function, then [f ]S is
the set of all functions g ∶ R → R such that g(x) = f (x) + C for some constant C.
We can say that ∫ 3x2 dx is [x3 ]S . That is, if we pick any function f in [x3 ]S (remember,
this is a set of functions) then we have dx d
f (x) = 3x2 . And if we pick any function that’s not in
[x ]S , then its derivative (if it is has one!) will not be 3x2 .
3

The advantage of this is that we can work with several indefinite integrals without having to
write several different letters for constants of integration, but while still being reminded by the
notation [f ]S that when it matters we should put them in. We formalised the idea that x3 + 5
and x3 + 10 are ‘kind of the same’ as far as being indefinite integrals is concerned: what it means
is they’re both representatives of [x3 ]S .
In this case, we don’t really gain a lot by introducing this equivalence relation. It is also
pretty easy to write +C whenever we want to write an indefinite integral. What we are doing
when we write +C every time, is simply writing out explicitly the definition of the equivalence
relation S on every line. Whatever equivalence relation you are asked to work with, you can
always write out the definition on every line instead of talking about equivalence classes.

However, if the equivalence relation we want to work with is complicated, it very quickly
gets painful to write out the definition every time, and the mass of written-out-definition makes
it hard to see what’s important. Let’s see an (important!) example.
CHAPTER 6. EQUIVALENCE RELATIONS AND THE RATIONAL NUMBERS 92

6.4 Rational numbers

You’re used to working with rational numbers—that is, the fractions, the set Q, the numbers
we can write as pq where p and q are integers (and q is not zero). You know how to add and
multiply fractions, and so on. But why does all that makes sense? We can certainly think of the
natural numbers as representing a physical concept—these are the numbers you use to count
things (plus zero), the natural numbers N. It makes sense to say that 2 + 3 = 3 + 2, because either
way, if we’re counting apples we’ve counted to five apples. You’re probably used enough by now
to debt to feel that the concept of −1 is a natural enough thing (even if you are maybe not so
happy with a negative bank balance), so you are presumably happy with the integers Z. But
what about fractions? What does it mean to multiply them?
Think about this for a moment—you might say you can make sense of −1 5 in terms of reality:
it means I owe you a fifth of an apple; if there are five people who each owe you a fifth of an
apple, then you are owed one apple, and this all sounds good. But what does it mean to multiply
5 ? Why is 5 ⋅ 5 equal to 25 ? You can probably come up with some sentence involving
1
by −1 −1 −1

apples, but I’m not sure it will be very convincing—try it!

Another thing you can think about: you know 17 and 14 2
are the same fraction. But why? If you
plug both into your calculator, then you’ll get the same sequence of digits on the display—but
this is only part of a complicated infinite sequence of digits; maybe they are different somewhere
later off your screen? The answer is something like ‘because you can cancel twos’. How can we
make that formal?

6.4.1 An important equivalence relation

Rational numbers are simply the fractions you already studied in primary school. You’ll certainly
be aware that there are many ways of representing a given rational number. For instance, 25
4
represents the same number as 10 . We can capture these sorts of equivalences more formally by
using an equivalence relation on pairs of integers (m, n), where n ≠ 0. So let X = Z × (Z ∖ {0})
be the set of all pairs (m, n) where m, n ∈ Z and n ≠ 0, and define a relation Q on X by:

(m, n) Q (m′ , n′ ) ⇐⇒ mn′ = m′ n.

You should quickly check that this relation Q does what you think it should do: if (by
m′
your school-style calculation) the fractions m n and n′ are the same, then indeed we have
(m, n)Q(m′ , n′ ). However so far in this course we did not define ‘division’ nor ‘fraction’—that’s
exactly what we want to do now. The relation Q only uses the properties of Z which we are
already happy with.
Let’s pause for a moment to prove that Q is indeed an equivalence relation.
Q is Reflexive: (m, n)Q(m, n) because mn = nm.
Q is Symmetric: (m, n)Q(p, q) means mq = np. Rearranging we get pn = qm, which by
definition is the same as (p, q)Q(m, n).
Q is Transitive: Suppose (m, n)Q(p, q) and (p, q)Q(s, t). Then mq = np and pt = qs. So,
(mq)(pt) = (np)(qs) and, after cancelling qp, this gives mt = ns, so (m, n)Q(s, t).
But, wait a minute: can we cancel pq? Sure, if it’s nonzero. If it is zero then that means p = 0
(since we know that q ≠ 0). But then mq = 0, so m = 0; and qs = 0, so s = 0. So, in this case also
we get mt = ns (both sides are zero) and so (m, n)Q(s, t).
CHAPTER 6. EQUIVALENCE RELATIONS AND THE RATIONAL NUMBERS 93

6.4.2 Rational numbers as equivalence classes

We usually write the equivalence class [(m, n)]Q as m n . For example, we then have the (familiar)
fact that 5 = 10 which follows from the fact that [(2, 5)]Q = [(4, 10)]Q (i.e. these two sets are
2 4

equal). Why are they equal? You could prove it from the definition of set equality, but it’s
easier to use the theory we developed. We know (2, 5)Q(4, 10), because 2 × 10 = 4 × 5. And now
Theorem 6.10 tells us [(2, 5)]Q = [(4, 10)]Q .
We can now say what the set of fractions, written Q, is. It is the set of all the equivalence
classes of the relation Q. So 25 is a fraction. So is 21 . So is 10
4
, and (as we just saw) in fact 25 is
4
the same thing as 10 .
What we have done here is to find a way of making sense of fractions, and being able to say
when two fractions are the same, without ever having to define ‘division’. We are only relying
on the properties of Z—adding and multiplying integers—that you are already happy with.
Logically, this is the ‘right thing to do’.
To see why, think about how you might try to define ‘division’ without using fractions. You
can do that if you try hard enough (for example, you might write out exactly what you mean
by decimal long division) but you will end up with a complicated definition that’s hard to work
2 3
with (why is it true that 14 and 21 have the same decimal expansion..?). Whereas if we define
fractions, we can rather easily define division: we’ll just say that dividing by n means multiplying
by n1 . (We didn’t say what multiplying fractions means yet—we’ll get to that in Lent Term!)
and this turns out to be much easier to work with.
What about simply saying ‘obviously fractions exist and we can just work with them’ ? The
problem with this is: what if there is something we’re missing? It’s a bit hard to explain what
exactly the problem might be here: you are so used to fractions that you probably cannot
5 ⋅ 5 = 25 ; it’s hard to
1
imagine what could possibly be a problem. But it is a bit funny that −1 −1

explain what that should mean in ‘real world’ terms. The last time we saw something a bit
funny was in Section 3.6, and there, we did run into problems with a way of defining sets that
at first looks perfectly reasonable.
What we are doing here—we’re currently part-way through—is to construct the fractions
from the integers. So far, we defined an equivalence relation, and we created a set Q which is the
set of equivalence classes. A set on its own isn’t very interesting: we want to do things with it.
We would like to define addition and multiplication of fractions: which we will do in Lent Term.
This is of course something you know very well how to do from school, and in Lent Term what
you will see is nothing other than what you know from school. However, there is an important
point that you probably never thought about at school—briefly, why exactly is it that 17 + 82 and
14 + 12 come out to the same answer?
2 3

By the way, the rational numbers are described as such because they are (or, more formally,
can be represented by) ratios of integers.
CHAPTER 6. EQUIVALENCE RELATIONS AND THE RATIONAL NUMBERS 94

6.4.3 Non-examinable: Fields

We are going to continue to work with Q for the rest of this term, even though we did not
yet talk formally about how to add and multiply fractions. Once we do that, you will be able
to check that these operations have certain properties (that you know and used often before).
Specifically, Q is a field. What does that mean?

Definition 6.12 (Field axioms). F is a field, with operations + and ×, if we have:

(F1) Closure under addition and multiplication: for each a, b ∈ F both a + b and a × b are in F.

(F2) Commutative addition and multiplication: for each a, b ∈ F we have a + b = b + a and

a × b = b × a.

(F3) Associative addition and multiplication: for each a, b, c ∈ F we have (a + b) + c = a + (b + c)

and (a × b) × c = a × (b × c).

(F4) The distributive law: for each a, b, c ∈ F we have (a + b) × c = a × c + b × c.

(F5) Additive and multiplicative identity: there are two different elements 0 and 1, such that
for each a ∈ F we have a + 0 = a and a × 1 = a.

(F6) Additive and multiplicative inverses: for each a ∈ F there is an element −a such that
a + (−a) = 0, and if a ≠ 0 there is an element a−1 such that a × a−1 = 1.

If you want to in LT, you can do this check for Q without too much trouble: it’s (very) long,
but not hard. You’ll need to use the definition of Q together with its operations, and you’ll need
to know things like that multiplication of integers is commutative, and the distributive law for
the integers.
You should notice that the axioms for a field are all things you knew long ago are true for
the rational numbers, or for the real numbers. They are all part of the ‘doing algebra as normal’
that you learned in school. You’ll probably notice, too, that there are some properties that the
rational numbers have which aren’t listed. For example the rational numbers have an order
and you know how to do algebra with the < sign, too. We could have written down some more
axioms saying that there is an order and saying how ‘algebra as normal’ works with an order
(things like: if a < b then a + c < b + c). That would give us an ordered field. You will see later in
the course that some fields are not ordered—for example there is no way to put a sensible order
on C; no matter what you try, some piece of ‘algebra as normal with <’ won’t work.
This is your first (brief, and non-examinable!) introduction to the axiomatic approach to
mathematics, which we will see a good deal more of next term. What is the point?
Well, look to your MA100 notes on linear algebra (if you got that far—if not, it will start
in the next week or two) and look at a few statements. Matrix addition is commutative; both
addition and multiplication are associative. There isn’t really an explicit proof given for these
statements, but you can probably see how to check at least the statements for addition. You’ll
notice that all the things you do in your proof are using the field axioms. What that means is:
those statements are true not just for matrices of real numbers, but for any field. In fact, that’s
true for most of the linear algebra in MA100.
That turns out to be incredibly useful. There are many different fields in mathematics, and
some of them have practical applications too. For one example, the way that this document
was transmitted from my computer to yours makes use of linear algebra over the field Z2 .
Linear algebra over Z2 is the basis of coding theory, which is (part of) what makes Internet
communication work reliably.
CHAPTER 6. EQUIVALENCE RELATIONS AND THE RATIONAL NUMBERS 95

But you do not have to learn a whole new MA100 course to understand coding theory. All
you need to know is what Z2 is (next term) and to quickly check which results in MA100 use
only the field axioms. Then you can simply use what you learned in MA100 (and when you
want to do linear algebra over Z3 , you don’t even need to check which results are allowed, since
you already did that for Z2 ).
Not everything in MA100 does use only the field axioms. For √example, you can’t find a unit
vector in the direction (1), because the length of that vector is 2, which (we said earlier) isn’t
1

rational.
If we tried to do mathematics without the axiomatic approach to structures, we’d either
spend forever re-proving (and trying to learn!) dozens of almost-identical theorems for all the
different fields we want to do linear algebra over (and so on), or we would have to say something
vague like ‘matrix addition is commutative whenever we can do algebra as normal’. The first
of these is a waste of time and energy, the second is dangerous: what is algebra as normal?
Does it include taking square roots? Either you have to guess, or you have to check the proof
for square-rooting every time you want to use a theorem (which is again a waste of time and
energy).
With the axiomatic approach, we can simply say ‘matrix addition is commutative over any
field’: easy to learn and precise, once you remember what a field is. Quite a lot of mathematics
is like this: you need to learn some definition or concept, which at first looks like formality for
the sake of it (you don’t do algebra by thinking explicitly about the field axioms!) but the payoff
is that in the long run it will make your life easier.

Don’t be fooled into believing that checking the field axioms for Q from our definition
(remember, Q is a set of equivalence classes with some funny way to define addition and
multiplication) is somehow ‘automatically’ going to work. It’s reasonable to think: we start with
integers (where multiplication is commutative), we construct something with pairs of integers
and define ‘multiplication’, of course the multiplication will be commutative.
That’s not a valid argument. To see why, think about 2-by-2 matrices (with integer entries, say).
We build the set of 2-by-2 integer matrices by starting with the integers (where multiplication is
commutative) and define ‘multiplication’ of matrices, but the multiplication is not commutative.

6.5 Sample exercises

Exercise 6.1. Define a relation R on Z by: for x, y ∈ Z, x R y ⇐⇒ x2 = y 2 . Prove that R is an
equivalence relation, and describe the corresponding equivalence classes.

Exercise 6.2. Define the relation R on the set N by x R y if and only if there is some n ∈ Z
such that x = 2n y. Prove that R is an equivalence relation.

Exercise 6.3. Let X be the set of n × n real matrices. Define a relation ∼ on X by:

M ∼ N ⇐⇒ ∃ an invertible P ∈ X s.t. N = P −1 M P.

Prove that ∼ is an equivalence relation.

Exercise 6.4. Suppose that f ∶ X → Y is a surjection. Define the relation R on X by x R y ⇐⇒

f (x) = f (y). Prove that R is an equivalence relation. What are the equivalence classes? Let C
denote the set of equivalence classes [x] for x ∈ X. Prove that if [x] = [y] then f (x) = f (y). This
means that we can define a function g ∶ C → Y by: g([x]) = f (x). Prove that g is a bijection.

Exercise 6.5. Prove that the set {x ∈ Z ∣ x is a multiple of 4} has no lower bound.
CHAPTER 6. EQUIVALENCE RELATIONS AND THE RATIONAL NUMBERS 96

6.6 Solutions to exercises

Solution to Exercise 6.1. R is reflexive because for any x, x2 = x2 . R is symmetric because
x2 = y 2 ⇐⇒ y 2 = x2 . To show R is transitive, suppose x, y, z ∈ Z and x R y and y R z. Then
x2 = y 2 and y 2 = z 2 , so x2 = z 2 , which means x R z. Thus R is an equivalence relation. Given
any x ∈ Z, the equivalence class [x] consists precisely of those integers y such that y 2 = x2 . So
[x] = {x, −x}.
Solution to Exercise 6.2. R is reflexive because for any x, x = 20 x. R is symmetric because if
x R y then ∃n ∈ Z with x = 2n y. This means that y = 2−n x and hence, taking m = −n, ∃m ∈ Z
such that y = 2m x. So y R x. To show R is transitive, suppose x, y, z ∈ Z and x R y and y R z.
Then there are m, n ∈ Z such that x = 2n y and y = 2m z, so x = 2n y = 2n (2m z) = 2m+n z which,
since m + n ∈ Z, shows that x R z. Thus R is an equivalence relation.
Solution to Exercise 6.3. For any M , M = I −1 M I where I is the identity matrix, so M ∼ M . For
matrices M, N ∈ X, if M ∼ N then there’s an invertible P with N = P −1 M P and so M = P N P −1 ,
which can be written as M = (P −1 )−1 M P −1 . So there is an invertible matrix Q (equal to P −1 )
such that M = Q−1 N Q and hence M ∼ N . This shows the relation is symmetric. Suppose M ∼ N
and N ∼ R. Then there are invertible matrices P and Q such that N = P −1 M P and R = Q−1 N Q.
We therefore have

R = Q−1 (P −1 M P )Q = (Q−1 P −1 )M (P Q) = (P Q)−1 M (P Q),

so there is an invertible matrix T = P Q so that R = T −1 M T and hence M ∼ R, establishing that

∼ is transitive. It follows that ∼ is an equivalence relation. (We used here the fact that (P Q)−1 =
Q−1 P −1 . This follows from the fact that (Q−1 P −1 )(P Q) = Q−1 (P −1 P )Q = Q−1 IQ = Q−1 Q = I.)
Solution to Exercise 6.4. x R x because f (x) = f (x). If x R y then f (x) = f (y) so f (y) = f (x)
and hence y R x. If x R y and y R z then f (x) = f (y) and f (y) = f (z), so f (x) = f (z) and x R z.
Hence R is an equivalence relation.
For x ∈ X, [x] is the set of all y ∈ X with f (y) = f (x), so, since f is a surjection, the
equivalence classes are exactly the sets Cz for each z ∈ Y , where Cz = {x ∈ X ∣ f (x) = z} is the
set of elements of X mapped onto z by f .
The fact that [x] = [y] implies f (x) = f (y) follows directly either from this description of
equivalence classes, or from the fact that [x] = [y] implies x R y, which implies f (y) = f (x).
Let g be as defined. It is surjective because for each z ∈ Y , there is some x ∈ X such
that f (x) = z (since f is surjective) and hence g([x]) = f (x) = z. Also, g is bijective because
g([x]) = g([y]) implies f (x) = f (y), which means x R y and hence that [x] = [y].
Solution to Exercise 6.5. We can prove this by contradiction. Suppose that the set S = {x ∈ Z ∣
x is a multiple of 4} has a lower bound, l. Then, for all x ∈ S, x ≥ l. Now, one of l − 1, l − 2, l − 3, l − 4
must be a multiple of 4. So one of these numbers is in S. However, each is less than l, contradicting
the fact that l is a lower bound on S.
 Real and complex numbers
7
The material in this chapter is also covered in:

• Biggs, N. L. Discrete Mathematics. Chapter 9.

• Eccles, P.J. An Introduction to Mathematical Reasoning. Chapters 13 and 14.

The treatment in Biggs is probably better for the purposes of this course.
Neither of these books covers complex numbers. You do not have to know very much about
complex numbers for this course, but because this topic is not in these books, I have included
quite a bit of material on complex numbers in this chapter.
You can find useful reading on complex numbers in a number of books, including the following
(which you might already have, given that it is the MA100 text).

• Anthony, M. and M. Harvey. Linear Algebra: Concepts and Methods. Cambridge University
Press 2012. Chapter 13.

7.1 Introduction
In this chapter, we explore real numbers and complex numbers.
We are going to stick, mainly, to your intuition and what you already know about numbers
from school—which means we are not going to formally construct the real numbers.

7.2 Rational numbers and real numbers

So far, you probably never really saw a need for numbers which are not rational. You can add,
subtract, multiply and divide rational numbers and you always get a rational number—why do
we need more?
√
Theorem 7.1. The real number 2 is irrational. That is, there are no positive integers m, n
2
with ( m
n
) = 2.

Proof. Suppose, for a contradiction, that there were such m, n.

If m, n are divisible by some d > 1, we may divide both m and n to obtain m′ , n′ such that
the rational number m′ /n′ equals m/n. So we may assume that m, n have no common divisors
greater than 1. In particular, we can assume that they are not both divisible by 2.
Now, the equation (m/n)2 = 2 means m2 = 2n2 . So we see that m2 is even. We know (from
Chapter 2) that this means m must be even. So there is some m1 such that m = 2m1 . Then,

97
CHAPTER 7. REAL AND COMPLEX NUMBERS 98

m2 = 2n2 becomes 4m21 = 2n2 , and so n2 = 2m21 . Well, this means n2 is even and hence n must
be even. So m and n are both divisible by 2. But this is a contradiction; we just said we can
assume they are not both divisible by 2.
So our assumption that (m/n)2 = 2 must have been wrong and we can deduce no such
integers m and n exist.
Isn’t this theorem a thing of beauty?

Activity 7.1. Make sure you understand that this is a proof by contradiction, and that you
understand what the contradiction is.

What this theorem tells us is that, at least if we want to solve equations like x2 = 2, then the
rational numbers are not enough; we need more. Of course, we could just invent new symbols
and define them to satisfy the equations we want. But this is a dangerous thing to do—we
might be assuming something exists which doesn’t in fact exist; whose existence leads to a
contradiction. We’d better rather construct the reals.

7.2.1 Non-examinable: what are the real numbers exactly?

There is a simple way to define the real numbers, which you already saw in school. We just say
that these are all the things you can get by writing down a decimal number: 123.4124581... for
example. Except that we say 0.9999 and 1.000 are to be considered the same (and similarly for
any other decimal which after some point consists only of nines). How do we formalise that?
Well, it’s easy enough; we can write a decimal as consisting of (for example) an integer n
together with a string of digits after the decimal: (123, 4, 1, 2, 4, 5, 8, 1, ...). This is a member of the
set N × {0, 1, . . . , 9} × {0, 1, . . . , 9} × . . . . (There’s nothing wrong with an infinite product of sets.)
And we can easily write down an equivalence relation which says that (n, a1 , a2 , . . . , ak , 9, 9, 9, . . . )
is equivalent to (n, a1 , a2 , . . . , ak +1, 0, 0, 0, . . . ) (whenever ak is not 9); and the relation is reflexive—
we just did it. The real numbers are then the equivalence classes of this relation.
It is a pain to define addition and multiplication formally. It is not hard, it is just annoying.
You simply write out the details of how you would in practice add or multiply two numbers by
hand (as you learnt to do in primary school). It’s not hard, but it is painful, to check that it
doesn’t matter which representative of an equivalence class you take.
But there is something a bit worrying about this definition: it depends on the fact that we
do arithmetic in base 10. Maybe the Martians (who have six fingers, at least in the B-movies I
watch) will have a different set of real numbers? That would be terrible—they would certainly
attack us if they found out. And in any case, it’s not even obvious this definition fixes the
problem—is there a decimal whose square is equal to 2?
In fact, these concerns turn out not to be real problems. It doesn’t make a difference what
base you use, and there is such a decimal. But nevertheless, mathematicians tend to prefer one
of two different constructions. These two constructions are really different; depending on what
you want one or the other might look ‘better’, and some mathematicians will get very angry
if you don’t like their favourite construction. However—and we will not prove it!—it doesn’t
really make a difference which construction you use; you still get a set which behaves the way
you think the real numbers should.
CHAPTER 7. REAL AND COMPLEX NUMBERS 99

The first way is called the ‘Dedekind cut’ construction.

The idea is the following: if I pick a point on the number line, intuitively it separates the
number line into the smaller points and the points at least as large. I can’t really make formal
sense of that idea—it’s recursive: I’m talking about the number line in order to define the
number line. But I can also talk about separating the fractions into two sets, and I know how
to work with those. Formalising this, I can say a set S of rational numbers is a Dedekind cut
if for every rational number x, either x is in S and there is an element s of S which is larger
than x, or alternatively x is larger than all elements of S. And then I can say a real number
is the same thing as a Dedekind cut. I can easily define addition: it’s not hard to check that
if S and S ′ are Dedekind cuts, then the set {s + s′ ∶ s ∈ S, s′ ∈ S ′ } is a Dedekind cut. More or
less the same idea works for multiplication (but you have to be rather careful with negative
numbers!). If q is a rational number, then {x ∈ Q ∶ x < q} is a Dedekind cut which we can easily
check behaves like the rational number q (in the same way that the rational number 21 behaves
√
like the integer 2). Now, this definition at least solves the 2 problem: {q ∈ Q ∶ q < 0 or q 2 < 2}
is a Dedekind cut, and its square is 2. This is a nice clean definition, but it only works because
we have a definition of ‘order’ < on Q.

The second route is the ‘Cauchy sequence’ construction. This is rather more complicated.
The idea is the following: if I want to specify a real number, I’ll give a sequence of rational
numbers which get closer and closer to the number I want (the formal term is ‘Cauchy √ sequence’;
it’ll be defined later), such as longer and longer parts of the decimal expansion of 2; I might
give the sequence
(1, 1.4, 1.41, 1.414, ...) .
It’s easy to add such sequences—I just add the terms:

(a1 , a2 , a3 , ...) + (b1 , b2 , b3 , ...) = (a1 + b1 , a2 + b2 , a3 + b3 , ...) .

Multiplication works similarly (this time negative numbers aren’t a problem). So far this looks
rather like the ‘decimals’ construction. But of course I might have many possible sequences of
rational numbers which get closer and closer to say 0 (or any other real number), so I should
write down an equivalence relation which says two such sequences are equivalent. And then the
real numbers are the equivalence classes of this relation.
By the end of this term, you will have studied sequences in sufficient detail to make sense of
this ‘Cauchy sequence’ construction, and prove that it really works. It’s worth trying—this is
fairly hard work, but it is also a good test of the Analysis you’ll learn.

7.2.2 Real numbers: a ‘sketchy’ introduction

For the time being, you can just think of the real numbers R as given; they are all the points
on the number line, or equivalently they are all the decimal numbers (bearing in mind that
0.4999... = 0.5000...). Let’s think for a bit about these decimals, and (again a little bit informally)
let’s write down some properties they have.
First, let’s note that if a0 ∈ N ∪ {0} and ai ≤ 9 for 1 ≤ i ≤ n, then the (finite) decimal expansion

a0 .a1 a2 . . . an

represents the rational number

a1 a2 an
a0 + + +⋯+ .
10 (10) 2 (10)n
CHAPTER 7. REAL AND COMPLEX NUMBERS 100

For example, what we mean by 1.2546 is the number

2 5 4 6
1+ + + + .
10 100 1000 10000
Every positive real number can be represented by an infinite decimal expansion

a0 .a1 a2 a3 . . . ai . . . ,

where ai ∈ N ∪ {0} and ai ≤ 9 for i ≥ 1. We allow for ai to be 0, so, in particular, it is possible that
ai = 0 for all i ≥ N where N is some fixed number: such an expansion is known as a terminating
expansion. Given such an infinite decimal expansion, we say that it represents a real number a
if, for all n ∈ N ∪ {0},
a0 .a1 a2 . . . an ≤ a ≤ a0 .a1 a2 . . . an + 1/(10)n .
This formalism allows us to see that the infinite decimal expansion 0.99999 . . . , all of whose
digits after the decimal point are 9, is in fact the same as the number 1.0000000 . . . .
For example, two infinite decimal expansions are

3.1415926535 . . .

and
0.1833333333333 . . . .
(You’ll probably recognise the first as being the number π.) Suppose, in this second decimal
expansion, that every digit is 3 after the first three (that is, ai = 3 for i ≥ 3). Then we write this
as 0.183 (or, in some texts, 0.183̇). We can extend this notation to cases in which there is a
repeating pattern of digits. For example, suppose we have

0.1123123123123 . . . ,

where the ‘123’ repeats infinitely. Then we denote this by 0.1123.

CHAPTER 7. REAL AND COMPLEX NUMBERS 101

7.2.3 Rationality and repeating patterns

You probably have heard stories of strange, obsessive mathematicians working out the expansion
of π to millions and millions of decimal places. (This has been the subject of a novel, a play, a
film, and a song!) This is relevant because the digits of π have no repeating pattern, which you
might think quite remarkable. In fact, it turns out that a real number will have an infinitely
repeating pattern in its decimal expansion (which includes the case in which the pattern is 0, so
that it includes terminating expansions) if and only if the number is rational.
Let’s look at part of this statement: if a number is rational, then its decimal expansion will
have a repeating pattern (which might be 0). Let’s look at an example.

Example 7.2. We find the decimal expansion of 4/7 by long division.

0.5714285 ⋯
7 4.0000000
3.5
.50
.49
10
7
30
28
20
14
60
56
40
35
50
So,
4/7 = 0.571428.
Notice: we must have the same remainder re-appear at some point, and then the calculation
repeats. Here’s the calculation again, with the repeating remainder highlighted.

0.5714285 ⋯
7 4.0000000
3.5
.50
.49
10
7
30
28
20
14
60
56
40
35
50
CHAPTER 7. REAL AND COMPLEX NUMBERS 102

We can formalise this very easily:

Theorem 7.3. If pq = a0 ⋅ a1 a2 a3 . . . in decimal, where p and q > 0 are integers, then there exist
some natural numbers N and k such that for each n ≥ N we have an+k = an .

The idea here is that the first few digits might not fit the ‘repeating pattern’ (as is the case
for, for example, 16 = 0.16666....) but from digit N onwards, the repeating pattern starts, and
the length of the repeating block of digits is k.
Rather than just jumping into a proof of this theorem, let’s think about how we can get to
it. This is about the right level of difficulty for a (moderately hard) exam question (or it would
be if it wasn’t in the notes..!) and so you might want to close the notes for a while and try to
solve it yourself.

We’ve seen in an example how we get to a repeating pattern. When we do long division to
work out 47 as a decimal, at some point the remainder repeats and after that point the calculation
will repeat forever. Maybe the same statement is true if we replace 47 by pq ? Then we would be
done.
So we have two things to prove. First, at some point the remainder repeats. Second, after
that point the calculation repeats forever.
Why should the remainder repeat at some point? Intuitively, this is almost obvious. The
remainder on division by q is an integer between 0 and q − 1 inclusive. There are q such integers,
so after at most q + 1 steps we surely have to repeat. That is not quite a formal proof, but ‘once
we have more steps than possible remainders we have to repeat’ should sound like a special case
of something you know. That something is the Pigeonhole Principle, so we should be using the
Pigeonhole Principle. In order to avoid talking about ‘the first remainder’, it will help to give it
a name. Let’s say that r1 is the first remainder, i.e. when we try to divide p by q, we get the
quotient a0 and remainder r1 . Then r2 is the second remainder; when we try to divide 10r1 by
q, we get the quotient a1 and remainder r2 , and so on. The Pigeonhole Principle should tell us
that there exist N and k such that rN = rN +k .
Why does the calculation repeat from this point? Again, this is almost obvious. We know
that aN is the quotient when we try to divide 10rN by q, and rN +1 is the remainder. And we
know that aN +k is the quotient when we try to divide 10rN +k by q, and rN +k+1 is the remainder.
But that is the same calculation, so aN = aN +k and rN +1 = rN +k+1 .
Well, now we know that rN +1 = rNk +1 , we can use exactly the same argument to show
aN +1 = aN +k+1 and rN +2 = rN +k+2 . And so on... in other words, this is an induction with base
case N .
Let’s write that formally.
Proof. We define two sequences recursively. We let a0 be the quotient when we try to divide p
by q, and r1 be the remainder. Then, for each integer i ≥ 1, we let ai be the quotient when we
try to divide 10ai by q, and ri+1 be the remainder.
Since each ri is an integer such that 0 ≤ ri ≤ q−1, we can define a function f from {1, 2, . . . , q+1}
to {0, 1, . . . , q − 1} by setting f (i) = ri . Since the domain is larger than the codomain, by the
Pigeonhole Principle there exist i, j ∈ {1, 2, . . . , q + 1} which are distinct such that f (i) = f (j).
Suppose that i < j, and define N = i and k = j − i. Then f (N ) = f (N + k), i.e. rN = rN +k .
We now try to prove by induction that for each n ≥ N we have the statement P (n), where
P (n) is ‘an = an+k and rn+1 = rn+k+1 ’.
The base case is n = N . We know aN is the quotient when we try to divide 10rN by q, and
rN +1 is the remainder. And we know that aN +k is the quotient when we try to divide 10rN +k by
q, and rN +k+1 is the remainder. Since 10rN = 10rN +k , this is the same calculation, so aN = aN +k
and rN +1 = rN +k+1 as required.
CHAPTER 7. REAL AND COMPLEX NUMBERS 103

Now let s ≥ N , and suppose the induction hypothesis P (s) holds. In particular, we have
rs+1 = rs+k+1 . We know as+1 is the quotient when we try to divide 10rs+1 by q, and rs+2 is the
remainder. And we know that as+k+1 is the quotient when we try to divide 10rs+k+1 by q, and
rs+k+2 is the remainder. Since 10rs+1 = 10rs+k+1 , this is the same calculation, so as+1 = as+k+1 and
rs+2 = rs+k+2 . That is P (s + 1), so we proved the induction step. By the Principle of Induction,
we have P (n) for all n ≥ N .
In particular, we have an = an+k for all n ≥ N , which proves the theorem.
We’re calling it ‘obvious’ that when we divide p by q in the above, there is only one possible
answer for the quotient and remainder. If you’re not happy about that—maybe you shouldn’t
be—you will see a proper proof that this is true in Lent Term.

Next, we think about the second part of the statement: that if the decimal expansion repeats,
then the number is rational.
Clearly, if the decimal expansion is terminating, then the number is rational. But what about
the infinite, repeating, case? We’ve given two examples above. Let’s consider these in more
detail.
Example 7.4. Consider a = 0.183. Let x = 0.003. Then 10x = 0.03 and so 10x − x = 0.03 − 0.003 =
0.03. So, 9x = 0.03 and hence x = (3/100)/9 = 1/300, so
18 1 55 11
0.183 = 0.18 + 0.003 = + = = ,
100 300 300 60
and this is the rational representation of a.
Example 7.5. Consider the number 0.1123. If x = 0.0123, then 1000x = 12.3123 and 1000x − x =
12.3. So 999x = 12.3 and hence x = 123/9990. So,
1 1 123 1122
0.1123 = +x= + = .
10 10 9990 9990
In general, if the repeating block is of length k, then an argument just like the previous two,
in which we multiply by 10k , will enable us to express the number as a rational number.
Activity 7.2. Formalise this argument.

7.2.4 Irrational numbers

A real number is irrational if it is not a rational number. So, given what we said above, an
irrational number has no infinitely repeating pattern in its decimal expansion.
What’s clear from above is that any real number can be approximated well by rational
numbers: for the rational number a0 .a1 a2 . . . an is within 1/(10)n of the real number with infinite
decimal expansion a0 .a1 a2 . . . .
√ We can, in some cases, prove that particular numbers are irrational. √ We already saw that
2 is irrational, and in general for any natural number n, either n is irrational or it is an
integer (i.e. it is never a rational number which is not an integer).
√
Activity 7.3. Prove that if n is any natural number then either n is an integer or it is
irrational.
Many other important numbers in mathematics turn out to be irrational. I’ve already
mentioned π, and there is also e (the base of the natural logarithm). It’s not easy to prove either
of these numbers is irrational.
What about π + e, or πe? We don’t know if those are rational. I think every mathematician
believes neither is rational—but we don’t know how to prove it in either case. Rather amazingly,
though, we do know that at least one of π + e and πe is irrational.
CHAPTER 7. REAL AND COMPLEX NUMBERS 104

7.2.5 ‘Density’ of the rational numbers

As we’ve seen, some important numbers in mathematics are not rational. An intuitive question
that arises is ‘how many real numbers are rational’ and this is a difficult question to answer.
There are infinitely many real numbers and infinitely many rationals, and infinitely many real
numbers are not rational. More on this next term!
For the moment, let’s make one important observation: not only are there infinitely many
rational numbers, but there are no ‘gaps’ in the rational numbers. If you accept the view of
real numbers as (possibly) infinite decimal expansions, then this is quite clear: you can get a
very good approximation to any real number by terminating its decimal expansion after a large
number of digits. (And we know that a terminating decimal expansion is a rational number.)
The following theorem makes sense of the statement that there are no ‘rational-free’ zones in
the real numbers. Precisely, between any two rational numbers, no matter how close together
they are, there is always another rational number.

Theorem 7.6. Suppose q, q ′ ∈ Q with q < q ′ . Then there is r ∈ Q with q < r < q ′ .

Proof. Consider r = (1/2)(q + q ′ ). Details are left to you!

Activity 7.4. Complete this proof.

7.3 Complex numbers

7.3.1 Introduction
Consider the two quadratic polynomials,

p(x) = x2 − 3x + 2 and q(x) = x2 + x + 1

If you sketch the graph of p(x) you will find that the graph intersects the x-axis at the two real
solutions (or roots) of the equation p(x) = 0, and that the polynomial factors into the two linear
factors,
p(x) = x2 − 3x + 2 = (x − 1)(x − 2)
Sketching the graph of q(x), you will find that it does not intersect the x-axis. The equation
q(x) = 0 has no solution in the real numbers, and it cannot be factorised (or factored) over the
reals. Such a polynomial is said to be irreducible over the reals. In order to solve this equation,
we need to define the complex numbers.
If you met the complex numbers in school, then probably you were told to accept ‘there is a
symbol i which means the square root of −1’ and you did arithmetic with it. This isn’t a very
satisfactory way of doing things: why can we assume there is such a symbol? We could equally
well invent a symbol (say E) to be the result of trying to divide 1 by 0 and do arithmetic with
it—and if you do, you’ll find you can ‘prove’ 1 = 2. (Try it!)
What we will do is instead to write down a new number system, explain how to do arithmetic,
and then show that we can find a ‘square root of −1’ in this new system.
CHAPTER 7. REAL AND COMPLEX NUMBERS 105

7.3.2 Complex numbers: a formal approach

To start with, let’s formally construct the complex numbers from the real numbers. Recall that
in the last chapter, we constructed the rational numbers from the integers, by explaining how to
view rational numbers as equivalence classes of pairs of integers. We did not explain how to
do addition and multiplication: this is difficult (so we do it in Lent Term) precisely because a
fraction (an equivalence class) has many different representatives (pairs of integers) and we need
to see why we get the same answer for the addition of two fractions whatever representatives we
pick.
The complex numbers turn out to be easier.
We define the set C of complex numbers to be the set of all ordered pairs (x, y) of real
numbers, with addition and multiplication operations defined as follows:

(a, b) + (c, d) = (a + c, b + d) , (a, b) × (c, d) = (ac − bd, ad + bc) .

You should check that these definitions really work, that is, that (for example) the multipli-
cation is commutative, and that the distributive law holds. More generally, you should check
that C satisfies (F1)–(F6), i.e. it is a field: we can do all the algebra we’re used to. (What C
doesn’t have is an order: there isn’t any way of defining an order < on C which plays nicely with
addition and multiplication in the way that the order plays nicely in N, Z, Q or R.)
You can also check that the complex numbers of the form (x, 0) behave like the real numbers,
in other words that (x, 0)+(y, 0) = (x+y, 0), and (x, 0)×(y, 0) = (xy, 0), which is what you expect
for adding and multiplying real numbers. Finally, let’s remember why we began this: we wanted
to be able to solve the equation x2 + 1 = 0. Well, that means we want a complex number (a, b)
such that (a, b) × (a, b) + (1, 0) = (0, 0). And we can find such a number: (0, 1) × (0, 1) = (−1, 0),
so we are done.
Let’s return briefly to the 10 bad example from the last section. Suppose you try to construct
a new number system—maybe by taking pairs or triples or whatever of numbers, maybe with
some equivalence relation to say when two pairs are ‘equivalent’ (as we did to construct the
rationals). To do arithmetic with your new number system, you need to explain how to add and
to multiply, and (if you have some equivalence relation involved) you need to show that the
addition and multiplication you wrote down are well-defined. And you would like that there is
something like ‘subtraction’ and ‘division’ that are inverse operations, and you would like it to
be true that addition distributes over multiplication, and so on.
There are in fact lots of things you might come up with that make sense—not just the rational,
real and complex numbers—these other things are also fields and they are very important in
mathematics (and some of them turn out to be very important in modern technology). What
you will not find is a field that contains a solution to the equation 0 × x = 1, in the way that the
complex numbers we just defined contain a solution to the equation x2 + 1 = 0. This is why √ we
cannot invent a symbol E = 10 and do arithmetic with it, but we can invent a symbol i = −1.

7.3.3 Complex numbers: a more usual approach

Rather than the ordered pairs approach outlined above, it is more common to define the complex
numbers as follows. We begin by defining the imaginary number i which has the property
that i2 = −1. The term ‘imaginary’ is historical, and not an indication that this is a figment of
someone’s imagination—but historically the reason for the name is that some mathematicians
didn’t believe the complex numbers make sense: ‘imaginary’ is a term of Descartes, and he
meant it as an attack on the idea.
This symbol i is simply a nicer way of writing the pair (0, 1) of real numbers; it’s easier to
write on the board in calculations (in the same way that it’s easier to write ab for the rational
CHAPTER 7. REAL AND COMPLEX NUMBERS 106

rather than the equivalence class [(a, b)]R of the relation R we defined in Section 6.4.1). We
can then say what we mean by the complex numbers.
Definition 7.7. A complex number is a number of the form z = a + ib, where a and b are real
numbers, and i2 = −1. The set of all such numbers is

C = {a + ib ∶ a, b ∈ R} .

If z = a + ib is a complex number, then the real number a is known as the real part of z,
denoted Re(z), and the real number b is the imaginary part of z, denoted Im(z). Note that
Im(z) is a real number.
If b = 0, then z is a real number, so R ⊆ C. If a = 0, then z is said to be purely imaginary.
The quadratic polynomial q(z) = x2 + x + 1 can be factorised over the complex numbers,
because the equation q(z) = 0 has two complex solutions. Solving in the usual way, we have
√
−1 ± −3
x= .
2
√ √ √ √ √
We write, −3 = (−1)3 = −1 3 = i 3, so that the solutions are
√ √
1 3 1 3
w =− +i and w =− −i .
2 2 2 2
Notice the form of these two solutions. They are what is called a conjugate pair. We have the
following definition.
Definition 7.8. If z = a+ib is a complex number, then the complex conjugate of z is the complex
number z = a − ib.
We can see by the application of the quadratic formula, that the roots of an irreducible
quadratic polynomial with real coefficients will always be a conjugate pair of complex numbers.

Addition, multiplication, division

Addition and multiplication of complex numbers are defined as for polynomials in i using i2 = −1.
Example 7.9. If z = (1 + i) and w = (4 − 2i) then

z + w = (1 + i) + (4 − 2i) = (1 + 4) + i(1 − 2) = 5 − i

and
zw = (1 + i)(4 − 2i) = 4 + 4i − 2i − 2i2 = 6 + 2i
You should check that this is really exactly the same as the definitions we gave when we
formally constructed the complex numbers: the only difference is the way we’re writing complex
numbers.
If z ∈ C, then zz is a real number:

zz = (a + ib)(a − ib) = a2 + b2 .

Activity 7.5. Carry out the multiplication to verify this: let z = a + ib and calculate zz.
z zw
Division of complex numbers is then defined by = since ww is real.
w ww
Example 7.10.
1+i (1 + i)(4 + 2i) 2 + 6i 1 3
= = = + i
4 − 2i (4 − 2i)(4 + 2i) 16 + 4 10 10
CHAPTER 7. REAL AND COMPLEX NUMBERS 107

Properties of the complex conjugate

A complex number is real if and only if z = z. Indeed, if z = a + ib, then z = z if and only if b = 0.
The complex conjugate of a complex number satisfies the following properties:
• z + z = 2 Re(z) is real
• z − z = 2i Im(z) is purely imaginary
• z=z
• z+w =z+w
• zw = z w
z z
• ( )=
w w
Activity 7.6. Let z = a + ib, w = c + id and verify all of the above properties.

7.3.4 Roots of polynomials

Are we really done with construction? We invented the symbol i because we wanted to have a
solution to x2 + 1 = 0. But I also want a solution to x6 + 10x2 + 17 = 0. Do I need a new symbol
for that? It turns out the answer is No.
The Fundamental Theorem of Algebra asserts that a polynomial of degree n with complex
coefficients has n complex roots (not necessarily distinct), and can therefore be factorised into n
linear factors. Explicitly, any equation
an z n + an−1 z n−1 + ⋯ + a1 z + a0 = 0
where ai ∈ C has n solutions z ∈ C. Contrast this with the difficulty of solving polynomial
equations in R. So, the introduction of i enables us to solve all polynomial equations: there’s no
need to introduce anything else. A fancy way of saying this is: ‘The field of complex numbers is
algebraically closed.’
If the coefficients of the polynomial are restricted to real numbers, the polynomial can be
factorised into a product of linear and irreducible quadratic factors over R and into a product
of linear factors over C. The proof of the Fundamental Theorem of Algebra is beyond the scope
of this course (and this time not because it’s long and boring, but because it is genuinely quite
hard). However, we note the following useful result.
Theorem 7.11. Complex roots of polynomials with real coefficients appear in conjugate pairs.
Proof. Let P (x) = a0 + a1 x + ⋯ + an xn , ai ∈ R, be a polynomial of degree n. We shall show that
if z is a root of P (x), then so is z.
Let z be a complex number such that P (z) = 0, then
a0 + a1 z + +a2 z 2 ⋯ + an z n = 0
Conjugating both sides of this equation,
a0 + a1 z + a2 z 2 + ⋯ + an z n = 0 = 0
Since 0 is a real number, it is equal to its complex conjugate. We now use the properties of the
complex conjugate: that the complex conjugate of the sum is the sum of the conjugates, and
the same is true for the product of complex numbers. We have
a0 + a1 z + a2 z 2 + ⋯ + an z n = 0,
CHAPTER 7. REAL AND COMPLEX NUMBERS 108

and
a0 + a1 z + a2 z 2 + ⋯ + an z n = 0.
Since the coefficients ai are real numbers, this becomes

a0 + a1 z + a2 z 2 + ⋯ + an z n = 0.

That is, P (z) = 0, so the number z is also a root of P (x).

Example 7.12. Let us consider the polynomial

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

√
1 3
If w = − + i , then
2 2
x3 − 2x2 − 2x − 3 = (x − 3)(x − w)(x − w)

Activity 7.7. Multiply out the last two factors above to check that their product is the irreducible
quadratic x2 + x + 1.

7.3.5 The complex plane

The following theorem shows that a complex number is uniquely determined by its real and
imaginary parts.

Theorem 7.13. Two complex numbers are equal if and only if their real and imaginary parts
are equal.

There are two ways to prove this. We can do it directly, using the fact that the complex
numbers are a field:
Proof. Two complex numbers with the same real parts and the same imaginary parts are clearly
the same complex number, so we only need to prove this statement in one direction. Let z = a + ib
and w = c + id. If z = w, we will show that their real and imaginary parts are equal. We have
a+ib = c+id, therefore a−c = i(d−b). Squaring both sides, we obtain (a−c)2 = i2 (d−b)2 = −(d−b)2 .
But a − c and (d − b) are real numbers, so their squares are non-negative. The only way this
equality can hold is for a − c = d − b = 0. That is, a = c and b = d.
The other, much shorter (by now!) way to prove this is simply to observe that the complex
numbers are the same as pairs of real numbers (with addition and multiplication as we defined
them when we formally constructed the complex numbers) and pairs of real numbers are by
definition equal if and only if both parts—which are precisely the real and imaginary parts—are
equal.
As a result of this theorem, we can think of the complex numbers geometrically, as points in
a plane. For, we can associate the vector (a, b)T uniquely to each complex number z = a + ib,
and all the properties of a two-dimensional real vector space apply. A complex number z = a + ib
is represented as a point (a, b) in the complex plane; we draw two axes, a horizontal axis to
represent the real parts of complex numbers, and a vertical axis to represent the imaginary parts
of complex numbers. Points on the horizontal axis represent real numbers, and points on the
vertical axis represent purely imaginary numbers.
CHAPTER 7. REAL AND COMPLEX NUMBERS 109

(a, b)

7
z = a + ib 

i 


θ
(0, 0) 1

Complex plane or Argand diagram

√
Activity 7.8. Plot z = 2 + 2i and w = 1 − i 3 in the complex plane.

7.3.6 Polar form of z

If the complex number z = a + ib is plotted as a point (a, b) in the complex plane, then we can
determine the polar coordinates of this point. We have

a = r cos θ, b = r sin θ
√
where r = a2 + b2 is the length of the line joining the origin to the point (a, b) and θ is the
angle measured anticlockwise from the real (horizontal) axis to the line joining the origin to the
point (a, b). Then we can write z = a + ib = r cos θ + i r sin θ.
Definition 7.14. The polar form of the complex number z is

z = r(cos θ + i sin θ).

√
The length r = a2 + b2 is called the modulus of z, denoted ∣z∣, and the angle θ is called the
argument of z.
Note the following properties:
• z and z are reflections in the real axis. If θ is the argument of z, then −θ is the argument of
z.

• ∣z∣2 = zz.

• θ and θ + 2nπ give the same complex number.

We define the principal argument of z to be the argument in the range, −π < θ ≤ π.
√
Activity 7.9. Express z = 2 + 2i, w = 1 − i 3 in polar form.
Describe the following sets of z: (a) ∣z∣ = 3, (b) argument of z is π4 .
Multiplication and division using polar coordinates gives

zw = r(cos θ + i sin θ) ⋅ ρ(cos φ + i sin φ)

= rρ(cos(θ + φ) + i sin(θ + φ))

z r
= ( cos(θ − φ) + i sin(θ − φ))
w ρ
Activity 7.10. Show these by performing the multiplication and the division as defined earlier,
and by using the facts that cos(θ+φ) = cos θ cos φ−sin θ sin φ and sin(θ+φ) = sin θ cos φ+cos θ sin φ.
CHAPTER 7. REAL AND COMPLEX NUMBERS 110

DeMoivre’s Theorem
We can consider explictly a special case of the multiplication result above, in which w = z. If we
apply the multiplication to z 2 = zz, we have

z2 = zz
= (r(cos θ + i sin θ))(r(cos θ + i sin θ))
= r2 (cos2 θ + i2 sin2 θ + 2i sin θ cos θ)
= r2 (cos2 θ − sin2 θ + 2i sin θ cos θ)
= r2 (cos 2θ + i sin 2θ).

Here we have used the double angle formulae for cos 2θ and sin 2θ.
Applying the product rule n times, where n is a positive integer, we obtain DeMoivre’s
Formula

Theorem 7.15.
(cos θ + i sin θ)n = cos nθ + i sin nθ

Proof.
n
zn = z⋯z = (r(cos θ + i sin θ))
±
n times

= rn ( cos(θ + ⋯ + θ) + i sin(θ + ⋯ + θ))

´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶ ´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶
n times n times

7.3.7 Exponential form of z

Functions of complex numbers can be defined by the power series (Taylor expansions) of the
functions:
z2 z3
ez = 1 + z + + + ⋯ z∈C
2! 3!
z3 z5 z2 z4
sin z = z − + −⋯ cos z = 1 − + −⋯
3! 5! 2! 4!
If we use the expansion for ez to expand eiθ , and then factor out the real and imaginary
parts, we find:

(iθ)2 (iθ)3 (iθ)4 (iθ)5

eiθ = 1 + (iθ) + + + + +⋯
2! 3! 4! 5!
θ2 θ3 θ4 θ5
= 1 + iθ − − i + + i − ⋯
2! 3! 4! 5!
θ 2 θ 4 θ3 θ5
= (1 − + − ⋯) + i (θ − + − ⋯)
2! 4! 3 5!
From which we conclude:
Euler’s Formula: eiθ = cos θ + i sin θ
If you’re being careful, you might notice something a bit strange here—what exactly do
I mean by these funny infinite sums? and why am I allowed to rearrange the terms in them?
CHAPTER 7. REAL AND COMPLEX NUMBERS 111

Sure, I know addition is commutative, but that will only let me change places of finitely many
terms in the sum (which I don’t quite understand anyway), and I still have infinitely many more
things which I need to change places. The answer to that objection is: we’ll explain properly
some of it later this term, and some next year in MA203 Real Analysis. For now, take it on
faith that it does actually make sense.
Definition 7.16. The exponential form of a complex number z = a + ib is
z = reiθ
where r = ∣z∣ is the modulus of z and θ is the argument of z.
In particular, the following equality is of note because it combines the numbers e, π and i
in a single expression: eiπ = −1.
If z = reiθ , then its complex conjugate is given by z = re−iθ . This is because, if z = reiθ =
r(cos θ + i sin θ), then
z = r(cos θ − i sin θ) = r(cos(−θ) + i sin(−θ)) = re−iθ .

We can use either the exponential form, z = reiθ , or the standard form, z = a + ib, according
to the application or computation we are doing. For example, addition is simplest in the form
z = a + ib, but multiplication and division are simpler in exponential form. To change a complex
number between reiθ and a + ib, use Euler’s formula and the complex plane (polar form).
Example 7.17. √
2π
3 + i sin 3 = − 2 + i 2 .
ei 3 = cos 2π 2π 1 3
√ √ √ √
e2+i 3 = e2 ei 3 = e2 cos 3 + ie2 sin 3.
Activity 7.11. Write each of the following complex numbers in the form a + ib:
π 3π 3π 11π
ei 2 ei 2 ei 4 ei 3 e1+i e−1
√ π √ π
Example 7.18. Let z = 2 + 2i = 2 2 ei 4 and w = 1 − i 3 = 2e−i 3 , then
√ π
w6 = (1 − i 3)6 = (2e−i 3 )6 = 26 e−i2π = 64
√ π π √ π
zw = (2 2ei 4 )(2e−i 3 ) = 4 2e−i 12
and
z √ i 7π
= 2e 12 .
w
Notice that in the above example we are using certain properties of the complex exponential
function, that if z, w ∈ C,
ez+w = ez ew and (ez )n = enz for n ∈ Z.
This last property is easily generalised to include the negative integers.
Example 7.19. Solve the equation z 6 = −1 to find the 6th roots of −1.
Write z 6 = (reiθ )6 = r6 ei6θ , −1 = eiπ = ei(π+2nπ)
Equating these two expressions, and using the fact that r is a real positive number, we have
π 2nπ
r=1 6θ = π + 2nπ, + θ=
6 6
This will give the six complex roots by taking n = 0, 1, 2, 3, 4, 5.
Activity 7.12. Show this. Write down the six roots of −1 and show that any one raised to the
power 6 is equal to −1. Show that n = 6 gives the same root as n = 0.
Use this to factor the polynomial x6 + 1 into linear factors over the complex numbers and
into irreducible quadratics over the real numbers.
CHAPTER 7. REAL AND COMPLEX NUMBERS 112

7.4 Sample exercises

√
Exercise 7.1. Prove that 5 is irrational.
1 + 2i
Exercise 7.2. Express the complex number in the form a + bi.
4 − 5i
Exercise 7.3. Solve the equation x2 − 2ix + 3 = 0.
Exercise 7.4. Write each of the following complex numbers in the form a + ib:
π 3π 3π 11π
ei 2 ei 2 ei 4 ei 3 e1+i e−1 .
√ √
Exercise 7.5. Express 1 + 3i in exponential form. Hence find (1 + 3i)30 .

7.5 Comments on selected activities

√
Comment on Activity 7.3. The obvious thing to do is to try mimicking the proof that 2 is
irrational. So let’s try. Suppose for a contradiction that there are integers a and b such that
2
( ab ) = n. As before, we can assume n does not divide both a and b. We get

a2 = nb2

and it follows that a2 is divisible by n. But it doesn’t follow that a is divisible by n, in general.
For example 62 = 36 is divisible by 18, but 6 is certainly not divisible by 18). In order to get
further, it helps to think about the prime factorisation of n—this is something we will meet in
MA103 next term.
Comment on Activity 7.7. We have

(x − w)(x − w) = x2 − (w + w)x + ww.

Now, w + w = 2 Re(w) = 2(− 12 ) and ww = 41 + 34 so the product of the last two factors is x2 + x + 1.

2i ● z = 2 + 2i

(0, 0) 1 2
−i
√
● w =1−i 3
−2i
Comment on Activity 7.8.
Comment on Activity 7.9. Draw√ the line from the origin
√ to theπpoint z inπ the diagram above.
Do the same for w. For z, ∣z∣ = 2 2 and θ = 4 , so z = 2 2( cos( 4 ) + i sin( 4 )). The modulus of
π

w is ∣w∣ = 2 and the argument is − π3 , so that

π π π π
w = 2( cos(− ) + i sin(− )) = 2( cos( ) − i sin( )).
3 3 3 3
The set (a) ∣z∣ = 3, is the circle of radius 3 centered at the origin. The set (b), argument of z
π
is4,is the half line from the origin through the point (1,1).
CHAPTER 7. REAL AND COMPLEX NUMBERS 113

Comment on Activity 7.12. The roots are:

π 3π 5π
z1 = 1 ⋅ ei 6 , z2 = 1 ⋅ ei 6 , z 3 = 1 ⋅ ei 6 ,
7π 9π 11π
z4 = 1 ⋅ ei 6 , z5 = 1 ⋅ ei 6 , z 6 = 1 ⋅ ei 6 .
i 13π i π6
These roots are in conjugate pairs, and e 6 =e :
5π π π
z4 = z 3 = e−i 6 , z5 = z 2 = e−i 2 , z6 = z 1 = e−i 6 .

The polynomial factors as

x6 + 1 = (x − z1 )(x − z 1 )(x − z2 )(x − z 2 )(x − z3 )(x − z 3 ),

√
Using the a + ib form of each complex number, for example, z1 = 23 + i 21 , you can carry out the
multiplication of the linear terms pairwise (conjugate pairs) to obtain x6 + 1 as a product of
irreducible quadratics with real coefficients:
√ √
x6 + 1 = (x2 − 3 x + 1)(x2 + 3 x + 1)(x2 + 1).

7.6 Solutions to exercises

√ √
Solution to Exercise 7.1. Suppose we have 5 = m/n where m, n ∈ Z. Since 5 > 0, we may
assume that m, n > 0. (Otherwise, both are negative, and we can multiply each by −1.) We
can also suppose that m, n have greatest common divisor 1. (For, we can cancel any common
factors.) Then (m/n)2 = 5 means that m2 = 5n2 . So 5 ÷ m2 . Now m can, by the Fundamental
Theorem of Arithmetic, be written as a product of primes m = p1 p2 . . . pk . Then m2 = p21 p22 . . . p2k .
If no pi is 5, then 5 does not appear as a factor in m2 and so 5 does not divide m2 . So some pi
is equal to 5. So 5 ÷ m. Now, this means that m = 5r for some r ∈ N and hence m2 = (5r)2 = 25r2
and so 25r2 = 5n2 . Then, n2 = 5r2 , so 5 ÷ n2 . Arguing
√ as before, 5 ÷ n. So 5 is a common factor if
m and n, which contradicts gcd(m, n) = 1. Hence 5 is not rational.
Solution to Exercise 7.2. We have
1 + 2i 1 + 2i 4 + 5i
=
4 − 5i 4 − 5i 4 + 5i
(1 + 2i)(4 + 5i)
=
(4 − 5i)(4 + 5i)
4 + 8i + 5i + 10i2
=
16 − 25i2
−6 + 13i
=
41
6 13
= − + i.
41 41
You can check that this is the correct answer by calculating the product
6 13
(− + i) (4 − 5i)
41 41
and observing that the answer is 1 + 2i.
Solution to Exercise 7.3. To solve the equation x2 − 2ix + 3 = 0, we could use the formula for
the solutions of a quadratic equation. Or we could note that the equation is equivalent to
(x − i)2 = −4, so the solutions are given by x − i = 2i and x − i = −2i, so they are x = 3i and x = −i.
CHAPTER 7. REAL AND COMPLEX NUMBERS 114

Solution to Exercise 7.4. We have

1 1
eiπ/2 = i, ei3π/2 = −i, ei3π/4 = − √ + i √ ,
2 2
√
1 3
e =e
i(11π/3)
= −i
−i(π/3)
, e1+i = e1 ei = e cos(1) + i e sin(1),
2 2
e = e + 0i is real, so already in the form a + ib.
−1 −1
√
Solution to Exercise 7.5. To express z = 1 + 3i in exponential form, we first note that
√
√ 1 3
1 + 3i = 2 ( + i)
2 2

and this is r(cos θ + i sin θ) when r = 2, θ = π/3. So z = 2eπi/3 . Then,

√ 30 30
(1 + 3i) = z 30 = (2eπi/3 ) = 230 e30πi/3 = 230 e10πi = 230 .
 Analysis: the real numbers
8
8.1 What is analysis?
Analysis is the theory behind real numbers, sequences, and functions. The word ‘theory’ is
important. You might, for example, have a good idea of what we mean by a ‘limit’ of a convergent
sequence of numbers, or the notion of a ‘continuous’ function, but in this part of the course we
aim to formalise such notions.

8.1.1 Analysis and calculus: the history

Historically, mathematicians did not formalise the concept of a ‘function’ as we have done it
until the 1800s. Prior to this, mathematicians generally thought of a function as ‘something
defined by a formula’. This is what Isaac Newton and Gottfried Leibniz were thinking of when
they independently developed the ‘infinitesimal calculus’. What they did starts with what you
learned at school: how to integrate and differentiate, and what those two operations mean.
The roots of the calculus are much older—Pierre de Fermat had some systematic ideas in this
direction, and in ancient times Eudoxus and Archimedes in Greece, and independently Liu Hui
in China, developed geometric methods which are rather close to integration.
As Newton and Leibniz formulated it, the calculus is a collection of rules for calculating
things—this collection of rules starts with what you learned at school. You can use the calculus
to find a formula for, to take an example, the volume of a sphere of radius 1. Here is one way to
do that—it’s an argument due to Archimedes.

115
CHAPTER 8. ANALYSIS: THE REAL NUMBERS 116

h h

Figure 8.1: A hemisphere, cone and cylinder, with a slice through at height h

Take a sphere of radius 1, and cut in in half through the centre to obtain a hemisphere.
Place the hemisphere with its flat base on the table, and put next to it a cone of height
1, with its base a circle of radius 1, standing vertically on its point. And next to that, put a
cylinder of radius and height 1.
Now, the critical observation: if you take a slice through this picture at any height h between
0 and 1, you slice a circle out of each of the hemisphere, cone and cyclinder. The area of the circle
sliced from the hemisphere, plus the area of the circle sliced from the cone, is equal to the area
of the circle sliced from the cylinder. This is easy to check using the Pythagoras theorem and
the formula for the area
√ of a circle (which Archimedes knew). The slice through the hemisphere
is a circle of radius 1 − h2 for an area π(1 − h2 ); the slice through the cone is again a circle,
this time of radius h for an area πh2 ; the slice through the cylinder is a circle of radius 1 for an
area π.
So, (in modern language) if we integrate we see that the volume of the hemisphere plus the
volume of the cone is equal to the volume of the cylinder. Archimedes, of course, did not say
‘integrate’, but he had a similar conception.
Putting it another way, the volume of the hemisphere is the volume of the cylinder minus
the volume of the cone: π − 31 π = 23 π. (Archimedes knew the volume of a cone, too.)
So the volume of the sphere is 34 π.
Why did I go through this argument? Well, because you can try to do something similar
to find the surface area of the sphere. Here is one way, which I’ll give in terms of modern
integration.
Look again at the hemisphere √ lying on the table, and take a slice through it at√height h.
That gives us a circle of radius 1 − h2 , as before, and the length of that circle is 2π 1 − h2 .
Integrating, we find the area of the curved bit of the hemisphere is
1 √
∫h=0 2π 1 − h dh = 2 π .
2 1 2

So the surface area of the whole sphere is π 2 .

But you maybe know a formula for the surface area of a sphere—and it doesn’t agree with
the calculation we just did! The correct answer is slightly larger; it is 4π. Archimedes knew this
(he had a different method), and he also knew why the above method does not work.

Activity 8.1. Why did we underestimate the surface area—what are we missing?

Liu Hui developed quite a number of methods of this nature in China, and (importantly) he
gave intuitive explanations of why certain methods work and others do not; following his ideas,
CHAPTER 8. ANALYSIS: THE REAL NUMBERS 117

one can solve rather a lot of geometrical problems accurately. But this is an unpleasant situation
for a mathematician to find themselves in, with methods which might or might not work.
While mathematicians were using calculus-like methods to find answers to geometrical
problems, these intuitive explanations were considered good enough. But Newton and Leibniz,
and the mathematicians who came later, were soon using the calculus as you know it to find
answers to problems which aren’t obviously geometrical. Newton himself did not really trust
the calculus (or at least, he didn’t expect others to trust it); while he used it to solve problems,
once he knew the answer he generally looked for a geometrical proof before publishing. But
doing this is hard work (and sometimes you will get completely stuck and be unable to do it at
all), and if you trust the calculus, it is a waste of time.
But you cannot always trust the calculus, as we have seen!
The solution to this—developed by Cauchy and Weierstrass in the 1800s—is to formalise all
of this properly, remove the appeals to intuition, and provide some clear rules: if you do this
it will always work, and we can prove it. This is where analysis begins. If that was all there
is to it, it would be a small part of mathematics: but in fact, analysis provides us with many
surprising tools, going far beyond the calculus of Newton and Leibniz, and it is one of the two
major branches of pure mathematics, along with Algebra (which MA103 will introduce you to
in Lent Term).
We will not get that far in this course. But we can at least lay the foundations.

8.1.2 Greek letters, conventions and inexact calculation

In mathematics, we often use particular letters for particular things. When you see the letter n
in a proof, you probably expect that it is going to be an integer (we always used it that way
so far). The letter x, on the other hand, you maybe expect to be some real number, and f is
usually a function.
There isn’t any rule that says we have to do this; it’s perfectly fine to talk about n = 2.75, or
the function x ∶ R → R defined by x(f ) = f 2 . But it does help to have these conventions—you
probably find it hard even to read the function definition!
In analysis, there are two new letters we use from the Greek alphabet: epsilon ε and delta δ.
Conventionally, these are always positive reals and you should think of them as tiny. In the next
three chapters, we will always insist that ε and δ are positive real numbers. We will not always
insist that they are tiny, but the ‘interesting case’ will always be when they are very small.
In particular, you should think of ε2 or ε3 as being much smaller than ε. Why is this relevant?
In this part of the course, we often do not care about exact answers. This is something you have
not seen in maths before: in the past, whenever you were supposed to work something out, you
were expected to give an exact answer.
In this bit of the course, you might see a question like ‘Find a real number which is bigger
than f (x) = sin(x) − (x − 2)2 for all x ∈ R’. Before turning the page, think about how you can
solve this problem.
CHAPTER 8. ANALYSIS: THE REAL NUMBERS 118

You probably started thinking about the following: differentiate, set the derivative to zero
and solve...
What you are doing, if you do that, is too much work. You’re trying to find the maximum
value of f (x) on R. It turns out to be a pain to do this, and it’s not what the question asked
for. If you do it (correctly!) then you do have a right answer, but you have also spent a lot of
time getting it.
It’s much easier to observe that −1 ≤ sin x ≤ 1 is true for all x ∈ R (you know this from school,
I hope!), and −(x − 2)2 is never positive. So f (x) ≤ 1 − 0 = 1. So we can answer ‘1 is such a
number’ and write this very short proof, and we are done. The number 2022, by the way, would
have done just as well; there are no points for some ‘best possible’ answer.
For another example, if you are supposed to ‘choose ε > 0 such that 17ε + 283ε2 < 0.1’ then
you don’t need to start trying to solve a quadratic equation in ε (which is a pain). You can say:
I will choose some ε ≤ 1, so that ε2 ≤ ε. Then I get

17ε + 283ε2 ≤ 17ε + 283ε = 300ε

and if I choose ε = 1
6 000 (which is indeed ≤ 1) then I am done.
A final comment: some of you most likely heard something about funny names or symbols
like ‘infinitesimals’ or ‘dx’ before, and maybe have some idea that ε is ‘really’ the ‘dx’. This is
not true; ε is a real number and (whatever those funny things are) they are not real numbers.

8.2 The real numbers

The rational number system is inadequate for many purposes. For instance, there is no rational
number q such that q 2 = 2, and the set

S = {q ∈ Q ∣ q 2 ≤ 2}

does not have a largest element in Q. So we see that the rational number system has “gaps”.
The real number system R includes numbers to fill all of these gaps. Thus the set

T = {x ∈ R ∣ x2 ≤ 2}

has a largest element. This is a consequence of a very important property of the real numbers,
called the least upper bound property. Before we state this property of R, we need a few
definitions.
CHAPTER 8. ANALYSIS: THE REAL NUMBERS 119

8.2.1 Bounded sets; least upper bound

Definition 8.1. Let S be a subset of R.
1. An element u ∈ R is said to be an upper bound of S if, for all x ∈ S, x ≤ u. If S has an upper
bound, then we also say that S is bounded above.

2. An element l ∈ R is said to be a lower bound of S if, for all x ∈ S, l ≤ x. If S has a lower

bound, then we also say that S is bounded below.

3. The set S is said to be bounded if it is bounded above and bounded below.

Example 8.2.
(i) The set S = {x ∈ R ∣ 0 ≤ x < 1} is bounded. Any real number y satisfying y ≥ 1 (for instance
1, π, or 100) is an upper bound of S, and any real number z satisfying z ≤ 0 (for instance
0, or −1) is a lower bound of S.

(ii) The set S = {n ∣ n ∈ N} is bounded below; indeed, any real number x ≤ 1 serves as a lower
bound. It has no upper bound—we shall give a formal proof of this shortly—and so it is
not bounded above, and therefore also not bounded.

(iii) The set S = {(−1)n ∣ n ∈ N} is bounded. Note that S = {−1, 1}. It is bounded above by 1
and bounded below by −1.
More generally, any finite set S is bounded.

(iv) The set S = { n1 ∣ n ∈ N} is bounded. Any real number x satisfying 1 ≤ x is an upper bound,
and 0 is a lower bound.

(v) The sets Z and R are neither bounded above nor bounded below.

(vi) The set T = {x ∈ R ∣ x2 ≤ 2} is bounded. Any real number x satisfying x2 ≤ 2 also satisfies
x2 < 4. Therefore, x2 − 4 < 0, that is, (x − 2)(x + 2) < 0. It follows that

either x − 2 < 0 and x + 2 > 0,

or x − 2 > 0 and x + 2 < 0.

The second case is impossible. Thus the first case has to hold, that is −2 < x < 2. It follows
that T is bounded above by 2 and bounded below by −2.

(vii) The set ∅ is bounded. For instance, 1 is an upper bound for ∅, since the condition “for
every element x of ∅, x ≤ 1” is satisfied: there is certainly no element x of ∅ that doesn’t
satisfy x ≤ 1. Indeed, every real number is an upper bound for ∅, and similarly every real
number is a lower bound for ∅.
We now introduce the notions of a least upper bound (also called supremum) and a greatest
lower bound (also called infimum) of a subset S of R.
Definition 8.3. Let S be a subset of R.
1. An element u∗ ∈ R is said to be a least upper bound of S (or a supremum of S) if

(a) u∗ is an upper bound of S, and

(b) for each upper bound u of S, it holds that u∗ ≤ u.

2. An element l∗ ∈ R is said to be a greatest lower bound of S (or an infimum of S) if

CHAPTER 8. ANALYSIS: THE REAL NUMBERS 120

(a) l∗ is a lower bound of S, and

(b) for each lower bound l of S, it holds that l ≤ l∗ .

Example 8.4. Let S = {x ∈ R ∣ x < 1}. Show that the supremum of S is 1.

Proof. Clearly 1 is an upper bound of S.

Now we show that if u is another upper bound, then 1 ≤ u. Suppose not, that is, suppose
that there is an upper bound u of S with u < 1. Then we have
u+1
u< < 1,
2
where both inequalities follow using u < 1. From the second inequality, it follows that the number
u+1
2 belongs to S. The first inequality above then shows that u is not an upper bound for S, a
contradiction. Hence 1 is a supremum.
Next we show that 1 is the only supremum. Indeed, if u∗ is another supremum, then in
particular u∗ is also an upper bound, and the above argument shows that 1 ≤ u∗ . But 1 < u∗ is
not possible, since 1 is an upper bound for S, and as u∗ is a supremum, u∗ must be less than or
equal to the upper bound 1. So it follows that u∗ = 1.
In the above example, there was a unique supremum of the set S. In fact, this is always the
case and we have the following result.

Theorem 8.5. If a least upper bound of a subset S of R exists, then it is unique.

Proof. Suppose that S has a least upper bound u∗ . Suppose that u′∗ is also a least upper bound
of S. Then in particular u∗ and u′∗ are also upper bounds of S. Now since u∗ is a least upper
bound of S and u′∗ is an upper bound of S, it follows that

u∗ ≤ u′∗ . (8.1)

Furthermore, since u′∗ is a least upper bound of S and u∗ is an upper bound of S, it follows that

u′∗ ≤ u∗ . (8.2)

From (8.1) and (8.2) we obtain u∗ = u′∗ .

Thus it makes sense to talk about the least upper bound, or the supremum, of a set. Similarly,
the infimum of a set S (if it exists) is also unique.

Definition 8.6.

1. The least upper bound of a set S (if it exists) is denoted by sup S.

2. The greatest lower bound of a set S (if it exists) is denoted by inf S.

When the supremum and the infimum of a set belong to the set, we give them the following
familiar special names:

Definition 8.7.

1. If sup S ∈ S, then sup S is called a maximum of S, denoted by max S.

2. If inf S ∈ S, then inf S is called a minimum of S, denoted by min S.

CHAPTER 8. ANALYSIS: THE REAL NUMBERS 121

Example 8.8.

(i) If S = {x ∈ R ∣ 0 ≤ x < 1}, then sup S = 1 ∈/ S and so max S does not exist. But inf S = 0 ∈ S,
and so min S = 0.

(ii) If S = N, then sup S does not exist, inf S = 1, max S does not exist, and min S = 1.

(iii) If S = {(−1)n ∣ n ∈ N}, then sup S = 1, inf S = −1, max S = 1, min S = −1.

(iv) If S = { n1 ∣ n ∈ N}, then sup S = 1 and max S = 1. We show below (after Theorem 8.12)
that inf S = 0. So min S does not exist.

(v) For the sets Z and R, none of sup, inf, max, min exist.

(vi) For the set ∅, none of sup, inf, max, min exist.

8.2.2 The least upper bound property

In the above examples, we can see that if S is non-empty and bounded above, then its supremum
exists. In fact this is a fundamental property of the real numbers, called the least upper bound
property of the real numbers, which we now state:
The Least Upper Bound Property.
If S is a subset of R that is non-empty and bounded above, then S has a least upper bound.
In other words, for a subset S ⊆ R, if

1. S ≠ ∅, and

2. S has an upper bound,

then sup S exists.

While we are formalising the calculus, we want to avoid making more assumptions than we
need. So what we are going to assume is very simple. There is a set R of numbers. We can do
arithmetic with these numbers as you are used to (formally, these numbers form a field, which
we defined precisely—but we won’t in this part of the course use the field axioms to justify
arithmetic), there is an order < on them which behaves ‘as you expect’ (again, we could write
axioms saying how < interacts with + and ×, and formally we should), and the least upper bound
property holds.
Everything else we need, we will prove from these assumptions. It’s best to avoid thinking
about ‘what exactly are the real numbers’ ? If you want something concrete to think about, think
about the usual infinite-decimal representation that you learned in school, and we discussed in
the last chapter.
But it turns out we will not need to care; we can just work with the assumptions above.
CHAPTER 8. ANALYSIS: THE REAL NUMBERS 122

Example 8.9. Use the least upper bound property to show that there exists a number s ∈ R
such that s > 0 and s2 = 2.
Since the function x → x2 is increasing on the non-negative reals (i.e. if x ≥ 0) and it ‘doesn’t
have jumps’ (draw the graph!), the intuition is that we trace the number line (the x axis) from
0 upwards, until we reach the desired point x = s where x2 gets to 2. When x > s we will have
x2 > 2.
This is of course not a proof. We don’t know what ‘doesn’t have jumps’ means formally
(we’ll get to that!) and ‘trace the number line’ is not part of the least upper bound property.
But this idea does suggest something: we can split the reals into two parts: the part
S = {x ∈ R ∣ x2 < 2} of numbers that are ‘too small’ and the rest. Again, draw the graph and
look at where S is on the x axis. We would really like to say that s is the number ‘at the end of’
S. This informal idea is exactly what ‘least upper bound’ is supposed to formalise.
So we would like to say that s = sup S exists and that it satisfies s2 = 2. Let’s now formalise
this.
Proof. To begin with, we justify that x → x2 is an increasing function for x ≥ 0. That is, if
0 ≤ y < z then we want to prove y 2 < z 2 . That is the same as proving z 2 − y 2 > 0, and we can
factorise z 2 − y 2 = (z − y)(z + y) which is positive because both factors are positive.
Let S = {x ∈ R ∣ x2 < 2}. We now want to show sup S exists. This is what the least upper
bound property is for. We just need to show S is not empty and it has an upper bound.
First, 1 ∈ S (by definition), so S is not empty. And for example 3 is an upper bound for S.
This needs some justification: why is it that everything in S is at most 3? In other words, why
is everything bigger than 3 not in S? Well, 32 = 9, and so (because of the increasing property) if
we are given any x with 3 < x then 32 < x2 . That means in particular x2 > 2, so x is not in S.
Since we now know S is not empty and has an upper bound, by the least upper bound
property sup S exists. Let s = sup S. We can notice that s ≥ 1 since 1 ∈ S.
Finally, we need to prove s2 = 2. We do this by showing that each of the two alternatives
s2 > 2 and s2 < 2 leads to a contradiction.
Suppose first that s2 > 2. Intuitively, s is ‘too big’. We should be trying to contradict the
‘least’ part of ‘least upper bound’; we want to find an upper bound of S that is smaller than s.
That is, we want to find some small ε > 0 such that s − ε is an upper bound for S. By the
increasing property, that is the same as finding a small ε > 0 such that (s − ε)2 > 2.
2
We choose ε = s 2s−2 ; this formula comes out positive since s2 > 2 and s ≥ 1. Calculating, we
get
2
(s − ε)2 = s2 − 2sε + ε2 > s2 − 2sε = s2 − 2s s 2s−2 = 2 .
Here the inequality is since ε2 > 0. We are done in this case.
Now suppose s2 < 2. Again, intuitively s is ‘too small’ so we should be trying to contradict
the ‘upper bound’ part of ‘least upper bound’; we want to find something in S which is bigger
than s.
That is, we want to find some small ε > 0 such that s + ε is in S, that is such that (s + ε)2 < 2.
2−s2 2−s2
We choose ε = min ( 12 , 2s+1 ). That is, ε is whichever is smaller out of 12 and 2s+1 . Again, since
s < 2 and s ≥ 1 this formula comes out positive.
2

Now we can calculate:

2
(s + ε)2 = s2 + 2sε + ε2 = s2 + (2s + ε)ε < s2 + (2s + 1)ε ≤ s2 + (2s + 1) 2s+1
2−s
= 2.
Here the < comes because we replaced ε with 1: we know ε ≤ 12 so in particular ε < 1. And the ≤
2−s2
is since we know ε ≤ 2s+1 . What this calculation says is that s + ε ∈ S. This case is done.
The only remaining possibility is that s2 = 2. To conclude, we have shown the existence of
s ∈ R such that s > 0 and s2 = 2.
CHAPTER 8. ANALYSIS: THE REAL NUMBERS 123

In the ‘checking that s2 = 2’ part of this proof, there are two big ‘magic steps’ where I just
pulled a weird formula for ε out of somewhere and it turned out to work. Of course, these steps
cannot really be magic. Where does the formula come from?
The first one is easier. We suppose s2 > 2, so s is ‘too big’ and we can think about removing
some tiny ε > 0.
We knew from the start we wanted to get (s−ε)2 > 2. That’s the same as saying s2 −2sε+ε2 > 2.
This looks like a quadratic in ε, which is nasty: let’s try to avoid solving a quadratic. Let’s work
backwards.
It’s enough to get s2 − 2sε ≥ 2, because then we can add ε2 to both sides (and 2 + ε2 > 2).
This is now a linear inequality, which is easy. Rearranging, it’s enough to get
s2 −2
ε≤ 2s .

Well, but we can choose ε, so in particular we can choose ε to satisfy this inequality. Maybe the
2
easiest option was the choice we made; though really any ε satisfying 0 < ε ≤ s 2s−2 would work.
The other calculation needs a bit more explanation. The idea is the same. Since we assume
s2 < 2, we think s is ‘too small’ and we want to look at s + ε: think of ε as being tiny, and work
backwards.
We want (s + ε)2 < 2, or equivalently s2 + 2εs + ε2 < 2. This time the ε2 doesn’t help us. But
if ε is tiny, then ε2 should be even smaller. In particular, we should be able to write ε2 < ε. So
let’s do that.
It’s enough to get s2 + 2εs + ε ≤ 2, because ε2 < ε. But this is (again) linear and we can
2−s2 2−s2
rearrange it: it’s enough to get ε ≤ 2s+1 . And again—we can choose ε; we can choose ε = 2s+1 .
This time, though, we need to be a bit careful. We said ε is tiny so ε2 < ε. Our argument
2−s2
relied on it! But what if this funny fraction 2s+1 isn’t tiny? If it is 1 or bigger, then our ‘ε2 < ε’
assumption would go wrong and the proof would not work.
2−s2
This is why we chose ε = min ( 12 , 2s+1 ). We insist ε is at most 12 so that we can write ε2 < ε.
2−s 2
And then we insist ε is at most 2s+1 so the rest of the argument works.

Example 8.10. Prove the ‘greatest lower bound property’: if S is a non-empty subset of R
that is bounded below, then S has a greatest lower bound.

This is an exercise.

8.2.3 The Archimedean property of the real numbers

The least upper bound property of the real numbers tells us (intuitively!) that the real numbers
don’t have ‘gaps’, which is certainly nice.
But why did we insist on saying that 0.9999... is the same real number as 1? Why don’t we
say that they are two different numbers, differing by something ‘infinitesimal’ ? It turns out that
we can’t—it’s not compatible with the least upper bound property. This is the content of the
next theorem, called the Archimedean property of the real numbers.
Remark 8.11. Some mathematicians do study number systems which contain ‘infinitesimals’,
but the price to pay is that they do not behave as nicely as we would like. We will not go into
this.
CHAPTER 8. ANALYSIS: THE REAL NUMBERS 124

Theorem 8.12 (Archimedean property). If x, y ∈ R and x > 0, then there exists an n ∈ N such
that y < nx.

Proof. Suppose we are given x and y such that the conclusion is false. That is, that there does
not exist an n ∈ N such that y < nx. This means that for all n ∈ N, y ≥ nx. In other words, y is
an upper bound of the non-empty set S = {nx ∣ n ∈ N}.
By the least upper bound property of the reals, S has a least upper bound u∗ . Note that
u∗ − x is not an upper bound of S, since u∗ − x is smaller than u∗ (x is positive) and u∗ is the
least upper bound. Hence there exists an element mx ∈ S (with m ∈ N) such that u∗ − x < mx,
that is, u∗ < (m + 1)x. But (m + 1)x is also an element of S, and we just said (m + 1)x > u∗ : this
contradicts the fact that u∗ is an upper bound of S.
Let’s see why the Archimedean property tells us that 0.999... = 1. Well, if not, then x =
1 − 0.999... cannot be zero; suppose x > 0. (This is the obvious thing to assume, but formally
we should consider the ‘other’ case that it is negative.) And let y = 1. Then the Archimedean
property says that there is some natural number n such that 1 < nx, i.e. x > n1 . Putting it another
way, we have
n + 10 + 100 + 1000 + ⋅ ⋅ ⋅ < 1
1 9 9 9

for some fixed natural number n. But this is not possible. To see that, observe that
n
1
n + 10
9
+ 100
9
+ 1000
9
+ ⋅⋅⋅ > 1
n + 10
9
+ 100
9
+ ⋅ ⋅ ⋅ + 109n = 1
n + 1010n−1

where the first inequality is simply because we’re leaving out all the (infinitely many, positive)
terms of the series after 109n , and the equality uses the formula for the sum of a geometric series.
But n < 10n is true for every natural number n, so n1 > 101n , so the right hand side of the above
is bigger than one.
As a consequence of the Archimedean property we are now able to prove that the set N of
natural numbers is not bounded above.

Example 8.13. Show that the set N is not bounded above.

Proof. Suppose that N is bounded above. Then N has an upper bound y ∈ R. Since 1 ∈ N, 1 ≤ y,
and in particular, y > 0. Let x = 1. By the Archimedean property (Theorem 8.12), there exists
an n ∈ N such that y < nx = n. This contradicts the fact that y is an upper bound of N.
Warning 8.14. It’s rather common for students to say ‘the natural numbers are bounded above,
by ∞’. This is wrong. The symbol ∞ is a very convenient thing to use—and we will use it
repeatedly in what follows—but it is not a real number. Trying to treat ∞ as a real number is
one of the quickest ways to get to a wrong answer. Not just wrong in that your proof doesn’t
make sense, but wrong in that you get the wrong number out of your calculations, you end up
calculating things like 3 − 1 = 0.

Example 8.15. Set S = { n1 ∣ n ∈ N}. Show that inf S = 0.

Proof. We know that 0 is a lower bound of S. Suppose that l is a lower bound of S such that
l > 0. By the Archimedean property (with the real numbers x and y taken as x = 1 (> 0) and
y = 1l ), there exists n ∈ N such that 1l = y < nx = n ⋅ 1 = n, and so n1 < l, contradicting the fact
that l is a lower bound of S. Thus any lower bound of S must be less than or equal to 0. Hence
0 is the infimum of S.
CHAPTER 8. ANALYSIS: THE REAL NUMBERS 125

8.2.4 Intervals and absolute values

Definition 8.16. An interval (in R) is a non-empty subset S of R with the property: if x, y ∈ S
and x ≤ z ≤ y, then z ∈ S.

For instance, the set S = {x ∈ R ∣ x < 2} is an interval: if x and y are both in S, and x ≤ z ≤ y,
then in particular z ≤ y < 2 so z ∈ S.
An interval may or may not have an upper bound: if it does have an upper bound, then it
has a supremum which may or may not be in the interval. Similarly, an interval may or may not
have a lower bound: if it does have a lower bound, then it has an infimum which may or may
not be in the interval. In the example above, S has an upper bound, and the supremum is not
in S, while S has no lower bound. There are thus three possible forms for the “lower” end of
an interval, and three possible forms for the “upper end”, making nine forms in all. These are
listed in the figure below, along with the notation for each type of interval.

(a, b) = {x ∈ R ∣ a < x < b}

a b
[a, b] = {x ∈ R ∣ a ≤ x ≤ b}
a b
(a, b] = {x ∈ R ∣ a < x ≤ b}
a b
[a, b) = {x ∈ R ∣ a ≤ x < b}
a b
(a, ∞) = {x ∈ R ∣ a < x}
a
[a, ∞) = {x ∈ R ∣ a ≤ x}
a
(−∞, b) = {x ∈ R ∣ x < b}
b
(−∞, b] = {x ∈ R ∣ x ≤ b}
b
(−∞, ∞) = R

Figure 8.2: Intervals in R

An interval of the form (−∞, b), (a, b) or (a, ∞) is called an open interval. An interval of the
form (−∞, b], [a, b] or [a, ∞) is called a closed interval.
Thus in the notation for intervals used in Figure 8.2, a parenthesis ‘(’ or ‘)’ means that the
respective endpoint is not included, and a square bracket ‘[’ or ‘]’ means that the endpoint is
included. For example, [0, 1) is the set of all real numbers x such that 0 ≤ x < 1. (Note that the
use of the symbol ∞ in the notation for intervals is simply a matter of convenience and is not
be taken as suggesting that there is a real number ∞.) We do not give any special name to
intervals of the form [a, b) or (a, b].
CHAPTER 8. ANALYSIS: THE REAL NUMBERS 126

In analysis, in order to talk about notions such as convergence and continuity, we will need
a notion of ‘closeness’ between real numbers. This is provided by the absolute value ∣ ⋅ ∣, and the
distance between real numbers x and y is ∣x − y∣. We give the definitions below.
Definition 8.17.
1. For a real number x, the absolute value ∣x∣ of x is defined as follows:
⎧
⎪
⎪ x if x ≥ 0,
∣x∣ = ⎨
⎪
⎩−x if x < 0.
⎪

2. The distance between two real numbers x and y is the absolute value ∣x − y∣ of their difference.
Note that ∣x∣ ≥ 0 for all real numbers x, and that ∣x∣ = 0 if and only if x = 0. Thus ∣1∣ = 1,
∣0∣ = 0, ∣−1∣ = 1, and the distance between the real numbers −1 and 1 is equal to ∣−1−1∣ = ∣−2∣ = 2.
The distance gives a notion of closeness of two points, which is crucial in the formalization of
the notions of analysis.
We can now specify regions comprising points close to a certain point x0 ∈ R in terms of
inequalities in absolute values, that is, by demanding that the distance of the points of the
region, to the point x0 , is less than a certain positive number δ, say δ = 0.01 or δ = 0.0000001,
and so on. See Figure 8.3.

I
³¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹· ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ µ

x0 − δ x0 x x0 + δ

Figure 8.3: The interval I = (x0 − δ, x0 + δ) = {x ∈ R ∣ we have ∣x − x0 ∣ < δ} is the set of all points
in R whose distance to the point x0 is strictly less than δ (> 0).

The following properties of the absolute value will be useful.

Theorem 8.18. If x, y are real numbers, then

∣xy∣ = ∣x∣ ∣y∣ and (8.3)

∣x + y∣ ≤ ∣x∣ + ∣y∣ . (8.4)

Proof. We prove (8.3) by exhausting all possible cases:

x = 0 or y = 0.
Then ∣x∣ = 0 or ∣y∣ = 0, and so ∣x∣ ∣y∣ = 0. On the other hand, as x = 0 or y = 0, it follows that
xy = 0 and so ∣xy∣ = 0.
x > 0 and y > 0.
Then ∣x∣ = x and ∣y∣ = y, and so ∣x∣ ∣y∣ = xy. On the other hand, as x > 0 and y > 0, it follows
that xy > 0 and so ∣xy∣ = xy.
x > 0 and y < 0.
Then ∣x∣ = x and ∣y∣ = −y, and so ∣x∣ ∣y∣ = x(−y) = −xy. On the other hand, as x > 0 and y < 0,
it follows that xy < 0 and so ∣xy∣ = −xy.
x < 0 and y > 0.
This follows from the previous case by swapping x and y.
CHAPTER 8. ANALYSIS: THE REAL NUMBERS 127

x < 0 and y < 0.

Then ∣x∣ = −x and ∣y∣ = −y, and so ∣x∣ ∣y∣ = (−x)(−y) = xy. On the other hand, as x < 0 and
y < 0, it follows that xy > 0 and so ∣xy∣ = xy.

This proves (8.3).

Next we prove (8.4). First observe that from the definition of ∣ ⋅ ∣, it follows that for any real
x ∈ R, ∣x∣ ≥ x: indeed if x ≥ 0, then ∣x∣ = x, while if x < 0, then −x > 0, and so ∣x∣ = −x > 0 > x.
From (8.3), we also have ∣ − x∣ = ∣ − 1 ⋅ x∣ = ∣ − 1∣∣x∣ = 1∣x∣ = ∣x∣, for all x ∈ R, and so it follows that
∣x∣ = ∣ − x∣ ≥ −x for all x ∈ R. We have the following cases:

x + y ≥ 0.
Then ∣x + y∣ = x + y. As ∣x∣ ≥ x and ∣y∣ ≥ y, we obtain ∣x∣ + ∣y∣ ≥ x + y = ∣x + y∣.
x + y < 0.
Then ∣x + y∣ = −(x + y). Since ∣x∣ ≥ −x and ∣y∣ ≥ −y, it follows that ∣x∣ + ∣y∣ ≥ −x + (−y) =
−(x + y) = ∣x + y∣.

This proves (8.4).

The second of these inequalities, (8.4), is often called the triangle inequality. If you draw
a triangle in the plane, with sides of length a, b and c, then c ≤ a + b; to go from one vertex
of a triangle to another, it’s never longer to go down the side connecting them than via the
other two sides. The equation ∣x + y∣ ≤ ∣x∣ + ∣y∣ is just the special case when all three points of
the triangle are on a line.
Remark 8.19. We won’t go into it in this course, but this is going in the direction of why analysis
is such an important part of mathematics. It’s often useful to be able to make precise things
like: this shape is more like a circle than that shape; the distance you have to travel in the space
of shapes to get to the circle is shorter. Or the same thing with similarity of shapes replaced by
similarity of functions, or of many other things. One real-world application: if you use Amazon,
it recommends you products which were bought by people whose purchase history is similar
to yours. But it would take far too much computing time to find all users on Amazon whose
purchase history is similar to yours and look up what they bought that you don’t have, every
time you log on.
What is much quicker is for Amazon to keep track of a few ‘model users’ whose purchase
histories are fairly different, and for each model user a list of what people similar to the model
user bought. Then when you log on, Amazon just has to look up the one or two model users
whose purchase history is closest to yours, and output the list of what people similar to them
bought (minus the things you already have). Why does that work? Well, because of the triangle
inequality. The distance (in purchase history space) between your purchase history and a model
user M is maybe 3 units, and the similarity between M and any one of the people they are
similar to, say N , is at most 5 units. Why is something that N bought likely to be a good
suggestion for you? Well, because you and N cannot be more than 3 + 5 = 8 units apart: in this
‘purchase history space’ the triangle inequality holds. You have similar preferences to N , so you
might well like something they bought.
In MA203 you can study ‘metric spaces’—the abstract idea of a space together with an
idea of the distance between two points in the space. All one assumes is that two points are at
zero distance if and only if they are the same point (otherwise the distance is positive) and the
triangle inequality: the distance between any two points is never more than going via a third.
Amazingly, you can say quite a lot about these spaces without assuming anything more.
CHAPTER 8. ANALYSIS: THE REAL NUMBERS 128

8.3 Sample exercises

Exercise 8.1. Fill in the following table for each of the sets S = A, B, .., E below.
An A If sup S exists, If inf S exists,
Is S
upper lower sup S inf S then is sup S then is inf S max S min S
bounded?
bound bound in S? in S?

(a) A = {x ∈ R ∣ 0 ≤ x < 1} (d) D = {0, 2, 5, 2016}

(b) B = {x ∈ R ∣ 0 ≤ x ≤ 1}
(c) C = {x ∈ R ∣ x ∈ N and x prime} (e) E = {(−1)n (1 + n1 ) ∣ n ∈ N}

Exercise 8.2. Let A and B be non-empty subsets of R that are bounded above and define
A + B = {x + y ∣ x ∈ A and y ∈ B} .
(a) Show that sup A + sup B is an upper bound for A + B.
Deduce that sup(A + B) exists and that sup(A + B) ≤ sup A + sup B.
(b) For any real number ε > 0, show that (sup A − ε) + (sup B − ε) is not an upper bound for
A + B.
Deduce that sup(A + B) ≥ sup A + sup B − 2ε, for every ε > 0.
(c) Show that sup(A + B) = sup A + sup B.
Exercise 8.3. Let S be a non-empty set of positive real numbers, and define S −1 = { x1 ∣ x ∈ S}.
(a) Show that, if inf S = 0, then S −1 is not bounded above.
(b) Show that, if inf S > 0, then S −1 is bounded above and sup S −1 = 1
inf S .

Exercise 8.4. In this exercise we define the floor and ceiling functions. For any x ∈ R define
the set Sx = {n ∈ Z ∣ n ≤ x}.
(a) Show that, for any x ∈ R, the set Sx is non-empty and bounded above.
Hint: To show that Sx ≠ ∅, you will need the Archimedean property.
(b) Show that sup Sx exists and sup Sx ∈ Sx for any x ∈ R. Explain why we obtain as a
consequence that the following gives a proper definition of the floor function ⌊⋅⌋ ∶ R → Z
⌊x⌋ = max {n ∈ Z ∣ n ≤ x} , x ∈ R.

(c) Show that x − 1 < ⌊x⌋ ≤ x, for every x ∈ R.

(d) Adapt (a)) and (b)) to explain why the following gives a proper definition of the ceiling
function ⌈⋅⌉ ∶ R → Z
⌈x⌉ = min {n ∈ Z ∣ n ≥ x} , x ∈ R.
√
(e) Show that ⌊ k 2 + k⌋ = k for all k ∈ N.
Exercise 8.5. Let S be a subset of R that does not have a maximum. Show that, for every
x ∈ S, there is y ∈ S with y > x.
Hint: Deal separately with the two cases: (i) S has a supremum; (ii) S has no supremum.
Exercise 8.6. Prove that if x, y are real numbers, then ∣∣x∣ − ∣y∣∣ ≤ ∣x − y∣.
CHAPTER 8. ANALYSIS: THE REAL NUMBERS 129

8.4 Comments on selected activities

Comment on Activity 8.1. What is missing is that we don’t take into account that the surface
is slanted. By the logic we used, if we have a flat 1 by 1 sheet, standing vertically, we should
estimate its area by integrating from 0 to 1 the length of the line which is a horizontal slice
1
through, namely 1. So the surface area should be ∫x=0 1 dx = 1, which is correct; this sheet isn’t
slanted.
But now put the sheet at a π4 angle (45 degrees) to the vertical. A horizontal slice through is
still a length 1 line, so now (by the logic we used) the surface area should be
√
1/ 2
∫x=0 1 dx = √1
2
,
√
since the vertical height of the sheet is now 1/ 2. Of course this is nonsense; the sheet hasn’t
really become smaller just because it is slanted. The same error has been made in calculating
the area of the sphere.
And now you should decide if you are really convinced by Archimedes’ argument for the
volume—if so, why?

8.5 Solutions to exercises

Solution to Exercise 8.1.
(a) 1 is one upper bound, and is in fact the supremum of A. 0 is a lower bound, and the
infimum of A. The set is bounded. As 0 is an element of A, but 1 is not, the set has a minimum
(namely 0), but no maximum.
(b) As (a), except that now the supremum (1) is in B, so B has a maximum.
(c) There is no upper bound (this is equivalent to saying that there are infinitely many prime
numbers). There is a lower bound, and the infimum and minimum are both equal to 2, the
smallest prime number. The set is not bounded.
(d) Any finite set of real numbers has a maximum and a minimum, and the rest of the answer
follows. In this case, the maximum is 2016, and the minimum is 0.
(e) It’s worth evaluating the first few elements of the sequence (−1)n (1 + n1 ), to get an idea
of what’s going on. These elements are −2, 3/2, −4/3, 5/4, . . . . This set E has a minimum and a
maximum, namely −2 and 3/2 respectively, and the answers to all the other questions follow.
Solution to Exercise 8.2.
(a) A general element of A + B is of the form x + y for some x ∈ A and y ∈ B. For any such
pair of elements, we have that x ≤ sup A and y ≤ sup B (since sup A is an upper bound for A and
sup B is an upper bound for B) and hence x + y ≤ sup A + sup B. Therefore indeed sup A + sup B
is an upper bound for A + B.
This proves that A + B is bounded above, and the set is non-empty since both A and B are.
Hence A + B has a supremum, which is certainly at most the upper bound sup A + sup B.
(b) Since sup A − ε is not an upper bound for A, there is some x ∈ A with x > sup A − ε.
Similarly, there is some y ∈ B with y ≥ sup B − ε. Now x + y is an element of A + B with
x + y > sup A − ε + sup B − ε = sup A + sup B − 2ε. Hence sup A + sup B − 2ε is not an upper bound
for A + B. Moreover, no real number less than sup A + sup B − 2ε is an upper bound for A + B,
so the supremum of A + B (which is an upper bound for A + B) is at least sup A + sup B − 2ε.
(c) We have seen in (a) that sup(A + B) ≤ sup A + sup B. If sup(A + B) < sup A + sup B, then
there is a positive real number ε such that sup(A + B) < sup A + sup B − 2ε (to be specific, set
ε = 13 (sup A+sup B−sup(A+B))), but this contradicts (b). Hence also sup(A+B) ≥ sup A+sup B,
and the equality follows.
CHAPTER 8. ANALYSIS: THE REAL NUMBERS 130

Solution to Exercise 8.3. One key idea is that, provided we stay within the realm of positive
real numbers, a larger element of S corresponds to a smaller element of S −1 .
(a) We prove the contrapositive: if S −1 is bounded above, then inf S =/ 0. Take an upper
bound M for S −1 . [This is what we gain: we can get our hands on a specific object M , and
reason about it.] Note that M > 0, as S contains at least one positive real number. Then x ≤ M
for every element x ∈ S −1 , which is equivalent to saying that 1/z ≤ M for every element z ∈ S.
[Because the elements of S −1 are exactly the elements of the form 1/z for z ∈ S.] In turn, this is
equivalent to saying that z ≥ 1/M for every z ∈ S. [We use here that both M and z (an element
of S) are positive.] This means exactly that the positive real number 1/M is a lower bound for
S. The infimum of S is then at least 1/M , and therefore greater than 0.
(b) Suppose now that s = inf S > 0. We claim that 1/s is an upper bound for S. [It’s important
to have an idea of what is going on, so that you can see that this is what is likely to be true.]
Indeed, as s is a lower bound for S, we have s ≤ z for every z ∈ S, and therefore 1/z ≤ 1/s for
every z ∈ S. This means that x ≤ 1/s for every x ∈ S −1 , i.e., that indeed 1/s is an upper bound
for S −1 .
What do we have left to prove? We have seen that S −1 is bounded above, and that 1/ inf S = 1/s
is an upper bound for S −1 . We still need to show that 1/s is the least upper bound for S −1 .
Suppose then that 0 < u < 1/s, i.e., 1/u > s. As s is the infimum of S, it follows that 1/u is not
a lower bound for S, in other words there is some z ∈ S with z < 1/u, or equivalently u < 1/z.
But now 1/z is an element of S −1 , so u is not an upper bound for S −1 . Hence indeed 1/s is the
supremum of S −1 , as required.
(By the way, normally we do not attach any meaning to S −1 when S is a set: sets don’t have
“inverses”. If, in the future, you want the notation to mean what it does here, you either have
to define it afresh, or you could say “where, for a set S of positive real numbers, S −1 is as in
Exercise 8.3”.)
Solution to Exercise 8.4.
(a) The set Sx is certainly bounded above by x, so the main issue is to show that Sx is
non-empty. If x ≥ 0, then 0 ∈ Sx , so we can assume x is negative. Once you’ve realised that,
you should see that the Archimedean property is exactly what we need: given any (negative) x,
there is an integer m such that −m ≥ −x, and so m ≤ x. The integer m is then in the set Sx , so
Sx is non-empty.
(b) For each x, the set Sx is non-empty and bounded above, so has a supremum sup Sx .
Moreover, the set Sx is a set of integers. You have seen in the first half of the course that a set of
integers bounded above has a maximum, so the set has a maximum element (which is the same
as the supremum). Thus, for every real number x, max{n ∈ Z ∣ n ≤ x} is a well-defined integer.
(c) The fact that ⌊x⌋ ≤ x is immediate from the definition. To see the other inequality,
suppose that ⌊x⌋ ≤ x − 1. Then m = ⌊x⌋ + 1 is an integer with m ≤ (x − 1) + 1 = x, so m is in the
set {n ∈ Z ∣ n ≤ x}, and m > ⌊x⌋, contradicting the choice of ⌊x⌋ as the maximum of this set.
(d) Similar to (a) √
and (b). √
(e) Evidently k ≤ k 2 + k, so k ∈ S√k2 +k . On the other hand, k + 1 > k 2 + k: both sides are
non-negative, so this inequality is equivalent to (k + 1)2 > k 2 + k, which is indeed true.√ Thus no
integer greater than k is in S k2 +k , and hence k is the maximum of S k2 +k , i.e., k = ⌊ k 2 + k⌋,
√ √

whenever k ∈ N.
Solution to Exercise 8.5.
Recall the definition: if S has a supremum s, and s ∈ S, then s is the maximum of S.
(Otherwise S has no maximum.)
So there are two ways that S could fail to have a maximum: (i) S has a supremum, s, but s
is not a member of S; (ii) S has no supremum.
In case (i), take any element x of S. Then x ≤ s because s is an upper bound for S, but
x =/ s since x is in S and s isn’t. Thus x < s, and so x is not an upper bound for S. That means
CHAPTER 8. ANALYSIS: THE REAL NUMBERS 131

exactly that there is some y ∈ S with x < y.

In case (ii), either S is empty or S is not bounded above. In the first case there is nothing to
check (i.e., there is no element x in S that could possibly be a counterexample). In the second
case, any element x of S is not an upper bound for S, and this again means that there is some
y ∈ S with x < y.
Actually, you don’t need to divide into cases: you can argue as follows. Suppose S has no
maximum, and let x be any element of S. Then x is not a supremum of S. Certainly x is less
than any upper bounds of S, so it must be that x is not itself an upper bound of S. This means
exactly that there is some y ∈ S with x < y.
Solution to Exercise 8.6.
One can do this by breaking into cases, depending on the signs of x and y. Here’s an
alternative approach, based on the inequality ∣a + b∣ ≤ ∣a∣ + ∣b∣, valid for all real numbers a and b,
that we proved in lectures. Applying this with a = x − y and b = y gives ∣x∣ ≤ ∣x − y∣ + ∣y∣, and so
∣x∣ − ∣y∣ ≤ ∣x − y∣. Similarly ∣y∣ − ∣x∣ ≤ ∣y − x∣ = ∣x − y∣. Now, ∣∣x∣ − ∣y∣∣ is equal to either ∣x∣ − ∣y∣ or
−(∣x∣ − ∣y∣) = ∣y∣ − ∣x∣, and so it follows that ∣∣x∣ − ∣y∣∣ ≤ ∣x − y∣.
 Analysis: Sequences and limits
9
If we have a function f ∶ R → R, what does f ′ (0), the derivative of f at 0, mean? You probably
would say something like ‘the slope of the tangent to f (x) at x = 0’. Well, yes—but how can
we find what that is, or even say if it exists? If the function has some funny kink at zero, like
f (x) = ∣x∣, then this phrase ‘the tangent..’ doesn’t make much sense. Any line you draw doesn’t
look much like a tangent.
One way is to imagine drawing a line from (0, f (0)) to (h, f (h)) where h is ‘small’. Intuitively,
the slope of this line—which is f (h)−f
h−0
(0)
—should be more or less what we want. If we make h
smaller and smaller, we should get closer to the right answer. That is, basically, the definition of
‘derivative’. But in order to make sense of it, we need to formalise what it means to ‘get closer to
the right answer’. In other words, we need to say what it means for a ‘sequence’ of real numbers
to ‘converge to a limit’.

9.1 Sequences
The notion of a sequence occurs in ordinary conversation. An example is the phrase “an
unfortunate sequence of events”. In this case, we envision one event causing another, which in
turn causes another event and so on. We can identify a first event, a second event, etcetera.
A sequence of real numbers is a list

a1 , a 2 , a 3 , . . .

of real numbers, where there is the first number (namely a1 ), the second number (namely a2 ),
and so on. For example,
1, 12 , 31 , . . .
is a sequence of real numbers. The first number is 1, the second number is 12 and so on. (There
may not be a connection between the numbers appearing in a sequence.) If we think of a1 as
f (1), a2 as f (2), and so on, then it becomes clear that a sequence of real numbers is a special
type of function, namely one with domain N and co-domain R. This leads to the following
formal definition.

Definition 9.1. A sequence of real numbers is a function f ∶ N → R.

Only the notation is somewhat unusual. Instead of writing f (n) for the value of f at a
natural number n, we write an . The entire sequence is then written in any one of the following
ways:
(an )n∈N , (an )∞
n=1 , (an )n≥1 , (an ).

132
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 133

In (an )∞
n=1 , the ∞ symbol indicates that the assignment process 1 ↦ a1 , 2 ↦ a2 , . . . continues
indefinitely. In these notes, we shall normally use the notation (an )n∈N . In general, the terms of
a sequence need not be real numbers, but in these notes we shall only be dealing with sequences
whose entries are real numbers, so we shall simply refer to them as sequences from now on.
The n-th term an of a sequence may be defined explicitly by a formula involving n, as in the
example given above:
an = n1 , n ∈ N.
It might also sometimes be defined recursively. For example,
a1 = 1, an+1 = n
n+1 an for n ∈ N.
(Write down the first few terms of this sequence.)
Example 9.2.
(i) ( n1 )n∈N is a sequence with the n-th term given by n1 , for n ∈ N. This is the sequence

1, 21 , 31 , . . . .

(ii) (1 + n1 )n∈N is a sequence with the n-th term given by 1 + n1 , for n ∈ N. This is the sequence

2, 32 , 43 , 45 , 65 , 76 , . . . .

(iii) ((−1)n (1 + n1 ))n∈N is a sequence with the n-th term given by (−1)n (1 + n1 ), for n ∈ N. This
is the sequence
−2, 23 , − 43 , 45 , − 65 , 67 , . . . .

(iv) ((−1)n )n∈N is a sequence with the n-th term given by (−1)n , for n ∈ N. This sequence is
simply
−1, 1, −1, 1, −1, 1, . . .
with the n-th term equal to −1 if n is odd, and 1 if n is even.
(v) (1)n∈N is a sequence with the n-th term given by 1, for n ∈ N. This is the constant sequence
1, 1, 1, . . . .

(vi) (n)n∈N is a sequence with the n-th term given by n, for n ∈ N. This is the strictly increasing
sequence
1, 2, 3, . . . .
(vii) ( 111 + 212 + 313 + ⋅ ⋅ ⋅ + n1n )n∈N is a sequence with the n-th term given by 1
11 + 212 + 313 + ⋅ ⋅ ⋅ + n1n ,
for n ∈ N. This is the sequence of ‘partial sums’
1 1 1 1 1 1
, + , + + , ....
11 11 22 11 22 33

(viii) (n1 000 000 2−n )n∈N is the sequence whose nth term is n1 000 000 2−n . Its first term is 12 , its
second term is a huge integer with about 30 000 decimal digits, its third term is even
bigger, and if you keep calculating, the terms will just keep getting bigger and bigger as
long as you have patience to keep going.
n
(ix) ( n2 + (−1)
n
)n∈N is the sequence whose terms are
1
1 , 23 , 31 , 34 , 15 , 63 , . . .
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 134

9.2 Limit of a convergent sequence

A sequence can be graphed. For instance, the first 7 points of the graph of the sequence ( n1 )n∈N
are displayed in Figure 9.1.

1
0.5
0
0 1 2 3 4 5 6 7

Figure 9.1: First 7 points of the graph of the sequence ( n1 )n∈N .

This portion of the graph suggests that the terms of the sequence ( n1 )n∈N “tend toward 0” as
n increases. This is consistent with the idea of convergence that you might have encountered
before: a sequence (an )n∈N converges to some real number L, if the terms an get “closer and
closer” to L as n “increases without bound”. Symbolically, this is represented using the notation

lim an = L,
n→∞

where L denotes the limit of the sequence. If there is no such finite number L to which the
terms of the sequence get arbitrarily close, then the sequence is said to diverge.
The problem with this characterization is its imprecision. Exactly what does it mean for the
terms of a sequence to get “closer and closer”, or “as close as we like”, or “arbitrarily close” to
some number L? Even if we accept this apparent ambiguity, how would one use the definition
given in the preceding paragraph to prove theorems that involve sequences? Since sequences are
used throughout analysis, the concepts of their convergence and divergence must be carefully
defined.
For example, the terms of (1 + n1 )n∈N get “closer and closer” to 0 (indeed the distance to 0
keeps decreasing), but its limit is 1:

2
1.5
1
0.5
0
0 1 2 3 4 5 6 7

Figure 9.2: First 7 points of the graph of the sequence (1 + n1 )n∈N .

CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 135

Some of the terms of ((−1)n (1 + n1 ))n∈N get “as close as we like” or “arbitrarily close” to 1,
but the sequence has no limit:

2
1
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
−1
−2

Figure 9.3: First sixteen points of the graph of the sequence ((−1)n (1 + n1 ))n∈N .

The terms of (n1 000 000 2−n )n∈N , despite all appearances, do eventually get smaller. The first
few terms—indeed, the first few million terms—are enormous. But eventually (when n is very
large) the nth term is guaranteed to be very close to 0.
n
Finally, the terms of ( n2 + (−1)n
)n∈N don’t always get closer to 0. If you look at the nth
term where n is very large, that term will be very tiny—it will be either n1 or n3 depending on
whether n is odd or even—but the even-numbered terms are almost three times as big as the
odd-numbered term before: it keeps on getting further from 0:

1
0.5
0
0 1 2 3 4 5 6 7 8 9 10 11

n
Figure 9.4: First 11 points of the graph of the sequence ( n2 + (−1)
n
)n∈N .

So which of the sequences from Example 9.2 converge to 0? The answer is: the ones where
(eventually, maybe when n is very large) the terms are guaranteed to be close to 0. These are
examples (i), (viii) (even though the first few terms are enormous and growing), and (ix) (even
though each even term is further from 0 than the previous odd term). Sequence (ii) does converge,
but to the limit 1 not 0 (even though the terms are always getting closer to 0). Sequence (iii)
doesn’t converge to any limit (even though lots of terms are very close to 1, and lots more to
−1). The same is true for sequence (iv). The sequence (v) (obviously!) converges to 1.
The sequence (vi) ‘obviously’ doesn’t converge; it just keeps getting bigger, so it can’t possibly
stay close to any fixed real number (whatever candidate limit you pick, when n is large enough
the nth term of the sequence is going to be much bigger than your candidate limit). And finally
the sequence (vii) does converge, but to a number bigger than 1. It’s obvious the terms are all
at least 1—but how do I know it converges? More on that later.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 136

More generally, we want to say a sequence (an )n∈N converges to the real number L if, when
n is very large, an is guaranteed to be close to L. The following, which formalises that idea, is
the key definitio for this chapter.
Definition 9.3. The sequence (an )n∈N is said to converge to L if for every real number ε > 0,
there exists an N ∈ N (possibly depending on ε) such that for all n > N ,

∣an − L∣ < ε.

Then we say that (an )n∈N is convergent with limit L and write

lim an = L .
n→∞

We may also say that (an )n∈N tends to L.

If there does not exist a number L such that lim an = L, then the sequence (an )n∈N is said
n→∞
to be divergent.
This definition is complicated; there are several quantifiers to deal with. Written in logical
notation, we say (an )n∈N converges to L if

∀ε > 0 , ∃N ∈ N such that ∀n > N , ∣an − L∣ < ε .

Remember, you need to understand this definition not just memorise it—and the English
text version is easier to understand!
Warning 9.4. As we saw back in Chapter 3, if you swap around the order of quantifiers you can
change the meaning of a logical statement. This is the case here: if you change around the order
of the quantifiers in ‘converges to L’ then you will get a statement which means something, but
not any more what you want!
You can try to prove a sequence converges by following the general strategies in Chapter 3.
That is, the first quantifier in the definition is ∀ε > 0, a universal statement. So the first line of
the proof should be ‘Let ε > 0 be given.’ and then you need to prove the statement

∃N ∈ N such that ∀n > N , ∣an − L∣ < ε .

where now you know that ε is some fixed positive real number. Now, this statement is an
existential statement: the easiest way to prove it is to find an N which works. In other words,
the next line of the proof is going to be something like ‘We choose N = ..’. Having chosen N ,
you need to prove
∀n > N , ∣an − L∣ < ε .
Back to a universal statement! So: ‘Fix n > N .’ And finally you need to prove

∣an − L∣ < ε

which is just a calculation.

The obvious question is: But how should we choose N ..? The answer is pretty similar to
what we saw in the proof of Exercise 8.9. That is, you need it to be big enough that the final
calculation works. At some point when you are proving ∣an − L∣ < ε, you will say ‘because n > N
and we chose N = ...’. You will not be able to write in the value for N until you try to do this
calculation: you’ll need to work backwards.
Activity 9.1. Try to prove, for each of the sequences in Example 9.2, that the comment above
is accurate, i.e. (i) converges to 0, and so on.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 137

Some of these sequences are harder to work with than others. If you couldn’t do any of them,
the worked examples that follow should help. If you could do all of them except (vii) and (viii),
you’re doing very well. (If you think you have proofs for (vii) and (viii), then either you are
doing exceptionally well, or you saw this material before, or you assumed something unjustified!)
Note that ∣an − L∣ < ε if and only if an ∈ (L − ε, L + ε). Hence pictorially, for a convergent
sequence with limit L, this definition means the following, as illustrated in Figure 9.5.
Pick any ε > 0, and consider the shaded strip of width ε around the horizontal line passing
through L. Then one can find an N ∈ N, large enough, such that all the terms an of the sequence,
for n > N , lie in the shaded strip.

L+ε

L−ε

... ...
N N +1 N +2 N +3

Figure 9.5: Convergence of a sequence with limit L.

It is definitely worth keeping this picture in mind for the rest of the chapter. Of course you
can do everything in Analysis just by sticking to the algebra and logic without ever drawing a
picture, but (at least for most people) trying to do this is a good way to get confused and make
errors.

9.2.1 Proving convergence of a sequence

Example 9.5. Show that ∃N , ∀n > N , ∣ 2n+51sin n − 0∣ < 1
1000 .

This example is what we would have to show in a proof of convergence if we happened to be

given ε = 1000
1
; of course, if you want to prove convergence, your proof needs to work for every
ε > 0.
Proof. The denominator 2n + 5 sin n is painful to work with. But sin n is always between 1 and
−1, so if n ≥ 5 then we have n + 5 sin n ≥ 0.
Adding n to both sides, we see that if n ≥ 5 then we have 2n + 5 sin n ≥ n, so 2n+51sin n ≤ n1 . So
(finally) we should choose N = 1000. Then for each n > N we have
∣ 2n+51sin n − 0∣ = 1
2n+5 sin n ≤ 1
n < 1
N = 1
1000

which is what we wanted.

Let’s notice two things about this proof. One, it is quite short and easy. It does not find ‘the
best’ N , which (in fact) is about 500. But that is irrelevant: it works, that is all we need, and
finding ‘the best’ N is a waste of time because you’re not asked to find it.
Two, we don’t actually know what N we should choose until we do some calculations. So
even though a ‘model’ proof of ‘there exists N such that...’ should start with ‘Let N be..’ and
then go on to check that this particular choice works, we’re never going to come up with a proof
in that order. The ‘model proof’ is:
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 138

Proof. Let N = 1000.

Then for all n > N , we have n + 5 sin n ≥ 0, so 2n + 5 sin n ≥ n, so

∣ 2n+51sin n − 0∣ = 1
2n+5 sin n ≤ 1
n < 1
N = 1
1000

which is what we wanted.

But if you want to write a proof that looks like this, the only way to do it is to leave a blank
space to fill in the ‘Let N = ...’ later, then start thinking about the calculation. That will be a
feature of all our proofs of convergence.
Let’s now prove that some sequences converge (and that some other ones do not). These
proofs will look a bit complicated at first: but when we want to prove a sequence (an )n∈N
converges to a limit L, the proof structure will be what is outlined above. It will start ‘Given
ε > 0’, and then there will be a choice of N as in the model proof above, and the rest of the
proof will look like the model proof: it will argue that ∣an − L∣ < ε is true for all n > N . Usually,
the choice of N will depend on the real number ε you are given.

Example 9.6. Show that ( 2n+51sin n )n∈N converges to 0.

We’ll prove this by starting with the model proof above, and changing it so that it works for
any ε > 0 and not just 1000
1
.
Proof. Given ε > 0, choose N to be the smallest integer which is at least as big as both 5 and 1ε .
Then for all n > N , we have n + 5 sin n ≥ 0, so 2n + 5 sin n ≥ n, so

∣ 2n+51sin n − 0∣ = 1
2n+5 sin n ≤ 1
n < 1
N ≤ε

which is what we wanted.

One important thing to notice here is that, while you should generally think of the ε you are
given at the start of the proof as ‘a very small number’, it doesn’t have to be. You might be
given ε = 100, and your proof has to work in that situation. We need to say that N should be at
least 5 in order to deal with this situation—to make sure n + 5 sin n ≥ 0 is true—even though
‘usually’ 1ε is much bigger than 5.
In order to avoid complicated phrases like ‘the smallest integer which is at least as big as
both 5 and 1ε ’ we make the following definitions.

Definition 9.7. The ceiling of a real number x, written ⌈x⌉, is the smallest integer at least as
big as than x; the floor of x, written ⌊x⌋, is the largest integer which is not bigger than x.
Given a collection of real numbers s1 , s2 , . . . , st , we write max(s1 , ..., st ) for the largest of
them, and min(s1 , ..., st ) for the smallest.

So ‘the smallest integer which is at least as big as both 5 and 1ε ’ is simply max (5, ⌈ 1ε ⌉).
You should now be able to do the following.

Activity 9.2. Use the definition of the limit of a sequence to show that ( n1 )n∈N is a convergent
sequence with limit 0.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 139

Here are some slightly harder worked examples.

Example 9.8. Use the definition of the limit of a sequence to show that (1 + 2n21−n )n∈N is a
convergent sequence with limit 1.

Proof. Given ε > 0, we choose N = ⌈ 1ε ⌉.

Suppose n > N . We notice that whatever natural number n is, we have 2n2 − n ≥ n. So we
can write
1 1 1
∣1 + 2 − 1∣ = 2 ≤ < N1 < ε ,
2n − n 2n − n n
which is what we wanted.
Let’s point out that (again) we get to this proof in a different order to the way it’s written
on paper. We know we want to eventually write down
1
∣1 + − 1∣ ≤ ... < ε ,
2n2 − n
and we need to figure out what the . . . steps in the middle are. We can simplify the left-hand-side,
so we do that. We get something which is still not very simple, 2n21−n , but we notice that we can
replace that with something bigger that still gets small when n is large, namely n1 . That is why
we write that 2n2 − n ≥ n is true for all natural numbers n—to justify that indeed 2n21−n ≤ n1 .
And we know n1 < N1 because we supposed that n > N . Finally, we choose N (really at this time,
even though we write it at the top of the proof) so that we can write N1 ≤ ε and be done.
The critical step here is to notice that we can get rid of the complicated 2n21−n by replacing
it with something bigger that we can still show is smaller than ε in the end. There are lots of
different choices we could have make; none is ‘best’, and this one is just the first one I happened
to think of.
In Analysis, there are generally several ‘right answers’ and your job is just to pick one of
them. In the proof above, we could just as well have noticed that n2 − n ≥ 0 is true for all natural
numbers, and so 2n21−n ≤ n12 . If we’d done that, most likely we would have chosen N = ⌈ √1ε ⌉ in
order to get our ε at the end. That’s also fine; it’s not better or worse, it is an equally good
different proof.
So how do you make the choice? Think about it and try something. There are three
possibilities.
One, you find that what you tried isn’t actually bigger (or you can’t see how to prove it). Well,
then you chose something too small, go back and try something bigger.
Two, you find that no matter how you calculate you cannot get the ε out at the end, you
keep getting something big (like 1). Well, then you chose something too big, go back and try
something smaller.
Three, your proof works. You don’t need to care about whether you could’ve made another
choice, you’re done.

Example 9.9. Use the definition to show that (an )n∈N = ((−1)n (1 + n1 ))n∈N is a divergent
sequence.

There is only one reason why a sequence might converge: the terms get close to a limit and
stay there as you look at larger and larger n. But there are a few different ways a sequence
(an )n∈N can fail to converge. It might be that an bounces around crazily all over the place,
sometimes very big and sometimes very small, no matter how big you make n. It might be that
it just keeps getting bigger and bigger and eventually gets too big for any candidate limit (or
the same thing but in the negative direction). It might be that the sequence doesn’t look too
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 140

crazy, but it jumps between being close to different real numbers. This (see Figure 9.3) is an
example of the last: the odd-numbered terms get close to −1, and the even-numbered terms get
close to 1.
It’s very tempting to say ‘we prove that −1 is not a limit. Then we prove that 1 is not a limit.
And now we are done’. But this is not enough. If you say this, you didn’t rule out the possibility
that 0 is a limit, or π, or any other real number; you need to rule out all the real numbers.
So the proof has to start ‘Given a real number L..’ and then go on to show that L cannot be
a limit. Let’s see how that goes.
Proof. Given a real number L, we need to show that ∀ε > 0 , ∃N ∈ N , ∀n > N , ∣an − L∣ < ε is a
false statement.
Working from the start, that means we need to find a counterexample to the ‘for all ε > 0’.
That is, one particular ε > 0 such that ∃N ∈ N , ∀n > N , ∣an − L∣ < ε is false. We’ll do that for
ε = 1.
Now we need to show that ∃N ∈ N , ∀n > N , ∣an − L∣ < 1 is a false statement. So: given any
N ∈ N, we need to show ∀n > N , ∣an − L∣ < 1 is a false statement, so we need to come up with
some particular number n > N such that ∣an − L∣ < 1 is false.
At this point, we need to look at the terms an and what L is.
Case 1: L ≥ 0. Choose n to be any odd integer greater than N , then we have

∣an − L∣ = ∣L − an ∣ = L − (−1)n (1 + n1 ) = L + 1 + n1

and since L ≥ 0, this is strictly bigger than 1.

Case 2: L < 0. Choose n to be any even integer greater than N . Then we have

∣an − L∣ = (−1)n (1 + n1 ) − L = 1 + n1 − L

and since L < 0 this is strictly bigger than 1.

Since these two cases are exhaustive, whatever L is given we proved that it is not a limit. So
the sequence has no limit: it is divergent.
Again, this proof is not written in the order we think of it. I got to this proof by staring at
Figure 9.3 and noticing that all the odd-numbered terms are smaller than −1 (so they’re not
within distance one of any number L which is ≥ 0) and all the even-numbered terms are bigger
than 1 (so they’re not with distance one of any L which is ≥ 0). For this part, it really helps to
look at the picture.
Well, but any L is either ≥ 0 or ≤ 0 (or both, if L = 0). So (whatever L is) we can choose
ε = 1. Anything less than 1 would work too. (But anything bigger than 1 would not work: we
would run into trouble with L = 0 and some other numbers.)

Activity 9.3. Check that for any real number L, any sequence (an )n∈N , and any ε > 0, we have:
‘ ∃N ∈ N , ∀n > N , ∣an − L∣ < ε is a false statement’ if and only if ‘there are infinitely many n
such that ∣an − L∣ ≥ ε’.

We can (and generally we would) shorten the proof of Example 9.9 a bit:
Proof. Given a real number L, we choose ε = 1. We want to show there are infinitely many n
such that ∣an − L∣ ≥ 1.
Case 1: L ≥ 0. For every odd integer n, we have

∣an − L∣ = ∣L − an ∣ = L − (−1)n (1 + n1 ) = L + 1 + n1

and since L ≥ 0, this is strictly bigger than 1.

 CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 141

Case 2: L < 0. For every even integer n, we have

∣an − L∣ = (−1)n (1 + n1 ) − L = 1 + n1 − L

and since L < 0 this is strictly bigger than 1.

Since these two cases are exhaustive, whatever L is given we proved that it is not a limit. So
the sequence has no limit: it is divergent.
Try one yourself. Hint: in this case, any ε > 0 you choose will work.

Activity 9.4. Prove that (n)n∈N is divergent.

We should also remember (from Chapter 3) that when we want to use a ‘for all’ statement,
what we will do generally doesn’t look like proving a ‘for all’ statement. As we’ve just seen, the
first line of proving the statement lim an = L is generally going to be ‘Given ε > 0, . . . ’. What
n→∞
do we do if we are given that (an )n∈N is a convergent sequence with limit L, and we want to
prove something about (say) L?
The notation lim an suggests that the limit is unique. But is this actually well-defined, or
n→∞
could it be that there is a convergent sequence with two different limits?
Warning 9.10. We saw the sequence ((−1)n (1 + n1 ))n∈N before. It’s tempting to say ‘yes, this
sequence has two limits, 1 and −1’. But this is false: this sequence doesn’t tend to any limit at
all; it is divergent, as we just proved.

Theorem 9.11. A convergent sequence has a unique limit.

The proof of this is a good example of how we can use the fact that a given sequence is
convergent with a certain limit.
Proof. Formally, to prove ‘a convergent sequence has a unique limit’, we need to show two
things:

(i) A convergent sequence has a limit;

(ii) A convergent sequence cannot have two different limits.

Here, (i) is true by the definition of convergence, so we only have to prove (ii).
In other words, if lim an = L1 and lim an = L2 , then we have to prove L1 = L2 .
n→∞ n→∞
Suppose that (an )n∈N is a sequence which is convergent with limit both L1 and L2 .
If L1 = L2 , then there is nothing to prove. So suppose for a contradiction that this is not the
case.
We choose ε = 13 ∣L2 − L1 ∣, which is positive since L2 ≠ L1 .
Because lim an = L1 , there is some N1 ∈ N such that if n > N1 then an is guaranteed to be
n→∞
within ε of L1 . And since lim an = L2 there is a (maybe different) N2 ∈ N such that if n > N2
n→∞
then an is guaranteed to be within ε of L2 .
Now pick an n which is bigger than both N1 and N2 (for example n = N1 + N2 + 1). Then we
have ∣an − L1 ∣ < ε and ∣an − L2 ∣ < ε. So by the triangle inequality, we have

∣L2 − L1 ∣ ≤ ∣L2 − an ∣ + ∣an − L1 ∣ < 2ε = 23 ∣L2 − L1 ∣ < ∣L2 − L1 ∣ .

But this is a contradiction—∣L2 − L1 ∣ cannot be smaller than itself.

CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 142

You should notice that this proof, where we use the assumption that a sequence tends to a
limit, looks nothing like proving that a sequence tends to a limit. We get to choose our favourite
ε > 0, and then we are given N1 and N2 . I want to stress that the choice we made, ε = 13 ∣L2 − L1 ∣,
isn’t ‘obvious’ at the point in the proof where we write ‘we choose...’. Again, if you were not
just reading this proof but trying to think it up, you’d leave a blank space here to fill in later,
once you see (at the second to last line) what you actually need: 2ε shouldn’t be bigger than
∣L2 − L1 ∣. If you think a bit, in fact ε = 12 ∣L2 − L1 ∣ would actually work as well (because we have
strict inequalities) but it is better to write something which ‘works easily’; you’re less likely to
make a mistake.
Again, it’s not too easy to see where this proof comes from just by looking at the algebra.
How did I find it? Well, I drew Figure 9.5, with the ‘first’ limit L1 :

L1 + ε

L1

L1 − ε

... ...
N1 N1 + 1 N1 + 2 N1 + 3

Then I drew in L2 on the y axis with its ε-sized band and N2 .

...
N2 N2 + 1
L2 + ε

L2
L1 + ε
L2 − ε
L1

L1 − ε

...
N1 N1 + 1 N1 + 2 N1 + 3

Now, what does this picture mean? In order for lim an = L1 to be true, all the points after N1
n→∞
have to be in the grey box, which they are. And for lim an = L2 to be true, all the points have
n→∞
to be in the hashed box after N2 . Which they are not in this picture—but they could have been:
they could have all been in the grey-hashed overlap. We want to get rid of that possibility—we
chose ε too big, we should choose a smaller value so the boxes don’t overlap:
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 143

...
N2 N2 + 1

L2 + ε
L2
L2 − ε
L1 + ε
L1
L1 − ε

...
N1 N1 + 1 N1 + 2 N1 + 3

And now we are happy: if N is at least as big as both N1 and N2 , for lim an = L1 to be true
n→∞
all the points an with n > N need to be in the grey box; but for lim an = L2 to be true all the
n→∞
points an with n > N need to be in the hashed box. That can’t be: the boxes don’t overlap!
The proof we saw really came from drawing this final picture. It tells us how we need to
choose ε > 0. Then N1 and N2 are given to us by the definition of ‘converges to L1 ’ and ‘converges
to L2 ’ respectively. Then the picture tells us to choose N = max(N1 , N2 ).
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 144

Let’s finally prove one important fact about limits of sequences: if all the terms of a convergent
sequence are contained in a closed interval [a, b], then so is the limit.

Theorem 9.12. Suppose [a, b] is any closed interval, and (xn )n∈N be a convergent sequence with
limit L, where xn ∈ [a, b] for all n ∈ N. Then L is also in [a, b].

Proof. We prove this theorem by contradiction. Suppose for a contradiction that L ∈/ [a, b]. Then
either L > b, or L < a.
Case 1: Suppose L > b. Choose ε = L−b2 , which is positive. Since (xn )n∈N converges to L, there
exists N ∈ N such that for all n > N we have

∣xn − L∣ < ε = L−b

2 so xn > L − ε = L+b
2 > b,

where the final inequality is since L > b. In particular, for n = N + 1 we have xn > b. But this is a
contradiction to our assumption xn ∈ [a, b].
The second case is very similar.
Case 2: Suppose L < a. Choose ε = a−L 2 , which is positive. Since (xn )n∈N converges to L, there
exists N ∈ N such that for all n > N we have

∣xn − L∣ < ε = a−L

2 so xn < L + ε = L+a
2 < a,

where the final inequality is since L < a. In particular, for n = N + 1 we have xn < a. But this is
a contradiction to our assumption xn ∈ [a, b].
In either case, we found a contradiction, so the theorem is proved.
It’s worth noticing that this theorem would be false for an open interval; we’ve already seen
that ( n1 )n∈N is a convergent sequence with limit 0. All the terms of this sequence are in (0, 2),
but the limit is not.

9.2.2 Writing analysis proofs: formal versus informal

There are two styles for writing analysis proofs.
One, the classical formal style, is: write down the choices of all the constants at the beginning,
and then complete the proof. This is what the proofs above did. Here is another example. We
want to prove the sequence (an )n∈N given by an = 1 + √n−π 1
tends to the limit 1.

Proof. Given ε > 0, let N = max (400, ⌈4ε−2 ⌉).

Suppose n > N . Then we have
RRR R R RRR RRR RRR RRR R R R
R
R 1 RRRR RRRR 1 R
R R
R 1 R
R R
R 1 RRRR RRRR 1 RRRR
∣an − 1∣ = RRR √ RR = RR √ RR ≤ RR √ n √ √
RR < RR n RR < RR √ RR = ε
RRR n − π RRRR RRRR 2 4 − π RRRR RRRR 4 + 100 − π RRRR RRRR 4 RRRR RRRR 4ε4−2 RRRR
n

which is what we wanted.

Let’s quickly notice where this long sequence of inequalities above comes from. The first
equality is just putting in the definition of an and noticing that the 1s cancel. For the next, we
just take a factor of 2 out of the square root. The reason we do this is that
√ we can notice that,
√n √
4 =
400
since N is bigger than 400, and n is bigger than N , so 4 is bigger than 100. So when
√n √
we replace a 4 with 100 in the denominator, we make the denominator smaller, and that
makes√the fraction bigger. Why do we do this? Well, because π is about 3, in particular √ it’s less
than 100 = 10. So again, we can make the denominator smaller by taking out the 100 − π,
and again that makes the fraction bigger. Finally, we notice that n is bigger than N which is
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 145

bigger than 4ε−2 , so (yet again!) we can replace n with 4ε−2 to make the denominator smaller
and the fraction bigger. And simplifying we have the ε we wanted: we proved ∣an − 1∣ < ε.
This proof works fine. It’s easy to check. It is not obvious how to think of it. We make a
‘magical’ choice of N at the beginning, and it just happens to be exactly what we need to get a
pretty ε at the end.
Of course, the truth is that I didn’t really make this choice of N at the beginning. I left a
blank space, and filled it in later after realising what I needed. Later on, we’ll see proofs which
are more complicated, and there you might have several ‘magical’ choices made at the beginning
of the proof.
Some people (and some textbooks too) prefer a more informal style, where we don’t choose
N at the beginning, but simply write down that it is to be chosen later and then write down
what we need as we go. Here’s the same proof, written that way.
Proof. Given ε > 0, we will choose an integer N later.
Suppose n > N . Then we have
RRR R R RRR
R
R 1 RRRR RRRR 1 RRR
∣an − 1∣ = RRR √ RRR = RRR √ n
RRR n − π RRR RRR 2 4 − π RRRRR
√ √ √ √
We will need N ≥ 400 in order to write 2 n4 − π < n4 + 100 − π < n4 . Putting this in, we get
RRR R
1 RR
∣an − 1∣ < RRRRR √ n RRRRR .
RRR 4 RRR

We need the right side of this to be less than ε, so we need N ≥ 4ε−2 for that to work. Putting
this in, we get
RRR R
R
R 1 RRRR
∣an − 1∣ < RRR √ R=ε
RRR 4ε−2 RRRRR
4

which is what we wanted. So we should choose N = max (400, ⌈4ε−2 ⌉).

This proof is a bit longer, but that’s fine. It is certainly more obvious why we made this
particular choice of N . In general, I don’t like ‘magical choices’ in proofs; it feels like the lecturer
showing off, and, worse, it probably makes some students think they will never be able to come
up with these proofs themselves (which is false).
However, I am going to stick to the formal style for the rest of these notes, and the reason is
that if you choose the informal style, you need to pay attention to the following.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 146

Warning 9.13. If you choose N as you go in a proof, you might end up writing something like
‘we choose N bigger than (7 − ε)(n − π)’. This looks all fine, and there certainly is an integer n
all over the place in the proof: why not, if it makes the inequalities come out the way you want?
Let’s try to prove that ( n−π
n
)n∈N tends to 7.
Proof. Given ε > 0, we will choose an integer N later.
Suppose n > N . When n > 5, we have 6n > 30 > 7π, so 6n − 7π > 0, so 7(n − π) > n, so n
n−π < 7.
We want this last inequality, so we will choose N > 5. Then we have
n n
∣ − 7∣ = 7 − .
n−π n−π
We should choose N bigger than (7 − ε)(n − π), because then we can make the numerator smaller
(and so the RHS becomes larger) by writing

n n N (7 − ε)(n − π)
∣ − 7∣ = 7 − <7− <7− = 7 − (7 − ε) = ε ,
n−π n−π n−π n−π
which is what we wanted. So we should choose N = max (5, ⌈(7 − ε)(n − π)⌉).
If you try writing this proof in the formal style, you’ll see something is fishy rather fast:
Proof. Given ε > 0, choose N = max (5, ⌈(7 − ε)(n − π)⌉).
Wait—what is n? Something is wrong.
When you read the formal proof (as when you read any proof) one thing you should be
thinking is: do I know what each quantity is as it comes? With ‘given ε > 0’ there is no problem;
that means that ε is allowed to be any positive real number, from this point on we fix one
particular choice, and the rest of the proof should work whatever positive real number it happens
to be. Then ‘choose N to be..’ means that we are trying to define a quantity N . We want it
bigger than 5; no problem. And bigger than a formula. Well, the formula contains ε—we know
what that is, we just fixed it. And it contains π—that’s about 3.14. And it contains n. What
is n? We haven’t seen it before, we don’t know what it is—how should we work out what this
formula is? ERROR ERROR! COMPUTER SAYS NO!
What is wrong is not just a formality; this is not me being picky for the sake of it. The
sequence ( n−π
n
)n∈N does not tend to 7, in fact (as you can convince yourself by working out a few
values, and as you will be able to prove easily by the end of this chapter) the sequence tends
to 1. If you want to use the informal style of writing a proof, you need to check that if you
would write it out in the formal style, then you wouldn’t ever try to use some letter in a formula
before you actually say what that letter is. If you’re trying to prove a sequence converges to a
limit, that means that when you choose N you can refer to ε, but not to n (or to anything that
depends on n!). Otherwise, you may ‘prove’ completely wrong statements, like the one above.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 147

9.3 Bernoulli’s inequality and the sequence (xn)n∈N.

In this section, we prove that, whenever ∣x∣ < 1, the sequence (xn )n∈N is convergent, with limit 0.
This is a basic, and perhaps “obvious” result, but we will want to use it repeatedly: let’s prove
it.
We proceed by first proving a useful result known as Bernoulli’s Inequality.

Theorem 9.14. For all real x ≥ −1 and all n ∈ N,

(1 + x)n ≥ 1 + nx.

Proof. We prove this result by induction on n. Note first that, for n = 1, the inequality states
that 1 + x ≥ 1 + x, which is certainly true.
Suppose now that, for some n ∈ N, (1 + x)n ≥ 1 + nx. Now we have

(1 + x)n+1 = (1 + x)(1 + x)n ≥ (1 + x)(1 + nx) = 1 + nx + x + nx2 ≥ 1 + (n + 1)x .

So the inequality holds for n + 1. Hence, by induction, the inequality holds for all n ∈ N.
You might wonder where we used the assumption x ≥ −1 in this proof. The answer is: we
multiplied the induction hypothesis (1 + x)n ≥ 1 + nx through by 1 + x and didn’t change the
direction of the inequality because x + 1 ≥ 0.
Now we use Bernoulli’s Inequality to show that, whenever ∣x∣ < 1, (xn )n∈N is convergent with
limit 0.

Theorem 9.15. Let x be any real number with −1 < x < 1. Then (xn )n∈N is a convergent sequence
with limit 0.

Proof. Given x with ∣x∣ < 1, we separate two cases.

First note that, if x = 0, then xn = 0 for every n ∈ N, and so certainly limn→∞ xn = 0. So we
may assume that x =/ 0.
In this case, we have that 1 < ∣x∣1
, and so h = ∣x∣
1
− 1 > 0. Now, by Bernoulli’s Inequality,
Theorem 9.14, for any natural number n we have
1
= (1 + h)n ≥ 1 + nh > nh,
∣x∣n

and so
1
0 ≤ ∣x∣n ≤.
nh
Remembering this inequality, let’s begin the proof of convergence.
Given any ε > 0, we take N = ⌈ εh
1
⌉. Now, for n > N , we have

1 1
∣xn − 0∣ = ∣x∣n ≤ < ≤ ε.
nh N h
Hence lim xn = 0.
n→∞

Let’s check quickly that we did not fall into the ‘trap’ of Warning 9.13. We chose N = ⌈ εh1
⌉.
That depends on ε, which is fine: we were given ε already. And it depends on h—what is h? We
1
defined it to be ∣x∣ − 1. What is x? That was given to us at the start of the proof too, so that’s
fine. What is important is that the choice of N we make doesn’t depend on n (because we don’t
know what n is yet, we only introduce it on the next line).
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 148
1
Example 9.16. Use Bernoulli’s Inequality to show that lim 2 n = 1.
n→∞
1 1 1
Proof. 2 > 1 and so 2 n > 1 (for otherwise 2 = (2 n )n ≤ 1, a contradiction). Let an ∶= 2 n − 1 ≥ 0.
Then Bernoulli’s inequality says 2 = (1 + an )n ≥ 1 + nan , and so
1
0 ≤ an ≤ .
n
Now, given any ε > 0, choose N = ⌈1/ε⌉. Then, for n > N ,

1 1 1
∣2 n − 1∣ = ∣an ∣ = an ≤ < ≤ ε.
n N
√ √ √ √
Therefore the sequence 2, 2, 3 2, 4 2, 5 2, . . . is convergent with limit 1.

I’ve so far been rather careful to make sure N is always chosen to be an integer (because
that’s what it is declared to be in Definition 9.3, the definition of convergence). This is why I’ve
put in these ⌈⋅⌉ symbols ‘the smallest integer at least..’.
However, it’s rather common to simply write ‘choose N ≥ 1/ε’ rather than ‘choose N = ⌈1/ε⌉’,
and leave it implicit that N is supposed to be an integer. I’ll be happy with you producing
either.
You might wonder why we don’t simply change the definition of convergence and allow N to
be any real number to avoid these issues. The answer to that is that it often is convenient to
assume N is a natural number in proofs; we can write things like aN and be assured that that
is actually a term of our sequence.

9.4 Bounded and monotone sequences

It is cumbersome to check from the definition if a sequence is convergent or not. Furthermore,
it’s very hard to use the definition to show that a sequence is convergent if we don’t already
know what the limit is. This is one reason why it’s hard to show sequence (vii) from Example 9.2
converges to a limit—we don’t have any idea what the limit might be, and looking at the terms
of the sequence doesn’t suggest any nice number that might be the answer (because the answer
is, as far as we know, not a nice number!).
In this section, we will study a condition under which we can conclude that a sequence is
convergent even without knowing its limit! We will prove that if a sequence is both ‘bounded’
as well as ‘monotone’, then it is always convergent.

Definition 9.17. A sequence (an )n∈N is said to be bounded if there exists a real number M > 0
such that
for all n ∈ N, ∣an ∣ ≤ M. (9.1)

Note that a sequence is bounded if and only if the set S = {an ∣ n ∈ N} is bounded (this is an
exercise).
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 149

Example 9.18.

(i) (1)n∈N is bounded, since ∣1∣ = 1 ≤ 1 for all n ∈ N.

(ii) ( n1 )n∈N is bounded, since ∣ n1 ∣ = 1

n ≤ 1 for all n ∈ N.

(iii) (1 + n1 )n∈N is bounded, since ∣1 + n1 ∣ = 1 + n1 ≤ 2 for all n ∈ N.

(iv) ((−1)n (1 + n1 ))n∈N is bounded, since ∣(−1)n (1 + n1 )∣ = 1 + n1 ≤ 2 for all n ∈ N.

Example 9.19. Show that the sequence (an )n∈N defined by

1 1 1 1
an = 1
+ 2 + 3 + ⋅⋅⋅ + n, n ∈ N
1 2 3 n
is bounded. (This is sequence (vii) from Example 9.2.)

Proof. We will prove that ∣an ∣ ≤ 3

2 is true for all n ∈ N. Since all the terms of this sequence are
positive, we can write:

∣an ∣ = an
1 1 1 1
= 1 + 2 + 3 + ⋅⋅⋅ + n
1 2 3 n
1 1 1 1
< 1 + 2 + 3 + ⋅⋅⋅ + n
1 2 2 2
1 1 1 1 1 1 1
= 1 + (1 − ) + 2 (1 − ) + ⋅ ⋅ ⋅ + n−1 (1 − )
1 2 2 2 2 2 2
1 1 1 1 1 1 1
= 1 + − 2 + 2 − 3 + 213 − 4 + ⋅ ⋅ ⋅ + n−1 − n
2 2 2 2 2 2 2
1 1
=1+ − n
2 2
3
< .
2
Thus all the terms are bounded by 32 , and so the sequence is bounded.
As usual, I did not know 32 would turn out to be an upper bound at the start of the proof;
I left a blank space and filled it in once I found it at the end. As with convergence, what’s
important is that whatever number I write for an upper bound has to be a number which does
not depend on n.
You might be a bit unhappy with the proof above. If so:
1 1
Activity 9.5. Write down a detailed proof that ∣an ∣ ≤ 1 + − using induction on n.
2 2n
Example 9.20. Show that the sequence (an )n∈N given by an = n for n ∈ N, is not bounded.

Proof. Given any M > 0, there exists an N ∈ N such that M < N (Archimedean property with
y = M and x = 1). Thus

¬[∃M > 0 such that for all n ∈ N, ∣an ∣ = ∣n∣ = n ≤ M ],

and so (n)n∈N is not bounded.

CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 150

convergent bounded
sequences sequences

Figure 9.6: All convergent sequences are bounded.

The sequences (1)n∈N , ( n1 )n∈N , (1 + n1 )n∈N are all convergent, and we have shown above that
these are also bounded. This is not a coincidence, and in the next theorem we show that the set
of all convergent sequences is contained in the set of all bounded sequences, as illustrated in
Figure 9.6.

Theorem 9.21. If a sequence is convergent, then it is bounded.

Proof. Let (an )n∈N be a convergent sequence with limit L. Set ε = 1. Then, using the definition
of convergence for this value of ε, we see that there exists N ∈ N such that, for all n > N ,

∣an − L∣ < 1.

Hence for all n > N ,

∣an ∣ = ∣an − L + L∣ ≤ ∣an − L∣ + ∣L∣ < 1 + ∣L∣.
Let M = max (∣a1 ∣, . . . , ∣aN ∣, 1 + ∣L∣). Then for all n ∈ N

∣an ∣ ≤ M

and so (an )n∈N is bounded.

The above theorem can be used to prove the divergence of a sequence. Indeed, in contrapositive
form it asserts that unbounded sequences are not convergent. Thus, one way to prove that a
sequence is divergent is to prove that it is unbounded.

Example 9.22. Show that the sequence (n)n∈N is divergent.

Proof. We have seen above that (n)n∈N is unbounded. It follows from Theorem 9.21 that (n)n∈N
is not convergent.
Keep in mind that some divergent sequences are not bounded, but some other divergent
sequences are bounded, such as sequence (iv) of Example 9.2.

Definition 9.23. A sequence (an )n∈N is said to be

monotonically increasing (or simply increasing) if for all n ∈ N, an ≤ an+1 ,

strictly increasing if for all n ∈ N, an < an+1 ,
monotonically decreasing (or simply decreasing) if for all n ∈ N, an ≥ an+1 ,
strictly decreasing if for all n ∈ N, an > an+1 ,
monotone if it is either monotonically increasing or monotonically decreasing.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 151

Thus a sequence (an )n∈N is monotonically increasing if a1 ≤ a2 ≤ a3 ≤ . . . ,

strictly increasing if a1 < a2 < a3 < . . . ,
monotonically decreasing if a1 ≥ a2 ≥ a3 ≥ . . . ,
and strictly decreasing if a1 > a2 > a3 > . . . .

Example 9.24.
monotonically strictly monotonically strictly
Sequence monotone?
increasing? increasing? decreasing? decreasing?
( n1 )n∈N No No Yes Yes Yes
(1 + n1 )n∈N No No Yes Yes Yes
((−1)n (1 + n1 ))n∈N No No No No No
(1)n∈N Yes No Yes No Yes
(n)n∈N Yes Yes No No Yes
( 111 + 1
22
+ 1
33
+ ⋅⋅⋅ + 1
)
nn n∈N Yes Yes No No Yes

The following theorem can be useful for showing that sequences converge without knowing
the limit beforehand (or with less work than using the definition).

Theorem 9.25. If a sequence is monotone and bounded, then it is convergent.

Again, to see how to prove this it helps to draw a picture. We’ll draw the picture for the
case ‘monotonically increasing’.

2
1.5
1
0.5
0
0 1 2 3 4 5 6 7

The black line at 2 is an upper bound for the sequence. But it doesn’t look like 2 is a good
candidate for the limit: it’s too big.
The dashed line at 1 isn’t an upper bound for the sequence. This proves 1 can’t be the limit,
it is too small (because the sequence is increasing, it can never go back down from above 1 to
get arbitrarily close to 1).
The dotted line at 1.5 looks like a better candidate for the limit. It is an upper bound for
the sequence, and this proves that 2 can’t be the limit: the sequence can never get above 1.5 so
it cannot get arbitrarily close to 2.
But if there is a smaller upper bound than 1.5 then that would prove 1.5 cannot be the limit.
What we are looking for is the least upper bound of the set {an ∣ n ∈ N}. Since this is a non-
empty, bounded set of real numbers, the least upper bound property says that L = sup {an ∣ n ∈ N}
exists. We just need to prove it is the limit. Let’s formalise that.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 152

Proof. We first consider the case that our sequence is monotone increasing, then do the monotone
decreasing case.
Let (an )n∈N be a monotonically increasing sequence. Since (an )n∈N is bounded, it follows that
the set
S = {an ∣ n ∈ N}
has an upper bound and so L = sup S exists. We show that in fact (an )n∈N converges to L.
Given ε > 0, we want to find N such that L − ε < an < L + ε for all n > N . The right-hand
inequality is going to be easy: by definition of ‘upper bound’, an ≤ L < L + ε is true for all n ∈ N.
So the difficulty is to get L − ε < an .
Because L is the least upper bound of S, it follows L − ε is not an upper bound of S. That
means there exists some N such that aN > L − ε.
But because the sequence is increasing, we have for any n > N the fact

L − ε < aN ≤ aN +1 ≤ aN +2 ≤ ⋅ ⋅ ⋅ ≤ an

which is what we wanted.

We should really write the ‘convergence’ part of this proof formally:
Given ε > 0, choose N such that aN > L − ε (because L − ε is not an upper bound of S). Then if
n > N we have
L − ε < aN ≤ aN +1 ≤ ⋅ ⋅ ⋅ ≤ an ≤ L < L + ε
so in particular L − ε < an < L + ε as desired.
Written like this, it’s clear the choice of N doesn’t depend on n: we don’t even mention n
until after we chose N .
Warning 9.26. Rather often, students read the above theorem as ‘if and only if’. This is false.
Any bounded monotone sequence is convergent, but in the other direction all we can say is that
a convergent sequence is bounded (Theorem 9.21). There are sequences which are convergent
(and so bounded) but not monotone, such as sequence (ix) of Example 9.2.

Example 9.27. The following table gives a summary of the valid implications, and gives
counterexamples to implications which are not true.
Question Answer Reason/Counterexample

Is every convergent sequence bounded? Yes Theorem 9.21

Is every bounded sequence convergent? No ((−1)n )n∈N is bounded,
but not convergent.
(−1)n
Is every convergent sequence monotone? No ( n )n∈N is convergent, but not
1 1
monotone: −1 < 2 and 2 > − 13 .
Is every monotone sequence convergent? No (n)n∈N is not convergent.
Is every bounded AND monotone Yes Theorem 9.25
sequence convergent?

The following activity is slightly tricky: think about how to put together the theorems you
already saw in order to prove it.

Activity 9.6. Let (an )n∈N be a sequence. We say it is a Cauchy sequence if the following is
true: For every δ > 0, there is an M ∈ N such that if m, m′ > M then we have ∣am − am′ ∣ < δ.
Prove that a sequence is convergent if and only if it is a Cauchy sequence.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 153

Let’s finally give an example of how to use Theorem 9.25.

Example 9.28. Prove that the sequence (an )n∈N defined by
1 1 1 1
an = + + + ⋅ ⋅ ⋅ + , n∈N
11 22 33 nn
is convergent.
Proof. We showed in Example 9.19 that this sequence is bounded. It is strictly increasing since
1
an+1 − an = >0
(n + 1)(n+1)
for all n ∈ N.
So by Theorem 9.25, this sequence is convergent.

9.4.1 Series
We won’t really talk about series in this course, but it is worth giving the definition.
Given real numbers a1 , a2 , . . . , when we write the ‘series’
∞
∑ an
n=1

what we mean is the limit of the sequence of partial sums, that is

∞ n
∑ an = n→∞
lim ∑ ai .
n=1 i=1

As we’ve seen, some sequences converge and other sequences diverge; that’s equally true for
sequences of partial sums. So some series converge (we can write down a real number which is
the ‘infinite sum’) and some diverge (the ‘infinite sum’ doesn’t make sense).
What Example 9.28 shows, in this language, is that
∞
1
∑ n
n=1 n

is a convergent series; it makes sense to say that this ‘infinite sum’ is a real number.
It would be reasonable to guess that there is perhaps some nice formula for the limit of this
series that lets us find out what it is more easily than adding up infinitely many terms. But
we don’t know any such formula. Not only that, we don’t even know if the limit is a rational
number or not! This is a long open problem: in 1697, Johann Bernoulli proved that
∞
1 1 1
∑ n
= ∫ dx ,
n=1 n 0 xx

but this doesn’t help us, either with calculating the limit or finding out if it is rational.
One can fairly easily (with a computer) find out what the first few (or few million) digits of
the limit are, and from this calculation we can show that if the limit is rational, then the fraction
p
q which is the limit needs to have a very large denominator: q has to be into the millions. So
our best guess is that no such fraction exists: probably the limit is irrational.
This is about all I want to say about series in this course. If you take MA203, you’ll return
to the topic there. The only thing I have left to say is a warning. There is a reason that we
invented the new word ‘series’ rather than just say ‘infinite sum’. The reason is that ‘infinite
sum’ sounds friendly and well-behaved. You can do all kinds of things in a sum, like rearrange
the terms (because addition is commutative).
Series are not friendly and well-behaved. If you rearrange the terms you get a different series,
which might have a completely different limit.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 154

9.5 Algebra of limits

It’s ‘obvious’ that if you have a couple of sequences (an )n∈N and (bn )n∈N , both of which are
convergent with limits respectively A and B, then (an + bn )n∈N should be a convergent sequence
too, with limit A + B. Why is it ‘obvious’ ? Well, if n is large then we are guaranteed ∣an − A∣ is
very close to 0, and ∣bn − B∣ is very close to 0, so ∣an + bn − A − B∣ must still be pretty close to 0
(by the triangle inequality).
The above is a good way to think about limits of sequences: try to keep in mind the intuitive
meaning, without having to write down all the quantifiers in the definition all the time. But you
also need to be able to go from this kind of intuitive idea to a formal proof, which means you
need to figure out how to use the definition with all the quantifiers.

Activity 9.7. Prove that if (an )n∈N and (bn )n∈N are both are convergent, with limits respectively
A and B, then (an + bn )n∈N is a convergent sequence too, with limit A + B.

However, once you have done this once, you will not learn much from doing the same thing for
(say) (an − bn )n∈N or (an bn )n∈N . The purpose of this section is to do that work for you. We’ll see
that a sequence which looks ‘complicated’ can often be broken down into ‘simple’ sequences by
algebraic operations like addition and subtraction, and we can find the limit of the complicated
sequences by doing the same algebra with the limits of the simple sequences (which we will
hopefully already know or be able to look up). This work-saving device is called the Algebra of
Limits.

Example 9.29. Show that the sequence (an )n∈N defined by

4n2 + 9
an =
3n2 + 7n + 11
converges to 34 .

Proof. We could do this by going back to the definition of convergence, and writing half a page
of algebra.
But it is much easier to write:
n2 (4 + n92 ) 4 + n92
an = = ,
n2 (3 + n7 + n112 ) 3 + n7 + n112

where the terms n92 , n7 , 11

n2 all have limit 0, and by a repeated application of Theorem 9.30 given
below, we obtain that
9 9
lim (4 +
) lim 4 + lim
n→∞ n 2 n→∞ n→∞ n2 4+0 4
lim an = = = = .
n→∞ 7 11 7 11 3 + 0 + 0 3
lim (3 + + 2 ) lim 3 + lim + lim 2
n→∞ n n n→∞ n→∞ n n→∞ n

Theorem 9.30 (Algebra of Limits). If (an )n∈N and (bn )n∈N are convergent sequences, then the
following hold:

(a) For all α ∈ R, (αan )n∈N is a convergent sequence and lim αan = α lim an .
n→∞ n→∞

(b) (∣an ∣)n∈N is a convergent sequence and lim ∣an ∣ = ∣ lim an ∣.

n→∞ n→∞

(c) (an + bn )n∈N is a convergent sequence and lim (an + bn ) = lim an + lim bn .
n→∞ n→∞ n→∞
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 155

(d) (an bn )n∈N is a convergent sequence and lim an bn = ( lim an ) ( lim bn ).

n→∞ n→∞ n→∞

k
(e) For all k ∈ N, (akn )n∈N is a convergent sequence and lim akn = ( lim an ) .
n→∞ n→∞

1 1
(f ) If for all n ∈ N, bn ≠ 0 and lim bn ≠ 0, then ( ) is convergent and moreover, lim =
n→∞ bn n∈N n→∞ bn
1
.
lim bn
n→∞

(g) For all k ∈ N, (an+k )n∈N is convergent and lim an+k = lim an .
n→∞ n→∞

√ √ √
(h) If for all n ∈ N, an ≥ 0, then ( an )n∈N is convergent and lim an = lim an .
n→∞ n→∞

That was a long theorem—which you should think of as good: there are lots of algebraic
operations you can do and you are guaranteed to get the right answer.
Before proving it (the proof comes in eight parts, so it is long, but no part is hard) it’s
probably best to highlight what the Algebra of Limits does not let you do.
Warning 9.31. The Algebra of Limits only works if the sequences (an )n∈N and (bn )n∈N are
convergent. If they are not, sometimes you will end up with a nonsensical answer (like ‘infinity
minus infinity’, and at least you know something is wrong. Sometimes you will get a nice real
number, but it happens to be the wrong real number.
The Algebra of Limits lets you add up (or subtract, multiply, et cetera) two sequences. By
using it repeatedly, you can also add up three sequences, or four, and so on. We’ll normally do
that without comment (as we did in Example 9.29). But let’s recall that (1)n∈N converges to 1,
and ( n1 )n∈N converges to 0. So can we write

lim 1 = lim1
+ ⋅ ⋅ ⋅ + n1 = lim n1 + ⋅ ⋅ ⋅ + lim n1 = 0 + ⋅ ⋅ ⋅ + 0 = 0 ..?
n→∞ n→∞ n
´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶ n→∞ n→∞ ´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹¸¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶
n times
´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶ n times
n times

Of course not, because that would say 1 = 0. The problem is the second equality, which looks
like the Algebra of Limits. It’s not. This is not a fixed number of sequences, and we’ve just seen
a way to misuse the Algebra of Limits to get the wrong answer. If you’re paying attention, you
will notice that the next two formula don’t make sense: what should the n under the bracket at
the bottom actually be? n is supposed to be some natural number, but which one?
What I mean by this is that in the first two formulae, n is a bound variable—that is, it is a
placeholder, it only makes sense inside of the ‘lim’ symbol. So for example, the formula lim n1
n→∞
means exactly the same as lim z1 , which means the same as ‘the limit of the sequence whose
z→∞
terms are 1, 12 , 13 , 14 , ...’. Any time you start a sequence of statements or equations with an n
(or some other letter) as a bound variable (inside a limit, or a quantifier) and at the end it’s
popped out to become a free variable (summing up n lots of zero) you can be fairly confident
that you have made a mistake.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 156

Proof of Theorem 9.30. Throughout this proof, we assume (an )n∈N and (bn )n∈N are convergent
sequences, and their limits are La and Lb , respectively.

(a) If α = 0, then lim αan = lim 0 = 0. So we can now assume α ≠ 0.

n→∞ n→∞
Let ε > 0 be given. By the definition of lim an = La , there exists an N ∈ N such that for all
n→∞
n > N,
ε
∣an − La ∣ < .
∣α∣
Then
ε
∣αan − αLa ∣ = ∣α∣ ∣an − La ∣ ≤ ∣α∣ = ε,
∣α∣
and (again by the definition of a convergent sequence) (αan )n∈N is convergent with limit
αLa , that is,
lim αan = αLa = α lim an .
n→∞ n→∞

(b) Given ε > 0, let N ∈ N be such that for all n > N ,

∣an − La ∣ < ε.

Then (as you will prove in an exercise) we have for all n > N :

∣∣an ∣ − ∣La ∣∣ ≤ ∣an − La ∣ < ε.

Hence (∣an ∣)n∈N is convergent with limit ∣La ∣, that is,

lim ∣an ∣ = ∣La ∣ = ∣ lim an ∣ .

n→∞ n→∞

(c) You should have already proved this, in Activity 9.7.

(d) Before beginning this, let’s quickly notice why it is a bit tricky. We want to argue that if
an is close to La , and bn is close to Lb , then an bn is close to La Lb .
The easiest way to do this is to argue in two steps: first, an bn is close to La bn , then second
La bn is close to La Lb . If we can do that, then the triangle inequality tells us an bn is close
to La Lb .
The second part is about the same as what we already did in (a), and we can copy the
proof over. For the first part, the difficulty is that if bn is huge, then an might be close to
La but still an bn is not very close to La bn . To deal with this, we use Theorem 9.21 to say
that (bn )n∈N is bounded, which gives us an upper bound on how huge bn can be.

First, since (bn )n∈N is a convergent sequence, by Theorem 9.21 it is bounded. Let M > 0
be1 such that ∣bn ∣ ≤ M for every n ∈ N.

1
Even if bn = 0 for all n, the definition of ‘bounded’ insists that we choose a bound which is strictly positive—for
example we could set M = 1 in this situation.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 157

Given ε > 0, we choose Na such that for all n > Na we have ∣an − La ∣ < 2Mε
, which we
can do since (an )n∈N converges to La . We choose Nb such that for all n > Nb we have
∣bn − Lb ∣ < 2∣Lεa ∣+1 . And finally we let N = max(Na , Nb ).
Now suppose n > N .
Step 1: We want to show ∣an bn − La bn ∣ < 2ε .
We have ∣an − La ∣ < ε
2M , so multiplying both sides by ∣bn ∣ we get

∣bn ∣∣an − La ∣ < 2M ∣bn ∣

ε
≤ ε
2M M = 2ε .

Since ∣bn ∣∣an − La ∣ = ∣an bn − La bn ∣ by Theorem 8.18, that’s what we wanted for Step 1.
Step 2: We want to show ∣La bn − La Lb ∣ < 2ε .
We have ∣bn − Lb ∣ < ε
2∣La ∣+1 , so multiplying both sides by ∣La ∣ we get

∣La ∣∣bn − Lb ∣ < 2∣La ∣+1 ∣La ∣

ε
= 2ε ⋅ ∣La∣L∣+1/2
a∣
< 2ε ,

which again is what we want for Step 2.

Putting the two Steps together, and using the triangle inequality (Theorem 8.18) we have

∣an bn − La Lb ∣ ≤ ∣an bn − La bn ∣ + ∣La bn − La Lb ∣ < 2ε + 2ε = ε ,

which is what we needed to verify to show lim an bn = La Lb .

n→∞
Remark 9.32. There are some more ‘magical choices’ in this proof. Why do we aim for < ε
2
in each of Steps 1 and 2? Well, because we’re going to add these up to get ε.
How did I know to ask for ∣an − La ∣ < 2M ε
; why this particular funny number on the right
hand side that (by the end of the proof) just turns out to be exactly what we need? The
answer is, of course, I didn’t know. I originally wrote ∣an − La ∣ < δ and2 did some algebra
with δ in it, until at the end of Step 1 I got to ∣an bn − La bn ∣ < δM . Then I saw that I should
choose δ = 2M ε
, and went back and replaced all my δs with this quantity.
Finally, how did I end up with ∣bn − Lb ∣ < 2∣Lεa ∣+1 ? Some logic similar to what I just said
(working things through with a δ) got me to the idea that 2∣Lε a ∣ looks good. But what if
La = 0? I don’t want to divide by zero, so add one to the denominator. That makes the
denominator bigger, so the fraction a bit smaller; it’s insisting that bn is a bit closer to Lb .
That can only help. It doesn’t really matter what we add; 1, or 42, or 0.001.

(e) This can be shown by using induction on k and from part (d) above. It is trivially true
with k = 1. Suppose that it holds for some k, then (akn )n∈N is convergent and
k
lim akn = ( lim an ) .
n→∞ n→∞

Hence by part (d) above applied to the sequences (an )n∈N and (akn )n∈N , we obtain that the
sequence (an ⋅ akn )n∈N is convergent and
k k+1
lim an akn = ( lim an ) ( lim akn ) = ( lim an ) ( lim an ) = ( lim an ) .
n→∞ n→∞ n→∞ n→∞ n→∞ n→∞

Thus (ak+1
n )n∈N is convergent and

k+1
lim ak+1
n = ( lim an ) .
n→∞ n→∞
2
delta, δ is another Greek letter traditionally used for small quantities.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 158

(f) This time, what could be tricky is if bn is very close to 0. To avoid that, let N1 ∈ N be such
that, for all n > N1 ,
∣Lb ∣
∣bn − Lb ∣ < ,
2
which we can do since (bn )n∈N converges to Lb .
For all n > N1 , we have

∣Lb ∣
∣Lb ∣ − ∣bn ∣ ≤ ∣∣Lb ∣ − ∣bn ∣∣ ≤ ∣bn − Lb ∣ < ,
2

and so ∣bn ∣ ≥ ∣Lb ∣

2 . This is our ‘bn is not close to 0’ guarantee.
Given ε > 0, let N2 ∈ N be such that for all n > N2 ,

ε∣Lb ∣2
∣bn − Lb ∣ < ,
2
which exists since (bn )n∈N converges to Lb . Now we let N = max{N1 , N2 }.
Suppose n > N . Then we have

1 1 ∣bn − Lb ∣ ε∣Lb ∣2 2 1
∣ − ∣= = ∣bn − Lb ∣ ⋅ ∣bn ∣−1 ∣Lb ∣−1 < = ε.
bn Lb ∣bn ∣ ∣Lb ∣ 2 ∣Lb ∣ ∣Lb ∣
1 1 1 1
So ( ) is convergent and lim = = .
bn n∈N n→∞ bn Lb lim bn
n→∞
ε∣Lb ∣2
Remark 9.33. Of course the funny number is something we got to by saying ‘choose
2
N2 such that if n > N2 then ∣bn − Lb ∣ < δ’, then doing algebra with δ (as in (d) to figure out
how we should choose δ). We get to

∣ b1n − L1b ∣ ≤ ∣bn − Lb ∣ ⋅ ∣bn ∣−1 ∣Lb ∣−1 < δ∣bn ∣−1 ∣Lb ∣−1

and we know we need the right hand side of this to be at most ε.

A standard mistake at this point is to choose δ = ε
2∣bn ∣∣Lb ∣ . What’s the problem?—the algebra
works.
The problem is that δ needs to be a quantity that doesn’t depend on n—so it can’t have
a bn in it. At the point where we first use it in the proof, there is no n around, we only
say what n should be on the next line, ‘Suppose n > N ’. This is again what we saw in
Warning 9.13.
When you’re figuring out a proof like this, and you get to needing δ∣bn ∣−1 ∣Lb ∣−1 ≤ ε, what
you should think is: whatever n > N we happen to be given, we need the left side to be
guaranteed small. This would be a problem if ∣bn ∣−1 was huge for some values of n—we
need to show that ∣bn ∣−1 cannot be too big, whatever n > N is. So we need to show ∣bn ∣
cannot be too close to 0.
And at this point (and not, despite the way it’s written above, at the start!) we realise that
we need N to be big enough that we can be sure ∣bn ∣ ≥ ∣L2b ∣ if n > N .

(g) This is homework.

CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 159

(h) Since an ≥ 0 for each n, from Theorem 9.12 we have La ≥ 0.

The case La = 0 is an exercise.
In the case La > 0, we use a technique called “rationalising the numerator”: note that
√ √
√ √ √ √ an + L a an − L a
an − La = ( an − La ) √ √ =√ √ .
an + L a an + La
√
Now, given ε > 0, choose N ∈ N so that, for n > N , ∣an − L∣ < ε La . Then we have, for
n > N, √
√ √ ∣an − La ∣ ε La
∣ an − La ∣ = √ √ ≤ √ = ε.
an + La La
√ √
Thus ( an )n∈N is convergent, with limit La .

Activity 9.8. Show the remaining part of (h), i.e. that if (an )n∈N is a sequence of nonnegative
reals, converging to 0, then
√
lim an = 0 .
n→∞

Example 9.34. Determine whether the following sequence is convergent and find its limit.

n2 − 24n3 + 3n4 − 12
( )
1 + 7n + 21n4 n∈N

1
Proof. By Activity 9.2 on page 138 we know that lim = 0. We now use Theorem 9.30 after
n→∞ n
first factorizing out n4 in both the numerator and denominator:

n2 − 24n3 + 3n4 − 12 n4 1
− 24
n + 3 − n4
12
lim = lim ⋅ n2
n→∞ 1 + 7n + 21n4 n→∞ n4
n4 + n3 + 21
1 7

2 4
( lim n1 ) − 24 lim 1
+ 3 − 12 ( lim n1 )
n→∞ n→∞ n n→∞
= 4 3
( lim n1 )+ 7 ( lim n1 ) + 21
n→∞ n→∞
0 − 24 ⋅ 0 + 3 − 12 ⋅ 04
2
=
04 + 7 ⋅ 03 + 21
3 1
= = .
21 7
Example 9.35. Determine whether the following sequence is convergent and find its limit.
2n + 3n + 1
( )
3n+1 + 3 n∈N
Proof. Divide numerator and denominator by the fastest growing term appearing, which is 3n ,
and use that limn→∞ xn = 0 for ∣x∣ < 1:

2n + 3n + 1 (2/3)n + 1 + (1/3)n
lim = lim
n→∞ 3n+1 + 3 n→∞ 3 + 3(1/3)n
lim (2/3)n + lim 1 + lim (1/3)n
n→∞ n→∞ n→∞
=
lim 3 + 3 lim (1/3)n
n→∞ n→∞
0+1+0 1
= = .
3+0 3
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 160

Remark 9.36. It follows from Theorem 9.30.(c) that if we have three convergent sequences
(an )n∈N , (bn )n∈N , (cn )n∈N , then their sum (an + bn + cn )n∈N is also convergent with limit

lim (an + bn + cn ) = lim an + lim bn + lim cn .

n→∞ n→∞ n→∞ n→∞

This is also true for the sum of four convergent sequences, the sum of five convergent sequences,
and by an easy induction proof, the sum of any fixed number of convergent sequences.
In general, we can apply any fixed number of algebraic operations via the Algebra of Limits,
as indeed we did in Example 9.35 (three additions and one division, all in one step).
But remember (Warning 9.31) that it doesn’t work (or make sense) for n (or anything not
fixed) sequences.

9.6 The Sandwich Theorem

Another theorem that is useful in proving that sequences are convergent and in determining
their limits is the so-called Sandwich theorem. Roughly speaking, it says that if a sequence is
sandwiched between two convergent limits with the same limit, then the sandwiched sequence is
also convergent with the same limit.
Theorem 9.37 (Sandwich theorem). Let (an )n∈N , (bn )n∈N be convergent sequences with the
same limit, that is,
lim an = lim bn .
n→∞ n→∞

If (cn )n∈N is a third sequence such that

for all n ∈ N, an ≤ cn ≤ bn ,

then (cn )n∈N is also convergent with the same limit, that is,

lim an = lim cn = lim bn .

n→∞ n→∞ n→∞

Proof. Let L denote the common limit of (an )n∈N and (bn )n∈N :

lim an = L = lim bn .
n→∞ n→∞

Given ε > 0, let N1 ∈ N be such that for all n > N1 , ∣an − L∣ < ε. Hence for n > N1 ,

L − an ≤ ∣L − an ∣ = ∣an − L∣ < ε,

and so L − an < ε, that is,

L − ε < an .
Still for the same given ε > 0, let N2 ∈ N be such that for all n > N2 , ∣bn − L∣ < ε. So for n > N2 ,
bn − L < ε, that is,
bn < L + ε.
Thus for n > N ∶= max{N1 , N2 }, we have

L − ε < an ≤ cn ≤ bn < L + ε,

and so L − ε < cn < L + ε. Consequently, cn − L < ε and −(cn − L) < ε, and so

∣cn − L∣ < ε.

This proves that (cn )n∈N is convergent with limit L.

CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 161

n
Example 9.38. Use the Sandwich theorem to show that lim = 0.
n→∞ 10n

Proof. It can be shown by induction that for all n ∈ N, n2 < 10n .

Consequently, we have
n n 1
0≤ n ≤ 2 = .
10 n n
1
Since lim 0 = 0 = lim , from the Sandwich theorem it follows that the sequence ( 10nn )n∈N is
n→∞ n→∞ n
convergent and
n
lim = 0.
n→∞ 10n
1 2 3 4
Thus the sequence 10 , 100 , 1000 , 10000 , . . . is convergent with limit 0.
1
Example 9.39. Use the Sandwich theorem to show that for any a, b ∈ R, lim (∣a∣n + ∣b∣n ) n =
n→∞
max(∣a∣, ∣b∣).

Proof. Without loss of generality, suppose max(∣a∣, ∣b∣) = ∣a∣. (That is, 0 ≤ ∣b∣ ≤ ∣a∣ holds.)
We have ∣a∣n ≤ ∣a∣n + ∣b∣n ≤ ∣a∣n + ∣a∣n = 2∣a∣n . Taking nth roots of this, we see that for all n,
1/n
∣a∣ ≤ (∣a∣n + ∣b∣n ) ≤ 21/n ∣a∣ .

Now (∣a∣)n∈N converges to ∣a∣, and (21/n ∣a∣)n∈N converges to ∣a∣ as well, by Example 9.16 and
the Algebra of Limits.
So using the Sandwich theorem, it follows that
1
lim (∣a∣n + ∣b∣n ) n = ∣a∣ = max(∣a∣, ∣b∣) .
n→∞

1
In particular, with a = 24 and b = 2005, we have that lim (24n + 2005n ) n = 2005, that is, the
n→∞
sequence
2029, 2005.1436, 2005.001146260873, . . .
is convergent with limit 2005.
n n n
Example 9.40. Show that lim ( + 2 + ⋅⋅⋅ + 2 ) = 1.
n→∞ +1 n +2
n2 n +n
n n
Proof. There are n terms in the sum: the smallest is 2 and the largest is 2 . Thus, for
n +n n +1
all n ∈ N, we have
n2 n n n n2
≤ + + ⋅ ⋅ ⋅ + ≤ ,
n2 + n n2 + 1 n2 + 2 n2 + n n2 + 1
and since
n2 n2
lim 2 = 1 = lim 2 .
n→∞ n + n n→∞ n + 1

it follows from the Sandwich theorem that

n n n
lim ( + 2 + ⋅⋅⋅ + 2 ) = 1.
n→∞ n2+1 n +2 n +n
This last is an example where you might want to try to use the Algebra of Limits somehow
to split up the limit of the sum of n terms into n separate limits. If you do that... you will get
zero as the answer, which is wrong. See (again) Warning 9.31.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 162

9.7 Subsequences and the Bolzano-Weierstrass theorem

In this section we prove an important result in analysis, known as the Bolzano–Weierstrass
theorem, which says that every bounded sequence has a convergent ‘subsequence’. We begin
this section by defining what we mean by a subsequence of a sequence.

Definition 9.41. Let (an )n∈N be a sequence and let (nk )k∈N be a strictly increasing sequence of
natural numbers. Then (ank )k∈N is called a subsequence of (an )n∈N .

Another way to think about this is: a subsequence is what you get from a sequence by
crossing out some terms (but not rearranging anything).

Example 9.42.

(i) ( 2n
1
)n∈N , ( n12 )n∈N , ( n!1 )n∈N and ( n1n )n∈N are all subsequences of ( n1 )n∈N .

(ii) Let pn be the n-th prime number. (Thus p1 = 2, p2 = 3, p3 = 5, p4 = 7, etc.) Then the
sequence (an )n∈N defined by an = p1n is a subsequence of ( n1 )n∈N .

(iii) The sequence

1
2, 1, 31 , 14 , . . .
is not a subsequence of ( n1 )n∈N .

(iv) The sequence ((−1)2n )n∈N , that is, the constant sequence

1, 1, 1, . . .

and the sequence ((−1)2n−1 )n∈N , that is, the constant sequence

−1, −1, −1, . . .

are both subsequences of ((−1)n )n∈N .

Theorem 9.43. If (an )n∈N is a convergent sequence with limit L, then any subsequence of
(an )n∈N is also convergent with the limit L.

Proof. Let (ank )k∈N be a subsequence of (an )n∈N .

Given ε > 0, let N ∈ N be such that for all n > N , ∣an − L∣ < ε.
Since 1 ≤ n1 < n2 < . . . and all the nk are integers, we have nk ≥ k for each integer k.
Suppose k > N . Then nk ≥ k > N , so ∣ank − L∣ < ε. This gives us, by definition,

lim ank = L .
k→∞

This theorem lets us build new convergent sequences from old ones; it also gives us a new
way to prove divergence of sequences.

Example 9.44.

(i) ( 2n
1
)n∈N , ( n12 )n∈N , (2−n )n∈N , ( n!1 )n∈N and ( n1n )n∈N are convergent sequences with limit 0.

(ii) The sequence ((−1)n )n∈N is divergent, since the subsequence 1, 1, 1, . . . has limit 1, while
the subsequence −1, −1, −1, . . . has limit −1.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 163

Let’s now state the Bolzano-Weierstrass theorem.

Theorem 9.45. (Bolzano3 –Weierstrass4 theorem.) Every bounded sequence has a convergent
subsequence.
It probably isn’t at all obvious why this is an interesting result at this point. But we will see
later that it is very useful.
There are several different ways to prove this theorem. We’ll give maybe the easiest, which
lets the theorems we already proved do most of the work.
The idea is the following. We start with an interval [−M, M ] which contains all the terms of
the sequence (which we can, because the sequence is bounded). We’ll write s1 = −M and t1 = M .
Then we look at the two intervals [−M, 0] and [0, M ], which together cover [−M, M ] = [s1 , t1 ].
One of these two intervals has to contain infinitely many terms of the sequence, call it [s2 , t2 ].
(It might be both; then pick the first half).
Then we similarly split [s2 , t2 ] into two halves, and pick one, [s3 , t3 ] that contains infinitely
many terms of the sequence. And so on.
Now (sk )k∈N is a monotone increasing bounded sequence. So by Theorem 9.25, this sequence
converges to a limit L. It’s not hard to believe (since the intervals get shorter and shorter) that
also the sequence (tk )k∈N converges to L.
Finally, we can construct the subsequence of (an )n∈N that we want. Choose n1 = 1, then an1
is in [s1 , t1 ]. Now choose n2 > n1 such that an2 is in [s2 , t2 ]. This is possible because infinitely
many terms of the sequence are in [s2 , t2 ]. And so on; in general we choose nk > nk−1 such that
ank is in [sk , tk ].
Now for each k ∈ N, we have ank ∈ [sk , tk ], i.e. sk ≤ ank ≤ tk . That means the sequences (sk )k∈N
and (tk )k∈N sandwich (ank )k∈N , and by the Sandwich Theorem, (ank )n∈N is convergent with limit
L—and we’re done!
Let’s now fill in the details of a formal proof.
Proof. Given a bounded sequence (an )n∈N , by definition there is a real number M > 0 such that
[−M, M ] contains each an with n ∈ N.
Step 1: We let s1 = −M and t1 = M , so infinitely many terms of (an )n∈N are in [s1 , t1 ] and
t1 − s1 = 2M .
We now start defining real numbers sk and tk such that infinitely many terms of (an )n∈N are
in [sk , tk ], and tk − sk = 22−k M , and sk−1 ≤ sk < tk ≤ tk−1 , recursively, for each k ≥ 2 as follows.
If there are infinitely many terms of (an )n∈N in [sk−1 , 12 (sk−1 + tk−1 )], we choose sk = sk−1 and
tk = 12 (sk−1 + tk−1 ).
Otherwise, there are only finitely many terms of (an )n∈N in [sk−1 , 12 (sk−1 + tk−1 )], but there
are infinitely many terms in [sk−1 , tk−1 ], so there must be infinitely many terms of (an )n∈N in
[ 12 (sk−1 + tk−1 ), tk−1 ]. So we choose sk = 21 (sk−1 + tk−1 ) and tk = tk−1 .
Note that in both cases, tk −sk = 21 (tk−1 −sk−1 ) = 21 ⋅22−(k−1) M = 22−k M and sk−1 ≤ sk < tk ≤ tk−1 .
Step 2: By construction, the sequence of real numbers (sk )k∈N is monotone increasing, and it is
bounded (with bound M ). So by Theorem 9.25, there is a number L such that limk→∞ sk = L.
We now show that also limk→∞ tk = L.
Given ε > 0, let K1 ∈ N be such that for all k > K1 we have ∣sk − L∣ < 2ε ; this exists since
limk→∞ sk = L. Let K2 ∈ N be such that for all k > K2 we have 22−k M < 2ε ; this exists since
limk→∞ 2−k = 0 (Example 9.44) and the Algebra of Limits. Now let K = max(K1 , K2 ).
Suppose k > K. Then we have

∣tk − L∣ ≤ ∣tk − sk ∣ + ∣sk − L∣ < 22−k M + 2ε < 2ε + 2ε = ε

3
Bernhard Bolzano (1781–1848)
4
Karl Weierstrass (1815–1897)
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 164

which is what we need for the definition of limk→∞ tk = L.

Step 3: Let n1 = 1. Note that an1 ∈ [s1 , t1 ] = [−M, M ] by definition of M . Now for each k ≥ 2,
recursively, we choose nk > nk−1 such that ank ∈ [sk , tk ]. For each k, we can do this because there
are infinitely many terms of (an )n∈N within [sk , tk ], and at most nk−1 of them have index nk−1
or smaller.
By definition, we have sk ≤ ank ≤ tk for each k ∈ N. So by the Sandwich Theorem (Theo-
rem 9.37) we have limk→∞ ank = L, and this is our desired convergent subsequence.
√
Example 9.46. Consider the sequence (an )n∈N of fractional parts of integral multiples of 2,
defined by √ √
an = n 2 − ⌊n 2⌋, for n ∈ N,
where, for x ∈ R, ⌊x⌋ is the floor function of x. Show that this sequence has a convergent
subsequence.
The terms of the sequence (an )n∈N are as follows:
√
2 = 1.414213 . . .a1 = 0.414213 . . .
√
2 2 = 2.828427 . . .a2 = 0.828427 . . .
√
3 2 = 4.242640 . . .a3 = 0.242640 . . .
√
4 2 = 5.656854 . . .a4 = 0.656854 . . .
√
5 2 = 7.071067 . . .a5 = 0.071067 . . .
√
6 2 = 8.485281 . . .a6 = 0.485281 . . .
⋮

Proof. The sequence (an )n∈N is bounded: indeed, 0 ≤ an < 1 for every n ∈ N. So by the Bolzano–
Weierstrass theorem this sequence has a convergent subsequence.
We have seen that if (an )n∈N is convergent with limit L, then any subsequence also converges
to L (Theorem 9.43). We’ve also seen examples of divergent sequences (an )n∈N for which there
are exactly two limit points, that is numbers p ∈ R such that there is a subsequence of (an )n∈N
converging to p. Both sequences (iii) and (iv) from Example 9.2 have limit points 1 and −1, and
nothing else. It’s easy to give (in both cases) a subsequence which converges to 1, and another
one that converges to −1. Why are there no other limit points?
Activity 9.9. Show that, if p ∈ R is not equal to 1 or −1 then there is no subsequence of either
((−1)n )n∈N or of ((−1)n (1 + 1/n))n∈N that converges to p.
A sequence which is not bounded doesn’t have to have a convergent subsequence (because
the condition of the Bolzano-Weierstrass theorem isn’t satisfied).
Activity 9.10. Find a sequence (an )n∈N which is not bounded and which has no convergent
subsequence.
Find another sequence (bn )n∈N which is not bounded and which does have a convergent
subsequence.
It’s not too hard to come up with sequences which have two, or three, or ten, different limit
points. But there can be many more.
Activity 9.11. Show that for any real number x and any ε > 0, there are infinitely many rational
numbers in the interval (x − ε, x).
Suppose that (an )n∈N is any sequence such that every rational number is a term of the sequence.
Prove that for every real number x, there is a subsequence of (an )n∈N which converges to x.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 165

You could reasonably object to the above: but maybe there isn’t any such sequence (an )n∈N ?
But there are in fact such sequences. Here is an example. For any rational number pq written
in lowest terms (i.e. q is positive, and p and q have no common factor bigger than 1) say the
weight of pq is ∣p∣ + q. For any w ∈ N, there are at most 2w rational numbers of weight w: we have
to choose 1 ≤ q ≤ w, and then we are left with two possibilities, either p = w − q or p = −(w − q).
So we can make a sequence listing all the rational numbers by first writing down all the ones
with weight 1 (there is only one, 01 ) and then weight 2, weight 3, and so on.
In fact, the sequence from Example 9.46 has a similar property: for any x ∈ [0, 1], there is a
subsequence converging to x. This is rather harder to prove, though!

9.7.1 Non-examinable: another proof of Bolzano-Weierstrass

The proof we gave of the Bolzano-Weierstrass theorem is not the only way to prove it. Another
very natural approach is to notice that if (an )n∈N is bounded, so is any subsequence. And we
already know that any monotone bounded sequence converges—so it is enough to prove the
following.

Theorem 9.47. Every sequence has a monotone subsequence.

The proof of this splits up into the following two activities.

Activity 9.12. Given a sequence (an )n∈N , let n1 be an index such that an1 = max(a1 , a2 , . . . ), if
it exists. Let n2 be an index such that an2 = max(an1 +1 , an1 +2 , . . . ), if it exists, and so on: given
k ≥ 3 and nk−1 , let nk be an index such that ank = max(ank−1 +1 , ank−1 +2 , . . . ) if it exists.
Prove that either we obtain a monotone decreasing subsequence (ank )k∈N of (an )n∈N , or there
is some K ∈ N such that (aK , aK+1 , aK+2 , . . . ) has no biggest term.

Activity 9.13. Given a sequence (bn )n∈N which has no biggest term, let n1 = 1 and for each
k ≥ 2, given nk−1 , let nk be the smallest index such that bnk > bnk−1 , if it exists. Prove that this
gives a subsequence (bnk )k∈N which is strictly increasing.

Finally—and this is easier—put the pieces together.

Activity 9.14. Use the statements you proved in Activities 9.12 and 9.13 to prove Theorem 9.47.
Use Theorems 9.25 and 9.47 to prove Theorem 9.45.

9.8 Sample exercises

Exercise 9.1. Prove, directly from the definition, that the sequence ( 3n−1 )
n+2 n∈N is convergent,
and find its limit.

Exercise 9.2. (a) Let (an )n∈N be a convergent sequence with limit L, and let M be some real
number with M =/ L. Show that the set {n ∈ N ∣ an = M } is bounded above.

(b) Prove that the sequence ((−1)n )n∈N is divergent.

Exercise 9.3. Use the definition of limit to prove directly that 1 is not a limit of the se-
quence (1/n)n∈N .

Exercise 9.4. In each of the cases listed below, give an example of a divergent sequence (an )n∈N
that satisfies the given conditions.

(a) For every ε > 0, there exists an N such that, for infinitely many n > N , ∣an − 1∣ < ε.
 CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 166

(b) There exists an ε > 0 and an N ∈ N such that for all n > N , ∣an − 1∣ < ε.

Exercise 9.5. A sequence (an )n∈N is said to be a Cauchy sequence if for every ε > 0, there
exists an N ∈ N such that for all n, m > N , ∣an − am ∣ < ε.
Show that every convergent sequence is Cauchy.
Hint: ∣an − am ∣ = ∣an − L + L − am ∣ ≤ ∣an − L∣ + ∣am − L∣.

Exercise 9.6. Let (an )n∈N be a sequence defined by

2n + 1
a1 = 1 and an = an−1 for n ≥ 2.
3n
Prove that (an )n∈N is convergent.

Exercise 9.7. Suppose that the sequence (an )n∈N is bounded. Prove that the sequence (cn )n∈N
defined by
a3 + 5n
cn = n2
an + n
is convergent, and find its limit.

Exercise 9.8. Recall the definition of a Cauchy sequence from Exercise 9.5, where we have
already seen that every convergent sequence is Cauchy. Use the Bolzano–Weierstrass theorem to
prove the converse: if a sequence is Cauchy, then it is convergent.
Hint: Proceed as follows. Let (an )n∈N be a Cauchy sequence. Show that (an )n∈N is bounded. By
the Bolzano–Weierstrass theorem, it follows that (an )n∈N has a convergent subsequence, say
(ank )k∈N with limit L. Prove (using the fact that (an )n∈N is Cauchy), that then (an )n∈N is itself
convergent with limit L.

Exercise 9.9. Sample Exam Question. 2007 Q5.

(a) What does it mean to say that a sequence (an )n≥1 is convergent? Use this definition to show that
if (an )n≥1 is convergent, then (an+1 )n≥1 is also a convergent sequence and lim an = lim an+1 .
n→∞ n→∞

Let b be a real number with 2 < b < 3. We define a sequence (bn )n≥1 by

b1 = b and bn+1 = b2n − 4bn + 6 for every n ∈ N.

(b) Show that 2 < bn < 3 for every n ∈ N.

(c) Prove that (bn )n≥1 is a monotone sequence.

(d) Explain why lim bn exists and find its value.

n→∞

(e) Let S = {bn ∣ n ∈ N}. Find sup S, inf S, max S, min S. Justify your answers.

9.9 Comments on selected activities

Comment on Activity 9.1.

(i) For this, you should find that given ε > 0, choosing N = ⌈1/ε⌉, or anything bigger, will
work. If your N is smaller, then your proof is wrong.

(ii) As (i).
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 167

(iii) You need to consider all possible values of L ∈ R, and rule out all of them. If you only
consider L = 1 and L = −1, the ‘obvious’ limits, then you haven’t shown that (for example)
this sequence doesn’t converge to 0. Whatever L is, you should find that the definition of
convergence, with ε = 1, fails. Any smaller ε will also fail (but you only need to show that
some one ε is a counterexample). If you tried to use ε > 1, then your proof will not work
for L = 0.

(iv) As (iii).

(v) For any ε > 0, you can simply choose N = 1. But if you wrote something bigger (e.g. that
N = ⌈ 1ε ⌉) your proof works, it’s just a bit more complicated than necessary.

(vi) This is a bit tricky. Given L ∈ R, we want to rule out that (n)n∈N converges to L. We will
use ε = 1. This is a counterexample to the definition of convergence to L for the following
reason.
Whatever N is given, it is not true that for all n > N we have ∣n − L∣ < 1. Indeed, we can
choose n = max(N + 1, ⌈∣L∣⌉ + 1). Now by definition we have n > N , and by definition we
have n ≥ ∣L∣ + 1, so ∣n − L∣ ≥ 1.

(vii) This is really the ‘wrong time’ to try to prove that this sequence is convergent; we will
prove it after we develop some tools that help us find when sequences are convergent.

(viii) Again, this is really the wrong time to look at this sequence. The idea is that (eventually!)
the exponential will tend to zero much faster than the polynomial grows, so it will win
in the long run. But formally proving this needs a bit of algebra, and it helps to know
Bernoulli’s Inequality.

(ix) For this, you should find that given ε > 0, choosing N = ⌈4/ε⌉, or anything bigger, will work.
If your choice of N tries to take into account whether N is odd or even, then probably
your proof is wrong. If it takes into account whether n is odd or even, then your proof is
definitely wrong: see Warning 9.13.

Comment on Activity 9.2. Given ε > 0, we choose N = ⌈ 1ε ⌉.

Given n > N , we have
∣ n1 − 0∣ = n1 < N1 ≤ ε ,
which is what we wanted to show for the definition of convergence to 0.
If your N isn’t as big as the one above, your proof is wrong. If there is an n in your definition
of N , your proof is wrong.
If your N is bigger than the one above, then most likely things are fine.
Comment on Activity 9.3. If there does not exist any N ∈ N such that ‘ ∀n > N , ∣an − L∣ < ε’
then that means that for any given N ∈ N, there is some n > N such that ∣an − L∣ ≥ ε.
We will use this assertion infinitely many times, for different choices of N .
First (with N = 1) there is some integer, which we call n1 , such that n1 > 1 and ∣an1 − L∣ ≥ ε.
Now (with N = n1 ) there is some integer n2 , such that n2 > n1 and ∣an2 − L∣ ≥ ε.
Recursively, suppose for some k ≥ 2 we have integers n1 < n2 < ⋅ ⋅ ⋅ < nk . Then (with N = nk )
there is some integer nk+1 such that nk+1 > nk and ∣ank − L∣ ≥ ε.
The set {n1 , n2 , . . . } is infinite. This is because there is an (obvious!) bijection between this
set and N, namely nk ↔ k for each k ∈ N, and we already proved that N is infinite. To check
this bijection is well-defined (i.e. it really is a bijection) we need to observe that if k < k ′ are in
N, then they are in correspondence with two different members of {n1 , n2 , . . . }, namely nk and
nk′ , which by construction satisfy nk < nk′ .
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 168

Comment on Activity 9.4. See Activity 9.1(vi).

Comment on Activity 9.5. We have an = ∑nk=1 k1k . We want to prove by induction that an ≤ 32 − 21n .
The base case is a1 ≤ 32 − 211 , which we check by calculation is true (with equality).
Suppose as an induction hypothesis that for some given s ∈ N we have as ≤ 32 − 21s .
Then we have

as+1 = as + (s+1)
1
s+1 ≤ 2 − 2s + (s+1)s+1 ≤ 2 − 2s + 2s+1 = 2 − 2s+1 ,
3 1 1 3 1 1 3 1

which is what we wanted for the induction step. By induction, we conclude the desired inequality
holds for all n.
Comment on Activity 9.7. Suppose lim an = A and lim bn = B.
n→∞ n→∞
Given ε > 0, let Na be such that for all n > Na we have ∣an − A∣ < 2ε , and let Nb be such that
for all n > Nb we have ∣bn − B∣ < 2ε . Both Na and N − b exist by the definition of convergence. We
choose N = max(Na , Nb ).
Suppose n > N . Then we have

∣(an + bn ) − (A + B)∣ = ∣(an − A) + (bn − B)∣ ≤ ∣an − A∣ + ∣bn − B∣ < 2ε + 2ε = ε ,

which is what we wanted to prove.

Comment on Activity 9.8. Suppose (an )n∈N is a sequence of nonnegative reals converging to 0.
√
We want to show ( an )n∈N converges to 0.
Given ε > 0, there exists N ∈ N such that for all√n > N we have ∣an ∣ < ε2 . Taking square
roots
√ of both sides, we see that for all n > N we have ∣an ∣ < ε, and since an is positive we have
√ √
∣an ∣ = ∣ an ∣. So for all n > N we have ∣ an − 0∣ < ε, which is what we wanted.
Comment on Activity 9.9. This is a little bit tricky because you have to consider all possible
subsequences, and you don’t know what these might look like. However, the point is that if
(ank )k∈N converges to some p ∈ R, then for any ε > 0, from some point on all the terms of the
subsequence are within ε of p. This is infinitely many terms. So those infinitely many terms
have to also be present in the original sequence:
If (an )n∈N has a subsequence converging to p ∈ R, then for every ε > 0 there exist infinitely
many terms of (an )n∈N which are in (p − ε, p + ε).
And now you just need to figure out how to choose ε, given p which is not 1 or −1, such that
(p − ε, p + ε) doesn’t contain infinitely many terms of the given sequence. One choice that will
work is ε = ∣∣p∣ − 1∣.
Comment on Activity 9.10. For (an )n∈N , one choice that works is (n)n∈N . We’ve already seen
that this sequence isn’t bounded, and almost exactly the same proof shows no subsequence is
bounded either, so it can’t have a convergent subsequence.
For (bn )n∈N , we can take bn equal to either n (if n is odd) or zero (if n is even). This clearly
isn’t bounded, but the subsequence of even-numbered terms converges to 0.
Comment on Activity 9.11. If x is rational, then x − n1 is rational for each n ∈ N and (if n > 1/ε)
bigger than x − ε.
If x − ε is rational, then similarly x − ε + n1 is rational for each n ∈ N and if n > 1/ε smaller
than x.
So suppose x and x − ε are both irrational: then both of them have a unique decimal
representation (i.e. neither finishes with an infinite string of 9s). Since x − ε < x, there must be a
first decimal place where the expansions differ. Let y be the number we get by terminating the
decimal expansion of x at that place. Then y is by construction smaller than x, and larger than
x − ε, and it is rational.
Now y + n1 is rational and smaller than x for all n > x−y 1
.
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 169

In each case, we found the desired infinitely many rational numbers between x and x − ε.
Now suppose that (an )n∈N is a sequence whose terms contain all the rational numbers. Given
any real number x and any ε > 0, we just proved that there are infinitely many terms of the
sequence in (x − ε, x).
In particular, we can construct a subsequence (ank )k∈N as follows. Let n1 be any index such
that x − 1 < an1 < x (using the above observation with ε = 1, such an index exists).
Now for each k ≥ 2 in turn, given nk−1 , look at all the terms of (an )n∈N which are in (x − k1 , x).
There are infinitely many, and only finitely many have an index less than or equal to nk−1 . So
we can choose nk > nk−1 such that x − k1 < ank < x.
The sequence (ank )k∈N is sandwiched by (x − k1 )k∈N and (x)k∈N , so by the Sandwich Theorem
we have limk→∞ ank = x.
Comment on Activity 9.12. If at each stage in the construction the ‘if it exists’ is true, then we
obtain a subsequence (ank )k∈N . By construction an1 is the biggest term in the entire sequence;
in particular it is at least as big as all the following terms, so whatever n2 > n1 we choose we
will have an1 ≥ an2 . Similarly, for each k ≥ 2, when we choose ank it is at least as big as all of
the following terms, and in particular whatever n3 we choose we get ank ≥ ank+1 . So this is a
monotone decreasing subsequence.
What is left is the possibility that at some stage k in the construction the ‘if it exists’ is
false. That is, (ank−1 +1 , ank−1 +2 , ank−1 +3 , . . . ) has no maximum element. Letting K = nk−1 + 1, that
is precisely saying that (aK , aK+1 , . . . ) has no biggest term.
Comment on Activity 9.13. The point here which you need to see is the following. If for some
k ≥ 1 we have followed the construction up to stage k—that is, we have constructed nk —then
we know that bnk is not a biggest element in the sequence; there is certainly, somewhere in the
sequence, a bigger element. But we need to be sure that there is a bigger element which comes
after bnk . This is the reason for choosing the smallest index every time: we know bnk is (by
construction) bigger than all the terms from bnk−1 to bnk −1 , and so (by an induction argument)
it is the biggest term in {b1 , b2 , . . . , bnk }. So whatever the bigger term is that we know exists, it
has to come after bnk .
This shows that the ‘if it exists’ will always be true—there will always be such a term—and
then by construction we get a strictly increasing subsequence.
Comment on Activity 9.14. The proof of Theorem 9.47 is now the following. By Activity 9.12,
either (an )n∈N has a monotone decreasing subsequence, or it has a subsequence which has no
biggest term. But then by Activity 9.13, that subsequence has a subsequence which is strictly
increasing—and this is a subsequence of a subsequence of (an )n∈N , so by definition it is a
subsequence of (an )n∈N . Either way, we found a monotone subsequence.
Now to prove the Bolzano-Weierstrass theorem, suppose (an )n∈N is a bounded sequence. Then
any subsequence is also bounded, and by Theorem 9.47 there is a monotone subsequence (ank )k∈N .
Now (ank )k∈N is a bounded monotone sequence, so by Theorem 9.25 it is convergent.

9.10 Solutions to exercises

Solution to Exercise 9.1.
The first thing to do here is to figure out what the limit is. A thought process is “for large n,
the numerator is about 3n and the denominator is about n, so the fraction is about 3”. Hence
we expect 3 to be the limit. Let’s now prove it.
Given ε > 0, we set N to be some natural number such that . . . (leave this blank for the
moment, and come back and fill it in later).
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 170

Now, for n > N , we have

3n − 1 3n − 1 − 3n − 6 7 7
∣ − 3∣ = ∣ ∣= < .
n+2 n+2 n+2 N
We now want to finish off by deducing that 7/N < ε. So what do we want from N ? Evidently, it
suffices to take N > 7/ε. So now we go back and fill in: “N to be some natural number such
that N > 7ε.” Then we can indeed conclude that, for n > N , ∣ 3n−1
n+2 − 3∣ < N < ε. So the sequence
7

tends to 3, as claimed.
Solution to Exercise 9.2.
(a) How are we going to use the fact that the sequence (an )n∈N converges to L? One idea you
might have is to say that, for large enough N , the elements of the sequence have to be “close
to L”, so that they cannot be equal to M . How close should close be: closer than the distance
∣L − M ∣ between L and M .
So set ε = ∣M − L∣. By the definition of a limit, there is some N ∈ N such that, for n > N ,
∣an − L∣ < ∣M − L∣, and that certainly implies that an =/ M .
That means that, if n has the property that an = M , then n ≤ N . In other words, N is an
upper bound for the set {n ∈ N ∶ an = M }, as required.
(b) We can use part (a): that’s the point. Suppose the sequence does tend to a limit, and call
the limit L. If L =/ 1, then {n ∈ N ∣ (−1)n = 1}, which is the set of even numbers, is bounded above.
That is false, which leaves only the possibility that L = 1. But then the set {n ∈ N ∣ (−1)n = −1},
which is the set of odd numbers, is bounded above, which again is false. We conclude that the
sequence does not converge to any limit, i.e., it is divergent.
Solution to Exercise 9.3.
To show that the sequence (an )n∈N does not tend to the limit L, we need to show that there
exists ε > 0 such that, for all N ∈ N, there is some n > N with ∣an − L∣ ≥ ε.
We can take ε = 1/2 here. Now, whatever N ∈ N is proposed, take n > max(2, N ). We see
that an < 1/2, and so ∣an − 1∣ > 1/2. Thus indeed an does not converge to 1.
Solution to Exercise 9.4.
(a) Take, for instance, the sequence (an )n∈N where an = 1 for n odd, and an = 0, for n even.
Given any ε > 0, take N = 1: then ∣an − L∣ = 0 < ε for all odd n > 1.
(b) Take the same sequence (why not?), and take ε = 2 and N = 1. Then indeed ∣an − L∣ < 2
for all n > 1.
[A sequence satisfies (a) if L = 1 is a limit point of the sequence. A sequence satisfies (b)
(whatever L is) if and only if the sequence is bounded.]
Solution to Exercise 9.5.
Let (an )n∈N be a convergent sequence, with limit L.
The idea is that, if we go far enough down the sequence, all the terms beyond that point are
close to L, and so they are all close to each other. More precisely (this is what the hint tells us),
if all the terms are within ε/2 of L, then the distance between any two terms is at most ε. We
now make this idea into a formal proof.
Take any ε > 0. As (an )n∈N converges to L, there is some N ∈ N such that, for n > N ,
∣an − L∣ < ε/2. Hence, for any n, m > N , we have (as in the hint)
ε ε
∣an − am ∣ ≤ ∣an − L∣ + ∣am − L∣ < + = ε.
2 2
Hence the sequence (an )n∈N is a Cauchy sequence, as claimed.
Solution to Exercise 9.6.
You might want to generate a few terms of the sequence (an ) in order to get a feel for what
is going on. Whether or not this helps, you need to have the idea that the terms of the sequence
 CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 171

(an ) get smaller, while staying positive, and so that perhaps the sequence is decreasing and
bounded below. In fact, you might (correctly) suspect that the sequence converges to 0, but
you’re not asked to prove that.
Indeed, it is clear that since the first term a1 is positive, and each subsequent term is a
positive multiple of the previous term, all terms are positive. (By now, you should be confident
that you can write a formal induction argument if you really needed, but as far as I’m concerned
this will suffice here.)
Also, since 2n + 1 ≤ 3n for all n ≥ 2, it follows that an ≤ an−1 for all n ≥ 2, in other words that
(an ) is a decreasing sequence, bounded below by 0. Therefore the sequence is convergent.
Solution to Exercise 9.7.
There are various ways to tackle this exercise. But they have one thing in common: the first
thing to do is to understand how the sequence (cn )n∈N behaves for large n. Since the sequence
(an )n∈N is bounded, say with ∣an ∣ ≤ M for every n ∈ N, then also ∣a3n ∣ ≤ M 3 and a2n ≤ M 2 . So the
numerator “behaves like” 5n, while the denominator “behaves like’ n, in the sense that the other
terms are “of smaller order” for large n. So we expect the limit to be equal to 5.
Here are three ways to show this.
a3n −5a2n
(a) Write cn − 5 = a2n +n , so
∣a3n − 5a2n ∣ M 3 + 5M 2
∣cn − 5∣ ≤ ≤ .
n n
Now, given ε > 0, we can choose N > (M 3 + 5M 2 )/ε, so that, for n > N , we have
M 3 + 5M 2 M 3 + 5M 2
∣cn − 5∣ ≤ ≤ < ε.
n N

(b) Use the Algebra of Limits. Write

5 + a3n /n
cn = .
1 + a2n /n
Now show that a3n /n and a2n /n both tend to zero as n → ∞. One way to do this is to use the
Sandwich Theorem: −M 3 /n ≤ a3n /n ≤ M 3 /n, and by the Algebra of Limits the two sandwiching
sequences tend to zero as n → ∞, so a3n /n also does; similarly for a2n /n. Then we have that
3
limn→∞ cn = 5+lim n→∞ an /n
1+limn→∞ a2n /n = 5.

(c) Use the Sandwich Theorem directly Note that

−M 3 + 5n M 3 + 5n
≤ cn ≤ .
n n
Now argue that both the sequences on the outside tend to 5, and hence so does (cn )n∈N .
Solution to Exercise 9.8.
We naturally follow the hint. Let (an )n∈N be a Cauchy sequence. To show that it is bounded,
we apply the definition of Cauchy sequence with ε = 1. There exists some N ∈ N such that for
all n, m > N we have ∣an − am ∣ < 1. In particular, letting m = N + 1, we have ∣an − aN +1 ∣ < 1
for all n > N , and by the triangle inequality we get ∣an ∣ < ∣aN +1 ∣ + 1 for all n > N . This
is almost the bound we need; we just need to take care of the terms an with n ≤ N . Let
B = max(∣a1 ∣, ∣a2 ∣, . . . , ∣aN ∣, ∣aN +1 ∣ + 1). Then by definition of B, we have ∣an ∣ ≤ B for all n ∈ N.
This proves that (an )n∈N is a bounded sequence.
By the Bolzano-Weierstrass Theorem, it follows that (an )n∈N has a convergent subsequence,
say (ank )k∈N with limit L.
Now, what have we got? If we go far enough down the sequence, all the terms of the
subsequence are “close to L”, and all the terms of the sequence are close to each other, and so
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 172

in particular close to some term of the subsequence. Therefore all the terms of the sequence are
close to L. Let’s take that vague plan and make it into a proof.
Take any ε > 0. As ank → L, there is some K such that, for k > K, ∣ank − L∣ < ε/2. Also, as
(an )n∈N is a Cauchy sequence, there is some N ∈ N such that, for all m, n > N , ∣an − am ∣ < ε/2.
Now, choose a natural number k > K so that also nk > N . Note that ∣ank − L∣ < ε/2. For
n > N , we have ∣an − ank ∣ < ε/2 (as both n and nk are greater than N ), and so
ε ε
∣an − L∣ ≤ ∣an − ank ∣ + ∣ank − L∣ < + = ε.
2 2
We have now shown that an → L, as required.
(Imagine writing the displayed calculation first, then filling in the proof.)
Solution to Exercise 9.9. Since this was an exam question, I’ll give the markscheme too.
(a) [2pts] A sequence (an )n≥1 is convergent if there is L ∈ R so that for every ε > 0 there is
N ∈ N such that for all n > N we have ∣an − L∣ < ε.
[3pts] Suppose that (an )n∈N is convergent with limit L. Take any ε > 0. By the definition of
convergence, there is some N ∈ N such that, for all n > N , ∣an −L∣ < ε. Now, for any n > N , we also
have n + 1 > N and so ∣an+1 − L∣ < ε. Therefore, (an+1 )n≥1 is convergent and lim an = lim an+1 = L.
n→∞ n→∞

(b) [5pts: 1 for using induction and noting that the base case is valid, 2 each for the lower
and upper bound in the induction step]
We proceed by induction on n. For n = 1, we have 2 < b = b1 < 3. Suppose that for n = k, we have
2 < bk < 3. Then, for n = k + 1, we obtain

bk+1 − 2 = b2k − 4bk + 4 = (bk − 2)2 > 0,

as bk > 2. Hence, bk+1 > 2. Similarly,

3 − bk+1 = 3 − (b2k − 4bk + 6) = −b2k + 4bk − 3 = (3 − bk )(bk − 1) > 0,

as 3 > bk > 2. Hence, bk+1 < 3.

(c) [4pts: 2 for the plan and 2 for the details]
For all natural numbers n, we have

bn+1 − bn = b2n − 4bn + 6 − bn = b2n − 5bn + 6 = (bn − 2)(bn − 3).

From part (b), we have bn − 2 > 0 and bn − 3 < 0, hence bn+1 − bn < 0 and (bn )n≥1 is a decreasing
sequence.
(d) [7pts: see breakdown in square brackets]
Parts (b) and (c) show that (bn ) is a bounded, monotone sequence. By Theorem 9.25, its limit
lim bn = B exists. [2] By part (a), lim bn+1 = B as well. [1] Hence
n→∞ n→∞

B = lim bn+1 = lim b2n − 4bn + 6 = B 2 − 4B + 6, [2]

n→∞ n→∞

where we applied the results about the algebra of limits. The above quadratic equation has
solutions B = 2 and B = 3. [1]
However, since (bn )n≥1 is a decreasing sequence, we have bn ≤ bn−1 ≤ ⋅ ⋅ ⋅ ≤ b1 = b < 3 for all
natural numbers n. So B = lim bn ≤ b < 3. Hence B = 2. [1]
n→∞

(e) [4pts: 1 for each of max/sup/inf/min] max S = sup S = b1 = b, inf S = B = 2, and min S
does not exist because 2 < bn for all n ∈ N and, therefore, inf S = 2 ∈/ S.

Notes: A mark of 10/25 is equivalent to a bare pass; a mark of 17/25 is a First. As an exam
setter, I am asked to ensure that there are at least 10 relatively easy marks available: here I
CHAPTER 9. ANALYSIS: SEQUENCES AND LIMITS 173

would claim that (a) and (e) are certainly in this category, and at least some of the marks in
(b). On the other hand, some parts of each question are supposed to be hard, and here I would
nominate (d). In an ideal world, most marks would fall between 10 and (say) 20: in practice,
the range on any one question is always wider than this.
There is an alternative approach to the question, pointed out to me by a class teacher based
on some student answers. Set cn = bn − 2, and show that cn+1 = c2n . Then show by induction that
cn = (c1 )2 for every n ∈ N. Therefore bn = 2 + (b1 − 2)2 for each n. From this exact formula,
n−1 n−1

one may read off all the required results. Neat!

 Analysis: Continuity
10
A function f ∶ R → R is a rule of correspondence that assigns to each real number a unique real
number. The functions we are most familiar with are actually unusually well-behaved, and most
functions are impossible to describe fully, let alone to work with. Many bizarre functions make
their appearance in analysis, and in order to avoid falling into pitfalls with simplistic thinking
based on our experience with “nice” functions, we need our definitions and the hypotheses
(assumptions) of theorems to be stated carefully and clearly.
Within the huge collection of functions, there is an important subset: the continuous functions.
Continuous functions play a prominent role in analysis since they include all the most familiar
and useful functions, and because they share many useful properties.
In this section we give the formal definition of a continuous function and prove two of
the most important properties of continuous functions: the Extreme Value theorem and the
Intermediate Value theorem.

10.1 Definition of continuity

f (c)

Figure 10.1: A function with a break at c. If x lies to the left of c, then f (x) is not close to f (c),
no matter how close x comes to c.
In everyday speech, a ‘continuous’ process is one that proceeds without gaps of interruptions
or sudden changes. What does it mean for a function f ∶ R → R to be continuous? The common
informal definition of this concept states that a function f is continuous if one can sketch its
graph without lifting the pencil. In other words, the graph of f has no breaks in it. If a break
does occur in the graph, then this break will occur at some point. Thus (based on this visual
view of continuity), we first give the formal definition of the continuity of a function at a point.
Next, if a function is continuous at each point, then it will be called continuous.
If a function has a break at a point, say c, then even if points x are close to c, the points
f (x) might not get close to f (c), as illustrated in Figure 10.1.

174
CHAPTER 10. ANALYSIS: CONTINUITY 175

f (c) + ε
f (x)
f (c)

f (c) − ε

c−δ c x c+δ x

Figure 10.2: The definition of continuity of a function at point c. If the function is continuous at
c, then given any ε > 0 (which determines a strip around the line y = f (c) of width 2ε), there
exists a δ > 0 (which determines an interval of width 2δ around the point c) such that whenever
x lies in this interval (so that x satisfies c − δ < x < c + δ, that is, ∣x − c∣ < δ), then f (x) satisfies
f (c) − ε < f (x) < f (c) + ε, that is, ∣f (x) − f (c)∣ < ε.

This motivates the following definition of continuity, which guarantees that if a function
is continuous at a point c, then we can make f (x) as close as we like to f (c), by choosing x
sufficiently close to c. This is illustrated in Figure 10.2.

Definition 10.1.

1. Let I be an interval in R and let c ∈ I. A function f ∶ I → R is continuous at c if for every

ε > 0, there exists a δ > 0 such that for all x ∈ I satisfying ∣x − c∣ < δ, ∣f (x) − f (c)∣ < ε.

2. If f is not continuous at c, we say that f is discontinuous at c.

3. A function f ∶ I → R is continuous on I (or just continuous if I is clear from the context) if

for every c ∈ I, f is continuous at c.

4. If f is not continuous on I, we say that f is discontinuous on I (or just discontinuous if I is

clear from the context).

These definitions are (even) a bit more complicated than the definition of convergence of
a sequence you met in the last chapter. You’ll see that you need to work with ‘continuous at
a point’ in much the same way as ‘convergence to a limit’; the proof that some function is
‘continuous at a point c’ is rather like the proof that a sequence is convergent to a limit.
The statement that a function is continuous on a whole interval I is, written out with
quantifiers:

∀c ∈ I , ∀ε > 0 , ∃δ > 0 , ∀x ∈ I such that ∣x − c∣ < δ we have ∣f (x) − f (c)∣ < ε .

This is the most complicated statement you’ve seen so far. It is important to keep in mind
what it is supposed to mean intuitively: name a point c, and a ‘how close’ ε, and then if x is
‘close enough (δ)’ to c, we’re guaranteed that f (x) is close to f (c). This will help you remember
CHAPTER 10. ANALYSIS: CONTINUITY 176

what order the quantifiers above come in. It matters. You can (as you can guess from the
sentence above) swap the first two ‘for all’ quantifiers without changing anything, but it is very
important that first you name c and ε, then you decide on how small δ has to be, and only after
that comes a second point x. In particular, the formula you write for δ cannot depend on x (it
can, and usually will, depend on c and ε.
Let’s see an example. As with convergence, I’ll write it out in the formal style in order that
this issue of ‘what does δ depend on?’ is easy to keep clear.

Example 10.2. Show that the function f ∶ R → R given by f (x) = x for all x ∈ R is continuous.

Proof. Given c ∈ R and ε > 0, we choose δ = ε.

Now given x ∈ R such that ∣x − c∣ < δ, we want to show ∣f (x) − f (c)∣ < ε. We have

∣f (x) − f (c)∣ = ∣x − c∣ < δ = ε .

This is what we need for the definition of ‘f is continuous at c’, and since we proved it for an
arbitrary c ∈ R, we conclude that f is continuous at c for all c ∈ R, i.e. f is continuous on R.
As usual, we do not know what we should choose δ to be when we write the line ‘we choose
δ = ...’, we leave it blank at first and fill it in later once we see what works.

Example 10.3. Show that the function f ∶ R → R given by f (x) = 2x + 1 for all x ∈ R is
continuous.

Proof. Given c ∈ R and ε > 0, we choose δ = 12 ε.

Now given x ∈ R such that ∣x − c∣ < δ, we have:

∣f (x) − f (c)∣ = ∣(2x + 1) − (2c + 1)∣ = 2∣x − c∣ < 2δ = ε .

Again, this is what we need for the definition of ‘f is continuous at c’, and since we proved it
for an arbitrary c ∈ R, we conclude that f is continuous on R.

Example 10.4. Show that the function f ∶ R → R given by f (x) = 1 for all x ∈ R is continuous.

Proof. Let c ∈ R = (−∞, ∞). We have to prove that f is continuous at c. Let ε > 0 be given. In
this case, any positive choice of δ will work; for instance, let δ = 1. Then if x ∈ R and ∣x − c∣ < δ = 1,
we have:
∣f (x) − f (c)∣ = ∣1 − 1∣ = ∣0∣ = 0 < ε.
So f is continuous at c. Since the choice of c ∈ R was arbitrary, it follows that f is continuous
on R.
CHAPTER 10. ANALYSIS: CONTINUITY 177

Example 10.5. Show that the function f ∶ R → R given by

⎧
⎪
⎪0 if x = 0,
f (x) = ⎨
⎪
⎩1 if x ∈ R ∖ {0},
⎪
is discontinuous at 0 and continuous at all c ∈ R ∖ {0}.

0 x

Figure 10.3: A function continuous everywhere except at 0.

Proof. Suppose that f is continuous at 0. Then for any given ε > 0 there exists a δ > 0 such that
whenever ∣x − 0∣ < δ, ∣f (x) − f (0)∣ < ε.
We just need to find one example of ε > 0 for which the above statement fails (one coun-
terexample to the ‘for all’). One example that will work is ε = 12 .
To show ε = 12 is a counterexample, we need to show there does not exist δ > 0 such that
whenever ∣x − 0∣ < δ, ∣f (x) − f (0)∣ < 12 .
So let δ > 0 be given; we need to show ‘whenever ∣x − 0∣ < δ, ∣f (x) − f (0)∣ < 12 ’ is false. In
other words, we need to find one counterexample x, i.e. x with ∣x∣ < δ such that f (x) is not
within 12 of f (0). We can take x = 2δ . To see that this choice of x is a counterexample, we need
to observe that indeed ∣x∣ < δ, and furthermore
∣f (x) − f (0)∣ = ∣1 − 0∣ = 1
which is not smaller than 12 .
So f is not continuous at 0.
Next we show that for all c ∈ R ∖ {0}, f is continuous at c. Let ε > 0 be given. Take δ = ∣c∣
2 > 0.
Then if x ∈ R and ∣x − c∣ < δ, we have
∣c∣
∣c∣ − ∣x∣ ≤ ∣∣c∣ − ∣x∣∣ ≤ ∣c − x∣ = ∣x − c∣ < δ =
2
and so
∣c∣
∣x∣ >
> 0.
2
Thus x ≠ 0 and so f (x) = 1. Hence if x ∈ R and ∣x − c∣ < δ, we obtain
∣f (x) − f (c)∣ = ∣1 − 1∣ = ∣0∣ = 0 < ε.
Consequently f is continuous at c.
In the above, note that the proof of continuity follows the same pattern as the previous
examples. It’s easy to get the proof of discontinuity at 0 wrong. If you are not sure what to
do, write out the statement of ‘f is continuous at 0’ clearly, with all the quantifiers, and then
negate it (i.e. follow the rules from Chapter 3), and check that the negation is a true statement.
This is what we did above. If you are happy with this logic, then you can afford to shorten it:
‘To prove f is not continuous at 0, pick ε = 12 . Given δ > 0, pick x = 2δ . Then ∣x∣ < δ, but
∣f (x) − f (0)∣ = 1 > ε.’
CHAPTER 10. ANALYSIS: CONTINUITY 178

Example 10.6. Show that the function f ∶ (0, ∞) → R given by f (x) = 1

x for all x ∈ R is
continuous.
2
Proof. Let c ∈ (0, ∞). Given ε > 0, let δ = min { 2c , εc2 } (which is positive). Then if x ∈ (0, ∞) and
∣x − c∣ < δ, we have
c c
c − x ≤ ∣c − x∣ < δ ≤ , and so x > > 0.
2 2
Consequently, if x ∈ (0, ∞) and ∣x − c∣ < δ,

1 1 ∣c − x∣ 1 1 2 1 2δ
∣ − ∣= = ∣x − c∣ ⋅ ⋅ < δ ⋅ ⋅ = 2 ≤ ε.
x c xc x c c c c
So f is continuous at c. Since the choice of c ∈ (0, ∞) was arbitrary, it follows that f is continuous
on (0, ∞).
It’s very easy in this last proof to make a mistake, especially if you write it informally ‘let
δ > 0 be chosen later’ and only at the last line put in a value for δ. Much as in Warning 9.13, it
is very tempting to write ‘let δ = εxc’, since that would make the algebra work: we have

∣f (x) − f (c)∣ = ∣x−c∣

xc < δ
xc .

But this ‘choice’ of δ doesn’t make sense: it depends on x, and (at the point where we choose it
in the formal proof above) there is no x around. What we need is to give some real number for
δ which guarantees that xc δ
will be at most ε whatever x gets chosen such that ∣x − c∣ < δ. That
would be difficult if x were very tiny (because then x1 is huge) and this is why we choose δ small
enough that we can write x ≥ 2c , to rule out it being very close to 0.

10.2 Continuous functions preserve convergent sequences

We now give an alternative characterisation of continuity. You should have recognised that
the proof that f is continuous at some given point c resembles rather closely the proof that a
sequence converges to a given limit. The reason is that something rather similar is going on:
and it turns out to be useful to recognise this fact formally.

Theorem 10.7. Let I be an interval in R and let c ∈ I. Suppose that f ∶ I → R is a function.

Then f is continuous at c if and only if

for every convergent sequence (xn )n∈N contained in I with limit c,

(10.1)
(f (xn ))n∈N is convergent and lim f (xn ) = f (c).
n→∞

Proof. “Only if ” (Ô⇒) direction: Assume that f is continuous at c ∈ I and let (xn )n∈N be a
convergent sequence contained in I with limit c. We have to show that (f (xn ))n∈N converges to
f (c).
Let ε > 0 be given.
Since f is continuous at c ∈ I, for the given ε > 0, there exists δ > 0 such that for all x ∈ I
satisfying ∣x − c∣ < δ we have ∣f (x) − f (c)∣ < ε. Now since (xn )n∈N is convergent with limit c, by
the definition of convergence there exists N ∈ N such that for all n > N we have ∣xn − c∣ < δ. This
is the N we will use to verify the definition of lim f (xn ) = f (c).
n→∞
Suppose n > N . Then by choice of N we have ∣xn − c∣ < δ. Since xn ∈ I (because (xn )n∈N is
assumed to be contained in I), by choice of δ that means we have ∣f (xn ) − f (c)∣ < ε, which is
what we wanted to show.
CHAPTER 10. ANALYSIS: CONTINUITY 179

“If ” (⇐Ô) direction: Suppose that (10.1) holds. We have to show that f is continuous at c.
We prove this by contradiction. Assume that f is not continuous at c, that is,

¬ [∀ε > 0 ∃δ > 0 such that ∀x ∈ I such that ∣x − c∣ < δ, ∣f (x) − f (c)∣ < ε,]

or equivalently,

∃ε > 0 such that ∀δ > 0 ∃x ∈ I such that ∣x − c∣ < δ but ∣f (x) − f (c)∣ ≥ ε .

So let ε > 0 be our counterexample to continuity of f at c, i.e. ε is such that

∀δ > 0 ∃x ∈ I such that ∣x − c∣ < δ but ∣f (x) − f (c)∣ ≥ ε (10.2)

is a true statement.
For each n ∈ N, we want to choose a number xn . We do this as follows. First, choose δ = n1 .
Since this is a positive number, by (10.2) there exists x ∈ I such that ∣x − c∣ < δ = n1 and
∣f (x) − f (c)∣ ≥ ε. We let xn be any such x. That is, we choose xn such that ∣xn − c∣ < n1 and
∣f (xn ) − f (c)∣ ≥ ε.
This gives us a sequence (xn )n∈N .
Claim 1: The sequence (xn )n∈N is contained in I and is convergent with limit c.
Indeed, we have for all n ∈ N, xn ∈ I. Furthermore, given any ζ > 0, we can find N ∈ N such
that ζ1 < N (Archimedean property), that is, N1 < ζ. Hence for n > N , we have ∣xn − c∣ < N1 < ζ.
So (xn )n∈N is convergent with limit c.
Claim 2: The sequence (f (xn ))n∈N does not converge to f (c).
Recall that lim f (xn ) = f (c) means (by definition):
n→∞

∀ζ > 0 , ∃N ∈ N , ∀n ∈ N with n > N we have ∣f (xn ) − f (c)∣ < ζ .

We now show ζ = ε
2 is a counterexample to this statement. That is, for any N ∈ N,

∀n ∈ N with n > N we have ∣f (xn ) − f (c)∣ < ζ

is false. It’s enough to find one counterexample n. We choose n = N + 1, which is an integer

bigger than N .
Recall that we constructed xn (for this particular n) such that ∣f (xn ) − f (c)∣ ≥ ε. Since ε > ζ,
in particular ∣f (xn ) − f (c)∣ < ζ is false, which proves Claim 2.
Claims 1 and 2 show that (10.1) does not hold, a contradiction. Hence f is continuous at
c.
It’s worth noticing that our choice δ = n1 is just one of many choices that work. What’s going
on in the ‘if’ part of the proof is that we want to show that if f is not continuous at c, then there
is a sequence of points (xn )n∈N which witnesses it—that is, the points xn get arbitrarily close to
xn but the function values don’t get close to c. We could have written δ = 2−n , or δ = log(1+n) 1
,
equally well—we just need that if n is large then xn is guaranteed to be close to c.

Activity 10.1. Consider the function f ∶ [0, ∞] → R defined by

⎧
⎪
⎪0 if x = 0
f (x) = ⎨ 1 .
⎪
⎪ if x ≠ 0
⎩sin x
Show that the sequence ( πn
1
)n∈N converges to 0 and that lim f ( πn
1
) = f (0). Is f continuous at 0?
n→∞
CHAPTER 10. ANALYSIS: CONTINUITY 180

The point of Theorem 10.7 is that it lets us ‘translate’ the Algebra of Limits to show that
doing algebraic operations with continuous functions gives us continuous functions; this is
Theorem 10.9 below. As with the Algebra of Limits, it is painful to prove from the definition that
a given function is continuous, and we would like tools that tell us that many of the functions
we want to study are indeed continuous.
Before stating it, we introduce some convenient notation.
Definition 10.8. Let I be an interval in R. Given functions f ∶ I → R and g ∶ I → R, we define
the following:
1. If α ∈ R, then we define the function αf ∶ I → R by (αf )(x) = α ⋅ f (x), x ∈ I.
2. We define the absolute value of f to be the function ∣f ∣ ∶ I → R given by ∣f ∣(x) = ∣f (x)∣, x ∈ I.
3. The sum of f and g is the function f + g ∶ I → R defined by (f + g)(x) = f (x) + g(x), x ∈ I.
4. The product of f and g is the function f g ∶ I → R defined by (f g)(x) = f (x)g(x), x ∈ I.
5. If k ∈ N, then we define the k th power of f , to be the function f k ∶ I → R given by f k (x) =
(f (x))k , x ∈ I.

6. If for all x ∈ I, g(x) ≠ 0, then we define the function 1

g ∶ I → R by ( g1 ) (x) = 1
g(x) , x ∈ I.
Theorem 10.9. Let I be an interval in R and let c ∈ I. Suppose that f ∶ I → R and g ∶ I → R
are continuous at c. Then:
(a) For all α ∈ R, αf is continuous at c.
(b) ∣f ∣ is continuous at c.
(c) f + g is continuous at c.
(d) f g is continuous at c.
(e) For all k ∈ N, f k is continuous at c.
(f ) If for all x ∈ I, g(x) ≠ 0, then 1
g is continuous at c.
Proof. Suppose that (xn )n∈N is a convergent sequence contained in I, with limit c. Since f and g
are continuous at c, from Theorem 10.7, it follows that (f (xn ))n∈N and (g(xn ))n∈N are convergent
with limits f (c) and g(c), respectively. Each one of the statements now follows easily from
Theorem 9.30 and a second application of Theorem 10.7. For example, consider statement (d)
(the other cases may be proved as exercises).
By Theorem 9.30, (f (xn )g(xn ))n∈N is convergent with limit f (c)g(c), that is, ((f g)(xn ))n∈N
is convergent with limit (f g)(c). Since (xn )n∈N is arbitrary, this proves (10.1) for f g.
Thus by Theorem 10.7, f g is continuous at c.
Example 10.10. Every polynomial function (that is, every function p ∶ R → R defined by
p(x) = a0 + a1 x + ⋅ ⋅ ⋅ + ak xk , where k ∈ N and a0 , . . . , ak ∈ R) is continuous.
Proof. Since f ∶ R → R given by f (x) = x for x ∈ R is continuous (see Example 10.2), it follows
that for all k ∈ N, the function x → xk is continuous by Theorem 10.9(e). Thus given any
scalars a0 , a1 , . . . , ak in R, it follows that the functions a0 ⋅ 1, a1 ⋅ x, . . . , ak ⋅ xk are continuous
by Theorem 10.9(a). Summing these up, using Theorem 10.9(c) k times, we see
p(x) = a0 + a1 x + ⋅ ⋅ ⋅ + ak xk , x ∈ R
is continuous.
CHAPTER 10. ANALYSIS: CONTINUITY 181

10.3 Restrictions and compositions of functions

Definition 10.11. If f ∶ I → R is a function on an interval I, and J is an interval contained in
I (that is, J ⊆ I), then the restriction of f to J, is the function f ∣J ∶ J → R defined by
f ∣J (x) = f (x), x ∈ R.
The following theorem implies (completely as expected) that the restriction of a continuous
function is continuous.
Theorem 10.12. Let I and J be intervals such that J ⊆ I and let c ∈ J. If f is continuous at c,
then f ∣J is continuous at c.
This is an exercise.
The converse of the above theorem is not true unless J is an open interval (this is part of
the same exercise).
Recall from Section 5.2.3 that normally we only define the composition g ○ f of two functions
f and g if the codomain of f is equal to the domain of g. In Analysis, it is convenient to ‘cheat’
a bit: we will still talk about g ○ f when the codomain of f is R and the domain of g is only
some interval I ⊆ R. However, be careful: for this to make sense, we need to be sure that the
range of f is contained in I.
That is, we say: if f ∶ I → R and g ∶ J → R are functions such that for all x ∈ I, f (x) ∈ J, then
we write g ○ f ∶ I → R for the function defined by
(g ○ f )(x) = g(f (x)) .
The following theorem implies that the composition of continuous functions is continuous.
Theorem 10.13. If f ∶ I → R is continuous at c ∈ I and g ∶ J → R is continuous at f (c) (with
f (x) ∈ J for all x ∈ I), then g ○ f is continuous at c.
Proof. Take ε > 0. Then, as g is continuous at f (c), there is some η > 0 such that y ∈ J and
∣y − f (c)∣ < η imply that ∣g(y) − g(f (c))∣ < ε. As also f is continuous at c, there exists δ > 0 such
that x ∈ I and ∣x − c∣ < δ imply ∣f (x) − f (c)∣ < η.
Now we assemble what we know. Suppose that x ∈ I and ∣x − c∣ < δ. Then f (x) ∈ J and
∣f (x)−f (c)∣ < η, from which we deduce that ∣g(f (x))−g(f (c))∣ < ε, i.e., ∣(g○f )(x)−(g○f )(c)∣ < ε.
We conclude that g ○ f is continuous at c.
Example 10.14. Show that the function f ∶ (1, ∞) → R defined by
√
x2 − 1
f (x) = , x > 1,
x2 + 1
is continuous on (1, ∞).
√
Proof. Let g ∶ (0, ∞) → R be defined by g(x) = x. In an exercise you will prove that g is
continuous. Let h ∶ R → (0, ∞) be defined by
x2 − 1
h(x) = , x ∈ R.
x2 + 1
We have already seen that all polynomial functions are continuous. Therefore, p1 , p2 ∶ R → R
defined by p1 (x) = x2 − 1 and p2 (x) = x2 + 1 are continuous. Since p2 (x) ≠ 0 for all x ∈ R, it follows
1 1
by Theorem 10.9((f)) that is continuous. Then by Theorem 10.9((d)), p1 ⋅ = h is continuous.
p2 p2
Note that the restriction of h to (1, ∞), h∣(1,∞) assumes only positive values: if x ∈ (1, ∞)
then x > 1, therefore x2 − 1 > 0, and since x2 + 1 > 0 anyway, h(x) > 0. This shows that the
composition g ○ h∣(1,∞) = f is properly defined. By Theorem 10.12, h∣(1,∞) is continuous. Finally,
by Theorem 10.13, g ○ h∣(1,∞) = f is continuous at each c > 1, and therefore, continuous.
CHAPTER 10. ANALYSIS: CONTINUITY 182

10.4 Intermediate Value theorem

We now prove one of the most fundamental (and obvious!) theorems on continuous functions: a
continuous function cannot “hop over” intermediate values. For instance, if the top of a mountain
is 1976 meters above sea level and the foot of the mountain is 341 meters above sea level, then
given any number between 341 and 1976, say 1001, there must exist a point on the mountain
that is exactly 1001 meters above sea level.
y
f (b)

y
f (a)

a c x
b

The Intermediate Value theorem was first proved by Bernhard Bolzano in 1817.

Theorem 10.15. (Intermediate Value theorem). If f ∶ [a, b] → R is continuous and y is such

that f (a) ≤ y ≤ f (b) or f (b) ≤ y ≤ f (a), then there exists c ∈ [a, b] such that f (c) = y.

The theorem statement says ‘if f (a) ≤ y ≤ f (b) or f (b) ≤ y ≤ f (a)’. This is two separate
(almost—unless f (a) = f (b) ) cases. It’s easy to prove the theorem if either f (a) = y or f (b) = y,
because we just take c = a or c = b respectively. So let’s suppose that y is not equal to f (a) or
f (b), but rather strictly in between them. And let’s look at the first case, f (a) < f (b), which is
the picture above.
As we can see from the above picture, it might be the case that f (x) = y occurs at several
points; in the figure, it occurs twice (once at c, and once earlier at what looks like a local
maximum of f ). We would like to avoid confusing the different possibilities, so we pick out some
special c; maybe the easiest is to pick the biggest. That is, we want to prove there exists c ∈ [a, b]
such that f (c) = y, and f (x) > c for all c < x ≤ b. Intuitively, the c we want should just be the
biggest element of {x ∶ f (x) ≤ y}; what we need to do is prove that that exists and works.
The ‘idea’—the reason that continuity comes in to the proof—is the following. We know f is
continuous at c. That is, if x is close to c then f (x) is close to f (c).
If f (c) is smaller than y, then so is f (x) for all x close to c. In particular, if we look at x
just a little bit bigger than c, then f (x) will be smaller than y—but then x should be in S, so c
isn’t an upper bound for S, a contradiction.
If f (c) is bigger than y, then so is f (x) for all x close to c. In particular, if we look at any x
just a little bit smaller than c, then f (x) is bigger than y—but then c isn’t a least upper bound
for S, again a contradiction.
We saw an argument like this before, in the proof of Example 8.9. This is really the same
argument as there, just written out in general.
Let’s now put in the formal details.
CHAPTER 10. ANALYSIS: CONTINUITY 183

Proof. If y = f (a) then c = a satisfies the statement of the theorem. Similarly if y = f (b) then
c = b satisfies the statement of the theorem. So suppose y is not equal to f (a) or f (b).
Case 1: Suppose that f (a) < y < f (b).
Let S = {x ∈ [a, b] ∣ f (x) ≤ y}. We want to prove that

1. sup S exists, and

2. if we set c = sup S, then c ∈ [a, b] and f (c) = y.

Since f (a) ≤ y, it follows that a ∈ S, so S ≠ ∅. And S is clearly bounded above by b. Therefore,

by the least upper bound property of the reals, sup S exists.
Write c = sup S. Since a ∈ S, by definition a ≤ c. Since b is an upper bound of S, c ≤ b.
Therefore, c ∈ [a, b], and it remains to prove that f (c) = y.
Suppose for a contradiction that f (c) ≠ y, and let

∣f (c) − y∣
ε= ,
2
which is positive. Since c ∈ [a, b], so f is continuous at c. That means that there exists δ > 0
such that
∣f (c) − y∣
∣f (x) − f (c)∣ < for all x ∈ [a, b] such that ∣x − c∣ < δ . (10.3)
2
Fix δ > 0 such that (10.3) holds.
If f (c) < y: In this case, we have c < b since f (b) > y. Now observe that either c + 2δ ≤ b, or
∣b − c∣ < δ (or both). In particular, there is an x > c such that x ∈ [a, b] and ∣x − c∣ < δ; fix any
such x. Now by (10.3), we have

∣f (c) − y∣
f (x) < f (c) + <y
2
and so by definition of S, we have x ∈ S, and we already know x > c. But c is assumed to be an
upper bound of S, which is a contradiction.
If f (c) > y: In this case, we have c > a since f (a) < y. As before, observe that either c − 2δ ≥ a,
or ∣a − c∣ < δ (or both) and so there is a z < c such that z ∈ [a, b] and ∣z − c∣ < δ; fix any such z.
Now given any x ∈ [z, c], we have x ∈ [a, b] and ∣x − c∣ < δ, so by (10.3) we have

∣f (c) − y∣
f (x) > f (c) − > y,
2
and in particular x is not in S. That is, no element of [z, c] is in S, and (because c is an upper
bound of S) no element of (c, ∞) is in S. Putting these together, no element of [z, ∞) is in S,
so z is an upper bound of S. But we know z < c, which contradicts our assumption that c is a
least upper bound of S.
Assuming f (c) ≠ y, we considered both possibilities, f (c) < y and f (c) > y, and in either
case got to a contradiction. So the only possibility is that f (c) = y, which is what we wanted.
Case 2: Suppose that f (a) > y > f (b).
It would be easy to modify the above proof—just swap signs—to handle this case too. But it
is neater (and less work!) to reduce this case to Case 1. By Theorem 10.9, since f is continuous,
so is (−f ). Since f (a) > y > f (b), so (−f )(a) < −y < (−f )(b).
So by applying Case 1 to −f , there is some c ∈ [a, b] such that (−f )(c) = −y, and this tells
us f (c) = y, which is what we wanted.
CHAPTER 10. ANALYSIS: CONTINUITY 184

Example 10.16. Show that every polynomial of odd degree with real coefficients has at least
one real root.
The idea here is simple. Suppose that our odd-degree polynomial is p(x) = a0 + a1 x + ⋅ ⋅ ⋅ + ak xk ,
where k is odd. We know ak is not zero (that’s what it means to say the degree is k). Dividing the
whole polynomial by ak doesn’t change the roots, so let’s assume p(x) = a0 +a1 x+⋅ ⋅ ⋅+ak−1 xk−1 +xk .
When ∣x∣ is very big, xk will be much bigger than all the other terms; the graph of p(x) and
of xk will look pretty similar. So when ∣x∣ is very big and x is negative, we should find p(x) is
also negative. So there is some a such that p(a) < 0. Similarly, if ∣x∣ is big and x > 0, then p(x)
will also be positive. So there is some b such that p(b) > 0. And then applying the Intermediate
Value theorem on [a, b], with y = 0, says that there is some c ∈ [a, b] such that p(c) = 0, which is
what we want.
Proof. Suppose that p is a polynomial with degree k, where k is an odd natural number. Then
the coefficient of xk in p(x) is not zero. Since p(x)/s has the same roots as p(x) for any non-zero
s, we can assume that the coefficient of xk in p(x) is 1. That is, we have

p(x) = a0 + a1 x + ⋅ ⋅ ⋅ + ak−1 xk−1 + xk .

We want to first justify that there is some b such that p(b) > 0 (the algebra is a bit simpler
than for a). We will actually choose b to be a positive integer. Observe that
p(n)
lim = lim na0k + nak−1
1
n +1=1
+ ⋅ ⋅ ⋅ + ak−1
n→∞ nk n→∞

where the first equality is just writing out p(n) and simplifying, and the second equality is by
the Algebra of Limits.
p(n)
Using the definition of lim k = 1, with ε = 12 , there exists some N such that for all n > N
n→∞ n
we have
RRR R
RRR p(n) − 1RRRRR < 1 p(n) 1
> 2.
RRR nk RRR 2 and so
nk
RR RR
In particular, we can let b = N + 1 and obtain p(b) > 12 bk > 0, which is what we wanted.
We do something similar for a (which will be a negative integer). We have
p(−n)
lim = lim (−n)
a0
k + (−n)k−1 + ⋅ ⋅ ⋅ + −n + 1 = 1
a1 ak−1
n→∞ (−n)k n→∞

and again using the definition of the limit, with ε = 12 , there is N such that for all n > N we have
RRR R
RRR p(−n) − 1RRRRR < 1 p(−n) 1
> .
RRR (−n)k RRR 2 and so
(−n)k 2
RR RR
In particular, we can let a = −N − 1, and this gives p(a) < 12 (−N − 1)k < − 12 (N + 1)k < 0, which is
what we wanted.
Now we apply the Intermediate Value theorem to p(x), on the interval [a, b], with y = 0.
The function p is continuous by Example 10.10, and f (a) < 0 = y < f (b), so the conditions of the
Intermediate Value theorem are satisfied. We conclude there is some c ∈ [a, b] such that p(c) = 0,
which is what we wanted.
It’s worth noticing that a bit more is true. If we want to know where a polynomial (or any
continuous function!) has roots, it’s enough to find a point where the polynomial takes a value
smaller than 0 and another close by where it takes a value bigger than 0. Then there has to be
a root in between.
CHAPTER 10. ANALYSIS: CONTINUITY 185

Example 10.17. The polynomial p(x) = x3 − x2014 + 399

1
has a real root in [−1, 1].

Proof. p(1) = 3991

> 0 and p(−1) = −2 + 1
399 < 0 and so by the Intermediate Value theorem
∃c ∈ [−1, 1] such that p(c) = 0.
Remark 10.18. You’ve no doubt already noticed doing MA100 that computers can be rather
useful for solving ‘methods’ problems. Wouldn’t it be nice if they could also help us prove things?
Well, sadly we still do not know how to make a computer do all of the inventive clever stuff you
need in order to prove interesting statements. But you can get the computer to do the algebra
for you. And, courtesy of this observation about roots of functions, you can also get a computer
to prove for you that a polynomial that shows up in your proof has a root smaller than 1 (which,
we’re imagining, will be useful to you..!). Or to show that a 50 by 50 matrix is positive definite,
i.e. all its eigenvalues are positive (why should you care? well, because it is a useful thing to
know about a matrix). In short, you can get a computer to do a lot of numerical work for you,
give you an answer which is some funny decimal, and prove that this decimal is correct up to
ten (or a hundred, if you want) places. Quite a lot of modern proofs in mathematics have parts
which are numerical work done by a computer, but which the computer can prove is correct.
Why should you care about√ being able to prove a computer’s numerical work is correct?
Here’s an example: evaluate eπ 163 . With a calculator, you can check it is equal to the integer
640 3233 + 744. Try again, with a computer, to check it’s true to five or so decimal places, and
you should be convinced.
In fact, this number is not an integer; it’s off by about 10−11 from an integer. It would be
rather embarrassing if you thought it was an integer and assumed that in a proof!

Example 10.19. Show that at any given time, there exists a pair of diametrically opposite
points on the equator which have the same temperature.

Proof. Let T (Θ) denote the surface temperature at the point at longitude Θ. See Figure 10.4.
(Note that Θ(0) = Θ(2π).) Assuming that T is a continuous function of Θ, it follows that the

Figure 10.4: The point on the equator with longitude Θ.

function f ∶ [0, π] → R defined by f (Θ) = T (Θ) − T (Θ + π) is continuous as well. If f (0) = 0, then

it follows that the temperatures at 0 and 180○ longitude are the same. If f (0) ≠ 0, then since
f (0) and f (π) = −f (0) have opposite signs, by the Intermediate Value theorem, it follows that
f must vanish at some point, and so the claim follows.
CHAPTER 10. ANALYSIS: CONTINUITY 186

10.5 Extreme Value theorem

We’ve seen that functions in general can do all kinds of unpleasant things, jumping around all
over the place and so on. Continuous functions don’t behave this badly, but we’ve still seen
continuous functions which go off to infinity, such as f ∶ (0, 1) → R defined by f (x) = x1 ; this
gets very big as x gets close to 0.
There are also continuous functions which do some rather strange things, such as the
continuous function g ∶ R → R defined by
⎧
⎪
⎪0 if x = 0
g(x) = ⎨ ,
⎪
⎪
1
if x ≠ 0
⎩x sin x
which has infinitely many local maxima and minima, even if you only look at the interval [−1, 1]
(see Figure 10.5).
Another ‘funny function’ is h ∶ (0, ∞) → R defined by

h(x) = (1 − x) sin x1 ,

which also has infinitely many local maxima (as x gets closer to 0), where the function values
get closer and closer to 1, yet no global maximum exists. This function is also continuous—but
notice that it is not defined at 0, and in fact if we try to change the function to one defined on
[0, ∞) (i.e. we give the function a value at 0) then it will not be continuous at 0.

0.8 1

0.6 0.5
0.4
0
0.2

0 −0.5

−0.2 −1
−1 −0.5 0 0.5 1 0 0.1 0.2 0.3 0.4 0.5

Figure 10.5: The functions g(x) and h(x).

Activity 10.2. Prove that g(x) as defined above is continuous on R (you should assume the
sine function is continuous).
Prove that there is no continuous function h∗ ∶ [0, ∞) → R such that h∗ (x) = h(x) for all
x ∈ (0, ∞).

You might well imagine that there are continuous functions on closed intervals which have
some similar nasty behaviour: maybe they go off to infinity somewhere in the middle of the
interval, or they have infinitely many local maxima which (like h(x) above) keep getting bigger
and bigger so that even if the function stays bounded, there is no global maximum?
In fact, surprisingly, this is not the case. Any continuous function f on any closed interval is
bounded, and it has at least one global maximum and minimum (the extreme values).
CHAPTER 10. ANALYSIS: CONTINUITY 187

d
a c x
b

In this figure the extreme value are both attained once each, at c and d. For a function like
x → cos x, on the interval [0, 1000], there are many global maxima (and minima!)—0, 2π, 4π
and so on are all global maxima. There can even be infinitely many global maxima. A trivial
example is a constant function on I: every point is a global maximum.
Activity 10.3. Find a function f ∶ [0, 1] → R which is continuous, not constant, and has
infinitely many global maxima.
Theorem 10.20 (Extreme Value theorem). Let [a, b] be any closed and bounded interval and
let f ∶ [a, b] → R be a continuous function. Then there exists c ∈ [a, b] such that

f (c) = sup {f (x) ∣ x ∈ [a, b]} , (10.4)

and there exists d ∈ [a, b] such that

f (d) = inf {f (x) ∣ x ∈ [a, b]} . (10.5)

Since c, d ∈ [a, b] in the above theorem, the supremum and infimum in (10.4) and (10.5) are
in fact the maximum and minimum, respectively. This proof is a beautiful application of the
Bolzano–Weierstrass theorem—actually two beautiful applications.
We’ll only prove the ‘maximum’ half of the theorem; the ‘minimum’ half is an exercise.
The idea is the following. Suppose for a contradiction that sup {f (x) ∣ x ∈ [a, b]} does not
exist. How can that happen? This is a non-empty set of real numbers; if it has an upper bound
then the least upper bound principle says it has a supremum. So the only possibility is that
this set doesn’t have an upper bound. Well, that would mean that there are points x1 , x2 , . . . in
[a, b] where the function values f (x1 ), f (x2 ), . . . are growing bigger and bigger: the sequence of
function values isn’t bounded above. But this looks suspiciously like a contradiction. We know
that if lim xn = c, then lim f (xn ) = f (c) by Theorem 10.7. But this can’t be: the sequence of
n→∞ x→∞
function values is unbounded, so it cannot be convergent.
But this isn’t quite a contradiction: we don’t know that (xn )n∈N is a convergent sequence—
what if that sequence bounces around in [a, b] without tending to a limit? The answer is: apply
the Bolzano-Weierstrass theorem to find a convergent subsequence (xnk )k∈N , and repeat the
above argument with this convergent subsequence—this time it works.
That argument shows that M = sup {f (x) ∣ x ∈ [a, b]} exists. But why is there actually a value
c ∈ [a, b] such that f (c) = M ? Well, M is supposed to be a least upper bound of {f (x) ∣ x ∈ [a, b]}.
That means that for every ε > 0,

M − ε isn’t an upper bound, i.e. there is some x ∈ [a, b] such that M ≥ f (x) > M − ε . (10.6)

Note that we can write that M ≥ f (x) since M is an upper bound for {f (x) ∣ x ∈ [a, b]}. We
want to use this to construct a sequence (xn )n∈N such that the function values converge to
M , i.e. limn→∞ f (xn ) = M . We’ll use the same trick we saw before: use (10.6) for a sequence
of ε values which tends to zero. The obvious ‘sequence tending to zero’ is ( n1 )n∈N , so let’s use
that. By (10.6), for each n ∈ N there is xn ∈ [a, b] such that M ≥ f (x) > M − n1 . The Sandwich
Theorem tells us lim f (xn ) = M . If lim xn = c exists, then by Theorem 10.7 we have M = f (c),
n→∞ n→∞
which is what we want.
CHAPTER 10. ANALYSIS: CONTINUITY 188

Again, there is no reason why lim xn should exist: if f has several global maxima, then perhaps
n→∞
(xn )n∈N bounces around between them without converging. But, again, the Bolzano-Weierstrass
theorem rescues us.
Proof. We prove the first half of the theorem, leaving the second half as an exercise.
Let S = {f (x) ∣ x ∈ [a, b]}. We first want to prove M = sup S exists, then we want to show
that there is some c ∈ [a, b] such that f (c) = M .
To show that sup S exists, we use the least upper bound property of R. We just need to
show that S is not empty and that it has an upper bound.
S is not empty because f (a) ∈ S.
Now suppose for a contradiction that S has no upper bound. Then in particular, for each
n ∈ N, the integer n is not an upper bound for S. That means there is an element of S which is
bigger than n, i.e. a function value of f which is bigger than n. There is some xn ∈ [a, b] such
that f (xn ) > n.
This gives us a sequence (xn )n∈N which is bounded: it is in [a, b]. By the Bolzano-Weierstrass
theorem, there is a convergent subsequence (xnk )k∈N , and by Theorem 9.12 the limit s of this
subsequence is in [a, b].
Since 1 ≤ n1 < n2 < . . . by the definition of a subsequence, we have nk ≥ k. Thus f (xnk ) > nk ≥ k
is true for each k ∈ N, and so (f (xnk ))k∈N is not a bounded sequence. By Theorem 9.21 it is
therefore not a convergent sequence. But this is a contradiction to Theorem 10.7, which says that
lim f (xnk ) = f (s), since lim xnk = s ∈ [a, b] and f is continuous on [a, b] (and so in particular
k→∞ k→∞
at s). This contradiction shows that our assumption—that S has no upper bound—is false.
Thus S has an upper bound, and we already observed that it is non-empty. So the least
upper bound property of the reals says that sup S exists.
Now we show that M = sup S is M = f (c) for some c ∈ [a, b].
For each n ∈ N, we let xn be a point in [a, b] such that f (xn ) > M − n1 . This exists since M
is a least upper bound of S, and so in particular M − n1 is not an upper bound of S.
Now the Sandwich Theorem says that, since M − n1 < f (xn ) ≤ M holds for each n ∈ N, so
lim f (xn ) = M .
n→∞
By the Bolzano-Weierstrass theorem, (xn )n∈N has a convergent subsequence (xnk )k∈N . Let
c = lim xnk ; by Theorem 9.12 we have c ∈ [a, b].
k→∞
The sequence (f (xnk ))k∈N is a subsequence of (f (xn ))n∈N (which we just saw is convergent
with limit M ), so by Theorem 9.43 it is convergent with limit M .
Now by Theorem 10.7, since c ∈ [a, b] and so f is continuous at c, we have f (c) = M . This
finishes the proof of the first half of the theorem.
The proof that there exists d ∈ [a, b] such that f (d) = inf S is left as an exercise.
Something we should immediately notice at this point is the following. We have just proved
a theorem which isn’t obviously true, and the proof is not really all that easy. But all the work
is being done by the theorems we proved already! We are proving things about continuous
functions using convergent sequences, yet we never actually needed to use the definitions of
either concept. Nor are we producing any ‘pages of equations’ as you might have expected from
your school maths. This is good, because working with the definitions is painful and writing
pages of equations is boring and easy to mess up.
This is your first taste of how modern mathematics looks. We want to build up a beautiful
palace of a theorem, but we generally will not go all the way down to the bricks-and-mortar of
working with all the basic definitions. Rather, we want to play architect: we outline how the
proof should go, and call upon theorems we (or, usually, others!) already proved to serve as the
walls and roof.
CHAPTER 10. ANALYSIS: CONTINUITY 189

10.6 Sample exercises

Exercise √ 10.1. Prove, using the definition of continuity, that the function f ∶ [0, ∞) → R given
by f (x) = x is continuous at 4. √
Hint: First show that ∣x − 4∣ ≥ 2∣ x − 2∣ for every x ∈ [0, ∞).
Prove that the function f ∶ R → R given by
⎧
⎪
⎪x if x is irrational,
f (x) = ⎨
⎪
⎪
⎩0 if x is rational,
is continuous at 0.

Exercise 10.2. Prove that if f ∶ R → R is continuous and f (x) = 0 whenever x is rational, then
f (x) = 0 for all x ∈ R.
Hint: Given any real number c, there exists a sequence of rational numbers (qn )n∈N that
converges to c.

Exercise 10.3. (a) Let J = (a, b) be an open interval contained in another interval I. Let
f ∶ I → R be a function. Let c ∈ J and assume that f ∣J is continuous at c. Prove that f is
continuous at c.

(b) Give an example of intervals J and I with J ⊆ I, c ∈ J and a function f ∶ I → R such that
f ∣J is continuous at c, but f is not continuous at c.
(This shows that it is necessary to assume that J is an open interval in (a)).)

Exercise 10.4. Show that the statement of Theorem 10.20 does not hold if [a, b] is replaced by
[a, b).

Exercise 10.5. A function f ∶ R → R is a periodic function if there exists T > 0 such that for
all x ∈ R, f (x + T ) = f (x). If f ∶ R → R is continuous and periodic, then prove that f is bounded,
that is, the set S = {f (x) ∣ x ∈ R} is bounded.
Give an example of a periodic function g such that g is not bounded.

Exercise 10.6. Exam Question, 2009 Q5.

(a) Let I be an interval in R and let c ∈ I.

What does it mean to say that a function f ∶ I → R is continuous at c ? What does it mean
to say that a function f ∶ I → R is continuous on I?

(b) Suppose that f ∶ R → R, g ∶ R → R and h ∶ R → R are functions such that

(1) f and g are continuous on R,

(2) f (0) = g(0), and
(3) for every x ∈ R, f (x) ≤ h(x) ≤ g(x).

Show that h is continuous at 0.

(c) State the Intermediate Value Theorem.

Suppose that the function f ∶ [0, 2] → R is continuous on the interval [0, 2] and f (0) =
f (2) ≥ f (1). Prove that there exist numbers a, b ∈ [0, 2] such that ∣a − b∣ = 1 and f (a) = f (b).
Hint: Consider the function g ∶ [0, 1] → R given by g(x) = f (x + 1) − f (x).
CHAPTER 10. ANALYSIS: CONTINUITY 190

10.7 Comments on selected activities

Comment on Activity 10.1. We have

lim 1 = 1
lim n1 =0
n→∞ πn π n→∞

using the Algebra of Limits and the known fact that lim 1/n = 0.
n→∞
Since f ( πn
1
) = sin(πn) = 0, we have

lim f ( πn
1
) = f (0) .
n→∞

However, this function f is not continuous at 0. To see this, observe that the sequence

( 2πn+π/2
1
)
n∈N

is also convergent with limit zero, but

f ( 2πn+π/2
1
)=1

for all n ∈ N. Since 1 ≠ f (0), we see that (10.1) does not hold. So by Theorem 10.7, f is not
continuous at 0.
Comment on Activity 10.2. We’ll need to cheat a bit to prove that g is continous on R. We
don’t know that the sine function is continuous, and proving it is rather hard given that we do
not really know what the function is at all! We sort-of know that it comes from Euler’s identity,
but we don’t really know how to work with infinite series. So let’s just assume it is a continuous
function (which is true).
Now for any c ≠ 0, we can find an interval [a, b] which contains c but not 0. And on that
interval, g(x) is a composition of two continuous functions, so it is continuous.
Proving continuity at 0 is easier, assuming we are happy to say we know ∣sinx∣ ≤ 1 is true
for all x ∈ R. For any ε > 0, choose δ = ε. Then for any ∣x − 0∣ < δ we have

∣g(x) − g(0)∣ = ∣x sin x1 ∣ = ∣x∣ ⋅ ∣ sin x1 ∣ ≤ ∣x∣ < δ = ε ,

which is what we wanted.

If a continuous h∗ existed, then necessarily we would have, for any (xn )n∈N which is convergent
with limit 0 and all of whose terms are positive, the statement

h∗ (0) = lim h∗ (xn ) = lim h(xn )

n→∞ n→∞

by Theorem 10.7. In particular, there cannot be two choices of (xn )n∈N where the limits of
function values are different. But the two sequences in the solution above to Activity 10.1 do
give different limits, namely 0 and 1 respectively (though proving the second sequence gives
function values converging to 1 needs a little bit more work).
Comment on Activity 10.3. We can ‘cheat’ by for example letting
⎧
⎪
⎪1 if x ≤ 1
f (x) = ⎨ 3 2
.
⎪
⎩2 − x if x > 1
⎪ 2

But this only has infinitely many global maxima because it’s constant on part of [0, 1]. Can we
find a function which isn’t constant on any part of [0, 1] that does the job?
We can—try modifying one of the examples you saw before the Extreme Value theorem.
CHAPTER 10. ANALYSIS: CONTINUITY 191

10.8 Solutions to exercises

Solution to Exercise 10.1.
First, √we show√the inequality
√ given √ in the hint.
√ We have that, for every non-negative x,
∣x − 4∣ = ∣( x − 2)( x + 2)∣ = ∣ x − 2∣ ∣ x + 2∣ ≥ 2∣ x − 2∣.
Now we show that the square root function √ at 2. 1Take any 1ε > 0, and set δ = ε.
√ is continuous
√
Then, if x ∈ [0, ∞) with ∣x − 4∣ < δ, we have ∣ x − 4∣ = ∣ x − 2∣ ≤ 2 ∣x − 4∣ < 2 δ < ε, as required.
Solution to Exercise 10.2.
Let c be any real number (rational or irrational), and – following the hint – take some
sequence (qn ) of rational numbers with qn → c. As f is continuous at c, we have f (qn ) → f (c),
by Theorem 10.7. As each f (qn ) is equal to 0, so is f (c). Thus f (c) = 0 for all real c.
Solution to Exercise 10.3.
(a) If we choose t = 12 min(c − a, b − c), we see that inside our open interval (a, b), there is an
open interval (c − t, c + t) with c in the middle. That means that every point within distance t of
c is in (a, b).
Now, let’s prove that f is continuous at c. Given ε > 0, we know that there is some δ > 0
such that, for every x ∈ J with ∣x − c∣ < δ, we have ∣f (x) − f (c)∣ < ε. We need the same conclusion
without the “∈ J”, and that would be no problem at all if δ ≤ t, since then all points x within
distance δ of c would be in J = (a, b). The key point is that we are allowed to take a smaller δ, as
long as we keep it positive. So here the natural thing to do is to set δ ′ = min(δ, t) > 0. Now every
x such that ∣x − c∣ < δ ′ satisfies both that x ∈ J and tha ∣x − c∣ < δ, and therefore ∣f (x) − f (c)∣ < ε.
(b) We can follow the hint and take f (x) = ⌊x⌋. This takes the value 1 for x ∈ [1, 2), and so
f is certainly continuous on J = [1, 2). However, f is not continuous on (0, 2), as (for instance)
f (1 − 1/2n) = 0, for all n ∈ N, so f (1 − 1/2n) does not tend to 1 = f (1), even though 1 − 1/2n → 1.
Solution to Exercise 10.4.
We need to find an example of a continuous function f ∶ [a, b) → R such that there is no
c ∈ [a, b) with f (c) = sup{f (x) ∶ x ∈ [a, b)}. In other words, we want a function that does not
have a maximum on [a, b). If you think about it, you will see that any strictly increasing function
will do, for instance f (x) = x on [1, 2): here sup{f (x) ∣ x ∈ [1, 2)} = sup[1, 2) = 2, but there is
no c ∈ [1, 2) with f (x) = x = 2. It is even possible to come up with a function f ∶ [1, 2) → R
that is not bounded above (and hence the supremum does not exist) – one easy example is
f (x) = 1/(2 − x).
Solution to Exercise 10.5.
The Extreme Value Theorem tells us that the restriction of f to [0, T ] is bounded: there is
some M such that ∣f (x)∣ ≤ M for all x ∈ [0.T ].
Now note that, for any x ∈ R, there are k and t such that x = kT + t, and 0 ≤ t < T . (We
can take k = ⌊x/T ⌋.) As f is periodic with period T , we have f (x) = f (t), so ∣f (x)∣ ≤ M . The
conclusion is that ∣f (x)∣ ≤ M for all x ∈ R, and so f is bounded.
For an example of a periodic function which isn’t bounded, of course we will need to look at
discontinuous functions. One example is to consider the function g defined by setting, for each
0 < x ≤ 1 and each integer n,
g(n + x) = x1 .
This is well-defined, because there is exactly one way to write any given real number y in the
form y = n + x where n is integer and 0 < x ≤ 1. It is obviously periodic with period 1. But it is
not bounded, because for each natural number m we have

g( m1 ) = m .
CHAPTER 10. ANALYSIS: CONTINUITY 192

Solution to Exercise 10.6.

(a) A function f ∶ I → R is continuous at c if for every ε > 0 there exists δ > 0 such that for every
x ∈ I, ∣x − c∣ < δ, we have ∣f (x) − f (c)∣ < ε. [3pts]
A function f ∶ I → R is continuous on I if it is continuous at every point c ∈ I. [2pts]
(b) From (2) and (3), we have f (0) = g(0) = h(0). [1pt]
We must show that for every ε > 0 there exists δ > 0 such that for every x such that ∣x − 0∣ < δ,
we have ∣h(x) − h(0)∣ < ε. By (1), we know that
(i) for every ε > 0 there exists δf > 0 such that for every x such that ∣x − 0∣ < δf , we have
∣f (x) − f (0)∣ < ε, and
(ii) for every ε > 0 there exists δg > 0 such that for every x such that ∣x − 0∣ < δg , we have
∣g(x) − g(0)∣ < ε.
So, for given ε > 0, let δ = min{δf , δg }. For every ∣x − 0∣ < δ, we have ∣x − 0∣ < δf and ∣x − 0∣ < δg ,
hence ∣f (x) − f (0)∣ < ε and ∣g(x) − g(0)∣ < ε.
So,
h(0) − h(x) = f (0) − h(x) ≤ f (0) − f (x) ≤ ∣f (x) − f (0)∣ < ε
and
h(x) − h(0) = h(x) − g(0) ≤ g(x) − g(0) ≤ ∣g(x) − g(0)∣ < ε.
[7pts for the bulk of the proof: 4 for the structure and idea, and 3 for the details]
(c) Intermediate Value Theorem: If f ∶ [a, b] → R is continuous and y is such that either
f (a) ≤ y ≤ f (b) or f (b) ≤ y ≤ f (a), then there exists c ∈ [a, b] such that f (c) = y. [3pts]
The function j ∶ R → R given by j(x) = x + 1 is continuous by the Algebra of Limits. Since f
is continuous on [0, 2] ⊇ [1, 2], the composition h, h(x) = f (x + 1) = f (j(x)), is continuous on
[0, 1]. Therefore the function g(x) = f (x + 1) − f (x) = h(x) − f (x) is continuous on [0, 1]. [3pts,
including 1 just for saying that g is continuous]
We see that g(1) = f (2) − f (1) ≥ 0 and g(0) = f (1) − f (2) ≤ 0. By the Intermediate Value
Theorem, there exists c ∈ [0, 1] such that g(c) = f (c + 1) − f (c) = 0. Take a = c, b = c + 1 and
observe that 0 ≤ c = a < b = c + 1 ≤ 2, ∣a − b∣ = 1 and f (a) = f (b). [6pts: 3 for the idea and 3 for
the detail]