Dr Luke Heaton graduated with first class honours in

Mathematics at the University of Edinburgh before going

on to take an MSc in Mathematics and the Logical
Foundations of Computer Science at the University of
Oxford. After spending a year making mathe� matically
inspired art, he gained a BA in Architecture at the
University of Westminster, and was briefly employed as
an architectural assistant. He then returned to Oxford,
completing a DPhil in Mathe� matical Biology. He is
currently employed by the University of Oxford as a post-
graduate research assistant in the Department of Plant
Sciences. Heaton’s research interests lie in matheÂ�matics
and the mathe�matical modelling of biological phenomena,
the history and philosophy of mathe�matics, morphogenesis
and biological pattern formation, network theory,
biophysics, and the statistical properties of efficient trans-
port networks. He has published several papers on the
biophysics of growth and transport in fungal networks.
 Recent titles in the series

A Brief Guide to James Bond

Nigel Cawthorne

A Brief Guide to Secret Religions

David V. Barrett

A Brief Guide to Jane Austen

Charles Jennings

A Brief Guide to Jeeves and Wooster

Nigel Cawthorne

A Brief History of Angels and Demons

Sarah Bartlett

A Brief History of Bad Medicine

Ian Schott and Robert Youngston

A Brief History of France

Cecil Jenkins

A Brief History of Ireland

Richard Killeen

A Brief History of Sherlock Holmes

Nigel Cawthorne

A Brief History of King Arthur

Mike Ashley

A Brief History of the Universe

J. P. McEvoy

A Brief History of Roman Britain

Joan P. Alcock

A Brief History of the Private Life of Elizabeth II

Michael Paterson
 A BRIEF HISTORY OF

Mathe�matical
Thought
Luke Heaton

1
 1
Oxford University Press is a department of the University of Oxford. It furthers the
University’s objective of excellence in research, scholarship, and education by publishing
worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and
certain other countries.
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America.
© Luke Heaton 2017
First published in Great Britain by Robinson.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, without the prior permission in
writing of Oxford University Press, or as expressly permitted by law, by license, or under
terms agreed with the appropriate reproduction rights organization. Inquiries concerning
reproduction outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above.
You must not circulate this work in any other form
and you must impose this same condition on any acquirer.
Library of Congress Cataloging-in-Publication Data
Names: Heaton, Luke.
Title: A brief history of mathematical thought / Luke Heaton.
Description: New York, NY : Oxford University Press, [2017] | Includes
bibliographical references and index.
Identifiers: LCCN 2016004417 | ISBN 9780190621766
Subjects: LCSH: Mathematics—History | Math anxiety.
Classification: LCC QA21 .H43 2017 | DDC 510.9—dc23 LC record available at
http://lccn.loc.gov/2016004417

135798642
Printed by Sheridan Books, Inc.,
Printed in the United States of America
 CONTENTS

Introduction 1

1 Beginnings
1.1 Language and Purpose 13
1.2 Human Cognition and the Meaning of Math 16
1.3 Stone Age Rituals and Autonomous Symbols 20
1.4 Making Legible Patterns 24
1.5 The Storage of Facts  28
1.6 Babylon, Egypt and Greece 31
1.7 The Logic of Circles 35
1.8 The Factuality of Math 37

2 From Greece to Rome

2.1 Early Greek Mathe�matics 42
2.2 Pythagorean Science 44
2.3 Plato and Symmetric Form 48
2.4 Euclidean Geometry 53
2.5 The Euclidean Algorithm 56
2.6 Archimedes 58
2.7 Alexandria in the Age of Rome 61

3 Ratio and Proportion

3.1 Measurement and Counting 64
3.2 Reductio Ad Absurdum67
3.3 Eudoxus, Dedekind and the Birth of Analysis 70
3.4 Recurring Decimals and Dedekind Cuts 77
3.5 Continued Fractions 81
3.6 Quadratic Equations and the Golden Ratio 84
3.7 Structures of Irrationality 87
3.8 The Fibonacci Sequence 90

4 The Rise of Algebra

4.1 Zero and the Position System 98
4.2 Al-Khwarizmi and the Science of Equations 104
4.3 Algebra and Medieval Europe 107
4.4 Fermat’s Little Theorem 112
4.5 How to Make a Mathe�matical Padlock 114

5 Mechanics and the Calculus

5.1 The Origins of Analysis 119
5.2 Measuring the World 124
5.3 The Age of Clocks 127
5.4 Cartesian Coordinates 129
5.5 Linear Order and the Number Line 134
5.6 Isaac Newton 137
5.7 The Fundamental Theorem of Calculus 141
5.8 From Algebra to Rates of Change 147

6 Leonhard Euler and the Bridges

of Königsberg
6.1 Leonhard Euler 152
6.2 The Bridges of Königsberg 154
6.3 On Drawing a Network 156
6.4 The Platonic Solids Revisited 158
6.5 Poincaré and the Birth of Topology 161

7 Euclid’s Fifth and the Reinvention

of Geometry
7.1 Measurement and Direction 167
7.2 Non-Euclidean Geometry 173
7.3 The Curvature of Space 175
7.4 The Unity and Multiplicity of Geometry 180
7.5 Symmetry and Groups 186
7.6 The Oddities of Left and Right 193
7.7 The Möbius Strip 196

8 Working with the Infinite

8.1 Blaise Pascal and the Infinite in Math 199
8.2 Reasoning by Recurrence 202
8.3 The Mathematics of the Infinitely Large 205
8.4 Cantor’s Pairs 206
8.5 The Diagonal Argument 211

9 The Structures of Logical Form

9.1 The Formal Logic of AND, OR and NOT 215
9.2 Classical Logic and the Excluded Middle 219
9.3 Mechanical Deductions 221
9.4 Quantifiers and Properties 222
9.5 Inputs for Predicate Calculus 225
9.6 Axiomatic Set Theory 227

10 Alan Turing and the Concept

of Computation
10.1 From Mechanical Deductions to Programmable
Machines232
10.2 Depicting Calculation 236
10.3 Deterministic Language Games 238
10.4 Church’s Thesis 242
10.5 Decision Problems 243
10.6 Figure and Ground 247
10.7 Semi-Decidable Problems 249

11 Kurt Gödel and the Power of Polynomials

11.1 Matiyasevich’s Theorem 253
11.2 Kurt Gödel 258
11.3 Searching for Solutions 260
11.4 The Incompleteness of Arithmetic 261
11.5 Truth, Proof and Consistency 266

12 Modelling the World

12.1 Science and the Uses of Models 273
12.2 Order and Chaos  277
12.3 Theoretical Biology 280
12.4 Interactions and Dynamical Systems 284
12.5 Holism and Emergent Phenomena 288
13 Lived Experience and the Nature of Facts
13.1 Rules and Reality 292
13.2 The Objectivity of Math 298
13.3 Meaning and Purpose 304

Further Reading308
Acknowledgements312
Index313
 INTRODUCTION
‘Mathematics is the gate and key of the sciences. …
Neglect of mathe�matics works injury to all knowledge,
since he who is ignorant of it cannot know the other
sciences or the things of this world. And what is
worse, men who are thus ignorant are unable to
perceive their own ignorance and so do not seek a
remedy.’
Roger Bacon, 1214–1292

The language of mathe�matics has changed the way we

think about the world. Most of our science and technology
would have been literally unthinkable without mathe�matics,
and it is also the case that countless artists, architects,
musicians, poets and philosophers have insisted that their
grasp of the subject was vital to their work. Clearly mathe�
matics is important, and in this book I hope to convey
both the poetry of mathe�matics and the profound cultural
influence of various forms of mathe�matical practice. For
better or worse, you can’t comprehend the influence of
math until you have some understanding of what mathe�
maticians actually do. By way of contrast, you don’t need
to be an engineer to appreciate the impact of technological
change, but it is hard to comprehend the power and
2 MATHEMATICAL THOUGHT

�
influence of mathe� matical thought without an under-
standing of the subject on its own terms.
Most people are numerate, and have learned a handful
of rules for calculation. Unfortunately the arguments and
lines of reasoning behind these techniques are much less
widely known, and far too many people mistakenly believe
they cannot hope to understand or enjoy the poetry of
math. This book is not a training manual in mathe�matical
techniques: it is an informal and poetic guide to a range
of mathe�matical thoughts. I disregard some technicalities
along the way, as my primary aim is to show how the
language of math has arisen over time, as we attempt to
comprehend the patterns of our world. My hope is that
by writing about the development of mathe�matical ideas
I can inspire some of my readers, shake up some lazy
assumptions about pure and applied mathe�matics, and
show that an understanding of math can help us to arrive
at a richer understanding of facts in general.
Mathematics is often praised (or ignored) on the
grounds that it is far removed from the lives of ordinary
people, but that assessment of the subject is utterly
mistaken. As G. H. Hardy observed in A Mathematician’s
Apology:

Most people have some appreciation of mathe�matics,

just as most people can enjoy a pleasant tune; and
there are probably more people really interested in
mathe�matics than in music. Appearances suggest the
contrary, but there are easy explanations. Music can
be used to stimulate mass emotion, while mathe�matics
cannot; and musical incapacity is recognized (no doubt
rightly) as mildly discreditable, whereas most people
are so frightened of the name of mathe�matics that
they are ready, quite unaffectedly, to exaggerate their
own mathe�matical stupidity.
 INTRODUCTION3

The considerable popularity of sudoku is a case in point.

These puzzles require nothing but the application of
mathe�matical logic, and yet to avoid scaring people off,
they often carry the disclaimer ‘no matheÂ�matical knowÂ�
ledge required’! The matheÂ�matics that we know shapes
the way we see the world, not least because mathe�matics
serves as ‘the handmaiden of the sciences’. For example,
an economist, an engineer or a biologist might measure
something several times, and then summarize their mea�
surements by finding the mean or average value. Because
we have developed the symbolic techniques for calculating
mean values, we can formulate the useful but highly
abstract concept of ‘the mean value’. We can only do this
because we have a mathe� matical system of symbols.
Without those symbols we could not record our data, let
alone define the mean.
Mathematicians are interested in concepts and patterns,
not just computation. Nevertheless, it should be clear to
everyone that computational techniques have been of vital
importance for many millennia. For example, most forms
of trade are literally inconceivable without the concept
of number, and without mathe� matics you could not
organize an empire, or develop modern science. More
generally, mathe�matical ideas are not just practically
important: the conceptual tools that we have at our
disposal shape the way we approach the world. As the
psychologist Abraham Maslow famously remarked, ‘If
the only tool you have is a hammer, you tend to treat
everything as if it were a nail.’ Although our ability to
count, calculate and measure things in the world is prac-
tically and psychologically critical, it is important to
emphasize that mathe�maticians do not spend their time
making calculations. The real challenge of mathe�matics
is to construct an argument.
Pythagoras’ famous Theorem provides an excellent
example of how the nature of mathe�matical thought is
4 MATHEMATICAL THOUGHT

widely misunderstood. Most educated people know that

given any right-angled triangle, we can use the formula
a 2 + b 2 = c 2 to find all three lengths, even if we have only
been told two of them. As they have been asked to repeat-
edly perform this kind of calculation, people mistakenly
conclude that mathe�matics is all about applying a given
set of rules. Unfortunately, far too few people can give a
convincing explanation as to why Pythagoras’ Theorem
must be true, despite the fact that there are literally
hundreds of different proofs. One of the simplest argu-
ments for showing that it’s true hinges around the following
diagram:

a a2 a

b 2
= b
c2
b

b a a b

Pythagoras: The shapes on either side of the equals sign

are contained inside a pair of identical squares, whose sides
are a + b units wide. The one on the left contains a square
a units wide, a square b units wide plus four right-angled
triangles. The one the right contains a square c units wide
plus four right-angled triangles. We can convert the picture
on the left into the one on the right simply by moving the
four triangles, and moving a shape does not change its
area. Since the white area is the same in each of the draw-
ings, this demonstrates that a 2 + b 2 = c 2 for any right-angled
triangle.

Sceptic: How can you be certain that we always get a

square on the right-hand side? More specifically, how do
 INTRODUCTION5

you know that your triangles always meet at a point, what-

ever the values of a and b?

Pythagoras: Both drawings are of equal height (a + b units

high). This tells us that the two triangles that are just to
the right of the equals sign must touch at a point, because
they only just manage to fit inside the containing square.
Similarly, the two triangles on the bottom of the right-hand
side touch at a point, because the total length along this
side is a + b, which equals b + a (the width of the containing
squares).

Sceptic: OK, but how do you know that the triangles on

the right-hand side always meet at right angles? In other
words, how do you know that the shape on the right-hand
side is really a square?

Pythagoras: You agree that we have four sides of equal

length, and all four corners are the same?

Sceptic: Yes. Rotating the picture on the right by 90°, 180°

or 270° leaves the diagram unchanged.

Pythagoras: And despite these facts you still aren’t

convinced that it’s a square? No wonder they call you a
sceptic!

My aim in writing this book is to show how the language

of mathe�matics has evolved, and to indicate how mathe�
matical arguments relate to the broader human adventure.
This book is related to the work of various philosophers
(particularly Ludwig Wittgenstein), but it is not a history
of non-mathe�matical ideas, or an attempt to draw battle
lines between conflicting ‘big pictures’ from the philosophy
of math. I will have succeeded if my writing provokes
thought, but I have also tried to argue against the idea that
6 MATHEMATICAL THOUGHT

mathe�maticians discover facts about abstract objects, just

like scientists discover facts about physical objects. Mathe�
matical language does not make sense because abstract
objects existed before mathe�maticians! Contrariwise, we
can only become cognisant of abstract objects because
mathe�matical language is something that humans can actu-
ally use.
It is fundamental to human understanding that our theo-
ries or accounts of the physical world are expressed through
language. People make statements of fact, and the reflective,
systematic study of our ability to make statements leads
us into the world of math. Indeed, our understanding of
mathe�matics always begins with a clear, comprehensible
case, from which we form a notion of the abstract prin-
ciples at play. For example, children learn the counting
song, and they are then initiated into the practice of
counting actual, physical objects. This concrete experience
grounds our sense of number, as we abstract away from
a particular experience of counting things, justifiably
believing that we could set about counting any collection
of objects. That is to say, number words become meaningful
for an individual as they use those words on some particular
occasion, in the presence of actual, countable objects, but
once that person has acquired a language, the language
itself enables them to think in terms of number, whatever
they might wish to count.
Some people mistakenly believe that to do mathe�matics,
we simply need to follow certain rules. I suspect that people
arrive at this erroneous position because in order to satisfy
their teachers and examiners, all they need to do is apply
some rules correctly. In fact, higher mathe�matics is an
essentially creative pursuit that requires imagination. That
said, rules are never far behind our creative insights, because
in order to contribute to the body of mathe�matical know�
ledge, mathe�maticians need to be able to communicate
their ideas. The formal discipline that we require to fully
 INTRODUCTION7

state our arguments is an essential constraint on the shape

of mathe�matical knowledge, but the mathe�matics that we
know also reflects the problems, challenges and cultural
concerns that have motivated the various members of the
mathe�matical community.
I hope that reading this book persuades you that mathe�
maticians are explorers of patterns, and formal, logical
proofs that can be methodically checked are the ultimate
test of mathe�matical validity. The clarity of a strictly formal
proof is a beautiful thing to comprehend, and I think it
is fair to say that an argument is only matheÂ�matical if it is
apparent that it can be formalized. However, while we
can gain a sense of understanding by learning to use a
particu�lar formal scheme, it is certainly possible to check
each step in a formal argument without understanding
the subject at hand. Indeed, a computer could do it, even
though a computer is no more a mathematician than a
photocopier is an artist.
I have taken a more intuitive approach as my aim is not
to train the reader in the appropriate formal techniques,
but simply to make the heart of each argument as compre-
hensible as possible. That said, the subject matter of this
book is subtle and sophisticated, so there is no escaping
the need to take certain arguments carefully and slowly.
Mathematics is a subject where you must read the same
sentence several times over, and as with poetry, you must
read at an appropriate pace.
Over the course of my book I trace out a history of
mathe�matical practice, with a focus on conceptual innova-
tions. I do not claim to have covered all of the key ideas,
but I have tried to sketch the major shifts in the popular
understanding of math. The book is structured by a combi-
nation of historical and thematic considerations, and its
thirteen chapters can be grouped into four main sections.
I begin by discussing the number concept, from a specu-
lative and rhetorical account of prehistoric rituals to
8 MATHEMATICAL THOUGHT

mathe�matics in the ancient world. I examine the relation-

ship between counting and the continuum of measurement,
and try to explain how the rise of algebra has dramatically
changed our world.
The first section ends with ‘matheÂ�matical padlocks’ of
the modern era, but in the second section I step back in
time. More specifically, I discuss the origins of calculus,
and the conceptual shift that accompanied the birth of
non-Euclidean geometries. In short, I try to explain how
modern mathe� matics grew beyond the science of the
Greeks, the Arabs, or other ancient cultures.
In the third section I turn to the most philosophically
loaded terms in mathe�matics: the concept of the infinite, and
the fundamentals of formal logic. I also discuss the genius
of Alan Turing, and try to elucidate the subtle relation-
ship between truth, proof and computability. In particular,
I focus on a proof of the infinite richness of addition
and multiplication (as demonstrated by Matiyasevich’s
Theorem), and examine Kurt Gödel’s celebrated theorems
on the Incompleteness of Arithmetic.
In the final section I consider the role of mathe�matics
in our attempts to comprehend the world around us. In
particular, I describe the importance of models, and the
role of mathe�matics in biology. I conclude by taking a step
back from any particular theorem, and try to use what we
have learned about mathe�matical activity to think about
thinking in general.
One of the challenges in writing this book was doing
justice to the weight of simple, teachable statements.
Some statements are like paper darts: you can follow
them with a lightness of contemplation, if you know to
where they float. If your only guide is to cling to the
words themselves, they cannot carry you, as their target
has not been spoken. Other statements possess gravitas,
as in their accessible simplicity they act like stones,
pulling us down to what can and has been said.
 INTRODUCTION9

Unfortunately, people tend to underestimate the value

of simple, understandable statements, as we more often
praise ideas by suggesting they are hard to grasp. As the
great thinker Blaise Pascal remarked in The Art of
Persuasion, ‘One of the main reasons which puts people
off the right way they have to follow is the concept they
first encounter that good things are inaccessible by being
labelled great, mighty, elevated, sublime. That ruins
everything. I would like to call them lowly, commonplace,
familiar. These names befit them better. I hate these
pompous words …’
The great edifice of mathe�matical theorems has a crys-
talline perfection, and it can seem far removed from the
messy and contingent realities of the everyday world.
Nevertheless, mathe�matics is a product of human culture,
which has co-evolved with our attempts to comprehend
the world. Rather than picturing mathe�matics as the study
of ‘abstract’ objects, we can describe it as a poetry of
patterns, in which our language brings about the truth
that it proclaims. The idea that mathe�maticians bring about
the truths that they proclaim may sound rather myste-
rious, but as a simple example, just think about the game
of chess. By describing the rules we can call the game of
chess into being, complete with truths that we did not
think of when we first invented it. For example, whether
or not anyone has ever actually played the game, we can
prove that you cannot force a competent player into
checkmate if the only pieces at your disposal are a king
and a pair of knights. Chess is clearly a human invention,
but this fact about chess must be true in any world where
the rules of chess are the same, and we cannot imagine
a world where we could not decide to keep our familiar
rules in place.
Mathe�matical language and methodology present and
represent structures that we can study, and those structures
or patterns are as much a human invention as the game of
10 MATHEMATICAL THOUGHT

chess. However, mathe�matics as a whole is much more

than an arbitrary game, as the linguistic technologies that
we have developed are genuinely fit for human purpose.
For example, people (and other animals) mentally gather
objects into groups, and we have found that the process
of counting really does elucidate the plurality of those
groups. Furthermore, the many different branches of
mathe�matics are profoundly interconnected, to art, science
and the rest of mathe�matics.
In short, mathe�matics is a language, and while we may
be astounded that the universe is at all comprehensible,
we should not be surprised that science is mathe�matical.
Scientists need to be able to communicate their theories,
and when we have a rule-governed understanding, the
instructions that a student can follow draw out patterns
or structures that the mathematician can then study. When
you understand it properly, the purely mathe�matical is
not a distant abstraction – it is as close as the sense that
we make of the world: what is seen right there in front
of us. In my view, math is not abstract because it has to
be, right from the word go. It actually begins with linguistic
practice of the simplest and most sensible kind. We only
pursue greater levels of abstraction because doing so is a
necessary step in achieving the noble goals of modern
mathe�maticians.
In particular, making our mathe�matical language more
abstract means that our conclusions hold more generally,
as when children realize that it makes no difference
whether they are counting apples, pears or people. From
generation to generation, people have found that numbers
and other formal systems are deeply compelling: they can
shape our imagination, and what is more, they can enable
comprehension. The story of math is fascinating in its
own right, but in writing this book I hoped to do more
than simply sketch a history of mathe�matical ideas. I am
convinced that the history and philosophy of math provide
 INTRODUCTION11

an invaluable perspective on human nature and the nature

of facts, and I hope that my book conveys the subject’s
cultural, aesthetic and philosophical relevance, as well as
the compelling drama of mathe�matical discovery.
 Chapter 1:
BEGINNINGS
‘There can be no doubt that all our knowledge begins
with experience. … But though all our knowledge
begins with experience, it does not follow that it all
arises out of experience.’
Immanuel Kant, 1724–1804

Language and Purpose

Researchers working with infants and animals have found
compelling evidence that we have an innate sense of quan-
tity. More specifically, humans, birds and many other
animals can recognize when a small collection has changed
in size, even if they do not observe the change taking
place. For example, birds can recognize when one of their
eggs is missing, even if they did not witness the egg’s
removal. Similarly, many animals will consistently pick
the larger of two collections when they are given a choice.
Presumably, this sensitivity to quantity is a necessary
precondition for the development of math, and it is inter-
esting to note that some animals are quicker than humans
at intuitively sensing differences in quantity. Nevertheless,
although such abilities constitute evidence for animal intel-
ligence, it is rather inaccurate to claim that ‘birds count
their eggs’.
14 MATHEMATICAL THOUGHT

I would argue that ‘proto-matheÂ�matical’ thinking can

only begin once we have developed language, and that this
kind of understanding is fundamental to many types of
human behaviour, not just what we ordinarily think of as
‘understanding math’. Of course, any account of the life-
style of our Stone Age ancestors is bound to be highly
speculative, but despite the lack of conclusive evidence, I
think it is helpful to imagine how our ancestors first devel-
oped rational capabilities, and the enormously complex
thing that we call language.
Humans are not the only animals to use tools, and for
millions of years our primate ancestors extended their
abilities by utilizing what was found at hand. Sticks,
stones, fur, leaves, bark and all manner of food stuffs
were used in playful ways that we can only guess. Flesh
was scraped from fur, sticks were sharpened and adapted
to a purpose, and stones were knapped to produce effec-
tive butchery kits. Most importantly, about 1.8 million
years ago Homo erectus started using fire to cook, which
reduced the amount of energy needed for digestion,
making it possible to grow larger brains and smaller diges-
tive tracts.
As human intelligence evolved, our vocalizations and
patterns of interaction developed into something that
deserves to be called language. One very plausible specu-
lation is that more intelligent hominids were more
successful in making the most from the complex dynamics
of their social situation, providing the selective advantage
that led to increasing intelligence. In any case, communi-
cative aspects of modern language are common to many
animals. For example, many animals can convey a state of
panic when they see a predator. It is therefore clear that
complex, communicative forms of interaction massively
predate the development of the proto-mathe�matical, or
any conception that language might be the thing of interest,
rather than the people who were making the sounds.
 BEGINNINGS15

This idea is worth elaborating, so as an example of

how a culture of interactions can evolve to something
greater, imagine a woman who lives in a community with
a particular culture of responses: Men who give me tasty
food will hear me hum, but those who grab me without
giving me tasty food will hear me growl. If a man was
trying to establish a sexual relationship with this woman,
he would want to hear the humming sound, because a
woman who hums is much more likely to be interested
in sex than a woman who growls. Consequently, the man
would wilfully do a bit of cooking prior to any sexual
advance, acting to establish the circumstances that he
associates with the humming sound.
By living in such a social context, we came to feel the
sense of our own actions. In other words, we responded
to changing social occasions with increasingly sophisticated
forms of motivated strategy, and were mindful of complex
goals whose achievement required actions beyond those
in the immediate present. For example, the occasion of
preparing an especially tasty meal is not the same occasion
as hearing a woman make the humming sound, but we see
that one is motivated by the other.
Social norms and the vocabulary of praise and blame
are both potent forces for shaping the imagination. It is
absolutely fundamental that we find words with which to
judge our actions, and our judgements work with words.
An example of this endlessly subtle process can be found
in the following conversation:

‘Let’s break into that house.’

‘I don’t know, that seems like a bad idea.’
‘Go on, don’t be chicken.’

We are fearful that our reasoning will compel us to name

ourselves cowards, idiots, or many other kinds of undesired
utterance. The will to avoid such experience is part of our
16 MATHEMATICAL THOUGHT

humanity, as is the compelling nature of the reasons that

we find. As Blaise Pascal observed, we are most compelled
by the reasons of our own devising, but such complexities
can be closely shared and instinctively taught to others.
The caveman is compelled by the fact that the woman has
established reasons for growling, if he fails to meet the
expectations of an established practice. It is a process of
judgement that he has a feeling for, and the weight of the
utterance is that it is not felt to be arbitrary. Similarly, our
potential thief is pulled by the fact that he too can reason
himself a coward, and does not wish to do so.
However, it is crucial to note that in each of these exam-
ples, the significance of an utterance is inseparable from
the fact that another person has decided to say the state-
ment in question. In other words, we reached the point
of very sophisticated communication long before we ever
considered ‘language’ as a thing in itself, separate from the
people who were speaking.

Human Cognition and the Meaning of Math

The literature of mathe�matics is largely composed of argu-
ments of the form ‘If A and B are true, then it follows
that C is also true’, and it is worth pausing to wonder
how it is that humans developed the capacity for deductive
reasoning. We are not the only animals who are alert to
the range of possible consequences of our actions, and we
might suppose that our grasp of logical consequence is
only possible because we have evolved the cognitive abili-
ties needed to predict the practical consequences of the
things that we might do. For example, imagine a hungry
ape looking at another ape with some food. It might think
to itself, ‘If I grab the food, that big guy will hit me. I
don’t want to be hit, so therefore I should restrain myself
and not grab the food.’
The fact that we use language fundamentally changes
the character of our reasoning, but it is easy to believe
 BEGINNINGS17

that imagining the consequences of potential actions is

an ancient ability that confers an evolutionary advantage.
However, it is hard to see how the evolution of this kind
of ‘reasoning’ about actions and their consequences could
enable abstract thought. After all, the scenario I have
described is all about judging the way to behave in a
complex context, where any new information might
change our prediction of what will happen next, and we
ought to be open to noticing further clues. For example,
if our ape saw the other ape make a friendly gesture, it
might be wise to grab the food instead of letting it go.
That is very different from working out logical conse-
quences, where one thing follows from another,
regardless of any further information that could plausibly
come our way.
Because the social cunning of animals depends on their
grasp of entire contexts, where there are always further
clues, it is difficult to see how that kind of understanding
could provide the cognitive abilities that a mathematician
requires. In contrast, our capacity for spatial reasoning
is much less open ended, and human beings do not need
to be trained to make valid spatial deductions. For
example, suppose that there is a jar inside my fridge.
Now suppose that there is an olive inside the jar. Is the
olive inside the fridge? The answer is yes, of course the
olive is inside the fridge, because the olive is in the jar,
and the jar is in the fridge. Now imagine that the jar is
in the fridge but the olive is not in the fridge. Is the
olive in the jar? Of course not, because the jar is in the
fridge, and I have just told you that the olive is not in
the fridge.
In reasoning about the location of the olive, it is suffi-
cient to bear a thin skeleton of facts in mind. Additional
information will not change our thinking, unless it contra-
dicts the facts that form the basis of our deduction. Also
note that in order to make these deductions, we do not
18 MATHEMATICAL THOUGHT

need to be initiated into some or other method of symbol-

izing. All humans can reason in this way, so it is plausible
to claim that there are innate neural mechanisms that
underpin our grasp of the logic of containers. Of course,
in order to pose these questions I need to use some words,
but humans (and other animals) find it very easy to under-
stand that containers have an inside and an outside, and
this kind of understanding provides a structure to our
perceptual world.
There is strong empirical evidence that before they learn
to speak, and long before they learn mathe�matics, children
start to structure their perceptual world. For example, a
child might play with some eggs by putting them in a
bowl, and they have some sense that this collection of eggs
is in a different spatial region to the things that are outside
the bowl. This kind of spatial understanding is a basic
cognitive ability, and we do not need symbols to begin to
appreciate the sense that we can make of moving something
into or out of a container. Furthermore, we can see in an
instant the difference between collections containing one,
two, three or four eggs. These cognitive capacities enable
us to see that when we add an egg to our bowl (moving
it from outside to inside), the collection somehow changes,
and likewise, taking an egg out of the bowl changes the
collection. Even when we have a bowl of sugar, where we
cannot see how many grains there might be, small children
have some kind of understanding of the process of adding
sugar to a bowl, or taking some sugar away. That is to
say, we can recognize particular acts of adding sugar to a
bowl as being examples of someone ‘adding something to
a bowl’, so the word ‘adding’ has some grounding in
physical experience.
Of course, adding sugar to my cup of tea is not an
example of mathe�matical addition. My point is that our
innate cognitive capabilities provide a foundation for our
notions of containers, of collections of things, and of
 BEGINNINGS19

adding or taking away from those collections. Furthermore,

when we teach the more sophisticated, abstract concepts
of addition and subtraction (which are certainly not
innate), we do so by referring to those more basic, phys-
ically grounded forms of understanding. When we use
pen and paper to do some sums we do not literally add
objects to a collection, but it is no coincidence that we
use the same words for both mathe�matical addition and
the physical case where we literally move some objects.
After all, even the greatest of mathe�maticians first under-
stood matheÂ�matical addition by hearing things like ‘If you
have two apples in a basket and you add three more, how
many do you have?’
As the cognitive scientists George Lakoff and Rafael
Núñez argue in their thought-provoking and controversial
book Where Mathematics Comes From, our under�standing
of mathe�matical symbols is rooted in our cognitive capa-
bilities. In particular, we have some innate understanding
of spatial relations, and we have the ability to construct
‘conceptual metaphors’, where we understand an idea or
conceptual domain by employing the language and patterns
of thought that were first developed in some other domain.
The use of conceptual metaphor is something that is
common to all forms of understanding, and as such it
is not characteristic of matheÂ�matics in particular. That is
simply to say, I take it for granted that new ideas do not
descend from on high: they must relate to what we already
know, as physically embodied human beings, and we
explain new concepts by talking about how they are akin
to some other, familiar concept.
Conceptual mappings from one thing to another are
fundamental to human understanding, not least because
they allow us to reason about unfamiliar or abstract things
by using the inferential structure of things that are deeply
familiar. For example, when we are asked to think about
adding the numbers two and three, we know that this
20 MATHEMATICAL THOUGHT

operation is like adding three apples to a basket that

already contains two apples, and it is also like taking
two steps followed by three steps. Of course, whether
we are imagining moving apples into a basket or thinking
about an abstract form of addition, we don’t actually
need to move any objects. Furthermore, we understand
that the touch and smell of apples are not part of the
facts of addition, as the concepts involved are very
general, and can be applied to all manner of situations.
Nevertheless, we understand that when we are adding
two numbers, the meaning of the symbols entitles us to
think in terms of concrete, physical cases, though we are
not obliged to do so. Indeed, it may well be true to say
that our minds and brains are capable of forming abstract
number concepts because we are capable of thinking
about particu�lar, concrete cases.
Mathe�matical reasoning involves rules and definitions,
and the fact that computers can add correctly demonstrates
that you don’t even need to have a brain to correctly
employ a specific, notational system. In other words, in a
very limited way we can ‘do matheÂ�matics’ without needing
to reflect on the significance or meaning of our symbols.
However, matheÂ�matics isn’t only about the proper, rule-
governed use of symbols: it is about ideas that can be
expressed by the rule-governed use of symbols, and it
seems that many mathe�matical ideas are deeply rooted in
the structure of the world that we perceive.

Stone Age Rituals and Autonomous Symbols

Mathematicians are interested in ideas, not just the manipu-
lation of ‘meaningless’ symbols, but the practice of
mathe�matics has always involved the systematic use of
symbols. Mathe�matical symbols do not merely express
mathe�
matical ideas: they make mathe� matics possible.
Furthermore, even the greatest mathe�maticians need to be
taught the rules before they can make a contribution of
 BEGINNINGS21

their own. Indeed, the very word mathe�matics is derived

from the Greek for ‘teachable knowledge’. The question
is, how and why did human cultures develop a system of
rules for the use of symbols, and how did those symbols
change our lives?
It seems fair to claim that the most fundamental and
distinctive feature of human cognition is our boundless
imagination. We don’t just consider our current situation,
we imagine various ways that the future could pan out,
and we think about the past and how it could have been
different. In general, we inhabit imaginable worlds that
follow certain principles, and compared to other animals,
our thoughts are not overly constrained by our current
situation or perceptions. In particular, we can think about
objects that are not ready at hand, and it is reasonable to
assume that in the distant past, our ancestors would feel
distressed if their desire to act was frustrated by the absence
of some object or tool.
As an animal might express the presence of predators,
our ancestors would gesture, ‘I am missing a flint’. By using
their vocal cords, facial expressions and bodily posture,
they would express their motivated looking. Fellow primates
would respond to this signal in a manner appropriate to
the occasion, having an empathic grasp of what it is to
search in such a fashion. Over countless generations, our
ancestors must have developed ways of conveying a desire
for certain objects, even though those objects were currently
out of sight. Furthermore, at some point our ancestors
must have made the vital step of imbuing those expressive
gestures with an essentially mathe�matical meaning. This
remarkable feat was not achieved by the discovery of
abstract objects: it was achieved by developing rituals.
Suppose, for example, that there was a pre-existing form
of expression that conveyed the speaker’s irritation over
missing a flint. Now imagine the earliest people running
their hands over their treasured tool kit of flints. As a
22 MATHEMATICAL THOUGHT

person checked their tools time and time again, they may
have expressed their familiarity with these objects by
reciting a sequence of names. As each tool was touched
in turn, we can imagine our ancestors repeating a distinc-
tive sequence of rhythmic speech, with one word for each
tool, like someone saying ‘Eeny-meeny-miney-mo’. If this
ritual was left unfinished by the time there were no more
objects left to touch, Stone Age humans could see that
they had a reason for making the gesture ‘I am missing a
flint’.
This is not the same as counting with an abstract concept
of number, as whether or not they deserve to be called
mathe� maticians, even the youngest of children will not
mistake ‘Eeny-meeny-miney’ for ‘Eeny-meeny-miney-mo’.
Sensitivity to the incompleteness of a habitual action is
clearly innate, and this is very close to the sing-song voice
of baby talk and our natural sense of rhythm. When our
ancestors expressed this failure of correspondence between
the present tools and the familiar ritual, the other people
would also know that something was missing, because they
recognized that the ritual had been performed correctly. In
other words, it is the ritual that tells us that a flint is missing,
and not just the individual who performed it.
By possessing such a clear sense of justified speech,
people could find grounded meanings in their utterances,
and strategically approach the issues that concerned them
(namely, ‘Is it the case that all the familiar devices are
present?’). In this way, expressive gesture came to signify
more than an immediate cultural resonance, and the primal
gestures that conveyed the sense ‘I am irate over a
missing flint’ become something deeper. The common
sense of valid reasoning gave new weight to our commu-
nications, as by means of the common practice a statement
of fact can be established. In particular, note that the
fact’s appearance in the world is dependent on the prac-
tice itself, not the individual who carried it out. This
 BEGINNINGS23

process of language speaking for itself emphatically does

not require the abstract concept of number, and I would
argue that proto-� mathe�matical thought had a gradual
evolution that predates counting by many tens of thou-
sands of years.
The origin of number words as we understand them
today isn’t known for certain, but there are some interesting
theories supported by linguistic evidence. It may be that
practices closely related to counting arose spontaneously
throughout the world, more or less independently from
place to place. However, the mathematician and historian
of science Abraham Seidenberg (1916–1988) proposed that
counting was invented just once, and then spread across
the globe. Number words are often related to words for
body parts, and Seidenberg claimed that the similarities in
number words from very distant places constitutes evidence
for his theory. He also made the intriguing observation
that in almost every numerate culture, there is an ancient
association between the odd numbers and the male, while
the even numbers are female.
There is certainly plenty of evidence that animals are
aware of who is first in the pecking order, who is second,
third and so on. Seidenberg suggests that counting origi-
nated in rituals based on rank and priority, arguing that
counting ‘was frequently the central feature of a rite, and
that participants in the rite were numbered’. Whether the
first numbers or number-like words were applied to an
ordered sequence of people, or used to assess the plurality
of a collection of tools, it is clear that the human mind
has been capable of learning how to count for tens of
thousands of years.
It is important to note that mathe�matics is not a universal
human trait, as some cultures have no words for numbers
larger than three. Furthermore, some people have a highly
cultured sense of quantity even though they cannot really
count, as their language has too few number words.
24 MATHEMATICAL THOUGHT

For  example, the Vedda tribesman of Sri Lanka used to

‘count’ coconuts by gathering one twig for each coconut.
The people who did this clearly understood that there was
a corresponding plurality between the twigs and the coco-
nuts, but if asked how many coconuts a person had
collected, they could only point to the pile of twigs and
say ‘that many’.
The first expressions of quantity are lost deep in the
mists of time, but it is surely safe to assume that long
before the advent of abstract number words, people had
one word for ‘hand’, and a different word for ‘pair of
hands’. The move from words that convey quantities of
specific physical things to an abstract or universally appli-
cable language of number is an example of logic at work.
That is to say, once we have a sequence of words for
‘counting’ something or other, it is possible to recognize
that it is the words themselves that form an ordered, rule-
governed sequence, and they need not be bound to any
particular thing that people are used to counting.

Making Legible Patterns

As human beings we live in a world of people and things,
sights and sounds, tastes and touches. We don’t see a pattern
on our retina: we see people, trees, windows, cars, and
other things of human interest. This relates to the fact that
we use language to think about our world, doing things
like naming objects, or creating accounts of people or situ-
ations. My point is that human beings conceptually
structure the perceptual flux in which we live, so our use
of symbols, images and words is central to making sense.
For example, imagine a young child drawing a picture
of Daddy: a stick-man body with a circle for a head, two
dots for eyes and a U-shape for a smile. It is significant
that each part of this drawing can be named, as we under-
stand, for example, that two dots can represent the eyes.
The art of 30,000 bc was probably somewhat similar to a
 BEGINNINGS25

child’s drawing, not because our ancestors were simple

minded, but because drawing nameable things is such a
basic, human skill. Indeed, we can say that children’s draw-
ings are understandable precisely because we can talk our
way about them.
Our ancestors decorated caves with vivid illustrations
of large mammals, but they also used simpler marks
(arrangements of dots, V-shapes, hand prints, etc.). Just as
a child might not need to draw ears and a nose before
their marks become a face, so the caveman artist may have
drawn some tusks and already seen a mammoth. Such
stylized, intelligible drawings are not the same as writing,
but there is a related logic of meaningful marks, and it is
surely safe to assume that our ancestors talked about their
drawings. As another example of Stone Age pattern making,
archaeologists in central Europe found a shinbone of a
wolf marked with fifty-seven deeply cut notches. These
marks were arranged in groups of five, and carbon dating
indicates that this bone is over 30,000 years old.
People and animals alike are good at spotting patterns.
In particular, many birds and mammals are demonstrably
sensitive to changes in quantity. Time and time again
humans have discovered a basic technique for clearly
showing quantity: we group elements together in a regular
way, so that a single ‘counting’ operation is broken into
a combination of simpler assessments. For example, we
can recognize four as a pair of pairs, or ten as a pair of
fives. This means that even before one can count, it is
easier to assess the plurality of things if they are arranged
in regular groups, rather than scattered in a disordered
fashion. Furthermore, once we have words for numbers,
this practical idea can lead us to the concept of multiplica-
tion. This suggests that by the time that our ancestors had
articulated an abstract concept of number, people were
counting by dividing their things into regular groups and
counting off five, ten, fifteen, twenty (say). In other words,
26 MATHEMATICAL THOUGHT

the times tables may be just as old as abstract number

words themselves.
Mathematics has been described as the language of
patterns, and there is a deep relationship between our
innate tendency to recognize patterns, and our cultured
sense of shape and number. Long before we developed
proper number words, ancient peoples must have recog-
nized the reality of patterns, and explored some formal
constraints. Very ancient peoples must have known that
triangles can be arranged to produce certain shapes or
patterns, but there are some shapes (e.g. a circle) that
cannot be made from triangles. People have been
exploring patterns for tens of thousands of years, using
their �material ingenuity (e.g. pottery and basket weaving),
music, dance and early verbal art forms. For example, it
is an evident truth that if you clap your hands every
second heartbeat, and stamp your feet every third one,
there necessarily follows one particular combined rhythm
and not others.

The meaningfulness of mathe�matical statements did not

appear from nowhere, and we don’t need ‘proper’ matheÂ�
matics to first be aware of quantity and shape. Before
people developed counting or abstract number words, they
might have used a phrase like ‘as many bison as there are
berries on a bush’, or shown a quantity with an artistic
abundance of marks. After all, our artist ancestors surely
strove to be eloquent, and someone must have been keen
to show the size of a massive herd. Many generations later,
the advent of counting gave birth to the concept of number:
a great advance in our ability to conceptually distinguish
between different pluralities.
At first our ancestors must have only used their counting
 BEGINNINGS27

words in particular situations, but over time we realized

that in order to count any collection of objects, we do not
need to keep our eyes on the qualities of the objects them-
selves. In a sense we can count any collection of objects,
so long as we can give each object a name (e.g. by attaching
labels). From that point on we can play the counting game
simply by reviewing the sequence of names that we
ourselves have given, even if our labels become detached
from their associated objects.
Many, many social needs require calculation and number,
and over the long arc of prehistory mathe�matics continued
to evolve along with the social systems that supported
mathe�matical techniques. In return, more sophisticated
mathe�matics enabled more complex social structures. For
example, an inheritance cannot be distributed unless certain
facts about division are known, and at a more sophisticated
level, tax rates and monetary systems are literally incon-
ceivable without the concept of number.
The development of agriculture revolutionized our
ways of life, and according to many ancient historians,
geometry (Greek for earth-measuring) came into being
as people needed to speak authoritatively and unconten-
tiously about the size of fields. In particular, every year
Egyptian mathe�maticians needed to replace the property
markers that were washed away from the flood plains of
the Nile. The story of geometric techniques arising with
the need to measure fields is certainly plausible, but there
were also prehistoric traditions for communicating
specific plans for temples and other buildings, which
necessarily involves a language of shape, so this claim
may not actually be correct. What is known for certain
is that by the third millennium bc, civilizations with
sophisticated mathe�matical practices had developed along
the fertile banks of many of the world’s great rivers. The
Nile, the Tigris and Euphrates, the Indus, the Ganges,
the Huang He and the Yangtze all provided ground for
28 MATHEMATICAL THOUGHT

these new ways of life. Furthermore, mathe�matics had a

central role to play in the emergence of large-scale civi-
lizations, not least in the development of trade, measured
and planned construction, administrative techniques,
astronomy and time keeping.

The Storage of Facts

You might guess that the oldest recorded numbers would
be fairly small, but in fact archaeologists have found that
some of the oldest unambiguously numerical records refer
to the many thousands of cattle that Mesopotamian kings
had claimed through war. It is also interesting to note that
no civilization has ever become literate without first
becoming numerate, and almost every numerate civilization
is known to have used some kind of counting board or
abacus. In other words, tools for recording the counting
process are much, much older than tools for recording
speech. Greeks and Romans used loose counters, the
Chinese had sliding balls on bamboo rods, and the Ancient
Hindu mathe� maticians used dust boards, with erasable
marks written in sand.
Because of geographical distance, it is assumed that the
development of mathe� matics in the Americas was
completely independent from that of Europe and Asia. It
is therefore remarkable to note that around 1,500 years
ago, the Maya were employing number symbols much as
we do today. The Incas were a more recent civilization
than the Maya, and although the Incas never developed a
system for making records of the spoken word, they could
record information by using a system of knotted cords
called quipus. These were colour coded to represent the
various things that were counted, and scribes would read
the clusters of knots by pulling the cord through their
hands. Each cluster of knots represented a digit from one
to nine, and a zero was represented by a particularly large
gap between clusters.
 BEGINNINGS29

Quipus with as many as 1,800 cords have been found,

and as different-coloured threads signified different kinds
of information, these fascinating objects demonstrate that
sophisticated record keeping is not the exclusive domain
of the written word. Very few quipus have survived, and
given that they are far more distant than the Incas, it is
plausible that many prehistoric civilizations possessed
long-lost means of embodying data. It is worth remem-
bering that only a tiny subset of equipment survives the
ravages of time, and if people kept records by arranging
pebbles or making scratches on bark, we might have no
way of knowing. As the mathe�matical historian Dirk Struik
has suggested, the builders of ancient monuments like
Stonehenge must have had some idea of what it was that
they were building. Many of the regular features of this
construction cannot be accidental, and it seems highly
unlikely that the builders didn’t know what they would
do with the stones until they got them on site. Their
means of communicating intent may well have involved
physical artefacts that embodied data unambiguously. For
example, they may have made shadow casting models that
showed exactly how many stones were going to be
arranged, and their orientation in relation to the sun’s
path.
The ancient civilizations of Asia used bamboo, bark and
eventually paper to keep records of numerical information.
Although the origins of their mathe� matical knowledge
remain obscure, certain pieces of Chinese mathe�matics have
been faithfully passed down through hundreds of genera-
tions. For example, consider the following ‘magic square’,
known as the Lo-Shu. Legend has it that this mathe�matical
pattern emerged from the Yellow River on the back of a
giant turtle about 4,000 years ago. We can’t really be certain
about the true age of the Lo-Shu, but we do know that it
was considered to be truly ancient knowledge as far back
as the Han dynasty (206 bc–ad 220).
30 MATHEMATICAL THOUGHT

Every row, column and diagonal contains numbers that

add up to fifteen. The central cross of odd numbers ensures
that every line contains either one odd number or three
odd numbers (the ‘yang’ of the Lo-Shu). The four corners
contain even numbers (the ‘yin’ of the Lo-Shu), completing
this sacred symbol of cosmic harmony and balance.
Perhaps the most famous book of Ancient Chinese
mathe�matics is the Nine Chapters on the Art of Calculation.
This book was written c.â•›200 bc, and the 246 problems it
contained were used to test and train potential civil
ser�vants. Arguably the most impressive feature of this
mathe�matical tradition is the fact that the Ancient Chinese
routinely solved systems of linear equations. For example,
suppose we have two different kinds of weights, coloured
red and blue. If two reds and three blues weigh eighteen
units, while two reds and two blues weigh sixteen units,
how much does one red weigh? In modern notation we
have:

2r + 3b = 18 and 2r + 2b = 16.

The left-hand side of the first equation minus the left-hand

side of the second is equal to b. The right-hand side of
the first equation minus the right-hand side of the second
is equal to 2. Therefore b, the weight of one blue is 2
units. Substituting this value into either equation shows
that r╛=╛6. Modern mathe�maticians and the Ancient Chinese
alike solve this kind of problem by adding and subtracting
 BEGINNINGS31

equations (or equation-like forms) in order to eliminate

unknowns, and the same principle can be naturally
extended to situations with more than two unknowns.
The Chinese have been solving problems like this for over
2,000 years, but remarkably, this extremely useful tech-
nique was not known in the West until the beginning of
the nineteenth century, when it was independently
invented by one of the giants of modern mathe�matics,
Carl Freidrich Gauss.

Babylon, Egypt and Greece

Around 1650 bc, a scribe named Ahmes copied out a
text from the Twelfth Dynasty of Egypt (c.â•›1990–1780 bc).
It seems that Ahmes was mathe�matically educated, which
makes him the most ancient mathematician whose name
we know. The ‘Ahmes Papyrus’ contains eighty-five prob-
lems, and it demonstrates the Egyptians’ ability to solve
problems that involve unknown quantities, as well as
their systematic use of fractions of the form 1/n. For
example, the ‘Ahmes Papyrus’ contains the line ‘a heap
plus a quarter of that heap again makes fifteen’. Through
trial and error Ahmes realized that the heap must be size
twelve (because twelve plus a quarter of twelve is fifteen).
The manuscript also states that the area of a circle of
diameter nine can be taken to be equal to the area of a
square of width eight. This represents an error in the
estimate of π of only 0.6%.
Although many modern teenagers know significantly
more mathe�matics than any Ancient Egyptian, it is clear
that the Egyptians knew how to achieve a fairly rich variety
of computational tasks. In particular, they had a long tradi-
tion of measuring the position of the stars and planets,
and after the annual flooding of the Nile they were more
than capable of replacing the property markers that had
been washed away. A particularly ancient measurement
technique involves a form of set square, made from a loop
32 MATHEMATICAL THOUGHT

of rope marked into twelve equal lengths. The Egyptians

and other ancient peoples knew that if you stretched out
a rope with one side of 3 units, one side of 4 units and
one side of 5 units, you invariably produced a right-angled
triangle. Indeed, the word ‘hypotenuse’ is derived from
the Greek for ‘stretched against’, reflecting this ancient
technology. The Egyptians also knew that 32 + 42 = 52, and
like other ancient peoples they were familiar with many
other Pythagorean triangles.
The mathe�maticians of Mesopotamia were probably
more advanced than their Egyptian contemporaries. By
the time that Hammurabi became king of Babylon (c.â•›1750
bc), his people had developed powerful methods for finding
areas and volumes. Indeed, the Babylonians were familiar
with the empirical content of Pythagoras’ Theorem over
a thousand years before Pythagoras himself was born. Our
main sources of evidence concerning Babylonian mathe�
matics are the many clay tablets that have survived,
preserving the math homework of young scribes from
 BEGINNINGS33

over 3,500 years ago. Like the Egyptians, the Babylonians

would sometimes phrase questions in terms of unknown
quantities, and remarkably, they knew effective methods
for finding positive solutions to quadratic equations. More
specifically, they would use their knowledge of the area
of squares and rectangles to answer questions like the
following: ‘A rectangular field has an area of 77m2 and
one side is 4m longer than the other. How long are the
two sides?’

Mesopotamian mathe�matics co-evolved with a desire to

study the night sky (a feature of the natural world that
had great religious significance for many ancient peoples),
and it also co-evolved with the practical tasks of taxation,
trade and measurement. Without a mathe�matical language,
tax records could not have been kept, nor would it have
been possible to record accurately the passage of the stars.
Mathematics enabled the bureaucratic procedures that are
required to rule a large state and, conversely, the novel
problems that confronted the first cities must have provided
a potent breeding ground for newly elaborated forms of
mathe�matics.
The range of ancient problems and exactitude of some
of the surviving calculations strongly suggests that symbolic
calculation was deeply valued, and cultivated as a broadly
applicable skill, transcending its role in any concrete, prac-
tical task. Pythagoras’ Theorem in particular has been
applied to a huge range of both practical and poetic prob-
lems, and it is clear that ancient peoples relished the fact
that the same numerical relationship holds whether we are
talking about the length of fields, spears, shadows or build-
ings. For example, an Ancient Egyptian manuscript asks
‘If a ladder of 10 cubits has its foot 6 cubits from the wall,
how high does the ladder reach?’, and this is clearly not
a question that needed to be answered for pressing practical
reasons.
34 MATHEMATICAL THOUGHT

It has often been argued that the oldest mathe�matical

procedures were simply something that you did: a kind
of ritual of computation that did not stand in need of
justification. For example, the Ancient Egyptians were
acutely concerned with the concept of Ma’at. This concept
was personified as a goddess, whose name can be translated
as truth, order or justice. The natural world, the state and
the individual were all parts of the domain of Ma’at, and
for several millennia, rulers would flatter themselves by
being described as ‘Lords of Ma’at’, or preservers of the
divine order.
For the Ancient Egyptians, following the time-honoured
ways of their ancestors was literally believed to be a matter
of life and death. The divine order and cosmic harmony
of Ma’at could turn to chaos and violence if the ruler or
his people did not adhere to their traditions and rituals.
To put it bluntly, Ancient Egyptians believed that they
had a sacred duty to do things as they had ‘always’ been
done, and a priest who was writing mathe�matics would
not have thought it a good idea to invent a new way of
doing things. After all, if a time-honoured way with
symbols remains eloquent, why argue with your teachers?
In particular, people would be shocked and angry if you
suggested changing the way that property markers were
replaced after the flooding of the Nile, as people’s confi-
dence in the process quite rationally depended on the fact
that it was done the same way every year.
In contrast, later generations of Greek historians were
justifiably proud of the fact that in Greece learned men
debated mathe�matical truths, and actively tried to develop
new forms of mathe�matics. Instead of merely having a
ritual of computation, they engaged in arguments, and
tried to deduce an expanded vision of the truth. Of course,
earlier cultures had also made mathe�matical deductions.
The difference is that when people in earlier cultures were
initiated into mathe�matical practices, they were taught
 BEGINNINGS35

through the presentation of examples. Teachers may well

have made quite abstract, elucidatory remarks as to why
particular techniques were effective, but the emphasis on
mathe�matical proof, and the rigorous articulation of logical
principles, are characteristically Greek innovations.

The Logic of Circles

As the Greeks debated, they articulated well-defined
conceptual schemes, and deduced a range of very general
truths. This process led to a significantly more abstract
form of mathe�matics, as Greek thinkers made a point of
stressing that they were interested in conceptual principles,
and not simply working with particular collections of
actual, countable things, or actual measured objects. Their
radical innovation was to construct arguments that defini-
tively settled the facts of various mathe�matical matters,
and their geometric deductions were made in the presence
of describable, labelled diagrams.
This final point is very important. As we shall see in
later chapters, modern mathe�matical arguments often hinge
on a kind of calculation that essentially rests on the defini�
tive properties of symbols. Ancient Greek mathe�matics
was somewhat different. It did involve abstract symbolism
(e.g. a diagram might include an arbitrary length ‘AB’ or
an angle ‘ABC’), but the matheÂ� matical objects they
discussed were all idealizations that were not too far
removed from our experience with actual objects. For
example, in classical geometry a curve is said to be ‘a path
traced by a moving point’. The movement involved is
understood to be metaphorical, as a circle (for example)
does not literally need to appear over time, drawn by a
point moving at some or other speed. We simply describe
what is meant by ‘a curve’ by talking in terms of motion,
because we all know what it is like to follow a path with
the mind’s eye. Indeed, this metaphor of motion is also
found in non-matheÂ�matical language, as when we say ‘the
36 MATHEMATICAL THOUGHT

road runs through the wood’, or ‘the road goes over the
hill’, we do not mean to imply that the road is literally
moving.
Similarly, in classical geometry the term ‘straight line’
refers to something abstract, but the meaning of the word
is nevertheless grounded in experience. In particular, it
seems fair to say that our grasp of the concept of straight
lines is partly informed by our experience with taut strings.
Of course, this doesn’t mean that our idea of a straight
line is merely some mental image of a taut string, as even
the finest string has some thickness, while a ‘straight line’
is an abstract, mathe�matical form that by definition has
no thickness at all. For many centuries European mathe�
matics was completely dominated by geometric concepts
of this sort, and it is worth pausing for a moment to
remember our initiation into the language of shape. As
children we are taught to recognize named shapes, and we
quickly take this game for granted. However, drawing a
loop in the sand and naming it ‘circle’ is a truly remark-
able act. Almost every aspect of the drawing is irrelevant,
because by seeing the drawing as a circle (that is, by seeing
the rightness of our description), we encounter the very
sense that the geometer is interested in.
To look for such forms is to engage in the process of
abstraction: trying to see the thing in its most elemental
of characteristics, and no more. We simply do not and
should not care whether the circle is perfectly drawn, or
whether it has been traced in sand or scratched on wood.
This is in marked contrast to something like a portrait,
where there is always more to talk about than the name
of the person painted. The spherical symmetry of sun and
moon suggests that mankind has been cognisant of circu-
larity for a very long time indeed, having the capacity to
draw the attention of others to the visual property of
roundness. Of course, it is comparatively easy to learn
what a circle looks like. More difficult is the task of stating
 BEGINNINGS37

the defining characteristic, namely that all the points on

the edge are equally distanced from the centre.
Identifying the essential character of circles enables
comprehension. For example, the fact that circles have this
defining centre explains the shape of ripples in a pond.
Before the stone hits, the water’s surface is flat, and every
point on the pond is very much like all the others. The
impact of the stone disrupts this equilibrium, and points
that are an equal distance from the disruption experience
the same effects (rising and falling at the same time). In
other words, because the stone marks a centre and distance
is the relevant feature (not the particular direction from
the centre), circular ripples are the evident consequence
of dropping a stone in a pond. Similarly, if we ignore the
effects produced by rotation, a liquid planet will form a
sphere, so that each point on the surface is equally close
to the centre of gravity.

The Factuality of Math

We have seen that the Ancient Babylonians could find the
length of a rectangular field, given the area of the field,
and the difference between its length and breadth. If these
people only had a narrow, practical interest in the length
of fields, they could have simply measured them. After
all, in what bizarre circumstance does a person know the
difference between a field’s length and breadth, without
actually knowing either the length or the breadth? As this
example indicates, it is clear that our ancestors’ passion
for mathe� matics ran deeper than narrowly practical
concerns.
People are naturally drawn into investigating the world
that our language evokes, and in many ways the mathe�
matics of the ancient world is still accessible to us today.
However, even though we can replicate the procedures of
ancient mathe�maticians, our attitude towards mathematic
facts has become very different. Up until the nineteenth
38 MATHEMATICAL THOUGHT

century matheÂ�matics was considered to be a realm of ‘hard

facts’. MatheÂ�matical axioms were taken to be absolutely
true, and any valid deductions based on those axioms were
accepted as being facts about the universe. In other words,
up until the nineteenth century mathe�matics could be accu-
rately described as the science of shape and number, and
nobody had thought to divide the subject into ‘pure’ and
‘applied’.
The facts of mathe�matics are different from the facts
of physics, but when we talk about straight lines, our
words are intelligible because (among other things) we
have encountered ‘lines’ in the real world. There is a long
tradition of somewhat mystical rhetoric about how mathe�
matical lines are ‘perfectly’ straight, unlike the pieces of
string we might use to measure a field. Nevertheless,
mathe�matical truths are of a piece with the only world
we can know, even if they are further abstracted from
experience than most of the truths we are interested in.
After all, every statement that humans are capable of
understanding must have some kind of grounding in our
cognitive capabilities and our experience of the world,
even if our interest is in the abstract concept rather than
its particular instantiations.
Many people are rather confused about the relationship
between physical reality and mathe�matical truth, as they
forget that physical reality is not the same thing as our
conception of it. For example, it is sometimes said that
‘matheÂ�matics is true in every possible world’, but that is
only true in the sense that in every conceivable world you
cannot force checkmate against a lone king with only a
king and a pair of knights. Similarly, it is sometimes said
that ‘the universe runs according to matheÂ�matical laws’,
but that does not mean we ought to accept the quasi-
religious belief that mathe�matics itself somehow forces the
world to behave as it does. For example, it is a mathe�
matical truth that the circumference of a circle is π times
 BEGINNINGS39

its diameter, but that does not mean that the number π is
literally present in every circular object.
The empirical fact is that if we measure a circular object
we need a piece of string a little over three times as long
as the string we need to measure its diameter, but there is
no empirical proof that the ratio between the two is precisely
π. In fact, quantum theory tells us that there is a funda-
mental limit to the accuracy of any measurement of length,
so physical lengths can never determine the infin�itely precise
quantities of mathe�matics. My point is that a circle is a
concept, not something physical, and there is no empirical
evidence to support the claim that the physical world is
literally forced to obey the laws of mathe�matics. Of course,
physics is deeply mathe� matical, as we use mathe� matical
concepts to describe the regularities that are empirically
observed. In other words, the fundamental fact is the regu-
larity of the universe, but mathe� matics only enters the
picture when we try to comprehend, describe and explain
those regularities.
We cannot attempt to understand the world without
using concepts, and although the gulf between mathe�matical
and non-mathe�matical concepts is sometimes overstated,
from the very beginning people recognized that mathe�
matical truths have a distinctive quality. In particular,
number concepts are very abstract indeed, and we have a
strong sense that when we are talking about numbers we
are not merely talking about the actual things we happen
to be counting. The equivalence between counting a list
of names and counting the named objects means that as
mathe�maticians we are free to disregard whatever has been
counted, so as children learn quite quickly, it really doesn’t
matter whether you are counting apples, pears or people.
The abstractness of number concepts is very significant,
but we sometimes forget that many other everyday words
are similarly abstract. For example, we talk about the world
in terms of colour, and have the related concepts ‘same’
40 MATHEMATICAL THOUGHT

colour and ‘different’ colour. Talking this way suggests

the idea of a universal colour chart, though of course we
should take account of the fact that our perception of
colour is highly sensitive to context, and entirely dependent
on light conditions.
Any particular ‘universal colour chart’ will be a rather
arbitrary affair, as we can pick any degree of accuracy to
qualify as ‘the same colour’, which means that we can
squabble interminably over how many different colours
should be on our chart. The situation is much more satis-
factory in the case of numbers, as the way we count in
itself generates the representatives that we require. The
obvious possibility of translation means that it doesn’t
really matter whether we count in English or in French,
using fingers or some beads. It is very important that we
can use mathe�matical vocabulary in the marketplace or
other specific contexts. However, the facts about our mathe�
matical words come from the mathe�matical framework to
which those words belong, and not from any wider sense
of meaning that such words might gain by being practi-
cally employed.
As we shall see, mathe� maticians in the nineteenth
century developed non-Euclidean geometries. This imag-
inative feat subtly altered our basic conception of math,
as only modern matheÂ�maticians would say, ‘Let’s just
assume that these axioms are true, and then talk about
the things we can deduce.’ The art of proper, symbol-
based reasoning has always been central to mathe�matics.
What changed in the nineteenth century is that instead
of starting with a hard fact, mathe�maticians now felt free
to begin with ‘mere’ statements, which did not need to
be ‘actually true’. To put it another way, over the centu-
ries the idea has developed that scientists study the real,
physical world, while mathe� maticians study abstract
objects that might or might not have anything to do with
the real world.
 BEGINNINGS41

This subtle shift in the philosophical roots of mathe�

matics led to the invention of ‘pure’ and ‘applied’ math.
It is often assumed that these are fundamental, natural
categories, but until the late nineteenth century, no mathe�
matician would have known what you were talking about
if you asked whether they studied pure or applied mathe�
matics. Of course, there have always been practical
problems that require mathe�matical expertise. Indeed, most
great mathe�maticians have made a significant contribution
to science, and to this day, many of the most compelling
mathe�matical challenges directly relate to the languages
that we use when describing physical phenomena. Even
though such concerns are not a part of mathe�matics itself,
the problems that we are interested in solving have always
exerted a powerful influence on the development of math.
The fundamental point is that theory and abstract
language are inescapable parts of our attempts to describe
the world. After all, we cannot present an account of the
world without a language that expresses our theory.
Furthermore, in the case of scientific theories, it is imper-
ative that our language is capable of supporting precise,
unambiguous deductions. We don’t just want to say ‘this
is what the world is like’; we want to say ‘because this is
what the world is like, it follows that …’ Over time, scien-
tists and other theorists have refined our definitions, and
as we clarify the logical import of our statements, the
disciplines of science have moved closer and closer to
mathe�matics. As the uses of mathe�matics have changed,
our sense of what the subject is about has also shifted, but
ever since the Ancient Greeks, we have appreciated that
mathe�matical concepts have a certain autonomy, as all
mathe�matics is logical and essentially systematic.
 Chapter 2:
FROM GREECE TO ROME
‘In arithmetic we are not concerned with objects which
we come to know as something alien from without
through the medium of the senses, but with objects
given directly to our reason and, as its nearest kin,
utterly transparent to it.’
Gottlob Frege, 1848–1925

Early Greek Mathematics

We don’t know much about Greek matheÂ�matical prac-
tice in the pre-archaic period (before 650 bc), but we
do know that the civilizations in that part of the world
were all numerate, they were familiar with the concept of
measured distances, and they employed trained scribes to
keep numerical records. The negotiations of trade and the
invention of money must also have encouraged numeracy
and stimulated mathe� matical development. Indeed, the
legendary father of Greek mathe�matics Thales of Miletus
was said to have been a merchant.
With the older figures of Ancient Greece it is nigh-on
impossible to separate fact from fiction, but it is thought that
Thales visited Egypt and Babylon in the sixth century  bc,
learning something of their mathe�matical skills and engaging
in learned debate. Some people’s methodologies were seriously
 FROM GREECE TO ROME43

flawed, and distant mathe�maticians sometimes disagreed with

one another. For example, the Egyptians were confident of
their rule for finding the volume of a truncated pyramid, but
the Babylonians had a conflicting one. It was clear to Thales
that only one of them could be correct (in this case it was
the Egyptians), and the kind of argument that he used to
settle the fact of the matter could be considered the earliest
form of modern, mathe�matical proof.
The crucial point is that the Ancient Greeks were inter-
ested in developing a theory of mathe� matics, and the
construction of meta-narratives was characteristic of Greek
scholarship. They didn’t just write about the behaviour of
good or bad rulers, they wrote about the theory of politics.
Likewise, they didn’t just write about techniques for healing,
they wrote about the theory of medicine. It is also important
to note that for the Ancient Greeks (and many later thinkers),
the capacity for rational thought was considered to be one
of the cardinal virtues of humanity. Mathematics provided
an ideal arena in which to refine our logical abilities, and by
the end of the fifth century bc the basic principles of matheÂ�
matical deduction were firmly established.
For example, around 440 bc the Ionian philosopher
Hippocrates of Chios wrote about the area of a crescent
shape, and in doing so constructed arguments that logically
led the reader from truth to truth. In other words, he
didn’t simply tell people how to calculate the given areas,
he showed that his answers were logical consequences of
an explicit list of axioms. Not all Greek mathe�matics was
of the form ‘by definition statement A is true, and if A is
true B is true, and if B is true C is true …’, but the Greeks
were certainly keen to identify the starting points of mathe�
matical arguments, and their axiomatic method made a
powerful impression on countless generations of thinkers.
Before this time, people had said things like ‘to calculate
this, follow such and such a procedure’. What was new
about Greek mathe�matics was the fact that the entire means
44 MATHEMATICAL THOUGHT

of deduction were rigorously stated, using diagrams and

writing to bring the obvious to light. Indeed, the mathema-
tician’s well-justified aversion to unstated assumptions is
one of the great legacies of the Ancient Greeks. They did
not simply present a pattern and then say true things about
it: they defined their terms, and then used their definitions
to support deductive practice.
A system of rigid definitions is a surprisingly powerful
thing, and the Greeks could employ their logos or language
to remarkable effect. For example, consider the proof of
Pythagoras’ Theorem presented in the introduction. By
explicitly stating all the relevant facts, we can deduce
Pythagoras’ Theorem from the statements ‘squares have
four equal sides and four identical corners’, ‘moving a shape
does not change its area’, and ‘a + b = b + a’. Other ancient
cultures did not develop the same interest in elucidating
the way that one statement follows from another, but in
Greek mathematics the logical relationships between true
statements are the main focus of attentions. Consequently,
the simple and obvious statements that earlier people may
have taken for granted acquired a new status. Henceforth
such ‘axioms’ were not only considered to be true, they
were seen as the foundations of an entire discipline.

Pythagorean Science
As the founder of a mystical tradition, Pythagoras is
shrouded by legend. We know that he was born on the
island of Samos around 580 bc, and that he was at least
eighty years old when he died. It is said that he met with
Thales, who encouraged him to visit Egypt and Babylon,
to learn the things they had to teach there. Many of the
Greek schools emphasized the reality of change, but
Pythagoras and his followers looked to identify the eternal
features of the world, particularly those that related to
number. In search of such principles, Pythagorean scholars
 FROM GREECE TO ROME45

contemplated geometry, arithmetic, astronomy and music,

guided by their belief that the universe reveals itself in
matheÂ�matical form, and that ‘All is Number’.
The notion of putting the things of the world into some
kind of ordered scheme is very ancient, and very powerful.
Such ordered schemes or catalogues were central to early
science, including the Pythagoreans. As the philosopher
Jacob Klein explained in Greek Mathe�matical Thought
and the Origin of Algebra, ‘The general point of view
governing the efforts of the Pythagoreans might be sketched
out as follows: they saw the true grounds of the things in
this world in their countableness, inasmuch as the condi-
tion of being a “world” is primarily determined by the
presence of an “ordered arrangement” or taxis. [Conversely,
any] order rests on the fact that the things ordered are
delimited with respect to one another, and so become
countable.’
The Pythagoreans’ study of number was not entirely
mathe�matical (at least as we understand that term), as they
also proclaimed to find mystical meanings in numerical
relations. Many religious and divinatory traditions associate
particular qualities with particular numbers or patterns
(e.g. astrology, tarot cards, or the I Ching), and these asso-
ciations may feel far from arbitrary. For example, consider
the following pairs of words:

Light/Dark; Warm/Cold; Right/Wrong;

Present/Absent; Active/Passive; On/Off;
True/False; Heavy/Light; Wet/Dry.

Now consider the following trinities:

Beginning/Middle/End; Viewer/Viewing/Viewed;
Past/Present/Future; Mind/Body/Soul;
Father/Son/Holy Ghost; Brahman/Shiva/Vishnu.
46 MATHEMATICAL THOUGHT

The mathe�matical fact of this matter is entirely mundane:

counting the words in the first list proceeds ‘one, two’,
while for the second list we have ‘one, two, three’. But
there is also a potent metaphorical connection within each
group: the paired words have an oppositional character,
with one part being defined in opposition to the other. In
contrast, the trinities have a sense of interpenetration or
interdependence. That is to say, something of the entire
trinity is present in each part, and to separate the parts of
each trinity is to divide an authentic whole.
As the integers were sacred to the Pythagoreans, they
were motivated to grasp as much as they could of the
character of each number, combining a mystical or meta-
phorical approach, together with what is still considered
rational. Although it falls within a larger, extremely ancient
tradition, Pythagorean number science was quite distinct
from the knowledge that preceded it. A principle technique
was to separate the integers into specific and intricately
related categories, according to characteristics that can be
demonstrated geometrically. For example, the Pythagoreans
would speak about:

Odd numbers1

Even numbers

1 Modern mathe�maticians consider one to be an odd number, a square number, and

so on, but the Ancient Greeks would not have included it in such lists. For them,
numbers only began when there was a multitude of countable things.
 FROM GREECE TO ROME47

Composite numbers
An integer is said to be composite if it is the multiple of
two smaller integers (excluding one). Any composite
number of dots can be arranged into a rectangle.

Prime numbers
By definition prime numbers are not composite, so they
cannot be formed into any rectangle (other than a line).

Triangular numbers

Square numbers

The Pythagoreans’ belief in the supreme power of number

found a beautiful confirmation in their study of stringed
instruments. Halving the length of a plucked string raises
the note by one octave, and people can clearly hear the
harmony between two notes an octave apart. More gener-
ally, the pentatonic scale (black notes on a piano) can be
produced by using the following sequence of lengths:

1 (doh), 8/9 (ray), 4/5 (me), 2/3 (soh), 3/5 (la), 1/2 (doh)

Similarly, the diatonic scale (white notes on a piano) runs

as follows:
48 MATHEMATICAL THOUGHT

1 (doh), 8/9 (ray), 4/5 (me), 3/4 (fah), 2/3 (soh), 3/5
(la), 8/15 (te), 1/2 (doh)

These tunings are used because they generate a preponder-

ance of the ratios 1:2 (an octave), 2:3 (a musical ‘fifth’),
3:4 (a ‘fourth’) and 4:5 (a ‘third’). In general, the Law of
Small Numbers tells us that highly commensurate lengths
of vibrating strings produce pleasing, resonant harmonies,
while less commensurate lengths produce dissonance.

Plato and Symmetric Form

The words ‘academy’ and ‘academic’ are derived from
‘Hekademeia’: the place in Athens where Plato had his
school. The gates to the Hekademeia were emblazoned
with the words ‘Let no one ignorant of geometry enter
here,’ and Plato’s account of the world placed a special
emphasis on the ideal forms of mathe�matics. He believed
that abstract objects have a timeless, independent existence
that stands apart from any human activity, and his rhetoric
and philosophy encouraged the widespread belief that
mathe�matical forms are perfect and eternal. At the end of
this book I argue that this view of mathe�matics is funda-
mentally mistaken, but many people think that Plato was
right, and describe themselves as Platonists or Realists.
The claim that mathe�matical forms are perfect and eternal
has proved to be a potent and appealing notion. Indeed,
there is a direct, historical link from Plato’s rhetoric to the
many spherical domes in churches, synagogues and
mosques. More generally, it is striking that across the globe,
from Mexico to Tibet, depictions of the cosmos have almost
invariably been constructed along geometric lines. It would
seem that when we conceive of a scheme, and relay that
across generations, it tends to acquire an orderly, geometric
form. For example, in many disparate cultures we find
rings of hell; heavenly spheres; the four corners of the
earth, and so on.
 FROM GREECE TO ROME49

A fascination with symmetric, regular forms can be found

in many cultures. What made the Ancient Greeks so
remarkable was the systematic way that they investigated
the properties of mathe� matical shapes, cataloguing and
rigorously deducing their various properties. Long before
Plato, people spoke of triangles, squares and other ‘regular
polygons’, but as we shall see, our understanding of such
forms made great advances in the time of Plato. By defini-
tion, a polygon is any planar figure with an integer number
of straight sides. In a regular polygon, every side is the
same length and every corner is identical. Now, a particu-
larly ancient challenge is to find all of the different ways
that we can cover a flat surface using regular polygons as
tiles. In particular, imagine arranging a number of polygons
so that they touch corners at one point.
If the polygons are going to completely cover a flat
surface, the corners that meet at the given point must
contain angles that add up to 360°. It follows that if we
are to cover a flat surface with one type of regular polygon,
the number of angles in one corner of that polygon must
precisely divide 360°. Only three types of regular polygon
satisfy this condition:

1. Triangular tiling. Six triangles can meet at a point

because there is an angle of 60° between adjacent
sides of an equilateral triangle, and 360° = 6 × 60°.
2. Square tiling. Four squares can meet at a point
because there is an angle of 90° between adjacent
sides of a square, and 360° = 4 × 90°.
3. Hexagonal tiling. Three hexagons can meet at a point
because there is an angle of 120° between adjacent
sides of a regular hexagon, and 360° = 3 × 120°.

The angle between adjacent sides of a regular pentagon

is 108°, so where three pentagons meet we have a total
of 324°, and a gap of 36°. Only two polygons with more
50 MATHEMATICAL THOUGHT

than six sides can meet at a point, so triangles, squares

and hexagons are the only regular polygons that can be
used to tile the plane. If instead of using a palette of one
kind of regular polygon we allow any combination of
regular polygons while maintaining identical vertices (i.e.
insisting that each corner is touched by an identical
sequence of polygons), then there are exactly eight tiling
possibilities.
 FROM GREECE TO ROME51

Plato studied mathematics with his friend and fellow Athenian

Theaetetus, and it was Theaetetus who first identified all the
different ways of enclosing a finite volume of space within
a boundary composed from a single type of regular polygon.
A cube is the most familiar example of a finite volume with
a boundary composed of a single type of regular polygon,
but – as we shall see – there are several others. Because they
play a central role in Plato’s metaphysics, shapes of this kind
have somewhat unfairly acquired the name ‘Platonic solids’.
In order to find a complete list of these deeply regular shapes,
Theaetetus needed to consider all the different types of corner
that can possibly be constructed out of a single kind of
regular polygon. That is to say, he needed to calculate the
number and type of faces that could possibly meet at a point.
It should be clear that at each corner at least three poly-
gons must meet. Two polygons or fewer cannot make a
solid, volume containing form. Furthermore, when the
corners of some polygons meet at a point, the total number
of angles in those corners must be less than 360°. If the
angles added up to 360°, the polygons in question would
lie flat on the page, and make a tiling instead of a volu-
metric form. By falling short of 360° an arrangement of
polygons can close in on a space, and may fit together to
make one of the Platonic solids.

Given these constraints, there are only five possibilities

for any corner of a Platonic solid. Three triangles can meet
at each corner, and we can see that this is a genuine possi-
bility because 3â•›×â•›
60°â•›
=â•›
180°, which is less than 360°.
Similarly, four triangles can meet at each corner, and we
52 MATHEMATICAL THOUGHT

can see that this is a possibility because 4â•›×â•›60°â•›=â•›240°, which

is less than 360°. Likewise, we can have five triangles
meeting at each corner, but we cannot have six. If six
equilateral triangles touch corners they lie flat on the page,
which corresponds to the fact that 6â•›×â•›60°â•›=â•›360°. To recap,
if we use triangles to make the faces of a Platonic solid,
we can have three triangles meeting at each corner, four
triangles meeting at each corner, or five triangles meeting
at each corner, but those are the only possibilities. Using
square faces, the only possibility is that three of them meet
at each corner (giving 270°), as four squares that touch at
one corner lie flat on the page. We can also have three
pentagons meeting at each corner, as that gives us a total
of 324°, but that is the only remaining possibility. Hexagons
and other polygons with more than five sides cannot
produce a Platonic solid, as if you put three of these shapes
together you get an angle that is at least as big as 360°.
 FROM GREECE TO ROME53

This argument does not complete the proof that there are
precisely five Platonic solids. We have yet to prove that
there are in fact shapes with vertices of the kinds described.
Furthermore, we have yet to prove that there can only be
one shape whose vertices satisfy the previous descriptions.
For example, how do we know that there cannot be a
shape that has three squares meeting at every corner, but
which is different from a cube? Similarly, how do we know
that there is a one-and-only shape that has three pentagons
meeting at each of its corners? I will return to this point
later in the book, but for the moment I think it will suffice
to say that, greatly to their credit, the Greeks did not leave
this final step as an assumption.

Euclidean Geometry
On 20 January 331 bc, Alexander the Great was sailing
along the Egyptian coast, opposite the island of Pharos.
Recognizing the many natural advantages of this site, he
ordered that a city should be built in his name. Within
decades, Alexandria was home to Macedonians, Greeks,
Arabs, Babylonians, Assyrians, Italians, Carthaginians,
Persians, Egyptians, Gauls, Iberians and Jews. To help estab-
lish the new metropolis as a great cultural centre, Alexander’s
successor and half-brother ordered that an enormous library
be built. A great deal of money was spent, and scholars
were employed to gather an encyclo�paedic range of texts
into an open, secular institution, modelled on the Athenian
schools. It was in this newly founded city that Euclid
(c.â•›325–265 bc) started his school of matheÂ�matics. Euclid
himself trained at Plato’s Academy, but under the rule of
Alexander’s general, Ptolemy I, Alexandria rose to become
the scientific capitol of the world. Indeed, within Euclid’s
lifetime it became a greater centre of mathe�matical excellence
than anywhere in Greece and, what is more, it maintained
its pre-eminent position for five hundred years or so.
Euclid’s Elements is the most influential textbook ever
54 MATHEMATICAL THOUGHT

written, and only the Bible has been printed in a greater

number of editions. Although his book (or sequence of books)
has been a major influence for more than twenty-three centu-
ries, Euclid did not invent or discover the kind of geometry
that bears his name. Nor did he invent the axiomatic method,
whereby basic assumptions are listed and then deductions
made from a definitive list of axioms through a logically
structured sequence of theorems. The messy truth is that
there was a long and complex evolution behind the emergence
of Euclidean geometry, but Euclid’s particular summary of
the essential truths of Greek geometry is a tour de force,
largely because of its exceptionally clear logical structure.
His books start with five famous axioms, and by
employing these explicit statements we can logically clarify
our language for describing shapes. Instead of relying on
our intuitive interpretation of words like ‘point’, ‘straight
line’, ‘circle’ or ‘right angle’, Euclid stated five defining
principles for these things – the famous axioms of Euclidean
geometry. In order to prove all of Euclid’s theorems, we
only need to refer to the following definitive properties
of straight lines, points, right angles and so on.

1. There is precisely one shortest path between any

two points. We refer to any such finite path as a
straight line segment.
2. Any straight line segment can be extended indefi-
nitely, forming a straight line.
3. Every straight line segment can be used to define
a circle. One end of the segment is the centre of
the circle, and its length forms the radius.
4. All right angles are essentially identical, in that any
right angle can be rotated and moved to co�incide
with any other.
5. Given any straight line and a point that is not on
that line, there is precisely one straight line that passes
through the point, and does not intersect the line.
 FROM GREECE TO ROME55

The fifth axiom effectively specifies what Euclid meant by

the term ‘parallel’, and it also determines his interpretation
of the term ‘direction’. For millennia it was understood
that two straight lines point in the same direction if and
only if they do not cross (that is, if and only if they are
parallel). In a later chapter I shall return to the subject
of  Euclid’s fifth axiom, and elucidate the rethinking of
geometry that took place in the nineteenth century. In
particular, I shall discuss non-Euclidean geometry, and the
significance of the fact that there are geometries that are
fundamentally different to those described by Euclid.
First I want to point out that Greek geometry in general,
and Euclid’s Elements in particular, have exerted a profound
influence: an influence that has reached well beyond the
lives of mathe� maticians. Art, architecture, philosophy,
theology, science and countless other endeavours have all
been shaped by Euclid’s example. It is therefore under-
standable that many philosophers, theologians and other
constructors of arguments have echoed something of his
form, thinking it only proper to start an argument with
explicitly stated premises, and then use axioms to derive
further implications.
As well as being a paradigm-defining example of a thor-
oughly sound argument, the practice of geometry also had
an incalculable influence on the visual arts, particularly
architecture. It is impossible to imagine how our world
would have looked if it had been designed by people who
did not use rulers and compass, as apart from a few notable
exceptions, almost every blueprint, from medieval churches
to modernist blocks, can be drawn using these two instru-
ments. Furthermore, in classical geometry we often say
things like, ‘if we extend these lines we can see that they
meet at a point’. Architects and painters have often favoured
subtle kinds of order, where the generating pattern can be
seen more clearly by extending the lines involved.
Straight lines and right angles are so prevalent in architecture
56 MATHEMATICAL THOUGHT

that people who try using other shapes are often said to
be ‘experimenting with non-Euclidean geometry’. This is
a perfectly understandable abuse of the term, but, strictly
speaking, geometry is more fundamental and much harder to
rethink than our palette of shapes or forms. Our geometry
is present in the language that we use to describe our shapes,
but the shapes themselves are not enough to determine
which geometry we may be using. There is no problem
with using the same geometry to study squares, circles or
weird, irregular blobs. The geometry itself is characterized
by such fundamental things as the kinds of symmetry we
can possibly exhibit in the given space, or by the definitive
relationships between straight lines, angles, distances and so
on. In other words, a geometry is not defined by the shapes
that we typically or most easily talk about. We can describe
the same shapes using the language of Euclidean geometry
or other essentially different, non-Euclidean geometries: the
difference is not in the shapes, but in the meanings of the
terms we use to describe them.

The Euclidean Algorithm

The concept of proportion is fundamental to art, architec-
ture and mathe�matics. There is an ancient technique for
characterizing or determining the ratio between given
lengths, and although it predates Euclid this method is now
known as the Euclidean algorithm. The first point to appre-
ciate is that proportion is a more abstract concept than a
ratio of lengths, a ratio of areas, and so on. For example,
if we have a piece of string that is 20 cm long and another
that is 40 cm long, we have a ratio of 20 cm:40 cm, but we
also find it obvious that one is twice as long as the other.
If we consider the distances 17 miles and 34 miles, then
although the actual lengths are completely different, we
recognize that 20  cm:40  cm is in the same proportion as
17  miles:34 miles. In both cases we have the proportion
1:2, and we understand that the expression 1:2 stands as a
 FROM GREECE TO ROME57

�
representative for every equally proportioned ratio. Also
note that in writing a proportion we do not need units, so
the proportion 1:2 could equally well refer to a ratio of
lengths, areas, or any other quantified property.
In his famous book, Euclid described the well-known
method for determining the proportion between two given
lengths. The first step is to construct a rectangle whose
sides possess the lengths we are interested in. Having done
that, we can determine the ratio between the lengths of
those two sides by employing the following technique:

1. Draw the largest square that fits inside your

rectangle.
2. Squeeze in as many of these squares as you possibly
can. If you can fill the entire rectangle with your
squares, you are finished. Otherwise, there will be
a rectangular space that has not been covered by
squares. We now take the remaining rectangle and
apply step 1.

The Euclidean algorithm terminates when we find a square

that fits perfectly, filling in the entire rectangle. Squares of
this size can be used to make a grid that perfectly subdivides
the generating shape, and this will be the coarsest square grid
that can possibly be used to divide the rectangle into squares.
In other words, the width of the smallest square is equal to
the greatest common divisor of the lengths of the two sides.
58 MATHEMATICAL THOUGHT

For some rectangles, there simply isn’t any square grid

that will fit perfectly. In this case our algorithm can be
continued indefinitely. If the lengths of the sides are inte-
gers with no factors in common (e.g. 5  cm x 9  cm), the
final step of our algorithm uses 1 cm squares. If the numbers
have a common factor (e.g. 3  cm x 9  cm), we produce a
scaled up version of some co-prime pattern (e.g. we finish
with 3 cm squares):

(equivalent to 3:4) (equivalent to 3:5)

This technique makes the proportions of the generating

rectangle evident and intelligible, and it has inspired count-
less generations of mathe�maticians, artists and architects.

Archimedes
Archimedes of Syracuse (c.â•›287–212 bc) is arguably the
greatest mathematician to ever live. He has also been
described as the first mathe�matical physicist, as he devel-
oped theories that accurately predict when a thing will
balance, and when a thing will float. As well as his
astounding theoretical accomplishments, Archimedes was
a brilliant engineer: people have been using the Archimedean
screw to move water uphill for over twenty-two centuries,
and it is still used to this day. However, the Ancient Greek
historian Plutarch (c.â•›a d 45–125) tells us that Archimedes’
proudest achievement was carved onto his tombstone,
namely his stunningly beautiful theorem concerning the
ratio of volume between equally high cylinders, hemi-
spheres and cones:
 FROM GREECE TO ROME59

Archimedes became famous in his own lifetime, and we

know that at least one contemporary account was written
about his life and work (now sadly lost). Plutarch, Livy,
Cicero and Vitruvius all mentioned his life’s work, and he
was helped in acquiring legendary status by defending the
Sicilian city of Syracuse. Plutarch writes of the fear and
respect that Archimedes struck in the Roman invader’s hearts,
as he fought them from the city for over two years: ‘If they
only saw a rope or a piece of wood extending beyond the
city walls, they took flight, exclaiming that Archimedes had
once again invented a new machine for their destruction.’
His conceptual grasp of the physical literally changed
the world, and shaped the way that technology developed.
For example, when he wrote On the Equilibrium of Planes,
Archimedes gave a completely axiomatic treatment of
mechanics, enabling logical deductions for physical systems.
By such means he devised a powerfully general explanation
of the functioning of rollers, wedges, levers and pulleys,
using terms whose relevance clearly extends beyond the
particular contraptions that are physically present. His
strictly mathe�matical approach enabled the calculation of
centres of gravity, and he was in a position to make such
general statements as ‘The tipping point is where the centre
of gravity is directly above the edge of the base.’ Similarly,
he was the first person to calculate when two objects will
balance on a see-saw.
As if that wasn’t enough to ensure his immortal fame,
Archimedes was the first person to prove that a sphere of
radius r must have a volume of 4/3πr3, and a surface area
four times greater than a circle of equal radius (i.e. 4πr2).
He proved many fundamental results, but a parti�cu�larly
attractive example of his work can be found in the way that
he considered the area of a circle. By definition of the term
‘π’, the circumference of a circle of radius r is equal to 2πr.
Archimedes knew that if we slice a circle into equally
sized segments and rearrange those component shapes, the
60 MATHEMATICAL THOUGHT

total area will not change. He also knew that given any
circle, we can draw a hexagon that is just small enough to
fit inside the circle, and we can draw another hexagon that
is just large enough to contain the entire circle. The area of
our circle must be bigger than the area of the smaller hexagon,
and smaller than the area of the larger hexagon. Furthermore,
it is easy to calculate the area of these hexagons because
they are composed of triangles of known size. Archimedes
understood that in principle we can do the same for a polygon
of any number of sides, and as we consider polygons with
larger and larger numbers of sides, we find shapes whose
areas are closer and closer to the area of a circle. This method
can be used to calculate estimates of π. Indeed, Archimedes
managed to calculate π to two decimal places by using a
96-sided polygon. More importantly, Archimedes’ ingenious
arguments showed that the area of circle of radius r must
be πr2. Like all matheÂ�matical theorems, this fundamental
truth has many different proofs, including the following
beauty devised by Leonardo da Vinci:

If we slice a circle of radius r into an increasing numbers

of segments, those segments can be arranged into a shape
 FROM GREECE TO ROME61

that is increasingly similar to a rectangle of width πr and

height r.
In many of his proofs, Archimedes argued that a given
area or volume cannot be bigger than some given number,
and likewise it cannot be smaller than the given number.
For example, Archimedes proved that the area of a circle
cannot be bigger than πr2, and it cannot be smaller than
πr2. It follows that the area of a circle must be equal to πr2:
a form of argument known as ‘proof by exhaustion’. The
mathematician and astronomer Eudoxus (c.â•›408–355 bc) is
credited with being the first person to prove that one quan-
tity is equal to another by demonstrating that it cannot be
bigger or smaller, and over the following millennia, proof
by exhaustion showed itself to be a neat but rather unpro-
ductive idea. However, as we shall see in Chapter 5, the
thought of covering a shape with increasingly fine pieces
is an ancient ancestor of the most fruitful mathe�matical
idea of the modern era: the infinitesimal calculus.

Alexandria in the Age of Rome

In the centuries that followed the founding of Alexandria,
Rome grew to dominate all of its surroundings. As Rome’s
sphere of influence expanded, countless manuscripts were
‘donated’ or pillaged to furnish the libraries of eminent
Romans. For example, we know that Cleopatra gave Caesar
a gift of tens or hundreds of thousands of manuscripts
from the Royal Library of Alexandria. Fortunately, many
Romans revered the mathe�maticians of Greece, appreciating
both the practical and theoretical significance of mathe�
matical knowledge. There was a popular trade in busts
and portraits of the famous mathe�maticians, and Greeks
were often employed as tutors. The Roman Empire spread
the practice of mathe�matical education throughout Europe
(well beyond the ethnically Roman), but surprisingly the
cosmopolitan city of Alexandria remained dominant as the
centre for purely mathe�matical study.
62 MATHEMATICAL THOUGHT

Despite the unfortunate fate of the tragically flammable

Royal Library, the city of Alexandria remained a magnet
for mathe�maticians for many, many centuries. We have
good reason to believe that Archimedes studied there, and
it was in Alexandria that the Pythagorean scholar
Nicomachus of Geresa (c.â•›a d 60–120) wrote his Introduction
to Arithmetic, employing arithmetic notation and ordinary
language to explain topics that Euclid had described from
an essentially geometric perspective. This book was highly
influential, and remained the standard European textbook
on arithmetic for more than a thousand years. Some fifty
years after Introduction to Arithmetic was published,
Claudius Ptolemy (c.â•›a d 85–165) produced a work that
was even more influential: The Almagest, or the ‘Great
Collection’.
As a Roman citizen of Egypt working in Alexandria,
Ptolemy employed the detailed astronomical knowledge
of the Babylonians to make improvements on Eudoxus’s
model of planetary motion. The resulting masterpiece was
considered to be a definitive guide to the movement of
the heavens, and over the following millennia, countless
people used his system to predict successfully the apparent
location of the planets in the night sky. The Almagest also
spread the system of longitude and latitude, and it contained
many sophisticated arguments about the relationship
between measurements of angles and measurements of
length (what we call trigonometry).
Following the collapse of the Roman Empire, priests
and laymen continued to teach and study the mathe�matics
of Greco-Roman civilization. The diplomat and philoso-
pher Anicius Manlius Severinus Boethius (c.â•›a d 480–525)
deserves particular mention in this regard. His Institutio
Arithmetica was not a great piece of original mathe�matics,
but his reverence and love for the subject helped to ensure
that the Catholic Church would teach and preserve the
Pythagorean knowledge of number, following Boethius’
 FROM GREECE TO ROME63

educational quadrivium of arithmetic, geometry, astronomy

and music.
There is a certain poetry in the fact that while so many
practical points of reference were lost (e.g. the means of
making concrete), the most abstract of notions remained
unbroken. Indeed, a long line of Christian theologians,
including Saint Augustine, have supported the view that
God’s creation is matheÂ�matical. It is therefore understand-
able that despite the terrible upheavals following Rome’s
collapse, crucial fragments of mathe�matical science were
faithfully preserved. By the time of the Middle Ages, scho-
lastic writers were making tentative advances from the
body of established mathe�matical knowledge. However,
until Italian mathe� maticians started to make exciting
advances in the late fifteenth century, ancient works such
as Euclid’s Elements were still considered to be the pinnacle
of mathe�matical knowledge.
In later chapters we will see how Europe experienced a
renaissance in mathe�matics, building on the achievements
of Greek and Arab scholars. First, I want to focus on an
aspect of reality that has occupied mathe�maticians, artists
and architects for millennia. The notions of measure-
ment and proportion are exceptionally ancient, and in the
next chapter I will explain how counting and continuous
mea�surement have been used to make proportion evident.
In particular, we will see how the insights of Eudoxus and
Dedekind extended the concept of number, from integers
to fractions and on to irrational numbers.
 Chapter 3:
RATIO AND PROPORTION
‘The concept of number is the obvious distinction
between the beast and man. Thanks to number, the
cry becomes a song, noise acquires rhythm, the spring
is transformed into a dance, force becomes dynamic,
and outlines figures.’
Joseph de Maistre, 1753–1821

Measurement and Counting

In daily life, we don’t just count objects, we also measure
quantities such as length, area, weight and time. First we
select a unit of measurement: feet, acres, grams or hours
as the case may be. We assign the relevant unit quantity
the measure of one, and then count the number of unit
measures that make up the quantity to be measured. For
example, we might measure the length of a field by counting
the number of feet between one end and the other. In
general, the process of counting out units may not ‘come
out even’. For example, our field might be longer than
sixty feet, but shorter than sixty-one feet. In that case we
may measure the remaining distance by using some subunit
measure, obtained by subdividing our original unit into n
equal parts.
The Ancient Egyptians and numerous other cultures
 RATIO AND PROPORTION65

studied this kind of process. In ordinary language the

standard units and subunits are given their own names, as
feet are subdivided into inches, hours are subdivided into
minutes, and so on. In general we divide our unit measure
into n subunits, and when we measure something we need
to count out m of these subunits to make up the quantity
to be measured. In that case we say that we have measured
a fraction m/n. Note that the ‘denominator’ n tells us what
kind of subunit we are using, while the ‘numerator’ m tells
us how many of the subunits we have counted.
Over the course of several centuries, the ancient equiv-
alents to the modern symbols m/n gradually lost their
association with the process and units of measurement. In
other words, people began to consider fractions as ‘pure’
numbers, much like the integers. How do we justify
extending the word ‘number’ from the counting numbers
onto the fractions? Well, we might say that just as it doesn’t
matter whether you are counting apples, pears or people,
it doesn’t matter whether we are measuring fractional quan-
tities of distance, weight, time, etc. It is also absolutely
fundamental that we can add and multiply the fractions
alongside the counting numbers. Indeed, the rules for
adding and multiplying fractions can be summarized as
follows:
a c ad + bc a c ac
=
+ = ,
b d bd b d bd

a a c
= 1 and = if and only if ad = bc.
a b d
From a modern perspective we accept the fractions as a
legitimate number system because the rules for adding and
multiplying fractions satisfy the same, axiomatic laws as the
addition and multiplication of integers. More specific�ally,
the following statements are true whether p, q and r are
integers or fractions:
66 MATHEMATICAL THOUGHT

p + q = q + p, p + (q + r) = (p + q) + r, pq = qp,

p(qr) = (pq)r and p(q + r) = pq + pr.

For many centuries it was believed that the only conceiv-

able quantities were fractional or ‘rational’. However, as
I shall show in the next section, this plausible claim is not
in fact true. As an initial observation, consider the following
diagrams:

In each case we have combined a pair of identical shapes

to produce a similar shape with double the area. The sides
of these larger shapes are √2 times greater than the sides
of the original, component shapes. We can prove this fact
using Pythagoras’ Theorem. More generally, we can use
Euclid’s axioms to prove that rescaling a two-dimensional
shape by making the lengths n times longer changes the
shape’s area by a factor of n2. Because the second, larger
shapes have double the area (n2â•›=â•›2), the lengths of the
sides must have increased by a factor of √2. Similarly, we
can combine three hexagons to produce one with triple
the area, and sides √3 times that of the originals:
 RATIO AND PROPORTION67

Because the ratios 1:√2 and 1:√3 are geometrically construct-

ible, we intuitively feel that √2 and √3 must be numbers
of some kind. In other words, we readily accept that each
of these things has a single, definitive size. This notion is
extremely ancient, and hundreds of generations from across
the globe have pondered the question, ‘How long is the
diagonal of a square, given the length of the sides?’ Indeed,
a Babylonian clay tablet from c.â•›1800–1600 bc tells us that
this ratio of lengths can be expressed as:
24 51 10
1 to 1+ + +
60 60 2 60 3
(a solution that is exact enough for any practical purpose,
as the error in this approximation is only 0.00004%).
As we saw in the previous chapter, Pythagorean science
used the language of the integers, and in that language we
might say that a quantity simply is a ratio of two integers.
However, during the fifth century bc, the Ancient Greeks
proved that √2 cannot be expressed as a ratio of integers,
and nor can √3 or √p where p is any prime.

Reductio Ad Absurdum
In order to prove that √2 cannot equal any fraction, we
must first prove the Fundamental Theorem of Arithmetic.
Both of these proofs can be established by the mathemati-
cian’s gambit: reductio ad absurdum. As the mathematician
G.  H.  Hardy remarked, ‘It is a far finer gambit than any
chess gambit: a chess player may offer the sacrifice of a
pawn or even a piece, but a mathematician offers the game.’
I shall return to this idea, but for now let us just accept
that we do not say contradictory things about mathe�matical
objects. In short, we cannot work with a logically incon-
sistent system, so if a putative set of axioms is inconsistent
but we want to continue doing mathe�matics, we have to
change one of our assumptions. Somewhat surprisingly, this
negative principle (the ban on inconsistency) enables us to
68 MATHEMATICAL THOUGHT

prove positive new conclusions, including facts concerning

the relationship between fractions and measured lengths.
Euclid’s proof of the Fundamental Theorem of Arithmetic
begins with an axiomatic observation: ‘Every collection of
positive integers contains a smallest number.’ In particular,
suppose that there are integers that are neither prime nor
the multiple of primes (integers with property P, say). It
follows from our axiom that if there are any integers with
property P, there must be a smallest integer (called k) that
has the property P. In other words, if the Fundamental
Theorem of Arithmetic were false, there would have to be
an integer k, which is the smallest number that is neither
prime nor the multiple of primes. If k is not the multiple
of any two integers (apart from k x 1 of course), then by
definition k is prime, which would mean that it does not
have property P after all.
The other possibility is that there are two integers s and
t, such that s x t = k. In that case, k must be larger than
either s or t. This implies that s and t cannot have property
P, which means they must be prime or the multiple of
primes. This follows because by definition k is the smallest
number with property P. But the multiple of any two
multiples of primes is itself a multiple of primes, which
means that k cannot have property P. Since there cannot
be an integer k with property P, it must be true to say
that every integer can be factorized.
Reductio ad absurdum is a simple trick, but it reveals a
wealth of truth: every integer must either be prime or it
must be equal to the product of some primes. In other
words, given any positive integer N (other than N = 1),
there is a collection of prime numbers p1, p2, …, pn such
that Nâ•›=â•›p1,â•›×â•›p2â•›×â•›…â•›×, pn . Euclid also proved that for every
integer N, there can only be one collection of primes that
multiply together to give the number N. For example,
2â•›×â•›3  =  6 and 3â•›×â•›2 = 6, but every other multiplication of
primes equals something other than 6.
 RATIO AND PROPORTION 69

Using a similar method, we can prove that √2 is not

equal to any ratio of integers. Suppose for the sake of
argument that we have found a fraction N/M, which satis-
fies the equation N/M = √2. We can rewrite this equation
by multiplying both sides by M and then squaring both
sides, giving us the equation N2 = 2M2 . By the Fundamental
Theorem of Arithmetic, we must be able to rewrite the
integers N and M as a sequence of prime numbers. We
therefore have something of the form:

( p1 × p2 × … × pn)2 = 2(q1 ×q2 × … × qm)2, or equivalently

p12 × p22 × … × pn2 = 2 × q12 × q22 × … × qm2 .

If it is indeed the case that the left-hand side equals the

right, then the prime factors of the left-hand side must be
identical to the prime factors of the right. But it cannot
be true that both sides have the same prime factors, because
the factors of the left-hand side are two lots of p1, two
lots of p2, and so on up to two lots of pn, while on the
right we have two lots of q1, two lots of q2, and so on up
to two lots of qm, plus a single factor of 2. This solitary
factor of 2 cannot have a counterpart on the left-hand side,
which means that the two sides of the equation cannot be
equal to one another. We must therefore reject our assump-
tion that for some pair of integers N and M we have
N2 = 2M2. In other words, we must conclude that there
cannot be a fraction N/M = √2.
The proof that √R cannot equal any fraction when R is
any prime is very similar. First we suppose that we have
found a fraction N/M, which satisfies the equation N/M
= √R, which is equivalent to supposing N2 = RM2. The next
step is to rewrite N and M as a sequence of prime numbers,
giving us something of the form:

( p1 × p2 × … × pn)2 = R(q1 ×q2 × … × qm)2, or equivalently

70 MATHEMATICAL THOUGHT

p12â•›×â•›p22â•›×â•›…â•›× pn2â•›=â•›Râ•›×â•›q12â•›×â•›q22â•›×â•›…â•›×â•›qm2â•›.

As in the case where R = 2, the solitary factor of R on

the right-hand side cannot have a counterpart on the left-
hand side, which means that there cannot be a pair of
integers N and M such that N2â•›=â•›RM2. Hence, there cannot
be a fraction N/M = √R.
Once the Greeks had constructed this beautiful little
argument, they were faced with a considerable challenge.
How could they maintain the link between proportion
(measuring lengths against one another) and the familiar
laws of arithmetic (counting, adding and multiplying)?
Given that √2 is irrational (i.e. given that √2 cannot be
expressed as a ratio of integers), how can we justify calling
it a number? How can we justify the assumption that we
can add and multiply √2 together with the fractions? To
put it another way, how can we numerically express a
geometric length when we know that the length in ques-
tion does not equal any fraction?

Eudoxus, Dedekind and the Birth of Analysis

Thanks to the widespread use of graphs most people find
the concept of a number line highly intuitive. The basic
idea is that every point on a number line represents a
number: a subtle form of metaphor with a long and complex
history. Although graphs as we know them are a relatively
recent innovation, the mathematician and astronomer
Eudoxus (c.â•›408–355 bc) understood that although a contin-
uous number line necessarily contains more than just
fractions, it is made intelligible through a rational frame-
work. Eudoxus’s insight was one of the most pivotal events
in the history of mathe� matics, and to understand the
mystery from which this clarity emerged, let us return to
Pythagoras’ Theorem. We have already seen that whenever
we have a right-angled triangle whose sides are of length
a, b and c, it must be the case that a2â•›+â•›b2â•›=â•›c2. Conversely,
 RATIO AND PROPORTION71

every solution to the equation a2â•›+â•›b2â•›=â•›c2 corresponds to a

right-angled triangle with sides of length a, b and c.
We can be certain that every solution to this equation
corresponds to an ‘actual’ triangle because we can draw
two lines at right angles, and let the lengths of those lines
equal any pair of positive numbers a and b. When we
connect our endpoints, the hypotenuse that we draw must
have a length equal to our third number c, because we
have already specified that our triangle is right angled, and
so by Pythagoras’ Theorem the length of the hypotenuse
must satisfy the relationship a2â•›+â•›b2â•›=â•›c2.
People could draw and identify right-angled triangles
long before the time of Pythagoras. Indeed, some people
believe that as far back as 1800 bc the Babylonians knew
that if a triangle was formed by joining lines of length a,
b and c, then the triangle will be right angled whenever
a2â•›+â•›b2â•›=â•›c2. The main evidence for this claim is a clay tablet
known as Plimpton 322, which systematically lists fifteen
pairs of integers {a, c}. In every case there is some integer
b, such that a2â•›+â•›b2â•›=â•›c2. For example, the tablet contains
the pairs of integer {45, 75}, {1679, 2929} and {12709, 18541},
which form ‘Pythagorean triples’ with the integers 60, 2400
and 13500.
The connection between the geometric facts of right-
angled triangles and purely arithmetic facts such as
32â•›+â•›42â•›=â•›52 has greatly impressed hundreds of generations.
On first inspection there appears to be little connection
between arithmetic and geometric knowledge, for as John
Stillwell writes in his excellent Mathematics and its History:

Arithmetic is based on counting, the epitome of a

discrete process. The facts of arithmetic can be clearly
understood as outcomes of certain counting processes,
and one does not expect them to have any meaning
beyond this. Geometry, on the other hand, involves
continuous rather than discrete objects, such as lines,
72 MATHEMATICAL THOUGHT

curves, and surfaces. Continuous objects cannot be

built from simple elements by discrete processes, and
one expects to see geometric facts rather than arrive
at them through calculation.

Despite the many differences between arithmetic and

geometry, Pythagoras’ Theorem hints at a depth of inter-
connection. Indeed, his now legendary contribution was
to demonstrate that the following statements are logically
equivalent:

1. A triangle whose sides are of length a, b and c

(where c is the longest length) is in fact a right-
angled triangle if and only if
2. A square whose sides are of length c contains the
same area as a square of side a plus a square of
side b.

We could say and understand these individual sentences

long before we could prove their logical equivalence, so
Pythagoras’ achievement is a very striking one. It should
also be noted that as a formula the theorem reads as a state-
ment about the addition and multiplication of quantities,
because the lengths of the given sides are numbers that are
related by the formula a2â•›+â•›b2â•›=â•›c2. The Greeks certainly knew
how to add and multiply fractions, but the lengths mentioned
in the formula may well be irrational, as one can certainly
consider triangles with irrational sides. For this reason (and
many other related ones), it was imperative that the process
of adding and multiplying �arbitrary quantities was under-
stood, and related back to the standard methods for adding
and multiplying fractions.
This problem is far less daunting to the modern reader,
as most people think of a number as being something like
2.713… (a possibly infinite string of digits). With such a
representation, one can naturally extend the operations of
 RATIO AND PROPORTION73

addition, subtraction, multiplication and division in a rela-

tively straightforward way. For example, suppose that we
want to find x = 2.713…â•›×â•›3.425… It isn’t very difficult to
work out how to calculate each of the digits of x in turn:
you just need to be systematic about it. Indeed, just by
looking at the first digits we can deduce that x is bigger
than 2â•›×â•›3 and smaller than 3â•›×â•›4. By looking at the second
digits we can work out that x is bigger than 2.7â•›×â•›3.4 and
smaller than 2.8â•›×â•›3.5, and so on. Crucially, if you want to
find a finite number of digits of x, you only need to look
at finitely many digits to the right of the equals sign. My
point is that when we imagine doing sums with an infinite
string of digits, we simply keep on adding and multiplying
in the time-honoured way, but accept that when we are
using infinite decimals, there is quite literally no compu-
tational end in sight, even for a single multiplication.
That is all well and good, but the central problem
remains. Is it possible to prove that two infinite decimals
are equal (as they may be in Pythagoras’ Theorem)? To
make sense of this question, we must first be clear about
the defining relationships greater than, smaller than and
equal to. For fractions, we have a very simple criterion
for equality:
a c
= if and only if ad = bc.
b d
For example, we say that 1/2 = 3/6 precisely because
1â•›×â•›6 = 2â•›×â•›3. Similarly, we say that:
a c
< if and only if adâ•›<â•›bc.
b d
In other words, we can compare rational numbers by
multiplying and comparing integers. Eudoxus realized that
not only are there well-defined criteria for the comparison
of fractions, there is also a reasoned sense of measure that
applies to every ‘real number’. The crucial and deceptively
74 MATHEMATICAL THOUGHT

simple observation is that two real numbers (points on a

continuous number line) are said to be ‘different’ if and
only if there is a gap between them. If there is a gap, then
there must be a fraction that is bigger than one of the
numbers, and smaller than the other, because every stretch
of the number line contains fractions. This means that two
real numbers are in fact one and the same if and only if
it is impossible to find a fraction that is larger than one
number and smaller than the other.
It is highly significant that equality between real numbers
is defined in terms of computational failure (that is, the
failure to find a fraction that is bigger than one number
and smaller than the other). On a more positive note, we
can show that two different definitions pick out the same
real number by demonstrating that being larger than one
definition and smaller than the other is a logical impos-
sibility for any fraction. In other words, we can prove
equality between real numbers within a rational framework,
and we don’t need to worry about irrational numbers
messing things up.
The other crucial point about Eudoxus’s brilliant idea
is that in order to speak of a number’s size (that is, its
defining characteristic), all that we require is the ability
to say if it is bigger, smaller or equal to any given frac-
tion. For example, we can compare any fraction f to √2
by calculating fâ•›2 and comparing this number to 2. Our
ability to add, multiply and compare fractions is firmly
rooted in our ability to count, as the proper way to add,
multiply and compare fractions is strictly determined by
the proper way to add, multiply and compare integers.
We can therefore be absolutely certain that every fraction
rightly belongs in one of two sets: the set of ‘smaller than
√2’ fractions, and the set of fractions ‘at least as big as
√2’ (which is exactly the same as the set of fractions bigger
than √2).
The idea of partitioning the fractions into ‘smaller’ and
 RATIO AND PROPORTION75

‘larger’ is critically important. Among other things, a rule

for sorting any fraction into either ‘smaller than √2’ or
‘larger than √2’ provides us with a strictly deterministic
method for generating the decimal expansion of √2. The
procedure runs something like this:

12â•›=â•›1 which is less than 2, so 1 must be less than √2.

22â•›=â•›4 which is greater than 2, so 2 must be greater than

√2.

1.42â•›=â•›1.96 which is less than 2, so 1.4 must be less than

√2.

1.52â•›=â•›2.25 which is greater than 2, so 1.5 must be greater

than √2.

1.412â•›=â•›1.9881 which is less than 2, so 1.41 must be less

than √2.

1.422â•›=â•›2.0164 which is greater than 2, so 1.42 must be

greater than √2, …

There are many other techniques that we can use to effec-

tively sort any fraction into ‘smaller’ or ‘larger’. For
example, consider the following method for calculating
whether some fraction a/b is more or less than π, bearing
in mind that π is normally defined as the circumference
of any circle divided by the diameter of that circle, but π
is also equal to the area of a circle with radius one.
76 MATHEMATICAL THOUGHT

We can calculate π by drawing circles on grids of increas-

ingly tiny squares. These squares are coloured white if
they lie completely outside the circle, black if they lie
inside the circle, and grey if the circle crosses them. As
we use finer and finer grids, the black area gets closer and
closer to π, while the grey area becomes arbitrarily small.
If a fraction a/b is different to π (as every fraction is),
then one of these pixellated images will be detailed enough
to show this fact. As Eudoxus understood, ‘a/b is different
to π’ simply means ‘there is a detectable gap between a/b
and π’. That is to say, the sufficiently fine images will
demonstrate that one (and only one) of the following state-
ments is true:

1. The area of the black region is larger than a/b, so

a/b is smaller than π, or
2. The area of the black region plus the area of the
grey region is less than a/b, so a/b is larger than π.

Notice that in each case we are comparing our fraction

a/b against other fractions (the relevant areas). We can
correctly fill in any grid by using Pythagoras’ Theorem,
which tells us the distances between the corners in the
square grid and the centre. Hence our procedure for
checking the relative size of a/b and π depends on nothing
more or less than the knowledge of how to add and multiply
integers.
Another way to specify a particular real number is to
write it as the ‘limit case’ of an infinite sum. For example,
the Indian mathematician Nilakantha Somayaji (1444–1544)
demonstrated that 4 is smaller than 1, bigger than 1
1 ,
3
smaller than 1 1 3 + 1 5 , bigger than 1 1 3 + 1 5 1 7 , and
so on. This sequence of approximations defines a specific
real number because we can always use a finite number
 RATIO AND PROPORTION77

of these approximations to tell whether any given fraction

is greater or smaller than π. It should also be noted that
many different sequences can specify precisely the same
real number, as the following equations indicate:

Nilakantha Somayaji (c.â•›1500):

1 1 1 1 1 1 1
+ =1 + + + ...
3 54 7 9 11 13 15
François Viète (1593):

2 1 1 1 1 1 1 1 1 1
= + + + ...
2 2 2 2 2 2 2 2 2
John Wallis (1655):
2 2 4 4 6 6 8
= ...
2 1 3 3 5 5 7 7

Leonhard Euler (c.â•›1750):

2
1 1 1 1 1 1 1
= + 2 + 2 + 2 + 2 + 2 + 2 + ...
6 1 2
2
3 4 5 6 7
2
22 32 42 52 62
= ...
6 (2 2 1) (32 1) (4 2 1) (5 2 1) (6 2 1)

Recurring Decimals and Dedekind Cuts

When Eudoxus clarified the fundamental relationships
‘greater than’, ‘smaller than’ and ‘equal to’, matheÂ�matics
was essentially geometric in character. However, later
generations have taken a more algebraic approach, and by
the nineteenth century mathe�maticians were once again
wanting to clarify precisely what is meant by a ‘real
number’. Intuitively speaking, real numbers correspond
to points on a number line. As we shall see, this is a rather
subtle concept, despite its intuitive appeal. The modern
78 MATHEMATICAL THOUGHT

definition of the real number line is due to Richard

Dedekind (1831–1916), but the idea that numbers could
be identified as points on a line was first articulated by
René Descartes (1596–1650).
We know exactly how to do arithmetic with some of
the points on a number line (e.g. the integers), but the
problem with saying that every point on a number line is
actually a number is that it isn’t obvious how we are
supposed to do arithmetic using an arbitrary point on a
geometric line. Dedekind recognized that there is a better
way to define what is meant by a real number, which did
not appeal to the geometric notion that a line is the path
traced by a moving point. Dedekind insightfully observed
that every real number can be identified with a pair of
sets, namely the set of fractions smaller than x, and its
inverted twin, the set of fractions at least as big as x. This
complementary pair of sets is called a ‘Dedekind Cut’, for
obvious reasons.
It may be more intuitive to think of real numbers as
being points on a number line, but when rigour is called
for matheÂ� maticians clarify the notion of ‘a point on a
number line’ by turning to the formalism of Dedekind
Cuts. That is to say, specifying a real number x is under-
stood as being the same thing as specifying a way to divide
the set of fractions into two, such that every fraction in
the ‘smaller’ set is smaller than every fraction in the ‘larger’
set. By definition a number x is bigger than y if and only
if there is a fraction that is in the set ‘larger than y’, which
is also in the set ‘smaller than x’. Similarly, the number x
is equal to y if and only if the set ‘smaller than x’ is iden-
tical to the set ‘smaller than y’.
Now, when we first learn about fractions we tend to
think of one number being divided by another (fractions
as a kind of verb). By gaining experience of the addition
and multiplication of fractions, we learn to think of them
as nouns (‘completed things’ in their own right). We must
 RATIO AND PROPORTION79

make a similar conceptual transition from verb to noun

when we specify a number using recurring decimals (e.g.
0.3333…). The idea behind this notation is that we can
use an infinite sequence of terms (0.3, 0.33, 0.333, …) to
pick out one particular number. That is to say, we can
specify one particular Dedekind Cut by putting fractions
in the ‘smaller’ set if and only if they are smaller than one
of the terms 0.3, 0.33, 0.333, … This leads us to our next
question: are there any numbers that have regular, recur-
ring digits, but which do not equal any fraction? In
particular, could √2 have recurring digits, or must it
continue forever un-repeating?
We can answer this question by appreciating some of
the algebraic qualities of recurring decimals. Consider, for
example, the number 0.234523452345… By naming this
number x, we can make the following argument:

10000x = 2345.23452345… and

x = 0.23452345… therefore
9999x = 2345 (subtract the second equation from the
first).

Dividing both sides by 9999 gives us the equation

2345
x = 0.2345… = .
9999â•›
We can shift the recurring digits along by adding an appro-
priate number of zeroes to the denominator. For example,
2345
0.00234523435 …â•›999900â•›. We can also find an equivalent
fraction when there is a finite string of digits before the
recurring ones. For example,
66 2345 662279
0.6623452345 … =â•› â•›+â•› â•›=â•› .
100 999900 999900â•›
It should now be clear that every recurring decimal can
be written as a fraction. This means that √2 definitely
80 MATHEMATICAL THOUGHT

cannot have recurring digits, because unlike numbers with

recurring digits, √2 cannot be written as a fraction. One
particular application of this argument highlights the intui-
tion refining power of Eudoxus’s conception of equality.
When x = 0.999… we have:

10x = 9.9999… and

x = 0.9999… therefore
9x = 9 and so x = 1.

Many people are somewhat perturbed to find that the

recurring decimal 0.999… is equal to one. But we can see
that this is true, because if a fraction is smaller than one
of the numbers 0.9, 0.99, … , then it must also be smaller
than one. Furthermore, the number one is the smallest
fraction that is not smaller than any of the fractions 0.9,
0.99, 0.999, … This implies that one is the smallest member
of the set ‘larger than 0.999…’ Hence the symbols 1 and
0.999… indicate precisely the same Dedekind Cut, which
is another way of saying that they are one and the same
real number.
The notion that there is an infinitesimal difference
between 0.999… and 1 is not a part of ordinary matheÂ�
matics (though there are forms of mathe�matics that include
infinitely small quantities). In other words, the number
one is the only real number that 0.999… can equal. It may
seem strange to say that the difference between 1 and
0.999… is actually zero, but in the words of the great
mathematician Leonhard Euler, ‘To those who ask what
the infinitely small quantity in mathe�matics is, we answer
that it is actually zero. Hence there are not so many
mysteries hidden in this concept as they are usually believed
to be.’ More generally, our analysis shows that every recur-
ring decimal is equal to some fraction, and by considering
the process of long division we can see that the converse
of this statement is also true. In other words, every frac-
 RATIO AND PROPORTION81

tion has an infinitely repetitive decimal expansion, and

every recurring decimal is equal to some fraction.

Continued Fractions
Decimal notation is an extremely powerful and convenient
way to represent numbers. However, it is far from being
the only system, and there is an alternative form of repre-
sentation, which is particularly elegant and interesting. The
forms I am referring to are known as continued fractions,
and many historians believe that they were studied by the
Ancient Greeks (who were certainly familiar with the key
ideas). Although he didn’t claim to invent them, our earliest
unambiguous record of the use of continued fractions
comes from a book called L’Algebra, which was written
by Raphael Bombelli in 1572.
So what are these things called continued fractions? Well,
we are all familiar with the idea that one divided by two
is a number, which can be written in the form 12 . We are
also familiar with the idea that if 1 is a number, then so
1 2
is 2â•›+â•›2 . But since 2â•›+â•›12 is a number, does it not follow that
one divided by this number is also a number, namely
1
2â•›+â•›12 â•›? And why shouldn’t we continue adding integers and
dividing the resulting number into one? For example, it
1
should be clear that is also a specific number, it
3â•›+â•› 1 1
2â•›+â•›2
has just been written in an unusual manner.
Numbers of this form are called continued fractions. In
the case of a finite chain we can always rewrite our number
as an ordinary fraction, but that may not be the case if our
numbers form an infinite chain. As we are most used to
thinking of arbitrary numbers in terms of an infinite
sequence of decimals, it is instructive to think about how
we can convert a decimal number into the form of a
continued fraction. The fundamental observation is that
82 MATHEMATICAL THOUGHT

every real number x can be written in the form ⎣x⎦â•›+â•›Δ(x)

where ⎣x⎦ is the integer part of x and Δ(x) is some remaining
real number between 0 and 1. For example, the integer part
of 2.269 is 2, while the remainder ∆(2.269) is 0.269. A greater
insight comes from observing that for every non-integer
real number x, there is an integer part ⎣x⎦ but there is also
1
an integer part corresponding to the real number Δ(x) .â•›
For example, when x = 2.269 we have an integer part
1
2. The remainder is 0.269, and 0.269 â•›=â•›3.71 … The integer
part of 3.71 … is 3, and this tells us that the number 2.269
is close to 213 . By an iterative process of inverting a sequence
of remainders and taking integer parts, we can generate a
specific sequence of integers that represent our real number
x. In this example the first integer is 2 and the second
integer is 3, but more generally for every real number x
there is an integer ⎣x⎦ and a (possible infinite) sequence of
positive integers r1,r2, … such that:
1
x= x +
1
r1 +
1
r2 +
r3 + ...
Writing a number x in terms of the above notation is known
as ‘expressing x as a continued fraction’. Furthermore,
finding this sequence of integer parts is equivalent to
applying the Euclidean algorithm. In other words, given
a real number x, we can find the corresponding continued
fraction by drawing a 1:x rectangle, and then systemati-
cally filling in as many squares as we can, using the largest
possible square at each step.

When x = 2.269 we have ⎣x⎦â•›=â•›2, and we can fit two squares

inside the rectangle.
 RATIO AND PROPORTION83

We can fit three squares into the space that remains, which
tells us that r1â•›=â•›3.

We can only fit one square in the space that remains, which
tells us that r2â•›=â•›1.

If there is a length that exactly divides both sides of the

rectangle, then our procedure comes to a halt when we
find that length. This happens if and only if the number
x is expressible as the ratio of two integers. As a repre-
sentation of number continued fractions are rather like
decimal fractions, in the sense that the first numbers tell
us the most about the overall size of x. Furthermore,
the rational sequence 2, 2â•›+â•›13 , 2â•›+ 1 is comprised of
3â•›+â•›1
â•› 1
increasingly accurate approximations for our real number
x = 2.269, just as the sequence 2, 2.2, 2.26, … is composed
of increasingly accurate approximations for the underlying
number 2.269 (say).
We can rewrite the terms ⎣x⎦, ⎣x⎦â•›+â•›1r , ⎣x⎦â•›+â•› 1
1
r1â•›+â•›1r
2

as ordinary fractions, and as the sequence progresses the

terms become arbitrarily close to x. This is crucial,
because it means that sequences such as these determine
one parti� cular Dedekind Cut. Furthermore, the odd-
numbered terms in the sequence are all at least as small
as x. For example, our first estimate ⎣x⎦ must be at least
as small as x. Similarly, the rounding down in the third,
fifth and seventh approximations produces a number that
is at least as small as x. Conversely, all the even-numbered
terms are at least as big as x. This follows because ⎣x⎦ is
at least as small as x, which means that ⎣1x⎦ must be at
least as big as 1x â•›.
84 MATHEMATICAL THOUGHT

Quadratic Equations and the Golden Ratio

Finite continued fractions can be rewritten as ordinary
fractions, which means that finite continued fractions can
be used to specify precisely the same real numbers as recur-
ring decimals. But what kind of real number can be specified
by an infinitely recurring continued fraction? Consider
the following example:
1
1+
1
1+
1
1+
1+ ...

This is a highly exalted number, and in keeping with tradi-

tion I shall label it (or phi). By definition, the number
1
satisfies the equation = 1+ . If we multiply both sides
by 2â•›, we get
2
= + 1 . We can solve this equation
using the standard ‘quadratic formula’, which tells us
1+ 5
that = , or 1.618033989 … Alternatively, the
2
1
equation = 1+ can be used to generate its own solution
(as a continued fraction), by an iterative process of repeated
substitution. We simply replace each occurrence of with
1
1+ , and by doing this over and over again we generate
the continued fraction
1
= 1+
1
1+
1
1+
1+ ...
 RATIO AND PROPORTION85

The ratio 1:1.618… (or equivalently 0.618… : 1) is called

the ‘golden ratio’, and rather like π it seems to crop up
everywhere. Many natural patterns exhibit the golden ratio,
as do many of the world’s most famous works of art. For
example, the Ancient Greeks believed that the ideal human
form should be full of golden ratios (e.g. the ratio from
navel-to-toe to head-to-toe ‘should’ be golden), the Mona
Lisa was carefully proportioned by the repeated use of the
golden ratio, and the modernist Le Corbusier advocated
the use of the golden ratio when designing buildings or
furniture.
I will return to this particular example of a continued
fraction presently. First I want to point out a very general
connection between quadratic equations and continued
fractions with a repetitive structure. For example, consider
the number
1
x =2+
1
1+
1
2+
1+ ...

Because this continued fraction has a repetitive structure,

1
we can replace the second occurrence of 2 + with an
1+ ...
x, to give us the equation
1
x =2+ .
1+ x â•›
It is possible to prove that given any equation that describes
a repetitive, continued fraction, we can algebraically
rearrange that equation into the form ax2â•›+â•›bxâ•›+â•›câ•›=â•›0, where
a, b and c are ordinary fractions. Equations of this form
are known as ‘rational quadratic equations’. In our
particu�lar example of a repetitive continued fraction, some
86 MATHEMATICAL THOUGHT

elementary algebra can prove that x2â•›−â•›2xâ•›−â•›2â•›=â•›0, so we have

x = 1 + √3. In other words:
1
3 = 1+
1
1+
1
2+
1
1+
2 + ...
We have just seen that repetitive continued fractions can
solve rational, quadratic equations. When he was only
seventeen, the wildly original and tragically short-lived
Evariste Galois (1811–1832) demonstrated that there is also
a converse to this argument. In other words, it is an alge-
braic fact that a real number x can be the solution to a
rational quadratic equation if and only if x can be written
in the form of a repetitive continued fraction. For example,
suppose that we start with the equation x2â•›=â•›2. We are free
to add an x to both sides of the equation, because this
operation preserves equality. Furthermore, xâ•›+â•›x2â•›=â•›2â•›+â•›x is
equivalent to x(1â•›+â•›x)â•›=â•›(1â•›+â•›x)â•›+â•›1. Dividing both sides by
1
(1 + x) gives us xâ•›=â•› 1 + . By repeatedly substituting
1 1+ x
1+ for every occurrence of x, we effectively home in
1+ x
on the only positive solution to the equation x2â•›=â•›2, which
is 1 or 1.41421.
1+
1
2+
2 + ...
Galois’ other great achievement was even more remark-
able. The night before he was killed in a duel, he spent
his time writing an explanation of what is now known as
‘Galois Theory’. His conclusion was that equations that
involve a term x5 are fundamentally different to equations
that only involve x, x2, x3 or x4. As you may recall, if we
are given any equation of the form ax 2 + bx + c = 0 , we can
use a simple formula to find all the values of x that solve
 RATIO AND PROPORTION87

b ± b 2 4ac
our equation (namely x = ). Similarly, there
2a
is an equation for finding the solutions to cubic equations,
and there is also an equation for finding the solutions to
equations involving x4. Many mathe�maticians had tried to
find a general rule for solving equations in x5, but Galois
is famous for proving that an equivalent formula cannot
possibly exist.

Structures of Irrationality
If a real number is expressible as the ratio of two integers,
we say that it is a rational number. Every real number
is either rational or irrational, but some numbers are
more irrational than others. That is to say, some irrational
numbers are even less like fractions than other irra-
ti�onal numbers. To understand why this is so, imagine
picking a point on a number line, then trying to find a
rational approximation that is both simple and accurate.
For every integer n, we can find the best (or joint best)
estimate of the form m/n.

1/n 2/n x 3/n 4/n

1
It should be clear that every real number is at most
2n
away from one of these fractions, and this maximum error
m m +1
occurs when x is halfway between and . Also
n m n m +1
notice that if x is close to halfway between and ,
1
mâ•›+â•›2 2mâ•›+â•›1 n n â•›
that means it is close to â•›=â•› .
n 2n ╛╛
If we use larger and larger values of n, we can find
increasingly accurate approximations to our number x.
Furthermore, for every integer K there is a ‘best rational
approximation’ of the form m/n, where mâ•›≤â•›K. The previous
88 MATHEMATICAL THOUGHT

statement must be true because there are finitely many

‘sensible’ approximations of this form, so there must be
at least one approximation that is as close to x as any of
them.
For any particular integer K there will be some numbers
that are very close to their best rational approximation,
while other numbers will be relatively distant from their
best rational approximation (where we insist that any
approximation m/n is such that mâ•›≤â•›K). Later in this chapter
we will see that some numbers are poorly approximated
for every integer K. In the case of these highly irrational
numbers, every integer ratio is a relatively poor approxi-
mation for the actual number.
To understand the structure of irrationality a little
better, we need to appreciate a particular fact about the
sequence of approximations generated by the Euclidean
algorithm. Suppose that we generate the approximation
m/n by our method of taking integer parts. It is possible
to prove that for any fraction a/b, one of the following
must be true:

1. m/n is closer to x than a/b, or

2. a is greater than m.

In other words, given a real number x, the Euclidean algo-

rithm generates approximations of x that are the best
rational approximations. The next step is to appreciate the
pivotal relationship between the accuracy of these approxi-
mations, and the size of the next integer in the continued
1
fraction. For example, r1 = is large whenever ∆(x)
(x)
is small. Furthermore, if the remainder Δ(x) is small,
our initial approximation ⎣x⎦ must have been close to x.
More generally, the larger the value of rn , the smaller the
 RATIO AND PROPORTION89

1
difference between rn 1 and rn 1 + , and the smaller
rn
the difference between consecutive approximations.
As we work our way along the sequence of approxima-
tions generated by the Euclidean algorithm, we find an
under-estimate, followed by an over-estimate, followed by
an under-estimate, and so on. Because we alternate between
under- and over-estimates, the difference between the n’th
approximation and the true value of x must be less than
the difference between the n’th approximation and the
n+1’th approximation. It follows that if the integer is
large, the n’th remainder must be small, and the n’th
approximation must be very accurate.
1 22
As an example, 3 + = is a pretty good approxima-
7 7
tion of π. We might say that π is well approximated by
single-digit fractions, because most real numbers don’t
have such a simple but accurate approximation.
Correspondingly, the third fraction approximation of π
1
3+
is 1 . One fifteenth is fairly small, so improving
7+
15
our estimate by including this factor results in a relatively
small decrease in the overall size of our approximation.
This implies that our original estimate must have been
close to π (as indeed it was). On the other hand, π is
poorly approximated by fractions with two-digit denom-
inators, in the sense that most numbers can be
approximated more accurately. This corresponds to the
fact that the fourth digit in π’s continued fraction is a
one (as are almost half of the first million integers of π’s
continued fraction).
In a far more extreme case, it occurred to the Indian
mathematician Srinivasa Ramanujan (1887–1920) that the
90 MATHEMATICAL THOUGHT

number e 163 must be almost exactly an integer. In fact,

there are seventeen digits before the decimal point, followed
by thirteen zeroes. Correspondingly, the second term in
the continued fraction form of e 163 is a whopping
1,333,462,407,511.

The Fibonacci Sequence

Leonardo Fibonacci (c.â•›1170–1250), also known as Fibonacci
of Pisa, was an Italian merchant who travelled extensively,
studying with Arab scholars. His influential book Liber
Abaci was the first European textbook to cover decimal
notation, and the now familiar techniques of multiplication
and long division (operations that are much, much easier
to carry out with decimals than Roman numerals).
Fibonacci also helped to spread the basic ideas of algebra:
a development that we shall examine in the next chapter.
Ironically, most people have heard of Fibonacci because
in 1877 a number theorist called Edouard Lucas was stud-
ying the sequence 1, 1, 2, 3, 5, 8, 13, … , and he decided
to pay tribute to Fibonacci by calling that sequence the
Fibonacci sequence. Fibonacci himself said very little about
this sequence, but he would have recognized that it was
the solution to one of the more whimsical problems he
included in his book. That is to say, Fibonacci once posed
the following question: If we start with a single pair of
baby rabbits, how many pairs of rabbits will we have each
month assuming that:

1. There are no deaths, and

2. Every month each pair produces one new pair,
which becomes productive in the second month
after birth.

The answer for each successive month can be found in the

following sequence:
 RATIO AND PROPORTION91

At each stage we add the two preceding terms to get the

next in the sequence. The previous term accounts for all
the rabbits that are surviving, and the term before that is
equal to the number of mature pairs. As each mature pair
produces a new born pair each month, the total of these
two terms is equal to the total number of pairs of rabbits.
As we move along this sequence, the ratio between succes-
sive terms gets closer and closer to 1.618… (the golden
ratio). This is related to the fact that terminating the
1
continued fraction = 1+ after finitely many steps
1
1+
1+ ...
generates the following sequence of best rational approx-
imations for :
1 2 3 5 8 13 21 34
, , , , , , , , ...
1 1 2 3 5 8 13 21

Both the numerators and the denominators form the

sequence we found with Fibonacci’s rabbit problem, and
so the sequence of integers 1, 1, 2, 3, 5, 8, 13, 21, … is
known as ‘the Fibonacci sequence’. Returning to the golden
ratio, the above sequence of rational approximations to
have a fascinating and highly significant property. Because
1/1 is the largest fraction of the form 1/n, the infinite part
of our continued fraction (the bit we effectively ignore by
stopping after finitely many steps) makes as big a differ-
ence as possible. In other words, every single step changes
our approximation by the maximum possible amount, given
92 MATHEMATICAL THOUGHT

our method of taking integer parts. This tells us that we

have the maximum possible gap between every consecutive
pair of under- and over-estimates, which means that none
of our approximations are very good.
Because our method of generating continued fractions
gives us the best possible approximations (for their size),
the golden ratio really is the hardest number to approxi-
mate using fractions. This means that is the most irrational
number of all! To reveal the beauty of this truth, consider
a kind of automated drawing machine. These machines
operate in a step-by-step fashion, moving a single shrink-
able arm that pivots around the centre of each drawing.
At each step the arm rotates through some fixed angle R,
shrinks by some fixed proportion P, and then leaves a
mark. To make the point clearer, I have also drawn lines
around the edge of a containing circle to show the different
directions that the arm was pointing in as it left the marks.
If we use a fraction or an integer for our angle R the
1
marks lie in straight lines. Indeed, when R = 360° we
n
m
generate n radial lines. More generally, when R = 360°
n
(and m and n have no factors in common), we generate n
radial lines.
Using rational values of R generates straight lines, while
using an irrational angle guarantees that no two marks will
lie on a straight line through the centre. This follows
because finding two marks that lie on a straight line through
the centre is equivalent to finding three integers a, b and
c, such that:

(a R) = (b R) + (c 180°) .
c
But in that case we would have R = 180°, which
a b
can only be true if R is rational. Although there is this
 RATIO AND PROPORTION93

fundamental difference between the patterns produced by

rational and irrational values of R, similar values produce
similar patterns, so every finite drawing looks like a (more
or less) twisty version of some fraction. Reasonably
straight lines indicate the presence of a ‘good’ fraction
approximation:

R = 60.2°, just over one sixth of a circle.

If the length of the arm stays relatively constant (i.e. if P

is relatively close to one), there will be more points near
the edge of our containing circle, so the eye tends to pick
out a larger number of spiralling lines. This means that if
we increase P we tend to notice that R is similar to some
new fraction (one with a larger denominator). If R is very
similar to the fraction in question, then the spiralling lines
are relatively straight, and the marks around the edge of
our containing circle are closely bunched.
94 MATHEMATICAL THOUGHT

R = 2 60° for all three drawings, and P increases from

left to right.
The left most drawing has four spiral arms, illustrating
1
the approximation 2 60° = 84.85 360° . The third
4 4
drawing shows that 2 60° 360° , which is a much
17
better approximation.
When R = 360° (or equivalently, if R = 137.508 …),
we generate the pattern that is most different from any
fraction:

R = 137.5°, P = 0.98.

When R = 360° we can use any value for P and our

drawing will still spiral dramatically. That is because there
simply isn’t any good approximation for the eye to notice!
Furthermore, the lines around the edge of our containing
circle are always evenly spaced, no matter how many points
we add. Indeed, each new line around the edge falls into
the largest gap left by the previous marks.
 RATIO AND PROPORTION95

Many plants avoid growing branches, petals or leaves

directly above one another, and such arrangements
frequently contain golden ratios. Sunflowers, pineapples,
raspberries, daisies and pinecones all exhibit particularly
clear patterns that are based on the golden ratio. If you
count the number of clockwise spiral lines formed by the
pieces of these things, you (almost) always find a number
in the Fibonacci sequence. The number of anticlockwise
spiral lines will be an adjacent number in the sequence.
Because the total number of pieces is the same whichever
direction you count, the average number of points along
the clockwise spirals will be in a golden ratio to the average
number of points along the anticlockwise spirals.
People have marvelled at the natural preponderance of
golden ratios for millennia. In particular, since the early
nineteenth century biologists and mathe�maticians have been
intrigued by the phenomena of spiral phyllotaxis: the
golden arrangement of leaves, flowers or other lateral
organs sticking out of a central axis or stem. It is clearly
inefficient to grow a leaf directly above another leaf because
one would put the other in the shade, so we might expect
that plants would evolve non-overlapping leaf arrange-
ments. Furthermore, the non-overlapping patterns that
plants form ought to be ones that can be extended to
accommodate more and more leaves, as plants grow more
leaves over time.
As we have seen, growing each leaf at a golden angle
relative to the one below is an excellent strategy, as it
ensures that no matter how many leaves you grow, none
of them will be directly above the leaves you have already
grown. However, it is far from obvious that selective pres-
sure to avoid overlapping necessarily produces the golden
ratio, rather than one of the many other non-overlapping
arrangements. It is only in recent years that the key mech-
anisms determining the location of leaves have begun to
be understood, and, in particular, we now know that the
96 MATHEMATICAL THOUGHT

plant hormone auxin plays a critical role. Cells with high

levels of auxin tend to grow more rapidly, and high levels
of auxin tend to induce the production of a leaf.
Furthermore, cells actively pump the auxin they contain
though special export channels, and the export channels
of each cell tend to face whichever neighbouring cell
contains the most auxin.
As a result of this process cells that contain relatively
high levels of auxin tend to accumulate even more, while
the neighbouring cells tend to deplete their stores.
Furthermore, when a cell acquires higher and higher levels
of auxin, a leaf begins to form. This process effectively
drains the auxin from the neighbouring cells, inhibiting
the growth of other leaves in the immediate vicinity. As
the stem continues to grow, it is unlikely to grow a leaf
directly above another leaf, because that part of the stem
will have grown out of a region where the auxin concen-
tration is particularly low (since the cells in that region
are busy pumping auxin back towards the newly forming
leaf). The fact that some cells grow faster than others
means that there are asymmetric mechanical forces as
the growing cells push against their neighbours, and in
many plants the interplay between growth dynamics and
auxin transport means that the places where leaves will
grow tend to be related to one another by the golden
angle.
Of course, nothing in biology is mathe�matically exact,
but the patterns produced by many different plants closely
resemble the patterns of the golden angle. Indeed, through
a combination of computer models, physical models and
biological experiments, it is becoming increasingly clear
that the relationship between the golden angle and non-
overlapping works both ways. Not only does the golden
angle produce patterns with minimal overlaps, it is also
the case that growth procedures that avoid overlapping
have a strong tendency to produce golden angles.
 RATIO AND PROPORTION97

We will have more to say about the role of mathe�matical

models in biological science. But before we can appreciate
the importance of this exciting new branch of knowledge,
we should return to the beginnings of mathe�matical science.
In particular, we should examine two ideas that have funda-
mentally shaped the way we think about the world: decimal
notation and algebraic equations. Over the course of the
next chapter I shall discuss the invention of zero, and we
will see how the key ideas of algebraic number theory passed
from India to the Muslim world, before reaching Europe.
 Chapter 4:
THE RISE OF ALGEBRA
‘There are the nine figures of the Indians: 9 8 7 6 5
4 3 2 1. With these nine figures, and with the sign 0
which in Arabic is called zephirum, any number can
be written, as will be demonstrated.’
First lines of the Liber Abaci, written by
Leonardo Fibonacci in 1202

Zero and the Position System

The modern system of writing numbers as a sequence of
digits is so simple and convenient that we tend to take it
for granted, but it is difficult to overstate the importance
of this brilliant innovation. As the mathematician Alfred
North Whitehead (1861–1947) once wrote:

Before the introduction of [decimal notation], multi-

plication was difficult, and the division even of the
integers called into play the highest mathe�matical
faculties. Probably nothing in the modern world
could have more astonished a Greek mathematician
than to learn that, under the influence of compulsory
education, the whole population of Western Europe,
from the highest to the lowest, could perform the
operation of division for the largest numbers. This
 THE RISE OF ALGEBRA99

fact would have seemed to him a sheer impossibility.

… Our modern power of easy reckoning with
decimal fractions is the most miraculous result of a
perfect notation.

Crucially, you cannot have decimal notation without a

symbol for zero. The idea that nothing can be viewed as
a number is a rather strange and subtle concept, but it is
terribly important. Indeed, it is no exaggeration to say that
mathe�matics, and therefore science, technology and culture,
could not possibly be what they are today without the
number zero. The number zero as we understand it today
originated in India, but historians now believe that the
proper place to start the story of zero is in Ancient
Mesopotamia, where a succession of civilizations used clay
tablets to keep numerical records. A wide range of number
systems have been used over the ages, but a major advance
occurred around the third dynasty of Ur (c.â•›2100 bc), when
the Sumerians began to use position as part of their numer-
ical system.
The nameless scribe who first began to use positional
numeration was a real genius, and this exceptionally
convenient idea is central to the modern decimal system.
For example, we all know that ‘23’ represents two lots of
ten plus three, while ‘32’ represents three lots of ten plus
two. This demonstrates the fact that our notation is posi-
tional, which means that changing the relative position of
our digits yields a different number. The same digit ‘3’ can
stand for three (as in ‘23’), thirty (as in ‘32’), three hundred
(as in ‘302’), and so on. This contrasts with Roman numerals
or the notation used by the Ancient Greeks, where the
symbol v always stands for five, the symbol x always
stands for ten, and so on.
There is, however, one crucial difference between modern
numeration and the earliest position based systems: they
didn’t have a symbol like ‘0’. To understand how this
100 MATHEMATICAL THOUGHT

exceptionally important change occurred, it is instructive

to consider something like the counting boards or abaci
that almost every civilization has developed. That way we
can appreciate how people could do without such a symbol,
and how helpful it was when the symbol was first intro-
duced. The Sumerians used a system based around the
number sixty, rather than base ten. Indeed, the fact that
we divide circles into 360 degrees, hours into 60 minutes
and minutes into 60 seconds is part of our inheritance
from Mesopotamian civilizations. However, for the sake
of simplicity, let us consider the example provided by a
Madagascan method for counting armies.
What would happen is that soldiers were led in single
file, and as each one passed, a pebble was dropped into a
bowl. Once the bowl contained ten pebbles it was emptied,
and a single pebble would be dropped into a second bowl,
used for counting groups of ten soldiers. Similarly, once
the second bowl contained ten pebbles it was emptied, and
a single pebble would be dropped into a third bowl, used
for counting groups of a hundred soldiers. Now, imagine
that the authorities wanted to keep a record of how the
counting process had gone. This record might have read
‘The third bowl held six pebbles, the second bowl was
empty, and the first bowl held four pebbles.’ This account
would tell the reader that there were six hundred and four
troops. Note that whatever words or notation were used
to say that the second bowl was empty played an essential
role, and this role is now performed by the symbol ‘0’.
However, counting and record keeping in this manner does
not require us to think of zero as a proper number.
Even when this kind of number record becomes highly
abbreviated or systematized, we are not necessarily led to
the concept of zero. For example, suppose people wrote
‘6E4’ to indicate six pebbles in the third bowl, an empty
second bowl and four pebbles in the first. Our symbol ‘E’
can still be read as something rather like a punctuation
 THE RISE OF ALGEBRA101

mark, and not as something that is equivalent in kind to

the other numerals ‘6’ and ‘4’. Indeed, the same comment
could be made about modern decimal points. They appear
in numerical expressions alongside the digits 0 to 9, but
no one thinks the decimal point is a number.
Returning to the Ancient Sumerians, it is remarkable to
note that even though they developed positional notation,
and (somewhat later) a symbol comprised of two diagonal
lines that worked rather like the modern ‘0’, their system
did not spread to other civilizations. I say that this is
remarkable because compared with Roman numerals (say),
positional numeration is awesomely convenient. For
example, imagine trying to multiply lxxxvii by cxi, and
then think how much easier it is to multiply 87 by 111.
The crucial point is that Roman numerals, like the English
words for numbers, have very little connection to the work-
ings of an abacus. That is to say, adding groups of ten is
entirely akin to adding groups of a thousand, but this
mathe�matical truth is not well reflected in the English
language or in Roman numerals.
In particular, if you want to multiply a pair of numbers
that are represented as Roman numerals, you cannot simply
break down the problem by multiplying one digit at a
time. In contrast, positional notation naturally leads us to
the observation that 87â•›×â•›111â•›=â•›8,700â•›+â•›870â•›+â•›87. For similar
reasons, addition and subtraction are much easier with
positional notation, as like an abacus, the notation shows
us that we can add one numeral at a time. Furthermore,
positional notation makes it much easier to write large
numbers, as the same numerals are used for numbers of
every size. Indeed, in Roman numerals, even a number as
small as one million can be represented only by an unwieldy
and easily misread string of one thousand Ms.
Alexander the Great must have encountered important
records that were written with positional notation, and so
presumably the idea made its way back to Greece, but for
102 MATHEMATICAL THOUGHT

whatever reason it didn’t catch on, and was soon forgotten.

In any case, our modern number system doesn’t just require
the basic idea of positional notation. Negative numbers
are another essential concept, as it is hard to believe that
any civilization could even invent the number zero without
first using negative numbers. The use of negative numbers
started sometime around 200 bc, when the Chinese started
using red sticks on a counting board to indicate a credit,
while black sticks indicated a debt. Ancient textbooks for
Chinese civil servants included instructions that said ‘If
your board contains red sticks and black sticks, remove
an equal number of both.’ In other words, they understood
that equal-sized positive and negative numbers cancel each
other out.
Of course, using the integers to represent both credits
and debts is not the same thing as having a single number
system that incorporates both positive and negative
numbers, even if it is well understood that subtracting a
credit of n from someone’s ledger is equivalent to adding
a debt of n, while subtracting a debt of n is equivalent to
adding a credit of n. My point is that to have a proper
system of positive and negative integers, we need to be
able to add, subtract and multiply any pair of integers to
get another integer, and we cannot have a number system
of this type without first conceptualizing ‘nothing’ as a
number. This conceptual leap was a key event in the history
of ideas, and it took place in India.
Nobody knows who invented the symbol zero, but there
is little doubt that such a symbol was in existence in parts
of India as far back as 500 bc, although back then it was
not widely used. At first this symbol acted as a kind of
placeholder, which represented an empty column on a
counting board. It was a brilliant and convenient addition
to the local positional system, but for a thousand years or
so, the zero symbol did not really represent a number as
such. The first truly definitive remark concerning the
 THE RISE OF ALGEBRA103

number zero was made by a mathematician and astronomer

called Brahmagupta of Gujarat (c.â•›598–670). He said that
zero is the number you get when you combine a credit
and a debt of equal value (i.e. he observed that ‘nâ•›−â•›n =
0’). He also stated that one plus zero is one, one minus
zero is one, and one times zero is zero. These axioms were
crucially important, because they put zero into the same
conceptual domain as the other numbers. For example, we
might say six minus five leaves a credit of one, five minus
six leaves a debt of one, while six minus six leaves zero,
viewing all three statements as being of a similar kind.
Brahmagupta’s name for the number zero was ‘sunya’,
which means void or empty, and long before the time of
Brahmagupta, the word sunya was used to describe a
column on a counting board that did not contain any
markers. Sunya could also be translated as ‘space’, since
Ancient Indian architects noted that they did not design
walls so much as the space (sunya) between the walls. The
various uses of the word sunya before the invention of the
number zero are telling, and it is no coincidence that
the  matheÂ�maticians who developed the concept of zero
belonged to the Vedic tradition, where discussions of the
qualities of the void or ‘emptiness’ had been taken seri-
ously for thousands of years. Indeed, I think it is
significant that where Christians, Muslims and Jews tend
to associate the divine with the infinite, Hindus, Buddhists
and Jains more frequently associate the divine with a
conception of emptiness, as the very essence of nirvana is
the absence of desire.
When the Arabs adopted Indian numeration in the tenth
century, the word sunya was translated to the Arabic word
for empty, sifr. The English words cipher and decipher are
derived from the word sifr, as when the modern, decimal
system was introduced to Europe, many people needed to
decipher such numbers into Roman numerals before they
could understand them. The word zero comes from the
104 MATHEMATICAL THOUGHT

same root, as in the early thirteenth century the word sifr

was Latinized into zephirum, which later developed into
the word zero.

Al-Khwarizmi and the Science of Equations

The history of mathe�matics would be very different if it
were not for the influence of people from Iran and Iraq.
Before and after the rise of Islam, the ruling elite in that
part of the world took a serious and scholarly interest in
the intellectual developments of their neighbours, which
included the exceptionally sophisticated mathe�maticians
of Constantinople, Alexandria, India and China. Their great
legacy of exact science was based on a remarkably broad
assessment of humanity’s matheÂ�matical knowledge, with
scholars searching far and wide in order to further their
own intellectual interests. For example, translations were
made of the Siddhantas (important Hindu texts concerned
with mathe�matics and astronomy), and it is highly prob-
able that there were scholars of Brahmagupta’s masterpiece
the Brâhmasphutasiddhânta, which was the world’s most
advanced number-theoretic text when it was written back
in adâ•›628. As well as introducing zero and rules for manipu-
lating negative and positive numbers, this influential work
introduced a fully general form of the quadratic equation,
which we still use today. Most advanced of all, Brahmagupta’s
masterpiece also contained a technique for solving certain
‘Pell equations’, such as x 2 92y 2 = 1.
Arab scholars studied widely, acquiring mathe�matical
knowledge that was utterly unknown to Europeans. They
also preserved and studied the works of Ancient Greece,
maintaining traditions that Europe had lost. As well as
performing the invaluable service of pooling and consoli-
dating the world’s matheÂ�matical knowledge, the Arabs also
developed their own innovations, and crucially they intro-
duced Europeans to the many benefits of the decimal
position system. The most famous and influential of the
 THE RISE OF ALGEBRA105

Baghdadi matheÂ�maticians was Abu Ja’far Mohammed ibn

Musa al-Khwarizmi (c.â•›a dâ•›780–850), or al-Khwarizmi for
short. The word algorithm is a Latinization of his name,
and the word algebra is derived from the title of his most
famous book, Al-Jabr W’al Mûqabalah, or ‘Calculation
by Restoration and Reduction’.
It is hard to overstate the importance of the basic ways
that scientists and mathe�maticians manipulate the symbols
in equations to produce other, equally valid equations. For
example, following al-Khwarizmi, we learn that we can
collect a number of terms together on one side of the
equals sign, or multiply both sides of an equation by any
convenient number. Although al-Khwarizmi fully deserves
his fame, it is only fair to stress that his general methods
for finding unknown quantities were developments based
on a number of ancient traditions. In particular, some people
refer to Diophantus (c.â•›210–294) as the father of algebra.
Very little is known about Diophantus, but he lived in
Alexandria in the second century ad, and famously worked
to identify integer solutions to various kinds of polynomial
equation. That is to say, he studied equations involving
the addition and multiplication of integers together with
some unknown quantity, such as 2x2â•›+â•›10â•›=â•›60. His distinc-
tive contribution was not only to use a single symbol to
represent an unknown quantity x (as earlier authors are
thought to have done): he also used a symbol for the square
of an unknown quantity, as well as the cube of an unknown
quantity.
The mathe� matics of Diophantus was highly original
because he was the first person to relate unknown quantities
in equations that can have an infinite number of solutions.
However, the equations that he studied were always related
to geometric problems, and in that context his work did
not naturally lead to a radically new branch of math.
The fundamental ways that we can legitimately rearrange
or manipulate equations were essentially � identified by
106 MATHEMATICAL THOUGHT

al-Khwarizmi, and his Al-Jabr was accepted for centuries

as the definitive text on the science of handling arithmetic
problems involving unknown quantities.
The crucial difference between the Ancient Greeks and
the Arabs was that the Arabs considered linear equations,
quadratic equations, cubic equations and so on to be distinct
and fundamental categories of problem. The Greeks treated
such problems as part of their assault on geometry, while
the Arabs saw that the language of equations can be applied
to a whole range of mathe�matical areas of interest, including
geometry and number theory. They therefore recognized
that finding the legitimate ways to manipulate equations
is a worthy enterprise in its own right. For example,
although al-Khwarizmi worked with word equations, and
not a symbol like the modern x, he effectively stated that
an equation of the form xâ•›=â•›40â•›−â•›4x can be rewritten as
5xâ•›=â•›40. Furthermore, he made it clear that collecting terms
together was a fundamental and general principle, as his
principles of ‘Reduction’ and ‘Restoration’ effectively tell
us that we can do whatever we like to one side on an
equals sign so long as we do the exact same thing to the
other side as well. This basic technique can be used in all
manner of situations, as, for example, we can divide both
sides by 5 in order to reduce the equation 5xâ•›=â•›40 to the
simpler form xâ•›=â•›8.
The stated purpose of al-Khwarizmi’s celebrated work
was to show ‘what is easiest and most useful in Arithmetic,
such as men constantly require in cases of inheritance,
legacies, partition, lawsuits, and trade, and in all their deal-
ings with one another, or where the measuring of lands,
the digging of canals, geometrical computations, and other
objects of various sorts and kinds are concerned’.
Al-Khwarizmi clearly understood that the principles he
was explaining were exceedingly general, holding true
across all of the mathe�matical subject matters considered
by the ancients. Indeed, al-Khwarizmi rather charmingly
 THE RISE OF ALGEBRA107

remarked on this unity of mathe�matics with the simple

comment, ‘When I consider what people generally want
in calculating, I found that it is always a number.’
The notion that techniques for handling equations might
constitute an independent branch of mathe�matics was an
extremely important development, which the Arabs were
uniquely placed to make. As John Stillwell wrote in his
classic Mathematics and its History:

‘In Indian matheÂ�matics, algebra was inseparable from

number theory and elementary arithmetic. In Greek
mathe�matics, algebra was hidden by geometry. Other
possible sources of algebra, Babylonia and China, were
lost or cut off from the West until it was too late for
them to be influential. Arabic mathe�matics developed
at the right time and place to absorb both the geom-
etry of the West and the algebra of the East and to
recognize algebra as a separate field with its own
methods. The concept of algebra that emerged – the
theory of polynomial equations – proved its worth
by holding firm for 1000 years. Only in the nineteenth
century did algebra grow beyond the bounds of the
theory of equations, and this was at a time when most
fields of mathe�matics were outgrowing their estab-
lished habitats.’

Algebra and Medieval Europe

In Europe, the basic syllabus of early medieval science
could be found in a fairly small number of famous books,
each of which was written in Latin. To surpass their peers
in mathe�matics and science, scholars turned to Ancient
Greek texts, and the science of the Arab world. European
scholars were slow in learning from the Arabs, but it seems
clear that their influence was ultimately pivotal. A particu-
larly early example of European Christians learning Arabic
science can be found in the truly remarkable figure of
108 MATHEMATICAL THOUGHT

Gerbert of Aurillac (c.â•›946–1003). Gerbert was born in

France, but in 967 he moved close to Barcelona, studying
under the direction of Atto, Bishop of Vic. There was
conflict in Spain between the Christians and the Muslims
of al-Andalus, and in the face of a Christian defeat, Atto
was charged with delivering a request for a ceasefire.
Atto was received as an honoured guest, and soon found
himself mesmerized by the palaces in Cordoba. His brilliant
student Gerbert shared his fascination with the Arabs, whom
he greatly admired for their knowledge of mathe�matics,
astronomy and science. A prolific scholar and a gifted teacher,
Gerbert is credited with reintroducing Europeans to the
abacus and the armillary sphere: a kind of visual aid for
teaching mathe�matics and astronomy that had been lost to
Europe since the end of the Greco-Roman era. Remarkably,
Gerbert’s abacus incorporated Arabic numerals (many centu-
ries before modern, decimal notation was used elsewhere
in Europe), and it was rumoured that when he was young,
Gerbert would sneak out from the monastery at night to
study under the guidance of the Arabs.
Gerbert was appointed as a teacher to the Holy Roman
Emperor Otto II, and in 999 he reached the very head of
the Western Church, becoming the first French pope (Pope
Sylvester II). On 31 December he celebrated a solemn
mass as the people of Rome trembled before the impending
apocalypse, but contrary to expectations the sun rose on
the year 1000, and Gerbert remained as pope until his
death on 12 May 1003. In those turbulent times knowledge
of advanced science was rare, and for several centuries
very few Europeans were familiar with decimal numbers.
The spread of knowledge was much slower in those times,
but by the twelfth century Arabic texts were being trans-
lated into Latin and distributed across Europe.
Perhaps surprisingly, there was a gap of several centuries
between the introduction of decimal numbers and their
widespread use by merchants and administrators. The first
 THE RISE OF ALGEBRA109

widely read book on decimal numbers was Leonardo

Fibonacci’s Liber Abaci, written in 1202, but the popular
response to this innovation was largely hostile. Indeed, in
1299 the city of Florence went so far as to make the use
of decimal numbers a criminal offence! Generations of the
general public suspiciously dismissed them as a form of
financial trickery, and it was not until the end of the four-
teenth century that ever increasing numbers of people
abandoned Roman numerals for the superior decimal
system. For example, the largest and most powerful bank
in Europe, the Medici Bank, did not switch their account
books to the decimal system until 1439.
Decimal numeration is a very powerful system, and the
introduction of algebra also had a massive impact on the
development of European math and science. We have seen
that even in Ancient Babylon, people could answer ques-
tions such as: ‘A rectangle has an area of 77m2, and one
side is 4m longer than the other. How long are the two
sides?’ In modern notation, we would write x(xâ•›+â•›4)â•›=â•›77,
or x2â•›+â•›4xâ•›−â•›77â•›=â•›0. My point is that the form of writing
known as an equation is a relatively modern innovation,
though it can be used to rewrite ancient problems, and
this powerful innovation has Arabic roots. Indeed, in the
modern world, our uses and misuses of equations have
become central to our attempts to understand the world.
A critical point is that we can use the very same equa-
tions to summarize all manner of analogous things, not
just the geometric facts that the first ‘equations’ referred
to. As everybody knows, modern science is filled with
equations. Furthermore, Jacob Klein and other scholars
have convincingly argued that using equations to study
the integers has led to a subtle shift and abstract expansion
of our conception of number. For example, we know that
every integer or counting number is either even or odd.
If the integer N is even, it can be written in the form
Nâ•›=â•›2x, (where x is simply another integer). Similarly, if N
110 MATHEMATICAL THOUGHT

is odd it can be written in the form Nâ•›=â•›2xâ•›+â•›1. More gener-

ally, given any positive integer M, any integer N must be
expressible in one (and only one) of the following forms:

Nâ•›=â•›Mx (where x is some integer), or Nâ•›=â•›Mx+1,

or Nâ•›=â•›Mx+2, or Nâ•›=â•›Mx+3, … , or Nâ•›=â•›Mxâ•›+â•›M-1.

In other words, every integer N is either a multiple of M,

or it is one bigger than a multiple of M (so dividing N by
M leaves a remainder of one), or it is two bigger than a
multiple of M (so dividing N by M leaves a remainder of
two), and so on up to the largest possible remainder, M-1.
This kind of observation was explored by the eminent
lawyer and mathematician François Viète (1540–1603).
Viète wanted to understand the different kinds of pattern
that are produced when we divide various integers and
examine the remainders. He carried out this investigation
in a systematic fashion, introducing an algebraic notation
that is much like the one used today. More specifically, if
N divided by M leaves a remainder of n, we say that N is
equivalent to n modulo M, and write Nâ•›≡ n (mod M). For
example, 3â•›≡â•›1 (mod 2), because three divided by two leaves
a remainder of one. Similarly, 10â•›≡â•›1 (mod 3), because ten
divided by three leaves a remainder of one.
Now suppose that we have three integers A, B and M,
such that A divided by M gives a remainder a, while B
divided by M gives a remainder b. A profound question
to ask is whether it is always true that Aâ•›+â•›Bâ•›≡â•›aâ•›+â•›b (mod
M), and Aâ•›×â•›Bâ•›≡â•›aâ•›×â•›b (mod M). As an example, suppose
that Mâ•›=â•›10, Aâ•›=â•›53 and Bâ•›=â•›12. Dividing by 10 and taking
the remainder is equivalent to looking at the final digit of
the numbers A and B. If you want to know the final digit
of 53â•›×â•›12, you simply calculate 3â•›×â•›2. The larger digits only
affect the first digits of the answer – they cannot affect
the final digit.
 THE RISE OF ALGEBRA111

This observation suggests that if we divide by M and

take the remainder, then do our sums with a and b, we
must get exactly the same answer as when we do our sums
with the numbers A and B and then divide by M and take
the remainder. Indeed, this very general fact is easy to
prove using algebra. We simply write A in the form aâ•›+â•›xM
and B in the form bâ•›+â•›yM, where x and y are integers, and
a and b are non-negative integers smaller than M. We now
note that:

(a + xM) + (b + yM) = a + b + (x + y)M , and

(a + xM) (b + yM) = a b + (bx + ay + xyM)M .

This proves that if A ≡ a (mod M) and B ≡ b (mod M), then

Aâ•›+â•›Bâ•›≡â•›aâ•›+â•›b (mod M) and Aâ•›×â•›Bâ•›≡â•›aâ•›×â•›b (mod M). Hence
we can prove statements like ‘odd times odd is always
odd’ without needing to consider specific integer cases. We
simply note that an integer is odd if and only if it can be
written in the form 1â•›+â•›2x, and that:

(1+ 2x) (1+ 2y) = 1+ 2(x + 2xy + y).

In other words, we have proved that 1â•›×â•›1â•›≡â•›1 (mod 2).

Similarly, it is very easy to show that 1â•›+â•›1â•›≡â•›0 (mod 2).
The crucial point is that following the precedent set by
François Viète, one can conceive of settling facts such as
‘odd times odd is odd’ by exploiting a purely algebraic
system. By doing this we find ourselves systematically
dealing in ‘species’ of number (such as even or odd),
without needing to refer to the counted collections that
were the ultimate point of reference for ancient mathe�
maticians. Such equations correctly summarize the
individual facts where x and y are some definite, counted
out pair of integers, but they also state a more general,
abstract truth.
112 MATHEMATICAL THOUGHT

We can read the equation (1â•›+â•›2x)â•›×â•›(â•›+â•›2y)â•›=â•›1â•›+â•›2(xâ•›+â•›2xyâ•›

+â•›y) even when we interpret the symbols x and y as inher-
ently variable. In other words, we can read 2xâ•›+â•›1 as our
way of symbolically representing the general form called
‘odd’. What is more, the expression 2xâ•›+â•›1 is seen as having
an essentially numerical character precisely because it can
serve as an item within a computational system. That is
to say, we know how to add and multiply expressions such
as 2xâ•›+â•›1 and 2yâ•›+â•›1, and doing these kinds of sum does
not require us to ever substitute any specific integer values
for the variables x and y.

Fermat’s Little Theorem

By the seventeenth century European mathe�maticians were
taking full advantage of the basic techniques of algebra.
By working with symbols for unknown quantities and by
studying general forms of equation, mathe�maticians could
now prove new kinds of number-theoretic results. A
particularly beautiful example of this relatively modern
form of number theory is Fermat’s Little Theorem: one
of the many important results proved by Pierre de Fermat
(1601–1665). As Fermat stated in a letter from 1640, if we
take any prime number p and any integer n, we can be
certain that:

npâ•›≡â•›n (mod p).

In the next section we will see how this theorem underpins

the construction of the mathe�matical padlocks we use to
protect electronic business transactions. First, I want to
sketch a proof of remarkable theorem.
The fundamental observation behind Fermat’s Little
Theorem concerns the way that we multiply brackets filled
with additions. For example, a(bâ•›+â•›c)â•›=â•›abâ•›+â•›ac, and (aâ•›+â•›b)
(câ•›+â•›d)â•›=â•›acâ•›+â•›adâ•›+â•›bcâ•›+â•›bd. In general, when we multiply out
a sequence of bracketed terms, we take one thing from
 THE RISE OF ALGEBRA113

each bracket and multiply them together to form one of

the terms on the right-hand side. Once we have taken one
term from each bracket every different way we can, we
add up all the resulting terms to get our final answer.
In particular, consider how we would rewrite the
following expression without using brackets:
(x1 + x 2 + ... + x n ) 2 = (x1 + x 2 + ... + x n ) (x1 + x 2 + ... + x n ).

Taking the first term from each bracket and multiplying

them together gives us x12 . Similarly, taking the second
term from each bracket and multiplying them together
gives us x 22 , taking the third term from every bracket and
multiplying them together gives us x 32 , and so on. When,
for example, we take the first term from the first bracket
and the second term from the second bracket, we get x1 x 2╛╛.
If we take the second term from the first bracket and the
first term from the second bracket we also get x 2 x1 = x1 x 2 â•›.
Because we can pick the term x1 from either the first or
the second bracket, our final expression will contain the
term 2x1 x 2â•›. More generally, our final expression will have
the form

x12 + x 2 + ... + x n2 + 2 f (x1, x 2, ... , x n ).

where f (x1, x 2 , ... , x n ) is a polynomial expression involving

the variables x1 to x nâ•›. Now let’s consider how we would
rewrite the following expression without using brackets:

(x1 + x 2 + ... + x n ) p = (x1 + x 2 + ... + x n ) ... (x1 + x 2 + ... + x n ).

Taking the first term from every bracket and multiplying

them together gives us x1p, taking the second term from
every bracket and multiplying them together gives us x 2p╛╛ ,
and so on. When, for example, we take the first term from
the first bracket and the second term from every other
114 MATHEMATICAL THOUGHT

bracket, we get the expression x1 x 2p 1. Similarly, if we take

the first term from the second bracket, and the second
term from every other bracket, we also get x 2 x1 x 2p 2 = x1 x 2p 1â•›.
Because we can pick the term x1 from any of the p different
brackets and still get the same answer, our final expression
must contain the term px1 x 2p 1. More generally, we must
be able to write our final expression in the form
x1p + x 2p + ... + x np + pf (x1, x 2 , ... , x n ), where f (x1, x 2 , ... , x n ).
is a polynomial expression involving the variables x1 to x n â•›.

We have just argued that

(x1 + x 2 + ... + x n ) p = x1p + x 2p + ... + x np + pf (x1, x 2 , ... , x n ),

which nearly completes the proof of Fermat’s Little

Theorem. All we need to do is consider the case where
x1 = x 2 = ... = x n = 1. In this case we have

(1+ 1+ ... + 1) p = 1p + 1p + ... + 1p + pf (1, 1, ... , 1)

and so n p = n + pf (1, 1, ... ,1).

Since f is a polynomial equation with integer coefficients,

it follows that f (1, 1, ... ,1) must be an integer. Hence, we
have proved that for every integer n and every prime
number p, n p must equal n plus some multiple of p. In
other words, for every integer n and every prime number
p, n p is equivalent to n modulo p. If we divide both sides
by n we can see that n p 1 1 (mod p), and as we shall see
in the following section, this fact lies at the heart of
Internet security.

How to Make a Mathe�matical Padlock

For every composite number n, there is a very simple
proof of the fact that n is composite. All we require are
two numbers (a and b), such that a b = n. In contrast to
 THE RISE OF ALGEBRA115

this short and simple proof that the number n is composite,

a proof that a given number is prime is (necessarily?) large.
That is to say, every known method for factorizing numbers
involves a huge amount of brute calculation, and showing
that a number has no factors is similarly time-consuming.
Some astronomically large numbers have been shown
to be prime. In fact, the largest known primes are much
bigger than astronomical, as the largest known prime is
much, much bigger than our best estimates for the number
of atoms in the universe. Strangely enough, large prime
numbers have a financial value, because they are essential
for the construction of the mathe�matical padlocks that are
used to protect electronic business transactions.
These mathe�matical padlocks are rules that convert an
input into an output. Corporate entities publish these rules
(i.e. they freely distribute padlocks), and someone who
wants to encode the message ‘I will buy a million shares’
simply feeds their message into the rule, which converts
their message into some large number C. The person then
transmits the number C, and at the other end this number
is fed into a secret rule (the key to the padlock), which
converts the encoded message back into ordinary text.
Someone who eavesdrops on these transmissions can,
in principle, work out what any encoded message says
without being told the secret rule. For example, they could
try encoding every possible statement in turn, and compare
the results with the encoded message C (that is, the scram-
bled version of the message that they want to read). If one
of our hacker’s made-up messages happens to encode to
exactly the number C, that tells our hacker that their
made-up message is actually the same as the unscrambled
version of the message that they want to translate. Of
course, encrypting every possible message that a person
might have sent is not a practical way to eavesdrop, and
it is believed that successfully cracking these codes requires
an astronomical number of calculations (unless you know
116 MATHEMATICAL THOUGHT

the secret rule, of course). That is to say, finding the key

by examining the padlock is practically impossible.
One standard form of mathe�matical padlock is called
the RSA system, after the trio of mathe� maticians and
computer scientists who first published the algorithm.1
This kind of mathe�matical padlock takes a fixed number
of binary digits as its input. Any sequence of binary digits
can be interpreted as an integer M. Given such an integer
M, the output of the system is

C M e (mod n),

where e and n are fixed integers, specified by the people

who are handing out the padlock. Most real-life RSA
crypto�systems use values for e and n with over a thousand
digits. Because we are doing these calculations modulo n,
it is actually quite simple for a modern-day computer to
calculate the correct output. However, doing this procedure
‘backwards’ is riddled with difficulties and very, very time-
consuming. It is somewhat analogous to trying to piece
together something that has gone through a shredder – even
the person who put the message in can’t undo it. By way
of contrast, the secret rule is very simple: we just take the
encoded message C, and calculate C d (mod n).
The whole point of this system is that C d = (M e ) d = M ed
(mod n). To make this system work, all we require are
three numbers, e, d and n, such that for every message M,
we have M M ed (mod n). That way, if we run a message
through the padlock and then through the key, we get
back what we started with.

1 The RSA system is named after Ron Rivest, Adi Shamir and Leonard Adleman,
who devised the algorithm in 1977. In fact, an equivalent scheme was described
in 1973 by the British mathematician and cryptographer Clifford Cocks, but
since Cocks was employed by the Government Communications Headquarters
at the time, his work was classified top-secret, and did not enter the public
domain until 1998.
 THE RISE OF ALGEBRA117

Given two prime numbers p and q, we can find the

required numbers by putting n = p q. The integers e and
d must be smaller than (p − 1) times (q − 1), and it is also
essential that ed 1 (mod (p − 1)(qâ•›−â•›1)). In other words,
there must be some integer t such that edâ•›=â•›1â•›+â•›t(p-1)(q-1).
Given any two prime numbers p and q, it is actually quite
easy to find a pair of integers e and d with the desired
property.
Now, given any prime number p, Fermat’s Little Theorem
tells us that:

Mp 1
1 (mod p).

Raising both sides of this equation to the power s gives

us the following relationship:

M s(p 1)
1 (mod p), for every integer s.

In particular, M t(q 1)( p 1) = M ed 1 1 (mod p). In other words,

M ed 1 1 is a multiple of p. A similar argument shows that
M ed 1 1 is also a multiple of q. Since p and q are both
prime numbers, this tells us that M ed 1 1 must be a multiple
of n = p q. In other words, M ed 1 1 (mod n). By multi-
plying both sides of this equation by M, we get the result
we were looking for, namely that:

M ed M (mod n).

To recap, raising the message M to the power e gives us

the coded message C M e (mod n). Raising the coded
message C to the power d gives us M ed M (mod n), which
means that our secret rule correctly decodes messages. It
is theoretically possible to calculate the secret code number
d by using the public padlock numbers n and e. The only
difficult bit is factorizing the integer n, because everything
else can be calculated relatively quickly.
118 MATHEMATICAL THOUGHT

This means that the security of the world’s financial

markets rests on the assumption that factorizing vast
numbers is very difficult (or to be more accurate, that
cracking RSA is very difficult). It is reasonable to assume
that this is so, but it really is an assumption. Although
mathe�maticians can prove that huge classes of problems
are equally difficult (in a particular well-defined sense),
there are serious difficulties in proving that some problems
are harder than others! We can see that a particular method
for solving one problem is more time-consuming than some
particular method for solving a second problem, but that
is not the same thing at all. There is literally a million-
dollar reward for the person who can prove an inherent
distinction of difficulty, where we show that any possible
method for solving one class of problem necessarily takes
longer than some given method for solving a second class
of problem.
Time and again, great progress has been made when one
branch of mathe�matics has been brought to bear on another.
For example, Fermat’s Little Theorem and modern matheÂ�
matical padlocks are both examples of algebraic number
theory. As we shall see in the following chapter, the basic
techniques of algebra were also pivotal in enabling the
development of algebraic geometry. Without this new
approach to the ancient discipline of geometry, modern
science could not have developed, and we would not have
scientific equations as we know them today.
 Chapter 5:
MECHANICS AND THE
CALCULUS
‘The heart of matheÂ�matics consists of concrete exam-
ples and concrete problems. Big general theories are
usually afterthoughts based on small but profound
insights; the insights themselves come from concrete
special cases.’
Paul Halmos, 1916–2006

The Origins of Analysis

In this chapter I will examine the origins of calculus, and
the co-evolution of two very different branches of mathe�
matical science: analysis and mechanics. Roughly speaking,
mathe�matical analysis is the study of infinite, mathe�matical
sequences, while mechanics is the study of moving objects.
As we shall see, the science of mechanics played a critical
role in the development of analysis, particularly the infini-
tesimal calculus. On the other hand, many ideas relating
to infinite sequences are extremely old, predating our ability
to mathe�matically analyze motion. Indeed, we have already
encountered the oldest analytic concepts, such as the intui-
tive idea of generating an infinite sequence, and finding
the limit case.
For example, one of Archimedes’ most elegant proofs
120 MATHEMATICAL THOUGHT

concerns the area under a parabola. Imagine, for example,

that we have fired a projectile up into the air from a hori-
zontal firing range. The projectile will follow a parabolic
path, and we can draw a triangle between the firing point,
the highest point in the arc, and the point where the projec-
tile lands. It was obvious to Archimedes that if this triangle
has an area A, the parabola must have an area greater than
A (precisely because the triangle fits inside the parabola).
As a small part of the parabola was left uncovered, he
added two more triangles, and he did this in a clever way
so he could prove that those two triangles must have a
total area of A . After adding those triangles an even smaller
4
part of the parabola remained uncovered, so he added a
further four triangles with a total area of 4A2â•›. Furthermore,
he made it clear that you can continue adding smaller and
smaller triangles indefinitely, where at each stage the total
area of the additional triangles was one quarter of the
previous stage.

Archimedes could prove that the triangle marked A has

four times the area of the two triangles directly above A.
He could also prove that those two triangles have four
times the area of the four triangles directly above them,
and so on.
Up until the nineteenth century, the area under a curve
was taken to be a given quantity. In other words, Archimedes
simply assumed that a truncated parabola defines a definite
quantity, namely the area underneath it. Not every curve
has a well-defined area underneath it, but up until the time
 MECHANICS AND THE CALCULUS121

of Carl Friedrich Gauss (1777–1855), matheÂ�maticians only

studied curves that had well-defined areas underneath them.
Because it was intuitively obvious that the shapes in ques-
tion had a definite area, mathe�maticians were justifiably
confident when using an infinite sequence to calculate that
area. My point is that we don’t need to worry about whether
or not A(1â•›+â•› 14 â•›+â•›4â•›12╛╛+â•›4â•›13╛╛+â•›…) defines one specific quantity. Our
geometric intuition is enough to convince us of the fact that
the area under a particular parabola is equal to one particular
quantity, and this leads us to accept the fact that 1╛+╛14 ╛+╛4╛12╛╛+╛
â•›1â•›
43
â•›+â•›… picks out exactly one real number. Given that this
limit case is indeed a number, we can safely make the
following argument:

Since x = 1+ 1 4 + 1 2 + 1 3 + ...
4 4
it follows that 4 x = 4 + 1+ 1 4 + 1 +1 + ...
42 43
Subtracting the first equation from the second tells us that
3xâ•›=â•›4, which means that x = 4 3 .
1 1 1 4
Archimedes could see that 1+ 4 + 4 2 + 4 3 + ... = 3 ,
â•›
but he was understandably wary of any argument that
built on the idea of an infinite sequence. He therefore
established the validity of his answer by using Proof by
Exhaustion. In other words, he rigorously proved that the
area under a parabola cannot possibly be smaller than 4 A 3╛╛,
and it cannot be larger than 4 A 3╛, so it must equal 4 A 3╛╛ .
Although Archimedes’ proof does not depend on the prop-
erties of infinite sequences, the notion of slicing a shape
into thinner and thinner pieces is at least twenty-two centu-
ries old. We know that this is so because Archimedes
explained this ingenious idea in a work entitled The Method.
122 MATHEMATICAL THOUGHT

This remarkable text had been lost since the Middle

Ages, so the modern inventors of calculus certainly hadn’t
read it. Indeed, until very recently, everything we knew
about the mathe� matics of Archimedes was ultimately
derived from just two manuscripts (codices A and B).
However, in 1906 the Danish philologist J. L. Heiberg
discovered a third ancient source (codex C), which included
The Method. This particular collection of works of
Archimedes had been overlooked for so many generations
because it had been copied onto parchment in the tenth
century, but a couple of hundred years later the precious
material had been washed down, rebound, and inscribed
with prayers. The original text was almost invisible, but
in the Monastery of the Holy Sepulchre, Istanbul, it lay
waiting to be discovered. Fortunately for us, parts of the
original text could still be seen, and thanks to x-ray tech-
nology almost all of it has now been restored.
In the course of The Method Archimedes reveals that
many of his deepest and most original results were found
by using questionable infinitary arguments, and only later
did he go on to prove his results using the rigorous Proof
by Exhaustion. In other words, Archimedes’ constructed
arguments using infinite sequences, but as the logical foun-
dations of such arguments were unclear, he refused to rely
on them while stating his proofs. Nevertheless, The Method
tells us that ‘It is of course easier to supply the proof when
we have previously acquired some knowledge of the ques-
tions by the method, than it is to find it without any
previous knowledge.’ In other words, it is easier to write
a rigorous proof when you already know the right answer.
Since the 1970s scholars have also known about another
form of ancient analysis, which was hidden in the
astronomy of medieval India. In particular, the Kerala
region in the south-west of India produced one of the
great mathe�matical visionaries: Madhava of Sangamagrama
(c.â•›1350–1425). All the matheÂ�matical writings of Madhava
 MECHANICS AND THE CALCULUS123

have been lost, though some of his texts on astronomy

have survived. We know about his brilliant mathe�matical
work through reports from later centuries, most of which
were written in the regional language of Malayalam.
Trigonometric problems routinely arise while plotting the
course of the stars, and Madhava discovered and extended
rules for solving these kinds of problem. Crucially, his
technique involved infinite sequences. For example, he
knew that:
1 +1 1 + ...
4 =1 3 5 7

Madhava needed numerical answers to put in the ledgers

of astronomy, and, most impressively, when he used an
infinite sequence to generate an approximate answer, he
commented on the size of his margin of error. For example,
in the case of finding4â•›, he knew that using n terms gives
an approximation that is within a margin of error of 1 2n╛╛ .
In a remarkable passage describing the construction of this
kind of sequence, the mathematician-astronomer Jyesthadeva
(c.â•›1500–1575) states that care must be taken ‘otherwise the
correcting term [or margin of error] will not tend to the
vanishing magnitude’. This line indicates that Madhava’s
work prefigured the modern definition of a limit case by
four or five hundred years.
The crucial point is that some sequences do not have a
limit case, and we cannot always rely on geometric intui-
tion to reassure us that a limit must exist. For example,
the sequence 1, 1â•›−â•›1, 1â•›−â•›1â•›+â•›1, 1 – 1â•›+â•›1 – 1, … is funda-
mentally different to 1, 1 1 , 1 1 + 1 , … The first
3 3 5
sequence hops between 1 and 0, and never settles on a
limit. The second sequence narrows in on one particular
number, and the margin of error becomes arbitrarily small.
124 MATHEMATICAL THOUGHT

For this reason we say that 1 1 + 1 1 + ... is a conver-

3 5 7
gent series, which defines a real number. In contrast, the
string of symbols 1 1+ 1 1+ ... is said to be divergent, and
this second sequence does not define a real number.
Modern mathe� maticians say that by definition, a
sequence x1, x 2 , x 3 , ... converges to a limit L if and only if:

For every positive number , there is some number n

such that every term x n , x n +1, ... is bigger than Lâ•›−â•› and
smaller than Lâ•›+â•› .

In other words, a point L is the limit of a sequence if

and only if any ‘target region’ containing L contains all
but a finite number of the points in our sequence. As we
shall see in a later chapter, it is crucial that our definition
uses the logical words ‘and’, ‘every’ and ‘some’, because
this kind of vocabulary can support various forms of
logical deduction. Furthermore, we should note that
although certain analytic ideas have ancient origins, this
logical, axiomatic foundation for modern analysis did not
arise until the early nineteenth century. Before that time
the scope of calculus was not fully apparent, and in the
hands of less expert men, many more mistakes could have
been made.

Measuring the World

Most branches of mathe�matics have been influenced by
empirical science, but the mathe�matical ideas presented in
this chapter have a particularly close relationship with the
study of moving objects. It is a fundamental fact that
the study of moving objects requires the measurement of
time. For example, countless generations have tried to
calculate the time it takes for the planets to complete their
orbits across the heavens, and this kind of inquiry was
critical to the development of modern, mathe� matical
 MECHANICS AND THE CALCULUS125

science. In order to convey something of the intellectual

traditions that preceded Newtonian mechanics, let’s
consider the work of two masters of measurement: al-Biruni
and Galileo Galilei.
The Persian scholar Abu Rayhan Muhammad ibn Ahmad
al-Biruni (973–1048) was an exceptionally advanced scholar,
and we can use him to gain an impression of the best of
medieval science. A great polymath, al-Biruni was right at
the cutting edge of mathe�matics, science and the humani-
ties. His most famous book was Tarikh Al-Hind (or
‘History of India’), a remarkable account of the ritual,
mathe�matical and astronomical knowledge of Hindu India.
He was also an expert on Greek mathe�matics and science,
but unlike many of his contemporaries, he was prepared
to challenge Aristotle. For example, it was widely believed
that cooling an object makes it shrink, but al-Biruni chal-
lenged this assumption, noting that a glass full of water
will crack when the water freezes. An expert writer of
lists, he was also the first person to define explicitly the
concept of relative density, compiling tables of the empir-
ically determined densities of various gemstones and metals.
In Europe it took longer for mathe�matical physics to
develop, and until the time of Galileo Galilei (1564–1642),
Ancient Greek texts remained the pinnacle of science. The
route to further progress required a combination of theo-
retical and empirical work, but the defining spirit of the
scientific revolution could be summarized by Galileo’s
powerful slogan: ‘Measure what is measurable, and make
measurable what is not.’ Galileo was appointed to the chair
of mathe�matics in Pisa, and over the course of a long and
varied career, his careful measurements of moving bodies
led to a deeply mathe�matical understanding of motion. He
gave an accurate, mathe�matical account of the swinging of
a pendulum, and he was the first person to argue that
projectiles should follow a parabolic path. He was also
one of the first people to explore the night sky with a
126 MATHEMATICAL THOUGHT

telescope, identifying the moons of Jupiter, and observing

that the haze of the Milky Way is comprised of countless
stars.
Because of his many achievements, and the fact that
he explicitly elevated mathe�matical reasoning and empir-
ical data as the ultimate guides to nature, Galileo is
sometimes called the father of modern science. One of
his most impressive insights was a claim that was essen-
tial to the development of Newtonian mechanics. That
is to say, Galileo famously stated that under the force of
gravity alone, all objects accelerate at the same rate. In
other words, it is only air resistance that stops stones
and feathers from falling side by side. Galileo argued that
this statement must be true, not because of the results of
some physical experiment (though he was certainly a keen
experimentalist), but because of a mathe�matical thought
experiment.
This conceptual experiment began with the assumption
that heavy bodies fall faster than lighter ones, as Aristotle
claimed. If this were a fundamental truth (as most people
assumed), Galileo wondered what would happen if you
dropped a heavy cannon ball attached to a lighter musket
ball. According to Aristotle’s theory, the musket ball would
trail behind the cannon ball because of its lighter weight.
This would mean that if the two balls were attached to
one another, the slower moving, lighter ball would slow
down the faster moving, heavier ball, so the combination
of two balls would fall more slowly than the cannon ball
on its own.
On the other hand, the combined weight of a musket
ball and a cannon ball is greater than the weight of a
cannon ball on its own. Hence Aristotle’s theory also
implies that the combination of two balls should fall faster
than the cannon ball on its own. This analysis is devastating
for Aristotle’s account of gravity. It cannot possibly be
the case that a heavy ball falls both faster and slower than
 MECHANICS AND THE CALCULUS127

the even heavier system of a heavy ball attached to a light

ball. Now, given that Aristotle’s theory of falling bodies
has been shown to be unworkable, what should we put
in its place? The same line of reasoning would destroy
the absurd notion that lighter objects fall faster than heavier
ones, leaving only one readily conceivable option: the
effect of earth’s gravity is to make all objects fall at the
same rate.

The Age of Clocks

Ancient accounts of motion or change were largely focused
on classifying the form or ‘cosmic purpose’ of the various
kinds of change that people had observed. Properly quan-
tifying how quickly objects fall (for example) was a kind
of science that did not develop until the Renaissance, with
Galileo being a key, transitional figure. It is worth stressing
that mechanics is essentially concerned with the movement
of objects over time, and towards the end of the medieval
period, the common understanding of time itself underwent
a fundamental shift: a shift that enabled this new kind of
science. All people are aware that certain activities or events
have a characteristic duration, and life on this planet has
always followed the beat of an ancient rhythm. Day after
day the sun rises and sets, year after year the seasons are
repeated, and it is surely obvious to everyone that a day
is not the same as a year.
Ancient people could refer to periods of time (e.g. the
time taken to boil a certain quantity of water), but it was
the activities themselves that had a duration, and time
was not typically conceived as an abstract concept, sepa-
rate for the events whose duration we might care to
measure. It is a characteristically modern sensibility that
thinks of time as a fundamentally abstract sequence of
units (days, hours, seconds, etc.). After all, at its root a
clock is a mechanical representation of our planetary
system: the clouds may cover the sun, but the hands of
128 MATHEMATICAL THOUGHT

a clock will carry on ticking regardless, showing us the

moment when the sun should be overhead. Only a modern
person would ever think of saying that 24 hours is 24
hours, regardless of the rate at which the earth happens
to revolve!
Clocks were first used in the monasteries of Europe,
where the daily routine was highly regimented, and the
desire to stick to a strict schedule of prayers was reason
enough to keep careful track of time. Eventually the use
of clocks spread from monasteries, to clock towers, and
out into the cities beyond. This was a mixed blessing, as
accurate time keeping enabled the rationing of time, and
it permanently changed the way human affairs are organ-
ized. After all, clocks not only measure time; they are also
used to synchronize the actions of men, telling us when
to work, when to eat, and when the play begins.
In a world of clocks it seems self-evident that the passage
of time is an empirical fact, and we can even come to
believe that the facts that a clock can verify are somehow
more real than our direct, subjective experience of the
passage of time. In short, the psychological significance
of the clock is that it separates time from human activity,
or the events we can observe in Nature. As the writer
and historian of technology, Lewis Mumford, remarked
in Technics and Civilization, the spread of clocks ‘disso-
ciated time from human events and helped to create the
belief in an independent world of mathematically measur-
able sequences: the special world of science.’ This ‘special
world of science’ is one in which individual objects of
study can be measured and probed in isolation, and the
relevant features for understanding are taken to be those
that can be agreed upon by any suitably qualified panel
of experts.
Objective measurement is central to the scientific enter-
prise, and in the words of the computer scientist Joseph
Weizenbaum:
 MECHANICS AND THE CALCULUS129

This rejection of direct experience was to become one

of the principal characteristics of modern science. It
was imprinted on western European culture not only
by the clock but also by the many prosthetic sensing
instruments, especially those that reported on the
phenomena they were set to measure by means of
pointers whose positions were ultimately translated
into numbers [e.g. barometers, thermometers or
scales]. Gradually at first, then ever more rapidly and,
it is fair to say, ever more compulsively, experiences
of reality had to be represented as numbers in order
to appear legitimate in the eyes of the common
wisdom.

It was in this unquestionably measurable world that great

thinkers such as Descartes, Newton and Leibniz made
their mark. As we shall see in the following section, the
youngest of these men, René Descartes, developed a philos-
ophy and an approach to mathe� matics that were both
radical and influential. This was a critical step in developing
a science that went beyond that of the ancients, though as
the preceding quotations should make clear, this move to
modernity was not without a price.

Cartesian Coordinates
For the last four hundred years or so the study of geometry
has become increasingly tied to algebraic methods, and it
is this development that enabled Isaac Newton to describe
the paths of moving objects in terms of an equation. This
cross-fertilization between geometry and algebra was a
massively important event, and not only because it gave
rise to the use of equations in our descriptions of motion.
In the words of the great mathematician Joseph Lagrange
(1736–1813), ‘As long as algebra and geometry travelled
separate paths, their advance was slow and their applica-
tions limited. But when these two sciences joined company,
130 MATHEMATICAL THOUGHT

they drew from each other fresh vitality and then forward
marched at a rapid pace.’ There are many matheÂ�matical,
historical and philosophical factors behind this momentous
shift, but François Viète (1540–1603) and René Descartes
(1596–1650) are clearly pivotal figures.
As a man of undoubtable talent, Descartes was unusu-
ally hostile towards the other intellectual giants of his age.
He was happy to work in relative isolation, and it seems
that his preferred audience was one that would understand
him without daring to challenge his sometimes erroneous
views. For the most part Descartes’ lifestyle was simple
but leisurely, and up until the year of his death he refused
to get out of bed before eleven o’clock. Unfortunately he
agreed to teach Queen Christina of Sweden who wanted
her philosophy lessons to begin at 5 a.m., and after a few
weeks of early mornings in the depths of the Swedish
winter, Descartes contracted a fatal case of pneumonia.
Like many great thinkers of his age, Descartes believed
that we should rely on a ‘general method of reason to find
the truth in the sciences’. I would be tempted to dismiss
this goal as absurd and unattainable, were it not for the
fact that his attempts to summarize the general principles
of valid reasoning revolutionized mathe�matical thought.
One of Descartes’ most influential innovations (which
Pierre de Fermat independently conceived) was the idea
of using coordinates, as when we draw a graph with an
x- and a y-axis.
The story goes that as Descartes was lying in bed, he
realized that the apparent location of the fly buzzing above
his head could be described using two numbers: its distance
from the wall behind his head, and its distance from the
wall that stood to his left. This was a simple observation,
but Descartes’ genius was to recognize important ways in
which this idea could be employed. Coordinate-based
graphs are enormously convenient, as can be appreciated
from their ubiquity in the modern world. Coordinates are
 MECHANICS AND THE CALCULUS131

also crucial for another, fundamental reason: they enable

the practice of drawing a shape, while describing that shape
through the use of a function (that is, a rule that takes a
number as an input and gives a number as an output). In
other words, coordinate systems make it possible to think
of curves as functions, and to think of functions as curves.
For example, imagine drawing a graph of the function
y = x 2. Once we have this graph, we can gain insights about
the shape (a parabola) by studying and manipulating the
equation. The Ancient Greeks had achieved the converse
of this feat: they gained insights into equations by making
geometric arguments. However, Ancient Greek ‘equations’
were not the concise, formal statements of modern mathe�
matics. Instead, they wrote out relationships like yâ•›=â•›x2
using complete sentences. What is more, they considered
such relationships to be properties of the underlying curves,
and they did not consider the corresponding equations as
mathe�matical objects in their own right. Consequently,
they did not develop the basic algebraic techniques that
Descartes had inherited from the Arabs.
For the Greeks, a curve was something that was traced
by a moving point, and particular curves were produced
by using a ruler and compass, by slicing a cone, or through
some other idealization of a physical act. Following his
idea of adopting a coordinate system, Descartes took the
novel step of defining curves through his use of equations.
Crucially, once we have defined a pair of curves yâ•›=â•›f(x)
and yâ•›=â•›g(x), it is rather obvious that we can also consider
the curve yâ•›=â•›fâ•›+â•›g, where the value of the function fâ•›+â•›g at
any point x is simply f(x)â•›+â•›g(x). Likewise, we can also
consider the curves, yâ•›=â•›fâ•›−â•›g, yâ•›=â•›fâ•›×â•›g and yâ•›=â•›fâ•›÷â•›g, though
the last of these curves will be undefined at any point
where g(x)â•›=â•›0. Note that the addition, subtraction, multi-
plication and division of functions or curves are entirely
new concepts. These important ideas are a natural exten-
sion of arithmetic using numbers, but the Ancient Greeks
132 MATHEMATICAL THOUGHT

did not use symbols in this way, and they did not imagine
that you could add or multiply curves.
Mathematics is now dominated by our use of discrete,
algebraic symbols, but until the modern period, European
mathe�matics was dominated by geometric reasoning.
Descartes was an influential figure in bringing about this
fundamental change, because of his philosophy as well as
his contributions to mathe�matical technique. Because he
was sceptical of the evidence of our senses, he thought
that the real essence of mathe�matical statements lay in their
‘rational’ character. So, for example, if we were to discover
that we are just brains floating in a jar, almost everything
we thought we knew about the world would be wrong.
Nevertheless, when we imagine this bizarre scenario we
can also imagine doing math in the same old way, and
Descartes took this thought experiment as proof that the
rule-governed use of symbols is our most certain form of
knowledge.
Before Descartes’ time, questions concerning linear equa-
tions were understood as being substantial precisely because
those equations could be understood in terms of physical
lengths. Similarly, quadratic terms could be understood in
terms of areas, while cubic terms referred to volumes. We
can still make those connections, but generally speaking,
we don’t. For example, we are happy to employ algebraic
techniques to solve the equation ‘ x 2 + x = 110 ’. Very few
modern people would worry that x 2 represents an area,
while x represents a length, and it doesn’t make much
sense to add an area to a length.
In contrast the Greeks, like the Egyptians and Babylonians
before them, oriented their mathe�matics around geometric
interpretations. For them, the equation ‘ a 2 + b 2 = c 2 ’ was
invariably interpreted as a true description of the relation-
ship between the areas of three different squares. Indeed,
they didn’t even write the equation that we know so well:
they simply used words and diagrams to state the relation-
 MECHANICS AND THE CALCULUS133

ship between the relevant areas. Students today often think

of Pythagoras’ Theorem as primarily a recipe for generating
a number given two other numbers. Of course, they know
that the numbers in question refer to the lengths of a triangle,
but in looking at the equation, they may not think of areas
at all.
In short, Descartes made us recognize the fact that we
can take a collection of lengths (or other measured quan-
tities) as purely symbolic statements, divorced from any
worldly sense of measurement. In other words, we can
systematically name a length ‘2’ instead of ‘2  cm’ (say).
The validity of this step marks the dividing line between
mathe�matics and physics, because as far as abstract mathe�
matics is concerned, the meaning of our units really isn’t
relevant. As a mathematician, Descartes was happy to
consider adding x to x 2 or any other power, as we can
remain oblivious to any intuitions that might motivate our
use of any numerical symbol. For example, the expression
x 2 does not have to be read as the area of a square of side
x. It could just as easily apply to a line of length x 2 , and
as a person performing a given sum, we do not need to
care which of these is ‘really’ the case.
This emphasis on the use of formal symbols is charac-
teristically modern. Indeed, I would argue that for modern
mathe�maticians it is the symbolic representability of concepts
that seems to mark them out as grounded or ‘properly
defined’, being parts of a symbolic system that others can
employ. For example, the ancients had considerable concep-
tual difficulty in reconciling something like √2â•›=â•›1.414… with
their notion of number. I believe that there has been a subtle
but significant shift in our sense of number, as most people
today find it natural to think of a number as any sequence
of digits whatsoever. In other words, symbolic forms such
as ‘1.414…’ are part of our system of numerals, and learning
to use this system of numerals is at the heart of our initia-
tion into the number concept.
134 MATHEMATICAL THOUGHT

At this point I want to clarify the history of coordinates

and number lines, as I don’t want you to get the impres-
sion that they were all invented in a single stroke. The use
of coordinates, and the concept of longitude and latitude,
had been used by astronomers and cartographers since
Hipparchus (c.â•›150 bc). In fact, the idea of using coordi-
nates while studying geometric curves dates back to Nicole
Oresme (1323–1382), but Oresme did not develop this
idea in quite the same radical fashion. What was new in
Descartes’ time was the use of axes together with equa-
tions, expressed in a concise, symbolic form. Descartes’
understanding of the word ‘line’ was essentially classical,
taking it to refer to the paths traced by moving points.
His flash of insight was to realize that we can identify
points in geometric space by their distance from the axes.
The Greeks would have had difficulty developing this idea,
as they did not have a concept of zero. They thought of
numerical quantity in terms of counting and ratios between
lengths, so they did not think of a line with a point corre-
sponding to zero. It was Newton who added the
innovation of having positive numbers on one side and
negative numbers on the other (Descartes’ graphs just had
positive numbers), and as we shall see the modern concept
of the number line was not fully formed until the end of
the nineteenth century.

Linear Order and the Number Line

Number lines are a rather subtle concept, as it is far from
obvious what numbers have to do with geometric lines.
The most basic facts relate to a deep rooted, physically
grounded understanding that applies whenever we have
three objects, A, B and C, lined up in a row. We know
that if object A is to the left of B, and B is to the left of
C, it must also be the case that A is to the left of C.
Similar statements also apply when we replace the words
‘to the left of’ with the words ‘bigger than’, ‘higher than’,
 MECHANICS AND THE CALCULUS135

‘hotter than’ or ‘heavier than’. We approach the world

with the knowledge that we can put things in order of
size, height, temperature or weight, and we understand
that every such order is metaphorically akin to the spatial
form of linear order.
This axiomatic truth concerning linear order is not some
arbitrary convention like the rules of chess: the world
seems to work that way, and it is very hard for us to
imagine it being otherwise! Indeed, to use the number
concept properly is to accept (among other things) that if
x is smaller than y and y is smaller than z, then x must be
smaller than z, just like the case of physical objects arranged
in a line. That is to say, real numbers and points on a line
both satisfy the following axioms:

1. Given a pair of numbers x and y, either xâ•›<â•›y, or

xâ•›>â•›y, or xâ•›=â•›y. Likewise, if we are given a couple
of points on a line, either the first point is before
the second, or the first point is after the second,
or both points are in fact the same point.
2. If xâ•›<â•›y and yâ•›<â•›z, then xâ•›<â•›z. Likewise, if point x is
to the left of y and y is to the left of z, then x is to
the left of z.

In short, the relationships between numbers are struc-

tured in the same way as the relationships between points
on a line. On the other hand, numbers and the points
on a line are fundamentally different because geometric
lines are continuous, with no gaps between their parts,
while numbers are essentially discrete. Although the
number line is ‘continuous’ in the sense that it is not
missing any real numbers (i.e. the number line does
not have a gap located at any real number), the number
line is not continuous in the geometric sense, because
the parts of a number line are points, and these parts do
not touch. Points on a number line cannot touch, because
136 MATHEMATICAL THOUGHT

if there is no distance between a pair of real numbers

(as there would be if they touch), they are in fact one
and the same number/point.
Crucially, once we have identified one point as ‘being’
the number 0 and another point as ‘being’ the number 1,
every point on a line determines a unique real number.
More specifically, there is some distance between the point
in question and the point 0, and there is some other distance
between the points 0 and 1. Given any point, the ratio
between those two lengths determines a quantity in the
classical sense, and we can think of that number as being
the distance between the point and the origin (that is, the
point that we identify with 0). Conversely, we can use the
concept of ‘distance from the origin’ to associate each real
number with some or other point on the line, with different
real numbers picking out different points in space. It was
Richard Dedekind who explicitly made the final conceptual
leap, when he asserted that we can define the number line
as being the set of real numbers, where the point identified
by the number r is r units from the origin.
This new definition says that fundamentally, a line is
an infinite set of points. That is subtly but significantly
different from the classical, geometric conception of a line,
which says that a line is the path traced by a moving point.
In the case of the classical, geometric line, the lines them-
selves are primary, and points on a given line are merely
infinitely precise locations on that line, not the thing that
the line is made of. Furthermore, in classical geometry the
lines themselves are understood to have certain intrinsic
properties. For example, a line that forms a closed loop
constitutes the boundary of a region in space, and we
understand that this is so because of our conception of
lines, rather than the points that are on that line.
By Dedekind’s definition the points are always the
primary entities. Lines are redefined as being infinite sets
of discrete points, and in this framework the properties
 MECHANICS AND THE CALCULUS137

of lines need to be redefined in terms of relationships

between those points. In fact, because points have zero
length, there is no end to the number of points that we
can fit onto a geometric line, and we are not obliged to
assume that there is one and only one point for every real
number. However, by adopting Dedekind’s definition we
acquire a system in which the arithmetic of points on a
line is fully determined, and, what is more, we can prove
all of classical geometry within this new, fundamentally
discrete system.

Isaac Newton
Newton’s father was an illiterate free-hold farmer who died
months before his birth. When his mother Hannah
Ayscough remarried some three years later, Newton’s stepÂ�
father insisted that she leave her son with the Ayscough
family. Newton grew up to be a rather neurotic and secre-
tive individual, prone to fits of rage. He never married,
loathed criticism, and was intensely curious about the Bible
and the world around him. Indeed, his curiosity about the
nature of man’s vision almost drove him blind. Among
other things, he carefully described what happens when
you stare at the sun all day, and how your field of vision
distorts when you squeeze your eyeballs with a wooden
stick. When the plague struck Cambridge in 1665, a young
and largely unnoticed Isaac Newton (1642–1727) left his
lodgings at Trinity College and moved back to Woolsthorpe,
the Lincolnshire village of his birth. Although Newton was
only in his early twenties, he at his creative peak, there is
a strong case for arguing that no scientist ever achieved
more in a two year period than Newton did between 1664
and 1666. In this brief period Newton conceived of three
of his greatest insights (any one of which would have earned
him a prominent place in history): the Law of Gravity, the
idea that white light is physically comprised of all the colours
of the rainbow, and the beginnings of the calculus.
138 MATHEMATICAL THOUGHT

Unlike his more speculative predecessors, Newton did

not try to explain why things moved, he simply focused
on giving a mathe�matical account of how things move, in
the language of force, mass and acceleration. While
Aristotle and his followers had tried to distinguish between
heavenly motion and the motion of things on earth,
Newton boldly stated that all matter attracts all other
matter, and this ‘gravitational’ attraction is a truly universal
force. That is to say, what Newton realized in a flash of
insight was that the sun and the earth are not the only
things that exert a gravitational pull, as an apple or a lump
of soil or rock attract mass just like any other object (of
the given mass). Furthermore, given any pair of objects,
the size of the force of gravity will be proportional to
each of the masses, and inversely proportional to the square
of the distance between them. In other words, where the
first object has mass m, the second object has mass n, and
they are a distance r units apart, the force F pulling the
two objects together will be
mn
F =G
r2
where G is a number known as the ‘universal gravitation
constant’. This simple statement is almost a complete
summary of Newton’s famous theory. All we are missing
is a definition of what happens when an object experiences
a force of a given size. The question ‘What is a force?’ is
far from trivial. Nevertheless, to understand the success
of Newton’s theory, we only need to understand the motion
that forces produce. By definition, if an object of mass m
is subject to a force of size F, that object will accelerate
at a rate F/m.
Newton’s Law of Gravity has been justifiably described
as the greatest generalization achieved by the human mind.
This simple ‘inverse square law’ has many implications,
the first of which is easy to deduce. Suppose that we have
 MECHANICS AND THE CALCULUS139

two objects of different mass, located in the same place

on earth. Newton’s Law implies the result that Galileo
first suggested: both objects will accelerate at the same
rate, namely Gm r 2 , where m is the mass of the earth, and
r is the distance from the objects to the centre of the earth.
The very same law can also be used to explain the
motion of the planets, the path taken by a projectile,
the existence of the tides, the eccentric orbits of comets, the
precession of the earth’s axis, and many other measurable
phenomena. In particular, the German astronomer and
mathematician Johann Kepler (1571–1630) made a famous
study of some very detailed observations of the motion
of the planets across the night sky. He tried to summarize
this information in a number of ways, and in 1609 he
finally formulated a simple scheme that fitted his data
extremely well. Kepler’s three famous rules are as follows:

1. As the planets move about the sun, they make the

shape of an ellipse, with the sun at one of the foci.
2. If you draw a line from a planet to the sun, and
measure the area that is swept out in a fixed period
of time, you always get the same answer.
3. If you square the amount of time it takes for a
planet to complete an orbit, and divide this number
by the cube of the width of the planet’s ellipse,
you always get the same number (a constant that
depends on the weight of the sun). Among other
things, this means that the length of a planet’s year
can be calculated from its average distance from
the sun, and vice versa.

Famously, all of these results are logical consequences of

Newton’s Law of Gravity. Indeed, it should be emphasized
that Newton’s singular achievement was not so much
conceiving the Law of Gravity, as elucidating its mathe�matical
140 MATHEMATICAL THOUGHT

consequences. Indeed, we know that in 1684 Edmond

Halley, Sir Christopher Wren and Robert Hooke met in
a London coffee shop to discuss the idea that the motion
of planets is governed by an inverse square law, and at
that time Newton’s ideas could be found only in his private
notebooks. Halley, Wren and Hooke tried and failed to
calculate the path that a planet will follow if it is subject
to an inverse square law, and later that same year Halley
asked Newton if he knew how to solve the problem.
Newton immediately responded that an inverse square law
produces elliptical motion, as he had known that result
for nearly twenty years. Halley was delighted by Newton’s
response, but when he asked Newton for a rigorous proof,
it took him nearly three months to reconstruct this argu-
ment from his youth. Halley recognized the supreme
importance of Newton’s work, and thanks to his encour-
agement and financial backing, Newton was persuaded to
spend the next year and half writing the most influential
physics book ever written: the Principia Mathematica.
Newton could not have written his masterpiece without
developing calculus. In light of this fact, it is somewhat
surprising to learn that Principia does not directly employ
the language of calculus. Instead, Newton relied on geometric
arguments that were very much in the tradition of Ancient
Greece, combined with physical intuitions about motion,
and appeals to geometric notions of a limit case. That is to
say, as far as possible, Newton used very traditional means
for finding tangent lines or the areas under curves.
Nevertheless, to develop the new science of mechanics,
Newton needed to answer questions that were beyond the
Ancient Greeks. In particular, he needed a way of handling
the notion of a rate of change. For example, it is clear that
if an object moves four metres in one second, it is moving
with a speed of four metres per second. Similarly, if an
object moves two metres in half a second, it is moving with
a speed of four metres per second. But how can we define
 MECHANICS AND THE CALCULUS141

the speed of an object in an instant, when it doesn’t have

time to move any distance at all?

The Fundamental Theorem of Calculus

As far back as the fourteenth century, the French mathemati-
cian Nicole Oresme realized that we can represent the motion
of an object with changing position and velocity through
the use of a graph. For example, we can use the x-axis to
represent time, while the y-axis represents distance from the
origin. As we move from left to right, our curve can go up
or down, representing the changing location of the object
in question. The faster our object moves, the steeper our
curve will be. Indeed, given such a representation of motion,
we can define the instantaneous speed of our object to be
the gradient of this curve. This idea connects mechanics with
one of the two fundamental procedures of the calculus:
finding the gradient of a curve (differentiation). The other
fundamental procedure of the calculus is called integration.
In many cases we can identify the process of integration
with the process of finding the area under a curve, though
strictly speaking that is not how integrals are defined.
Archimedes and other ancient mathe� maticians were
familiar with the concept of a tangent. They also knew
how to find the areas underneath certain curves. However,
their proofs were quite specific to the shape in question,
and you couldn’t easily adapt their arguments when
confronted with a new curve. Newton and Gottfried
Wilhelm Leibniz (1646–1716) independently developed
calculus, which is a much more general method. Over the
centuries calculus has been used in countless ways, but
the simplest thing we can use it for is to calculate the areas
and gradients of a curve, given an equation that describes
that curve.
As a simple example, imagine that we wish to find the
gradient at a point on a parabola. Because our graph is
perfectly described by the equationâ•› f (x) = x 2 , we can find
142 MATHEMATICAL THOUGHT

the gradient by examining the equation. This is in contrast

to the kinds of geometric reasoning that would have
occurred to the Greeks. The graph becomes increasingly
steep, so we can find a line that will be slightly steeper
than the tangent by drawing a line between the points
(1, 1) and (1+ d,(1+ d) 2 ) . Some simple algebra shows that
this line has a gradient of 2â•›+â•›d (for any number d greater
than zero), which means that the tangent line must be
smaller than any number of this form.

2
(1 + d)
2 2
(1 + d) – 1 = 2 d + d

1
d

1 1+d

To draw a line between these two points, we move 2d + d 2

units up for every d units that we move across. Therefore
the gradient of the line between these two points is:
2d + d 2
=2+ d.
d
Similarly, the tangent line must have a gradient greater
than 2â•›−â•›d, for any number d greater than zero. Therefore,
the tangent line has a gradient of 2 (no more, no less).
More generally, the tangent line at the point (x, x2) has a
gradient less than 2xâ•›+â•›d but greater than 2xâ•›−â•›d, which
implies that the gradient is equal to 2x.
In other words, the gradient at the point (x, x2) is 2x.
We now have a second equation, namely f '(x) = 2x , which
tells us the gradient at any point on our original curve
 MECHANICS AND THE CALCULUS143

f (x) = x 2. The equation f ' is called the derivative of f, and

as we shall soon see, the equation f is called the integral
of f '. Note that we have found f ' by analyzing f: making
a deduction based on the way that we multiply out xâ•›+â•›d2.
There is also a second connection between the equations
f (x) = x 2 and f '(x) = 2x :
y y
y = 2x y = 2x
2b
2a

2
b
a2

x x
a b
The area between the x-axis, the y-axis, the line yâ•›=â•›2x and
the line xâ•›=â•›a is equal to a2 for all values of a. In the notation
of Leibniz, we write:
a
2xdx = a 2.
0

In 1684, Leibniz published the Fundamental Theorem of

Calculus. In other words, he proved that finding the area
under a graph (integration) is the converse of finding the
gradient of a graph (differentiation). We will examine
Leibniz’s theorem below. First I want to mention that
Newton had been familiar with this result for about
eighteen years. Newton was able to make this step because
he had immersed himself in the new mathe�matics of alge-
braic geometry. In particular, he was a virtuoso at working
with ‘power series’. That is to say, he learned that certain
geometric and trigonometric relationships were expressible
in terms of sequences of non-negative powers of x. For
144 MATHEMATICAL THOUGHT

example, like the mathematician-astronomers of Medieval

Kerala, Newton knew that:
x3 x5
sin(x) = x + ...
3 2 1 5 4 3 2 1

Newton also knew another fundamental fact: if the func-

tion f integrates to F, and the function g integrates to G,
then the function fâ•›+â•›g integrates to Fâ•›+â•›G. This is intuitively
obvious, as by definition we find the function fâ•›+â•›g by
adding together f and g. It follows that if we want to know
the area under the function fâ•›+â•›g, we can simply find the
area under f, and add it to the area under g. Similarly, if
the function F differentiates to f, and the function G differ-
entiates to g, then the function Fâ•›+â•›G differentiates to fâ•›+â•›g.
Because Newton knew how to differentiate and integrate
every term in a power series, he realized that he knew
how to differentiate and integrate any function that can
be expressed as a power series. At least, that is how I
would describe his approach: Newton himself did not use
the words function, integrate or differentiate. Nevertheless,
in 1666 Newton prepared a manuscript entitled On Analysis
by Equations with an Infinite Number of Terms, and he
showed it to a couple of his peers. Several years later
Newton was ready to publish his ideas on calculus as a
technical appendix to a book concerning optics, but
following a dispute with his hated rival Robert Hooke,
Newton withdrew the entire work.
Newton was obsessed with the ideas that gripped
his  mind, and he alternated between being suspicious,
indifferent or disdainful of what the wider world might
think. In contrast, the great polymath Leibniz always had
one eye open for an appropriate audience, and he would
happily work on any problem that might earn him money
or enhance his reputation. Like Newton, Leibniz found
a way to use the principles of arithmetic in thinking about
 MECHANICS AND THE CALCULUS145

rates of change, but there were significant differences in

the forms of calculus that the two men developed. Newton
argued geometrically, looking at the limit case of a sequence
of tangents. In contrast, Leibniz’s argument explicitly
referred to a ratio of infinitely small quantities.
From a modern perspective, the concept of an infinitely
small quantity is perfectly respectable, even though (by
definition) such infinitesimals are smaller than any real
number. I say that infinitesimals are perfectly respectable
because they have a clear definition in set theoretic terms,
yielding a number system called the ‘hyperreals’, and a form
of matheÂ�matics called ‘non-standard analysis’. However, as
that name suggests, mathe�maticians do not usually admit
infinitesimals as part of their number system. By using the
concept of a limit case we can do calculus without referring
to infinitely small ‘numbers’, and since that is the standard
method, modern mathe�maticians follow Newton rather
than Leibniz. On the other hand, it is Leibniz’s superior
notation that we use today.
Leibniz’s proof of the Fundamental Theorem of Calculus
is particularly elegant, and it exemplifies his remarkable
ability to spot the crucial point. We begin by imagining a
continuous, unbroken curve described by an equation f(x).
Any mathe�matical rule that takes a number as input and
gives a number as output is called a ‘function’, and our
equation f(x) is a function because for each point a there
is unique output, namely f(a). As shown below, we can
turn this equation into a drawing of a curve, where the
height of the curve at the point a is f(a). Note that at each
point, there is a certain amount of distance between the
x-axis and our curve. Furthermore, we can enclose a finite
area between the x-axis, the y-axis, the curve yâ•›=â•›f(x) and
the line xâ•›=â•›a. Leibniz denoted this area with the symbols
a
f (x)dx , but for simplicity’s sake, I will use the symbols
0
F(a) to denote the area between 0 and a.
146 MATHEMATICAL THOUGHT

y y
f (x) f (x)
f (a) f (a)

F(a) F(a + d x)

x x
a a + dx

If we vary the constant a, we change the area in question.

In other words, our original curve also determines a second
function, as for any positive number a there is a unique
number F(a), which is equal to the area under the curve
up to the point a. Now, if we change a to aâ•›+â•›dx (where
dx is an ‘infinitesimal’ amount), the area in question gets
infinitesimally larger. Assuming that f(a) is positive,
F(aâ•›+â•›dx) is larger than F(a), because the area F(aâ•›+â•›dx)
contains an additional strip of height f(a) and width dx.
By definition, the gradient of F(x) at the point xâ•›=â•›a is
given by a ratio: the change in F(x) as we increase x from
a to aâ•›+â•›dx, divided by the change in x (a change that equals
dx by definition). In other words, the gradient of F(x) at
F(a + dx) F(a)
the point xâ•›=â•›a is . We have just seen that
dx
F(a + dx) F(a) = f (a)dx . Hence the gradient of F(x) at
xâ•›=â•›a is:
F(a + dx) F(a) f (a)dx
= = f (a) .
dx dx

Since F(x) is the integral of our arbitrary, continuous func-

tion f(x), it follows that if we integrate a function and then
differentiate the result, we get back to our original func-
tion. In other words, the Fundamental Theorem of Calculus
holds true.
 MECHANICS AND THE CALCULUS147

From Algebra to Rates of Change

In the previous section, I recounted Leibniz’s proof of the
Fundamental Theorem of Calculus. Apart from being in
English, my version of the argument is very similar to his
original proof. There is, however, a subtle but significant
difference in the way that we follow the argument. Unlike
seventeenth-century mathe�maticians, modern readers will
be familiar with the concept of a function. Strictly speaking
a function is a kind of set, where each acceptable input is
paired with a unique output. Even if we are not familiar
with this formal definition, we are used to the idea of rules
that map inputs to outputs. For example, when we learn
how to square numbers, we might be told that fâ•›=â•›x2 ‘is a
function’. Consequently, we now think of functions as
basic mathe�matical objects. For earlier generations, finding
the area of a square of side x was a process, while the
objects were the numbers (or areas) themselves.
Leibniz was an important figure in bringing about this
change. As a philosopher, he was fascinated by the idea
that we can capture knowledge in the form of a rule, and
I think it is fair to say that in some sense, Leibniz had a
function-based perspective. Nevertheless, the modern,
totally general definition of a function was not put forward
until 1755, when Leonhard Euler wrote his Institutiones
Calculi Differentialis. Indeed, one could argue that the
concept was not fully developed until the work of David
Hilbert (1862–1943). In particular, Hilbert wrote about
‘operators’, which are rules that take a function as an input,
and give a function as an output. From his modern, highly
abstract perspective, the set of all real-valued functions is
a typical mathe�matical object, much like the set of all real
numbers.
Integration and differentiation are both operators, as
when we differentiate a function, we produce another func-
tion. For example, we have seen that the function f (x) = x
2

differentiates to f '(x) = 2x, and that f '(x) = 2x integrates

148 MATHEMATICAL THOUGHT

to f (x) = x 2. It is also worth mentioning that other basic

calculus results were known to mathe� maticians before
Newton and Leibniz. For example, the Dutch mathemati-
cian Johann van Waveren Hudde (1628–1704) knew that
if we differentiate the function x k, we get kx k 1. Of course,
he didn’t use the term ‘differentiate’, but he calculated the
tangents of some algebraic, geometric curves, including
ones of the form y = x k . As I have mentioned, this kind
of equation-based approach to geometry was the form of
mathe�matics that immediately preceded the invention of
calculus, and Leibniz and Newton were both familiar with
Hudde’s work.
An even older result is that the area under a curve x k
x k +1
is equal to (provided that k is not equal to −1). We
k +1
can deduce this fact by combining the previous result and
the Fundamental Theorem of Calculus. Also note that
particular cases of this result correspond to standard,
geometric problems. For example, finding the area under x 2
corresponds to the problem of finding the area under a
parabola. Indeed, the special cases where kâ•›=â•›1, 2, 3 or 4
were known to the Arab mathematician Ibn al-Haytham
(c.â•›965–1039). Six centuries later, the Italian Bonaventura
Cavalieri reinvented Archimedes’ idea of slicing a shape
into infinitely many pieces, and proved this result for values
of k up to 9. He conjectured that it is true for all integers
k, and in the 1630s the French mathe�maticians Fermat,
Descartes and Roberval proved that his conjecture was
correct. Before calculus was even invented, Fermat had
a
a k +1
effectively proven that x k dx = for fractional values
0 k +1
of k.
In the centuries that followed the invention of calculus,
European science underwent a massive expansion. As
Newton had forcefully demonstrated, calculus and the new
 MECHANICS AND THE CALCULUS149

mathe�matics could prove invaluable to science. As ever,

the developing mathe�matical theory co-evolved with the
questions people sought to answer. In particular, later
generations of mathe�maticians developed the theory of
differential equations. By definition, a differential equation
is one that relates a function f to its derivative f '. For
x
example, f (x) = f '(x) is a differential equation (though
2
not a very interesting one), and we can see that f (x) = x 2
is a solution. Differential equations play a prominent role
in engineering, physics, economics, chemistry, biology and
many other disciplines. In some ways this is not surprising.
Whenever we have a rule that tells us how a quickly
continuous quantity changes over time, we almost always
end up producing a differential equation.
For example, if we have a few bacteria in a dish, it is
reasonable to assume that the number of bacteria ‘born’
per minute will be proportional to the number of bacteria
that are currently in the dish. In other words, if there is
no shortage of resources, it is reasonable to model the
population size by saying that the derivative fâ•›' is propor-
tional to f, where f(t) is the population size at time t. This
is one of the simplest differential equations, and if the
constant of proportionality is a positive number, the solu-
tion to this equation is exponential growth. In other words,
our assumption about the rate of change fâ•›' implies that if
the number of bacteria has increased from n to 2n by time
t, there will be 4n bacteria at time 2t, 8n at time 3t, 16n
at time 4t, and so on.
A more interesting differential equation is described
when we specify that taking the derivation of the deriva-
tive of our function f produces an equation that is
proportional to our original equation f. This kind of differ-
ential equation can be used to describe a plucked string,
as when we pluck a guitar string, it accelerates back towards
its starting point at a rate that is proportional to the distance
150 MATHEMATICAL THOUGHT

by which the string has moved. So, for example, parts of

the string that have moved a distance 2x from their starting
point accelerate towards their starting point at twice the
rate of those parts that have been moved a distance x. This
particular differential equation is known as the ‘Wave
Equation’, and it is one of the most important equations
in the physical sciences. The initial questions that inspired
this equation came from the study of vibrating strings, as
people wanted to describe and explain all the different
ways that a string could be made to vibrate. More specifi�
cally, finding ‘a solution to the Wave Equation’ meant
finding one possible form of vibration, while ‘solving the
Wave Equation’ meant finding every possible solution.
Several mathe� maticians tried to summarize all the
possible forms of vibration, and arguments over the best
approach helped to extend and clarify the concept of a
function. For example, if we look at my account of Leibniz’s
proof of the Fundamental Theorem of Calculus, it isn’t
really clear how free we can be in picking our function f.
The argument certainly holds true for polynomial equa-
tions, but can we imagine that f is any rule for pairing an
input with an output? To put it another way, just how
arbitrary is an arbitrary, continuous function? Similarly,
until the language of analysis was refined, it wasn’t clear
whether every ‘function’ could be expressed in terms of a
power series.
We now know that the Wave Equation has all manner
of applications. It plays a crucial role in describing and
predicting the behaviours of liquids, gasses, electromagnetic
phenomena and many other things. Indeed, it may be fair
to say that vibrating strings have inspired more mathe�
matics than any other object. Many different branches of
math, from partial differential equations, to Fourier series
and set theory, have deep roots in the study of vibrating
strings. It is also remarkable to note that with our current
know�ledge, it has become clearer than ever that if you
 MECHANICS AND THE CALCULUS151

want to study the general phenomenon of vibration and

oscillatory form, stringed instruments are the best place
to start (much easier, say, than studying the vibrations of
the skin of a drum).
The fact that mathe�maticians have been studying stringed
instruments for thousands of years takes on particular
significance when we reflect on the supreme importance
of selecting the objects of study that yield the mathe�matical
models of the greatest simplicity. After all, finding a
particular, easily managed case, analyzing it, and using that
analysis to develop a broadly applicable framework or
theorem is a central and recurring theme in all of our
attempts to understand the world. That is because when
we are faced with a problem that we do not know how
to solve, we usually make progress by recognizing that
the problem at hand is somehow akin to some other
problem that we do know how to solve. Because of this
pervasive pattern, relatively simple, exemplary cases are
an essential part of every kind of expertise (including mathe�
matics), as our general sense of how the world works is
deeply grounded by our understanding of particular,
concrete cases.
The importance of simple, concrete cases is sometimes
underplayed in the case of mathe�matics, as we might reason-
ably claim that if matheÂ�matics is ‘about’ anything, it is
about general principles rather than particular, concrete
cases. However, that does not mean that special cases are
anything less than essential. We might always want to move
on to the most general statement we can make, but as
Hilbert remarked, ‘The art of matheÂ� matics consists in
finding that special case which contains all the germs of
generality.’
 Chapter 6:
LEONHARD EULER AND THE
BRIDGES OF KÖNIGSBERG
‘True eloquence consists in saying all that is required
and only what is required.’
François de La Rochefoucauld, 1613–1680

Leonhard Euler
Leonhard Euler (1707–1783) is widely recognized as being
one of the greatest and most prolific mathe�maticians to
ever live. He is famous for many reasons, and has often
been called the father of modern mathe�matics. The mathe�
matical notation that we use today is largely due to Euler
(pronounced ‘oiler’), and because Euler was one of the
first people to understand the power of the calculus, he
was able to make many major contributions to physics,
engineering and astronomy.
During Euler’s lifetime, the Grand Prix of the Paris
Academy of Sciences was the most prestigious and lucra-
tive award a scientist could win. When he was only
nineteen, the Paris Academy posed the following question:
‘What is the best way to arrange the sails on a ship?’ Euler
was born in Basel, and had never left the land-locked nation
of Switzerland. Despite only having seen pictures of large,
sailing ships, the young Euler conceived of the concept of
 LEONHARD EULER AND THE BRIDGES OF KÖNIGSBERG153

‘an equivalent sail’. Essentially, this is the position that a

single mast and sail would need to be in if they were to
match the net propulsive force of some collection of masts
and sails.
At the time, the mathe� matical study of forces and
Newton’s laws of motion were novel innovations, but
Euler was supremely confident in the new science of
mechanics. His ‘Thoughts on a Nautical Problem’ would
surely have shocked many an experienced shipbuilder, as
the teenager boldly stated that ‘I did not find it necessary
to confirm this theory of mine by experiment because it
is derived from the surest and most secure principles of
mechanics, so that no doubt whatsoever can be raised on
whether or not it be true and takes place in practice.’ Sure
enough, Euler’s analysis was sound, and his ideas were
incorporated into the next generation of English and
French naval ships.
Euler continued to publish a stream of brilliant ideas
right up until the day he died, even though he was
completely blind for the last twelve years of his life. He
was uniquely willing to share his preliminary guesses and
partial proofs with anyone who was interested, and no
mathematician in history has been more prolific.
Unlike many mathe�maticians, he constantly sought to
explain himself as simply and directly as possible, often
working to prove and prove again already established
results. He wrote the most influential textbooks since
Euclid, and did more to create our efficient, modern nota-
tion than any other author. For example, his particular
way of writing the facts of trigonometry is that taught
today: he started the practice of writing f(x) to denote a
function of a variable x, and he introduced the symbols e
for 2.718… and i for 1.
In Euler’s time it was just about possible for one indi-
vidual to learn the entire body of European mathe�matical
knowledge. This is what Euler did, and he possessed an
154 MATHEMATICAL THOUGHT

exceptional ability to cut through the clutter of centuries’

of mathe�matical practice, revealing the necessary concepts
through particularly perspicuous representations. Indeed,
I think it is fair to characterize his genius by saying that
more than any mathematician that preceded him, Euler
lived by the principle that Wittgenstein articulated: ‘The
mathematician has no right to be surprised.’ When experiÂ�
mental scientists put nature to the test, they must keep
their eyes open for surprising results. By way of contrast,
mathe�maticians are struck by the fact that established
concepts may be far more fruitful and rich in patterns than
anyone had realized.
For example, in the case of Pythagoras’ Theorem we
may start by being surprised by the regularity in the ratio
of sides, because we don’t yet understand the result.
However, the effect of the proof is to make us see the
result as obvious, as we might say: ‘It’s no surprise that
the area of a2 plus b2 equals c2, since all you have done
is rearrange the triangles.’ Many of the greatest matheÂ�
maticians have used their understanding to penetrate a
mystery, and presuming themselves exceptionally able,
they have left the result as something difficult to see.
Euler’s genius was to realize that if matheÂ�maticians are
convinced of a particular conclusion, it is wise to have
faith that the relevant reasoning can be stated with a
simple precision. Time and time again he searched for
more ever direct ways of proving already established
results, and he is credited with crisply formalizing much
of the mathe�matics and physics of his day.

The Bridges of Königsberg

Königsberg was a town in Prussia, but it is now an enÂ�-
clave of the Russian Federation, and has been renamed
Kaliningrad. The old town of Königsberg was famous for
its seven bridges, and in the early eighteenth century,
people sometimes posed a challenge: ‘Can you walk
 LEONHARD EULER AND THE BRIDGES OF KÖNIGSBERG155

through town crossing every bridge once and only once?’

Euler’s analysis of the problem is a beautiful example of
a mathematician ignoring the irrelevant, saying all that is
required and only what is required. More specifically, he
realized that it doesn’t matter how you wander about
within each of the districts in town. We can draw each
bridge as a line and each district as a point, and the re-
sulting map captures all the relevant facts. In other words,
we can move about the town crossing each bridge once
if and only if we can move about the following simplified
map by crossing each line once:

A A
B B
C C
D D

We have to start in a district, and we have to stop in a

district, and there are four districts in Königsberg. This
tells us that there are at least two districts where we neither
start nor finish. Imagine someone trying to find a path
that crosses each bridge once. As they visit one of the
districts where they neither start nor finish, they must
leave on a different bridge to the one they used to enter,
so each ‘entry’ bridge must be paired with a different ‘exit’
bridge. This tells us that any region where we neither start
nor finish must have an even number of bridges. In other
words, to complete the task of crossing each bridge once,
the districts where we neither start nor finish must contain
an even number of bridges. All four regions in Königsberg
have an odd number of bridges, which means that the task
must be impossible.
As well as being a charming little puzzle in its own
right, Euler’s solution pointed towards some new and
156 MATHEMATICAL THOUGHT

beautiful math. Indeed, when Euler published his thoughts

about the problem, he entitled his paper, ‘The Solution
of a Problem Relating to the Geometry of Position’. This
is a very suggestive title, and in the next chapter we will
see how mathe�maticians developed the idea of multiple
geometries. First, I want to explore the concept of a
network, and the notion that there is a subset of geometric
truths, namely the truths where distance is not relevant.
Euler had some understanding of this powerful idea, but
as we shall see in a later section, a rigorous and fully
general presentation of the basis of ‘topology’ had to
wait until Henri Poincaré, who tackled the subject in
1895.

On Drawing a Network
A very basic matheÂ�matical idea is that of a ‘network’ or
‘graph’, which is the name we give to a collection of labelled
points that are connected together by lines. To qualify as
a single, connected network, we must be able to reach
every point by crossing the lines of the network.
Furthermore, the end of every line counts as a point, as
does every intersect point (where two or more lines meet).
To develop this idea further, we need to notice that we
can draw any network (starting with a point) using three
different types of move:

1. Draw a new line to a new point.

2. Draw a new point on an old line.
3. Draw a new line between two old points.

Now, imagine any connected network. A simple way to

characterize that network is to count the number of speci-
fied points (giving a number P), and count the number of
lines L. We can also count the number of regions R that
a drawing contains. Intuitively speaking, a region is just
an area contained within a set of lines. So, for example, if
 LEONHARD EULER AND THE BRIDGES OF KÖNIGSBERG157

our network was a square with a point at every corner,

we would have Pâ•›=â•›4, Lâ•›=â•›4 and Râ•›=â•›1.
We now have three numbers which we can use to describe
any given network. To gain a further insight, let’s look
back at our three network–generating rules. After move
(I), we have one extra point and one extra line, so our
total number of points P goes up to Pâ•›+â•›1, while the total
number of lines L goes up to Lâ•›+â•›1. The same is true of
move (II). As you can verify with a pencil and a piece of
paper, move (III) always gives us one extra line and one
extra region, so L goes to Lâ•›+â•›1 and R goes to Râ•›+â•›1. We
can now make a startling observation: after any of our
three moves, the number Pâ•›−â•›Lâ•›+â•›R remains the same. In
other words, no matter what we draw on our piece of
paper, Pâ•›−â•›Lâ•›+â•›R must always equal one!

P 5 6 8 9
L –4 –9 –8 – 10
R +0 +4 +1 +2
=1 =1 =1 =1

All of the networks we have considered have two things

in common. First, they have an ‘Euler number’ P –
Lâ•›+â•›Râ•›=â•›1, and, second, they are all drawn on a piece of
paper. Suppose that we had drawn them on a sphere
instead. In that case we would start our drawing with one
region and one point, as opposed to no regions and one
point. This tells us that the Euler number of a drawing
on a sphere is 2. This is true of every drawing on a sphere.
The argument is almost exactly the same as for networks
on a piece of paper, it is just that the drawing we start
with (a single point) has an Euler number of 2 instead of
158 MATHEMATICAL THOUGHT

1. In particular, we can draw the edges of a cube onto a

sphere: just imagine a cubic balloon that has been over-
inflated into a sphere. The important point is that a cube
that has been inflated into a sphere has the same number
of faces, corners and edges as any other cube. Our theorem
also applies to the other Platonic solids. These kinds of
shape have their own vocabulary, so instead of counting
‘numbers of points’ we talk about ‘number of corners’,
‘number of lines’ becomes ‘number of edges’, and ‘number
of regions’ becomes ‘number of faces’. Our analysis of
drawing networks shows that all manner of faceted solids
have an Euler number of 2.

4â•›−â•›6â•›+â•›4â•›=â•›2 6â•›−â•›12â•›+â•›8â•›=â•›2 20â•›−â•›30â•›+â•›12â•›=â•›2

The Platonic Solids Revisited

We saw earlier in the book that there are five Platonic
solids. That is to say, there are five ways to enclose a
volume V, where the surface is composed of multiple
copies of a regular polygon P. At each vertex the total
angle of the polygons must add up to less than 360º, and
there are only five cases that can satisfy this demand.
However, we have not yet proved that there is only one
regular shape for each of the following types of corner
arrangement:

3 Triangles 4 Triangles 5 Triangles 3 Squares 3 Pentagons

 LEONHARD EULER AND THE BRIDGES OF KÖNIGSBERG159

The completed shapes are described by three numbers: the

number of corners, the number of edges, and the number
of faces. There are three more numbers that are relevant:
the total number of polygons (denoted R), the number of
polygons that meet at each corner (denoted N) and the
number of corners or sides on each of the polygons
(denoted S). If we scatter the pieces on the ground, we
have a total of RS edges and RS corners (R polygons, each
of which has S edges). On the completed shape, every edge
is shared by two different polygons. This means that the
total number of edges must be RS 2 . Furthermore, each
corner is shared by N different polygons, so the number
of corners on our completed shape is equal to RS (the
total number of corners on the polygons) divided by N
(the number of polygons is takes to make one corner of
the finished shape). We now have three equations, which
tell us the number of corners or points (P), edges or lines
(L) and faces or regions (R):
RS RS RS RS
L= , P= and R = 2 + L P = 2 + .
2 N 2 N
There is only one possibility for the number of faces,
because the number of faces (R) is uniquely determined
by the numbers S and N. For example, using pentagons
at the corners we have Sâ•›=â•›5 and Nâ•›=â•›3, so we know that
the number of faces on the completed shape satisfies the
following conditions:
5 5
R =2+ R R and this implies that Râ•›=â•›12.
2 3
Since we now know the value of R, our equations also tell
5 5
us that P = 12 = 20 and L = 12 = 30 .
3 2
Given that we can construct the five Platonic solids, we
can see that there are precisely five solids with surfaces
160 MATHEMATICAL THOUGHT

consisting of a number of identical polygons, meeting at

identical corners.

Number of Number of Number of

Faces Corners Edges
Tetrahedron â•⁄4 â•⁄4 â•⁄6
Cube â•⁄6 â•⁄8 12
Octahedron â•⁄8 â•⁄6 12
Dodecahedron 12 20 30
Icosahedron 20 12 30

Notice that cubes and octahedrons both have twelve edges,

but one has six faces and eight corners, while the other
has eight faces and six corners. This observation is closely
related to a beautiful fact: if we draw lines connecting the
centre points of each of the faces of a cube, we generate
an octahedron, and if we connect the centres of an octa-
hedron, we generate a cube:

Because of this intimate relationship, we say that these

shapes are dual. As the cube is dual to the octahedron, so
the dodecahedron is dual to the icosahedron, while the
tetrahedron is dual to itself:
 LEONHARD EULER AND THE BRIDGES OF KÖNIGSBERG161

The concept of duals is well worth investigating, as can

be appreciated by a simple thought experiment. If you
pick up a regular cube, there are certain ways that you
can rotate it around so that when you place it back, it
looks just like it did before you moved it. You can also
reflect it in a mirror, and get something that looks like the
original. In other words, cubes possess certain forms of
symmetry. The same can be said of the octahedron, and
when you move the cube to produce an apparently identical
form, the octahedron inside it experiences an equivalent
transformation. Because of their connection as duals, it
follows that every symmetry of the cube is also a symmetry
of the octahedron, and vice versa. Similarly, the dodeca-
hedron has the same symmetry as the icosahedron.
This shows that the five Platonic solids fall into three cate-
gories of completely regular three-dimensional symmetry.
We shall return to the concept of symmetry in the next
chapter, where I examine the advent of non-Euclidean
geometries, and the resulting shift in our conception of
math. First, let’s have a quick look at the basics of modern
topology, and see how Euler’s ideas were developed by
Henri Poincaré (1854–1912), and the less well-known Simon
Lhuilier (1750–1840).

Poincaré and the Birth of Topology

The great French mathematician Henri Poincaré was a
worthy successor to Euler. He was a successful writer of
popular science books, and is sometimes referred to as ‘the
162 MATHEMATICAL THOUGHT

Last Universalist’. He made significant contributions to

all the major fields of mathe�matics, hopping from topic
to topic with unusual rapidity. As well as his work in
matheÂ�matics, Poincaré also carried out important research
concerning celestial mechanics, fluid mechanics, the special
theory of relativity and the philosophy of science.
Interestingly, he often developed his papers, books and
lectures by laying out the simplest, most basic concepts
he could think of, preferring to start once again from the
beginning, rather than jumping into a complex, cutting-
edge discussion with experts from each field.
In 1895 Poincaré published a book called Analysis Situ,
which was an early systematic treatment of topology,
particularly the fundamental concept of a ‘continuous func-
tion’. I won’t describe the formal definition of a continuous
function, but, as the term suggests, it is the nature of
continuous things to have no gaps or sudden jumps. A
line is said to be continuous if you can draw it without
taking your pencil off the paper. Similarly, we can imagine
a continuous function as a rule for gradually distorting a
shape, without any tears or jumps. We can, for example,
continuously morph a cube into a sphere, or continuously
morph a map of Königsberg into our simplified network.
When we are looking at the topology of solid shapes, it
is rather like the shapes are made from an infinitely stretchy,
shrinkable piece of rubber. For this reason, topology is
sometimes known as ‘rubber-sheet geometry’.
We have already seen that for the Platonic solids, the
Euler number is P L + R = 2. This result was known to
both Euler and Descartes, and Euler’s proof of 1752 showed
that this result is very general indeed: all manner of faceted
shapes have an Euler number 2. However, not every shape
has an Euler number 2. It was Poincaré who identified the
two topological properties that are logically equivalent to
having an Euler number 2:
 LEONHARD EULER AND THE BRIDGES OF KÖNIGSBERG163

Property 1: All the surfaces of the shape must be

connected. Not every shape has property 1. For example,
consider a cube with a smaller cube removed from its
interior. This shape has an Euler number 4, as the surface
of this shape is composed of two disconnected parts,
each of which makes a contribution of 2 to the total
Euler number.

Property 2: If we cut the shape, it should fall into two

pieces. The above drawing illustrates a shape that does not
have property 2. If we cut this shape along the marked
164 MATHEMATICAL THOUGHT

loop of edges, it will not fall into two pieces. This is related
to the fact that this shape has an Euler number of 0.

All the Platonic solids and countless other shapes share

these two topological properties. This is related to the fact
that we can continuously morph one such shape into
another. Continuous functions can distort a shape in many
ways, but they cannot add or remove holes. This follows
because making a hole requires moving neighbouring points
apart, which contradicts the definition of continuous. As
far as topologists are concerned, spheres and cubes are
effectively the same shape, because each can be continu-
ously distorted into the other. For this reason, we say that
spheres and cubes have the same topological identity.
Similarly, a doughnut has different topological properties
to a sphere, but a doughnut and a coffee cup have the
same topological identity. To understand this idea a little
better, consider what happens when we draw a network
on a doughnut shape.
By drawing these two lines, we effectively cut our shape
(called a torus or doughnut) into a flat surface:

From this point on, steps (I), (II) and (III) cannot affect
the Euler number of our network. In other words, the
hole enables us to draw exactly two lines ‘for free’. The
Euler number of a shape is defined to be the minimum
Euler number of any network you can draw on that shape.
We can see that the Euler number of a sphere is 2, while
the Euler number of a torus is 0.
The first general equation concerning topological iden-
tities was called ‘Euler’s Polyhedral Formula’. Despite the
 LEONHARD EULER AND THE BRIDGES OF KÖNIGSBERG165

name, this formula was first written down by the Huguenot

Simon Lhuilier, in 1813. When he was a young man, Lhuilier
declined a large fortune because he wanted to study math,
and didn’t want the career in the church that was a precon-
dition for receiving his relative’s cash. Lhuilier toyed with
Euler’s work for most of his life, and was taught matheÂ�
matics by a student of Euler’s. He was particularly
interested in the argument concerning the bridges of
Königsberg, and one of the highlights of Lhuilier’s career
was spotting a relationship between the Euler number of
any shape and the ‘genus’ of any shape.
Intuitively speaking, the genus of a shape is just the
number of holes it contains. So, for example, a sphere has
genus 0, while doughnuts and coffee cups both have genus
1. More specifically, we say that a shape has genus 0 if and
only if any loop on the surface of the shape defines two
separate regions: ‘inside’ the loop and ‘outside’ the loop.
Notice that a doughnut does not have genus 0, because it
is possible to draw a loop on the surface of a doughnut
and still leave a single, connected region.
Also notice that we can construct a shape of genus 2
by gluing together two shapes of genus 1, and this process
can be extended in the obvious fashion:

E=0 E=0

E = –2
166 MATHEMATICAL THOUGHT

Examine the pieces we are gluing together (the circles

made from two points and two lines that can be found on
each of the doughnuts). The four lines, four points and
two regions of the first drawing make a contribution of 2
to the total Euler number. The corresponding piece of the
second drawing consists of two lines, two points and no
regions. This is true because the glued part is inside our
new shape, so it does not count as a region. Two lines,
two points and no regions makes a 0 contribution to the
Euler number, which is two less than we had before. In
other words, given two shapes with Euler numbers E(1)
and E(2), we can glue them together to produce a shape
with Euler number E(1)â•›+â•›E(2)â•›−â•›2. It follows that gluing a
doughnut onto any shape reduces the Euler number by
two, and in general the Euler number E is related to the
genus g by Euler’s Polyhedral Formula, Eâ•›=â•›2â•›−â•›2g.
 Chapter 7:
EUCLID’S FIFTH AND THE
REINVENTION OF GEOMETRY
‘Insofar as the propositions of matheÂ�matics give an
account of reality, they are not certain; and insofar
as they are certain they do not describe reality. But
it is, on the other hand, certain that mathe�matics in
general and geometry in particular owe their existence
to our need to learn something about the properties
of real objects.’
Albert Einstein, 1879–1955

Measurement and Direction

The first age of geometry can be characterized by the
Ancient Egyptians, and their ritual of replacing boundary
lines on the flood plains of the Nile. The second age of
geometry can be characterized by Euclid, proving geometric
truths as a consequence of carefully stated axioms. The
third age of geometry did not dawn until the nineteenth
century, and in this chapter we will see how mathe�maticians
made this radical shift.
The geometry of 1800 was remarkably similar to the
geometry of Euclid, and up until the mid-nineteenth
century, geometric statements were understood as being
true descriptions of ideal objects in actual, physical space.
168 MATHEMATICAL THOUGHT

Progress had been made, but Euclid’s Elements was still

the major textbook for students of the subject, and in
comparison to the other branches of mathe�matics, geometric
arguments were much as they had always been. Indeed,
arguments that developed in the newer fields of analysis
and calculus were often translated into geometric problems,
precisely because geometric arguments were seen as trusty,
time honoured and irreproachably sound.
The first sign of a radical break came from long-
standing arguments about the proper role of algebraic
symbols in geometric proof. However, our conception
of geometry did not undergo a fundamental change until
people began to explore the matheÂ�matics of ‘projective
geometry’, ‘non-Euclidean geometry’ and ‘curved spaces’
(or manifolds). To understand the nature of these radical
innovations, and the corresponding shift in our concep-
tion of geometry, we must first recall the following
five  axioms, which Euclid used as the basis for all his
geometric deductions:

1. There is precisely one shortest path (or straight

line segment) connecting any two points.
2. Any straight line segment can be extended indefi-
nitely, forming a straight line.
3. Every straight line segment can be used to define
a circle. One end of the segment is the centre of
the circle, and its length forms the radius.
4. All right angles are essentially identical, in that any
right angle can be rotated and moved to coincide
with any other.
5. Given any straight line and a point that is not on
that line, there is precisely one straight line that
passes through the point, and does not intersect
the line.
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY169

These statements (plus some other, basic logic) define the

concepts of Euclidean geometry. They are related to one
another, and together they form a conceptual scheme.
Formally speaking, all five axioms share the same logical
status. That is to say, within Euclidean geometry, all five
axioms are definitively true. Nevertheless, historically
speaking, the first four statements are closer to the bedrock
of humanity’s geometric understanding. If you spend your
time using a set square and a tape measure, the first four
statements are obviously true. We can demonstrate that
the first four statements describe the world of set squares
and tape measures, but it isn’t clear how you can practicÂ�
ally show the reality of Euclid’s fifth. Clearly, we cannot
physically examine two infinite lines, and confirm that
they don’t intersect!
In the Introduction, I presented a proof of Pythagoras’
Theorem. For those who have eyes to see, this kind of
informal proof is utterly convincing. Indeed, an under-
standing of the truth of Pythagoras’ Theorem predates
Euclid’s system of definitions by many centuries. Given
that this is so, it seems fair to say that the Ancient Greeks
articulated Euclid’s fifth, and drilled it into future genera-
tions because of two, intimately related reasons: they saw
that it is ‘true’, and it enabled them to prove Pythagoras’
Theorem, and other, basic results.
The fifth axiom listed earlier is more accurately referred
to as the ‘Playfair axiom’, after the Scottish mathematician
John Playfair (1748–1819). Euclid’s original fifth axiom
was quite enough to make one’s eyes glaze over, as Playfair
stated that: ‘If a line A and a line B intersect a third line
C in such a way that the angle from A to C plus the angle
from C to B is less than two right angles, then the lines
A and B must intersect. Furthermore, this point of inter-
section will be on that side of the line C where the inner
angles add up to less than two right angle[s], and not on
the side where they add up to more than two right angles.’
170 MATHEMATICAL THOUGHT

Euclid’s fifth axiom was sufficiently messy to repel

generations of mathe�maticians. Some, like Playfair, found
more elegant formulations. For example, the great English
mathematician John Wallis (1616–1703) showed that
Euclid’s fifth is logically equivalent to saying that any
triangle can be blown up or shrunk (e.g. by halving the
length of each side), without changing any of the angles.
Other matheÂ�maticians tried to do away with Euclid’s fifth
altogether by proving this statement using axioms (I) – (IV)
As we shall see, that task is provably impossible. Now,
when I said that Playfair found a more elegant formula-
tion, I meant that Euclid’s original fifth axiom is logically
equivalent to the Playfair axiom, so we can happily make
a switch. That is to say, we can use Euclid’s five axioms
to prove the Playfair axiom, and, conversely, if we assume
axioms (I) – (IV) together with the Playfair axiom, we
can prove Euclid’s fifth. The following statement is another
logically equivalent alternative to Euclid’s fifth: Straight
lines parallel to the same straight line are parallel to one
another. As any of these choices of axiom enables us to
prove exactly the same theorems, our list of defining
axioms is somewhat arbitrary, but aesthetic considerations
have meant that Playfair’s statement is considered the
standard form.
As well as providing an axiomatic basis for their
geometric deductions, the Ancient Greeks tried to explain
the meaning of words such as ‘point’ and ‘line’. For
example, Euclid says things like: ‘A point is that which
has no part’; ‘A path is that which is traced out by a
moving point’; and ‘A line is a breadth-less length’. We
might find such comments intuitively vivid, as they paint
a picture of the things that we are talking about.
Nevertheless, we really don’t need to worry about defining
our most basic terms since their mathe�matical meaning
can be fixed by our axioms, or by any other set of logically
equivalent axioms. After all, it is precisely our axioms that
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY171

tell us what we are entitled to deduce about the things

that our proofs are about. To put it another way, there is
no point in defining a basic word if our proofs do not
actually make use of the definition. A definition other than
that given by the axioms might help students see the kind
of thing that the teacher wants to reason about, and if we
really wanted to we could use more words to define the
words that we use in our definitions, but as Blaise Pascal
advised in The Art of Persuasion, ‘Do not try to define
anything so obvious in itself that there are no clearer terms
in which to explain it.’
Among other things, Euclid’s five axioms define our
commitments when we say ‘these lines are parallel’. We
could also say that Euclid’s fifth relates to the concept of
direction, since we commonly imagine that parallel lines
‘point in the same direction’. After two thousand years of
following Euclid’s scheme, people finally realized that
Euclid’s implicit definition of direction is not the only
mathe�matically valid one, and our ordinary notions of
straight lines and so forth are not the only things that are
consistent with axioms (I) – (IV). As we shall soon see,
we can work equally well by accepting axioms (I) – (IV)
together with a completely different alternative to Euclid’s
fifth. Furthermore, the various geometric deductions that
people had made without referring to Euclid’s fifth now
make sense (and hold true) in a context that is far broader
than was previously appreciated.
Rather poetically, this broadening of the horizons of
geometry first occurred in a prisoner-of-war camp. When
Napoleon was forced out of Russia, the Frenchman Jean-
Victor Poncelet (1788–1867) was captured from among the
dead, and marched for five months to a camp on the Volga.
He spent two years in Russia, and by the time he was
allowed back to France, in September 1814, Poncelet had
completed his Treatise on the Projective Properties of
Figures. I am not going to discuss projective geometry in
172 MATHEMATICAL THOUGHT

this book, but I will describe the kinds of non-Euclidean

geometries that were independently developed by Carl
Friedrich Gauss, Nikolai Ivanovich Lobachevsky, János
Bolyai and Ferdinand Karl Schweikart.
Carl Friedrich Gauss (1777–1855) never published his
early thoughts on the subject, but as early as 1824 he wrote
a letter to a friend that included the following sentence: ‘The
assumption that the sum of the three angles in a triangle is
less than 180º leads to a curious geometry, quite different
from ours but thoroughly consistent …’ The first publica-
tion of the idea that Euclid’s fifth was somehow optional
came in 1829, when Nikolai Ivanovich Lobachevsky (1792–
1856) presented a paper that a provincial journal agreed to
publish, after the St Petersburg Academy of Science rejected
the paper as ‘outrageous’. Although a number of matheÂ�
maticians toyed with the idea, the notion of non-Euclidean
geometries didn’t really become respectable until the 1850s.
Indeed, we might date the change rather precisely, to 10
June 1854, which is when Gauss asked the 27-year-old
Bernhard Riemann (1826–1866) to deliver a lecture entitled
‘On the Hypotheses that Lie at the Foundations of
Geometry’. As we shall see later, the full significance of this
lecture wasn’t widely appreciated until the work of Einstein,
but Gauss certainly enjoyed it. The rock of Riemann’s
argument was the observation that the basis of geometry is
not an intuitive or conceptually necessary knowledge about
‘universal space’, but simply the capacity to speak properly
about measurements. Where we measure and formally speak
about the lengths and angles of lines on a flat surface,
Euclidean geometry is just the tool. Where we measure
lengths and angles on a curved surface, we are still doing
geometry, but it is no longer ‘Euclidean’.
Given that the word geometry derives from the Greek
for ‘earth-measuring’, the foundational significance of
measuring lengths and angles may sound ludicrously
obvious. However, since the Greeks, mathe�maticians have
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY173

had the aristocratic pleasure of knowing all the facts about

angles in a triangle and so forth, without having to use
anything so crude as an actual, physical ruler. To reassert
the appropriate priority of metric theory required a deli-
cate skill: pruning back to the essentials, so that further
branches might grow. The elderly Gauss was absolutely
delighted by Riemann’s approach, and spoke with unchar-
acteristic enthusiasm about the depth of Riemann’s
thoughts. Gauss coined the term ‘non-Euclidean geometry’,
which Riemann was too modest to announce, and from
that time non-Euclidean geometries were no longer consid-
ered as logically curious monstrosities. Instead, the truths
of non-Euclidean geometry were accepted along with
Euclid’s, and the third age of geometry was born.

Non-Euclidean Geometry
By definition, a line between two points is ‘straight’ if
and only if it is as short as any other line that might
connect the two points. This fact is fundamental to the
meaning of the term ‘straight’, and it is true in all geom-
etries, not just Euclidean geometry. Euler realized that
you can find the shortest path between two points on any
convex surface by making a model, poking holes at the
two points in question, threading through a length of
string and pulling it taut. Because pulling the string makes
it as short as possible, it follows a shortest path or
‘geodesic’.
The shortest paths between points on a sphere are
always ‘great circles’, which is the name given to circles
with the same radius as the sphere in question. If you
head straight towards a particular point (e.g. the North
Pole), and just keep on going, your path around the earth
will draw out a great circle. If two people both head
north, their paths cross at the North Pole. Hence the
direction ‘North’ is not what Euclid had in mind while
he was writing the fifth axiom, because he specified that
174 MATHEMATICAL THOUGHT

parallel lines never cross. Nevertheless, we can produce

a perfectly consistent non-Euclidean geometry by
reinterpreting our basic terms like ‘point’ and ‘straight
line’. We are free to do this because the matheÂ�matical
meaning of these terms is specified by the axioms alone,
and not any other intuitions we may have about the
perceived subject matter.
In particular, we can picture a kind of geometry known
as ‘circular elliptic geometry’. We do this by interpreting
the term ‘point’ to mean ‘a pair of antipodal points on a
sphere’ (that is, points directly opposite one another),
‘straight line’ as meaning ‘great circle’, and so on. Axiom
(I) is perfectly compatible with this new conception of
direction, because there can only be one great circle through
any two pairs of antipodal points (and this really is the
shortest path):

Similarly, axiom (II) is satisfied because any shortest path

connecting a pair of points on a sphere can be extended
to form a great circle. Axioms (III) and (IV) are also satis-
fied in this model: we simply interpret the terms ‘right
angle’ and ‘circle’ by reading them as ‘right angle on a
sphere’ and ‘circle on a sphere’. The only somewhat tech-
nical problem is that because our sphere is finite, there is
an upper limit on the size of line or circle we can draw.
This limitation of our model can be overcome, but there
is another fundamental difference between this new kind
of geometry on a sphere and traditional, Euclidean
geometry.
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY175

The fundamental difference is that if we draw a great

circle and a point not on that circle, we cannot draw any
great circle through that point without crossing our original
circle. In other words, under this novel interpretation of
the terms ‘point’, ‘line’ and ‘circle’, axiom (V) is utterly
false, as in this geometry there are no such things as parallel
lines. Indeed, if we want to do elliptic geometry instead of
Euclidean geometry, we have to adopt the following axiom:

Given any straight line and a point not on that line,

there are no straight lines which pass through the
point, but do not intersect the line.

The fundamental point is that we are free to interpret the

basic terms ‘point’, ‘line’ and ‘circle’ however we like, so
long as our interpretation satisfies our axioms. After all,
we are supposed to be relying on nothing but the axioms
when making our deductions. Furthermore, we can use
the axioms of Euclidean geometry to prove that lines on
the surface of a sphere satisfy the five axioms of elliptic
geometry. This tells us that elliptic geometry is at least as
consistent as Euclidean geometry, as there is a Euclidean
object that satisfies the axioms in question.
This argument shows that when people tried to use
axioms (I) to (IV) to prove Euclid’s fifth, they were
attempting the impossible. Euclid’s fifth is consistent with
axioms (I)-(IV), but so are statements that contradict
Euclid’s fifth. It follows that if axioms (I)-(IV) are consistent
(which they surely are), Euclid’s fifth cannot possibly be
a logical consequence of axioms (I)-(IV). That is to say,
Euclid’s fifth tells us something that we simply cannot
deduce from the other axioms.

The Curvature of Space

Familiar terms such as ‘triangle’ and ‘square’ are perfectly
well defined in elliptic geometry:
176 MATHEMATICAL THOUGHT

If we add up the angles in the corners of these triangles, we

find that the total is more than 180º. The sum of the internal
angles is greater than 180º because the boundary lines bulge
out, increasing the angles at each of the corners. Triangles
that cover a small proportion of a sphere have a sum of
internal angles that is very close to 180º, as small parts of a
sphere are very close to being flat. For example, the surface
of a tub of water is not quite flat as it follows the curvature
of the Earth, but this curvature is undetectably small unless
the tub is enormous. On the other hand, if we draw a triangle
that covers a significant proportion of a sphere, the sum of
the internal angles will be significantly larger than 180º.
Indeed, our second example of a triangle on a sphere contains
three right angles, which gives us a sum of 270º. Hence it is
possible to detect curvature by measuring the angles in a
triangle, and when we change the size of a triangle on a
sphere, the angles in that triangle change.
It was Gauss who first developed a rigorous definition
of curvature. The definition of the curvature of a line is
really very simple. It rests on the fact that for any three
points on a plane, there is a unique circle that passes through
all three points (unless all three points lie on a straight line).
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY177

The curvature of a straight line is said to be zero, while

the curvature of a circular arc of radius r is said to be 1/r.
Given a smooth, continuous line, we define the curvature
of each point p by finding a ‘limit case’: an idea that relates
to the infinitesimal calculus. Essentially, Gauss imagined
a line through a point p, together with a point on the line
before p and another point on the line after p. There is
precisely one circle that passes through any three points,
and we can find the value of 1/r, where r is the radius of
the circle that passes through all three points. If we pick
a closer pair of points before and after p, we get another
value of 1/r. In the case of smooth curves with no corners,
the value of 1/r gets closer and closer to some fixed value
as the points before and after p move closer and closer to
p. By definition, we say that the curvature at p is the limit
case 1/r.
More importantly, Gauss also developed a definition
for the curvature of a surface, which relies on nothing
more than measurements of angle and distance within the
surface itself (that is to say, Gaussian curvature does not
depend on the space in which our surface is embedded).
The basic idea is that if we move along a surface, our line
of motion will have a curvature. If we start at a given
point and move in different directions, we will find lines
of different curvature, and in particular we will find one
direction of travel with the smallest curvature k1, and
another direction of travel with the largest curvature k2.
We also say that these curvatures are positive if we are
on the top of a hill, and negative if we are at the lowest
point of a bowl. Now, by definition, the Gaussian curva-
ture at the given point on a surface is equal to k1k2. Note
that the top of a hill and the bottom of a bowl both have
positive curvatures, as k1k2 is positive if k1 and k2 are both
negative. Also note that if either k1 or k2 is zero, the
Gaussian curvature is zero. Finally, note that some surfaces
have a negative curvature, as in the case of a ‘saddle point’,
178 MATHEMATICAL THOUGHT

where you move downhill if you go forwards or back-

wards, but you move uphill if you go left or right.
If a surface contains a (Euclidean) straight line, then the
points on that line must have a Gaussian curvature of zero.
For example, a cylinder and a cone both have Gaussian
curvature zero, as at each point the path with the least
curvature is a straight line, so we have k1â•›=â•›0, which means
that k1k2â•›=â•›0. In other words, we can identify shapes with
zero Gaussian curvature by placing them on a flat surface.
If there is a line of contact between the shape and the flat
surface, the surface must have zero Gaussian curvature. If
there is a point of contact (as we find in the case of a
sphere), the surface does not have zero Gaussian curvature.
Another way to identify surfaces with zero Gaussian curva-
ture is to draw a triangle, taking the shortest path along
the surface between three given points. If the surface has
zero Gaussian curvature the internal angles of this shape
will add up to 180º. If the surface has positive Gaussian
curvature the internal angles will add up to more than
180º, and if it has negative curvature the angles will add
up to less than 180º.
Surfaces with zero Gaussian curvature can be rolled out
flat, preserving all the geometric relationships between the
points on that surface, while surfaces with non-zero
Gaussian curvature cannot be rolled flat. For example, a
cylindrical roller with an engraved geometric pattern works
just fine, and can be used to transfer a pattern from the
roller onto a flat piece of paper. In contrast, the curvature
of a sphere means that it cannot be used as a roller, and
any flat map of the globe necessarily distorts at least one
geometric relationship. That is to say, it is provably impos-
sible to map the surface of a sphere onto a flat surface
without distorting either the lengths or the angles between
points on the surface of the sphere. At best you can produce
a map that distorts the lengths but not the angles, or the
angles but not the areas.
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY179

Gauss called his insight into curvature the Theorema

Egregium (Latin for ‘remarkable theorem’), and it explains
the effectiveness of a common pizza-eating strategy.
Because it is basically flat, a slice of pizza can be seen as
a surface with Gaussian curvature zero. You can bend and
fold a slice of pizza, but it is difficult to stretch, and that
means it is difficult to change its Gaussian curvature.
Similarly, you can bend or fold a piece of paper, but because
it is difficult to stretch a piece of paper, you can’t really
wrap a spherical present with flat wrapping paper. Now,
imagine wrapping a slice of pizza around an invisible
cylinder. In one direction the pizza is curved like a circle,
but because the pizza has zero curvature, in the perpen-
dicular direction it has to lie flat. The slice of pizza cannot
bend in both directions without having a non-zero Gaussian
curvature, and it cannot change its curvature without
stretching. In other words, folding a pizza creates rigidity
in the direction perpendicular to the fold, which is useful
because it stops the pizza flopping, and keeps the toppings
nicely in place. The same kind of reasoning also explains
why a corrugated sheet of metal is nowhere near as floppy
as a flat sheet of the same thickness.
While Gauss was the dominant figure in mathe�matics,
the remarkable French mathematician Marie-Sophie
Germain (1776–1831) made a number of significant contri-
butions, and in 1816 she became the first woman to win
the Grand Prix of the Paris Academy of Sciences. One
of her most brilliant ideas was to realize that Gaussian
curvature isn’t the only useful definition of curvature:
the quantity 12 (k1â•›+â•›k2) is also very informative. She called
this quantity the mean curvature of a surface (at the given
point), and she noted that the mean curvature of a surface
has a fascinating connection with its surface area. More
specifically, if a surface has zero mean curvature at every
point, you cannot distort that surface without increasing
its surface area. This means that materials that tend to
180 MATHEMATICAL THOUGHT

minimize their surface area will naturally form minimal,

zero mean curvature surfaces. For example, it is a phys-
ical fact that if you dip any shaped piece of wire into
some soapy water, the surface tension pulls the surface
area to a minimum. A soap bubble contains a volume of
slightly compressed air, and it minimizes its surface area
by making a sphere (a shape that has the same curvature
at every point on its surface). The air inside a soap bubble
is at a slightly higher pressure than the air outside the
bubble, but when a soap film is subject to a single pres-
sure, every point on its surface must have a mean
curvature of zero. This means that if we dip some wire
into soapy water, the soap film that spans the space
between the wires must have zero curvature, however
we bend the wire.

Also note that a soap film can span between the same wire
shapes in more than one way, though each point on the
surface will always have a mean curvature of zero.

The Unity and Multiplicity of Geometry

As well as the positive curvature of a sphere’s surface and
the associated elliptic geometry, there is also something
called ‘hyperbolic geometry’. In other words, geometricians
can also work in a space with negative Gaussian curvature.
In hyperbolic geometry the angles in a triangle add up to
less than 180º, but the most fundamental fact is that we
work with the following axiom:

Given any straight line and a point not on that line,

there are at least two straight lines which pass through
the point, but do not intersect the line.
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY181

Hyperbolic geometry is somewhat less intuitive than elliptic

geometry, but surfaces with negative curvature do occur
in nature. For example, coral reefs often have negative
Gaussian curvature, but given that the following illustra-
tions need to appear on a flat page, it is easiest to use the
model of hyperbolic space devised by Henri Poincaré. On
Poincaré’s circular map, objects in the space appear to
shrink as they move closer to the edge, but according to
the internal measure of hyperbolic space (which is the only
thing that matters), there is an infinite distance from the
centre of the map to the edge. In other words the circular
boundary is not a location within the space itself, as for
the inhabitants of the hyperbolic world it is an infinite
distance away.
A full account of hyperbolic geometry is beyond the
scope of this book, but Poincaré’s map conveys the basic
idea. On this map, a ‘straight line’ is either a diameter
across the circular map, or a circular arc that meets the
map’s boundary at right angles. Two lines are said to be
‘parallel’ if they intersect on the boundary. Finally, circles
in hyperbolic space look like ordinary circles, but their
centres are not where one would naively expect. Poincaré’s
map accurately represents the angles of hyperbolic space,
so we can see that the angles of a triangle add up to less
than 180º, while the angles of a quadrilateral add up to
less than 360º:

The significance of there being more than one geometry

is frequently misunderstood, but the development of non-
Euclidean geometry triggered major changes in the public
182 MATHEMATICAL THOUGHT

perception of math. Statements about the factual world  of

countable objects and measured spaces lie at the heart
of matheÂ�matics. Indeed, we might say that the language of
facts gives us the language of mathe�matics. It is an ancient
knowledge to describe a square field as a mathe�matical square,
and the truths of geometry are certainly facts of our world.
However, by the end of the nineteenth century, non-
Euclidean geometry had helped to create a popular image
of ‘pure matheÂ�matics’: a logical, deductive discipline disas-
sociated from worldly facts. The ancient technique of
reductio ad absurdum involves the description of things
that aren’t true, where you state a proposition and conclude
that you must reject it as false. What was new was the
notion of accepting and exploring axioms ‘whether or not
they are true’. This was a radical and disruptive idea, and
it changed the public perception of the nature of mathe�
matics. In particular, it gave rise to the idea that only some
mathe�matics was concerned with the real world, as up
until the late nineteenth century, absolutely no one divided
matheÂ�matics into ‘pure’ and ‘applied’. In the final chapters
of this book I shall reconsider the controversial relation-
ship between mathe�matics and the physical world. First I
want to discuss ‘absolute geometry’, and the significance
of having more than one form of geometry.
The first point to make is that the existence of non-
Euclidean geometries does not mean that Euclid was wrong,
or that mathe�matics is fractured into totally separate types.
It is true that Euclid specified one kind of geometry and
not others, but these various geometries are intimately
related to one another as parts of a broader conceptual
scheme. For example, given that we are motivated to iden-
tify straight lines with shortest paths, we cannot simply
draw a squiggle and claim that it defines a ‘direction’, or
some new kind of geometry. We have to have a measure
by which our ‘straight lines’ are ‘shorter’ than every other
way of moving between two points.
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY183

Because Euclid and his followers were careful to avoid

Euclid’s fifth whenever possible, many theorems apply to
any reasonable definition of direction. For example, if one
rotates a match through the internal angles of a three-sided
figure, its head switches side as it rotates through half a
loop. If you rotate through the internal angles of a four-
sided figure, the head of the match does not switch sides
as it rotates through a complete loop.

This observation is said to be part of absolute geometry,

because it can be proved on the basis of axioms (I) – (IV)
alone. This means that it holds true regardless of curvature,
in Euclidean, elliptic and hyperbolic geometry. On the
other hand, it is a distinctive characteristic of Euclidean
geometry that if we measure the angles at the corners of
these shapes, the total comes to 180º in the case of triangles,
360º in the case of quadrilaterals, and more generally the
internal angles of an N-sided polygon will total 180 (N 2).â•›
Each axiom operates in the context of the other axioms,
all of which contribute to the meaning of our words or
symbols. The way that axioms work together is absolutely
fundamental, and logical terminology plays an essential
role in enabling an axiom to be descriptive of a mathe�
matical concept. Mathe�matical logic is a subject we will
return to. At this point I just want to point out that in
order to make sense, collections of axioms must cohere as
184 MATHEMATICAL THOUGHT

a system. In particular, if we are to work from the axioms

to their logical consequences, the axioms must contain
words such as ‘and’, ‘or’, ‘not’, ‘every’ and ‘some’, which
can support logical deductions.
Euclid’s arguments are definitively geometric when they
refer to his five axioms. His arguments also involved the
concepts of ‘equals’, ‘plus’ and ‘minus’. For example, in a
proof of Pythagoras’ Theorem we might say ‘the area of
the largest square is equal to the sum of the areas of the
two smaller squares’. These concepts can be defined axio-
matically, and as we shall see in a later chapter, modern
mathe�maticians prove things about adding and subtracting
by referring to an axiomatic system described by Giuseppe
Peano (1858–1932). Euclid did things slightly differently,
by stating and accepting the following ‘common notions’,
which apply to numbers, lengths, areas and volumes:

1. Things that are equal to the same thing are equal

to one another.
2. If equals be added to equals, the wholes are also
equal (e.g. if Aâ•›=â•›B then Aâ•›+â•›Câ•›=â•›Bâ•›+â•›C).
3. If equals be subtracted from equals, the remainders
are also equal.
4. Things that coincide with one another are equal to
one another. In other words, shapes that lie on top
of one another have the same dimensions.
5. The whole is greater than the part. For example,
the area of a shape must be at least as large as the
area of any part of that shape.

As a final comment on the significance of multiple geom-

etries, let’s return to Riemann’s lecture ‘On the Hypotheses
that Lie at the Foundations of Geometry’. The first part
of his lecture was essentially concerned with geometry
itself. The second part of this now famous lecture posed
deep questions about geometry in the physical world,
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY185

asking about the dimension and geometry of actually

measured space. This would have been inconceivable in
an earlier age, when it was accepted that Euclid’s axioms
captured an immutable framework for spatial comprehen-
sion. In particular, notice that axioms (I)â•› −â•›
(IV) are
physically highly plausible. We can connect any two points
with a taut string (I). We can unroll some more string and
extend the line (II). We can use a length of string to define
a circle (III), and so on. Moving a physical shape really
doesn’t change its area, and axioms (I)â•›−â•›(IV) all capture
statements as physically obvious as this. Euclid’s fifth is
the only axiom that we cannot clearly demonstrate using
stretched threads or rays of light. This is because it makes
a claim about the whole extent of a line, as a pair of
parallel lines must never cross, no matter how far they
are extended.
Gauss actually bothered to make a physical check on
Euclid’s fifth, and as far as he could tell, measurements on
earth are in agreement with Euclidean geometry. For
example, if we measure the angles in a triangle of light
between three mountain tops, we cannot detect a difference
between the sum of those angles and 180º. At the time,
very few people appreciated the significance of the ideas
in the second part of Riemann’s lecture, but sixty years
later Riemann was dramatically vindicated by the general
theory of relativity. Crucially, Einstein realized that we
can give the word ‘straight’ a physical definition by saying
that the path of light in a vacuum will always be straight
(by definition), but his analysis indicated that we cannot
find the shortest path without considering the presence of
mass. In other words, Einstein predicted that light was
affected by gravity, so, in a sense, light rays ‘bend’ around
massive objects. In 1919, during an eclipse, astronomers
confirmed that when light from a star passed nearby the
sun, the pattern of stars was bent just as Einstein had
predicted.
186 MATHEMATICAL THOUGHT

If we want to measure a building on earth or calculate

where a rocket will land on the moon, Euclidean geometry
is just the tool we need. If we want to calculate the path
of light from a distant galaxy, we turn to Einstein’s geom-
etry instead. By analogy, we can detect the curvature of a
sphere by measuring the angles in a triangle. We just need
to draw a large enough triangle. Similarly, the astronomical
observations that made Einstein famous detected a slight
warping of angles. It isn’t easy to get your head around,
but we can describe this situation with perfect mathe�matical
accuracy by identifying gravity with a curvature of space-
time. In other words, the earth can ‘detect’ the presence
of the sun (and moves accordingly) because space itself is
curved. This idea is often represented by the image of a
bowling ball on a trampoline. The surface of the trampo-
line deforms and curves under the weight of the ball, and
if we place a marble on the trampoline, it will accelerate
towards the ball. Likewise, the presence of mass induces
a curvature in space-time, and the effect of that curvature
is to produce the acceleration due to gravity that we are
all familiar with.

Symmetry and Groups

The naïve concept of symmetry is very ancient indeed,
and in ordinary usage, something is said to be symmetrical
if its left half is the mirror image of its right half. For
several centuries now mathe�maticians have been rather
more precise, but also more general, in their use of the
word ‘symmetry’. For matheÂ�maticians, a symmetry is
understood as being a particular kind of transformation,
or rule for moving or changing a mathe�matical object
or shape. More specifically, an object is said to possess
a particular symmetry if and only if performing the given
transformation (or symmetry) produces an object that
is apparently identical to the original. So, for example,
an object that is bilaterally symmetric (i.e. an object
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY187

whose left half is the mirror image of its right half) is

indeed mathe� matically symmetric precisely because
reflecting the right half onto the left and the left half
onto the right yields a shape that is indistinguishable
from the original.
One of the big advantages of the mathe�matical definition
of symmetry is that it is extremely general. There are many
kinds of transformation other than reflection! For example,
it is mathe�matically useful to observe that every shape has
at least one symmetry, namely the identity transformation.
That is to say, if you perform the transformation ‘leave
every point in the shape exactly where it is’, the end result
is identical to the original shape. A more interesting kind
of symmetry is rotational symmetry, which can be found
in the following shape:

For obvious reasons, this shape is said to possess four-fold

rotational symmetry.

Note that if we effect several of these transformations in a

row, the net result is equivalent to one or other of the four
listed symmetries. If we rotate our shape through 180º, and
then rotate it through 90º, the net result is a rotation of
270º. Similarly, if we rotate our shape through 270º, and
then rotate it through 180º, the net result is a rotation of
450º, which is equivalent to a rotation of 90º. The basic and
fundamental fact is that because each of these transforma-
tions produces a shape that is indistinguishable from the
original, any combination of these transformations must also
produce a shape that is indistinguishable from the original.
188 MATHEMATICAL THOUGHT

To give another example of a symmetric shape, the

following is said to possess three-fold rotational symmetry,
as well as reflective symmetry:

As with our previous example, the end result of

performing any sequence of these transformations must
be equivalent to performing one or other of the six listed
symmetries. For example, if we reflect our shape in mirror
line 1 and then rotate it through 120º, the net result is
equivalent to a reflection in mirror line 2. Similarly, if
we reflect our shape in mirror line 2 and then reflect in
mirror line 1, the net result is equivalent to rotating our
shape through 240º.
Reflection and rotation are by no means the only trans-
formations that can be considered as symmetries. Perhaps
the simplest example of another kind of symmetry is what
matheÂ�maticians call ‘translation’, where we move a given
shape in a fixed direction by some fixed amount. For
example, if we have an infinitely long string of identical
beads we can shift each bead one place to the left and the
result is indistinguishable from the original configuration.
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY189

Likewise, we could shift each bead one place to the right,

two places to the left, seventeen places to the left, and so
on, and the result would be indistinguishable from the
original configuration.
As we have already seen, modern mathe�maticians use
something known as ‘group theory’ when they are
describing symmetries, and this branch of mathe�matics has
quite justly been described as the supreme art of mathe�
matical abstraction. All groups are comprised of a set of
elements together with an ‘operation’ or rule for combining
those elements. To qualify as a group, the elements and
operation must satisfy the four, definitive rules listed below.
The most familiar example of a group is the integers (our
set of elements), together with addition (our rule for
combining elements). Another example of a group is the
set of symmetries of any object where we combine pairs
of symmetries by performing one symmetry transforma-
tion followed by the other. Now, the properties that define
a group are as follows:

Closure: If we combine any two elements from our set,

the result must be another element from our set. For
example, if we add any two integers, the result is always
another integer. Similarly, recall that the symmetries of a
four-fold rotationally symmetric object are ‘rotate by 90º’,
‘rotate by 180º’, ‘rotate by 270º’ and the identity trans-
formation, ‘rotate by 0º’. If we combine any two elements
by doing one and then the other, the net result will be
equivalent to one of our four elements, which means that
we have ‘closure’.

Associativity: The operation must be associative. In other

words, if you combine any three of the elements, it cannot
matter whether you combine the first two and then
combine the result with the third, or combine the last
two and then combine the result with the first. For
190 MATHEMATICAL THOUGHT

example, addition is associative, as is shown by the

following example:

(2â•›+â•›5)â•›+â•›1â•›=â•›7â•›+â•›1â•›=â•›8, and likewise 2â•›+â•›(5â•›+â•›1)â•›=â•›2â•›+â•›6â•›=â•›8.

The operation of following one transformation by the other

is also associative, because, for example:

(‘rotate by 90º’ then ‘rotate by 180º’) then ‘rotate by

90º’â•›= … ‘rotate by 270º’ then ‘rotate by 90º’â•›=â•›

‘rotate by 360º’.

Likewise, ‘rotate by 90º’ then (‘rotate by 180º’ then

‘rotate by 90º’)â•›= ‘rotate by 90º’ then ‘rotate by 270º’â•›=â•›

‘rotate by 360º’.

Identity: The group must contain an identity element. In

other words there must be one element that, when it is
combined with any other element, leaves that element
unchanged. In the case of the integers, the identity element
is zero, because, for example, 5â•›+â•›0â•›=â•›0â•›+â•›5â•›=â•›5. In the case
of groups of symmetries, the identity transformation is
‘leave every point exactly where it is’.

Inverse: Every element in the group must have what is

known as an inverse. If you combine an element with its
inverse, then by definition the result is the identity element.
For example, the inverse of 5 is -5, because 5â•›−â•›5â•›=â•›0, and
zero is the identity element. Similarly, the inverse of ‘rotate
by 90º’ is ‘rotate by 270º’, because:

‘rotate by 90º’ then ‘rotate by 270º’â•›=â•›‘rotate by

360º’â•›=â•›‘rotate by 0º’.
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY191

Note that in some cases an element is the inverse of itself.

By definition the identity element has to be its own inverse,
and ‘rotate by 180º’ is another example of an element
which is its own inverse.
Group theorists summarize and in a sense completely
describe the groups they study by using multiplication
tables. These are square tables that contain a row and a
column for every member of the group. For example, we
look to the space in the row marked r and the column
marked s to find the result of combining r with s. The
point is that the multiplication table tells us everything we
need to know about combining the elements of our given
group. The axioms of group theory impose constraints on
the possible forms that such tables can take. For example,
the inverse axiom tells us that by definition, every row
and every column of every group’s multiplication table
must contain the identity element I.
The crucial fact is that there are limited numbers of
multiplication tables. For example, there is only one
abstract group containing three elements (the one whose
elements are r, r2 and r3â•›=â•›I), although this single kind of
group can be illustrated in countless different ways. We
might, for example, say that r is the transformation ‘rotate
by 120º’. Another way to consider the same abstract group
is to imagine that we have three playing cards in our hand,
and then say that r represents the operation ‘move the first
card to the back’. In this case r2 represents moving the
first card to the back and then repeating the operation. As
in the case where r is ‘rotate by 120º’, our group of card
shuffling operations implies that r3 is the identity trans-
formation ‘leave the cards where they are’. The point of
all this is that although our two groups are illustrated in
different ways, they have exactly the same multiplication
table, and so we say that they are two examples of exactly
the same abstract group.
To recap, the different kinds of abstract group can be
192 MATHEMATICAL THOUGHT

distinguished by their multiplication tables, and there is only

one multiplication table for groups containing three elements.
However, in some cases two completely different groups
have exactly the same number of elements. For example,
consider the following pair of groups, and note that only in
the second group is every element its own inverse:

In this illustration I denotes the identity transformation

while r denotes a rotation of 90º.

In this illustration I denotes the identity transformation,

r denotes a rotation of 180º and s denotes a reflection in
a vertical mirror line.

Both of these groups contain four elements, and so we

say that they are both groups of order four. Furthermore,
these two groups are the only ones that contain four
elements: every other multiplication table with four
elements fails to satisfy the axioms of a group. In other
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY193

words, there are precisely two abstract groups of order

four, though each one of these may have countless illus-
trating instances. Since 1980, mathe�maticians have had a
complete classification of all the different ‘simple finite
groups’. We know, for example, that if a group contains
p elements where p is a prime number, then there is only
one possible multiplication table for that group. In fact,
there must be some element r such that r, r2, r3, … , râ•›pâ•›=â•›I
are the only elements of the group. In other words, every
group with a prime order p is equivalent to the group
generated by r, where r is the transformation ‘rotate by
360º/p’. The situation is more complicated when a group
does not have a prime number of elements, as in that case
there may or may not be several possibilities for the multi-
plication table. For example, there are two different groups
that contain six elements, five different groups that contain
eight elements, but there is only one group that contains
fifteen elements.

The Oddities of Left and Right

Mirrors are intriguing objects. Perhaps the strangest thing
is that we can measure every facet of an object, from length,
surface area, volume, colour or texture, yet none of these
will help us to distinguish a left hand from the right. The
oddity is that in order to name left and right correctly we
cannot look to the object alone. As we shall see, this is
because left and right do not carry their differences with
them, but standing together we can establish an arbitrary
convention, calling this one ‘left’ and that one ‘right’. So
why are left and right so similar but different, and why
are there two distinct but equivalent options, and not some
other number?
There are two fundamental facts that underpin the
answer to these questions. The first point to note is that
there are two ways to travel along a line: ‘forwards’
and ‘backwards’. The second point to note is that the
194 MATHEMATICAL THOUGHT

mathematical operation of moving a shape does not change

the object’s shape or properties: it only changes the shape’s
location. As we shall see, these two facts are the key
to understanding why mirror images are the same but
different.
First, let’s consider two-dimensional shapes. If we cut
the following shapes out of a sheet of paper and keep them
flat on our desk, there is no way that we can move them
around so that one is on top of the other. Within the two-
dimensional space there are the two distinct shapes: a
left-hander and a right. However, if we pick one of them
up and turn it over, its front becomes its back, which
switches a left-hander to a right (or the other way around).

In other words, when we compare the outlines of these

shapes within a two-dimensional space, we can see that
they are different. One faces left while the other faces
right, and no amount of rotation will make the two outlines
the same. However, if we compare the outlines within a
three-dimensional space we can see that they are identical.
When we flip one of them over it looks just like the other,
and so by definition the two outlines have the same shape
(one is just a rotated version of the other). This observa-
tion can help us to understand the rather peculiar way that
reflecting an image in a mirror changes the image, but also
leaves it the same.
To put it another way, people sometimes wonder why
mirrors switch left and right. The answer to this question
is that a mirror changes parity (so left-handed gloves look
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY195

like right-handed gloves), but it is not left and right that

are switched. Imagine a person standing in front of a
mirror, with a glove on their left hand but not on their
right. The person’s head will be next to their reflection’s
head, and their gloved hand will be next to their reflec-
tion’s gloved hand. In other words, the up–down axis and
left–right axis remain unchanged. The person differs from
their reflection because mirrors switch forwards and back-
wards, so we stand face-to-face with our mirror image,
rather than seeing the back of our reflection’s head.
In contrast, imagine dressing two identical twins with
gloves on their left hands, and standing them face-to-face.
If one twin is facing North the other faces South, and if
one has a glove pointing West, the other has a glove pointing
East. The only dimension in which both twins are the
same is ‘up–down’, as both twins have their feet on the
ground and their head in the air. In other words, a mirror
reverses your image in just one dimension (‘front–back’),
but in the case of face-to-face twins, two dimensions have
been switched (‘front–back’ and ‘left–right’).
We can change the parity of two-dimensional objects
by rotating them in three-dimensional space. Similarly,
we can mathe�matically rotate three-dimensional objects
in a four-dimensional space, and when we do that, right-
handed objects become left-handed objects. As we cannot
directly observe such a transformation, it is tempting to
imagine that turning a right-handed glove into a left-
handed one requires a moment when the object goes ‘pop’,
as some essence is suddenly switched from type-left to
type-right. However, as we will see in the following
section, there is no sudden moment when the object is
switched. There is nothing truly intrinsic to being left-
handed or right-handed, and we find the change by
comparing a before-and-after version of the glove, not by
looking for a distinctive moment when the essence of our
shape is transformed.
196 MATHEMATICAL THOUGHT

The Möbius Strip

Sometimes the mathe�matical community is ripe for an idea,
and more than one person takes the same elementary step.
In 1858, Johann Listing and August Ferdinand Möbius
independently described the mathe� matical object now
called a ‘Möbius strip’. This fascinating shape can be used
to elucidate the fundamental equivalence between mirror
images, and it is worth the effort of making one yourself.
Simply take a strip of paper, twist one end through 180º,
and glue it to the other end. This shape has only one edge,
as if you run your finger along the side, it twists its way
around both top and bottom. If you take a pair of scissors
and cut the strip along the middle, a single loop emerges.
This new loop has two edges: what was once the middle,
and the single edge of the Möbius strip.

One edge cuts to Two edges cut to

form one ring. form two rings.

We can make a shape called the Möbius strip, but we

can also make a space called the Möbius space. We simply
take two strips of paper, twist the ends and stick them
together. The gap between the strips is a perfectly
ordinary two-dimensional portion of three-dimensional
 EUCLID’S FIFTH AND THE REINVENTION OF GEOMETRY197

space. Mathematicians skilfully ignore the edges of the

strip because it is irrelevant to the questions we are
interested in asking. We do this by imagining that the
‘two’ sides are curled up like a cylinder, although it isn’t
physically possible to join up the edges like this. The
important point is to imagine what it would be like for
a pair of two-dimensional objects that live in a Möbius
world.
On meeting each other, our pair of objects might agree
that they are ‘mirror images’. Now suppose that one of
them sets out for a walk, and comes back to its former
acquaintance. This time each one meets their identical
image, as a loop through the space effectively switches
left and right. It would now appear that they are ‘the
same shape’.

The shape on the right goes for a clockwise walk around

the Möbius space, and apparently comes back switched.

The one that has not moved says, ‘You have changed – we
used to be different but now we are the same.’ The other
replies, ‘I think I would have noticed if a change had taken
place. It is you that must have changed.’ Now suppose
that our two shapes draw identical pictures of themselves,
and agree that whoever matches the picture deserves the
name ‘RIGHT’. Let us follow the fate of the shape on
the left as it moves around the Möbius space. It continues
to match its picture of ‘RIGHT’, and finds its way back to
its companion.
198 MATHEMATICAL THOUGHT

Now they are different again, but both of them are

‘RIGHT’. This is because we are mistaken in thinking we
have said anything at all when we made our claims about
same and different. There is no essence to being of one
form or the other – it is in comparison that each claims its
place (being the ‘same’ or ‘different’ to one another). In
the Möbius space, left is identical to right, and there simply
is no parity to speak of.
Over the last few centuries our understanding of geo�
metry has been radically transformed, primarily because of
the growth of algebra and the infinitesimal calculus. When
explaining calculus people naturally refer to vanishingly small
quantities: a family of concepts that are fraught with philo-
sophical difficulties. However, over the course of centuries
our conception of the infinitely small has been successfully
tamed, as mathe�maticians have used formal logic to formu-
late explicitly the definition of the limit case. In the next
chapter we will see how mathe�maticians have tackled other
kinds of mathe�matical infinity, extending our concept of
the infinite, and proving results about infinite sets.
 Chapter 8:
WORKING WITH THE
INFINITE
‘What has been said once can always be repeated.’
Zeno of Elea, c.â•›490–430 bc

Blaise Pascal and the Infinite in Math

Given any number, we can always say ‘add one’. What
is more, there is nothing to stop us from repeating this
instruction. The unbounded or infinite character of the
integers is a profound and basic truth: there is no largest
number. We can also encounter the infinite by asking,
‘How many points on a line?’ Given any line we can
mathe�matically chop it in half by identifying the mid-
point. This operation leaves us with two ordinary lines,
so we can chop and chop again an infinite number of
times.
The mathe�matically infinite was a hot topic of debate
for Greek mathe�maticians and philosophers, as well as
later scholars. In particular, Aristotle famously proposed
that there are two kinds of infinity: the potentially infinite
and the actually infinite. An actual infinity is one that is
completed, definite and consists of infinitely many elements.
In contrast, a potentially infinite sequence is simply a finite
sequence that can be extended indefinitely. So, for example,
200 MATHEMATICAL THOUGHT

Aristotle described a line of finite length as potentially

infinite with respect to division, because a line can be
divided into two halves, which in turn can be divided into
four quarters, and so on ad infinitum. To this day many
people are prepared to acknowledge only the existence of
the potentially infinite, denying the reality of actual infin-
ities.
In the previous chapter we touched on the topic of
‘infinitesimals’. That is to say, at various points in history,
mathe�maticians have studied curves by considering an
interval that becomes smaller and smaller. This process
suggests the idea of an infinitely small interval, but mathe�
maticians do not need a special number that is infinitely
small: we simply use the formal notion of a limit case. To
put it another way, mathe�maticians skilfully avoided refer-
ence to the infinitely small. As we have seen, the real
number 0.999… is equal to 1, and there is no infinitely
small gap between them.
In this chapter we will see some other ways that mathe�
maticians have worked in the shadow of the infinite. In
particular, we will examine a form of proof first described
by Blaise Pascal, which invokes Zeno of Elea’s principle
of saying something and repeating. We will also see how
Georg Cantor established transfinite mathe�matics, taming
the actually infinite as a part of set theory. First, I want
to write about my favourite thinker: Blaise Pascal (1623–
1662).
Pascal thought very deeply about uncertainty, authority
and rules, and his insights permanently changed the way
we think about science and religion. No medieval person
would question whether belief in Christianity was ‘rational’
or not, as it was assumed that evidence for the existence
of God was all around us. Pascal was a deeply religious
man, but he was clear that we are led to Christ by the
heart, not by reason, and we should admit that accepting
the teachings of the church involves a leap of faith.
 WORKING WITH THE INFINITE201

Furthermore, Pascal was very clear that in some cases, but

not others, existing authorities have to be respected. For
example, you cannot understand law without looking to
see what the law courts have already decided, as the very
meaning of legal words depends on the history of law.
Similarly, the religious leap of faith is to accept the word
of prophets. However, Pascal was also capable of pene-
trating scepticism, and he was very clear that when it comes
to scientific knowledge, a public experiment carries more
weight than the word of Aristotle himself. Indeed, his
public demonstrations of the behaviour of barometers were
a real turning point in establishing experiment as the ulti-
mate scientific authority.
Pascal’s genius was evident from an early age, and when
he was only nineteen, he realized that his father’s job of
calculating taxes could be done by a clock-like device. For
example, you can represent the procedure of increasing an
integer by one by rotating a cog by one notch. Pascal was
inspired to construct a device that physically embodied
the basic laws of taxation, and he spent three years refining
one of the world’s first mechanical calculators. These ingen-
ious devices were known as ‘pascalines’, and although a
few curious individuals purchased them, the idea was liter-
ally 300 years ahead of its time, so in 1652 the production
of pascalines was halted.
Pascal also deserves to be celebrated as a founding father
of statistics: a discipline that lies at the heart of all of modern
science. Statistics is the art of reasoning from incomplete
information, and more than anyone who lived before him,
Pascal pondered the relationship between reason and uncer-
tainty. He helped to invent the mathe�matics of probability,
and he was also one of the key figures in developing the
modern understanding of air pressure, vacuums, and other
related phenomena. Not only was Pascal a first-rate scien-
tist and mathematician, he was also the author of one the
most vivid portraits of the human condition ever written,
202 MATHEMATICAL THOUGHT

the Pensées, which defends Christianity by examining the

roles of reason, emotion, faith and uncertainty in the life
of man. And if his many intellectual achievements were
not enough to win our admiration, he was also a remark-
ably charitable man. Indeed, when he decided to give
away most of his possessions, he spent some time thinking
about how his wealth could do the most good, which led
him to invent the idea of subsidized public transport.
Pascal’s proposal was to charge people a small sum for
catching a specially purchased coach, and all the profits
from this scheme were used to relieve the worst effects
of poverty.

Reasoning by Recurrence
As well as grounding their work by the analysis of concrete
cases, mathe�maticians have always searched for the most
general of truths. It is this impulse that drives mathe�matics
to ever increasing levels of abstraction. There are various
ways to establish general results. For example, the intro-
duction to this book contains an informal proof of
Pythagoras’ Theorem, where we saw that by rearranging
four right-angled triangles we can turn an ‘a-square’ and
a ‘b-square’ into a ‘c-square’. Because the argument refers
only to general properties of right-angled triangles, and
we don’t need to know the actual values of the lengths a,
b and c, we can correctly conclude that Pythagoras’
Theorem is true for every right-angled triangle, and not
just the particular one shown in our drawing.
There is another fundamental method for proving general
statements, which was first explicitly stated in Pascal’s The
Arithmetic Triangle (1654). This technique is rather
misleadingly known as ‘proof by induction’. To gain some
understanding of this general form of proof, let’s start by
considering the following demonstration that:
 WORKING WITH THE INFINITE203

52 + 5
1+ 2 + 3 + 4 + 5 = :
2

5
4
3
2 2
1
5 5
2 2

It should be clear that the same area is measured whether

we break our shape into component squares and count
them, or component triangles. This means that our equa-
tion must be true, but what about the more general formula
n2 + n
1+ 2 + 3 + ... + n = ? How can we be certain that this
2 â•›
statement is always true? In other words, how can we be
sure that our equation is correct for every integer n, and
not just for nâ•›=â•›5? Unlike our drawing for Pythagoras’
Theorem, we cannot simply say that our integer n is irrel-
evant, as salient features of the drawing depend on this
particular value. Despite this fact, it is still possible to
prove the general case.
When nâ•›=â•›1, our general equation reduces to the specific
12 + 1
case 1 = , which is certainly true. Furthermore, we
2
can prove that if our equation is true in the case where
nâ•›=â•›N, then it must also be true in the case where nâ•›=â•›Nâ•›+â•›1.
That is to say, if we assume that 1â•›+â•›2â•›+â•›…â•›+â•›Nâ•›=â•› N 2+â•›N , then
2â•›

it follows that:
1â•›+â•›2â•›+â•›…â•›+â•›Nâ•›+â•›(Nâ•›+â•›1)â•›=â•› N 2+â•›N â•›+â•›(Nâ•›+â•›1), and
2â•›

N2╛+╛N ╛+╛(N╛+╛1)╛=╛╛N2╛+╛2N╛+╛2╛+╛N╛╛=╛╛(N+╛1)2╛+╛(N+╛1) .
2 2 2 â•›
204 MATHEMATICAL THOUGHT

We have now shown that if our equation is true in the case

where nâ•›=â•›N, then it must also be true in the case where
nâ•›=â•›Nâ•›+â•›1. Because our equation is true for the case nâ•›=â•›1,
our general argument shows that it is also true for nâ•›=â•›2.
Because our equation is true for nâ•›=â•›2, we can see that it is
also true for nâ•›=â•›3. This means that the equation also holds
for nâ•›=â•›4, which means that it holds for nâ•›=â•›5, and so on for
every integer in turn. The way that this argument proceeds
from an initial case (nâ•›=â•›1) to the general case (n is any
integer) identifies it as a ‘proof by induction’. MatheÂ�matical
induction is a very general form of argument, which can
be applied to many different problems.
One of the most curious things about mathe� matical
induction is that although it was first explicitly used as
recently as 1654, the concept has a very long prehistory.
For example, Euclid’s proof that there are infinitely many
primes implicitly involves something very similar to the
modern form of proof by induction, but the fundamental
‘inductive’ principle was not explicitly identified. Indeed,
I suspect that the oldest form of argument closely related
to proof by induction is genuinely prehistoric, as we can
imagine someone in the distant past making the foolishly
claim that ‘I have counted as high as can ever be counted’.
We could show that they are mistaken by counting one
higher, perhaps adding one mark to their tally. If they then
said, ‘OK, I have counted one higher, and by doing so I
have reached the highest number that can be counted,’ we
should protest at their stupidity. After all, our ability to
add an extra mark to their tally does not depend on the
number of marks that are already present!
In other words, our objection can always be repeated,
and so there can be no integer that is larger than all the
others. A similar sense of conceptual necessity is at play
when we construct a proof by induction. We start by showing
that something holds in the particular case where nâ•›=â•›1, then
we use the general or inductive argument to prove that this
 WORKING WITH THE INFINITE205

implies the result for nâ•›=â•›2, and because we accept that this
second step of the argument can always be repeated, we are
logically compelled to conclude that the result holds for the
entire, infinite sequence of the natural numbers.
Although we have long appreciated the possibility of
endlessly repeating steps in our reasoning, Pascal deserves
most of the credit for explicitly developing the two-stepped
method of inductive proof that we are taught today, and
he used it to demonstrate some enormously significant
algebraic results. Another notable moment in the history
of mathe�matical induction came in 1713, when Jacques
Bernoulli used an inductive proof to bring some much
needed rigour to the rapidly developing field of mathe�
matical analysis. It may have been slow in coming to the
conscious forefront of mathe�matical thought, but by the
latter part of the eighteenth century induction was a widely
recognized tool in Europe’s matheÂ�matical armoury.

The Mathematics of the Infinitely Large

Although the infinite was an important and contentious
concept for the Greeks, it never occurred to them to
compare the infinity of integers with the infinity of points
on a line. The idea that there is more than one kind of
actually infinite number was first articulated in the second
and third centuries bc, by Jaina mathe�maticians working
in India. Jains believe that contemplating very large or
infinite numbers has a mind-expanding spiritual value, and
Jain cosmology was clearly a crucial factor in stimulating
discussion of the mathe�matically infinite. According to
G.  G. Joseph’s The Crest of the Peacock, Jain matheÂ�
maticians recognized five distinct kinds of infinite numbers:
‘Infinite in one direction, infinite in two directions, infinite
in area, infinite everywhere and perpetually infinite.’
It is hard to be certain, but the mathe�matical traditions
of Jainism probably didn’t impact on the European under-
standing of the mathe�matically infinite. There is a strong
206 MATHEMATICAL THOUGHT

argument for claiming that the crucial factor re-igniting

the question of infinity was the long, complex but ulti-
mately decisive shift from the study of numbers to the
study of number systems. Ancient peoples studied numbers
and the properties we can demonstrably ascribe to them,
but in the nineteenth century mathe�maticians began to
compare systematically entire number systems. For
example, it is a characteristically modern observation to
note that the integers and the real numbers are structurally
similar, in that adding or multiplying an integer by an
integer yields an integer, while adding or multiplying a
real by a real yields a real.
As well as demanding a level of formal precision and
rigour that was often lacking in earlier work, nineteenth-
century mathe� maticians distinguished themselves by
studying an ever-expanding and exotic range of mathe�
matical objects. This proliferation of abstract objects put
new pressure on the foundations of mathe�matics. The result
was a remarkable growth in the field of mathe�matical logic
(where fundamental work is done to this day), and the
development of set theory.
It was in this context of radical innovation that George
Cantor (1845–1918) began to rethink the way we approach
the infinite, constructing arguments and establishing proofs
that directly referred to actual, infinite sets. Previous mathe�
maticians had said that there are infinitely many integers,
but they did not think of there being an actual number
corresponding to the total number of integers, precisely
because no finite number could be big enough. Cantor’s
controversial claim was that there is a kind of number
which is equal to the total number of integers, but this
number is a transfinite number, and not a finite integer.

Cantor’s Pairs
George Cantor spent his early childhood in St Petersburg,
but his family moved to Germany when he was eleven.
 WORKING WITH THE INFINITE207

As a student he worked in some of the finest mathe�matics

departments in the world, but unfortunately his career as
a professor was much less successful than it could have
been. Stuck in a minor university, Cantor is now famous
for a series of six articles that were published between
1879 and 1884. These remarkable papers formed an intro-
duction to the radical new kind of set theory.
Cantor was gripped by the philosophical implications
of his work, and as a devout Lutheran, he believed that his
work on transfinite numbers was directly connected to the
mind of God. Cantor’s work was controversial, and was
bitterly opposed by many academics, including the great
Poincaré. The mathematician Leopold Kronecker (1823–
1891) was a particularly harsh critic, dismissing Cantor’s
work as ‘humbug’ and ‘matheÂ�matical insanity’. Like many
people before him, Kronecker believed that matheÂ�matics’
soundest and most fundamental base must be a number
theory derived from counting, as he thought that the only
legitimate mathe�matical concepts were those that can be
‘constructed’ in a finite number of steps. Kronecker was
even suspicious of certain Euclidean proofs, and his attitude
can be summarized by his famous remark, ‘God made the
integers; everything else is the work of man.’
Cantor suffered from depression, and was understand-
ably distressed by the vehement opposition that his work
encountered. He desperately wanted to teach in Berlin,
but Kronecker could block his appointment, and as the
editor of one of the two major mathe� matics journals,
Kronecker could also limit his opportunities to publish.
Cantor’s radical work on set theory and transfinite matheÂ�
matics was ultimately influential, but his ideas did not gain
widespread acceptance until the beginning of the twentieth
century. It is worth emphasizing that Cantor did not win
the argument on philosophical grounds (where the debate,
though shifted, continues to this day). His principles were
ultimately accepted because the set-theoretic framework
208 MATHEMATICAL THOUGHT

that his work inspired is incredibly useful. As David Hilbert

and many other mathe�maticians realized, almost all the
different branches of mathe�matics can be described in set-
theoretic terms, including the disciplines that were new at
the time, such as topology or real function theory.
Two of Cantor’s matheÂ�matical ideas are exceptionally
simple and profound. The first of these truly great ideas
was developed with one of the greatest logicians of all
time: Gottlob Frege (1848–1925). Essentially, the two men
constructed a mathe�matically fruitful definition for when
two sets contain the same number of elements, and they
did this by recognizing the deep significance of the
following kind of everyday situation. Imagine getting on
a bus where every seat has a different person in it, and
where there are some people without seats. This observa-
tion tells us that there are more people than seats without
counting either number. Cantor built on this by recognizing
the fundamental role that can be played by a one-to-one
mapping between sets. In our bus example we have a one-
to-one mapping from the set S (the seats) to the set P (the
people), precisely because every seat has a different person
sitting in it. To put it another way, we can make a rule
such that when we are given a seat as input, we return the
person sitting in it as an output. This rule is known as a
‘one-to-one mapping’ because every different input results
in a different output.
Every seat can be associated with a unique person
(namely the person sitting in that seat), but conversely it
is impossible to give a one-to-one mapping from P to S.
We cannot allocate every person a different seat when there
are more people than seats! We can generalize this principle
by saying that a set B is at least as big as a set A if and
only if there is a one-to-one mapping from A to B. In
other words, if every element of A can be paired up with
a different element of B, then we say that B is at least as
big as A. Similarly, A and B are said to be the same size
 WORKING WITH THE INFINITE209

if and only if A is at least as big as B, and B is at least as

big as A. It turns out that this definition is equivalent to
saying that two sets are the same size if and only if we
can pair up all of their respective elements.
This is a natural extension of the everyday notion of
size – if we have seven apples and seven oranges then we
can match them up one against another, and this observa-
tion is part of the meaning of ‘seven’. However, because
this notion of size does not involve counting we can also
apply it when reasoning with infinite sets. That is to say,
infinite sets can also be paired up. For example, every
integer n can be paired with a unique even number 2n, or
a unique square number n2.
The possibility of pairing up the integers with the square
numbers implies a truth that Galileo stated, long before
the work of Cantor: ‘Neither is the number of squares
less than the totality of numbers, nor the latter greater
than the former.’ Also notice that we can set about gener-
ating a list of every positive integer. Similarly, we can
generate a list of every (positive) even number. Since every
element appears somewhere in the defining list, we say
that these sets of numbers are countably infinite. Because
we can pair elements according to place (the first thing in
list one is paired with the first thing in list two, and so
on), we can correctly conclude that every countably infi-
nite set is the same size.
Many people are somewhat perturbed to find that a set
can be the same size as one of its subsets. For example,
there are as many integers as there are even numbers, even
though half the integers are odd. The expectation that a
subset must be smaller comes from our experience of
counting the finite, but these expectations do not apply at
the infinite level. After all, half an infinite set is still infi-
nite! So what about the set of all fractions? How does that
compare in size to the set of all the integers? If you try
to count the fractions on a number line by working from
210 MATHEMATICAL THOUGHT

the smaller fractions to the larger ones, it should be clear

that you have to miss most of them out. Whichever frac-
tion you count first will be larger than infinitely many
other fractions, and more generally every gap between
successive fractions on your list will contain infinitely many
rational numbers. Despite this fact, it is actually possible
to specify a list that contains every single fraction. In other
words, the rational numbers are countably infinite.

As we zigzag through the diagonals of this grid, we even-

tually hit every positive fraction. We can list every single
fraction by starting with 0, and then working our way
through the grid shown above, repeating every element in
both positive and negative form.

A similar image shows that if we have countably many

countable lists, we can combine them into a single count-
able list. All we need to do is replace row one with our
 WORKING WITH THE INFINITE211

first list, row two with our second list, and so on. Our
zigzag path along the diagonals then incorporates every
element from every list. Another fact that many people
find surprising is that the number of (real) points on a
line is the same as the number of points on a square. To
see that this is so, recall that each point on the number
line can be identified by its distance 0.a1a2a3a4 … from the
origin. According to Dedekind’s definition of a line, a line
is just the set of all such points, and given a particular
point (i.e., given a particular sequence of digits a1, a2, …),
we can pair the point in question with a point inside a
square, namely the point with coordinates (0.a1a3a5 … ,
0.a2a4a6╯…). Because the points on a line can be paired
with the points in a square, it follows that there are the
same number of points on a line as the number of points
in a square: an infinite number that is provably larger
than the number of integers.

The Diagonal Argument

At this point you would be forgiven for thinking that
every infinite set is the same size. To see that this is not
the case, we must consider the following question: ‘Is it
possible to list every real number in turn?’ Cantor’s way
of answering this question was the second of his excep-
tionally deep but simple ideas. His argument constitutes
a novel kind of proof, the basic form of which is frequently
borrowed by mathe�maticians. Perhaps the easiest way to
explain his argument is to tell a story, where we imagine
that George Cantor has been challenged to a guessing game
by Lister the Number Genie:

Lister: George, I am challenging you to a guessing game.

I have a list of real numbers, represented by their decimal
expansions. If you can guess a real number that isn’t on
my list, I will grant you a wish.
212 MATHEMATICAL THOUGHT

George: Very well, I accept. Can I see this remarkable list

of yours?

Lister: Of course, here it is:

0.12345…
0.18345…
0.67391…
0.23475…

George: This is ridiculous. Do you seriously expect me

to list an infinitely long real number?

Lister: No, no. I haven’t got all day you know. Just tell
me your general method for calculating the n’th digit, and
I will do the rest.

George: OK. The first digit of the first number on your

list is 1, so to ensure that my number is different from
the first number on your list, I am picking 3 as my first
digit. The second digit of your second number is 8 (i.e.
not 3 again), and so the second digit of my number is
another 3. The third digit of the third number on your
list is 3, so to ensure that my number is different from
the third number on your list, I shall pick a number whose
third digit is 2.

Lister: You’re making your number up as you go along,

you filthy cheat!

George: Are you sure it isn’t you that’s cheating? I mean,

you do have a definite fixed list don’t you?

Lister: Of course.

George: Well then, I have a definite fixed number, namely

the number you get by going down the diagonal of your
 WORKING WITH THE INFINITE213

list swapping 2s for 3s, and replacing every other digit

with a 3.

Lister: OK, OK – I give in. Your ‘diagonal number’ must

differ in at least one decimal place to every number on
my list. For example, it cannot equal the millionth number
on my list, because by definition the millionth digit of
your number is different from the millionth digit of the
millionth number on my list. I don’t suppose you fancy
a rematch?

George: Will I get another wish if I win?

Lister: Sure, but this time you have to provide the list.

George: Fair enough. Is it all right if instead of comparing

digits, we compare finite definitions of numbers, written
out in ordinary mathe�matical language?

Lister: That’s fine by me. Now tell me, what kind of list
are you going to use?

George: Alphabetical, and by order of length.

The set of definable numbers is countable, because every

collection of names or definitions can be listed systemati-
cally. You simply start with the shortest names or definitions
and move on to the longer ones, listing them alphabetically.
Although the set of definable real numbers must be count-
able, Cantor’s diagonal argument proves that every list of
real numbers is necessarily incomplete. He showed that
given any list whatsoever, we can define a number that
isn’t on the given list. We simply define a number that is
different to the first number in our list because it has a
different first digit. Similarly, by definition our number
must be different to the second number on the list because
214 MATHEMATICAL THOUGHT

it has a different second digit, and so on. On the other

hand, the set of finite definitions can be put into an alpha-
betical list. This argument shows that the set of real numbers
must be uncountable, and most real numbers cannot have
a finite definition.
Perhaps the most important point is that Cantor exhib-
ited two clear sizes of infinity. So, for example, the set of
all integers is countably infinite, while the set of all real
numbers is uncountably infinite. This hierarchy of size
can also be extended in a natural fashion to contain infi-
nitely many different sizes of infinity. More specifically,
Cantor showed that given any set, the set of all of its
subsets must be larger than the original set. This means
we can construct an infinite sequence of larger and larger
infinite sets: first we have the set of all integers (say), then
we have the set of all subsets of the integers, then we have
the set of all subsets of the set of all subsets of the integers,
and so on.
The other critical point is that because the idea of a set
is so very general, a vast range of mathe�matical structures
can be described in the language of sets. Partly for this
reason, the notion of a set has become central concept of
mathe�matical logic. The task of a logician is to think about
thinking. Thinking about thinking is very difficult, but
some of the most important intellectual developments in
human history have emerged from our attempts to char-
acterize the full extent of legitimate mathe�matical reasoning.
As we shall see in the following chapter, the first few
decades of the twentieth century witnessed an explosion
in mathe� matical logic, as many of the greatest mathe�
maticians strove to describe ‘the foundations of matheÂ�matics’.
In particular, the great logician Gottlob Frege helped to
bring set-theoretic concepts to the forefront of the mathe�
matical discourse, using the formalism of set-theory to
elucidate essential features of mathe�matical reasoning itself.
 Chapter 9:
THE STRUCTURES OF
LOGICAL FORM
‘Really good systems of logic, says Alembert, are of
use only to those who can do without them. Through
a telescope the blind see nothing.’
Georg Christoph Lichtenberg, 1742−1799

The Formal Logic of AND, OR and NOT

The study of logic is an ancient pursuit, but it is fraught
with the most serious of difficulties. The principle barrier
(or, rather, the unnerving lack of barriers) is the nature of
the subject matter itself. For example, a good legal argu-
ment must be ‘logical’, avoiding self-contradiction and
contradiction with established law, but the skill of devel-
oping such a case is part of the art of law itself. Similarly,
a good scientific argument needs to be logical, but the skill
involved is science itself. The difficulty for the logician is
that they aim to study the basis or framework for our valid
deductions, but they are interested in doing this for the
most general or abstract of contexts imaginable, where we
literally do not know what subject is under discussion.
It is difficult to consider logical deduction regardless of
the subject at hand, but that is precisely what logicians
have tried to do. As Aristotle observed in the Organon,
216 MATHEMATICAL THOUGHT

there are certain kinds of deductive pattern that crop up

time and time again. The basic metaphor underpinning
this form of deduction is that the things of this world can
be put into categories, and it seems that we think about
categories as though they were spatial containers. Cognitive
scientists believe we have an innate capacity to appreciate
‘container schemas’, such as the example of a jar inside a
fridge that we considered in Chapter 1. That is because
people who have never even heard of mathe�matics can
deduce that if a jar is inside a fridge, and an olive is inside
the jar, then the olive must be inside the fridge.
Since at least the time of Aristotle, we have understood
that the properties of a thing can place that thing in a
certain category (namely, the category of things with the
given property), and some categories are understood as
being contained within another, larger category. For
example, we might accept that the category of ‘men’ is
contained within the larger category of ‘mortal things’. It
follows that if Socrates is in the category of men, Socrates
must also be in the category of mortal things, just like an
olive that is inside the jar must also be inside the fridge.
Crucially, this example of deduction fits a general pattern,
and for millennia Aristotle’s account of proper deduction
was considered to be synonymous with valid, logical
thinking. Logical deductions lead us from one statement
to another, so every deduction begins with an initial
premise. For example, we might begin by accepting that
the jar really is contained within the fridge, or that the
category of men really is contained within the category of
mortal beings. Of course, in daily life we don’t always
explicitly refer to an image of one category contained within
another, or worry too much about what kinds of things
our categories contain. For example, consider the following
statement: ‘If it is snowing, then it must be cold outside.’
We might accept that this statement is true, and prefer this
phrasing to saying that the category of ‘occasions when it
 THE STRUCTURES OF LOGICAL FORM217

is snowing’ is contained within the larger category of ‘occa-

sions when it is cold’. After all, we aren’t really sure where
the category of ‘occasions when it is cold’ is meant to
begin or end, but in a sense that doesn’t matter. The essen-
tial point is that if I accept the statement ‘If it is snowing
then it must be cold outside,’ and I also agree that it is
snowing, I would be crazy to think that it is warm outside.
We understand that if I say that ‘It is snowing but it is
not cold outside’, I am contradicting my previously stated
belief that ‘If it is snowing, then it must be cold outside.’
Furthermore, it is the general form of the ‘if … then …’
assertion that enables us to make this deduction. Indeed,
that is the crucial point: to make the deduction, we do not
need to interrogate the meaning of the words ‘snows’ and
‘cold’, we simply need to understand the logical significance
of ‘if … then …’.
We can demonstrate this point with a second example.
If I were to say, ‘If Gobbledygook, then Flibbertigibbet,’
we can sensibly agree that the logical significance of my
‘if … then’ assertion is that whenever I accept the state-
ment ‘Gobbledygook’, I must also accept the conclusion
‘Flibbertigibbet’, precisely because I have already agreed
that ‘if Gobbledygook, then Flibbertigibbet’. In other
words, the following is a legitimate deduction, where the
conclusion (written below the line) is logically entailed by
the premises (written above the line):

‘If Gobbledygook, then Flibbertigibbet.’

‘Gobbledygook.’
----------------------------------------------------
Therefore ‘Flibbertigibbet’.

A few decades after Aristotle’s death, Euclid brilliantly

demonstrated the power of orderly, deductive reasoning.
His axiomatic approach showed that the large and complex
catalogue of geometric truths known in Ancient Greece
218 MATHEMATICAL THOUGHT

could all be deduced from a remarkably simple collection

of fundamental axioms. Many of Euclid’s deductions subtly
relied on geometric intuition, but deductions of the kind
demonstrated above can be carried out in an automatic,
‘mindless’ fashion, without the need to picture the subject
matter at hand.
Indeed, we can make valid deductions simply by spec-
ifying the proper way to use logical words such as ‘AND’,
‘OR’ and ‘NOT’. So, for example, we can make the working
assumption that our ‘atomic statements’ A and B are either
‘true’ or ‘false’. Now, by definition, we say that the state-
ment ‘A AND B’ is true if and only if the statements ‘A’
and ‘B’ are both true. That might sound like a circular
definition (‘AND’ means ‘and’), but the crucial point is
that we can use logical words to generate sentences from
other, smaller sentences. For example, given the statements
‘A’, ‘B’ and ‘C’, we can construct the statements ‘A AND
B’, ‘(A AND B) OR C’, ‘NOT ((A AND B) OR C)’, and
so on. Furthermore, the ‘truth’ of each of these compound
statements is entirely determined by the ‘truth’ of their
component pieces. In other words, we can draw up defin-
itive tables that tell us when we should assent to a
compound statement given nothing more than the truth
or falsity of the atomic statements ‘A’, ‘B’ and ‘C’.
Many of the statements that we can construct in this
fashion are said to be logically equivalent. We say that two
sentences are logically equivalent if they are true or false
in exactly the same cases. For example, ‘A AND A’ is true
if and only if ‘A’ is true, so we say that these sentences
are logically equivalent. Similarly, in classical logic the
statement ‘A’ is logically equivalent to ‘NOT (NOT A)’.
The assertion ‘if A is true, then B is true’ is particularly
important, and in formal logic this is typically written
‘A B’. This statement asserts that whenever ‘A’ is true,
‘B’ is also true. It follows that ‘A B’ is false if and only
if ‘A’ is true but ‘B’ is false. That is to say, there is only
 THE STRUCTURES OF LOGICAL FORM219

one state of affairs that contradicts the claim that ‘A B’,

namely the case where ‘A’ is true but ‘B’ is false. In other
words, in classical logic the statement ‘A B’ can be
rewritten in the logically equivalent form ‘NOT (A AND
(NOT B))’.

Classical Logic and the Excluded Middle

The statement ‘A OR (NOT A)’ can be represented by
the following diagram:

Our scheme presents two acceptable possibilities: the case

where ‘A’ is true, and the case where ‘NOT A’ is true. In
either case, ‘A OR (NOT A)’ must be true, by definition
of the word OR. Statements that are true in every case are
called ‘logical truths’ or ‘tautologies’. Because it does not
matter what the statement ‘A’ is, we can plug any formal
sentence into the form ‘A OR (NOT A)’, and get another
logical truth. For example,

‘((A AND B) OR C)’ OR ‘(NOT ((A AND B) OR C))’

is also a logical truth.

The crucial point to understand is that logical truth is a

testable property. That is to say, given any logical struc-
ture (constructed by using the words AND, OR and
NOT), we can plug in a ‘valuation’ of the form ‘A is
true, B is false, C is true, etc.’, and systematically work
our way along the sentence in question. We eventually
either get ‘true’ or ‘false’ as an output. Since there are
finitely many possible valuations, we can test and see if
220 MATHEMATICAL THOUGHT

any given statement is a logical truth. That is to say,

given any such sentence, we systematically check to see
whether or not every possible valuation gives the output
‘true’.
When we are using a formal language, it really doesn’t
matter what our atoms ‘A’ and ‘B’ assert, though it is
very significant that in every case we assume that our
atoms must be unequivocally ‘true’, or unequivocally
‘false’. This deep-rooted feature of classical logic is known
as the law of the excluded middle. There are modern
variations of mathe�matical logic that do not assume this
principle, and it is worth noting that in ordinary speech
we are not always motivated to grant such an assumption.
The example that is famous among philosophers is the
statement ‘Hamlet’s grandmother has blue eyes’. Since
we have no reason to assert that there are any facts what-
soever concerning Hamlet’s grandmother, it may seem
inappropriate to think that this statement must be either
‘true’ or ‘false’. However, even though Hamlet’s grand-
mother is an utter phantom (being an as-yet-unwritten
fictional character), we must admit that it is quite impos-
sible to cast a film tracing Hamlet’s ancestors without
coming to a decision as to whether or not she has blue
eyes.
Ordinarily we distinguish between factual accounts and
works of fiction by engaging with the wider world, looking
for the occasion to which our words might fit. In the
peculiar case of mathe�matics there is clearly a distinction
between the factual and the false, but the discipline is
unique in that mathe� maticians have no grounds for
sustaining a distinction between the factual and the fictional.
We cannot make such a distinction because the thing in
the world that testifies to mathe�matical truth is the mathe�
matical utterances themselves. In other words, mathe�matical
language brings into being the very truth that it proclaims.
 THE STRUCTURES OF LOGICAL FORM221

Mechanical Deductions
If we are given a sentence constructed out of the words
AND, OR and NOT, we can systematically check to see
whether or not our statement is true in each and every
case. So, for example, we can see that ‘A OR NOT A’ is
true in every case; ‘A OR B’ is sometimes true; while ‘A
AND NOT A’ is always false.
Furthermore, we can be very efficient in the way that
we identify logical truths. It is possible to make a machine
(e.g. a computer) that uses only a few different rules to
generate every logical truth in turn. The basic idea is that
every finite collection of statements can be used to generate
a finite number of ‘logical offspring’. For example, the
statements ‘A’ and ‘B’ can be used to produce the logical
offspring ‘A AND B’. Similarly, the statements ‘A OR B’
and ‘NOT A’ can be used to produce the logical offspring
‘B’. Notice that in the first example we have two shorter
sentences producing a longer one, while in the second we
have two longer sentences producing a shorter one.
The essential point is that given any collection of state-
ments we can systematically set about generating their
logical offspring. For example, we can feed a finite list of
statements into a computer, and mechanically generate a
further list of statements, which I have described as the
logical offspring of the original statements. If we start with
logical truths then we generate offspring that are also logical
truths. If we start with a statement such as ‘A’ (which is
not a logical truth), we will generate more than just the
logical truths. In fact, the statements that we generate are
precisely those statements that are true whenever statement
‘A’ is true (i.e. for every valuation that says ‘A is true’.)
For example, given the statement ‘A’, our machine will
eventually generate the statements

‘A OR B’ and ‘A AND (B OR (NOT B))’,

222 MATHEMATICAL THOUGHT

because both these statements must be true whenever ‘A’

is true. The inputs for our logic machine are called axioms,
while the statements our machine generates (given an
input A) are called ‘the logical consequences of A’.
Arguments in the language of AND, OR and NOT (or
AON for short) are usually called ‘derivations’. These
are just chains of statements, starting with the axioms,
such that every statement is the offspring of the state-
ments that precede it.
Arguments of this form show that if we believe the
axioms, then we should also believe every sentence in the
argument, precisely because we have accepted practice for
using the words ‘AND’, ‘OR’ and ‘NOT’. Indeed, this is
the heart of logical analysis, as we are able to make logical
deductions precisely because we use logical words when
we state our axioms. That is to say, it is the logical struc-
ture of our axioms that enables us to draw conclusions.
To put it another way, deduction is very important, but
when we make a valid deduction, we cannot learn anything
that wasn’t already implicit in the axioms themselves.

Quantifiers and Properties

For over two millennia, logic barely progressed beyond
the ideas of Aristotle. Two people who did make some
progress were Gottfried Leibniz (1646–1716) and George
Boole (1815–1864), who had the brilliant idea of replacing
the logical terms of ordinary language with symbols and
symbolic operations. In other words, Leibniz was pres-
cient enough to dream of turning logical deduction into
a form of calculation, while Boole successfully developed
a formal language equivalent to AON. This was a remark-
able innovation, but there is very little mathe�matics that
can be done if your only logical words are AND, OR
and NOT.
The idea that mathe�matics is or should be an exercise
in logic first became truly plausible with the work of
 THE STRUCTURES OF LOGICAL FORM223

Gottlob Frege (1848–1925). The pivotal moment occurred

in 1879, when Frege published the enormously influential
Begriffsschrift (or ‘Concept Script’). This famous work
contained an incredibly powerful system of formal logic,
which Frege and others developed over the following
decades. In essence, he recognized that we can produce a
perfectly effective automatic logic machine for formal
languages that are much more complex than AON. This
is crucial, because to rigorously state their arguments,
mathe�maticians need a formal language that includes the
phrases ‘EVERY thing’ and ‘SOME thing’. Without these
words we cannot ask difficult questions about the integers.
For example, we could not ask whether it is true that ‘for
EVERY integer n, there is SOME sequence of prime
number p1, p2 , ... , pm such that n = p1 p2 ... pm’.
One of the first things to note about these logical words
is that when we use the words ‘EVERY thing’ or ‘SOME
thing’, it doesn’t really matter what kind of ‘thing’ we are
talking about. This comment may sound rather peculiar,
but remember, in formal languages such as AON, it really
doesn’t matter what our atoms A, B and C refer to. That
is to say, we can understand statements in this language
as a finite collection of ‘meaningless’ symbols, because it
is the words AND, OR and NOT that do all the work
when we make our deductions. Similarly, in Frege’s more
advanced logical language, it is the words AND, OR,
NOT, EVERY and SOME that do all the work, so in a
sense we don’t need to worry about the meaning of any
other term.
As well as having symbols for EVERY and SOME,
mathe�maticians also require symbols that indicate proper-
ties, as an integer may or may not be even, square or
triangular (say). It is not that each new mathe� matical
concept requires a fundamentally new form of logic, it is
just that when we write our mathe�matical sentences in
formal language, we need some kind of symbol to express
224 MATHEMATICAL THOUGHT

the idea that a given thing has a given property. As far as

the logic is concerned it doesn’t really matter what these
properties are, but it is a fundamental truth that in this
language, EVERY thing either has property P or does
NOT have property P. If the property in question is a
bit fuzzy, so it does not make sense to assume that things
either have the property of not, Frege’s logic system is
not appropriate, and we need to use a different kind of
logical language.
The basic idea is that Frege’s logical language contains
the words AND, OR, NOT, EVERY and SOME, together
with an arbitrary number of ‘variables’ and ‘properties’
(or predicates). Not only did Frege provide axiomatic defi-
nitions for these basic logical terms, he also constructed a
strictly rule-governed, symbolic system for representing
logical deductions, rather like a modern computer language.
This formal system is called ‘predicate calculus’, and I shall
refer to it as PC for short. Strictly speaking this language
doesn’t contain the ordinary English words that I have
used to make sentences in PC more easily readable. For
example, we do not need to bother writing the words
‘thing’, ‘has’ or ‘property’, although some kind of basic
grammar is required. Another somewhat technical point
is that in the language PC, there is a correct way to use
brackets, and every grammatical sentence contains equal
numbers of left and right brackets. I am not going to run
through the details, but some of the key points of PC are
the following:

1. In PC, the statement ‘NOT (EVERY thing has

property P)’ is logically equivalent to the statement
‘SOME thing does NOT have property P’. Note
that when we are using the language PC, we are
not obliged to make any kind of claim about our
ability to actually find this thing (whatever it
may be).
 THE STRUCTURES OF LOGICAL FORM225

2. Throughout this book, whenever I am phrasing a

sentence in the formal language of PC, I will capi-
talize the words AND, OR, NOT, EVERY and
SOME. This is to remind the reader that the
sentence in question can be precisely phrased in
terms of a formal language, and it is the capitalized
words that are critical when we are making logical
deductions.
3. Given a list of grammatically correct sentences ‘I’,
we can imagine a PC machine that sets about gener-
ating an infinite list of their logical offspring, called
‘the logical consequences of I’. Inputs of the form
‘A OR (NOT A)’, or ‘EVERY thing (either has
property P OR does NOT have property P)’ are
called ‘logical inputs’, because these statements are
true for every possible valuation.
4. In his doctoral thesis of 1930, Kurt Gödel proved
that when we use logical inputs, PC generates every
logical truth in the given language. This important
fact is known as ‘the completeness of predicate
calculus’. In effect, Gödel proved that we can
systematically generate every statement which is
true for every collection of things, and every prop-
erty P. An example of one of these tautological
statements would be: ‘It is NOT the case that
EVERY thing has property P AND SOME thing
does NOT have property P.’

Inputs for Predicate Calculus

Logical inputs are not the only kind – we can also use
mathe�
matical axioms as inputs for PC machines. For
example, we might use the language of PC to state the
following eminently sensible axiom: ‘For EVERY integer
x there is SOME integer y such that y is bigger than x.’
This logical statement is what we mean when we say that
there are infinitely many integers. Similarly, we can use
226 MATHEMATICAL THOUGHT

the language of PC to define more sophisticated mathe�

matical concepts.
For example, in the early nineteenth century (some fifty
years before Frege published his works on logic), mathe�
maticians were eager to develop new branches of calculus.
In order to extend the fundamental ideas to the most general
case, they needed to specify or define what is meant by a
‘limit case’. Bernhard Bolzano (1781−1848) and Augustin-
Louis Cauchy (1789−1857) independently hit upon the
same definition: a sequence x1, x 2 , ... ‘converges to a limit
L’ if and only if ‘for EVERY positive number , there is
SOME number n such that EVERY term x n , x n +1, ... is bigger
than Lâ•›−â•› AND smaller than Lâ•›+â•› ’.
One cannot overstate the importance of defining mathe�
matical concepts in terms of a logical language, as such an
approach leaves us in no doubt about the kinds of deduc-
tion that we are entitled to make. As Jaakko Hintikka
wrote in The Principles of Mathematics Revisited, ‘We can
make the axioms of a typical mathe�matical theory say what
they say only by using [words such as “AND”, “OR”,
“NOT”, “EVERY” and “SOME”.] If matheÂ�matical prop-
ositions were not expressed in terms of logical concepts,
it would not be possible to handle their inferential rela-
tionships by means of logic.’ Of course, matheÂ�matics is
several millennia older than modern, formal logic.
Hintikka’s point is not that it is impossible to reason
without an explicit formal system, it is just that explicit
formal systems are of great benefit because they help to
elucidate the logical underpinnings that mathe�matical state-
ments necessarily possess.
If we adopt a system like PC, we can made valid deduc-
tions by shuffling symbols in an entirely mechanical fashion
without relying on any kind of understanding of what the
symbols refer to. In a sense, routine mathe�matics does not
require knowledge of whatever we are talking about: we
can articulate the relevant understanding of our subject
 THE STRUCTURES OF LOGICAL FORM227

matter by formulating axioms, and then apply a system

like PC in a mindless, mechanical fashion.
As well as elucidating our basic deductive practices,
formal languages like PC are enormously useful, because
they enable us to construct testable proofs. That is to say,
by translating an argument into PC we can be certain that
there are no hidden assumptions, and that our conclusion
really is a logical consequence of the axioms. In other
words, we can show that one should accept a chain of
reasoning from the axioms to the conclusion precisely
because of the formal practice associated with the words
AND, OR, NOT, EVERY and SOME. Furthermore, the
axioms that we use often seem to force themselves upon
us as inescapable, or ‘self-evidently true’. In such cases
our common sense compels us to accept their implications
quite literally once and for all.
It can be very difficult to construct a formal proof
even if the idea behind it seems relatively clear, but once
this has been done we can be absolutely certain that
there are no hidden assumptions, and that our conclusion
really is a logical consequence of the axioms. It is also
worth noting that the process of translating an intuitive
insight into a strictly formal proof can be very revealing,
but this is not always so. It is certainly possible to check
a formal proof but still be confused about the nature of
the argument. In many ways the most significant thing
for the mathe�matical community is that formal proof
provides a clear and comprehensible criteria for when a
piece of work is complete and valid, which means that
mathe�maticians can reach a truly exceptional level of
consensus.

Axiomatic Set Theory

Frege’s arguments were enormously influential, not least
because of their effect on the Vienna Circle, and the
philoso�phy of Wittgenstein. In the world of mathe�matics,
228 MATHEMATICAL THOUGHT

Frege’s work inspired the growth of two distinctively

modern branches. The first of these is mathe�matical logic.
Bertrand Russell and Albert North Whitehead wrote a
classic example of this kind of work: the three-volume
Principia Mathematica (1910, 1912 and 1913). In this
massive tome, the authors derive familiar mathe�matical
results using a highly formal and rigorous system of logic.
For example, they use set theory to confirm that ‘1â•›+â•›1â•›=â•›2’,
and they also prove things like Pythagoras’ Theorem, and
other more sophisticated results.
Over the next couple of chapters we will return to the
subject of mathe�matical logic, particularly the extraordinary
work of Alan Turing and Kurt Gödel. First I want to
mention the second branch of mathe� matics that Frege
inspired: axiomatic set theory. This branch of mathe�matics
grew from his work less directly, being shaped by devel-
opments elsewhere in the mathe�matical arena. In particular,
the study of sets gained new impetus as a result of George
Cantor, and his ideas about the infinite. Initially, mathe�
maticians presumed that the concept of a set was utterly
basic, as Cantor once described a set as ‘any collection
into a whole M of definite, distinct objects (that is, the
members of M)’.
As we shall see, this ultimately proved to be an inad-
equate way of articulating the concept of set. The problem
was that mathe�matical logicians, starting with Frege, were
interested in the incredibly general notion of a property,
where the basic idea is that we can either ascribe a prop-
erty to an object, or not ascribe that property to an
object. Frege was also concerned with the ‘extension’ of
properties, following the idea that every property deter-
mines the set of things that have that property. For
example, the property of being a dog can be thought to
determine the set of all dogs, while in the realm of mathe�
matical concepts, the property of being a prime number
is related to the set of prime numbers. At worst, it was
 THE STRUCTURES OF LOGICAL FORM229

believed that a well-defined property might have an

‘empty extension’, as, for example, there is not a single
number with the property of being an even prime number
larger than two.
In 1902 Russell posted a famous letter to Frege, saying:
‘I find myself in agreement with you in all essentials. …
There is just one point where I have encountered a diffi-
culty.’ He then went on to mention a property that a set
might have, namely the property of being a member of
itself. For example, there are infinitely many sets that have
the property of being infinite, and so we might imagine
that the set of infinite sets is a member of itself. Conversely,
the set of all dogs is not itself a dog, and so this set is not
a member of itself. Russell posed a question that devastated
Frege’s grandest scheme: given that we can conceive of the
property ‘x is not a member of itself’, what are we to make
of the corresponding set R, which consists of all sets x
such that x is not a member of itself?
The problem is that this so-called set is paradoxical. If
R is a member of itself, then by definition it must not be
a member of itself. Conversely, if R is not a member of
itself, then by definition it is a member of itself! Russell’s
paradox and other related conundrums provoked conster-
nation among many logicians, and the event is sometimes
described as a crisis in mathe�matics. However, the problem
was not with building up collections of familiar mathe�
matical objects, as that process has never led to any kind
of paradox. Rather, the problematic assumption that under-
pins Russell’s paradox is the naïve belief in a curious and
un-matheÂ�matical object known as the ‘set of all sets’. That
is to say, his paradox arises when we try to divide an
un-described totality into two distinct categories: every
set with property A, and every set without property A.
As Gödel sensibly remarked, ‘These contradictions did
not appear within mathe� matics but near its outermost
boundary toward philosophy.’
230 MATHEMATICAL THOUGHT

As a result of these debates, the naïve conception of a

set was formally refined. In particular, Ernst Zermelo
(1871–1953) and Adolf Abraham Fraenkel (1891–1965)
developed the modern, ‘iterative’ conception of sets, which
is described by a branch of matheÂ�matics known as ‘ZF set
theory’. The basic idea of ZF set theory is to build up new
sets from old. The first axiom simply states that there is
a set with no members, and we call this set the empty set.
The other axioms of ZF set theory tell us how to produce
new sets from old. For example, if A is a set then so is
{A}, (i.e. the set whose only member is A). Similarly, the
axioms of ZF set theory state that if A and B are both
sets, then so is A B (where a set is a member of A B
if and only if it is a member of A or a member of B). If
we adopt the eminently sensible axioms of ZF set theory,
the inherently contradictory ‘set of all sets’ is recognized
as not being a set at all, because it is decisively ruled out
by the iterative conception.
Because of this axiomatic approach, set theoretic logic
is not the same thing as predicate calculus. However, the
incredibly general concept of a property is closely linked
to that of a set. Indeed, the most important thing about
axiomatic set theory is that it unifies mathematics, as a
vast range of mathematical ideas can be rephrased in the
language of sets. For example, a line can be defined as
being a set of points, and the points where two lines inter-
sect is simply the set of all points which belong to both
the set that defines the first line and the set that defines
the second line. Similarly, the statement ‘the integer x is
smaller than y’ is equivalent to the statement ‘the pair of
integers (x, y) have the property xâ•›<â•›y’, which in turn is
equivalent to saying that ‘the pair of integers (x, y) belong
to a particular set, namely the set of pairs of integers where
the first integer is smaller than the second’. Of course, to
make these statements using nothing but the language of
set theory, we need to define the integers within the
 THE STRUCTURES OF LOGICAL FORM231

language of sets, but that is not difficult to do. We simply

identify ‘0’ with the empty set, while ‘1’ is identified with
the set that contains the empty set and nothing else, so by
this definition ‘1’ has one member. The number ‘2’ is iden-
tified with the set that contains two members (the empty
set ‘0’ and the set ‘1’), and so on.
 Chapter 10:
ALAN TURING AND THE
CONCEPT OF COMPUTATION
‘If you have clear concepts, you know how to give
instructions.’
Johann Wolfgang von Goethe, 1749–1832

From Mechanical Deductions to Programmable Machines

In the previous chapter we saw how mathe�matics at the
beginning of the twentieth century was characterized by
an extreme concern for formal rigour, enabled by signifi-
cant advances in formal, symbolic logic. Although great
progress was made, this era of mathe�matics was marked
by a belief in a pair of philosophical propositions that
have ultimately proved to be untenable. First, it was
believed that the essence of a subject matter can always
be given by a small set of axioms, and that those axioms
ought to provide a secure foundation on which the entire
subject can rest. As we shall, the truths of arithmetic cannot
be reduced to a finite set of axioms, so this idea is defi-
nitely mistaken. Second, it was believed that mathe�matical
reasoning is a form of calculation, and that all truths can
be calculated using mathe�matical logic. As I discuss in the
final chapter, this belief is not justified by the practices
of real �mathe�maticians. Nevertheless, we might consider
 ALAN TURING AND THE CONCEPT OF COMPUTATION 233

completely explicit, mechanically reproducible forms of

reasoning the ideal to which we ought to aspire.
The relationship between mathe�matics and logic has been
grist in the philosophical mill for many centuries, but the
genius of Gottlob Frege, and the possibilities presented
by his predicate calculus, led many of the brightest minds
into the challenge of ‘rebuilding’ the foundations of math.
In the beginning of the twentieth century, geometry,
number theory and other branches of math were rephrased
into the language of sets, so that the vagueness and ambi-
guity of natural language could be scrupulously avoided.
For example, the massive Principia Mathematica uses a
few axioms of set theory, together with a few rules of
inference, and on this basis the authors derive a substantial
proportion of ordinary mathe�matics.
In many ways, this work was the culmination of an
ancient scheme. At the very least, Leibniz had dreamed
of ‘a general method in which all truths of reason could
be reduced to a kind of calculation’. In Leibniz’s day
there was little progress in formal, mathe�matical logic,
but by 1910 the situation had changed. The many ‘truths
of reason’ contained in the Principia can all be reduced
to a kind of calculation. More specifically, we can adopt
the language of the Principia, and generate many true
statements, simply by following a few, clearly stated prin-
ciples. The extreme, systematic rigour of such works was
philosophically influential, and by trying to fit different
areas of study into a common framework, mathe�maticians
revealed many fundamental connections that had previ-
ously been obscured.
In 1928, David Hilbert cut to the heart of the matter
by posing a challenge known as Hilbert’s decision problem.
The underlying idea is that given a formal language, there
are certain statements we can make using the symbols of
that language. Hilbert’s challenge was to find a general,
mechanical procedure that takes as input any statement in
234 MATHEMATICAL THOUGHT

the given language. That is to say, our procedure ought to

begin with a simple string of symbols. To positively solve
Hilbert’s decision problem, our procedure must result in
the output ‘true’ whenever the statement (string of symbols)
is true, and it should result in an output ‘false’ whenever
the statement is false.
This was a provocative challenge, which inspired some
brilliant work. Intuitionist matheÂ�maticians, such as L. E. J.
Brouwer, took issue with certain rules of classical logic,
particularly the law of excluded middle. They also
cautioned against the prospect of mathe�matics being domi-
nated by mechanical deductions. After all, people become
mathe�maticians by engaging their imaginations. If we view
the mental life of mathe�maticians as the heart of mathe�
matics, we might follow the intuitionist line, and say that
formal notation, and the mechanical application of rules,
are ‘merely’ imperfect ways of communicating the true
heart of mathe�matics.
In contrast, Hilbert and his fellow formalists emphasized
the proper use of symbols. They argued that the system-
atic, rule-governed use of symbols is central to the meaning
of mathe�matical practice. There are many subtleties and
confusions that fuelled this debate, and I shall return to
the delicate question of the meaningfulness of mathe�matics
in the final chapter. At this point, I just want to stress that
whatever ideas or images we might personally entertain,
the meaning of mathe�matical symbols ultimately depends
on the way that those symbols are actually used. The
philosopher Michael Dummett put this point well in The
Philosophical Basis of Intuitionistic Logic:

An individual cannot communicate what he cannot be

observed to communicate: if an individual associated
with a mathe�matical symbol or formula some mental
content, where the association did not lie in the use
he made of the symbol or formula, then he could
 ALAN TURING AND THE CONCEPT OF COMPUTATION 235

not convey that content by means of the symbol or

formula, for his audience would be unaware of the
association and would have no means of becoming
aware of it. To suppose that there is an ingredient
of meaning which transcends the use that is made
of that which carries the meaning is to suppose
that someone might have learned all that is directly
taught when the language of a mathe�matical theory
is taught to him, and might then behave in every
way like someone who understood the language,
and yet not actually understand it, or understands
it only incorrectly.

Remarkably, these somewhat esoteric arguments in the

philosophy of math have had a very real and substantial
impact on the way we live today. As we shall see in the
following chapter, Hilbert’s decision problem inspired
Alonzo Church (1903–1995) and Alan Turing (1912–1954),
which directly led to the invention of computer programs.
A few years later, this fundamental concept inspired the
invention of real, programmable machines, which were
first used to crack the Nazis’ secret codes. The importance
of the code-breakers of Bletchley Park cannot be overstated,
as their work may well have changed the entire course of
the war. Of course, Turing did not crack the codes single-
handed! Over 10,000 people worked at Bletchley Park,
and no matter how intelligent our experts may have been,
reading the Nazi codes would have been a practical impos-
sibility without some fairly detailed knowledge of the
encryption process.
Fortunately, five weeks before the outbreak of war,
the Polish military intelligence service presented a gift to
their French and English counterparts. At a top secret
meeting in Warsaw, intelligence officers were shown a
replica of the German ‘Enigma’ machine. Polish matheÂ�
maticians had been working on methods of decryption
236 MATHEMATICAL THOUGHT

for several years, and a young Alan Turing quickly

mastered their techniques. He played a crucial role in
developing the British code-breaking routines, and, in
particular, he invented a machine called a ‘Bombe’. By
the end of 1940 these devices enabled the code-breakers
to read all messages sent by the Luftwaffe. Naval codes
were harder to break, but thanks to some captured infor-
mation, naval signals could be decoded for most of the
months of war.
Turing made a number of vital contributions to the war
effort, and we now know that he helped his old logic
teacher Max Newman to develop the world’s first program-
mable, electronic computer (the top-secret ‘Colossus’ of
Bletchley Park). Like most mathe�maticians, Turing pursued
a number of academic interests. In addition to his ground-
breaking work on computation, Turing developed some
remarkable, mathe�matical ideas relating to the growth of
biological forms, which we will return to in Chapter 12.
He was also a superb long-distance runner, and was almost
good enough to compete at the Olympic level. Tragically,
in 1952, he answered some policemen’s questions by telling
them he was gay. Turing didn’t think there was anything
wrong with being gay, but despite having the best character
witnesses imaginable, he only escaped prison by agreeing
to be injected with female hormones. As a result of this
‘treatment’ Turing developed breasts, and in 1954 he died
after biting a cyanide-coated apple.

Depicting Calculation
The modern, mathe�matical concept of computability dates
back to 1936. That was the year that Alonzo Church
developed his ‘lambda-calculus’, while Alan Turing was
devising his own approach to the challenge of finding a
mechanical procedure that identifies true statements
(Hilbert’s decision problem). Intuitively speaking, the
notion of a mechanical or automated procedure seems to
 ALAN TURING AND THE CONCEPT OF COMPUTATION 237

be fairly clear. In this day and age we are used to the idea
of computing machines, and we know that these machines
need a ‘program’. Indeed, everyone knows that computer
programs have significant financial value. The important
historical point is that the mathe� matical concept of a
‘program’ predates real, programmable machines. To make
dramatic progress with Hilbert’s decision problem, matheÂ�
maticians did not need actual computing machines. What
was needed was a form of symbolism that could be used
to represent programs.
Turing made his great advance by imagining a person
‘mindlessly’ carrying out a computational task. For
example, we might imagine a person adding together the
numbers on a list, systematically adding the digits in the
unit column, then the digits in the tens column, and so
on. Such a person starts their procedure with a finite
number of symbols printed on a piece of graph paper,
and they can carry out their task even if they only view
one digit at a time. We can imagine someone placing a
marker on the paper at a predetermined point (the top of
the units column). They then proceed in a step-by-step
fashion, with the marker moving one square at a time.
There are definite rules which the ‘player’ must follow,
and given a particular input, there is always one correct
thing to do. This may involve getting more paper (our
player is assumed to have an unlimited supply), and it
may involve writing some ‘working out’. This ‘working
out’ is an activity that takes place on the sheet of graph
paper, where it is recorded.
Adding a list of numbers is a specific example of a very
general occurrence. The nature of a routine calculation is
that we begin with a finite arrangement of symbols (the
problem), which are then changed one symbol at a time,
until we have another finite arrangement of symbols (the
answer). The crucial point is that in such a routine calcu-
lation, we must specify the one and only ‘correct’ thing
238 MATHEMATICAL THOUGHT

to do at every step. If we have done that, we say that the

person performing the calculation is playing a deterministic
language game.
Unfortunately, mathe� matics itself is often misun-
derstood as a deterministic language game (which it
conclusively cannot be, as we shall see later). In practice,
mathe�matical proofs of any real sophistication cannot
simply be checked by computers, so the activity of most
mathe�maticians cannot be summarized in terms of simply
describable routines. However, individual mathe�matical
procedures or calculations are language games, as mathe�
matical understanding tends to the point where we can
instruct others, and then check in an uncontroversial
manner whether or not a well-described technique has
been implemented properly.

Deterministic Language Games

Turing invented a standardized formal system for describing
deterministic language games. As we shall see, any deter-
ministic language game can be described in this manner,
no matter how strange its rules might be. A language game
that has been described using Turing’s system is usually
called a Turing machine. However, I want to emphasize
the fact that ‘programs’ are conceptually quite separate
from actual computing machines. Indeed, the designers of
modern, digital computers were inspired by their study
of mathe�matical logic: the math came first, and then the
physical machines. For that reason, I will describe the
mathe�matical objects in question as Turing cards, not Turing
machines.
Once again, the key idea is that we can use a pack of
Turing cards to summarize any fixed procedure that
converts an input of symbols into an output of symbols.
Perhaps the most important examples of such procedures
can be found in science, as in many cases we can describe
our scientific theories as a mapping between states of affairs.
 ALAN TURING AND THE CONCEPT OF COMPUTATION 239

Scientists and engineers often use descriptive data and a

well-stated theory (e.g. Newton’s laws) to produce a further
statement: the prediction of the stated theory. For example,
we might input a description of the location, speed and
mass of a star and planet, and output a prediction for the
planet’s orbit. We can describe this process as a rule-based
mapping from one set of symbols to another, so Turing
cards are just the kind of thing we need to summarize our
rule-based process.
The assertion that every deterministic procedure can
be put on a pack of Turing cards is called ‘Church’s
thesis’, and we will examine that important claim in the
following section. First let’s see just how easy it is to use
a pack of Turing cards. As we have seen, a deterministic
language begins with a piece of graph paper covered in
symbols. We can use any finite alphabet of symbols, and
as we only ever look at one symbol at a time, it is helpful
to imagine a movable marker that tells us where to look.
So what about the pack of Turing cards? What does that
look like?
Each card in the pack is labelled with a number, and
each card contains a list of instructions: one instruction
for each of the different symbols that might be written
on the piece of paper. When the game begins, we look
at the first symbol on the page (that is, the one with the
marker on it), and we turn to the card that is sitting at
the top of our pack. We then follow whichever instruc-
tion we are supposed to follow, given the symbol that is
written on the marked square of our graph paper. Turing
cards only ever contain two kinds of instruction. The
first kind says: ‘Erase the symbol on the graph paper,
replace it with the symbol x, and put card number n at
the top of your pack.’ The second kind of instruction
says: ‘Move your marker one square to the right (or one
square to the left, etc.), and put card number n at the top
of your pack.’
240 MATHEMATICAL THOUGHT

The game can come to an end because one of the cards

simply says, ‘Well done, you have finished.’ If the player
is instructed to turn to that card, the game does indeed
finish. Otherwise the player simply carries on following
one instruction after another in a deterministic manner.
The crucial point is that at every step in every deterministic
language game, the correct instruction to follow depends
on the answer to two questions: which card is at the top
of the pack, and what symbol is written on the marked
square?
A simple example of a deterministic language game takes
an input of the following form:

We start with n dots, a space, and then m dots. We begin

this particular game in ‘drawing mode’ (card number one),
which tells the player to move the marker to the right
until it hits the space. Once the marker hits the space, the
player fills in the space with a dot, and then switches to
‘scanning mode’ (card number two). This card tells the
player to move the marker to the right until it reaches
another blank, at which point they are instructed to switch
to ‘erasing mode’ (card number three). This tells the player
to move the marker back one square to the left, so it is
resting over the right-most dot of all. The player is then
instructed to erase this right-most dot, before finally being
told that the game has come to an end. What is left on the
piece of paper is something of the following form:
 ALAN TURING AND THE CONCEPT OF COMPUTATION 241

In other words, I have described a simple program for

addition, as when we are given n dots and m dots, the
game terminates once we have drawn nâ•›+â•›m dots.

In a sense, the heart of Turing’s argument is that any formal

procedure can be described. Such a description is a finite
collection of symbols, and any finite collections of symbols
can be considered as a mathe�matical object. We call these
abstract objects ‘programs’. The crucial point is that by
transforming a verb (the procedure) into a noun (the
program), we can make the intuitive notion of computa-
bility mathe�matically comprehensible.
Turing’s second great insight was to realize that a single
pack of cards can enable a person to carry out any compu-
tational procedure whatsoever! In other words, there
necessarily exists a ‘universal computing machine’, and
such a ‘machine’ only needs to be about as large and compli-
cated as an ordinary pack of cards. This follows because
it is perfectly possible to describe the proper process for
using an arbitrary pack of Turing cards. The rules that
govern the proper use of Turing cards can themselves be
summarized into a single, slender pack, which I shall call
pack U, for universal.
Now, imagine holding this pack U, which provides a
player with all the instructions needed to use any other
pack of Turing cards. If you want to carry out any compu-
tational procedure whatsoever, all you now require is the
appropriate input. For example, suppose that you want
to know what happens when a pack T is used, starting
with input I. You simply need to copy out some symbols
that describe the pack T, together with the input I. The
resulting combination of symbols {T, I} now becomes an
input for our universal pack U. Crucially, we can be certain
that using pack U on the input {T, I} produces exactly
the same output as using pack T on input I. Since pack T
and input I could summarize any deterministic language
242 MATHEMATICAL THOUGHT

game whatsoever,
� our universal pack U really can carry out
any computation we choose. Of course, using a universal
pack U might be an absurdly inefficient way to carry out
the calculation, but if we were given an endless supply of
time and paper, we would eventually write out the correct,
final answer.

Church’s Thesis
The claim that every well-defined, deterministic procedure
can be summarized on a pack of Turing cards is not a
strictly provable claim. The intuitive concept ‘matheÂ�matical
procedure’ is not itself matheÂ�matically defined, in contrast
to the set of all Turing packs. Despite this observation, we
have a very good reason for accepting Alonzo Church’s
thesis. Suppose that we have some kind of technique T,
which generates statements from other statements. For
example, imagine using Pythagoras’ Theorem to calculate
the size of the hypotenuse, given the lengths of the other
two sides as an input. There are two ways that we might
check that our procedure has been carried out correctly:

1. There is an algorithm for calculating the output,

which our technique T correctly follows.
2. There is a theorem that proves that input I corre-
sponds to output O. In this case we can prove
that the given relationship between input I and
output O follows as a logical consequence of our
theorem T.

Both of these things are step-by-step procedures, in which

we look at the input, write out working and look over
what we have already written (accessing information). But
these are precisely the activities that Turing cards are
designed to summarize!
In conclusion, if we follow the instructions on a pack
of Turing cards, and if we have an unlimited supply of
 ALAN TURING AND THE CONCEPT OF COMPUTATION 243

time and paper, we can (in principle) calculate the output

of any deterministic algorithmic whatsoever. Turing cards
can even carry out information processing procedures
where the program itself changes over time. On the other
hand, it is worth noting that modern computers often
perform tasks by responding to inputs while they are
already engaged in a computational procedure, and that
kind of interactive process isn’t well captured by the
formalism I have described.
Theoretical computers have many uses, and in particular,
we can use Turing cards to make decisions of a certain
kind. For example, a computer can effectively decide
whether or not a given integer is prime, and we can use
Turing cards to do the same thing. This idea is very impor-
tant, as we can use it to elucidate the subtle relationship
between mathe�matical truth and mechanical procedures.
That fundamental relationship is the focus of the next
chapter, but before we investigate these fascinating ques-
tions, we must first examine the concept of a ‘decision
problem’.

Decision Problems
In mathe�matics, we often say that one set of objects or
symbols have a given property, while another set of objects
or symbols do not. For example, we can divide the integers
into two sets: the numbers that are even, and the numbers
that are not. As we can represent the integers using ordi-
nary digits, we can also divide strings of digits into a pair
of sets: the strings of digits that represent even numbers,
and the strings of digits that do not.
If some things have a property P while other things do
not, it is natural to wonder how we might set about
answering the question, ‘Does this particular thing have
property P?’ Questions of this form are known as decision
problems. If there is a general, mechanical method that
provides an answer in every case, it must be possible to
244 MATHEMATICAL THOUGHT

summarize that method using a pack of Turing cards. When

using Turing cards to tackle a decision problem, we make
the safe assumption that our pack contains two special,
‘halting’ cards. One of these cards says: ‘Halt – the input
has property P.’ The other halting card says: ‘Halt – the
input does NOT have property P.’ None of the other cards
in the pack will tell us to stop the computation, though
they may instruct us to turn to one of the halting cards.
If a pack of Turing cards can correctly decide whether
or not a string of symbols has a certain property P, we say
that the problem is ‘decidable’. For example, the question
‘Is this integer prime?’ is a decidable problem, because we
only need finitely many axioms and finitely many rules of
deduction to determine whether or not a string of digits
represents a prime number. Because we can settle this
problem using a well-characterized finite system, there must
be a pack of Turing cards that can correctly decide whether
or not any given integer is prime. On the other hand, if a
problem cannot possibly be solved using a pack of Turing
cards, we say that the problem is ‘undecidable’. The first
and one of the most famous undecidable problems is the
halting problem. This problem was posed and analyzed by
Alan Turing in his classic paper ‘On Computable Numbers,
with an Application to the Entscheidungs problem’. He
asked, ‘If I use this pack of Turing cards on that input, will
my computation ever come to halt?’
If the combination does result in a finite computation,
we can definitely find this fact out. That is to say, there
is a well-defined test for every thing of this type, and every
thing of this type passes the appropriate test. In fact, the
test is very simple: we simply use the Turing cards on the
given input, and then check that we are eventually
instructed to halt. On the other hand, some programs
cannot be efficiently predicted. That is to say, in some
cases we can never know that a given computational proce-
dure will never come to a halt.
 ALAN TURING AND THE CONCEPT OF COMPUTATION 245

Turing proved that there is no general method for recog-

nizing non-halting combinations by considering ‘self
reflective’ language games. To understand the proof, the
first observation that we need to make is that given a pack
of Turing cards T, we can consider what happens when we
take, as an input, a standardized description of the pack T.
A language game like this proceeds much like any other
language game, as we use the instructions in pack T to
change the symbols of our description of T. This process
results in our description changing step-by-step into another
collection of symbols. Some packs produce self-reflective
language games that come to a halt, while other packs
produce a self-reflective language game that never halts.
Now suppose that we have a pack of Turing cards that
correctly determine whether or not a given input has some
property P. Recall that such a pack contains two halting
cards: one that says ‘Halt – the input has property P’ and
another that says ‘Halt – the input does NOT have prop-
erty P’. Given such a pack, we can construct a modified
version by removing the second halting card, and replacing
it with a ‘dead-end card’, which simply tells us to repeat
the same, pointless operation over and over again, without
ever moving a different card to the top of the pack. If we
use this modified pack there is only one way that the
computation can ever halt. If the input has property P, we
will eventually reach the card that says ‘Halt – the input
has property P’. As we removed the other halting card,
this is the only way that our computation can ever halt.
We are now ready to hear the final part of Turing’s
analysis. For the sake of argument, he supposed that
there is a pack of Turing cards that can recognize when
any given Turing pack T produces a non-halting, self-
reflecting game. Now, if there was a pack of Turing cards
R that can recognize non-halting, self-reflecting games,
we would find ourselves in the following situation for
every Turing pack T:
246 MATHEMATICAL THOUGHT

Halts if and only if Does NOT halt.

But in that case, what would happen if we used a descrip-

tion of the pack R as our chosen input T?

Halts if and only if Does NOT halt.

We have two copies of the same game, and if the first

version of the game reaches a halting card, the second copy
cannot ever reach a halting card. Similarly, if the first copy
does NOT halt, that very same game must halt!
Our argument has hit a paradox. The question is, where
have we gone wrong? The answer is, we were wrong to
assume that a pack like R exists. In other words, the
preceding argument compels us to conclude that there
cannot be a reliable, automatic method for recognizing
when a given Turing pack T produces a non-halting, self-
reflecting game. Given that we cannot find a general method
for recognizing when a self-reflecting game will halt, we
certainly cannot find a general method for identifying when
a given input and a given pack of Turing cards results in
an unending computation, as any method for identifying
non-halting computations could be used to identify non-
halting self-reflecting games. In other words, Turing’s
argument proves that we cannot always recognize when
a given computational procedure will never come to a halt.
 ALAN TURING AND THE CONCEPT OF COMPUTATION 247

This is a very important observation, as among other things,

it shows that there are problems which cannot be solved
using a finite, pre-stated system.

Figure and Ground

We have seen that a decision problem is a kind of chal-
lenge, where we consider a collection of things, and we
are asked to decide whether or not some arbitrary individual
has a particular property. For example, we might consider
the set of integers, and ask the decision problem ‘Is this
integer prime?’ If we can correctly answer a decision
problem ‘yes’ or ‘no’ by using a pack of Turing cards, we
say that the problem is decidable. Similarly, we say that a
problem is semi-decidable if we can recognize when a
thing has property P, but there is no general method for
recognizing when a thing does NOT have property P.
The existence of semi-decidable problems surprised
many mathe�maticians, who had implicitly assumed that
any definition for a figure necessarily contains the same
information as a definition for the ground (i.e. the points
which are inside the frame of reference, but not inside the
figure).

Some patterns have a constructive definition, and can be

generated by a finite system of rules. The process of gener-
ating a given pattern may resemble counting, in the sense
that it extends to infinity, but every part of the pattern is
248 MATHEMATICAL THOUGHT

completed after a finite amount of time. The somewhat

surprising fact is that the existence of a constructive defini-
tion for a pattern A does not imply the existence of a
corresponding, constructive definition for NOT A. Also
notice that the word NOT is being used in context with
a predetermined sense of thing, as is represented by the
box surrounding the previous drawing. For example, A
might be the set of square integers, which would make
NOT A the set of integers that are not square, rather than
the set of anything whatsoever that isn’t a square number.
Of course, there is an easy and finite test for finding
out whether or not an integer is square. However, things
get much more interesting when we consider the general
case of adding and multiplying an arbitrary string of integer
variables. For example, we can consider the following
examples of ‘Diophantine’ equations, which take an integer
or integers as input, and produce an integer output:

p(x) = x 3 , p(x) = x 2 4 or p(x, y) = 7xy 5 3x 2 y 3 .

Diophantine equations are named after the mathematician

Diophantus (c.â•›210–294), and they are basically just poly-
nomials, except that in a Diophantine equation, all of the
terms must be integers. In the next chapter we will examine
one of the deepest questions in matheÂ� matics: ‘Which
Diophantine equations have integer solutions?’ To get
some sense of why this question has such depth, imagine
trying to devise a method for sorting Diophantine equa-
tions into those that have an integer solution, and those
that do not. If a Diophantine equation has an integer
solution, we can certainly recognize that fact. For example,
p(x) = x 2 4 is equal to zero when xâ•›=â•›2, and we can be
sure that xâ•›=â•›2 is indeed a solution. Furthermore, the
process of finding such an integer, and proving that it is
indeed a solution, is an entirely computable procedure.
Theoretically speaking we can find the solution of any
 ALAN TURING AND THE CONCEPT OF COMPUTATION 249

solvable Diophantine equation just by trial and error: we

can simply try every integer in turn, as proving that a
putative solution is indeed a solution only requires the
ability to add and multiply integers correctly.
Since prehistoric times, mathe�maticians have explored
the solutions to polynomial equations. For example, the
Ancient Babylonians knew how to solve quadratic equa-
tions, and across the globe, over three or four millennia,
many other sophisticated techniques have been developed.
At this point, the important thing to note is that if a solv-
able equation has a solution, the solution itself stands as
proof that the equation has a solution. In contrast, consider
an equation that does not have a solution. What kind of
proof might we provide to show that the given equation
has no solution? What kind of statement can stand witness
to this kind of truth?

Semi-Decidable Problems
Recall that a property P is said to be ‘decidable over a set
S’ if it is theoretically possible to calculate whether or not
any given member of S has the property P. For example,
we say that the property of being even is decidable over
the integers because a single finite program can be used
to determine whether or not an arbitrary integer is even.
More interestingly, a property P is said to be semi-decidable
over a set S if given any member of S that has the property
P, there is a finite calculation that confirms the fact that
the given member of S does indeed have the property P.
For example, the property of ‘having an integer solution’
is semi-decidable over the set of all Diophantine equations,
because it is always possible to prove that a solvable equa-
tion does indeed have a solution.
Some problems are semi-decidable but not decidable. For
example, if we know how to use a pack of Turing cards,
we can certainly recognize that a given input and a given
pack ultimately produce a halting computation. However,
250 MATHEMATICAL THOUGHT

as Turing proved, there is no finite program for recognizing

when a given calculation will never halt. Remarkably, every
problem that is semi-decidable but not decidable has to
have certain characteristic features. To appreciate one of the
characteristic features of semi-decidable problems, imagine
that we have a property P and a machine that says ‘yes’ if
and only if we input a thing with property P. Furthermore,
suppose that we can look at any input I and come up with
a number t(I), such that if the machine is going to answer
‘yes’, we can be certain that it will do so in less than t(I)
units of time. In that case we could take any input I, calcu-
late the corresponding number t(I), plug our input into the
machine and wait for one of two things to happen:

1. The machine recognizes that the input I has the

property P.
2. A period of time t(I) passes, and our machine still
hasn’t said anything.

In the second case we know that our input does NOT

have property P, simply because the calculation has taken
too long. This observation tells us that the property in
question must be decidable, because we can recognize the
absence of P by the silence of our machine. This means
that if a property P is semi-decidable but not decidable,
there cannot be a way of finding a safe overestimate for
the time required to confirm that a given input has the
property P. By definition of semi-decidability we will even-
tually be able to confirm that a given input has property
P, but in general we cannot know how many computational
steps will be required to establish this fact.
Another necessary property of semi-decidable sets is
the following. Suppose that our set of things can be written
out as a particular list. For example, imagine listing the
integers in ascending order. If we can move along this list
ticking off EVERY thing with property P, then EVERY
 ALAN TURING AND THE CONCEPT OF COMPUTATION 251

thing we pass over without ticking must, by definition,

NOT have property P.

If property P is decidable, we can systematically move

along our list ticking the things that have the property P.
Conversely, if we can systematically tick off elements in
the order they appear in some list, then property P must
be decidable. This follows because if we could tick off the
things with property P in order, we would have a method
for recognizing things that do NOT have property P,
namely looking for things that have not been ticked, but
which are earlier in the list than something that has been
ticked. In other words, if a property P is semi-decidable
but not decidable, then for every list and every method
for ticking things with property P, our ticking pencil must
hop around an infinite amount, moving backwards as well
as forward.
Returning to a particular example of a semi-decidable
problem, let’s consider Turing’s halting problem. In that
example of a decision problem, we ask: ‘If I use these
Turing cards on that input, will my computation ever halt?’
In terms of logical truth, we are inclined to say that an
input/Turing card combination either halts or does NOT
halt. Furthermore, if an input/Turing card combination
halts we can prove that it halts, simply by carrying out
the calculation. On the other hand, just because we cannot
prove that a given calculation halts, that doesn’t necessarily
imply the existence of a proof that the given calculation
never halts. In short, there are cases where we cannot prove
a statement that we believe to be true.
Perceiving this gulf between the formally demonstrable
and the statements we accept as truths leads us to consider
252 MATHEMATICAL THOUGHT

the following question. What is the minimum difference

between a justification for:

1. Our belief in the truth of a statement that our

system of logic can prove, and
2. Our belief in the truth of a statement that our
system of logic cannot prove.

If we could show that there is always a difference between

such justifications, we might come to the conclusion that
we should drop the concept of unprovable truth altogether.
In other words, could it be the case that only provable
truths are really true?
 Chapter 11:
KURT GÖDEL AND THE
POWER OF POLYNOMIALS
‘The utility of systems lies not merely with their
making us think about something in an ordered way
according to a particular scheme, but in making us
think about it at all; the latter utility is incontestably
greater than the former.’
Georg Christoph Lichtenberg, 1742–1799

Matiyasevich’s Theorem
Before we return to the concept of unprovable truth,
we  need to sketch out a beautiful proof of Gödel’s
Incompleteness Theorems. More specifically, we shall prove
Gödel’s famous theorems by means of another result of
staggering significance, called Matiyasevich’s Theorem
(also known as the MPDR Theorem). This less well-
known theorem has its roots in the 1960s, and the work
of the American mathe�maticians Julia Robinson, Martin
Davis and Hilary Putnam. In effect, they made progress
by applying modern mathe�matical logic to the study of
Diophantine equations. More specifically, they were inter-
ested in the logical properties of Diophantine definitions.
A Diophantine definition is simply a standard way to
define a set of integers, by using a polynomial equation.
254 MATHEMATICAL THOUGHT

For example, the even numbers can be defined as the set

of integers n for which the following statement is true:

‘For SOME integer m, nâ•›−â•›2mâ•›=â•›0.’

Similarly, the square numbers can be defined as the set of

integers n where:

‘For SOME integer m, nâ•›−â•›m2â•›=â•›0.’

As I have already mentioned, if someone claims to have

found some integers that solve an equation P, it is very
easy to put that claim to the test. All you need is a defini-
tion of the equation P, plus a few simple rules that tell
you how to add, subtract and multiply. In fact, all we
really need are the following five axioms (due to Giuseppe
Peano, 1858–1932):

nâ•›+â•›(-n)â•›=â•›0, nâ•›+â•›0â•›=â•›n, nâ•›+â•›(mâ•›+â•›1)â•›=â•›(nâ•›+â•›m)â•›+â•›1,

n 0 = 0 and n (m + 1) = (n m) + n.

Now, suppose that we are considering some particular

Diophantine equation P. If there is a list of integers
n, m1, m2, ... , mk that solve the equation P, we can definitely
find that list in a finite amount of time. Even if we don’t
have a clever method for finding a solution, we could
simply try every possible combination of integers in turn.
The number of steps in this trial-and-error process might
be vastly greater than the number of electrons in the
universe, but so long as it is finite, we say that the proce-
dure is computable.
The logical structure of Diophantine definitions is admi-
rably clear, because in a more or less efficient way we can
set about looking for solutions of the form n, m1,m2 , ... , mkâ•›,
and in the case where such a solution exists, we know by
 KURT GÖDEL AND THE POWER OF POLYNOMIALS255

definition that the integer n has the property P. The inte-

gers m1,m2 , ... , mk are somewhat analogous to the working
out that a Turing pack might instruct us to write: they are
an essential part the calculation, but they aren’t part of
the final answer.
Up until 1970, only a few mathe�maticians suspected that
Diophantine definitions might be every bit as powerful as
Turing cards. In other words, only a few people guessed that
the addition and multiplication of integers might literally
generate the entire computable universe. The proof that
Diophantine equations are a kind of universal programming
language proceeded in a number of steps. By 1970, recursion
theorists could prove that Diophantine sets had almost all of
the necessary properties. In particular, Julia Robinson, Martin
Davis and Hilary Putnam had established the following facts:

1. Every finite list a, b, … , z has a Diophantine

definition. It is of the form (aâ•›−â•›n)(bâ•›−â•›n) …
(zâ•›−â•›n)â•›=â•›0, so the number n belongs on the list if
and only if it is one of the numbers a, b, … , z.
2. If the sets A and B have Diophantine definitions, so
does A OR B. If P(n, x)â•›=â•›0 and Q(n, y)â•›=â•›0 are the
respective Diophantine definitions for the sets A and
B, then P(n, x) Q(n, y) = 0 is a Diophantine defini-
tion for A OR B. For example, (n 2x)(n y 2 ) = 0
picks out the integers n, which are either even or
square.
3. Similarly, P(n, x) 2 + Q(n, y) 2 = 0 is a Diophantine
definition for A AND B.

Only one remaining property (called bounded universal

quantification) remained elusive. Recursion theorists real-
ized that if they could only establish the existence of a
single Diophantine set whose members grew exponentially
(like 1, 10, 100, 1â•›000, …), then every definable set would
necessarily have a Diophantine definition.
256 MATHEMATICAL THOUGHT

Most people thought that no such set could exist, but

on 4 January 1970, a twenty-two-year-old Russian mathe�
matician by the name of Yuri Matiyasevich proved the
doubters wrong. The complete proof of Matiyasevich’s
Theorem is much too involved for a book of this size or
nature, but roughly speaking his argument can be divided
into two parts. First, a parametric version of the Fibonacci
sequence is shown to have a wide range of properties,
each of which has a Diophantine definition. Second, he
showed that the combination of these Diophantine proper-
ties suffice to specify completely the exponential sequence
in question.
The fact that the prime numbers have a Diophantine
definition was a considerable surprise to many mathe�
maticians, but it really is a provable fact that multi-variable
polynomials have such a wide range of different solution
sets, that every definable set of integers has a corresponding
polynomial. Furthermore, as there are no major difficulties
in transliterating from one alphabet to another, we can
translate any collection of symbols into a sequence of digits.
This implies that every definable set of symbols has a
Diophantine definition. It is literally true to say that if
there is a finite, deterministic rule for generating a sequence
of symbols (even an infinite sequence of symbols), then
there must be a rule that generates those same symbols
simply by adding and multiplying.
A remarkable implication of this result can be found
by considering a ‘universal decision pack’. A player using
such a pack receives descriptions of decision packs as inputs,
together with a particular case that might or might not
have the property in question. The universal decision pack
reads the input pack P together with the particular input
N, and works to recognize whether or not the input N
has the property that is recognized by the decision pack
P. Since it does this in an entirely deterministic manner,
we have a well-defined subset of the set of all pack-input
 KURT GÖDEL AND THE POWER OF POLYNOMIALS257

pairs, namely those where the input does indeed have the
property encoded by the pack.
Given any particular alphabet that we might be using,
every possible input of symbols can be assigned a unique,
identifying number, because we can list all the possible
inputs alphabetically. I shall use the symbol n to denote
the integer that identifies a given input. Similarly, every
property can be assigned a unique number, because we
can list our descriptions of decision packs in alphabetical
order. I shall use the symbol p to denote the integer that
identifies a given decision pack. Matiyasevich’s Theorem
implies that for every universal decision pack, there must
be an equivalent, Diophantine equation U(n, p, x). In other
words, the polynomial ‘U(n, p, x)â•›=â•›0’ has a solution if and
only if the input n has the property p.
The polynomial U is just an ordinary polynomial,
involving nothing more complicated than the addition and
multiplication of some integer-valued constants with two
integer-valued parameters (n and p) and some fixed number
of integer-valued variables, which I have denoted x. They
tend to look rather messy when written out in full, but
the crucial point is that equations like U actually exist,
and they are called universal equations because by altering
a single, integer-valued parameter p we can use the equa-
tion U to generate every definable set of integers. In other
words, the solution sets of a universal equation literally
contain the entire computable universe!
If we are oblivious to calculation time, it does not take
much to attain the greatest level of computational power.
Even the basic rules of addition and multiplication can
ultimately generate any pattern whatsoever, provided that
the pattern can be computed in a deterministic or mechan-
ical manner. Also notice that the patterns generated by a
universal pack of Turing cards, or a universal equation,
are arbitrarily complex. That is to say, because the system
is universal, it can replicate the simplest patterns, the most
258 MATHEMATICAL THOUGHT

complicated patterns, and everything in between. A similar

observation holds for every other quality that a definable
list of numbers can be said to possess.

Kurt Gödel
Kurt Gödel (1906–1978) was born in Vienna, and when
he was six years old he contracted rheumatic fever. He
recovered to full health, but when he was eight he began
reading medical texts on the illness he had suffered, and
became convinced that he had a weak heart. This unfounded
suspicion was the beginning of a lifelong obsession with
his health; an obsession that became particularly problem-
atic when Gödel concluded that food was fraught with
danger, so people should eat as little as possible. As a
student at the University of Vienna, he was persuaded to
move from physics to mathe�matical logic by the gifted
Philip Furtwängler.
Furtwängler had to dictate everything he wrote, because
he was paralyzed from the neck down. The young Gödel
was greatly impressed by the image of a mind exploring
the world of numbers, unhindered by the lack of a working
body. A lifelong believer in a supernatural creator, Gödel
was also a fervent Platonist. In other words, he believed
that abstract, mathe�matical objects have a very real exist-
ence, independent from mathe�maticians, or the languages
that mathe�maticians can use. He was by all accounts a
brilliant and fastidious man, and he seems to have been
one of those people who are drawn to math not only
because of the intellectual challenge, but also because of
the unworldly, timeless and uncontaminated character of
its objects of study.
Gödel was disgusted by the Nazis, but even as war was
looming, he did not want to flee his native Vienna. When
the Second World War finally broke out, Gödel was judged
fit for military service, and he greatly feared being
conscripted into the German army. He finally decided to
 KURT GÖDEL AND THE POWER OF POLYNOMIALS259

flee Austria, and in 1940 he arrived in America. As a world

famous mathematician he was offered a prestigious post
at the Institute for Advanced Study in Princeton. Here he
became close friends with a man he had first met in 1933:
Albert Einstein. In a letter to his mother, Gödel wrote
that the two men would meet daily at Einstein’s house,
leaving at ten or eleven in the morning, and walking for
half an hour until they reached the institute. They would
then meet up again at one or two, before walking home
together, chatting in their native German. It is clear that
both men enjoyed their discussions on politics, philosophy
and physics, and towards the end of his life Einstein even
remarked that he only bothered going to his office ‘to
have the privilege of walking home with Kurt Gödel’.
Gödel was always drawn to problems of philosophical
interest, and in 1949 he made a remarkable foray into the
theory of relativity. More specifically, he showed that there
are solutions to Einstein’s equations that contain closed
loops. In other words, he showed that if you have enough
energy, time travel becomes a theoretical possibility! Gödel
was greatly saddened when his friend Einstein died in
1955, and as he grew older, Gödel became increasingly
concerned about his own physical condition. When his
beloved wife Adele suffered some serious health problems,
his difficulties with food and paranoia over poisoning
became even more acute. By the end of his life he was so
frightened of being poisoned that he refused to eat at all,
and in 1978 he effectively starved himself to death.
Although he made many important contributions to
matheÂ�matics and logic, Gödel is most famous for proving an
intellectual bombshell: the Incompleteness of Arithmetic. A
formal system is said to be complete if every grammatic�
ally correct statement in the given language can either be
proved or disproved from the axioms of the system. For
example, if we restrict the language of arithmetic by refusing
to use the words ‘EVERY’ and ‘SOME’, the Â�fragment that
260 MATHEMATICAL THOUGHT

remains is in fact complete. In other words, the few mathe�

matical statements that we are left with can all be proved
or disproved in a simple, mechanical manner, using a small
number of fundamental axioms, together with some basic
rules of deduction. As we shall see, the situation is much
richer and more interesting when we include the words
‘EVERY’ and ‘SOME’ into the vocabulary of math.

Searching for Solutions

Imagine a nation of people who prize the study of Diophantine
equations above all else. In the middle of their kingdom
stands an imposing stone monolith, covered in mathe�matics.
If the people knew that a statement was true of the integers,
they might carve it onto this monolith. Over the generations
many statements had been added, but the most revered carv-
ings were the laws for the addition and multiplication of
integers (e.g. Peano’s axioms), together with the laws of logic.
Other truths might also be added, but these people wouldn’t
carve any old statement onto their sacred stone.
When two men wanted to marry the same woman, she
would pick a Diophantine equation, and assign one of the
men ‘solvable’, and the other ‘NOT solvable’. The suitor
who was assigned ‘solvable’ would immediately set about
trying to find a solution, using the sacred instructions for
adding and multiplying integers. The suitor who was
assigned ‘NOT solvable’ had a somewhat more interesting
task.
It is easy to prove that some equations have no integer
solution. For example, x 2 + 1 = 0 has no integer solution,
because for every integer x, x2 is at least as big as zero.
The monolith would not need many axioms to be powerful
enough to prove that there is no integer n such that
x 2 + 1 = 0.â•› Other Diophantine equations require more
sophisticated methods to prove that they have no solution.
When a suitor thinks that the given equation has no
solution, he has to find something on the monolith that
 KURT GÖDEL AND THE POWER OF POLYNOMIALS261

tells him that his argument is valid (e.g. the monolith ought
to confirm the validity of the preceding argument). If
nothing on the monolith confirms the validity of a suitor’s
argument, the only remaining option is to persuade the
King to carve another statement onto the monolith. This
was considered a very serious business, because if the King
allowed inconsistent statements onto the monolith, the
logic sanctioned by the sacred stone would be utterly
useless. Among other problems, the people would face the
embarrassing situation in which both suitors would invar-
iably win the right to marry.
One day, the King’s daughter (an exceptional beauty
named Helen) was compelled to pick a Diophantine equa-
tion. It took her ten minutes to dictate the monstrous
thing, and the poor suitors were quite ashen by the time
she assigned them their respective tasks. The young man
Kurt was assigned the ‘NOT solvable’ task, while Bill got
‘solvable’. Bill was a bit of a computer whiz, and by the
end of the day he had checked that there were no solutions
with less than a billion digits. Meanwhile, Kurt grabbed a
pencil and tried to look for a pattern …

The Incompleteness of Arithmetic

Recall that axiomatic systems and the laws of logic can be
summarized on a computer program, and hence they can
also be summarized in Diophantine form. The fundamental
proof theoretic fact is that we can take any set of axioms,
or any countable axiom scheme with a finite definition,
and devise a program that finds each logical consequence
in turn. In particular, we can use such a system to find
proofs that might be used by a suitor whose task is to
demonstrate that a given equation doesn’t have any solu-
tions. Assuming that no extra axioms are carved onto the
monolith, this is a perfectly mechanical procedure.
Therefore Church’s thesis implies that we can summarize
this procedure using a pack of Turing cards.
262 MATHEMATICAL THOUGHT

Furthermore, we can list all the Diophantine equations

in alphabetical order, which is to say that we can associate
each equation with a unique number. Some numbers will
correspond to equations that can be mechanically proved
to be unsolvable. For example, we might use the monolith
to prove that x 2 + 1 = 0 has no integer solution, x 2 + 2 = 0
has no integer solution, and so on. Each of these equations
has a unique, identifying number, so our mechanical proce-
dure identifies a subset of the integers, namely those
integers that correspond to an equation p where we can
mechanically prove that p has no solution. Matiyasevich’s
Theorem implies that this subset of the integers must have
a Diophantine definition. With that in mind, we are ready
to return to the story of Kurt and the King’s daughter.
Kurt soon realized that Helen’s equation was related to
a universal Diophantine equation, and he spent many months
studying it, considering things that the integers might encode.
He was convinced that the King’s daughter (who was herself
a formidable mathematician) must have picked those
numbers for a reason. The more he looked, the more he
found, and eventually he picked out substructures that
encoded everything on the village monolith. The final break-
through came when he tried converting one of the constants
into a variable. He immediately recognized the equation
that this produced, because it captured the property of being
‘self-reflectively unsolvable’.
An integer n is said to be self-reflectively unsolvable
for U if and only if you can use the village monolith to
prove that U(n, n, x)â•›=â•›0 has no integer solutions. For
example, if U(1, 1, x)â•›=â•› x 2 + 1, we would say that 1 is self-
reflectively unsolvable for U (assuming that our axiom
system can prove that x 2 + 1 = 0 has no integer solution).
Notice that there are two facts involved. The first fact is
that when we set the two parameters of our universal
equation equal to 1, the resulting equation has no integer
solution. The second fact is that we can prove that this
 KURT GÖDEL AND THE POWER OF POLYNOMIALS263

equation has no integer solution just by using the axioms

on the monolith. The crucial point is that the formal proce-
dure that demonstrates this second fact can be encoded
into a pack of Turing cards, or into a Diophantine equation.
Kurt could hardly contain his excitement as he calculated
exactly which value for n Helen had used to produce her
equation. As he examined this enormous number, a sense
of déjà vu crept over him. A quick calculation confirmed
his suspicions: the number Helen had used encoded a
Diophantine equation which exactly mirrored the property
he had just been studying!
The point of this story is that Kurt is looking at an
equation that necessarily exists. That is to say, for every
system of axioms (which correctly adds and multiplies
integers), there is some Diophantine equation U and some
integer T with the following, remarkable property:

For every integer n,

1. U(T, n, x)â•›=â•›0 has a solution if and only if

2. The monolith is powerful enough to prove that
U(n, n, x)â•›=â•›0 has no solutions.

To repeat, we can be certain that the two statements above

are logically equivalent, and this is the heart of Gödel’s
Incompleteness Theorems. We know that it is true because
given the axiomatic system in play, there must be some
integer T that corresponds to a program for finding proofs
of the following kind of statement: ‘U(n, n, x)â•›=â•›0 has no
solution.’ Note that n is some integer which we are given,
while x is a finite string of integer-valued variables. Helen
could have picked any integer n to plug into the equation
U, but she decided to use T, the integer which corresponds
to the procedure for finding proofs of statements of the
form ‘U(n, n, x)â•›=â•›0 has no solution’.
Kurt took a deep breath, and considered what would
264 MATHEMATICAL THOUGHT

happen if statements 1 and 2 were true for the particular

case when nâ•›=â•›T. Statement 1 would tell us that U(T, T,
x)â•›=â•›0 has a solution, so we should be able to prove this
‘fact’ using our axioms for addition and multiplication.
The logically equivalent statement 2 tells us that we can
prove that U(T, T, x)â•›=â•›0 does NOT have a solution. In
other words, if Helen’s equation U(T, T, x)â•›=â•›0 has a solu-
tion, then the monolith must be inconsistent.
Returning to statements 1 and 2, if we put nâ•›=â•›T there
are two conceivable cases: U(T, T, x)â•›=â•›0 either has a
solution, or it doesn’t. If we suppose that U(T, T, x)â•›=â•›0
has a solution, then our PC logic machine will be able
to use the monolith’s axioms to generate the following
statements:

1. U(T, T, x)â•›=â•›0 has an integer solution

and
2. The monolith is powerful enough to prove that
U(T, T, x)â•›=â•›0 has no integer solution.

In this case we would have conclusive proof that the axioms

on the monolith are inconsistent, as we would be able to
prove that U(T, T, x)â•›=â•›0 has a solution AND U(T, T, x)â•›=â•›0
does NOT have a solution. But what if U(T, T, x)â•›=â•›0 does
not have a solution? In other words, what happens when
statements 1 and 2 are false? In that case we manage to
avoid paradox by observing the gap between truth and
proof:

NOT 1 U(T, T, x)â•›=â•›0 has no solution

and
NOT 2 The monolith is not powerful enough to
prove that U(T, T, x)â•›=â•›0 has no solutions.

If the monolith is indeed consistent, then Kurt will have to

have something added to it before he can complete his task.
 KURT GÖDEL AND THE POWER OF POLYNOMIALS265

He can be quite certain that in its current state, he cannot

use it to prove that Helen’s equation has no solution.
As things stand, Kurt is quite certain that he cannot
complete his allotted task. Indeed, for every consistent
set of axioms there are infinitely many true arithmetic
statements that we cannot prove using nothing but the
given axioms. That is what is meant by the Incompleteness
of Arithmetic.
Once he had completed this argument, Kurt headed
straight to the palace. As he waited for an audience with
the King, he wondered how to handle his peculiar predic-
ament. What axiom would he need to carve before he
could complete his task? And how would he persuade the
King to add it to the monolith?

Kurt: Tell me, your majesty, do you think that your

daughter should marry both her suitors?

King: How dare you suggest such a scandalous thing! I

should have you flogged for asking such an impertinent
question.

Kurt: I do apologize, I meant no offence. I agree that the

monolith is entirely consistent, and such a thing is quite
impossible. Unfortunately, I can’t find anything on the
monolith that tells me that.

King: What kind of madness is this? You don’t need the

monolith to tell you that its axioms are consistent. If my
ancestors so much as suspected that one of the axioms was
not consistent with the others, they would never have
allowed it to be carved onto the monolith.

Kurt: Does that mean I am allowed to make that assump-

tion, in my proof of unsolvability?
266 MATHEMATICAL THOUGHT

King: I don’t see why not. Every proof implicitly assumes

that the monolith is consistent – that assumption is the
very basis for our belief in the truth of the things we prove.
After all, if our axioms are inconsistent, then our theories
are worth no more than any other sentence! It is decided
then – I hereby decree, by royal command, that you may
assume that one and only one person has the right to
marry my daughter.

Kurt: In that case you had better fetch her. If she has to
marry one and only one person, then that person must
be me.

Truth, Proof and Consistency

People have always argued about the nature of truth, but
in some ways everybody understands the meaning of the
word. People say that a statement is ‘true’ if and only if
our understanding of the statement accords with our under-
standing of the subject matter or situation that the statement
is about. Of course, we sometimes think that a statement
is true, and then realize it is not. We therefore understand
that a statement is only true if it can withstand every kind
of relevant scrutiny, and we generally don’t know all the
different forms of scrutiny that could possibly show that
our statement is false. That is why it is hard to know when
a statement is really true, and why a ‘true’ statement is
more than a just a well justified statement. I also think it
is worth emphasizing that reality is present whether we
can talk about it or not, but ‘truth’ is fundamentally a
relationship between language and the world. Crucially, it
seems reasonable to say that when we subject potentially
true statements to every relevant form of scrutiny, we need
only consider the kinds of scrutiny that might be brought
to bear by people (real or idealized) who understand the
language that we use when we make our statements.
In the case of mathe�matics our choice of axioms largely
 KURT GÖDEL AND THE POWER OF POLYNOMIALS267

determines the meaning of our symbols. For example, if

I make a provable statement using the language of axiomatic
set theory, it doesn’t make sense to say that my statement
is actually false simply because you refuse to accept one
of my axioms. The axioms are part of what makes my
symbols mean what they mean, and if you don’t accept
my statement simply because you don’t accept one of my
axioms, that does not show that my statement (as originally
conceived) is actually false. It simply shows that you under-
stand my symbols differently from me, and at best your
argument might persuade me that it is a good idea to adopt
some new axioms. In that sense there really are mathe�
matical statements that are true by definition. Likewise,
we are the ones who determine the rules of chess, and we
really can know for certain that you cannot force check-
mate against a lone king with only a king and a pair of
knights. This kind of truth by definition is what makes it
possible to construct proofs in the first place, but that is
not the only kind of mathe�matical truth!
For example, I do not think that ‘1â•›+â•›1â•›=â•›2’ is merely
true by definition. It is also true in the usual, less certain
sense, as we can understand that this statement fits with
the way that we understand the relevant subject matters,
which are the concepts of number and addition. After all,
if you have a single object in your collection and you add
another object, you really do have two objects in your
collection. Likewise, if you take one step forward and then
take another step forward, you really have taken two steps
forward. These deeply understandable experiences are part
and parcel of the concepts of number and addition, and
you don’t understand my language if you don’t understand
that fact. We can see that the statement ‘1â•›+â•›1â•›=â•›2’ fits with
our common understanding of the concepts of number
and addition, and in that sense we can see that it is true.
Similarly, the statement ‘nâ•›+â•›0â•›=â•›n’ is true in the sense that
if I take n steps forward and then don’t take any more
268 MATHEMATICAL THOUGHT

steps, I have in fact taken n steps forward. It is also true

by definition, as this statement is one of the axioms
governing the proper use of the symbol ‘0’.
Of course, most mathe�matical statements are far too
complex for us to assess in such an intuitive manner.
Nevertheless, it is this kind of truth, or fidelity to an
underlying concept, that leads us to accept certain axioms
in the first place, making proof and ‘truth by definition’
possible. Mathematicians need to employ rule-governed
systems in order to do their work, but the question then
arises as to why we should trust those systems once they
start to outstrip our intuitions. In particular, how can we
be sure that our rule-governed, symbolic systems really
are consistent, and cannot lead us to contradict something
that we ought to accept as true? It is these mysteries that
Gödel helped to elucidate, and his insights came as a
surprise.
The first point to make is that some mathe�matical systems
really can prove their own consistency. For example, the
axioms of classical logic can be used to prove the consist-
ency of classical logic. That is to say, there is no way that
we can start with a statement A, apply the laws of logic,
and reach the conclusion NOT A. Some people argue that
there are better systems of logic than the classical one, but
there is no question that the language of AND, OR and
NOT is logically consistent.
On the other hand, arithmetic cannot prove its own
consistency. We can prove that fact because we know
that for every formal system capable of adding and multi-
plying integers, there is an equation that corresponds to
U(T, T, x)â•›=â•›0. In other words, for every axiom system
there is an equation that has a solution if and only if we
can prove that the equation has no solution. This means
that if our axiom system is consistent, the equation cannot
have a solution. If our formal system could prove its own
consistency, we could also prove that the equation U(T,
 KURT GÖDEL AND THE POWER OF POLYNOMIALS269

T, x)â•›=â•›0 has no solutions. But if we can prove that it does

not have solution, we know that it must have a solution,
which is a fact that we can prove! In other words, arith-
metic axioms that can prove their own consistency must
be inconsistent.
In light of this fact, we must be exceptionally careful
with formal assertions that are equivalent to the King’s
decree. We can claim that our formal system is consistent,
but we cannot make that claim an explicit feature of the
formal system itself. However, there is at least one way in
which we can consistently increase the power of the mono-
lith in light of the insight provided by Kurt’s formal
argument. At the very least, we can carve the following
statement onto the monolith: ‘U(T, T, x)â•›=â•›0 does NOT
have a solution.’
Adding this statement enables us to prove at least one
new thing (namely the statement itself). Furthermore, if
our new expanded system is inconsistent, then so was our
old system, because if you can prove that ‘U(T, T, x)â•›=â•›0
has a solution’, you not only contradict our new axiom,
you also demonstrate an inconsistency in the original axiom
system. The crucial point is that we can extend our axiom
system by adding the statement ‘U(T, T, x)â•›=â•›0 does NOT
have a solution’, and we might have a good reason for
wanting to do that. In contrast, we cannot consistently
extend our axiom system by adding the statement ‘U(T,
T, x)â•›=â•›0 has a solution’. If we accept that our original
system was consistent, then as members of the logical
community we must also accept that ‘U(T, T, x)â•›=â•›0 does
NOT have a solution’, even though we cannot prove that
statement without adding a new axiom.
Once we have increased the monolith’s power in this
fashion, the set of provably unsolvable Diophantine equa-
tions becomes slightly larger. The power of an axiom system
can always be increased further in this manner, but this
procedure is not itself computable. Each step requires a
270 MATHEMATICAL THOUGHT

further insight, as we must reconstruct Gödel’s argument

in the context of a new set of axioms. Also note that we
can have good reasons for adding such a statement to our
axioms even though this statement is not a self-evident
truth. The evidence or justification for this truth is found
in our comprehension of the other axioms, together with
Gödel’s argument (the context of our statement).
Our statement is logically independent from its context
(it is saying something new), but in some sense it is a
natural extension of the prior system. We can see that
‘U(T, T, x)â•›=â•›0 does NOT have a solution’ is true by
following the previous arguments, and by understanding
the axioms on the monolith. On the other hand, a mechan-
ical application of those same axioms cannot recognize
this truth. That is to say, if Kurt had used a PC machine
instead of his intellect, he could not have completed his
task. This is what I meant when I said that mathe�matics
is not a deterministic language game. However, it is not
clear that this way of extending our axioms requires an
inspiration that is beyond any finite description of a formal
methodology. We can’t use a single program to complete
our set of axioms, but we may be able to give strategic
advice that covers the relevant cases. By way of analogy,
a person may know no numbers but the integers, and using
the language of integers we can ask such a person ‘if 2xâ•›=â•›1,
what is x equal to?’ To answer this question we need an
expanded language that incorporates fractions, but this
question can show the sense in extending our language,
and persuade our audience that fractions make sense.
We could summarize Gödel’s achievement by saying
that he had the brilliant idea of using a very formal, mechan-
ical proof to show the limits of formal, mechanical proofs.
Make no mistake: he did not discover some mysterious
truth that formal language can never articulate. Similarly,
he did not prove that there are arguments that a human
being can follow but which no computer could ever
 KURT GÖDEL AND THE POWER OF POLYNOMIALS271

construct. What he showed is that for any particular formal

system (such as a computer might use) there is a Diophantine
equation with no integer solution, though the given
computer program cannot prove that this equation has no
integer solution. Another computer might be perfectly
capable of proving that the equation has no solution, but
in that case, there must be a second unsolvable Diophantine
equation, which falls through the web of proof for our
second computer. No matter how many formal principles
we cobble together, there will always be some Diophantine
equation whose unsolvability we cannot prove.
In an earlier section I described the incompleteness of
arithmetic as an intellectual bombshell. The grand project
that it really blew apart was known as ‘Hilbert’s program’:
the dream that all of mathe�matics could be formalized
within a finite, provably consistent axiomatic system.
Perhaps the most extreme example of logically explicit
matheÂ�matics was Russell and Whitehead’s Principia
Mathematica (published in 1910, 1912 and 1913), which
dotted every ‘i’ and crossed every ‘t’ for the bulk of ordi-
nary mathe�matics. Indeed, the authors were so careful to
spell out every logical assumption that it took them 362
pages to build up enough machinery to reach the conclu-
sion that ‘1â•›+â•›1â•›=â•›2’!
The logical system employed by Russell and Whitehead
was painstakingly spelled out, and in a sense their book
was a precursor to modern programming languages.
Everyone, including Gödel, was confident that all of their
conclusions were correct. The problem was that Gödel
proved that the logical system specified at the beginning
of the book couldn’t possibly prove every theorem that
the authors hoped it might prove. What is worse, he showed
that no amount of tinkering with the axioms would help:
there will never be a complete, definitive list of everything
you need to know in order to prove every arithmetic state-
ment that we think is true.
272 MATHEMATICAL THOUGHT

At this point, we need to reconsider the statement ‘Only

provable truths are really true’. We believe in provable
truths because we have good reason to trust our axioms
and our systems of argument. In particular, if we recognize
that a statement is true because of its proof, there is a
foundational belief in the consistency of our axioms, as
inconsistent axioms are profoundly useless. However, if
we formally state our implicit belief in the consistency of
our axioms, we produce an unprovable statement. In this
sense, if it isn’t ‘really true’ that our axioms are consistent
(because we can’t prove it), then our provable truths are
not ‘really true’ either!
 Chapter 12:
MODELLING THE WORLD
‘Models in [the life sciences] are not meant to be
descriptions, pathetic descriptions, of Nature; they
are designed to be accurate descriptions of our pathetic
thinking about Nature. … They are meant to expose
assumptions, define expectations and help us devise
new tests.’
James Black, 1924–2010

Science and the Uses of Models

Science is a complicated activity, involving the collection
of data and the critical evaluation of descriptions of events.
It involves a disparate patchwork of methodologies, not
some single ‘scientific method’, but all scientific claims are
made to be tested, and we are most interested in general
explanations, not mere descriptions of specific cases.
Scientific laws are an essential part of science (and they
are particularly important in physics), but you can certainly
work as a scientist without looking for new laws. Many
people over the centuries have been astounded by the fact
that scientific laws are mathe�matical in form, as what does
‘pure thought’ have to do with messy, contingent reality?
The extraordinary success of the scientific enterprise is
indeed remarkable, but the fact that science tends to become
274 MATHEMATICAL THOUGHT

mathe�matical is not so surprising if we think that mathe�

matics is the language of patterns, while the job of a scientist
is to try and identify regularities or patterns.
We shall return to the question of why science tends to
become mathe�matical, as that is central to understanding
the development of mathe�matical thought. First I want to
emphasize that we cannot understand the meaning or
significance of a general law without some sense of what
that law implies, and working out the implications of a
physical law is far from straightforward! For example, it
is one thing to know that gravity follows an inverse square
law, but understanding the implications of that law is
another thing altogether. Indeed, several contemporaries
of Newton independently imagined that gravity might
follow an inverse square law, but Newton is justly given
the credit because only he could deduce how an object
would move if it was subject to just such a system of
forces. That is to say, we can use Newton’s matheÂ�matical
and conceptual scheme to deduce that planets will follow
an elliptic orbit, a projectile will follow a parabolic path,
and so forth.
Everyone knows that physical laws are important, but
scientists and engineers don’t make predictions by simply
looking up the relevant rules. The application of physical
laws is not a neat, axiomatic process, but an art by which
practitioners choose and adapt the techniques that are
relevant to modelling a given situation. For example, it is
not at all obvious what the Law of Gravity tells us about
the motion of a paper dart, so we wouldn’t use the motion
of such a dart to check if Newton’s theory is correct.
Evaluating the contexts in which a given formalism is useful
or relevant is an essential part of the art of the various
sciences, and that discipline is not itself a strictly mathe�
matical one.
Biologists, medical researchers, physicists, engineers,
economists, social scientists and many other people make
 MODELLING THE WORLD 275

progress in their chosen field by constructing mathe�matical

models. A mathe�matical model is a logical machine for
converting assumptions into conclusions, and it is a striking
fact that seemingly esoteric forms of mathe� matics have
enabled countless insights into important, empirical events.
Since the objects of mathe�matics are governed by their own
rules, we can scrutinize mathe� matical claims by their
own, internal logic, and can be confident that our conclu-
sions really do follow from the given assumptions.
When we judge a matheÂ�matical model we don’t just
assess the validity of the mathe� matics. We also assess
whether mathe�matical accounts of actual events are apt
descriptions of the events in question, and we do this by
subjecting our claims to empirical tests, and by opening
our work to public scrutiny. Science is corrigible, and even
if we are reasonably confident that we know how some-
thing works, we cannot be certain that we have not missed
some critical fact. Nevertheless, there is no denying that
scientists really can show each other something, despite
the fact that our knowledge is always incomplete.
Many qualities can make a model useful or important,
but the best models are like the best theories: they show
that seemingly unrelated observations can all be explained
by a single mechanism. Mathe�matical models are simpli-
fications, not the complete truth, but in many cases we
can use them to obtain answers about what will happen
in the physical world. This form of excellence in a model
can be measured by a machine, as in the case where we
predict how much a beam will bend, and can empirically
demonstrate that the beam actually bends by the amount
predicted by our model.
We may be awestruck by the power of science, but I
do not think that we should be surprised that science is
mathe�matical. In a sense we have nowhere else to look, as
scientists require logical and mathe�matical terminology to
make precise, quantitative deductions from the theories
276 MATHEMATICAL THOUGHT

that they state. If the mathe�matics has already been worked

out, then finding the prediction of your theory will be a
form of calculation. Otherwise, the scientist will need to
work as a mathematician, developing mathe� matical or
formal systems that are inspired by the object of study.
One of the most successful and influential models is
where we assume that gravity is the only relevant force,
and represent a projectile as a mass that occupies a single
point in space. We can predict the motion of a cannon ball
or a star and a planet, but mathe�matical models are always
simplifications or idealizations, and it isn’t always clear
when we have missed some pertinent fact. Nevertheless,
simple, understandable models that get the basics right are
an invaluable source of insight, while more complex models
may be just as baffling as the real-world system that we
wish to study. The eminent physiologist Denis Noble put
it well in The Rise of Computational Biology: ‘Models are
partial representations. Their aim is explanation: to show
which features of a system are necessary and sufficient to
understand it. So, although we could try to understand
cardiac rhythm as the interaction of the few thousand
protein types that are in any cell, we can in fact understand
most of what we wish to know about pacemaker activity
from the interaction of around a dozen protein types. The
power of a model lies in identifying what is essential,
whereas a complete representation would leave us just as
wise, or as ignorant, as before.’
To make a prediction we need to construct a model.
Moreover, models shape our understanding of the natural
world. As the philosopher Nancy Cartwright convincingly
argued in How the Laws of Physics Lie, models play a role
in physics like that of fables in the moral domain: trans-
forming abstract principles into concrete examples.
Essentially, they enable us to present a comprehensible
and exemplary case of the relevant rule-governed behaviour.
For example, we can use Newton’s laws to construct a
 MODELLING THE WORLD 277

model of a projectile, and the fact that this model projec-

tile behaves like a real one is a principle reason for accepting
Newton’s laws. What is more, through their successes,
failures or limitations, models can help us to refine or
revise our theoretical framework. The way that a model
fails to capture a given phenomenon can show us our
limitations, and suggest a change to the theory that we
use. Models can also be used to help design experiments,
shaping the kinds of observations or measurements that
scientists think of making. In short, a good model does
more than fit the data: it helps to clarify the way to think
about the object we are modelling.

Order and Chaos

Mathe�matical systems are intrinsically ordered, and any
pattern with a regular structure can be described in the
language of math. Because of this fundamental association
between mathe�matics and order, the existence of a branch
of matheÂ�matics called ‘chaos theory’ may appear to be a
contradiction in terms. However, when mathe�maticians
study ‘chaos’, they are studying a system of explicitly stated
rules, and not the absence of rules that the word chaos
usually denotes. The distinctive and remarkable feature of
mathe�matical chaos is that the rules in question generate
behaviours that are inherently difficult to predict. To be
more precise, a dynamical system is said to be chaotic if
objects obeying the given set of rules always remain within
a finite region of space, but you cannot predict exactly
where the objects will be in the future, because making a
small alteration to an object’s location results in a series
of movements that are very different to what would have
happened if you had not made that change. In other words,
chaotic systems are governed by highly sensitive functions,
which produce very different outputs on the basis of very
similar inputs.
The process of kneading dough is an excellent example
278 MATHEMATICAL THOUGHT

of a chaotic system, as it shows how a very simple rule

can be extremely sensitive to the initial condition of the
system. As the following diagram illustrates, neighbouring
points rapidly become separated by the kneading process,
making it inherently difficult to predict where a given
speck will end up.

Every time we stretch the dough, the gap between nearby

points doubles in length. Any tiny error in our description
of a point’s original location will grow and grow at an
exponential rate. The gap between our estimate and the
actual location can double with every kneading motion,
which means you need to be arbitrarily accurate to predict
the future position of any given point. The fact that we
cannot predict the future location of a speck in some
kneaded dough demonstrates an important, general prin-
ciple. Chaos frequently emerges from extremely simple,
deterministic programs: a spontaneous emergence of the
unpredictable. This fact of life is something of a mixed
blessing. On the one hand, it means that even when we
know the laws that govern a physical system, we may not
be able to predict what will happen. On the other hand,
it shows that complex, unpredictable things can be governed
by a deep simplicity.
 MODELLING THE WORLD 279

The flow of fluids is a very familiar example of a phys-

ical system that can exhibit chaotic behaviour. We find
steady and regular movement when fluid is moving slowly,
but fast-flowing fluids are turbulent and chaotic. Many
systems exhibit more than one kind of behaviour, giving
stable or static behaviour for one range of parameter values,
periodic or repetitive behaviour for a second set of param-
eter values, and unpredictable or chaotic behaviour for
some third range of parameter values. A beautiful example
of a transition from one qualitative type of behaviour to
another can be found by blowing over a strip of paper. If
we blow gently, the Bernoulli effect lifts the paper to a
stable equilibrium, where the lift is balanced with gravity.
If we blow a little harder there is a certain point where
the paper will overshoot, and as long as we continue to
blow at the same rate, it flutters with a regular periodic
motion. The boun�daries between different, qualitative types
of behaviour are called bifurcation points. An example of
a bifurcation point is the smallest rate of blowing that will
make the paper flutter, as this marks the transition from
one qualitative kind of behaviour to another.
A second kind of bifurcation point is where we go from
a regular periodic behaviour to a chaotic one, as in the
case of a dripping tap. At a low rate of flow a tap will
produce a regular drip-drip-drip. If you loosen the tap a
little further, you can produce a two-step drip-drop-drip-
drop pattern. Further loosening can produce a four-step
pattern, followed by an eight-step pattern, and so on.
Very rapidly the sequence of drips becomes enormously
complex, irregular and inherently difficult to predict, and
at a certain point it ceases to be periodic. In practice this
is hard to see, because after you reach the chaotic regime,
it only takes a little more water for the drops to become
a stream. In theory at least, a dripping tap with a low
flow rate is ordered and predictable, but as we increase
the flow of water the rhythm of the drops becomes chaotic.
280 MATHEMATICAL THOUGHT

That point of transition, from periodic to chaotic, is a

second kind of bifurcation point.

Theoretical Biology
Everyone knows that theoretical physicists use a lot of
math. The exceptionally gifted physicist Freeman Dyson
put it well in his classic essay ‘Mathematics in the Physical
Sciences’: ‘For a physicist matheÂ�matics is not just a tool
by means of which phenomena can be calculated, it is the
main source of concepts and principles by means of which
new theories can be created.’ In contrast biology is far less
mathe�matical, and the role of mathe�matics in biological
science is much less widely known. Nevertheless, mathe�
matics and computing are increasingly essential to
biologists. A great deal of money and effort is being fed
into a process whereby mathe�matical models and experi-
mental data are used to develop ideas and hypotheses,
which can then be tested experimentally, leading to the
refinement and expansion of the original models.
The first point to make about the relationship between
math and biology is that careful and methodical observa-
tion is a central part of every science, and many poor
mathe�maticians have made hugely important contributions
to the life sciences. You don’t need to be a mathematician
to observe the forms and behaviours of living things, but
that does not change the fact that mathe�matics is absolutely
fundamental to modern biology. For example, consider
the most important idea in biology: Darwin’s theory of
evolution.
A modern definition of evolution is ‘the change in the
frequency of the different gene-types (alleles) in a popula-
tion over time.’ The basic idea is that given some
population of individuals, the number of individuals may
change over time, and the total number and kind of different
genes that are present in the population’s genomes will
also change over time. Some genes will become more or
 MODELLING THE WORLD 281

less prevalent, some children will have a slightly different

version of a gene compared with their parents (a change
from one allele to another), and so on. The word ‘evolu-
tion’ refers to the fact that over time, there are changes in
the proportion of the population that carries each of the
different alleles.
By this definition, no one can deny that evolution is a
real phenomenon. It is like the observation that a dropped
object falls towards the ground, not some general theory
of gravity that may turn out to be false. Religious funda-
mentalists may refuse to accept scientific versions of the
history of life on earth, but even the most intransigent
must surely agree that the frequency of different genes in
today’s population is different from the frequency of genes
in yesteryear’s population, if only because the human popu-
lation has increased from the original two! After all, in a
population of two there are only three possible frequencies
for any given gene: 0% (neither Adam nor Eve has the
gene), 50% (one person has the gene), or 100% (both
people have the gene). In contrast there are all manner of
different gene frequencies in the population of humans
who are alive today, which means that by definition evolu-
tion has occurred. More importantly, many of the observed
changes in the relative frequency of different alleles can
be explained by Darwin’s big idea: the process of natural
selection. That is to say, if the population in question
contains some genetic variation, changes in the frequency
of different alleles will occur whenever some genetic types
are more likely to survive and reproduce than others.
The concept of frequency or proportion is absolutely
essential, as what changes over time is the proportion of
individuals with some or other type of genetic inheritance.
The point I am trying to make is that we literally could
not phrase a theory of evolution without the ability to
count. More generally, modern biomedical science involves
all manner of sophistical mathe�matical techniques. There
282 MATHEMATICAL THOUGHT

is no question that scientists will continue to use computers

and mathe� matical analysis as a route to understanding
biological phenomena, from mathe�matical descriptions of
the interactions between biological molecules, up to the
scale of organ physiology, the study of development or
the modelling of population dynamics.
The story of how an egg becomes an organism is of
particular interest, and to understand such a process we
need to consider the genome, and much more besides. In
short, organisms are physical entities, and the interactions
that are essential to development occur in space and time,
subject to the laws of physics. As a simple example, when
a mammal produces milk we find spherical droplets of oil
suspended in an aqueous fluid. We don’t need a gene that
says ‘droplets in milk should be spherical’. We understand
that droplets will be spheres because spheres have the smallest
surface area for the given volume, and there are well under-
stood physical reasons why the surface area will minimize.
A moral of this story is that to gain a deeper under-
standing of the specific details that a biologist might study,
it is wise to turn to a more general, mathe�matical picture
of what is going on. As the theoretical biologist Hans
Meinhardt argued in Models of Biological Pattern Formation,
we require theory to make sense from observations.
Suppose, for example, that we possessed a perfect biochem-
ical scanner that could tell us the precise concentration of
every kind of molecule in every portion of space and time.
We might be able to measure the changing concentrations
associated with every developmental event, but we would
still lack insight as to what is essential to the process, and
what was merely incidental. Cause and effect would be
jumbled in the data, so although our perfect scanner could
tell us a great deal, we would still lack a sense of how the
system works. For example, we might not be able to predict
what would happen if we managed to change some or
other of the measured concentrations.
 MODELLING THE WORLD 283

In order to feel like we understand what is going on,

we need to call into being a hypothetical mechanism, or
model, which accounts for the data as well as we can
manage. Biology is incredibly complicated, and we often
make do with an informal word model: an unquantified
description of the forces at play. As is the case with any
verbal account, the implications and theoretical commit-
ments of such a description are often far from clear. Indeed,
at this stage in the development of biomedical science,
much of the progress we are making is of the form ‘gene
X is associated with process Y’, or ‘brain region X is asso-
ciated with process Y’. At the risk of sounding unduly
critical, observing a correlation between one thing and
another barely qualifies as a description of the process in
question, let alone an explanation of what is going on.
That is not to deny that observing a correlation can be
very useful: in some cases information of this kind can
even lead to the development of new and effective medical
treatments, and it is certainly worth knowing that a given
treatment correlates with recovery! It is just that spotting
correlations does not provide insight into how the system
actually works, unless those observations are helpful in
developing or critically evaluating some kind of model, or
hypothesized mechanism.
Given our current state of ignorance and the extreme
difficulty of measuring all that we might want to measure,
it is inevitable that any mathe�matical model of a biological
system is bound to be somewhat arbitrary. Nevertheless,
I would claim that precise, mathe�matical models that can
support detailed deductions are superior to informal word
models for several reasons:

1. Constructing a mathe�matical model forces us to

be clear about our working assumptions.
2. Proposed biological mechanisms should be consistent
with the known principles of physics and chemistry,
284 MATHEMATICAL THOUGHT

and our understanding of those subjects is profoundly

mathe�matical.
3. Systems of interacting parts can behave in counter-
intuitive ways, and we need mathe�matics to be
able to identify correctly the implications of
our assumptions. In other words, unlike vague
descriptions, mathe�matical models generate clear,
unambiguous predictions of both a quantitative
and qualitative nature.

Biology is advancing at an awesome pace, but it seems

implausible that mathe�matics will ever dominate biological
theory to the extent that it dominates theoretical physics.
Nevertheless, many of the recent advances in biology
(including the reading of genomes) have required increas-
ingly sophisticated collaborations between experi�mentalists
and mathe�matically sophisticated theorists. This is a recent
development, which partly reflects the fact that unlike their
predecessors, modern biologists often deal with very large,
computer-based data sets. For example, we cannot hope
to make sense of the vast catalogue of genetic data without
some sophisticated statistical techniques. However, in addi-
tion to providing computational tools and techniques for
biomedical research, I hope to persuade the reader that
simple mathematic arguments can also play an essential
conceptual role, as we try to speak aloud the logic of life.

Interactions and Dynamical Systems

Biologists have made huge progress by taking a reductionist
approach. For example, we can identify a gene, we can
identify the protein that cells produce when they transcribe
and translate that gene, and we can look to see how the
protein in question behaves. This kind of approach has
produced some fantastic science, but there are many biolog-
ical phenomena that cannot be understood until we put
Humpty Dumpty back together again. That is to say, if
 MODELLING THE WORLD 285

we are going to understand how a cell or an organism

functions, we need to know how the molecular components
interact over time.
Mathematical models of complex, dynamical systems
are an increasingly central feature of the scientific enter-
prise. The basic idea is to consider the behaviour of a
system of parts, each of which affects its neighbours in
accordance with predefined rules. Any particular model
of a complex system is likely to have many arbitrary
features, but despite the fact that we could have made
many other computer models of the same physical
phenomena, perspicuously constructed models can explain
a great deal. For example, models can explain why buses
come in threes, or at least, why buses tend to cluster.
In order to see why buses tend to bunch together, we
need to make three simple, empirical observations. The
first observation is that it takes time for people to move
from a bus stop onto a bus, and since they move onto
the bus one at a time, the more people there are at a bus
stop, the longer it takes for the bus to pick them all up.
The second observation is that the number of people
waiting at each bus stop increases over time, as more and
more people arrive at the bus stop. Third, people leave
the bus stop when a bus arrives. This combination of facts
implies that the presence of a second bus some short
distance in front of you speeds you up, because the bus
in front of you removes people from the stops, so your
bus passes the stops more quickly. Similarly, the absence
of a bus in front of you slows you down, as the bus stops
will contain more people, and it will take longer to pick
them all up.
It therefore follows that regular processions of buses
are inherently unstable. Small gaps will tend to get smaller
as the bus at the back will tend to catch up with the bus
in front of it, while conversely, large gaps will tend to get
larger. Clusters will naturally emerge, as small disruptions
286 MATHEMATICAL THOUGHT

to the regularity of bus arrivals become magnified over

time. Bus conductors minimize the clustering effect, as do
pre-purchased tickets.
As is the case with all of the models that I sketch out
in this chapter, we could be much more specific and quan-
titative about the assumptions that give form to our model.
For example, by being very specific we could use our
model to quantify the scale of the clustering effect.
However, we do not need to include every detail to under-
stand the principles involved, as questions about the real
world can be addressed at different levels of abstraction.
This example also demonstrates that not every truth about
the world is a mere consequence of ‘fundamental’ physical
laws, as there is more to physics than the rules that govern
the behaviour of particles. To put it another way, we can
understand why buses tend to cluster, and the fact that
the objects in question are composed of atoms is no more
relevant than the fact that they are painted red.
Another remarkable model of pattern formation has its
roots in one of Alan Turing’s best ideas. Back in 1952,
Turing was trying to understand the development of an
embryo. He knew that when an embryo consists of only
two cells, separating those cells results in the growth of
identical twins, while leaving them together produces a
single individual. Turing was fascinated by the develop-
mental process, and asked himself some difficult questions:
How do the two cells ‘know’ about each other, and how
can a group of cells self-organize to create a pattern?
Turing assumed that each cell must ‘know’ about the
other because certain molecules move between them.
He imagined an initially uniform chemical mixture, and
tried to think how it might spontaneously develop into
a pattern with different parts. Being a genius of the first
order, he realized that concentration gradients could
spontaneously emerge if two substances with different
diffusion rates react with one another. This is somewhat
 MODELLING THE WORLD 287

counter-�intuitive, as diffusion usually smoothes out any

differences in c� oncentration. Nevertheless, Turing showed
that if the diffusing chemicals are involved in reactions
of a given kind, the reaction-diffusion process will create
local regions with unusually high (or low) concentrations.
Few biologists paid much attention at the time, but the
mathematician Alfred Gierer and the theoretical biologist
Hans Meinhardt have developed Turing’s idea, identifying
a key principle in biological pattern formation.
The basic idea is that patterns can be generated by a
combination of ‘local activation’ and ‘global inhibition’.
There are many different variants of this theme, but as a
simple example imagine a flat desert with a few rocks
scattered in the sand. If there is no wind, our flat desert
could persist over time, and we would not have a pattern
of high points and low points. If there is wind, it will blow
the sand around. The rocks create a small area that is
sheltered from the wind, and sand will accumulate at those
points, because it is easier for the sand to be blown into
the shelter than it is to be blown out of the shelter. In
other words, the presence of wind energy can turn a small
rock into a large sand dune. This process of a large thing
getting larger, or a high concentration becoming even
higher, is what we call local activation.
If local activation was the only effect, we wouldn’t get
a spatial pattern: we would simply have a concentration
(or quantity of sand) that starts small but gets larger and
larger and larger. My point is that deserts have high points
and low points, and this relates to the fact that there is a
finite amount of sand blowing in the wind. If the sand is
accumulating in the shelter of the dunes, which means it
is not accumulating elsewhere. This long-range reduction
in the ‘chemical’ of interest is what we call global inhibi-
tion. As another example of pattern formation by local
activation and global inhibition, recall my account of the
formation of leaves. Leaves begin to form in the regions
288 MATHEMATICAL THOUGHT

of the growing shoot that contain a high concentration of

auxin, but in a given stem the concentration of auxin can
initially be quite uniform. As cells tend to pump their
auxin into the neighbouring cell with the highest auxin
concentration, high concentration areas tend to become
even higher (local activation), while the rest of the stem
is effectively drained of auxin (global inhibition).

Holism and Emergent Phenomena

Organisms, economies and ecosystems all contain large
numbers of different, interacting parts. If we want to under-
stand a complex system, having an accurate list of the parts
and their properties is certainly important. Nevertheless,
it is sometimes the case that knowing lots of details about
the individual, isolated parts does not tell you what you
want to know about the behaviour of the system. The
heart of the matter is that in countless cases, we cannot
understand the behaviour of the entire system unless we
pay careful attention to how the behaviour of one part
influences the behaviour of another.
There are many ways to represent one event being
coupled with another, and it is possible to discern some
important, general schemes. For example, we are all familiar
with the term ‘vicious circle’ (or its benevolent twin, a
‘virtuous circle’), and we have a good idea of when it is
appropriate to use this phrase. The notion of a vicious or
virtuous circle is closely related to that of ‘positive feed-
back’. Imagine a situation where an increase in the
concentration of A triggers an event that produces even
more of A, resulting in ever increasing concentrations. The
converse to positive feedback is negative feedback, where
an increase in A triggers an event that reduces the preva-
lence of A, while a decrease in A triggers an event that
tends to increase the concentration of A. Because low values
tend to rise and high values tend to drop, negative feedback
can be used to maintain a stable concentration.
 MODELLING THE WORLD 289

It is vital that our bodies’ internal environment remains

stable, so it is unsurprising that there are many physio-
logical examples of negative feedback. As we all know, if
our body temperature rises we cool ourselves down by
sweating, and if our body temperature drops we warm
ourselves up by shivering. Negative feedback can also be
used to produce oscillations, particularly if there are time
delays in the system. For example, if I can turn up or turn
down the heat in my house, but I am slow to make any
adjustments, my house might get very hot before I finally
turn down the thermostat. Having turned down the ther-
mostat my house might then become very cold, before I
finally get around to turning up the heat again. As a result
of these dynamics the temperature in my house might
oscillate between hot and cold, and a similar mechanism
of negative feedback with time delays can be used to
produce oscillations in the chemical concentrations in cells.
There are also important physiological examples of posi-
tive feedback. For example, the pituitary gland is located
at the base of the brain, and in women it sometimes secretes
a small amount of luteinizing hormone (LH). This stimu-
lates the ovaries to secrete oestrogens, and in certain
conditions a rise in oestrogen levels in the blood stimulates
the pituitary gland to produce more LH, which leads to
higher oestrogen levels, which leads to higher LH levels,
and so on. Because of this positive feedback a small initial
quantity of LH soon results in the production of large
concentrations of LH. This phenomenon is known as an
LH surge, and it triggers ovulation. Women’s blood doesn’t
always have a high concentration of LH because ovulation
temporarily inhibits the ovaries’ ability to secrete oestro-
gens. The resulting drop in oestrogen levels removes the
stimulus that produced the initial rise in LH levels, so
ovulation allows LH levels to drop back down to the
concentrations found before the LH surge.
More and more mathe� maticians are working with
290 MATHEMATICAL THOUGHT

biologists and other scientists to analyze the behaviour of

systems that are comprised of many interacting parts. One
of the really fascinating things about studying complex
systems is that when we build a model with many
interacting parts, we frequently find that it exhibits
‘emergent phenomena’: forms of behaviour that we cannot
possibly predict by looking at any of the individual parts.
For example, we cannot understand why buses tend to
cluster if we only study an individual bus, and so we might
say that the clustering of buses is an emergent phenomenon.
By the standards of mathe�matical science the notion of
emergence is somewhat vague, although a wide range of
theorists have worked to refine the concept. The practical
point is that in many kinds of enquiry, the property or
behaviour that we are interested in cannot be made manifest
in any single part: to see the thing in question, we need
to consider an entire system, including rules for how the
parts interact.
Such inherently holistic phenomena can be contrasted
with concepts like volume, as it is reasonable to say that
a shape’s volume precisely consists of the volume of its
parts. Interestingly, this claim about the relationship
between the whole and the parts of geometric forms cuts
to the heart of what is new and different about geometries
phrased in the language of sets. For millennia, people agreed
with Aristotle when he claimed that the parts of a contin-
uous thing are themselves continuous, so the parts are the
same kind of thing as the whole. For example, Aristotle
would say that we can identify a point that is inside a
sphere, but a sphere is not made of points. As far as he
was concerned, you can chop a sphere into smaller and
smaller parts, but anything that deserves to be called a part
of a sphere must have a finite, non-zero volume.
In contrast, modern mathe�maticians typically begin by
defining a sphere as the set of all points within a given
distance from the centre of the sphere. Hence the sphere
 MODELLING THE WORLD 291

is conceived as being an infinite set of points, which suggests

that the parts of a sphere are points, not volumetric regions.
Defining shapes and spaces in terms of sets and points,
where each point is identified by a set of real number
coordinates, is a very powerful move. It guarantees that
we can use algebraic techniques to tackle problems in
geometry, but it also leads to counter-intuitive results. In
particular, there is a theorem called the Banach-Tarski
paradox, which shows that if we chop a sphere into an
infinitely fragmented or non-continuous collection of
points, then those fragments can be rearranged to make a
sphere with a different volume! In other words, all of the
points in a small sphere can be rearranged to coincide with
all of the points in a large sphere. This is possible because
every sphere has the same, infinite number of points, and
points themselves have no volume.
Of course, if we chop a sphere into any finite collection
of continuous pieces, the total volume of those pieces
cannot change: it must always equal the volume of the
sphere. In other words, the volume of the whole is equal
to the total volume of the parts, and the property of having
a given volume is not something new that only emerges
when we glue all the pieces together. In contrast, the topo-
logical properties of a given shape could be described as
holistic properties, as it does not make sense to ask which
part of a circle makes it a closed loop. Being closed is not
a property of any one part of the shape; it is a property
of the whole. Similarly, the Euler number of a shape is
not something we can identify by looking at its separate,
component parts: the way that the whole thing is glued
together is absolutely fundamental. Indeed, if we do chop
our shape into separate parts, the Euler number changes!
 Chapter 13:
LIVED EXPERIENCE AND THE
NATURE OF FACTS
‘Those who judge by employing a rule are in regard
to others as those who have a watch are in regard
to others. One says, “Two hours ago”; the other says,
“It is only three-quarters of an hour.” I look at my
watch, and say to the one, “You are weary,” and to
the other, “Time gallops with you, for it has been
an hour and a half.” I laugh at those who tell me
that time goes slowly with me and that I judge by
imagination. They do not know that I am judging
by my watch.’
Blaise Pascal, 1623–1662

Rules and Reality

The word matheÂ�matics is derived from the Greek for ‘teach-
able knowledge’, and I think it is highly significant that
when we are engaged with mathe�matical science, there is a
distinctively close connection between what we know about
a given subject matter, and what we can say about it. An
insightful object of comparison can be found in Wittgenstein’s
Philosophical Investigations, where we are invited to consider
the difference between ‘knowing the height of Mount
Everest’, and ‘knowing what a cello sounds like’. You cannot
 LIVED EXPERIENCE AND THE NATURE OF FACTS293

be credited with knowing the height of Everest without

being able to tell someone. On the other hand, even a
master cellist may flounder when asked to show that they
‘know’ what their instrument sounds like.
There are lessons from our contact with reality that
cannot simply be packaged up and passed around. Not all
human understanding is factual in form! On the other
hand, the pursuit of scientific understanding is an essen-
tially public enterprise, as a scientist’s work largely consists
in articulating the knowledge that is gained from investi-
gating various objects. In the words of the physicist Niels
Bohr, ‘It is wrong to think that the task of physics is to
find out how Nature is. Physics concerns what we can say
about Nature.’ In my view, it is the necessity of explicitly
formulating and sharing our understanding that drives
empirical science to the point where it becomes mathe�
matical. To put it another way, nature could be there
without us, but until we have constructed a language, there
can be no facts. After all, we cannot have facts without a
language to express them!
The language of mathe�matics is an integral part of the
human adventure, and the history of math tells us some-
thing about the cultures in which it developed. On the
other hand, compared to other cultural artefacts, the world
of matheÂ�matics is strangely timeless: you don’t need to
know how the Ancient Greeks lived to understand
Euclidean geometry. The facts of mathe�matics can readily
be transmitted from one culture to another, and as they
have deep roots in our cognitive abilities, they seem to
belong in every world. Indeed, in and of themselves, the
facts of mathe�matics cannot tell us whether we are living
in a world with one history or another. Whatever the past
or present might be like, once we come alive to the possi-
bility of using a form of mathe�matical language, we cannot
imagine a world in which it would be impossible to think
in those terms! For example, I might be utterly deluded
294 MATHEMATICAL THOUGHT

about the world around me, but once I have learned how
to count, I cannot imagine being part of a reality in which
I cannot count.
As the quotation at the beginning of this chapter indi-
cates, it is important to appreciate that mathe�maticians,
scientists and engineers pursue their aims by employing
rules. As a principal example of a rule-governed activity,
consider the practice of counting. If a child does not count
like the other children, we are right to tell them they are
wrong, simply because they do not count like us. No other
justification is required: we do not need to refer to a fact
beyond the practice of counting itself. You might want to
protest that there is more to the integers than the proper
way to recite the counting song. That is no doubt true,
but we should be careful, for as Wittgenstein observed,
‘People can’t distinguish the importance, the consequences,
the application of a fact from the fact itself; they can’t
distinguish the description of a thing from the description
of its importance.’
In short, mathe�matics is not a natural phenomenon, but
it is experienced as though it were. For example, if I learn
to count and see that 2â•›+â•›2â•›=â•›4, I might want to insist that
this isn’t just true for me, or the people who count like
me, it is a truth of the universe itself. Such a claim is
eminently reasonable. Two stones and two stones do indeed
make four stones, and wishing that it were otherwise cannot
make a difference. However, by making this observation,
I am not denying that mathe�matics is something we create.
After all, it is us that see the stones as a group of objects,
and a statement of fact is not the same thing as a physical
state of affairs. To put it another way, our versions of how
the world works do not write themselves, even if our
accounts are genuinely faithful to the reality in which we
find ourselves.
In my view, mathe�matical truth comes into being with
the rule-governed use of symbols: without the explicit
 LIVED EXPERIENCE AND THE NATURE OF FACTS295

rules or principles set down by ourselves or our predeces-

sors, there can be no mathe�matical facts. On the other
hand, mathe�matics is more than merely shuffling symbols
according to a given set of rules, because once we have a
rule-governed system that we can gainfully employ, we
can view the world though the logical lens provided by
that language. And once a person uses a word or concept,
it acquires real meaning. This image of people using
symbolic systems might suggest a picture of mathe�matical
factuality that is comprised of two parts: the actual, mean-
ingful, historical concepts and symbols used by physically
embodied human beings, together with the ahistorical or
timeless symbolic facts which could in principle be verified
by any suitably programmed computer.
This two-fold image of mathe�matical truth has a lot to
recommend it, provided that we acknowledge the impos-
sibility of seeing a dividing line between the meaningful
mathe�matics that we know and use, and the computable,
symbolic systems that are consistent and legitimate, but
yet to be discovered. After all, the moment a person talks
about a possible computable system, it becomes a system
that has actually been used! As Wittgenstein remarked,
‘There is a feeling: “There can’t be actuality and possibility
in mathe�matics. Everything is on one level. And in fact,
is in a certain sense actual.” And that’s correct. For matheÂ�
matics is a calculus; and a calculus does not say of any sign
that it is merely possible; rather, a calculus is concerned
only with the signs with which it actually operates.’
Calculations and mathe�matical arguments can be used
to understand the world, and that is what makes mathe�
matics profoundly meaningful. On the other hand,
mathe�matical facts are not contingent on physical states
of affairs, as the language of math deals in generalities,
and is not supposed to be understood by reference to
specific or historical events. In contrast, scientists make
claims about the empirical world, and completely different
296 MATHEMATICAL THOUGHT

theories can all refer to the same empirical phenomena.

Consequently, scientists have to accept that new evidence
or a new approach might show that their theories were
wrong. Mathematicians often have hunches that turn out
to be wrong, but the results that they prove acquire a
certainty that science will always lack, precisely because
mathe�matical truth is always truth within a given system.
We might say that our language enables us to look at the
world through a particular mathe�matical lens, and however
our surroundings may change, we are free to keep our
logical glasses just the same.
Perhaps the most important thing is that mathe�maticians
articulate their concepts in a way that enables deduction.
This fundamental discipline constrains and shapes the body
of mathe�matical facts, but there is more to mathe�matics
than a tower of deductions. In short, mathe�maticians inves-
tigate describable, conceptual schemes. The same might be
said of other theorists, but the logical structures of mathe�
matics are distinctive because the point of interest is nothing
beyond our own definitive statements and symbolic forms
of practice (e.g. counting). Mathematicians are free to count
actual objects, but unlike other forms of theory, the terms
of matheÂ�matics don’t need to refer to any external objects.
I therefore conclude that the objective reality of mathe�
matics is not activity in the brain, or some magic realm of
matheÂ�matical forms. Like Pascal’s watch at the beginning
of this chapter, what matters are the statements that a
mathematician employs.
Although few people focus their mind on the purely
mathe�matical, I believe that our natural capacity to under-
stand such truth is a vital part of what it is to be human.
This is true of everyone, not just the well-trained experts.
For example, although children need to be initiated into
using the vocabulary of straight lines, squares, triangles
and other specific shapes, our capacity to recognize the
sense of such a language is innate, and we cannot imagine
 LIVED EXPERIENCE AND THE NATURE OF FACTS297

looking at the world without finding shapes within it.

This is not to say that human nature is explained by a
hidden mathe�matics, but we are deeply moved by the
truly obvious, and our mathe�matical traditions begin with
statements of the obvious. I would also say that when a
person makes the effort to assimilate the corpus of math,
they encounter the fact that as rational beings, we are
capable of experiencing truth. This is a fundamental fact
of human nature, and it is inseparable from our use of
language.
To account for the vitality or meaningfulness of the
symbols that they use, many mathe�maticians and some
philosophers are Platonists. Platonists believe that mathe�
maticians study abstract objects that exist independently
from our means of studying them. I find this claim of
independent existence unconvincing and ultimately hollow,
because it seeks to legitimize the things that people actu-
ally know how to do (e.g. doing sums or writing certain
kinds of proof) by appealing to something that is beyond
our knowing (that is, an imagined domain of transcendent
abstract objects).
As we’ll see in the following section, the status of matheÂ�
matical objects is a delicate matter, but in practice it seems
that it doesn’t really matter what matheÂ�matics is ‘about’.
Whether we imagine some pieces of string on the muddy
fields of Egypt or some perfect timeless shapes, the practice
of deducing the areas of shapes ultimately hinges on the
language we employ, not the pictures we have in our minds.
We might imagine our subject matter in many different
ways, but once we describe the object that we wish to
measure, it is the stated lengths and angles that matter, and
the possibility, or impossibility, of using such descriptions
together. More generally, what matters in mathe�matics is
how we can work from one statement to another, not the
objects that we think we might be talking about.
Mathematicians are free to study anything that can be
298 MATHEMATICAL THOUGHT

identified and characterized by means of a consistent

description. Chess, sudoku, and other rule-governed games
are essentially mathe�matical, though a game like chess is
much more arbitrary than the structures that mathe�
maticians choose to study. I say this because the concepts
that mathe�maticians investigate are closely related to a large
complex of other concepts, including ideas that are used
in science, or everyday accounts of what the world is like.
The interconnectedness of mathe�matics is truly profound,
and the subtle relationships between ideas provide a way
of valuing some mathe�matical insights over others. By way
of analogy, if a historian studies a particular event, it is
because they think that the event in question has bearing
on events in general, and an equivalent comment holds
true for mathe�maticians.
If we are interested in solving a particular problem, or
understanding how different mathe�matical ideas are related,
it is certainly not the case that studying one set of consistent
axioms is just as good as studying another. It is not like
choosing between chess and checkers! Using given axioms
to generate new theorems is indisputably important, but
mathe�matical progress also involves the formation of new
hypotheses and new conceptual frameworks, and novel
forms of mathe�matics can be judged by the light it sheds
on previously solved and unsolved problems. My point is
that even if every formal approach is equally mathe�matical,
some arguments are more memorable than others, and not
all of mathe�matics is deemed to be worth remembering.

The Objectivity of Math

What is wrong with saying that mathe� maticians study
abstract objects? After all, hundreds of generations of
mathe�maticians have studied patterns in the integers. Surely
it is safe to say that the integers are abstract objects? People
imagine the objects of mathe�matics: is our imagination the
realm of math? But don’t the integers possess their own
 LIVED EXPERIENCE AND THE NATURE OF FACTS299

form, regardless of what we think? Would they be there

without us? Is there a timeless, spaceless portion of reality
where mathe�matical objects exist?
If we are not careful, these bewitching images can stop
us from looking at the things that we actually know. We
do not need access to timeless, perfect objects. Instead of
casting meaning away from humanity to some ultimate,
all-seeing judge, we can suppose that mathe�matical language
brings about the reality it declares. Rather than thinking
of mathe�matical objects as the underlying bedrock of math,
I think we should concede that the irreducible fact of the
matter is that people are capable of employing the language
of mathe�matics. In particular, the brute fact about numeracy
is that in the medley of a lifetime we have assimilated the
basic concepts of the integers, and possess the ability to
employ them. As a result, people really can count, and are
capable of making true, arithmetic statements. So, you
might ask, if we experience truth by ‘assimilating matheÂ�
matical language’, are the facts of matheÂ�matics invented
or discovered?
The advent of novel mathe�matics is a unique kind of
event, but I am inclined to say that ‘invent’ is the better
metaphor. Like engineers building a conceptual device,
mathe�maticians need to be inventive, but we discover that
our invention works, and is intelligible to our colleagues.
By way of analogy, we invent the rules of chess, but the
game has a certain autonomy, and given its rules we cannot
change the fact that forcing checkmate with two knights
is impossible. Indeed, we might well say that we discover
this fact about our own invention. Likewise, mathe�matics
is a human creation, but we cannot simply bend our crea-
tions to our will, as mathe� matical objects and the
relationships between them are constrained by stated prin-
ciples. As in the case of chess, the partial autonomy of
mathe�matics derives from the fact that our creation is
governed by rules that we can state.
300 MATHEMATICAL THOUGHT

In short, mathe�matics is a language, and languages are

cultural artefacts. This is a subtle claim, as it does not
mean that the objects of mathe�matics are arbitrary invent�
ions, or only exist ‘in our head’. We can find meaning by
making a language our own (a process that enables us to
form certain kinds of thought), but language itself does
not belong to any individual, or exist in any one brain.
By its very nature language is sharable, with deep roots
in observable patterns of behaviour, so although we need
brains to think, we do not understand what sentences
mean by looking at each other’s brains! We may feel as
though our thoughts are private, and contained within
our own minds, but mathe�matical structures and symbols
are essentially public and sharable, even if our feelings
about them are not. As Heraclitus of Ephesus remarked
twenty-six centuries ago, ‘Although the forms of reason
(logos) are shared, most men live as though their thinking
were a private possession.’
We encounter the objects of mathe�matics through our
use of language (particularly the terms used in measure-
ment), and it is the use of rule-governed language that can
bring such objects to mind. Furthermore, our under-
standing of number is very deep rooted indeed, because
we naturally appreciate that objects can be added to a
group, or placed into containers. Indeed, when we are
teaching arithmetic we quite reasonably take it for granted
that children will be able to understand the process of
adding or taking away objects from a collection, even if
they don’t yet understand the symbols of arithmetic. By
building on the highly intuitive, even pre-linguistic concept
of a collection, we have been able to form highly robust
abstract number concepts. Such abstract concepts require
the use of rule-governed language, and as such they have
an essentially cultural component.
For example, we are justified in saying that there ‘is’
one and only one number ‘3’ precisely because it is
 LIVED EXPERIENCE AND THE NATURE OF FACTS301

actually possible to count to three, and a public can

assess whether or not someone has done this correctly.
The word is meaningful because we have an intuitive
understand of collections of countable objects, but the
facts about numbers cannot be separated from the
conventions for their use. To put it another way, we
should only accept a set of axioms if they are in accord
with the ideas that we associate with our symbols, but
it is the fixed axioms and other rules governed principles
that make mathe� matics precise, stable and sharable.
Furthermore, once we accept a set of rules and begin
to employ them, we can discover the patterns that are
drawn out by those rules.
Because mathe� matics is precise, stable, sharable and
rooted in human cognitive abilities, the objects of mathe�
matics have many things in common with nameable,
physical objects. However, numbers and other mathe�
matical objects are unlike physical objects in that the
existence we can claim for the integers (say) is not an
independent, isolatable one. In other words, number words
resemble names, but it is misleading to take the object ‘3’
too literally. We do not need to imagine some transcendent
object that has all the properties of three-ness, and none
of the mess or irrelevance of a particular system of nota-
tion. All we need is a series of number words, so we can
start to count. That is to say, the truly essential point is
that each number stands in a structured relationship to the
other numbers. Not only can we identify an item in a
particular sequence as ‘being the third’, it is also the case
that each integer can be characterized and identified by its
place within a definitive list of the integers. In that sense
number words are very much like names. However, the
essence of a ‘place’ is nothing other than its relation to
the other places.
For example, there is no special quality to being first
in line other than the definitive fact that being first means
302 MATHEMATICAL THOUGHT

being before the other places, without which there could

be no ‘first’. As Stewart Shapiro wrote in Thinking about
Mathematics, ‘Any small, movable object can play the role
of (i.e. can be) black queen’s bishop. Similarly, anything
at all can “be” 3 – anything can occupy that place in a
system exemplifying the natural number structure.’ It
would be absurd to try and make sense of the term ‘black
queen’s bishop’ without referring to the chess board, and
likewise, we cannot make sense of any particular number
without reference to our number system. In short, the
essence of an individual number is not something intrinsic
to an individual abstract object (whatever that might be):
the properties of numbers make sense only in the context
of a number system. For example, we know that ‘3 is odd’,
but that does not mean that oddness is a property intrinsic
to some special object known as 3. Another way of saying
‘3 is odd’ is to say that ‘there is SOME integer n such that
3â•›=â•›2nâ•›+â•›1’, so the oddness of the number 3 precisely consists
of a relationship between the number 3 and some other
numbers.
In short, the specific rules, definitions, and notational
systems that we adopt are central to mathe�matical practice,
as it is these things that specify the game that we are
playing. However, that does not mean that mathe�matics
is ‘a game played with meaningless symbols’, as the formal-
ists suggest. Our choice of rules is motivated by underlying
concepts, and although the things that computers can do
are central parts of mathe�matics, there is much more to
the subject than the valid application of formal languages
like PC. When I say that the irreducible fact of the matter
is that mathe� matical arguments exist, I am not only
thinking of computer checkable proofs and calculations.
I am also thinking about the informal arguments, images
and metaphorical statements that dominate this book, and
the broader play between various kinds of proof and
demonstration.
 LIVED EXPERIENCE AND THE NATURE OF FACTS303

Euclidean geometry is often described as a great

inverted pyramid: a vast tower of deductions resting on
a handful of explicitly stated principles. Despite the truth
in that seductive image, the development of mathe�matics
as a whole is much more than the orderly application of
deductive principles, with one statement following from
another. For starters, creative mathe�maticians not only
work forwards from axioms to theorems. They also work
backwards, starting with a problem and formulating
hypotheses that might enable them to deduce solutions
to the problem at hand. Indeed, as is the case with
Euclidean geometry, we often realize that our foundations
are good foundations precisely because a given conceptual
scheme enables us to tackle effectively a given set of
problems.
My point is that creative matheÂ�maticians don’t just make
obviously legitimate deductions: they try to spot patterns,
they make conjectures, they break problems into smaller
sub-problems, and they guess at general statements after
examining a few specific instances. They also try to solve
problems by employing analogies, as progress is often
made when someone says, ‘Maybe we can understand this
problem by thinking in the terms we use to solve that
other problem.’ These features of analytic thinking are
common to many fields of endeavour, and with experience
a skilled mathematician learns to put their finger on the
critical point.
Mathematicians make progress by thinking about the
challenges that face the mathe�matical community, and this
involves intuitive leaps of understanding, as well as formal,
deductive practice. It is certainly possible to develop mathe�
matical intuition (you just need to spend some time doing
mathe�matics), but mathe�maticians wisely trust their rules
and symbols, not their imaginations. Working mathe�
maticians may harness all manner of mental faculties, but
the distinctive discipline of mathe�matics is to aim at making
304 MATHEMATICAL THOUGHT

the relevant deductions completely formal and explicit, so

that others can follow the argument.

Meaning and Purpose

Nobody seriously believes that one piece of mathe�matics
is as good as any other, even though a dull or arbitrary
piece of math is every bit as correct as the most celebrated
proof. Mathematicians are motivated to engage in the
research that they do, and they have reasons for employing
the rules that they decide to follow. Although our feelings
towards mathe�matical statements cannot change the facts
themselves, our motivation or sense of purpose in doing
mathe�matics is absolutely essential. After all, people have
reasons for doing what they do, and that is an essential
part of the mathe�matical experience.
People are motivated to engage with mathe�matics for
all kinds of reasons. Important practical concerns require
mathe�matics, people want to solve famous problems, we
might wish to show connections between historically
�separate theorems. Most importantly of all, mathe�matical
patterns can simply capture our imagination, provoking
us to thought. As the mathematician G. H. Hardy percep-
tively remarked:

There are many highly respectable motives which

may lead men to prosecute research, but three which
are much more important than the rest. The first
(without which the rest must come to nothing) is
intellectual curiosity, desire to know the truth. Then,
professional pride, anxiety to be satisfied with one’s
performance, the shame that overcomes any self-
respecting craftsman when his work is unworthy of
his talent. Finally, ambition, desire for reputation,
and the position, even the power of money, which
it brings.
 LIVED EXPERIENCE AND THE NATURE OF FACTS305

Formal definitions and computer-checkable deductions are

absolutely central to mathe�matical science, and it is obvious
that computers do not need to have a sense of the motiva-
tions of mathe� maticians in order to do what they do.
Nevertheless, motivations are an essential part of the mathe�
matical experience, because the satisfaction of a compelling
argument does not simply consist of its publicly established
validity. My point is that the vitality of computer-checkable,
symbol-based methodologies is not inherent to the rules
alone, but rather depends on the endlessly mysterious and
unruly process whereby any kind of symbolic representa-
tion may come to engage with our imagination. To put it
another way, we may have no choice but to define a word
in its relation to other words, but our sense of meaning
is essentially bound to the apprehension of human purpose.
It is well known that mathe�maticians rely on clearly
stated definitions. Consequently, we might be inclined to
think that motivations are unimportant in math because
our feelings about the facts cannot change the facts them-
selves. However, that kind of insensitivity to underlying
motivations is far from being a unique or distinguishing
feature of mathe�matical science! It is physically inevitable
that people do not need to share the same sense of purpose
in order to use the same language. Indeed, even our language
for speaking about our own feelings or emotions cannot
be any better than our language for speaking about other
people’s feelings, despite the fundamental asymmetry
between our knowledge of self and other.
In the particular case of mathe�matics, the facts themselves
can be presented or represented by a formal scheme, and
we don’t need to refer to people or their cultures. But
there is more to the practice of mathe�matics than formal
definitions! As well as the statements of mathe�matics, there
are also mathe�maticians, who are cultured people that find
a palpable weight in their use of symbols. As is the case
with all kinds of language, we can sketch out definitions
306 MATHEMATICAL THOUGHT

in the abstract, but meaning can arise only in the context

of life, when flesh and blood human beings engage the
world with words.
People use mathe�matical concepts, and in my view our
very sense of reality is shaped by the fact that our words
can satisfy (or fail to satisfy) the motives that we hold. It
is manifestly the case that linguistic expressions can be
effective in ways that we ourselves are in a position to
confirm: we have enough in common for language to work.
For example, we might want to be successful in a highly
instinctive game called ‘name that object’, and more sophis-
ticated motives can likewise be fulfilled.
When it comes to assessing another person’s matheÂ�
matical work, we look to our definitions to answer the
basic question: ‘Has the ritual been performed correctly?’
However, mathe�matics involves much more than simply
using symbols in the same way as our peers, as we should
also consider the radically matheÂ�matical question, ‘What
is essential to the workings of my ritual, and what is simply
arbitrary?’ Fixed rules unfold as they must, but math is
more than a correctness of symbolic form, as such languages
may be fit for human purpose. In particular, the invention
of number words has given us a linguistic technology that
is genuinely capable of elucidating plurality: a presumably
ancient goal.
Our sense of mathe�matical purpose can be truly deep
and beautiful: grounded in the obvious, but endlessly other-
worldly. As I hope my book has shown, mathe�matical
arguments can be deeply striking, and they play a crucial
role in our best attempts at comprehending the world.
Furthermore, as esoteric as it may seem, I am convinced
that the philosophical contemplation of mathe�matical prac-
tice is profoundly worthwhile. A shift in our philosophy
will not change the facts themselves, but our attitude
towards the factual helps to shape our lives, and what
could be more important than that? With that point in
 LIVED EXPERIENCE AND THE NATURE OF FACTS307

mind, I cannot think of a saner final comment than these

lines from Circles, by Ralph Waldo Emmerson: ‘Every
ultimate fact is only the first of a new series. … No facts
are to me sacred; none are profane; I simply experiment,
an endless seeker with no past at my back.’
 Further Reading

The following lists are arranged by subject matter, with

more accessible books placed at the beginning of each list,
while more scholarly tomes are at the end of each list.

Introductions to Higher Mathematics

Alex’s Adventures in Numberland, by Alex Bellos
(Bloomsbury, 2010).
Imagining the Numbers (Particularly the Square Root of
Minus Fifteen), by Barry Mazar (Penguin, 2003).
Mathematics: A Very Short Introduction, by Timothy
Gowers (Oxford University Press, 2002).
The Mathe�matical Experience, by Philip J. Davis and
Reuben Hersh (The Harvester Press, 1981).
What is Mathematics? An Elementary Approach to Ideas
and Methods, by Richard Courant, Herbert Robbins
and Ian Stewart (Oxford University Press, 1996).
Concepts of Modern Mathematics, by Ian Stewart (Dover
Publications, 1995).
From Calculus to Chaos: An Introduction to Dynamics, by
David Acheson (Oxford University Press, 1997).

Histories of Mathematics
Number – The Language of Science, by Tobias Dantzig
 FURTHER READING309

(Macmillan Company, 1930, reprinted by Pearson

Education, 2005).
A History of Mathematics, by Carl Boyer (Wiley, 1991).
Euclid’s Window, by Leonard Mlodinow (Penguin, 2001).
Mathematics and Its History, by John Stillwell (Springer-
Verlag, 1989).
Pi in the Sky: Counting, Thinking and Being, by John
Barrow (Penguin, 1993).
Greek Mathe�matical Thought and the Origin of Algebra,
by Jacob Klein (Cambridge University Press, 1968).
The Crest of the Peacock: Non-European Roots of
Mathematics, by George Gheverghese Joseph (Penguin,
1991).
A Concise History of Mathematics, by Dirk Jan Struik
(Dover Publications, 1987).
Mathe�matical Thought from Ancient to Modern Times,
by Morris Kline (Oxford University Press, 1972).

Philosophy of Mathematics
Thinking about Mathematics, by Stewart Shapiro (Oxford
University Press, 2000).
What is MatheÂ�matical Logic?, by J.  N. Crossley, et al.
(Dover Publications, 1972).
Naturalism in Mathematics, by Penelope Maddy (Oxford
University Press, 1997).
Proofs and Refutations, by Imre Lakotos (Cambridge
University Press, 1976).
Where Mathematics Comes From, by George Lakoff and
Rafael Nunez (Basic Books, 2000).
Philosophy of Mathematics (Selected Readings), edited by
Paul Benacerraf and Hilary Putnam (Cambridge
University Press, 1983).
18 Unconventional Essays on the Nature of Mathematics,
edited by Reuben Hersh (Springer, 2006).
Reasons Nearest Kin, by Michael Potter (Oxford University
Press, 2000).
310 MATHEMATICAL THOUGHT

Tractatus Logico-Philosophicus, by Ludwig Wittgenstein.

I recommend reading this with Wittgenstein’s Tractatus,
by Alfred Nordmann (Cambridge University Press,
2005).
Philosophical Investigations, by Ludwig Wittgenstein
(Wiley-Blackwell, 4th Edition, 2009). I recommend
reading this with Wittgenstein’s Philosophical
Investigations: An Introduction, by David G. Stern
(Cambridge University Press, 2004).
Wittgenstein’s Lectures on the Foundations of Mathematics
(University of Chicago Press, 1939), Remarks on the
Foundations of Mathematics (Blackwell, 1978), and On
Certainty (Blackwell, 1993) are also well worth reading.

Symmetry, Geometry and Topology

Finding Moonshine, by Marcus du Sautoy (Harper
Perennial, 2008).
Fearful Symmetry, by Ian Stewart and Martin Golubitsky
(Penguin, 1992).
The Ambidextrous Universe, by Martin Gardner (Penguin
Books, 1964).
Mathematics and Optimal Form, by Stefan Hildebrandt
and Anthony Tromba (Scientific American Books, 1985).
Symmetry, by Hermann Weyl (Princeton University Press,
1952).

Gödel’s Theorems and Related Results

Gödel, Escher, Bach: An Eternal Golden Braid, by Douglas
R. Hofstadter (Basic Books, 1979).
Gödel’s Proof, by Ernest Nagel and James R. Newman
(New York University Press, Revised 2002).
Forever Undecided, A Puzzle Guide to Gödel, by Raymond
Smullyan (Random House, 1987).
On Gödel, by Jaakko Hintikka (Wadsworth Publishing,
1999).
 FURTHER READING311

Mathe�matical Biology and Emergent Phenomena

Nature’s Numbers, by Ian Stewart (Phoenix, 1998).
Emergence, by Steven Johnson (Penguin, 2001).
The Self-Made Tapestry: Pattern Formation in Nature,
by Philip Ball (Oxford University Press, 2001).
On Growth and Form, by D’Arcy Thompson (Cambridge
University Press, 1961).
Models of Biological Pattern Formation, by Hans Meinhardt
(Academic Press, 1982).
 Acknowledgements

My way of thinking about thinking has been through

matheÂ�matics, but I have also been influenced by my father’s
love of Wittgenstein. I owe countless intellectual debts,
but I owe particular thanks to my friends and family for
their love and support. I would also like to thank John
Woodruff and Hugh Barker for their helpful comments,
as well as the many mathe�maticians, historians, philoso-
phers and popular science writers who made this book
possible. For the sake of clarity and brevity I have not
always been able to credit the people whose work has
influenced my own, and I would like to apologize to anyone
who feels slighted by such an omission.
I might be described as a philosopher, a mathematician
or a theoretical biologist, but I would not claim to be a
historian. There is a list of mathe�matical histories at the
back of this book, and I would also like to recommend
an excellent online source: the University of St Andrew’s
website – www-history.mcs.st-and.ac.uk. Any historical
errors are my own, but if my account is faithful to actual
events, it is thanks to the scholarship of others. Finally, I
would like to dedicate this book to the memory of James
Wilson: he helped me stop and question what was obvious.
 Index

A differential equations
abacuses 28, 108 149–50
absolute geometry 182–3 Diophantus 105 (see
agriculture, development also Diophantine
of 27–8 equations and
Ahmes Papyrus 31 �definitions)
al-Biruni 125 Fermat’s Little
al-Haytham, Ibn 148 Theorem 112–14, 118
Al-Jabr W’al Mûqabalah Galois Theory 86–7
(al-Khwarizmi) 105 mathematical padlocks
al-Khwarizmi 105–7 114–18
Alexander the Great 53, 101 Pell equations 104
Alexandria, Egypt 53, 61–2 quadratic equations
algebra 109–11 84–6, 104
al-Khwarizmi and algorithms, Euclidean 56–8
Arabian mathematics Almagest (Claudius
104, 106–7 Ptolemy) 62
ancient cultures 30–1, analysis, origins of
33 Â�mathematical 119–24
Cartesian coordinates applied mathematics 41,
129–33 182
combining geometry Arabian mathematics
and 129–30 104–8, 109
314 MATHEMATICAL THOUGHT

Archimedes 58–61, 62, C

119–22, 141 calculus 119, 141–51
architecture 55 Cantor, George 206–8, 228
Aristotle 125, 126–7, diagonal argument
199–200, 215–16 211–14
armillary spheres 108 Pairs 206–11
associativity and Cartwright, Nancy 276
Â�symmetrical groups Catholic Church 62–3
189–90 Cauchy, Augustine 226
astronomy 33, 62, 122–3, Cavalieri, Bonaventura 148
125–6, 139–40 cave art 25
Atto, Bishop of Vic 108 chaos theory 277–80
axiomatic set theories children 13, 18
228–31 China 28, 29–31, 102, 107
Christians, European
B 107–8
Babylonians, Ancient 32–3, Church, Alonzo 237
37, 42–3, 67, 71, 107, 109 Church’s thesis 239,
Bacon, Roger 1 242–3
Banach-Tarski paradox 291 clocks 127–8
Begriffsschrift ‘Concept closure and symmetrical
Script’ (Frege) 223 groups 189
Bernoulli, Jacques 205 cognition, human 16–20
biology, theoretical 280–91 creation of symbols
Black, James 273 20–1
Boethius, Anicius 62–3 legible patterns 24–6
Bolzano, Bernhard 226 composite numbers
Bombelli, Raphael 81 defined 47
Boole, George 222 computers
Brahmagupta of Gujarat earliest programmable
103 236
Brâhmasphutasiddhânta nature of programs
(Brahmagupta) 104 237–8
bridges of Königsberg Turing cards 238–47
154–6, 165 continued fractions 81–6,
Brouwer, L. E. J. 234 91–2
 INDEX315

continuous functions 162, E

164 Egypt, Ancient 31–2, 33,
coordinates, Descartes and 34, 42–3, 53, 61–2
130–1, 134 Einstein, Albert 167, 185,
counting 23–4, 25–7 259
curves/ curvature 120–1, Elements (Euclid) 53–4
131–2, 134, 175–80, 186 Emmerson, Ralph Waldo
see also calculus 307
Enigma machine 235–6
D equations see algebra
Davis, Martin 253, 255 Euclid 68
debate, ancient Euclidean algorithms
Â�mathematical 34–5 56–8
decimal numbers 79–81, Euclidean geometry
98–104, 108–9 53–6, 168–70, 175,
decision problems 243–7 182–3, 184–6,
semi-decidable 217–18
Â�problems 247, 249–52 Eudoxus 61, 62, 70, 73–6
Dedekind, Richard 78, Euler, Leonhard 77, 80,
136–7, 211 147, 152–6, 165, 173,
Dedekind Cut 78–80, 291
83 Euler numbers 157–8,
Descartes, René 78, 162–6
130–2 Euler’s Polyhedral
deterministic language Formula 164–6
games 238–42 evolution, Darwin’s theory
diagonal argument, of 280–1
Cantor’s 211–14 evolution, human 14–16
differential equations see also language,
149–50 Â�development of
Diophantine equations and excluded middle, law of
definitions 248–9, 253–6, the 220
260–5
Diophantus 105 F
Dummett, Michael 234–5 factuality, mathematical
Dyson, Freeman 280 37–9, 40
316 MATHEMATICAL THOUGHT

Fermat, Pierre de 112, 148 132–3, 134 (see also

Fermat’s Little Euclidean geometry)
Theorem 112–14, 118 combining algebra and
Fibonacci, Leonardo 90–1, 129–30
98, 109 curvature of space
Fibonacci Sequence 90–1, 175–80
95, 256 Euclidean 53–6, 168–70,
fractions 31, 65–6, 73–7, 175, 182–3, 184–6,
78–81, 210 217–18
continued 81–6, 91–2 hyperbolic 180–1
Fraenkel, Adolf Abraham non-Euclidean 40,
230 172–86
Frege, Gottlob 42, 208, Plato 48–53
223–4, 227–9 theory of relativity
functions 147–8, 149–50, 185–6
162, 164 see also Archimedes
Fundamental Theorem of Gerbert of Aurillac 108
Arithmetic 67–9 Germain, Marie-Sophie
Fundamental Theorem of 179–80
Calculus 141–6 Gierer, Alfred 287
Furtwängler, Philip 258 global inhibition, patterns
and 287–8
G Gödel, Kurt 229, 258–60,
Galileo Galilei 125–7, 271
209 see also Incompleteness
Galois, Evariste 86–7 of Arithmetic
Gauss, Carl Freidrich 31, Theorem
121, 172–3, 176–8, Goethe, Johann Wolfgang
185 van 232
genus of shape 165 golden ratio 85, 91–6
geometry 167–8 graphs 70, 130, 134, 141–2,
absolute 182–3 156–8
agricultural develop- see also coordinates,
ment 27–8 Descartes and
Ancient Greece 35, Gravity, Law of 138–40,
48–53, 106, 107, 186
 INDEX317

Greeks, Ancient 28, 34–5, Inca civilization 28–9

42–4, 85, 106, 107, Incompleteness of
131–3, 134 Arithmetic Theorem
Archimedes 58–61 259–60, 261–72
Euclidean algorithms Indian mathematics 102–3,
56–8 107, 125, 205
Euclidean geometry infinite sequences 119–24
53–6, 169–70, 217–18 infinitesimal calculus see
Plato and symmetric calculus
thought 48–53 infinity 199–200, 205–6,
Pythagorean science 209–14, 228–9
44–8 Institutio Arithmetica
see also Pythagoras’ (Boethius) 62–3
Theorem Introduction to Arithmetic
group theory, symmetry (Nicomachus) 62
and 189–93 inverse property of
symmetrical groups
H 190–1
Halley, Edmond 140 irrational numbers 70, 72,
Halmos, Paul 119 87–90
Hardy, G. H. 2, 67, 304
Heiberg, J. L. 122 J
Hekademeia, Greece 48 Jain cosmology 205
Heraclitus of Ephesus 300 Jyesthadeva 123
Hilbert, David 147, 151,
208, 233–5, 237 K
Hilbert’s program 271 Kant, Immanuel 13
Hintikka, Jaakko 226 Kepler, John 139
Hippocrates of Chios 43 Klein, Jacob 45
Hooke, Robert 140 Königsberg bridges,
Hudde, Waveren 148 Prussia 154–6, 165
hyperbolic geometry 180–1 Kronecker, Leopold 207

I L
identity and symmetrical Lagrange, Joseph 129–30
groups 190 Lakoff, George 19
318 MATHEMATICAL THOUGHT

language, development of Madhava of Sangamagrama

14–16, 22–3 122–3
number words 23–4, magic squares 29–30
26–7 Maistre, Joseph de 64
Law of Gravity 138–40 Matiyasevich, Yuri 256
Le Corbusier 85 Matiyasevich’s Theorem
left and right, differences 253–8
of 193–8 Mayan civilization 28
Leibniz, Gottfried measurement and counting
Wilhelm 141, 143, 64–7
144–5, 147, 222, 233 measurement in Ancient
Lhuilier, Simon 165 Egypt 31–2
Liber Abaci (Fibonacci) measuring time 127–8
90, 98, 109 mechanical deductions
Lichtenberg, Georg 221–2
Christoph 215, 253 Medici Bank 109
limit cases 123–4, 226 Medieval Europe 107–8
linear order and number Meinhardt, Hans 282, 287
lines 134–7 Mesopotamia, Ancient 28,
Listing, Johann 196 32–3, 99, 100
Lo-Shu 29–30 The Method (Archimedes)
Lobachevsky, Nikolai 121–2
Ivanovich 172 mirror images 193–8
local activation, patterns Möbius, August Ferdinand
and 287–8 196
logic 215–20 Möbius Strip 196–8
logic machines 221–2 MPDR Theorem 253–8
predicate calculus multiplication tables 191–3
224–7 Mumford, Lewis 128
quantifiers and music 47–8
Â�properties 222–5
see also set theory N
Lucas, Edouard 90 nature and the golden
ratio 95–6
M negative numbers 102
Ma’at 34 networks, drawing 156–8
 INDEX319

Newman, Max 236 Plato 48–53

Newton, Isaac 129, 134, Platonic solids 51–3,
137–41, 143–4, 145, 274 158–61, 164
Nicomachus of Geresa 62 Platonists 297
Nine Chapters on the Art Playfair, John 169–70
of Calculation 30 Plimpton 322 71
Noble, Denis 276 Plutarch 58, 59
non-Euclidean geometries Poincaré, Henri 156,
40, 172–86 161–4, 181, 207
number lines, linear order Poncelet, Victor 171–2
and 134–7 positional numeration
number words 23–4, 26–7 99–100, 101–2
Núñez, Rafael 19 predicate calculus 224–7
prime numbers defined 47
O Principia Mathematica (B.
one-to-one mapping Russell and A. N.
208–11 Whitehead) 228, 233, 271
Oresme, Nicole 134, 141 Principia Mathematica (I.
Otto II, Holy Roman Newton) 140
Emperor 108 proof by induction 202–5
proof of exhaustion 61
P proportion and Euclidean
padlocks, mathematical algorithms 56–8
114–18 Ptolemy 62
Pascal, Blaise 9, 171, Ptolemy I 53
200–2, 205, 292 pure mathematics 41, 182
patterns Putnam, Hilary 253, 255
early humans and Pythagoras Theorem 3–4,
legible 24–6 32, 33, 70–2, 169, 184,
theoretical biology and 202
286–8 Pythagoreans 44–8
Peano, Giuseppe 184, 254
Pell equations 104 Q
Pi 38–9, 59–61, 75–6, 89 quadratic equations 84–6,
plants and the golden 104
ration 95–6 quipus 28–9
320 MATHEMATICAL THOUGHT

R Russell, Bertrand 228, 229,

Ramanujan, Srinivasa 271
89–90
rational numbers 66, 73, S
83, 87, 210 Saint Augustine 63–4
see also fractions science and mathematical
rational quadratic models 273–7
Â�equations 85–6 theoretical biology
real numbers, concept of 283–91
73–4, 76–8 Second World War 235–6
see also fractions Seidenberg, Adrian 23
reasoning, deductive semi-decidable problems
16–18 247, 249–52
see also logic set theories 228–31
recording numbers 28–9 Shapiro, Stewart 302
recurring decimals 79–81 Siddhantas 104
Reductio Ad Absurdum social structure, counting
67–70 and 23, 27
relativity, theory of 185 Somayaji, Nilakantha
religion and numbers 45–6, 76–7
62–3 space-time 186
Riemann, Bernhard 172–3, spirituality and numbers
184–5 45–6
right and left, differences statistics 201
of 193–8 Stillwell, John 71–2, 107
rituals, counting and 21–2, Stonehenge, Wiltshire
23 29
Robinson, Julia 253, 255 Sumerians 99, 100, 101
Rochefoucauld, François sunya 103
de La 152 Sylvester II, Pope 108
Roman numerals 101 symbols, creation of 20–1,
Romans 28, 61–3 25–6
Royal Library of number 28–9
Alexandria 61 symmetric form, Plato and
RSA system 116–18 48–53
 INDEX321

symmetry 186–93 Lanka 24

Syracuse, Greece 59 vibration, study of 150–1
Viète, François 77, 110,
T 130
Tarikh Al-Hind (al-Biruni)
125 W
taxation 33 Wallis, John 77, 170
Thales of Miletus 42–3, 44 Wave Equation 150
Theaetetus 51 Weizenbaum, Joseph
time 127–8 128–9
topology 162–6 Whitehead, Alfred North
trade in early civilizations 98–9, 228, 271
33, 42 Wittgenstein, Ludwig
transinfinite numbers 292–3, 294
207–8 Wren, Sir Christopher
trigonometry 62, 123, 153 140
Turing, Alan 235–6, 237,
286–7 Z
Turing cards 238–47 Zeno of Elea 200
Zermelo, Ernst 230
V zero 99–101, 102–4
Vedda tribesmen, Sri ZF set theory 230