Springer Undergraduate Mathematics Series

Daniel Grieser

Exploring
Mathematics
Problem-Solving and Proof
Springer Undergraduate Mathematics Series

Advisory Board
M. A. J. Chaplain, University of St. Andrews
A. MacIntyre, Queen Mary University of London
S. Scott, King’s College London
N. Snashall, University of Leicester
E. Süli, University of Oxford
M. R. Tehranchi, University of Cambridge
J. F. Toland, University of Bath
More information about this series at http://www.springer.com/series/3423
Daniel Grieser

Exploring Mathematics
Problem-Solving and Proof

123
Daniel Grieser
Institut für Mathematik
Carl von Ossietzky Universität Oldenburg
Oldenburg
Germany

ISSN 1615-2085 ISSN 2197-4144 (electronic)

Springer Undergraduate Mathematics Series
ISBN 978-3-319-90319-4 ISBN 978-3-319-90321-7 (eBook)
https://doi.org/10.1007/978-3-319-90321-7

Library of Congress Control Number: 2018939472

Mathematics Subject Classification (2010): 00-01, 00A07, 00A09, 97D50

Translation from the German language edition: Mathematisches Problemlösen und Beweisen by Daniel
Grieser, © Springer Fachmedien Wiesbaden 2013. All Rights Reserved.

© Springer International Publishing AG, part of Springer Nature 2018

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.

Printed on acid-free paper

This Springer imprint is published by the registered company Springer International Publishing AG
part of Springer Nature
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
For Ricarda and Leonard
Preface to the English edition

The classical curriculum for university mathematics education has

emphasized calculation and – on a higher level – proofs and system-
atic development of mathematical theories. In recent years there has
been a growing interest in supplementing this with a problem-solving
approach. It has become clear that this is not only fun but also very
useful as preparation for understanding the mathematical theories,
these tremendous advances that mathematics has made over the cen-
turies. These new developments have occurred in Britain, the United
States, Germany and certainly other countries as well.
In 2011, the University of Oldenburg introduced a new course
into the curriculum: ‘Mathematical problem-solving and proving’
(Mathematisches Problemlösen und Beweisen). It was aimed at first-
year students who have just finished Gymnasium (roughly equivalent
to A-levels in England and high school plus epsilon in the US) and
are starting a degree in mathematics or mathematics education. This
course has been very successful, both in bridging the gap between
high school and university, and in providing a fun introduction to the
higher mathematics taught at university. Similar courses have sprung
up at other universities in Germany, and in Britain and the US, and at
other places people are discussing the possibility of introducing them,
or at least introducing elements of such an approach into existing
courses.
This book arose from my teaching the course twice in Oldenburg,
and since many colleagues in English-speaking countries have ex-
pressed a great interest in it I decided to translate it into English.
There are many other excellent books on problem-solving. What
sets this book apart is that it starts at a very elementary level, but
then quite explicitly tries to be a bridge to higher mathematics: the
emphasis is on your discovery of mathematics by solving problems,
but along the way concepts, notation and terminology of higher
mathematics (e.g. sets and mappings) are introduced. In addition,
most chapters have a section with additional material (titled Going

7
8 Preface to the English edition

further) which provides a preview of where similar ideas are used in

advanced mathematical disciplines.
Contents

Introduction 13

1 First explorations 23
1.1 Cutting up a log . . . . . . . . . . . . . . . . . . . . . . . 23
1.2 A problem with zeroes . . . . . . . . . . . . . . . . . . . 24
1.3 A problem about lines in the plane . . . . . . . . . . . . 28
1.4 Toolbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

2 Recursion – a fundamental idea 41

2.1 Recursion in counting problems . . . . . . . . . . . . . 41
2.2 The number of subsets . . . . . . . . . . . . . . . . . . . 44
2.3 Tilings with dominoes . . . . . . . . . . . . . . . . . . . 50
2.4 Solving the Fibonacci recurrence relation . . . . . . . . 54
2.5 Triangulations . . . . . . . . . . . . . . . . . . . . . . . . 61
2.6 Toolbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

3 Mathematical induction 71
3.1 The induction principle . . . . . . . . . . . . . . . . . . 71
3.2 Colourings . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.3 Toolbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

4 Graphs 83
4.1 The Euler formula for plane graphs . . . . . . . . . . . 83
4.2 Counting in two ways for graphs . . . . . . . . . . . . . 90
4.3 Handshakes and graphs . . . . . . . . . . . . . . . . . . 94
4.4 Five points in the plane, all joined by edges . . . . . . . 95

9
10 Contents

4.5 Going further: Euler’s formula for polyhedra, topol-

ogy and the four colour problem . . . . . . . . . . . . . 98
4.6 Toolbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

5 Counting 107
5.1 The basic principles of counting . . . . . . . . . . . . . 107
5.2 Counting using bijections . . . . . . . . . . . . . . . . . 117
5.3 Counting in two ways . . . . . . . . . . . . . . . . . . . 123
5.4 Going further: double sums, integrals and infinities . . 128
5.5 Toolbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

6 General problem solving strategies 139

6.1 General problem solving strategies . . . . . . . . . . . . 139
6.2 The diagonal of a cuboid . . . . . . . . . . . . . . . . . . 143
6.3 The problem of trapezoidal numbers . . . . . . . . . . . 145
6.4 Going further: sum representations of integers . . . . . 153
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

7 Logic and proofs 157

7.1 Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
7.2 Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176

8 Elementary number theory 181

8.1 Divisibility, prime numbers and remainders . . . . . . 181
8.2 Congruences . . . . . . . . . . . . . . . . . . . . . . . . . 186
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191

9 The pigeonhole principle 195

9.1 The pigeonhole principle, first examples . . . . . . . . . 195
9.2 Remainders as pigeonholes . . . . . . . . . . . . . . . . 199
9.3 An exploration: approximation by fractions . . . . . . . 201
9.4 Order in chaos: the pigeonhole principle in
graph theory . . . . . . . . . . . . . . . . . . . . . . . . . 212
9.5 Toolbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Contents 11

10 The extremal principle 219

10.1 The general extremal principle . . . . . . . . . . . . . . 220
10.2 The extremal principle as problem solving strategy, I . 225
How to use the extremal principle . . . . . . . . . . . . 228
10.3 The extremal principle as problem solving strategy, II . 236
10.4 Going further: optimisation, mirrors and billiards . . . 240
10.5 Toolbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248

11 The invariance principle 253

11.1 The invariance principle, first examples . . . . . . . . . 253
11.2 How to use the invariance principle . . . . . . . . . . . 258
11.3 Further examples . . . . . . . . . . . . . . . . . . . . . . 260
11.4 Going further: knots, conserved quantities, and why
you should care about proofs of impossibility . . . . . 270
11.5 Toolbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275

A A survey of problem-solving strategies 281

B Basics on sets and maps 287

List of symbols 293

Glossary 295

Lists of problems, theorems and methods 303

Hints for selected exercises 305

References 317
Introduction

Tell me and I will forget.

Show me and I will remember.
Involve me and I will understand.
(Lao Tse)

With this book I invite you to come on a voyage of discovery. You

will discover mathematics from a completely new perspective: not
as a collection of formulas and rules, but as a world which you can
explore for yourself, in which you can develop your own ideas and
unearth hidden treasures. Don’t worry: you are well equipped for
your journey, and along the way you will add new tools to your
equipment that will help you master difficult stages. I will be your
guide. You can decide for yourself whether you want to travel alone,
accept a few hints, or let yourself be comfortably guided through the
terrain.
At the end you will return from your voyage enriched. Not only
will you have gained a new view of mathematics, but you will also
take home a wealth of experiences.
Three themes will recur throughout our voyage:

1. Mathematical problems and problem-solving strategies

2. Mathematical proofs: why do we need them, how do we find

them, how do we write them?

3. General ideas of mathematics

Problems and problem-solving strategies

Problems are the soul of mathematics.

The problems of how to calculate areas and predict the movement
of objects led to the invention of the calculus in the 17th century.
Fermat’s problem, whether the equation x n + yn = zn has positive
integer solutions for n > 2, has stimulated number theory up to the

13
14 Introduction

present day. The four colour problem, whether four colours suffice to
colour every conceivable map assuming that neighbouring countries
always get different colours, led to a new mathematical theory in the
19th century: graph theory. These are big mathematical problems.
But there are also countless smaller interesting problems which are
accessible for everyone, which you can explore for yourself and thus
experience what mathematics is about. You will find such problems
in this book.
Problem-solving is fun and creative.
Initially you are in the dark. You look at examples, make a sketch,
observe patterns, and by and by you realise what matters and what
doesn’t. Slowly the darkness lifts, you feel your way forward, develop
ideas, and suddenly: Eureka – I’ve found it! Understanding is deeply
satisfying. Being creative makes you happy.
Problem-solving can be learned.
During your voyage you will learn many problem-solving strategies.
Some of them are so general that they can also be used outside of
mathematics. Others are specific to mathematics. With every problem
that you solve you develop your creativity and enlarge your pool of
experience.
Problem-solving makes you curious about mathematical theories.
Theories give answers. You will really appreciate an answer only if
you have asked the question beforehand, and even more if you have
tried to answer it yourself and experienced difficulties. If you learn
new theories (like algebra or analysis) in this way, then you will be
able to use them well, and advance beyond them.

Proofs

Proofs are the heart of mathematics.

Mathematical proof is quite extraordinary: what is proved today is
true – today, tomorrow and in a thousand years. This distinguishes
mathematics from all other sciences. Many scientific theories were
considered valid for centuries but then had to be corrected. Mathemat-
ics also advances, new connections of hitherto unconnected worlds
are discovered all the time, problems are solved which had seemed
intractable for years; but what is proved once will always be true.
Introduction 15

Proofs tame infinity.

By experimenting you may discover that you can find more and more
prime numbers, no matter how many you have found already. But
does this continue for ever? Are there infinitely many primes? Only
a proof can give certainty.
Proofs give certainty.
Many a supposed fact turns out to be a fallacy – and if you have
experienced this then you will appreciate what proofs do for you.
Without a proof you can never be sure.1
Proofs help you to understand.
When you have thought in detail about why a mathematical assertion
is true, when you have removed any doubt by logical arguments, then
you will understand the assertion itself better, and also you will be
better at explaining it to others. And you will remember it better.
Finding proofs is problem-solving and therefore creative. And it can be
learned.
A proof is a logically complete argument. But how do we find a proof?
Here we need to be creative. Finding a proof is to mastering logic
as drawing a picture is to knowing about colours, or as composing a
symphony is to knowing about musical notes and harmonies. And
just like a picture or a symphony, a proof can be beautiful. There is
no general recipe for finding proofs, but just as for general problem-
solving there are recurring patterns and ideas. Getting to know these
will help you get better at finding proofs. You will find many of these
patterns in this book.

General ideas of mathematics

Often mathematics is divided into subdisciplines: geometry, algebra,

analysis and so on. However, many ideas occur in all disciplines. In
this book you will get to know such ideas by simple examples. You
will also learn about general methods of scientific work. You may be
surprised to notice that many of them will be familiar to you. We
merely name them so you can use them systematically.

1 SeeExercises E 1.10 and E 5.24 and the beginning of Chapter 7.2 for good
examples of this.
16 Introduction

Among the scientific methods are the cycle of exploring, making

hypotheses (also called conjectures in mathematics) and investigating
them systematically (in mathematics: proof or disproof); introducing
concepts and notation; and identifying essential aspects of a problem
and neglecting inessential ones (abstraction).
Among the general mathematical ideas are the extremal principle,
the invariance principle and the principle of counting in two ways.
You will learn about these ideas here as problem-solving strategies
and use them to find out surprising things. But when you delve
deeper into mathematics then you will meet these ideas again and
again in different guises. They will guide your way in the vast
landscape of mathematics.

Who is this book for?

This book is for everyone who likes mathematics and who likes to
think about problems that cannot be solved by simply applying a
given method. For all who want to improve their problem-solving
skills. For mathematically interested high school or secondary school
students. For students in colleges and universities who would like to
get a different view of mathematics, supplementing their standard
maths courses. For teachers in high or secondary schools, colleges and
universities who want to teach problem-solving in a more systematic
way than just by giving homework problems, maybe even to offer a
course that is method-oriented rather than topic-oriented. For leaders
of maths clubs or math circles.
As prerequisites you only need to know a bit about numbers, basic
geometry, how to rearrange equations. You learn these ideas quite
early on at high or secondary school. But even if you know some
higher mathematics already you will get your money’s worth: at many
places in the book you will find material on how the elementary ideas
introduced here are used in higher mathematics, including present-
day research.

Contents

The core of the book is the problems. Lots of problems, their investiga-
tion and solution. In the investigation we approach each problem step
Introduction 17

by step: What are we looking for? What could be a route of approach?

Not every route that we try will lead to a solution; sometimes we need
to backtrack and start again. Here you experience how mathematics
is born, invented, developed. While discovering mathematics you
learn how mathematicians think. Here you learn problem-solving
strategies. These are collected in the Toolbox sections. In Appendix A
you will find an overview of all the strategies introduced in the book.
What kind of problems do you find in this book? Mathematical
problems for which no straightforward method of solution offers
itself. Problems where you need an idea. Problems that lead to an
interesting mathematical concept. Some chapters introduce an area
of mathematics, for example graph theory in Chapter 4 or number
theory in Chapter 8. These topics are also introduced via a sequence
of problems, and at places where it is natural to state a theorem
and its proof, the proof is developed from the point of view of a
problem-solver.
Here is a more detailed outline. In Chapters 1 and 2 you will find
problems where you count things. Counting is the most basic mathe-
matical activity. By learning to advance step by step you will soon
have the wonderful experience of discovering mathematics yourself.
An important technique for counting is to use recurrence relations,
which are discussed in Chapter 2. The basic idea, called recursion,
is to reduce a problem to a smaller problem of the same kind. This
is also behind the principle of mathematical induction introduced in
Chapter 3. Using induction you will prove Euler’s famous formula
in Chapter 4 and then use it to show a remarkable impossibility
result: you cannot connect five points in the plane pairwise so that
the connecting curves don’t intersect. That we can prove such an
assertion rigorously is actually quite amazing: how can we control
the infinitely many ways to draw the points and curves? After this
excursion into graph theory we return to counting in Chapter 5, this
time formulating general systematic counting principles which are
often useful.
Chapters 6 and 7 are central: Chapter 6 introduces general problem-
solving strategies, that is, strategies which are useful beyond mathe-
matics. Chapter 7 is about logical foundations and the most important
general types of proof. These are illustrated by many examples. Chap-
ter 8 is about number theory, that is, prime numbers, divisibility etc.
18 Introduction

Here you can use the various types of proof to answer interesting
questions about numbers. Number theory also provides useful tools
and pretty examples for the following chapters. In Chapter 9 you
learn about the pigeonhole principle, a simple idea which spawns
amazing consequences when used aptly. Chapters 10 and 11 intro-
duce the extremal principle and the invariance principle. These are
versatile tools for solving mathematical problems, and also turn up
again and again in all exact sciences. Here you also learn about the
fundamental notions of permutations and their signature.
Sections titled Going further at the end of Chapters 4, 5, 6, 10 and
11 encourage you to delve deeper into the matters touched upon
in these chapters and give you a glimpse of the higher summits of
mathematics. Occasionally you may find here an unfamiliar term
without detailed explanation, but you should read on, just to get an
idea of what else mathematics has to offer. The references at the end
of the book guide you to literature that will be useful for further
exploration.
The mathematical topics treated in the book are elementary, that
is, they can be understood with very few prerequisites. However, we
take a higher perspective than you may be used to: often problems
are formulated in greater generality than is common at school, we
pay attention to logically complete arguments, and although the
tone is generally informal, we use modern mathematical terminology,
in particular the language of sets and maps. This is explained in
Appendix B.
The chapters are mostly independent of each other, so you may
read them in a different order. However, later chapters tend to be
more demanding than earlier chapters.

Hints for using the book

You learn problem-solving by solving problems, and by understand-

ing other people’s solutions. You learn proof by proving, and by
carefully reading and understanding proofs. The problems in the
text and the exercises at the end of each chapter give you ample
opportunity for this. Be an active reader! After reading a problem,
first try to solve it yourself. After reading part of the solution, put the
book aside and think about how to go on. The symbol
Introduction 19

Think about it!

is meant to remind you of this. Whenever you read a statement, ask

yourself whether you are convinced that it is true. Keep paper and
pencil handy. Ask yourself: why do it like that, couldn’t I do it better?
Explain your solution, or the solution you have just read, to a
friend. Let him or her play the devil’s advocate, who doesn’t let you
get away with the slightest imprecision, who tries to find gaps in your
argument, who asks after each step: “But what if . . . ”. After a while
you will learn to be your own devil’s advocate.
Sometimes you will have other ideas for solutions than what is
given in the text. Every problem has many solutions, and also the
same idea can be formulated in many ways. Try to work out whether
your solution has the same core idea or whether it is completely
different.
The exercises at the end of each chapter are a particularly important
part of the book. Work on them. It will pay off. You will find hints for
some of the exercises at the end of the book. But you will profit more
from an exercise if you don’t immediately succumb to the temptation
to look them up.
Each exercise has been assigned a level of difficulty. This should
be understood as a rough guideline, as there is no objective way to
judge difficulty. Whether and when you find a solution depends on
many factors. The levels are as follows:
1 – easy, often solvable in your head
2 – should be doable when you have worked through the chapter
3 – requires more commitment and ideas
4 – difficult

Most problems in the book are treated in four steps:

1. Understanding the problem

2. Investigating the problem

3. Writing up the solution properly

20 Introduction

4. Review

These are similar to the four steps of problem-solving introduced by

Pólya in his classic book How to solve it (Pólya, 2014). Steps 1. and
2. are closely interwoven and carried out in the sections marked by
the symbol G . The solution is marked by the symbol !, the review
by the symbol . The end of each of these sections is marked by the
same symbol, and the end of a proof is marked q. e. d. (quod erat
demonstrandum = which was to be proven).

Notes for the teacher

This book arose from a course whose main aims were to let students
see mathematics from a new perspective (not rules to be followed
but a world to be discovered), to give them confidence that they can
discover mathematics for themselves and to lead them to appreciate
proofs.2 Along the way students acquire tools for problem-solving,
get to know the main types of proofs, learn to write solutions and
proofs properly, learn about fundamental ideas that occur everywhere
in mathematics, and also pick up important concepts like graphs,
congruences, permutations and their signatures. These are introduced
in informal, often playful contexts, so they can be shared with friends
and will be remembered more vividly than if learned in more abstract
ways.
If you want to teach such a course my main piece of advice is: Less
is more. Don’t try to ‘cover’ a certain curriculum. Take time to solve
problems together with the students in class, to develop ideas for
solutions, to try out different approaches, to suffer the frustration
of getting stuck and enjoy the satisfaction of success together. Share
with the students your experience of how to approach a mathematical
problem, rather than presenting prefabricated solutions. Always start
with an easy problem. There will be enough challenges for everyone.

I am happy to acknowledge the help and valuable feedback that

2 This
course, titled Mathematisches Problemlösen und Beweisen (Mathematical
problem-solving and proving), was introduced in the mathematics curriculum at
Carl von Ossietzky University of Oldenburg, Germany, in 2011, with the goal
of smoothing the difficult transition from high school mathematics to university
mathematics.
Introduction 21

I got from many students, colleagues and readers of the first Ger-
man edition. I want to thank especially the student tutors Stefanie
Arend, Simone Barz, Karen Johannmeyer, Marlies Händchen, Stefanie
Kuhlemann, Roman Rathje, Kathrin Schlarmann, Steffen Smoor and
Eric Stachitz, and my colleague Andreas Defant. Sunke Schlüters
contributed many excellent ideas for additional problems and helped
me with many of the pictures in the book. Sophy Darwin made many
valuable suggestions beyond her thorough language editing. My
deepest gratitude is due to my wife Ricarda Tomczak for continuous
encouragement and innumerable discussions and suggestions.

Now I wish you many joyful hours of solving problems and think-
ing about proofs, of discovering the beauty of mathematics. If you
have any comments or suggestions please write me an email at
[email protected].

Oldenburg,
February 2018 Daniel Grieser
1 First explorations

We begin our journey into mathematics by investigating three prob-

lems. The first one is a simple warm-up exercise, but the other two
require some serious searching before we find a solution. During
this search we will observe ourselves: How do we proceed intuitively
when solving a problem? At the end of the chapter we will collect
our strategies into a toolbox. This will be of great value when we
attack harder problems.

1.1 Cutting up a log

? Problem 1.1
You have a log which is 7 metres long. How long does it take to cut the log
into pieces one metre long, assuming that a single cut takes half a minute?

This is easy! Your solution probably looks something like this.

! Solution
You need six cuts, therefore it takes six times half a minute, that is
three minutes. !
 Review
Let us have a look at our solution. We observe some basic strategies.

J We first found the number of pieces (seven) and then the number
of cuts (six). This is a simple example of an interim goal.

J A typical beginner’s mistake is to calculate like this: Divide 7

metres by 1 metres, therefore we need 7 cuts. But the correct
answer is that 7 pieces require 6 cuts. This is a typical shift by one.
What do you do when you are not sure whether you should shift
by one or not? A sketch will help, see Figure 1.1.

© Springer International Publishing AG, part of Springer Nature 2018 23

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7_1
24 1 First explorations

You should keep an eye out for shifts by one. See exercise E 1.1 for
more examples.

Figure 1.1 The number of cuts is one less than the number of parts

J To make this still clearer, we could simplify the problem: we

replace 7 by 2 or 3 and observe what happens: one cut for 2 metres,
2 cuts for 3 metres etc.

J We recognize a pattern: The number of cuts is always one less than

the number of parts.

J How can we be certain that this is always the case1 ? We need a

proof. This could run as follows.
Every cut has a part immediately to its left. All parts arise in
this way except the last one at the right-hand end. Therefore
the number of parts is equal to the number of cuts plus one.

J We could save some cuts by cutting first into parts 3 and 4 metres
long, putting these two pieces next to each other and then cutting
both simultaneously, and then continuing in a similar manner.
Although loggers certainly wouldn’t do this, it is still an interesting
problem what the smallest number of cuts is in this case. See
exercise E 1.5. 

1.2 A problem with zeroes

Let us try a harder problem.

? Problem 1.2
How many zeroes are at the end of 1 · 2 · 3 · · · · · 99 · 100?

1 Also for a million pieces, for example; we cannot draw a picture for this.

Although the problem was not worded in this generality, it is satisfying to recognize
general patterns.
1.2 A problem with zeroes 25

One writes 100! (in words: one hundred factorial) for the product
1 · 2 · 3 · · · · · 99 · 100. Zeroes at the end of a number we call trailing
zeroes.

Before reading on, think about the problem yourself.

 Investigation
G
We cannot calculate the product, it is too big even for a calculator.2 We
need an idea how to do it a different way. For a moment you might
think the answer is two: the two zeroes of the factor 100. Looking
more closely, you might discover that the product also contains the
factor 10, and also 20, 30 and so on, each contributing another trailing
zero. So we get 11 zeroes. Have we found them all? Can we be sure
that 11 is the solution?
To answer this we need to understand where the trailing zeroes
come from. We could proceed as follows.

 Simplify! This means that we consider the same problem for

smaller numbers than 100, in order to get a feel for the problem.
For example, 2! = 1 · 2 = 2, 3! = 1 · 2 · 3 = 6, 4! = 1 · 2 · 3 · 4 = 24,
5! = 1 · 2 · 3 · 4 · 5 = 120.
 For easier bookkeeping we make a table:

n 1 2 3 4 5 6
n! 1 2 6 24 120 720

The first trailing zero arises with 5!. Why? Write 5! = 1 · 2 · 3 · 4 · 5.

Where does the trailing zero of the result 120 come from? Think
about it!
A trailing zero means that 120 is divisible by 10. Now 10 = 2 · 5,
and both 2 and 5 are factors in 5!. That’s where the zero comes
from. In 4! = 1 · 2 · 3 · 4 there is no factor 5, therefore it has no
trailing zero.
Important insight: A number has a trailing zero if it is divisible by
2 and by 5.
2 Some modern calculators can do this. If yours does, try to make it solve the
same problem for 1000! or 10000!. Or Exercise E 1.6.
26 1 First explorations

 How about two trailing zeroes? Before reading on, try to predict
which is the smallest n whose factorial has two trailing zeroes. Use
the insight above.
To have two trailing zeroes, a number must be divisible by 100,
which is 10 · 10 = 2 · 5 · 2 · 5 = 22 · 52 . So we need two factors 5 and
two factors 2. Let us imagine continuing the table, without actually
doing any calculation. We only need to watch for fives and twos.
Where is the next factor 5? Not in 6, 7, 8, 9, but in 10 since 10 = 2 · 5.
There is a surplus of factors 2; each one of 2,4,6,. . . contributes
at least one. Just two of them would be enough. Therefore 10!
has two trailing zeroes. How does this continue? Try to solve the
problem now.3

Think about it!

 Maybe you recognize the pattern: Let k be any natural number.

For a number to have at least k trailing zeroes it must be divisible
by 10k = 2k · 5k , that is, it must contain k factors 5 and k factors 2.
 So we should count how often the factors 5 and 2 appear in 100!.
Let us first look at the factor 5: Which factors 1, 2, 3, . . . , 100 in 100!
contribute 5? Clearly 5, 10, 15, 20, 25, . . . , 100 do. That’s how many?
If we write these numbers as 1 · 5, 2 · 5, 3 · 5, . . . , 20 · 5 then we see
that there are 20 of them. Is this the solution?
Watch out! 25 = 5 · 5 contributes two factors 5. Which other factors
in 1 · 2 · 3 · · · · · 100 do that? Clearly the multiples of 25, that is, 25,
50, 75, 100. Each of them contributes another 5, that’s 4 additional
fives. (The factor 125 = 53 would contribute two additional fives,
but we don’t get this far in 100!.) All in all the factor 5 appears 24
times in 100!.
How about the factor 2? Clearly, each one of the 50 even numbers
2, 4, . . . , 100 contributes at least one factor 2, so we get more than
24 such factors. But the surplus twos (those beyond the 24th) do
not yield a trailing zero in 100! since they have no partner 5.
3Whenever you see the following symbol, put the book aside, take pen and
pencil and try to work out the problem, or at least the next step, by yourself.
1.2 A problem with zeroes 27

We obtain the solution: 100! has 24 trailing zeroes. 

G
To tell others about our solution we should write it up well. This
will also help us to check it for completeness. This could look as
follows.

! Solution of Problem 1.2

The number of trailing zeroes of 100! is the biggest integer k for which
100! is divisible by 10k . Because 10 = 2 · 5, this is the biggest integer k
for which 100! is divisible by 2k · 5k , that is, by 2k and by 5k .
The factor 5 arises as a factor in the 20 numbers 5,10,15,. . . ,100,
and twice in the 4 numbers 25,50,75,100. Therefore it arises in 100!
precisely 20+4=24 times, so k = 24 is the largest k for which 100! is
divisible by 5k .
The factor 2 arises as a factor in the 50 numbers 2,4,. . . ,100, in some
of them more than once. Therefore it arises in 100! more than 50
times. The precise number does not matter: we only need that the
number is at least 24.
Therefore the largest k for which 100! is divisible by 2k and by 5k is
k = 24. Therefore 100! has precisely 24 trailing zeroes. !
 Review of Problem 1.2
We were able to solve the problem by understanding the mechanism
that produces trailing zeroes. We made use of the following problem-
solving strategies.
First we got a feel for the problem by the following steps.

J Simplify the problem to make it tractable (consider 5! instead of

100!).

J Make a table for easier bookkeeping.

The breakthrough came with more considerations:

J Analyse the goal (where do trailing zeroes come from?), first in
the simpler case 5!.

J From the simpler problem work towards the original problem step
by step (when does the next trailing zero appear – at 10!); here we
focus on what matters (factors 5 and 2).
28 1 First explorations

J Recognize a pattern or rule (number of trailing zeroes = number

of factors 5 and 2).

Of course we always need to be alert not to overlook anything (addi-

tional fives in 25, 50, 75 and 100). 

You might like to look at the general question now: How many
trailing zeroes does n! have for an arbitrary natural number n? See
exercise E 1.6.

1.3 A problem about lines in the plane

? Problem 1.3
Suppose you draw n straight lines in the plane, no two of which are parallel
and no three of which meet at a point4 . These lines subdivide the plane into
regions. How many regions are there?
In this problem, we write n for an arbitrary natural number: n =
1, 2, 3, . . . . You might think that it is harder to consider the general
problem (any n) than the special problem (one fixed value of n, for
example n = 20).
But we will see that the more general problem is easier! Why?
Recall that in Problem 1.2 we made progress by first considering the
same problem with 100 replaced by 1,2,3,4,. . . Then we were able to
solve the special problem by understanding the general mechanism
which produces trailing zeroes. Here things will be similar.
For this reason we will pose many problems in a general form from
now on.

 Investigation
G
 Let us have a close look at the problem: Do we understand the
assumptions? Which configurations of lines are permitted, which

4 If
both of these conditions are satisfied then we say that the lines are in general
position.
1.3 A problem about lines in the plane 29

are not? We make a few sketches, look at examples. The config-

urations in Figure 1.2 are permitted, those in Figure 1.3 are not:
in the left-hand picture two lines are parallel, in the right-hand
picture three lines meet at a point.

Figure 1.2 Permitted configurations

Figure 1.3 Configurations which are not permitted

 What are we looking for? The number of regions created by the

lines. Let us count the regions in each of the configurations in
Figure 1.2, where n = 1, 2, 3, 4, and make a table.

n 1 2 3 4
an 2 4 7 11

Here we introduced the notation an for the number of regions, so

a1 = 2, a2 = 4 and so on (introduce notation).
To get more familiar with the problem we also look at the config-
urations in 1.3, although they are not permitted: In the left-hand
configuration we have n = 3 with 6 regions, in the right-hand
configuration we have n = 4 with 10 regions. These numbers differ
from the values of a3 , a4 in the table: Apparently, the assumptions
in the problem actually matter!
 Attention: It may well be that, even for permitted configurations,
the number of regions depends on the position of the lines, not
just on the number of lines. To check this out we draw a few other
30 1 First explorations

configurations with n = 4, see Figure 1.4. We see that we get 11

regions in each case, as before. This does not prove anything, but it
is reassuring: Maybe the number of regions is independent of the
position of lines after all (for permitted configurations, of course)!
It is part of the problem to find out whether this is always the case,
not just in these examples.

Figure 1.4 Two more configurations with four lines

 Look at the table for an . Can you see a pattern?

Think about it!

Maybe you discovered one of the following patterns:

– Every number an is the sum of the numbers directly to its left
and above it.
– As we go from left to right, the difference of consecutive
numbers an increases by one.
– Every an is the sum of the numbers in the first row, up to n,
plus one.
All are correct. There are always several ways to proceed.5 Maybe
you have discovered yet another pattern.
To proceed let us suppose you had found the first pattern. How
can we write it as a general formula? The number to the left of an
is an−1 , the number above it is n. So we observed that

a n = a n −1 + n for n = 2, 3, 4. (1.1)

5 Actually
these descriptions are closely related. Convince yourself that the first
two say essentially the same thing, and that the third one can also be related to them.
1.3 A problem about lines in the plane 31

Does this rule continue to hold for higher n? Of course we could

check a few more examples. But can we be sure that it holds for all
n? We need to find a general argument. This is an interim goal in
our quest to solve Problem 1.3.
 Let us try to find a reason why equation (1.1) should hold for all n.
What does this equation mean in terms of our original problem?
(We analyse our goal.)
Recall that an is the number of regions for n lines, so an−1 is the
number of regions for n − 1 lines. So (1.1) says: Whenever we
add a line to already existing n − 1 lines, the number of regions
increases by n. Why should this be true? Think about it!
If you don’t see it right away, it is useful to consider an example,
always looking out for a general rule. Suppose we start with two
lines and then add a third one, which we call . How do new
regions arise? For simplicity let us colour the previously existing
lines black and the new line red. Compare the pictures. Take your
time.

Think about it!

 Insight: The new regions arise because the new line  cuts through
some of the previously existing regions.
 How many previously existing regions are cut by ? Well, three,
but can we relate this to other aspects of the figure (remember to
look for general rules; ’three’ is not a general rule)? Is there another
’three’ in the picture? Yes:  is cut into three segments by the black
lines. And clearly this is a general rule: The number of regions
cut by  is the same as the number of segments of .
What’s next? Remember what we are after: Adding a new line to
32 1 First explorations

n − 1 existing (black) lines adds n regions. So we need to relate

the number of new regions to the number of black lines. We saw: The
number of new regions equals the number of segments of . Can
you relate the segments of  to the black lines? Think about it.
Of course! Each black line intersects , and the number of segments
of  must be one more than the number of intersection points! In
the example: two black lines, so two intersection points, so three
segments.6
 To get comfortable, let us check this with three black lines, adding
a fourth one: Three intersection points, four segments, therefore
four new regions.

 Clearly, this is a general rule: n − 1 existing lines yield n − 1

intersection points with , therefore n segments of , therefore 
cuts n ’old’ regions, hence we get n new regions. This proves
equation (1.1)!
 Does it really? Let us look at it in more detail. Why does  intersect
every existing line? Because they are not parallel!
 Did we use the assumptions? We have just seen where the first
assumption was used. How about the second assumption, i.e. no
three lines meet in a point? Answer: If three lines were allowed to
meet in a point then  might pass through the intersection point
of two black lines, so there would be less than n − 1 intersection
points, hence less than n new regions.
 So we have reached our interim goal: We proved equation (1.1).
This type of equation, which relates an to an−1 , is called a recur-
rence relation. A recurrence relation is quite useful, it allows us

6 This is just like the cuts and pieces of the log in Problem 1.1.
1.3 A problem about lines in the plane 33

to calculate a2 , a3 , a4 . . . etc. quickly (starting with a1 = 2), without

drawing pictures.7
 But really we would like a closed formula, i.e. a formula which
allows us to calculate an directly from n, without first calculating
a1 , a2 , . . . . We obtain this in two steps: In the first step, we plug the
recurrence relation into itself repeatedly: First note that replacing n by
n − 1 in (1.1)8 yields an−1 = an−2 + (n − 1), so

an = an−1 + n = an−2 + (n − 1) + n.

Continuing in the same manner we get

a n = a n −2 + ( n − 1 ) + n = a n −3 + ( n − 2 ) + ( n − 1 ) + n = . . .
= a1 + 2 + 3 + · · · + n
= 2+2+3+4+···+n.

(How do we know what to put in the second line? The previous two
expressions exhibit a pattern: The first number added is one more
than the index of the a before it, for example an−2 + (n − 1) + . . . ,
an−3 + (n − 2) + . . . . So when we are down to a1 then we start
a1 + 2 + . . . .)
 This last expression is still not satisfying because of the dots: For
example, if n = 100 then we would have to add one hundred
numbers to get a100 . So in a second step we use the following
famous trick9 10 for calculating 1 + · · · + n =: s which then yields

7 And, incidentally, it shows that the number of regions depends only on the
number of lines, not on their (permitted) position.
8 We are allowed to do this since (1.1) is true for all numbers n, so in particular

for the number n − 1.

9 You should put that in your bag of tricks! It is usually attributed to Carl

Friedrich Gauss (1777-1855), but most likely it is much older.

By the way, the original spelling of this name is Gauß. The German letter ß is a
sharp s.
10 Some people prefer to proceed as follows: In 1 + · · · + n first add the first and

last summand to obtain n + 1, then add the second and second to last summand to
get 2 + (n − 1) = n + 1 again and so on. This works fine except that you have to
think about what happens when you get to the middle, and this is different for even
and odd n. Doing each of these cases separately you get the formula in the text also.
But the method in the text is better since you don’t need to distinguish two cases.
34 1 First explorations

a quick formula for an . Write s two times and add:

1 + 2 + ... + ( n − 1) + n = s
n + ( n − 1) + ... + 2 + 1 = s
( n + 1) + ( n + 1) + ... + ( n + 1) + ( n + 1) = 2s
How many times does n + 1 appear in the last line? The first line
shows that the answer is n. Hence n(n + 1) = 2s, so
n ( n + 1)
1+2+···+n = , (1.2)
2
and to get an we simply add one. This is the closed formula we
were looking for. 
G
Let us write up the solution properly.

! Solution of Problem 1.3

Obviously a1 = 2. We first prove that for all n ≥ 2 the recurrence
relation
a n = a n −1 + n (1.3)
holds. To prove this, suppose n − 1 lines are given in the plane in
general position. Add another line  which is not parallel to any of
the given lines and does not pass through any of their intersection
points. Then  intersects every one of the n − 1 given lines, and we
get n − 1 intersection points on . These points subdivide  into n
segments. Each segment splits one of the regions formed by the n − 1
lines into two, so adding  produces n additional regions. This proves
equation (1.3).
Plugging the recursion into itself repeatedly and using equation
(1.2) we get the solution
n ( n + 1)
an = 2 + 2 + 3 + 4 + · · · + n = 1 + . (1.4)
2
!
 Review of Problem 1.3
As in Problem 1.2 we first got a feel for the problem with the help
of examples (in this case sketches) and a table. The examples also
helped us to understand the assumptions in the problem.
1.4 Toolbox 35

Then we found a pattern. Introducing some notation helped us

to express the pattern as an equation (1.1). To investigate whether
the pattern continues to hold for all n and all configurations, we
interpreted the equation in terms of the original problem (adding
a line). We focussed on the goal (where do the new regions come
from?). We could finally prove the formula by advancing in small
steps, always keeping the goal in mind. Here again it was useful to
look at an example, but we were careful to argue in a way that would
work in general. A look at the assumptions helped us to fill in small
gaps in the argument.
Finally we used the Gauss trick (a technique) to derive the solution
(1.4) from the recurrence relation (1.1). 

1.4 Toolbox

In this chapter you encountered the most important general problem-

solving strategies. We now collect them into a toolbox, which will be
supplemented in later chapters. You can find the complete tool box
in Appendix A.

1. Understand the problem

Read the problem carefully. What is given, what are we looking for?
What are the assumptions?
2. Investigate the problem
J Get a feel for the problem, get your hands on it.
Useful strategies for this:
– First consider a simpler problem
– Look at special cases, examples
– Make diagrams and tables

J Use the examples to find a general rule

J Look for patterns

J Set yourself interim goals

J What is essential?
 36 1 First explorations

J Work step by step

J Check if you used the assumptions

J Introduce notation11
3. Write up your solution properly
Pay particular attention to:
J Complete arguments

J Sensible structure

J Understandable writing (mention ideas, motivations)12

J Correct use of mathematical language

4. Review
What have we learned? Is the solution reasonable? Could it be done
differently, better?

Writing up the solution properly and reviewing helps you to check

your solution. Sometimes you notice gaps in your solution during
these steps. The review also helps for other problems.

Exercises

When working on the exercises, think about which problem-solving

strategies might be useful.

1 E 1.1 .Here are some examples of shifts by one. Or aren’t they? You
may use fingers to count, but also try to find a good argument.
• You see a row of five trees, each one 20 meters from the next.
How long is the row? What if there are 20 trees?
• I booked a hotel room from May 20 (arrival) to May 23 (depar-
ture). That’s how many nights? How many from May 3 to May
29?

11 That is, short identifiers, for example an for the number of regions in Problem
1.3.
12 Many people think that mathematical texts consist only of equations and

symbols. This is far from true. Explanations are as important as equations, and in
some mathematical texts there are no equations at all.
Exercises 37

• You see a row of 5 trees. The row is 100 meters long. What’s
the distance between successive trees, supposing it’s always the
same?
• I start work at 8 am, I work until 5 pm. That’s how many hours?
• The clock on my office wall chimes every hour, on the hour.
How often do I hear it chime during a working day?
• How many three digit natural numbers are there?
• How many integers n satisfy the inequalities 15 ≤ n ≤ 87 ? How
many satisfy −10 ≤ n ≤ 10 ?

E 1.2 Y. ou want to arrange matches into n squares as follows. How 1

many matches do you need? The side length of the squares is one
match.

1 2 n

. et n ∈ N. Consider the sum of the first n odd natural

E 1.3 L 2
numbers, that is,

1 + 3 + 5 + · · · + (2n − 1) .

Calculate the sum for some values of n. Can you see a pattern? Make
a conjecture and prove it.

E 1.4 L . et n ∈ N. Is it possible to draw n points P1 , . . . , Pn and a 2

straight line g on a sheet of paper such that none of the points lies on g,
but g intersects each of the line segments P1 P2 , P2 P3 , . . . , Pn−1 Pn , Pn P1 ?
Look at some examples, make a conjecture and prove it.

E 1.5 .Modify Problem 1.1 as follows: suppose that before any cut 3
you are allowed to put an arbitrary number of pieces obtained in
earlier cuts next to each other, and to make one straight cut through
all of them. Now how many cuts do you need? How many cuts for
a log of length 4, 8, 16, 32, 27, n? Give a reason why your answer is
best possible, i.e. why it cannot be done with fewer cuts.

E 1.6 .Let n ∈ N. Find a formula for the number of trailing zeroes 2

of n!.
 38 1 First explorations

E 1.7 T. he strategy ‘special cases, looking for patterns’ can be useful 2–3
for finding formulas. Here are some examples.
a) Find a closed formula for the sum

1 · 1! + 2 · 2! + · · · + n · n!

b) Find a closed formula for the sum

1 2 3 n
+ + +···+
2! 3! 4! ( n + 1) !

c) (With more calculation:) Compute the numbers

1 1 1 1 1 1
1+ + , 1+ 2 + 2 , 1+ 2 + 2 ,...
12 22 2 3 3 4

as fractions. What do you observe? Formulate and prove a

general formula. Use this formula to find a closed formula for
the sum
  
1 1 1 1 1 1
1+ 2 + 2 + 1+ 2 + 2 +···+ 1+ 2 +
1 2 2 3 n ( n + 1)2

1-2 E 1.8 .For each of these sequences find a pattern and formulate it in
words, then as a formula, where the numbers are a1 , a2 , . . . . Invent
more such problems.
Example: 1, 4, 9, 16, 25: square numbers, an = n2
Example: 7, 9, 12, 16, 21: the difference increases by one at each
step, starting at 2 from a1 to a2 ; formula: an = an−1 + n with
n ( n +1)
a1 = 7. Another (closed) formula is an = 2 + 6.

a) 3, 4, 5, 6, 7
b) 3, 9, 36, 180, 1080
c) 1, −1, 1, −1, 1
d) 1, 1, 1, 3, 5, 9, 17, 31
e) 2, 8, 24, 64, 160
 Exercises 39

3 E 1.9 F. or n ∈ N denote by sn the number of ways to write n as

ordered sum of natural numbers. Count n itself (one summand) as
one of these ways. For example

2 = 1 + 1 and 3 = 1 + 2 = 2 + 1 = 1 + 1 + 1,

so s2 = 2 and s3 = 4. Find s4 and s5 , conjecture a general rule and

prove it.

E 1.10 .Make a conjecture on the truth of the statement ’n2 + n + 41 is 2

a prime number for all n ∈ N.’ Then check your conjecture by giving
a proof or a counterexample for the statement.

E 1.11 .Investigate: Suppose you draw n straight lines in the plane, no 4

three of which meet in a point. How many regions do you get? Note
that lines are allowed to be parallel. The answer will not only depend
on n. What else will it depend on? Identify quantities associated
with a configuration of lines that allow you to calculate the number
of regions, and find a formula for this number in terms of these
quantities. In the special case of lines in general position you should
get the formula found in this chapter.
2 Recursion – a fundamental idea

Have you ever seen a Russian Matryoshka doll? When you open up
this wooden figure then you will find a smaller figure inside which
looks just like the first one. You can open this figure again and find a
yet smaller one, and so on.
Some mathematical problems can be tackled in a similar way: solve
the problem by reducing it to a smaller problem of the same kind.
This technique is called recursion. It is frequently used for counting
problems, and the equation which expresses this reduction is called a
recurrence relation. You have encountered this already in Problem
1.3. We will now explore the method systematically and see some
more examples.
For any given problem there are typically two tasks: First, find a
recurrence relation. Second, solve the recurrence relation; that is, find
a closed, non-recursive formula. In this chapter you will learn about
ways to tackle both steps.

2.1 Recursion in counting problems

Often the idea of recursion is expressed in terms of a recurrence

relation:

Definition A recurrence relation for a sequence of numbers

a1 , a2 , a3 , . . . is an equation which expresses any an in terms of
an−1 , an−2 , . . . , a1 , and n.

When solving Problem 1.3 we encountered the recurrence relation

an = an−1 + n. This expresses an in terms of an−1 and n only, the
previous values an−2 , . . . are not used. We call this a recurrence relation

© Springer International Publishing AG, part of Springer Nature 2018 41

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7_2
42 2 Recursion – a fundamental idea

of length one. Here are some more examples of recurrence relations:

a n = a n −1 + a n −2
an = nan−1
a n = 1 + a 1 + · · · + a n −1

When writing such equations we always mean that they should hold
for all n where all occurring indices are at least one. So the first
equation means

a3 = a2 + a1 , a4 = a3 + a2 , a5 = a4 + a3 etc.

Recurrence relations are a useful technique for solving counting

problems. In a counting problem we are interested in certain config-
urations or objects and want to know how many there are.

Examples of counting problems

J How many regions are formed by n straight lines in the plane

that lie in general position (see Problem 1.3)?

J In how many ways can you choose 6 numbers out of 1, . . . , 49?

J Suppose that n people meet, and everyone shakes hands with

everyone else – but only once. That’s how many handshakes?

J Suppose you have a board of size 2 × 10, and 10 domino

pieces. How many ways are there to cover the board with the
dominoes (see Problem 2.2)?

Often the problem involves a number n which indicates the ‘size’ of

the problem. Let us denote the number we are looking for by an . Then
we are looking for a formula that allows us to compute an directly
from n. The recursion technique runs as follows.

Technique: Recursion

1. Look for a recurrence relation: Can you reduce the problem

of size n to a smaller problem of the same kind? If yes then you
2.1 Recursion in counting problems 43

will obtain a recurrence relation for the sequence of unknown

numbers an .

2. Solve the recurrence relation: Use the recurrence relation to

find a formula for an . To do this you also need an initial
condition, that is, the values of an for some small n (depending
on the recurrence relation this could be one, two or more
values).

This is precisely how we proceeded in Problem 1.3: Removing one

of the n given lines, we reduced the problem to a smaller problem of
the same kind (with n − 1 lines). Arguing that adding the line back
in will produce n extra regions, we derived the recurrence relation
an = an−1 + n. Then we solved the recurrence relation using the
Gauss trick.
Looking for a recurrence relation is an example of setting an in-
terim goal: Instead of aiming for a formula for an directly, we content
ourselves first with the more modest goal of finding a recurrence
relation.
The two steps of the recursion technique are very different in
character, as the examples in the next sections will show. There is no
general technique for solving recurrence relations. Simple recurrence
relations can be solved simply by repeatedly plugging them into
themselves, as we did in Problem 1.3. For many recurrence relations
this does not help, however. An example is the famous Fibonacci
recurrence relation, and in Section 2.4 you will learn how to solve this
and similar recurrence relations. There are also recurrence relations
which cannot be solved at all by a closed formula.
Recursion is only one way to approach counting problems. You
will learn about other counting principles in Chapter 5.

Remark

When you are given a problem for some fixed number of objects
(for example, 100 lines in the plane), it may be useful to generalize
it (n lines in the plane). Only then you will be able to use the
recursion technique!
This is a curious phenomenon, typical for mathematics, and for
science in general:
44 2 Recursion – a fundamental idea

Sometimes a problem simplifies when we look at a more

general problem.
This sounds like a paradox since you would expect a more general
problem to be harder. But this is precisely how we were able
to solve Problem 1.2: Instead of looking at 100! as required, we
looked for the general way in which zeroes at the end of n! are
‘produced’.
This principle is the opposite of the problem-solving technique
of looking at special cases. Both are useful!

2.2 The number of subsets

Imagine you have three photographs and want to give some selection
of them to a friend. You have not decided yet how many and which
photos you want to give her. How many possibilities are there? You
could give away one photo (3 possibilities: photo 1 or 2 or 3) or two
(3 possibilities: photos 1,2 or 1,3 or 2,3), or all three or none at all
(one possibility each). That’s 3 + 3 + 1 + 1 = 8 possibilities all in all.
How many possibilities would you have with 4 or 5 or more photos?
Mathematically speaking, we want to find the number of subsets of a
set with three or more elements (your photos).
We will use the language of sets. If you are not familiar with this,
read Appendix B first.

? Problem 2.1
How many subsets does the set {1, 2, . . . , n} have?
We always count the empty set ∅ and the full set {1, . . . , n} among the
subsets1 . For example, the set {1, 2} has four subsets: ∅, {1}, {2}, {1, 2}.

 Investigation
G
 We first need to get a feel for the problem. Therefore we look at
a few examples and make a table. For simplicity we write, for

1 This may seem artificial at first. But you will see that in this way the solution

will be simpler and more elegant. If you don’t want to count the empty set then
simply subtract one from the result.
2.2 The number of subsets 45

example, 12 for the subset {1, 2}.2 We denote the desired number
by an .
n subsets of {1, 2, . . . , n} an
1 ∅, 1 2
2 ∅, 1, 2, 12 4
3 ∅, 1, 2, 3, 12, 13, 23, 123 8

 Maybe you see a pattern in the last column: a1 = 2 = 21 , a2 = 4 =

22 , a3 = 8 = 23 , the powers of 2. So we conjecture:

Conjecture: We have an = 2n for all n.

 Why should this be true? How can we be sure that this is still
true for n = 100, say? As a first attempt at an understanding let us
order the subsets by their size and see how many subsets there are
for each fixed size. For n = 2 the numbers of subsets with 0,1,2
elements are 1,2,1, respectively; for n = 3 the numbers of subsets
with 0,1,2,3 elements are 1,3,3,1, respectively. Adding up, we get 4
and 8 subsets; but there is no clear pattern which would help us
explain why we always (for all n) get a power of 2. We need a new
idea.
 Second attempt: Can we solve the problem recursively? Can we
reduce the problem of size n to the same problem of size n − 1?
Can we find a recurrence relation?
 Let us check the example n = 3: Can we relate the subsets of
{1, 2, 3} to the subsets of {1, 2}? Look at the table above.

Think about it!

 Observation: The subset of {1, 2} are among the subsets of {1, 2, 3}:
They are precisely those subsets of {1, 2, 3} which don’t contain
the element 3.

2 Such short notation can be very useful. It should be explained when it is used
the first time.
46 2 Recursion – a fundamental idea

 So let us consider the remaining subsets of {1, 2, 3}: those that do

contain 3 as an element. There are four of them. Can we relate
these also to the subsets of {1, 2}? Let us look at them:

3, 13, 23, 123 (2.1)

and compare them to the subsets of {1, 2}:

∅, 1, 2, 12 (2.2)

Do you notice anything? Stare at these two lines until you notice
something (beyond the fact that each line has four items, of course).

Think about it!

 Of course! Each set in row (2.1) arises from a set in row (2.2) by
adding the digit 3 to it.
 Let us summarize: We have divided the subsets of {1, 2, 3} in two
classes: those that contain 3 and those that do not. For each class
we showed that it has as many subsets in it as there are subsets of
{1, 2}. Therefore a3 = a2 + a2 = 2a2 .
 The main virtue of this argument is that it generalizes to any n: We
divide the subsets of {1, . . . , n} in two classes: those that contain n,
and those that don’t. Each class contains an−1 subsets, by the same
reasoning as above. Therefore we get the recursion

an = 2an−1 .

If we plug this into itself repeatedly and use a1 = 2 then we get

an = 2n for all n. G
We now write this up cleanly, with all the details. This will also
help us find out whether the argument has any gaps.

! Solution of Problem 2.1

Denote by an the number of subsets of the set {1, . . . , n}.
1. Claim: The recurrence relation an = 2an−1 holds for n = 2, 3, . . . .
2.2 The number of subsets 47

Proof.
We divide the subsets of {1, . . . , n} in two classes: the first class
contains those subsets that contain n, the second class contains
those subsets that do not contain n.
The subsets that don’t contain n are precisely the subsets of
{1, . . . , n − 1}. So the second class contains precisely an−1 sub-
sets.
For any subset A ⊂ {1, . . . , n} which contains n we consider
the subset B = A \ {n} of {1, . . . , n − 1}. Now any subset B ⊂
{1, . . . , n − 1} arises in this way from some A: simply set A =
B ∪ {n}. In fact, this is clearly the only choice of A which results
in B. Therefore the first class contains as many subsets A as there (*)
are subsets B of {1, . . . , n − 1}, i.e. an−1 subsets.
Summing up, we get an = an−1 + an−1 = 2an−1 for all n = 1, 2, . . . .
q. e. d.

2. We now solve the recurrence relation by repeatedly plugging it

into itself: If we replace n by n − 1 in the recursion formula we
obtain an−1 = 2an−2 . This is allowed for n ≥ 3 since the recursion
holds for all numbers 2, 3, . . . , so in particular for the number n − 1.
We plug in: an = 2an−1 = 2 · 2an−2 . Repeating in a similar vein we
get, using the initial condition a1 = 2,

an = 2an−1 = 2 · 2an−2 = · · · = 2n−1 a1 = 2n−1 · 2 = 2n . (2.3)

The factor 2n−1 arises since each time the index of a goes down by
1 we get another factor of 2. The index of a is reduced from n to 1
over all, hence by n − 1, hence there are n − 1 factors of 2, so we
get 2n−1 . !
We now write up the same solution again, but in very formal
mathematical language. This language is most useful when writing
up complex arguments, so for our problem it is really a bit of overkill.
So you may skip this part unless you want to practice formal language.
In the second part of the proof we use mathematical induction. This
will be introduced in Chapter 3.

! Formal solution of Problem 2.1

Let Tn = { A : A ⊂ {1, . . . , n}} be the set of subsets of {1, . . . , n} and
an = | Tn | the number of elements of Tn .
48 2 Recursion – a fundamental idea

1. Claim: We have the recurrence relation an = 2an−1 for n = 2, 3, . . . .

Proof.
Let

Un = { A ∈ Tn : n ∈ A}
Vn = { A ∈ Tn : n ∈ A} .

Then
Tn = Un ∪ Vn (disjoint union) . (2.4)

We now prove that Un and Vn have an−1 elements each. Obviously

Vn = Tn−1 , so
|Vn | = | Tn−1 | = an−1 .
In order to show that |Un | = | Tn−1 | we find a bijection u : Un →
Tn−1 . Define the map

u : Un → Tn−1 , A → A \ {n} .

In order to show that u is a bijection we write down another map

which we then prove to be the inverse of u. Define the map

v : Tn−1 → Un , B → B ∪ { n }.

We claim that v is the inverse of u, that is, that

u(v( B)) = B for all B ∈ Tn−1 ,v(u( A)) = A for all A ∈ Un .

(2.5)
The first equation holds because of

u(v( B)) = u( B ∪ {n}) = ( B ∪ {n}) \ {n} = B .

Here the first two equality signs are the definitions of v and u, and
the third one follows from n ∈ B. The second equation in (2.5)
follows from

v(u( A)) = v( A \ {n}) = ( A \ {n}) ∪ {n} = A

where in the last step we used n ∈ A.

Therefore we have proven (2.5), hence u is bijective. This implies

|Un | = | Tn−1 | = an−1 .

2.2 The number of subsets 49

Then (2.4) gives

an = | Tn | = |Un | + |Vn | = an−1 + an−1 = 2an−1 ,

which was to be shown. q. e. d.

2. Claim: We have an = 2n for all n.

Proof.
We use mathematical induction.
Base case: For n = 1 we have a1 = 2 = 21 .
Inductive hypothesis: Let n ∈ {2, 3, . . . } be arbitrary, and as-
sume an−1 = 2n−1 .
Inductive step: Using the recurrence relation we get

an = 2an−1 = 2 · 2n−1 = 2n ,
which was to be shown. q. e. d.

!
Remarks

J Both proofs are mathematically valid. It is a matter of taste

which one you prefer. Many mathematics books are written in
the formal language, so consider this a useful exercise.
J Note that one needs to argue carefully for the subsets A that
contain n: it does not suffice to say that A \ {n} is a subset of
{1, . . . , n − 1}. In addition, you need to show that every subset
of {1, . . . , n − 1} arises precisely once from this operation. In
the formal proof this is the statement that u is bijective.
J In order to prove bijectivity of u in the formal proof, one
could show that u is both injective and surjective, rather than
exhibiting an inverse map. These words are explained in
Appendix B. In the informal proof this proof was carried out
at the place marked by (*).
Injective: If A, A
∈ Un satisfy A = A
then u( A) = u( A
).
Surjective: For every B ∈ Tn−1 there is A ∈ Un satisfying
u( A) = B.
J If we interpret {1, . . . , n} for n = 0 as ∅ then the formula for
an is also true for n = 0, for the empty set has only itself as
subset, hence a0 = 1. So one could start the recursion at n = 1.
50 2 Recursion – a fundamental idea

 Review of Problem 2.1

We have used the general techniques from Chapter 1 (examples, table
etc.) again. In addition, we have searched for a recurrence relation.
We also encountered an important technique for counting problems:
We divided the objects that we wanted to count into classes. Each
of these classes for the counting problem of size n could be related
to the same counting problem of size n − 1, and this led us to the
recursion.
In order to find these classes we focussed on what distinguishes
the size n problem from the size n − 1 problem: the element n of
{1, . . . , n}. Observing how this element matters in the problem – it
can be contained in a subset or not – we found the solution. 
In Chapter 5 you will get to know another way to tackle the problem
of counting subsets.

2.3 Tilings with dominoes

? Problem 2.2
In how many ways can you tile a rectangle of size 2 × n with dominoes of
size 1 × 2?

Here ‘tiling’ the rectangle means covering it with dominoes so that no

two dominoes overlap and no domino extends outside the rectangle.

 Investigation
G
 To get a feel for the problem, sketch a few examples and try to
discover a rule. Squared paper is useful.

Think about it!

2.3 Tilings with dominoes 51

Figure 2.1 shows all possible tilings in the cases n = 1, 2, 3, 4.

We make a table for the number of tilings an for each n. Check
foryourself that a5 = 8.

n 1 2 3 4 5
an 1 2 3 5 8
Do you see a pattern? – How about this one: Each number in
the lower row is the sum of its two predecessors. That is, an =
a n −1 + a n −2 .
Does it go on like this? If yes, why?
 Based on the table we conjecture that the recursion an = an−1 +
an−2 holds for all n. How could we prove this? How can we relate
the tilings of the 2 × n rectangle (corresponding to an ) to tilings of
the 2 × (n − 1) and 2 × (n − 2) rectangles (corresponding to an−1
and an−2 )?

Think about it!

 What distinguishes a 2 × n rectangle from a smaller rectangle?

Something is added at the end. So let us look at the right-hand end
of the 2 × n rectangle, say. How can it be tiled? – Either there is
one vertical domino, or two horizontal ones, see Figure 2.2. Every

n=1

n=2

n=3

n=4

Figure 2.1 Tilings of 2 × n rectangles with dominoes

tiling is one of these two types. So we have divided the counting

problem into two classes again. Let us count the tilings of each
type separately.
52 2 Recursion – a fundamental idea

Either 2 ···

or 2 ···

Figure 2.2 Two kinds of tilings

 How many tilings of the first type are there? To the left of the
vertical domino there is a 2 × (n − 1) rectangle, and any tiling of
this rectangle yields a tiling of the 2 × n rectangle of the first type.
So there are an−1 tilings of the first type.
 Similarly, for a tiling of the second type we are left with a 2 × (n −
2)-rectangle, and any tiling of this rectangle yields a tiling of the
2 × n rectangle of the second type. So there are an−2 tilings of the
second type.
 Therefore, we get an = an−1 + an−2 .
 If you are not convinced, check the argument step by step for n = 4.

G
We have obtained the following solution.

! Solution of Problem 2.2 (derivation of a recurrence relation)

Let an be the number of domino tilings of a 2 × n rectangle.
Claim: The recurrence relation an = an−1 + an−2 holds for all n ≥ 3.
Proof.
Every tiling of a 2 × n rectangle is of one of the following two types
(see Figure 2.2):
1. There is a vertical domino at the right-hand end.
2. There are two horizontal dominoes at the right-hand end.
For any type-1 tiling, a 2 × (n − 1) rectangle remains to the left
of the vertical domino. Let us call the 2 × n rectangle R and the
2.3 Tilings with dominoes 53

2 × (n − 1) rectangle R
. The type-1 tilings of R correspond precisely
to all the tilings of R
: Any tiling of R
yields a type-1 tiling of R, by
simply adding the vertical domino at the right-hand end. Conversely,
removing the right-most vertical domino from a type-1 tiling of R
yields a tiling of R
.
Therefore, R has an−1 type-1 tilings.
Similarly, for any type-2 tiling of R a 2 × (n − 2) rectangle R

remains, and type-2 tilings of R correspond precisely to tilings of R

.
Therefore, R has an−2 type-2 tilings.
Altogether there are an−1 + an−2 tilings of the 2 × n rectangle, so
a n = a n −1 + a n −2 . q. e. d.

We have proved the recurrence relation. We will solve it in the next

section. !

 Review of Problem 2.2

Using a table we arrived at a conjecture for a recurrence relation. As
in the previous problem we were able to prove the recurrence relation
by dividing the objects to be counted (the tilings) into two classes
(corresponding to type-1 and type-2 tilings), and relating each class
to a smaller problem of the same kind.
With a little more experience we might have skipped the ‘conjecture’
part, and tried to find a recursion directly. The question “How can I
reduce the problem of size n to smaller problems of the same kind?”
and the focus on one end of the strip might have led us to the right
idea. This kind of thinking will be needed for problems where it is
hard to discover a pattern. We will discuss such a problem later, see
Problem 2.4.
What did we accomplish? The recurrence relation allows us to
extend the table quickly and compute, for example, a10 , see Table 2.1.
To find a10 without the recurrence relation would be very tiresome!
The an are called Fibonacci numbers.3 

3 Sometimesthe sequence of numbers with index shifted by one are called

Fibonacci numbers: f 1 = 1, f 2 = 1, f 3 = 2, f 4 = 3, in general f n = an−1 .
54 2 Recursion – a fundamental idea

n 1 2 3 4 5 6 7 8 9 10 · · ·
an 1 2 3 5 8 13 21 34 55 89 · · ·

Table 2.1 The Fibonacci numbers

2.4 Solving the F IBONACCI recurrence relation

We saw that the recurrence relation

a n = a n −1 + a n −2 (R)

allows us to compute an easily. However, we still need to do about n

additions to obtain an , and for large n this is tiresome. We now want
to solve the recurrence relation, that is, we want to find a formula
that allows us to compute an directly from n.
The initial conditions are a1 = 1, a2 = 2. In this case we need two
initial condition since each an is determined by the two previous
terms in the sequence. The calculations below are simplified if we
define a0 = 1, then (R) also holds for n = 2.

? Problem 2.3
Find a closed formula for the sequence a0 , a1 , . . . which is defined by the
recurrence relation (R) (valid for n ≥ 2) and the initial conditions

a0 = 1, a1 = 1. (IC)

A closed formula is an expression that allows us to compute an

directly from n. For example, the solution of the recurrence relation
an = 2an−1 , a0 = 1 is the closed formula an = 2n .

 ! Investigation and solution

G
 Can you guess a formula from Table 2.1? Does any of the usual
suspects work, for example n, n2 , 2n , n!, or a combination (sum,
product, . . . ) of those? Try it!
2.4 Solving the Fibonacci recurrence relation 55

Think about it!

 You didn’t find a formula? Don’t worry, to guess the formula is

quite impossible (as you will agree when you see it). However,
there is an ingenious method to find a formula. Rather than just
presenting how it works, I will develop it together with you from
simple principles.
 Let us use our problem-solving strategies: The problem is too
difficult. Simplify!
How? We need to satisfy two conditions: (R) and (IC). Let us first
forget about (IC) and focus on (R) only.
 So the question is: Can we find any expression in n which satisfies
the recursion (R) for all n? Let us play a little with examples: Try
an = n: Indeed, a3 = a2 + a1 is true since 3 = 2 + 1, but 4 = 3 + 2
shows that (R) is not true for n = 4. So an = n does not work.
Maybe this was too special. Let us try more general expressions, for
example an = rn + s (for fixed r, s ∈ R)4 , n2 , nk (for fixed k ∈ N),
. . . . You will find that none of this works. Let us try an exponential
again, but instead of the special 2n we leave the base undetermined.
That is, we make the5

Ansatz: an = αn for some real number α = 0

We plug this into the recurrence relation an = an−1 + an−2 and

obtain
α n = α n −1 + α n −2
or equivalently (divide by αn−2 )

α2 = α + 1 . (2.6)
4 Let’s do it: Plug an = rn + s into the recursion; this yields rn + s = r (n −
1) + s + r (n − 2) + s. A little algebra yields rn = 3r − s. Can this be true for all n?
Obviously not, unless r = 0 and s = 0. But then an = 0 for all n – indeed a solution
but an uninteresting one.
5 The German word Ansatz means attempt or approach. In this mathematical

context it is also used in English.

56 2 Recursion – a fundamental idea

This shows that an = αn satisfies the recurrence relation (R) for

all n if and only if α is a solution of equation (2.6). This is a
big step forward since now we only need to find the solutions
of one equation, not of infinitely many (an = an−1 + an−2 for
n = 2, 3, 4, . . . ).
Equation (2.6) is a quadratic equation for α which may be solved
 We rearrange to α − α − 1 = 0 and
by the well-known formula. 2

obtain the solutions 1

2 ± 1
4 + 1 or
√
1± 5
α= .
2
We introduce the notation
√ √
1+ 5 1− 5
β= , γ=
2 2
and call the corresponding sequences bn , cn . This will allow us to
continue to use the letter an as in the beginning, for the sequence
satisfying both (R) and (IC).
 What have we got? We have found two solutions for the recurrence
relation (R):
bn = β n , c n = γ n . (2.7)
This is a success: we have attained our interim goal, finding a
solution for the simplified problem. We even found two solutions,
and this will be very useful.
Now we need to look at the initial conditions. Let us check whether
one of our solutions satisfies (IC). Using β0 = γ0 = 1 we get

n 0 1 ...
known: bn 1 β ...
(2.8)
known: cn 1 γ ...
wanted: an 1 1 ···

We see that bn , cn satisfy the initial condition for n = 0 but not for
n = 1.
 We need another idea. Did we find all solutions of (R)? Or can we
produce more solutions from the two that we found?
2.4 Solving the Fibonacci recurrence relation 57

 New from old! There are two simple ways to produce new solu-
tions from old ones:
1. Multiply by a constant: If b0 , b1 , . . . solves (R) and B is any
real number then Bb0 , Bb1 , . . . also solves (R), because
bn = bn−1 + bn−2 ⇒ Bbn = Bbn−1 + Bbn−2
for all n ≥ 2.
2. Add: If b0 , b1 , . . . and c0 , c1 , . . . both solve (R) then so does
d0 , d1 , . . . defined by dn = bn + cn , because we can simply add
the equations:
bn = bn − 1 + bn − 2
cn = c n −1 + c n −2
+
bn + c n = ( bn − 1 + c n − 1 ) + ( bn − 2 + c n − 2 )
so dn = dn−1 + dn−2 for all n ≥ 2.
We can also combine both ideas: First apply 1. to the bn and the
cn , with two constants B and C, then add the results. This is called
the6
Superposition principle: If b0 , b1 , . . . and c0 , c1 , . . . solve (R)
and B, C are any real numbers then a0 , a1 , . . . defined by
an = Bbn + Ccn for all n
also solves (R).
The point is that since B and C are arbitrary, we get quite many
solutions of (R), so there is a chance that among them there is one
which also satisfies (IC).
 So now we need to find the numbers B, C such that these an also
satisfy (IC)7 . We write out the equation an = Bbn + Ccn for n = 0
and n = 1 and plug in the values from the table (2.8):
n=0: 1 = B + C
n=1: 1 = Bβ + Cγ
6 The word originates from physics where it describes, for example, the superpo-
sition of two waves to form a total wave. The superposition principle is always valid
for linear equations.
7 Before jumping into calculations let us think why this might work: We have to

satisfy two initial conditions (equations), and we have to determine two numbers B, C
(unknowns). So the number of equations is the same as the number of unknowns.
Rule of thumb: usually this will work.
58 2 Recursion – a fundamental idea

This is a linear system of equations for B, C and can be solved easily:

In order to eliminate one of the unknowns, say B, we multiply the
first equation by β and subtract the second equation.
β = Bβ + Cβ
1 = Bβ + Cγ
-
β−1 = 0 + C ( β − γ)
We obtain
β−1 γ−1
C= and then B = − ,
β−γ β−γ
where the second equation follows from the first using 1 = B + C
after a short calculation.
 We now use the values of B, C and bn , cn in the formula an =
Bbn + Ccn and obtain: The sequence of numbers
γ−1 n β−1 n
an = − β + γ ,
β−γ β−γ
satisfies both the initial condition (IC) and the recursion (R). Since
(IC) and (R) uniquely determine the sequence, this must be our
solution. √ √
We can write this a little more neatly: From β √= 1+2 5 , γ = 1−2 5
we get β − 1 = −γ, γ − 1 = − β and β − γ = 5, so
⎡ √  n +1  √  n +1 ⎤
1 1+ 5 1− 5
an = √ ⎣ − ⎦ (2.9)
5 2 2

This is the desired closed formula!

!
G
Let us summarize our method:
Method: Solving the Fibonacci recurrence relation

In order to solve the recurrence relation an = an−1 + an−2 with

initial condition a0 = 1, a1 = 1, proceed as follows:
1. Make the ansatz an = αn . This leads to the equation α2 = α + 1.
Denote its solutions by β, γ. Then βn , γn are solutions of the
recurrence relation.
2. Write an = Bβn + Cγn , plug in the initial conditions and solve
2.4 Solving the Fibonacci recurrence relation 59

the resulting linear system of equations for B, C. This yields

the desired formula for an .

Further remarks: Solving the Fibonacci recurrence relation

J Is the solution (2.9) reasonable? At first glance it does not seem

so: Clearly all an must be integers since the initial values
are and the recurrence relation then always produces inte- √
gers. However, the √
formula contains the irrational number 5.
1+ 5
For example, 2 = 1, 618 . . . . See Exercise E 5.16 for the
resolution of this apparent paradox.
J Let us state again why the method works: The right hand
side of formula (2.9) satisfies by construction both the initial
condition (IC) and the recurrence relation (R). Since the left
hand side also satisfies (IC) and (R), and since (IC) and (R)
determine the sequence uniquely, the formula must be correct
for all n.
J This proves the correctness of formula (2.9). You could provide
an additional proof using mathematical induction. You may
want to do this for exercise. But logically it is not needed.
J (Generalization) The same method works also for other initial
conditions, and even for other recurrence relations. More
precisely, it works for any recurrence relation

an = pan−1 + qan−2 (2.10)

where p, q are real numbers, with any two given initial val-
ues for a0 , a1 . These are called linear recurrence relations with
constant coefficients.
However, for some values of p, q the method needs to be
modified slightly, see the exercises.
The method also works for linear recurrence relations with
constant coefficients and length greater than two, for example,
an = an−1 − an−2 + an−3 . But now instead of the quadratic
equation for α you need to solve an equation of higher degree
(in the example α3 − α2 + α − 1 = 0), which may be difficult.
Of course one needs to give three initial values.
60 2 Recursion – a fundamental idea

J An important part of the method is the superposition prin-

ciple. This is a special case of the very general idea of linear
combination, a fundamental notion in linear algebra.
If you know some linear algebra then you can understand
the method as follows: In the vector space of all sequences
a0 , a1 , . . . the subset of those sequences that satisfy the
recurrence relation is a linear subspace. This subspace
is two-dimensional since any one of these sequences is
determined by two initial conditions. The special se-
quences β0 , β1 , . . . and γ0 , γ1 , . . . lie in this subspace and
are linearly independent, as follows easily from β = γ.
Therefore they form a basis of the subspace, therefore
the sequence a0 , a1 , . . . that we are looking for must be
representable as some linear combination of them.

J There is another elegant method for solving linear recurrence

relations with constant coefficients, using so-called generating
functions, see (Aigner, 2007).

By the way ...

To find the number of domino tilings for arbitrary rectangles is much

harder than in the case of 2 × n rectangles (see also Exercise E 2.12).
In 1961, both Kasteleyn and (independently) Temperley and Fisher
found a remarkable general formula: The number of domino tilings
of an m × n rectangle is
 1/4
m n
πj πk
∏∏ 2
4 cos
m+1
+ 4 cos2
n+1
.
j =1 k =1

The two big Pi’s are short notation for ‘product’. They tell us to plug
in all possible values of j and k in the parentheses and multiply all
the results. For example, if n = 2, m = 3 then this is a product of 6
factors.
This formula is very mysterious: Why do cosines appear? Why is
the product an integer? The reason why Kasteleyn and Temperley-
Fisher were interested in this problem is that it appears in theoretical
physics.8
8 You
can find more on this, and a guide to the literature, on the Wikipedia page
on domino tilings.
2.5 Triangulations 61

2.5 Triangulations

We now consider a counting problem where you cannot easily find

a pattern or recurrence relation by simply staring at a few numbers.
But we will still be able to solve it by looking systematically for a
recurrence relation.

? Problem 2.4
Let n ≥ 3 and P ¸be a convex n-gon.9 A triangulation of P is a subdivision
of P into triangles, using non-intersecting lines connecting vertices of P,
see Figure 2.3. How many triangulations does P have? Find a recurrence
relation.

Not a
Triangulation Triangulation Triangulation
triangulation

Figure 2.3 What is a triangulation? Examples for n = 6

We will consider two triangulations as different even if we can

obtain one from the other by a rotation or reflection. For example, the
three triangulations in Figure 2.3 are all different.

 Investigation of Problem 2.4

G
 We are looking for the number of triangulations of a convex n-gon.
Let us call this number Tn .
 By trying all possibilities systematically (do it!) we find:

n 3 4 5 6
Tn 1 2 5 14

9 That is, a triangle, quadrilateral, pentagon, hexagon if n = 3, 4, 5, 6. For n-gon

we also say polygon. A polygon is convex if all lines connecting two of its vertices
lie in the polygon.
62 2 Recursion – a fundamental idea

There is no obvious pattern in the sequence 1, 2, 5, 14. So we try

to look for a recurrence relation directly from the problem. How
can we obtain triangulations of an n-gon from triangulations of
polygons with fewer vertices? Can we divide all triangulations
of the n-gon into classes which correspond to triangulations of
smaller polygons?

Think about it!

 There are many ways to approach this. Maybe you had one of the
following ideas.
First attempt: We observe that the pattern shown in Figure 2.4
appears in all triangulations – ‘cutting off’ a vertex.

Figure 2.4 6-gon with cut-off vertex

After cutting off the vertex we are left with a polygon that has n − 1
vertices, which we can triangulate in Tn−1 ways. Since we can cut
off any of the n vertices and then, in each case, have Tn−1 ways to
triangulate the rest, we conjecture the recurrence relation

?
Tn = n · Tn −1 (2.11)

So here we tried to divide all the triangulations of the n-gon into n

classes, corresponding to the cut-off vertex.
Looking at the table we see that the recursion is wrong. But why?
Where was the mistake in our argument?

Think about it!

2.5 Triangulations 63

In our argument we counted some triangulations multiple times.

For example, the triangulation in Figure 2.5 is counted twice. It
lies in two of the classes. How can we remedy this? We could try

−→ ←−

Figure 2.5 A triangulation which is counted twice

to subtract a suitable number from n Tn−1 corresponding to such

multiple counts. This is possible but a little tricky, so instead we
will follow another path.10 Anyway, we see that conjecture (2.11) is
wrong.
By the way: When proceeding like this we would still have to
prove what we observed in examples: That every triangulation
has at least one cut-off vertex. This is not hard, for example using
induction or using the extremal principle, see Exercise E 10.9.
 Second attempt: Let us look at Figure 2.3 again and draw some
more triangulations. Let us focus on one vertex, say the right-hand
vertex of the hexagon. We see that there is always a line (diagonal)
from the right vertex to some other vertex, or else the right vertex
is cut off.
Such a diagonal allows us to reduce the problem to smaller prob-
lems of the same kind since it cuts the hexagon into two smaller
polygons, see Figure 2.6. In this example, there are two quadran-
gles, having T4 triangulations each. Since we can combine any two
of them and obtain a triangulation of the hexagon, we get T4 · T4
triangulations. In addition, we get triangulations arising from other
ways to draw the diagonal, and we can count those in a similar
way.
But again we have the problem of multiple counts, see Figure 2.7.
 Third attempt: We have seen that we need to be careful: When

10 If you want to know more about how to deal with such multiple counts

systematically, look up ‘Inclusion-exclusion principle’ in a book on combinatorics or

on Wikipedia. Then try to use this principle here.
64 2 Recursion – a fundamental idea

Figure 2.6 Subdividing the polygon by a diagonal

−→ ←−

Figure 2.7 A triangulation which is counted twice in the second attempt

dividing the problem into classes then the classes must be disjoint.11
We need a disjoint subdivision of the problem.

Think about it!

Here is a new idea: We focus on one side of the n-gon, say the
bottom side in Figure 2.8. Any triangulation must have precisely
one triangle which contains this side. So we can classify the trian-
gulations according to which triangle this is. For n = 6 there are
4 possibilities for this triangle as in the figure, therefore there are
4 classes. Note that the triangle is determined by the location of
its third vertex, and this can be any vertex except the two bottom
ones; this explains why there are 6 − 2 = 4 possibilities.

Figure 2.8 Classes into which we subdivide our problem

This is a disjoint subdivision since in any triangulation there can

11 Two sets (or classes, which is the same) are called disjoint if they have no

common elements. Here this means that each triangulation must lie in precisely one
class.
2.5 Triangulations 65

only be one triangle containing the bottom side. Compare this

with the second attempt: There can be several diagonals leaving
the right-hand vertex, and this led to the double count there, i.e.
non-disjoint classes.

To the left and the right of the dashed triangle we get convex
polygons with fewer than 6 vertices. By counting their triangula-
tions and combining them we can count the triangulations of the
hexagon. 
G
We now carry out this idea for general n.

! Solution of Problem 2.4

Consider an arbitrary convex n-gon and number its vertices P1 , P2 , . . . , Pn
as in Figure 2.9. We divide the triangulations of the n-gon into classes,
where each class is determined by the third vertex of that triangle in
the triangulation which contains the side P1 Pn .
Call this triangle D and its third vertex Pk . Here k can take any of
the values 2, . . . , n − 1, see Figure 2.9. So we get one class for each
such value of k.

Pk

P2 Pn−1

P1 Pn

Figure 2.9 One of the classes into which we subdivide the problem

Let us count how many triangulations are in the class labelled

by k. To the left of D we have a k-gon and to the right of D we
have an (n − k + 1)-gon (check this!). We can triangulate each of
these polygons and obtain a triangulation of the n-gon. Since we can
combine every left triangulation with every right triangulation, we
66 2 Recursion – a fundamental idea

multiply their numbers. This gives us the number of triangulations

in the class k.
The cases k = 2 and k = n − 1 are special since there is no actual
polygon on the left or right, respectively, only a ‘2-gon’ which cannot
be triangulated.

left of D right of D triang.

k vertices triang. vertices triang. combined
2 no polygon – n−1 Tn −1 Tn −1
3 3 T3 n−2 Tn −2 T3 · Tn−2
4 4 T4 n−3 Tn −3 T4 · Tn−3
.. .. .. .. .. ..
. . . . . .
k k Tk n−k+1 Tn − k +1 Tk · Tn − k +1
.. .. .. .. .. ..
. . . . . .
n−2 n−2 Tn −2 3 T3 Tn−2 · T3
n−1 n−1 Tn −1 no polygon – Tn −1

Thus we have found the number of triangulations in each class. Since

the classes are disjoint, the total number of triangulations of the n-gon
is their sum. We get

Tn = Tn−1 + T3 · Tn−2 + T4 · Tn−3 + · · · + Tn−2 · T3 + Tn−1

n −2
= Tn −1 + ∑ Tk · Tn − k +1 + Tn −1 .
k =3

See the list of symbols for an explanation of the summation sign ∑.

The formula simplifies if we set T2 := 1:
n −1
Tn = ∑ Tk · Tn − k +1 (2.12)
k =2

This is the recurrence relation we were looking for. It allows us to

extend the table for the Tn easily. !

 Review of Problem 2.4

We were not able to detect a pattern from our table. Therefore we
tried systematically to find a recursion, i.e. a method of reducing the
problem to a smaller problem of the same kind.
2.6 Toolbox 67

The first two ideas for such a reduction (cutting off a vertex, con-
sidering diagonals from a fixed vertex) did not work since they led to
multiple counts.
We finally found a recurrence relation by shifting focus from ver-
tices to sides, which allowed us to subdivide the problem into dis-
joint classes and express the number of triangulations in each class
using triangulation numbers of smaller polygons. 

Remarks

J The numbers Cn = Tn+2 are called Catalan numbers. They also

occur in many other counting problems. For example, Cn+1 is
the number of bracketings of n symbols, see Exercise E 2.11.
See Wikipedia for more examples. The Catalan numbers sat-
isfy the (slightly prettier) recursion
n
Cn+1 = ∑ Ck Cn−k
k =0

which you can easily derive from (2.12) (exercise!).

J To solve the recurrence relation for the Tn (or the Cn ) is harder

than for the Fibonacci numbers. The main difference is that
this recurrence relation is non-linear while the Fibonacci
recurrence relation is linear12 . A very elegant method which
works is generating functions, see (Aigner, 2007), for example.

2.6 Toolbox

In this chapter you learned a few special strategies which are useful
for problems of the kind “For all n find the number an of objects or
configurations of a certain kind”:

J Subdivide into classes: Subdivide the set of configurations that

you want to count into classes which are easier to count. The
classes must be disjoint.

12 This is a general phenomenon in mathematics: non-linear problems are often

harder than linear problems.
 68 2 Recursion – a fundamental idea

J Recursion: If counting the classes involves the same type of count-

ing problem but for a smaller problem size, you get a recurrence
relation.
A recurrence relationallows you to calculate an quickly even for
values of n where counting the objects/configurations directly
would be difficult.

These strategies are special cases of the general strategies simplify

and interim goals.
You also learned a technique for solving linear recurrence relations
with constant coefficients.

Exercises

Exercises E 2.1 to E 2.4 serve to deepen your understanding of the

method of solving the Fibonacci recurrence relation.

2 . et an be the Fibonacci numbers and define β, γ as in (2.7).

E 2.1 L
β n +1
a) Show that an equals the integer nearest to √ .
5
a n +1
b) (Using a little bit of calculus) Show that an converges for n → ∞
√
1+ 5
to the golden ratio β = 2 .

1-2 E 2.2 S. how that the method for solving the Fibonacci recurrence
relation works for all recurrence relations (2.10) for which the equation
α2 − pα − q = 0 has two different solutions. Solve the recurrence
relation an = 2an−1 + 3an−2 with the initial condition a0 = 1, a1 = 3
and with the initial condition a0 = 1, a1 = 2.

2-3 E 2.3 T . he solutions of α2 − pα − q = 0 can also be complex numbers.

This is a place where complex numbers are really useful.√ If you know
about them, consider the recurrence relation an = 2an−1 − an−2 ,
a0 = 0, a1 = 1, calculate the first ten terms, observe a pattern and
then explain the pattern using the fact that the numbers α = √1 (1 ± i )
2
satisfy α8 = 1.13 Even more interesting is the recurrence relation
an = 12 an−1 − an−2 , a0 = 0, a1 = 1. Calculate a few terms. The

13 You
can check this by direct calculation. But it is nicer to prove it geometrically
using the geometric meaning of complex numbers and their multiplication.
Exercises 69

sequence does not grow unboundedly, but it is also not periodic.

Can you explain this? How is it related to the fact that the number
π arccos( 4 ) is irrational (and how can you prove that it is?)?
1 1

E 2.4 W. hat do you do if α2 − pα − q = 0 has only one solution α? 2-3

Investigate!
For example, if p = 2, q = −1 then the only solution is α = 1, so
the recursion an = 2an−1 − an−2 has the solution an = 1n = 1 for
all n. Find a second solution which is independent of the first (i.e.
where not all an are equal). Find the general solution (with any initial
conditions) for this recurrence relation. Make a conjecture how to
proceed in general when there is only one solution α. Prove your
conjecture.

E 2.5 S. olve the recurrence relations an = an−1 + 1, an = a2n−1 and 1

an = nan−1 (for all n ≥ 2), with initial condition a1 = 1.

E 2.6 F. old a paper strip n times, halving its length every time. Then 1
unfold again. How many folding edges do you get?

E 2.7 .Suppose that n people meet and shake hands, every person 1-2
exactly once with every other person. That’s how many handshakes?
Solve the problem recursively.

E 2.8 W. hat is the number of trains of length n whose coaches each 2

have length 1 or 2? How many of these trains are symmetric, that is,
they look the same from front and back? Here is a symmetric and an
asymmetric train of length 4:

E 2.9 D. etermine the number of sequences of zeros and ones of length 2

n which contain no consecutive ones. For example for n = 3 there are
5 such sequences: 000, 001, 010, 100, 101.

. ow many subsets of {1, . . . , n} do not contain any pair of

E 2.10 H 2
consecutive numbers?

E 2.11 F. ind a recurrence relation for the number of possible bracket- 2-3
ings of n factors abc · · · . Here a bracketing must be such that when
evaluating the expression only two factors are multiplied in each
 70 2 Recursion – a fundamental idea

step. For example, the possible bracketings of n = 4 factors are

( ab)(cd), a(b(cd)), a((bc)d), (( ab)c)d, a((bc)d), but ( abc)d would not
be a bracketing.

3 E 2.12 I.n how many ways can you tile a rectangle of size 3 × n with
dominoes of size 1 × 2?

3 E 2.13 H
. ow many natural numbers are there that have n digits, where
only the digits 1, 2 and 3 occur and where consecutive digits differ by
at most one?

3 E 2.14 N
. im is a game for two players. At the start, n matches are on
the table. Players take turns, and in each move a player can take 1, 2
or 3 matches. The player who takes the last match wins. For which
n can the first player force victory, that is play in such a way that
he/she wins, whatever the second player does?

2 E 2.15 . For n ∈ N denote by on the number of ways to write n as

ordered sum of odd natural numbers. Count n itself (one summand)
as one of these ways if n is odd. For example

3 = 1 + 1 + 1 and 4 = 1 + 1 + 1 = 1 + 3 = 3 + 1

so o3 = 2 and o4 = 3. Find on for n = 1, 2, 3, 4, 5, 6, state a conjecture

and prove it. Compare Exercise E 1.9.
3 Mathematical induction

Mathematical induction is one of the most important methods for

proving statements of the form “For all natural numbers . . . ”. At its
heart it is another instance of the idea of recursion: reduce the prob-
lem to a smaller problem of the same kind. Mathematical induction
implements this idea for proofs, while recurrence relations are used
in problems where you want to determine some quantity.
In this chapter we consider two problems where mathematical
induction can be used, with more examples to come in the following
chapters. We also discuss the limitations of induction. Along the way
you will encounter another important element of problem-solving
and any scientific work: introducing suitable terms or concepts.

3.1 The induction principle

Mathematical induction is based on the following principle:

Induction principle

Let A(n) be a statement about natural numbers n. Suppose

1. Base case (BC): A(1) is true.

2. Inductive step (IS): For all n ∈ N the following implication is

true:
If A(n) is true, then A(n + 1) is also true.
Then A(n) is true for all n ∈ N.

In the inductive step A(n) is called the inductive hypothesis and

A(n + 1) the inductive claim.
The induction principle can be visualized as the domino effect: put
a row of dominoes upright on the table in such a way that each one
will knock over the next one when falling. Now if you knock over the
first one then they will all fall, see Figure 3.1.

© Springer International Publishing AG, part of Springer Nature 2018 71

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7_3
72 3 Mathematical induction

Figure 3.1 Domino effect

Here is a simple inductive proof for illustration:

Claim: For all n ∈ N we have1
n
n ( n + 1)
∑k= 2
.
k =1

Proof.
We use mathematical induction. Here A(n) is the statement that the
formula holds for the value n.
Base case: The statement A(1) is: 1 = 1·2/2. This is obviously true.
Inductive hypothesis: Let n be arbitrary, and assume A(n) is true, so
∑nk=1 k = n(n+1)/2 holds.
+1
Inductive claim: A(n + 1) is true, that is ∑nk= 1 k =
(n+1)(n+2)/2.
Proof (inductive step): We use A(n) and compute:
 
n +1 n
∑k= ∑k + ( n + 1)
k =1 k =1
n ( n + 1) n  n+2
= + ( n + 1) = + 1 ( n + 1) = ( n + 1)
2 2 2
(n + 1)(n + 2)
= ,
2
which was to be shown. q. e. d.

Practical note. After a while the correct use of mathematical induc-

tion will become second nature to you. Until then it is useful to write
out the inductive hypothesis and inductive claim fully, in order to
avoid mistakes.
Often we use variants of the induction principle. Here are some
variations, which can also be combined. You will find examples in
this and the next chapters.

J IS replaced by: For all n ≥ 2 we have A(n − 1) ⇒ A(n).

1 From now on we use the summation notation. See the list of symbols.
3.1 The induction principle 73

J BC for n = 0, IS for all n ≥ 0. More generally: BC for n = n0 , IS

for all n ≥ n0 , where n0 is some natural number.2

J IS replaced by: For all n ∈ N we have

A(1), . . . , A(n) imply A(n + 1).

J BC: A(1) and A(2) are true.

IS: For all n ∈ N we have: A(n), A(n + 1) imply A(n + 2).

What induction cannot do: Mathematical induction (or induction

for short3 ) is a useful method of proof, but it has its limitations: It
allows us to prove the formula for ∑nk=1 k, once we conjecture it. It does
not help us to find the formula. If you are faced with the task of
finding a formula for the sum of squares, ∑nk=1 k2 , you must come
up with some idea. However, if someone tells you the formula
∑nk=1 k2 = n(n+1)(2n+1)/6 then you can prove it using mathematical
induction.
Induction does not tell you “where the formula comes from”4 .
An inductive proof will give you security, but will often leave you
unsatisfied that you haven’t really understood the problem.
So induction is like a crutch. If you don’t find a better argument
then it is very useful. But it’s better to do without it whenever
possible.
This limitation is also reflected in the relation of mathematical in-
duction and recursion: When you have derived a recurrence relation
for numbers an and guessed a closed formula for the an , then you can
prove this formula using induction. The recurrence relation will be
the core of the inductive step; the initial condition will be the base
case. You saw an example of this in the formal solution of Problem
2.1. But induction will not help you find a closed formula for an .

2 Of course this will prove A(n) only for n ≥ n0 .

3 In mathematics, induction always means mathematical induction. This is not
to be confused with inductive reasoning as it is used in most other sciences, which
means to use a few special cases to conclude a general rule. This is appropriate for
other sciences but is not a sufficient argument for a mathematical proof. See Chapter
7.2 for an example.
4 For n
∑k=1 k the Gauss trick is a way to find the formula, for ∑nk=1 k2 see Problem
5.9 for a hint.
74 3 Mathematical induction

3.2 Colourings

In the example above we used mathematical induction to prove a

formula. But it can also be used for many other kinds of problems.
Here is an example.

? Problem 3.1
A number of lines are given in the plane. They subdivide the plane into
regions. Prove that you can colour the regions using two colours so that
adjacent regions always have different colours.
Here we call two regions adjacent if they have a common border. A
common vertex is not enough. See Figure 3.2.

Figure 3.2 Example of a proper colouring

 Investigation
G
Let us call a colouring proper if adjacent regions always have different
colours. Also, let us call a colouring by two colours a 2-colouring.

 Draw a few examples in order to check that the claim has a chance
of being true, and to get a feel for the problem. Try to find a
reason why it always works.

 An idea: Look at an intersection point of the lines. No matter

how many lines run through it, the number of regions having the
point as a corner will be even. Therefore we can alternate colours
and get a proper 2-colouring for these regions. We can do this at
each intersection point. However, we need to make sure that these
‘partial’ colourings fit together, since most regions will have several
3.2 Colourings 75

corners. How could we prove this? It seems hard to describe in

general how the various intersection points fit together.
Since we don’t see how to proceed5 we try something else:
 Second idea: induction with respect to the number of lines. Denote the
number of lines by n. Let us think about how an inductive proof
might work. What would the inductive step be? Given n + 1 lines
we want to find a proper 2-colouring. In order to use the inductive
hypothesis we remove one line and 2-colour the regions formed by
the remaining n lines. Then we put the line back in. Now we get
new regions, and we have to colour them.
Therefore, the central question in the inductive step will be:
Central question: Suppose n lines and a proper 2-colouring of
the regions formed by them are given. Suppose we now add
another line l. How can we adjust the original colouring to a (∗)
proper 2-colouring of the regions formed by all n + 1 lines?
We need a general rule how to adjust the 2-colouring.

Think about it!

To get an idea, let us investigate the question in the special case

n = 2. We start out with the colouring in Figure 3.2 and add a line
l as in Figure 3.3. We can leave the regions below l coloured as
before. Above l we get three new regions, and their colours must
be chosen as shown. What could be a general rule?
First attempt at a general rule: In any region which is cut by l
change the colour on one side of l (always the same side, e.g.
above l).
(Since we are using only two colours, it is clear what ‘changing’
the colour means: use the other colour.)
 Is this a good rule? That is, will we always obtain a proper colour-
ing in this way? Figure 3.4 shows that this is not the case. We
see that it is not sufficient to change the colour of regions cut by l:
5 But see Exercise E 3.9 for a solution based on this idea.
76 3 Mathematical induction

Figure 3.3 New line, I

Figure 3.4 New line, II, improper colouring

We also need to change the colour of the top middle region as in

Figure 3.5.

Figure 3.5 New line, II, proper colouring

 So our first rule doesn’t work in general. We need to modify it.

What could be a rule which also works for the line l in the second
example?
3.2 Colourings 77

Think about it!

Second attempt at a general rule: Change the colour of every

region above l.
Check that this is what we did in the examples, and try to argue
that this will always lead to a proper colouring! The full argument
is written down below. G
We now write up the full proof by induction. Pay particular attention
to the words every and arbitrary.

! Solution of Problem 3.1

Let us call a configuration of lines in the plane and the regions formed
by them a map. Let A(n) be the statement: Every map formed by n
lines has a proper 2-colouring.
Base case: A(0) is obviously true: If there is no line then there is only
one region, and we only need one colour.
Inductive hypothesis: Let n ≥ 0 be arbitrary. Suppose every map formed
by n lines has a proper 2-colouring.
Inductive claim: We want to prove A(n + 1), that is: Every map formed
by n + 1 lines has a proper 2-colouring.
Inductive step: Let an arbitrary map M formed by n + 1 lines be given.
We choose one of the lines and call it l. When we remove l we obtain
a map with n lines, which we call M
. By the inductive hypothesis
M
has a proper 2-colouring. Let us call this colouring C
.
From C
we construct a 2-colouring C of M as follows (this is the
answer to question (∗) above).
l divides the plane into two half planes. We choose one of them and
call it H. Now we start with the colouring C
and change the colours
of all the regions contained in H. This results in a 2-colouring of M
which we call C.
Claim: The colouring C is proper.
78 3 Mathematical induction

Proof: The lines of M are divided into segments by other lines. Any
border between two regions is one such segment. We check for each
segment that the two adjacent regions have different colours in the
colouring C. There are three types of segments:

1. A segment of the line l. This divides a region of M

into two parts,
both of which are regions of M. One of them lies in H. Both
regions have the same colour in C
, but the region lying in H got
its colour changed in C. So the two regions have different colours
in C.

2. A segment which is not on l and not in the half plane H. Both

adjacent regions have the same colours in C as in C
, and since C

is proper, they have different colours in C.

3. A segment which lies in the half plane H. Here both adjacent

regions lie in H, so their colours were both changed. Since they
had different colours in C
they have different colours in C.

We have shown that the new 2-colouring C is proper. This concludes

the proof of the inductive step A(n) ⇒ A(n + 1). !

 Review of Problem 3.1

In the inductive step we showed A(n) ⇒ A(n + 1). However, the
argument did not start with A(n), but with “Let an arbitrary map M
formed by n + 1 lines be given”. Only later did we use A(n). Make
sure that you understand that this makes sense.
We have introduced several concepts: proper colouring, 2-colouring,
map. This allowed us to avoid lengthy statements such as ‘a colouring
which satisfies the conditions of the problem’ or ‘the regions formed
by the lines’. This not only simplifies the written solution but also
helps us in thinking about the problem.
This was a simple example of one of the central elements of the
science of mathematics: finding good concepts.
Similarly, using short notation was also useful: M, l, C etc. 

In Chapter 4.5 we will take up colourings again.

3.3 Toolbox 79

3.3 Toolbox

You have learned about mathematical induction as a general proof

principle. It is useful for proving statements of the form “For all n
. . . ”.
It is useful to plan the induction, especially if you are not sure
whether an inductive proof would work. You ask: What would we
have to show for the inductive step?
The basic idea is the same as for recurrence relations: in order to
solve the problem of size n (prove statement A(n)) we look for a way
to reduce it to a smaller problem of the same kind (to the statement
A(n − 1) or A(n − 2) or . . . ).
Introducing suitable concepts is a fundamental task of science.
Not only does it simplify writing up arguments, but also it helps you
to think about a problem.

Exercises

E 3.1 .Prove that you can put infinitely many pins into a suitcase :-). 1

E 3.2 .Solve the recurrence relation an = a1 + · · · + an−1 with initial 1-2

condition a1 = 1 by first calculating a few values of an , stating a
conjecture and then proving the conjecture by induction.

E 3.3 .Prove that any triangulation of an n-gon (cf. Problem 2.4) has 2
precisely n − 2 triangles and n − 3 diagonals.

E 3.4 .Consider a square board divided into 2n × 2n congruent little 2

squares. Suppose that one of the little squares is colored black, all
others are white. Prove that it is possible to tile the white area by
L-shaped tiles. Here a tile covers precisely three little squares as in
Figure 3.6.

E 3.5 . Pascal’s triangle is a triangular array of numbers. The first 2

six lines are shown in Figure 3.7. You put ones at the left- and right-
hand edges, and then each other entry is the sum of the two entries
diagonally above it. Calculate the sums of the numbers along the
dotted diagonals, find a pattern, state a general rule and prove it by
induction.
 80 3 Mathematical induction

Figure 3.6 Example for Exercise E 3.4 for n = 2

1
1 1
1 2 1
1 3 3 1
1 4 6 4 1
1 5 10 10 5 1

Figure 3.7 Pascal’s triangle, see Exercise E 3.5

2 E 3.6 .Analyse the following statement and ‘proof’.

Let a be a positive real number and n ∈ N ∪ {0}. Then
an = 1.

‘Proof’: We use induction. For n = 0 it is known that a0 = 1.

This is the base case.
We now assume that an = an−1 = 1 and will show that this
implies an+1 = 1. For this purpose we write

an · an (∗) 1 · 1
a n +1 = a n · a = = = 1,
a n −1 1
where (∗) follows from the inductive hypothesis.
By induction the claim follows for all natural numbers n.

3 E 3.7 . Let n ∈ N. Consider subsets A ⊂ {1, . . . , n} which do not

contain any pair of consecutive numbers. Let p A be the product of the
elements of A. We also let p∅ = 1. Prove that the sum of all the ( p A )2
over all A equals (n + 1)!. Table 3.1 shows the sets A and the values
of p A and ( p A )2 in the case n = 4. The sum of all ( p A )2 is 5! = 120.

3 E 3.8 .The ‘Tower of Hanoi’ is a puzzle. You have three vertical rods
A, B, C and several disks of different sizes which have holes so you
can slide them onto the rods. In the beginning all disks are on rod A,
Exercises 81

A ∅ {1} {2} {3} {4} {1, 3} {1, 4} {2, 4}

pA 1 1 2 3 4 3 4 8
( p A )2 1 1 4 9 16 9 16 64

Table 3.1 Example for Exercise E 3.7

ordered by size, the largest one at the bottom. Your goal is to move
all disks to rod C. In one move you can take a single disk from one
rod and put it on another – but only if the disk is smaller than all the
disks which are already present on the new rod.
Find a formula for the smallest number of moves which is needed
to accomplish this when you have n disks. Prove the formula by
induction.

E 3.9 . Give another solution for Problem 3.1 by working out the 3
following idea: We choose any region, call it A, and colour it red.
Then we colour the other regions as follows: Let B be any region.
Choose a path from a point in A to a point in B which does not meet
any vertices. Count how often the path intersects one of the given
lines (boundaries of regions). If this number is even then B will be
coloured red, otherwise blue.
Remark: In this way you can also prove a generalisation of the state-
ment of Problem 3.1: If every vertex of a plane graph has even degree
then the faces of the graph can be coloured properly with two colours.
(The notions graph, degree and face will be introduced in Chapter 4.)
4 Graphs

The graphs which you will encounter in this chapter are very simple
objects at first glance – so simple that most people wouldn’t asso-
ciate them with mathematics. They bear no relation to formulas
or equations, nor to geometry. But thinking about them leads to a
lot of interesting mathematics, and you will discover some of that
mathematics in this chapter. You will use mathematical induction in a
new context and learn some new techniques for problem-solving, like
counting in two ways and even/odd arguments. You will encounter
a proof of impossibility – a fascinating species of mathematical propo-
sition, about the limits of what can be done. Finally, with Euler’s
formula you will get a first glimpse of the intriguing mathematical
area of topology.

4.1 The E ULER formula for plane graphs

To understand what a graph is it is best to look at some examples,

see Figure 4.1. For the moment we will only look at plane graphs;
these are drawn in such a way that the edges don’t intersect.1 Here is
a formal definition.

Definition A plane graph G consists of:

1. a finite set of points in the plane, the vertices of G;

2. finitely many lines joining vertices, the edges of G. The edges
are not allowed to intersect (except at their endpoints, the
vertices).2

1 Be careful: The graphs we are talking about here have no relation to the

graphs of functions, that you study in calculus. Therefore they are sometimes called
combinatorial graphs.
2 In most examples the edges are straight lines. However, this is not required in

the definition.

© Springer International Publishing AG, part of Springer Nature 2018 83

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7_4
84 4 Graphs

G1 G2 G3 G4 G5
Figure 4.1 Five examples of plane graphs

We assume that there is at least one vertex, but there could be no

edges. A plane graph is called connected if you can travel from any
vertex to any other vertex along edges. A path is a sequence of edges
where each edge has a common vertex with the previous edge, so
that the edges can be traversed one after another. The graphs G1 to
G4 in Figure 4.1 are connected, G5 is not.
A plane graph G divides the plane into regions, called the faces of
G. The ‘exterior’ of G is also counted as a face. For example, G1 , G3
and G4 have two faces each, G2 has three faces and G5 has one face.
In the next problem you will encounter Euler’s formula, one of
the fundamental formulas of mathematics. Its proof is a nice exercise
in mathematical induction.

? Problem 4.1
Let G be a connected plane graph. Denote by v, e, f the numbers of vertices,
edges and faces of G, respectively. Prove that

v−e+ f = 2 (Euler’s formula) (4.1)

 Investigation
G
 Look at a few examples and check the formula. Also check that
the formula is false for disconnected plane graphs. In this way you
get a feel for the formula and also for the great variety of graphs.
Graphs with many vertices can be quite complex. How can we find
a proof for the formula in spite of this variety?
 Plan: We want to prove this inductively. What would the inductive
step be? It could run as follows: In order to prove the formula for a
graph G we construct from G a smaller graph G
, use the inductive
4.1 The Euler formula for plane graphs 85

hypothesis for G
and then conclude that Euler’s formula holds
for G.
How can we make a graph smaller? We could remove a vertex or
an edge. This would correspond to mathematical induction on v
or e, respectively.
 First attempt: Induction on v. Let us try to do the inductive step.
Inductive hypothesis: Let v ≥ 1 be arbitrary. Assume that Euler’s
formula holds for all connected plane graphs having fewer than v
vertices.
Inductive step: Let G be an arbitrary connected plane graph, having
v vertices, e edges and f faces. We want to prove that v − e + f = 2.
In order to use the inductive hypothesis we remove a vertex from
G; then we also need to remove all edges attached to this vertex.
How do v, e, f change?
Denote the removed vertex by V and the new graph by G
. Let
v
, e
, f
be the number of edges, vertices, faces of G
. Clearly
v
= v − 1, but how are e
and f
related to e and f ? To get
an idea what happens, we look at some examples, see Figure
4.2. However, our arguments must be generalizable to arbitrary
connected plane graphs.
By the inductive hypothesis we have v
− e
+ f
= 2. (Or not? See
below!) We want to conclude v − e + f = 2.
 The first example in Figure 4.2 is easy: We remove one vertex and
one edge, so the difference v − e does not change: v − e = v
− e
;
also, f = f
, so v − e + f = v
− e
+ f
= 2.
The second example shows that the number of faces can change
also. Things get complicated. How can we control the simultaneous
change of v, e and f ?
Idea: Remove the edges attached to V one by one, and check how
f changes in each step: It seems clear that the number of faces
decreases by one every time we remove an edge (so −e + f and
hence v − e + f remains constant), except when removing the last
edge. When removing the last edge we are in the situation of the
first example.
 This looks promising. But the third example shows that we must be
careful: By removing an edge the graph may fall apart, i.e. become
86 4 Graphs

G G
: v
= v − 1, e
= e − 1, f
= f

G G
: v
= v − 1, e
= e − 3, f
= f − 2

G G
: not connected

Figure 4.2 Removing a vertex

disconnected.
 Summary of first attempt: It is not easy to track what happens when
removing a vertex. Along the way it is reasonable to remove edges
one by one.
This suggest another idea: We could start out by removing edges,
not vertices. That is, we could do an induction on e, not on v. Try
it!

Think about it!

 Second attempt: Induction on e. Again let us do the inductive step

4.1 The Euler formula for plane graphs 87

first.
Inductive hypothesis: Let e ≥ 1 be arbitrary. Assume Euler’s for-
mula holds for all connected plane graphs with fewer than e edges.
Inductive step: Let G be an arbitrary graph with e edges, v vertices
and f faces.
Remove one of the edges, call it E and the remaining graph G
. Let
v
, e
, f
be the numbers of vertices, edges and faces of G
.
 Clearly, v
= v and e
= e − 1 since we remove only one edge.
(We don’t remove any vertices, even if they get separated from the
remaining graph by removing E.) This is simpler than in the first
attempt, where all three quantities v, e, f could change.
 But it is still possible that the graph falls apart when removing E.
So we must take this seriously. We consider two cases:
Case 1: G
is connected.
Case 2: G
is not connected.
We make a sketch, see Figures 4.3 and 4.4. The dashed lines
represent any path in G or G
, the dots more vertices or edges.
(The drawn edges on the left and right are examples, they need
not be there.)
 Let us consider case 1: How does f change? By removing E we
amalgamate two faces into one, so we have f
= f − 1. Using
k
= k − 1 we get e − k + f = e
− k
+ f
= 2 as required.
 In case 2 G
is not connected, so we cannot use the inductive

G G

Figure 4.3 Case 1: G

connected
88 4 Graphs

G G

Figure 4.4 Case 2: G

not connected

hypothesis. How can we proceed? One idea is to pull together the

two parts and make one vertex out of the two endpoints of E, as in
Figure 4.5. Then we obtain a connected plane graph G

. Now we

G
G

Figure 4.5 Case 2: Gluing the two (former) endpoints of E into one vertex

can use the inductive hypothesis on G

. See below for details. Try

it yourself first!
 In case 2 the same face borders on E from both sides, so f
= f .
This reminds us that we need to argue carefully. Could this also
happen in case 1? In the example of Figure 4.3 it is clearly not so.
Why not? The dashed line separates the upper and the lower side
of E. So this fact must play a role in the complete argument. G 
We now write up the complete argument and add all details.

! Solution of Problem 4.1

We use mathematical induction on the number of edges of G.

Base case: e = 0. A connected graph without edges has one vertex
and one face, so v = 1, e = 0, f = 1, hence v − e + f = 2.
4.1 The Euler formula for plane graphs 89

Inductive hypothesis: Let e ≥ 1 be arbitrary. Suppose Euler’s formula

holds for all connected plane graphs with fewer than e edges.
Inductive step: Let G be an arbitrary connected plane graph with e ≥ 1
edges, v vertices and f faces.
Remove an edge E from G, and denote the resulting plane graph by
G
. Then G
has e
= e − 1 edges and v
= v vertices. Let f
be the
number of faces of G
.
We distinguish two cases.
Case 1: G
is connected. We first show that E borders two different
faces of G. Let u, v be the vertices connected by E. If u = v then E will
encircle a face, and the claim is obvious. Now assume u = v. Since
G
is connected there is a path in G
from u to v. When combining
this path with E we obtain a closed path3 P. One face bordered by
E must lie in the region enclosed by P and one face must lie outside.
Therefore they must be different faces.
In G
the two faces bordered by E become one face, all other faces
remain unchanged. Therefore f
= f − 1. Because e
< e we can use
the inductive hypothesis for G
and obtain

2 = v
− e
+ f
= v − ( e − 1) + ( f − 1) = v − e + f .

Case 2: G
is not connected. Then both sides of the edge E border
the same face. Therefore

v
= v e
= e − 1, f
= f .

We form a new connected graph G

by gluing the two endpoints of E

into one new vertex. This does not change the number of edges or
faces, but one vertex disappears, therefore, with obvious notation,

= v
− 1 = v − 1, e

= e
= e − 1, f

= f
= f .

Because e

< e we can use the inductive hypothesis for G

and obtain

2 = v

− e

+ f

= (v − 1) − (e − 1) + f = v − e + f .

In each case we have proven v − e + f = 2. This concludes the

inductive step. !
3A closed path is a path whose first and last vertex are the same.
90 4 Graphs

 Review of Problem 4.1

We first tried induction on v. For the inductive step we removed one
vertex. Then we had to remove all edges attached to this vertex. If
there were several such edges we had to analyse the simultaneous
changes of v, e, f . This suggested that we try an induction on e instead.
This was simpler since v does not change when we remove an edge.

Stay flexible! Be prepared to change your route if you get a better

idea! 

Remark

Here is an alternative solution for the second case:

Since G was connected and we only removed one edge, G

consists of two connected parts, G1

and G2
(the left- and right-
hand pieces of G
in Figure 4.4). Also, both sides of E border the
same face (the ‘exterior’ face in the figure). Let v1
, e1
, f 1
be the
numbers of vertices, edges and faces of G1
and v2
, e2
, f 2
those of
G2
. Then
v1
+ v2
= v, e1
+ e2
= e − 1 .
In addition, we have f 1
+ f 2
= f + 1 since the exterior face is
counted both for G1
and for G2
.
Because e1
< e, e2
< e we can use the inductive hypothesis for
G1 and G2
. We add and obtain:

v1
− e1
+ f 1
= 2
v2
− e2
+ f 2
= 2
+
v − ( e − 1) + ( f + 1) = 4.

Subtracting 2 yields v − e + f = 2.

4.2 Counting in two ways for graphs

Counting in two ways is a simple powerful idea that you can use in
many contexts to derive interesting formulas.
4.2 Counting in two ways for graphs 91

? Problem 4.2
Let G be a plane graph. Put a mark (a ‘custom house’) at both sides of each
edge, see Figure 4.6. Count the number of custom houses in two ways.

Figure 4.6 How many custom houses?

! Solution
First count: There are two houses at each edge. Therefore there are 2e
houses.
Second count: Count the houses in each face (‘country’) and add. To
write a formula we use the notation

bF = the number of borders of the face (country) F.

Then the total number of houses is ∑ bF .

F face of G
Note that the single edge on the right is counted twice as a border of
the exterior face. So the exterior face has 5 borders, the other faces
have 3 borders each. !
What follows? Since we counted one thing in two ways, the results
must be the same, and we get the edge-face formula

2e = ∑ bF (4.2)
F face of G

In the example: 2 · 7 = 5 + 3 + 3 + 3. You will find interesting

applications of this formula in Problem 4.6 and in Exercise E 4.4.
If you don’t like custom houses, say ‘one side of an edge’ instead,
or even better argue like this:
92 4 Graphs

Alternative proof of formula (4.2): We count the pairs ( E, F ) where E

is an edge and F is a face bordered by E. When both sides of E border
the same face F we count ( E, F ) twice.
First count: For each edge E there are two such pairs (or one counted
twice). So the number of pairs is 2e.
Second count: For face F there are precisely bF pairs that have F as
second component. So the number of all pairs is ∑ bF .
F
We get formula (4.2) again. Here is an even shorter proof.
Third proof of formula (4.2): Count the borders of each face. You will
count each edge twice, once from each side. Therefore we get (4.2).
Since this worked so well we try it again.

? Problem 4.3
Let G be a plane graph. Draw an arrow wherever an edge leaves a vertex,
see Figure 4.7. Count the number of arrows in two ways and deduce a
formula.

Figure 4.7 How many arrows?

! Solution
First count: Each edge carries two arrows. Therefore there are 2e
arrows.
Second count: For each vertex V there are as many arrows as there are
edges containing V. We let

dV = number of edges containing V (4.3)

(where we count an edge twice if both its endpoints are the vertex V).
Then the number of arrows is ∑ dV . From the two counts we
V vertex of G
get the edge-vertex formula

2e = ∑ dV (4.4)
V vertex of G
4.2 Counting in two ways for graphs 93

!
Here it was not essential that the graph was plane. The argument
still works if edges intersect. It is also not important how the graph is
drawn. This motivates the general definition of a graph, which does
not refer to any concrete picture of it.

Definition A graph is given by a finite set V whose elements

we call vertices, a finite set E whose elements we call edges, and
a rule that associates to each E ∈ E two vertices V1 , V2 ∈ V . It
is permitted that V1 = V2 . We call V1 , V2 the endpoints of E and
say that E joins the vertices V1 , V2 . If V1 = V2 then we call E a
loop.
The number dV defined in (4.3) for a vertex V is called the
degree of V.

Often we visualise a graph as a diagram in which vertices are given

by points in the plane and each edge is represented by a line or curve
joining its endpoints but not meeting any other vertices. A plane
graph is a representation in which edges don’t cross.
Figure 4.8 shows two representations of the same graph: four
vertices, with any two vertices joined by a single edge. The left-hand
representation is a plane graph, the one on the right is not.
Graphs are a useful tool for visualising complex structures. You
will find examples in Problems 4.4 and 10.4.

Figure 4.8 Two representations of the same graph

94 4 Graphs

We summarize:

Graph formulas

For a graph with e edges we have

2e = ∑ dV .
V vertex of G

For a plane graph with e edges we have

2e = ∑ bF .
F face of G

For a plane connected graph with v vertices, e edges and f faces

we have Euler’s formula

v − e + f = 2.

You should know these formulas well. They can often be used for
problem-solving.

4.3 Handshakes and graphs

The following problem shows how to model a problem using graphs

and how to use formula (4.4).

? Problem 4.4
At a party some of the guests shake hands. Show that at any point in time
the number of guests who have shaken an odd number of hands is even.

! Solution
Consider the graph whose vertices are the guests, where two vertices
are joined if and only if the guests have shaken hands. For each
guest V the degree dV is the number of handshakes she has made. By
formula (4.4) the sum of all dV is 2e, an even number. Therefore the
number of vertices (guests) V for which dV is odd must be even. !
4.4 Five points in the plane, all joined by edges 95

4.4 Five points in the plane, all joined by edges

How can we decide whether a given graph can be represented as a

plane graph? Let us start with a little warm-up problem.

? Problem 4.5
In Figure 4.9, can you connect boxes A with A, B with B and C with C so
that the connecting lines don’t intersect?

C A

A B C

Figure 4.9 Looking for connecting lines

Think about it!

 ! Investigation and solution

G
At first it looks impossible. Let us try to simplify. But consider this:
first interchange the two upper boxes A, C, then connect the upper
boxes with the lower boxes by vertical lines and finally pull A and
C to their original positions (pulling the connecting lines along like
rubber bands). This shows that it is possible! G!
 Review
The problem shows that we must be careful not to jump to conclusions
of the kind “This is clearly impossible!”. 
Let us try a more difficult problem.
96 4 Graphs

? Problem 4.6
Can you draw 5 points in the plane with a line joining each pair of them, so
that the joining lines don’t intersect?
Put differently: Does the graph G given by 5 vertices, with any two
vertices joined by a single edge, have a plane representation? Recall
that with 4 vertices this is possible, see the left-hand picture in Figure
4.8.
B

C A

D E

Figure 4.10 A non-plane representation of G and the failed attempt at a

plane representation

 Investigation
G
 Try it! – After a while you will probably conjecture that it is not
possible. For example, in Figure 4.10 all lines except the one from
A to D are drawn, and it is clear that we cannot draw the missing
line without intersections. This is because A lies outside the closed
path BCEB while D lies inside. So we formulate the conjecture: It
is impossible.
 But watch out: The picture does not prove the impossibility. Per-
haps we chose the previous lines badly, and if we had started out
differently then it would have worked. Problem 4.5 showed us
that we need to be careful. We need a general argument, one that
works no matter how we try to draw the lines.
 What do we want to prove? An impossibility. How do you prove
that something is impossible? You use a proof by contradiction:
You suppose that it is possible, and from this you derive a contra-
diction or a conclusion which is obviously false.
4.4 Five points in the plane, all joined by edges 97

 So let us assume that we have a plane representation of G. What

can we say about it? Let us collect information about G. For
example: What are v, e, f ? How many borders do the faces have?

Think about it!

 We know that the number of vertices is v = 5. Also, we have e = 10

since there are 10 ways to choose two out of five vertices (see the
left-hand part of Figure 4.10).
 What next? Recall our graph formulas. Euler’s formula can
be used since G is a plane connected graph, and it tells us that
e − k + f = 2. From e = 5, k = 10 we get f = 7.
 What else could we use? For example the vertex-edge formula
(4.4). What are the vertex degrees? Each vertex is contained in 4
edges – one to every other vertex. So dV = 4 for all V. The formula
gives 2 × 10 = 4 + 4 + 4 + 4 + 4, that is 20 = 20. That’s no news.
 So let us try to use the edge-face formula (4.2). The left-hand side
is 2e = 20. Can we say anything about the right-hand side, i.e.
about the numbers bF ? What can bF be? Could we have bF = 2 for
example, i.e. could there be a face with two borders? – No. This
would mean that the two borders have the same endpoints, see
Figure 4.11. But there is only one edge joining any two vertices.
Similarly bF = 1 is impossible since it would mean that the face F
is surrounded by a loop; but there are no loops in G.

Figure 4.11 Faces with one or two borders

 Therefore we have bF ≥ 3 for all faces F. Putting everything

together we get

20 = 2e = ∑ bF ≥ ∑ 3 = 7 · 3 = 21 , (4.5)
F face of G F face of G
98 4 Graphs

where we used that G has f = 7 faces. So we derived the false

conclusion 20 ≥ 21 from the assumption that G has a plane rep-
resentation. Therefore this assumption must have been wrong.

G
Let us write up the argument cleanly.

! Solution of Problem 4.6

This is impossible, i.e. there is no plane graph with 5 vertices where

any two vertices are joined by an edge.
Proof.
We prove this by contradiction. Suppose there was such a plane
graph G. Let v, e, f be the numbers of vertices, edges and faces
of G. By assumption we have v = 5 and e = 10. Because G is
plane and connected, Euler’s formula v − e + f = 2 implies f = 7.
Because G has no loops and no double edges, every face has at least
three borders, see Figure 4.11. The edge-face formula now gives the
equations and inequalities (4.5), so 20 ≥ 21. Because this is obviously
false, our assumptions that there is such a plane graph G must have
been wrong. q. e. d.

!
 Review of Problem 4.6
This was a first example of a proof of impossibility:4 We were able to
prove that every one of the infinitely many possibilities of drawing
G must have intersecting edges. Our argument was indirect: By
assuming there were no intersections, we derived a false statement. 

4.5 Going further: E ULER’s formula for polyhedra,

topology and the four colour problem

Graphs and polyhedra

Euler’s formula for graphs is a generalisation of the following:

4 In Chapters 7 and 11 you will find more.

4.5 Going further 99

Euler’s formula for polyhedra: Let v, e, f be the number of

vertices, edges and faces of a convex polyhedron in space. Then

v − e + f = 2.

A convex polyhedron is a convex three-dimensional body bounded

by planar faces.5 Some examples of convex polyhedra are the cube,
the tetrahedron, the octahedron, the dodecahedron, the icosahedron,
pyramids . . . , but not a cylinder. Check for these examples that the
formula holds.
How are convex polyhedra related to plane graphs? Think of a wire
model of the polyhedron and position a light source just outside one
of the faces. It will cast a shadow on a wall which is parallel to that
face and on the other side of the polyhedron. This shadow is a plane
graph. Faces of the polyhedron correspond to faces of the graph,
with the face where the light source is positioned corresponding
to the exterior face of the graph. Therefore, Euler’s formula for
polyhedra follows from Euler’s formula for graphs. The left-hand
picture in Figure 4.8 shows the shadow of a tetrahedron (three-sided
pyramid). Draw the shadows of the cube and the dodecahedron
(twelve pentagons)!
You would get the same plane graph by making the surface of
the polyhedron out of rubber, cutting out one face and then pulling
everything flat.
By the same argument everything that we derived for plane graphs
is also true for graphs that can be drawn on a sphere (surface of a
ball) without intersections.

A short excursion into topology

Euler’s formula is the starting point for the mathematical area of

topology. The basic question of topology is: how can you distinguish
a sphere (surface of a ball) from a torus (bicycle tire)? More precisely:
How can you express mathematically that the torus has a ‘hole’ that

5 Convex polyhedra can also be characterized as: 1. The convex hull of finitely

many points in space, i.e. the smallest convex set containing these points. 2. The
intersection of finitely many half spaces, assuming this is bounded. The equivalence
of these conditions is intuitively clear but not easy to prove.
100 4 Graphs

you can put your arm through, while the sphere doesn’t – and that this
property is preserved when we deform these surfaces continuously?6
Euler’s formula allows us to find an answer to this question, as
follows:

1. We consider (connected) graphs that can be drawn on the sphere or

on the torus without intersections. As explained above graphs on a
sphere are like plane graphs, for example the formula v − e + f = 2
holds. However, for graphs on the torus this is not the case.

2. For connected graphs on the torus there is also a version of Euler’s

formula. But its right hand side is 0, not 2. Also, the formula only
holds for connected graphs on the torus that satisfy an additional
condition: Every face must be such that it can be drawn in the
plane.7 For such graphs on the torus we have

Euler’s formula for graphs on the torus: v − e + f = 0.

See Figure 4.12 for an example. This is to be understood as follows:

Consider the square of paper whose corners are the four points
marked A. Glue together the left and right sides (with the single
arrows), to get a cylinder. Then glue together the upper and lower
sides (with the double arrows) – stretching the paper as necessary
–, to get a torus. The four points marked A will be glued into a
single point of the torus, and the same is true for the two Bs and
the two Cs. Therefore the graph on the torus has 4 vertices, 8 edges
(four containing D, two connecting A and B – not four since the
upper and lower one are glued together – and two connecting A
and C) and 4 faces, so v − e + f = 4 − 8 + 4 = 0. See Exercise E 4.8
for more on this.

3. There are also versions of Euler’s formula for other surfaces. For
a ‘torus with two holes’ you get the formula v − e + f = −2, for
three holes (surface of a pretzel) you get v − e + f = −4. In general

6 Thinkof turning the sphere into a long sausage, then knotting the sausage.
Topologically this is still like a sphere, but what is the essential difference from a
torus, which may also be knotted?
7 For example this is not the case for the graph consisting of a single vertex and

no edges, or of two vertices connected by a single edge. When drawing this on a

torus the resulting face can not be flattened into the plane.
4.5 Going further 101

B
A A

D
C C

A A
B

Figure 4.12 A graph on the torus

for g holes you get v − e + f = 2 − 2g. The number that you get
on the right is called the Euler characteristic of the surface.

4. The Euler characteristic answers our question above: It is a well-

defined quantity that can be computed for any surface and that
‘measures’ a difference between sphere and torus.8 Also, it does not
change under deformations of the surface. Therefore, the Euler
characteristic is called a topological invariant.9

See the References section at the end of the book for further reading
on topology.

Colourings of plane graphs

? Problem 4.7
Suppose we want to colour the faces of a plane graph in such a way that any
two faces which share a border must get different colours. We call such a
colouring proper. What is the smallest number of colours which is enough
for a proper colouring of any plane graph?

8 You may find this unsatisfying: While the Euler characteristic distinguishes
sphere and torus it does not intuitively reflect the ‘hole’ in the torus. But it has the
advantage of being easily defined. There is a way to talk about holes mathematically.
It is called homology theory, but it takes a lot more effort to define it.
9 You will learn more about invariants in Chapter 11.
102 4 Graphs

J It is easy to see that you need at least 4 colours, for example for
the graph on the left of Figure 4.8, since every one of the four faces
borders on every other face. If you carefully try out some more
examples then you will notice that four colours always seem to
suffice (be enough).
J This leads to the 4-Colour Conjecture: Four colours suffice for any
plane graph. In other words: The countries in any plane map can
be coloured with four colours in such a way that countries which
share a border get different colours.
J It turned out to be extremely difficult to prove this conjecture. The
conjecture was formulated the first time in 1852, but in spite of the
efforts by many mathematicians the first proof was only obtained
in 1976 (so now it is the 4-Colour Theorem). Unfortunately part
of the proof consists in checking a large number of special cases,
too many to do by hand. So a computer was used for this. Even
now, no proof is known which does not rely on computer help.10
Exercises E 4.10 to E 4.14 allow you to retrace some of the steps on
the way to the four colour theorem. There you will also see that
the analogous problem for the torus is simpler. Here the smallest
number of colours is 7.

4.6 Toolbox

While thinking about Problem 4.1 you saw that it is important to stay
flexible, to change a chosen route if you get a better idea. Problem 4.5
showed that you need to be very careful not to jump to the conclusion
that “This is clearly impossible!” – often it is hard or even impossible
to have all possibilities in mind. Nevertheless, mathematics allows
us sometimes to give rigorous proofs of impossibility, for example
using the technique of proof by contradiction. You also learned the
technique of counting in two ways and some of its uses, for example
in the proof of impossibility in Problem 4.6.
Finally in Problem 4.4 you saw a simple example of how to use
graphs as representations of complex webs of relationships.

10 You find more on the interesting history of this problem in (Aigner, 1987), for
example.
Exercises 103

Exercises

E 4.1 .Can the sum of 111 odd numbers be even? 1

E 4.2 . Is it possible that in a group of 57 people each person has 1

exactly three friends in the group? We want to assume that the
friendship relation is symmetric, that is: if A is a friend of B then B is
a friend of A.

E 4.3 .Finish the first attempt at a proof of Euler’s formula (induc- 2

tion on v).

E 4.4 .Is there a polyhedron with exactly 7 triangular and no other 1-2
faces? (See Section 4.5 for the definition of a polyhedron.)

E 4.5 . Imagine a house that has a front door but no back door 2
or other entrance. Can you be sure that there is a room in the
house which has an odd number of doors? (Rooms include kitchen,
bathrooms, corridors etc.)

E 4.6 . Each of three houses needs to be connected to the power 2-3

plant, to the waterworks and to the gasworks. Is it possible to do this
so that none of the 3 × 3 = 9 connecting lines intersect, that is, none
of the wires or pipes cross above or below another?

E 4.7 .The surface of a (classical) soccerball is made of pentagonal 2-3

and hexagonal patches which are sewn together at their edges. At
each ‘vertex’ three seams meet. Using this information determine the
number of pentagons.
Can you also determine the number of hexagons?11

E 4.8 .Prove Euler’s formula for the torus which is mentioned in 3

Section 4.5.

E 4.9 .Find an alternative solution to Problem 4.6. A natural idea is 2

‘inside-outside’: If it was possible to do it then the connections from
A to B, B to C and C to A would form a closed path in the plane
which does not intersect itself. Such a path divides the plane into
two regions, the inside and the outside, and every curve from an
11 The
same graph structure as for a soccerball appears in the material fullerene, a
carbon molecule.
 104 4 Graphs

interior point to an exterior point must intersect this closed path.12

Now consider the possible positions of D and E.

1-2 E 4.10 . Prove that there is no plane graph with five faces each of
which borders on every other face. Why does this not immediately
imply the 4-Colour Conjecture?

2-3 E 4.11 .Let G be a plane graph all of whose vertices have degree at
least 3. Prove that there is a face with at most 5 borders.

3 E 4.12 .Use Exercise E 4.11 to prove the 6-Colour Theorem: the faces
of every plane graph can be coloured properly with 6 colours.

2 E 4.13 .Draw a graph on the torus which has 7 faces, each bordering
every other face.

3 E 4.14 . Exercise E 4.13 shows that there is a graph on the torus for
which you need 7 colours. Prove the 7-Colour-Theorem for the torus:
the faces of any graph on the torus can be coloured properly with 7
colours.

3 E 4.15 .Consider the left-hand graph in Figure 4.13. Try to draw it in

one go, that is, without lifting the pencil and so that you traverse each
edge exactly once.13 There are many ways to do this, but you can
only do it if you start at certain vertices. Which vertices are these?

Figure 4.13 The house of Santa Claus and variants

Investigate which of the other figures can be drawn in one go. Can
you find a criterion to decide for a given graph whether it can be
drawn in one go?

3 E 4.16 .A polyhedron is called regular if the same number of edges

12 This is intuitively clear, but it is not easy to give a rigorous proof. This fact is
known as the Jordan Curve Theorem. You may use it in your argument without
proof.
13 This is a well-known German children’s game. While you draw it you say ‘Das

ist das Haus vom Nikolaus’ – ‘This is the house of Santa Claus’.
Exercises 105

meet at each vertex and every face has the same number of ver-
tices. Prove that cube, octahedron, tetrahedron, dodecahedron and
icosahedron are the only regular convex polyhedra.

E 4.17 .Label the vertices of a triangle with the numbers 1,2,3. Now 3
introduce some additional vertices in the interior and on the edges,
and add edges so that every internal face is a triangle. Then label the
interior vertices 1, 2 or 3 however you like, and label each new vertex
on an edge of the big triangle with one of the numbers at the end
points of this edge.
Prove that there is a (small) triangle all of whose vertices have different
labels.
Note: This assertion is known as Sperner’s Lemma. It can be used
to prove quite amazing results, for example Brouwer’s Fixed Point
Theorem.
5 Counting

In how many ways can you arrange 5 objects in a row? How many
poker hands have two pairs? Counting is one of the original purposes
of mathematics. You can find counting problems in everyday life
and in calculating probabilities (how likely is it to have two pairs in
a poker hand?). You have already seen some counting problems in
previous chapters and learned about the recursion technique. In this
chapter we will take a systematic look at counting problems.
Counting is not only a goal in itself, it can also be used to achieve
other goals. For example, counting the same objects in two ways can
lead to many interesting formulas. Variations of the basic idea of
counting in two ways occur in many areas of mathematics, and some
of them are explained at the end this chapter.

5.1 The basic principles of counting

Most counting problems can be expressed most clearly in the language

of sets and tuples (ordered pairs, triples etc.). If you are not familiar
with these notions you should consult Appendix B.
A counting problem can always be stated as counting the elements
of a finite set X. We denote

| X | = number of elements of X .

This is sometimes called the cardinality of X.

The fundamental counting rules

The two fundamental rules of counting are the addition rule and the
multiplication rule:

© Springer International Publishing AG, part of Springer Nature 2018 107

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7_5
108 5 Counting

Fundamental counting rules

1. Addition rule: If X is the disjoint union of the sets X1 , X2 , . . . , Xr ,

X = X1 ∪ · · · ∪ X r , Xi pairwise disjoint,

then | X | = | X1 | + · · · + | Xr |.

2. Multiplication rule: If you can specify the elements of X

uniquely by first making one decision, with n possible out-
comes, and then making another decision, with m possible
outcomes (where m is independent of the outcome of the first deci-
sion) then | X | = nm.
More generally for s ≥ 2 decisions: If there are n1 possible
outcomes for the first decision, n2 for the second etc., where
these numbers are independent of all previous decisions, then | X | =
n1 , . . . , n s .

The addition rule is so obvious that is often used without comment.1

We have already used it in Chapter 2 when we divided the set of
things to be counted (e.g. domino tilings, triangulations) into disjoint
classes and then counted each class separately.
The multiplication rule follows from the addition rule: Say the
possible outcomes of the first decision are 1, . . . , n. Let Xi be the set
of those elements of X where the first decision has outcome i. Then
X is the disjoint union of X1 , . . . , Xn . Also, | Xi | = m for each i, so
| X | = | X1 | + · · · + | Xn | = m + · · · + m = nm. The rule with s > 2
decisions follows similarly by induction on s (exercise!).
An important special case of the multiplication rule is counting the
number of elements of the cartesian product of two sets: If A, B are
finite sets then
| A × B| = | A| · | B| , (5.1)
since in order to specify uniquely an element ( a, b) ∈ A × B, you first
choose a in | A| possible ways (first decision) and then for each of
these choices you can choose b in | B| possible ways (second decision).
Equation (5.1) is also obvious if you arrange the elements of A × B in
a rectangular array as in Appendix B.
1 If you want a formal proof try mathematical induction.
5.1 The basic principles of counting 109

Examples for the multiplication rule

? Problem 5.1

1. On a menu there are 3 appetisers, 5 main courses and 2 desserts. How

many 3-course meals can you put together?

2. How many 2-digit numbers are there?

3. How many pairs of numbers ( a, b) are there, where a, b ∈ {1, . . . , 20}

and a = b?

4. How many pairs of numbers ( a, b) are there, where a, b ∈ {1, . . . , 20}

and a < b?

Think about it!

It may be useful for you to write up the possibilities systematically.

! Solution
1. First decision: which appetiser. Second decision: which main
course. Third decision: which dessert. Each decision can be made
independently of the outcomes of all previous decisions. Therefore
there are 3 · 5 · 2 = 30 different 3-course meals.
2. For choosing the first digit there are the 9 possibilities 1, . . . , 9 (first
decision), for choosing the second digit there are the 10 possibilities
0, . . . , 9 (second decision). So there are 9 · 10 = 90 numbers.
Second solution: These are all numbers from 10 to 99, that is,
99 − 10 + 1 = 90 numbers.2

2 Not 99 − 10 since you need to include the first and the last one. Compare the
shift by one in Problem 1.1.
110 5 Counting

3. First decision: Choose a. There are 20 possibilities.

Second decision: Choose b. There are only 19 possibilities since we
require b = a. Therefore, there are 20 · 19 = 380 pairs.
Important: Which numbers are possible for b depends on the choice
of a. However, how many numbers are possible for b does not
depend on the choice of a. Therefore we can use the multiplication
rule, but we are not counting the elements of a product set (as in
the special case mentioned above).
Second solution: There are 20 · 20 = 400 pairs ( a, b) with a, b ∈
{1, . . . , 20}. From these we subtract the ‘forbidden’ ones, i.e. those
having a = b. There are 20 of them, so we get 400 − 20 = 380 pairs.
4. If we first choose a then the number of possibilities for b depends
on what a is: if a = 1 then there are 19 possibilities, if a = 2 there
are 18 etc. Therefore the multiplication rule can not be used. What
now? Of course we could calculate 19 + 18 + · · · + 1 + 0, but is
there a simpler way? Problem 3 was similar and could be solved by
the multiplication rule. How can we relate Problem 4 to Problem 3?

Think about it!

Let

X = {( a, b) : a, b ∈ {1, . . . , 20}, a < b}

Y = {( a, b) : a, b ∈ {1, . . . , 20}, a = b}

be the sets to be counted in Problems 4 and 3 respectively. For each

pair in X we have two corresponding pairs in Y; for example the
pair (2, 5) ∈ X corresponds to the pairs (2, 5), (5, 2) ∈ Y. Therefore
|Y | = 2| X |, hence | X | = |Y |/2 = 380/2 = 190. !
The idea for solving Problem 5.1.4 was to count Y instead of X, and
this amounted to double-counting X. This is often useful:

Multiple-counting principle

If you can’t count something, try multiple-counting.

5.1 The basic principles of counting 111

We can make this precise as follows: Suppose we want to count a

set X. Then we try to find a set Y having the following properties, for
some m ∈ N:

a) We can count Y.

b) To each element of X we can assign a group of m elements of Y

in such a way that the groups don’t overlap and Y is the union
of all groups.

Then | X | = |Y |/m.

Proof: There are | X | groups, whence |Y | = | X | · m, therefore

| X | = |Y |/m.

In the solution of Problem 5.1.4 we assign to ( a, b) ∈ X the elements

( a, b), (b, a) ∈ Y. See Figure 5.1 for the analogous problem with
a, b ∈ {1, 2, 3}.

(1, 2)
(1, 2)
(2, 1)

(1, 3)
(1, 3)
(3, 1)

(2, 3)
(2, 3)
(3, 2)

X Y

Figure 5.1 Double counting of the pairs ( a, b) with a, b ∈ {1, 2, 3} and a < b

Additional remarks: The assignment in b) is not a map in the

mathematical sense (see Appendix B) since a map assigns a
unique image element, not several, to each element of its do-
main of definition. But if we turn round the arrows in Figure 5.1
112 5 Counting

then we do obtain a map! Therefore we can formulate condition

b) as follows:
There is a map f : Y → X with the property that for each x ∈ X
the preimage f −1 ({ x }) := {y ∈ Y : f (y) = x } has precisely m
elements.
Then the ‘groups’ of b) are precisely these preimages. Note that
the other conditions in b) – the groups don’t overlap and their
union is all of Y – are satisfied automatically since f is a map.
As an exercise you should check this in detail.
How can we write down the map f for Problem 5.1.4? f maps
the pair ( a, b) ∈ Y to the pair (min{ a, b}, max{ a, b}) ∈ X where
min{ a, b} is the smaller and max{ a, b} is the bigger of the
numbers a, b.

Further examples of the multiple-counting rule

? Problem 5.2
A game uses tiles in the shape of equilateral triangles, whose top faces are
marked with three different numbers from the set {0, . . . , 5}. How many
different tiles are there?

! Solution

Imagine the tiles. We can put a tile in front of us in different ways:

1 2 3
1

1
2

2
3

, , are the same tile. Similarly we can turn any

other tile, placing it in three different orientations. So if we count

fixed triangles marked by three different numbers then we have

counted each tile three times.
Therefore we first count how many ways we can mark a triangle
with three different numbers: For the left vertex there are 6 possibili-
ties, then for the right vertex there are 5 and then for the top vertex
4 possibilities. Therefore there are 6 · 5 · 4 = 120 marked triangles.
Since we fixed the left, right and top vertex, the three triangles above
are counted as different.
5.1 The basic principles of counting 113

By the multiple-counting principle there are 120/3 = 40 tiles. In

this case X is the set of possible tiles and Y is the set of marked
triangles. !

? Problem 5.3
The game Triominoes uses tiles as in the previous problem, but the numbers
on the tiles need not be distinct. How many different tiles are there?

! Solution
Again we first count how many ways we can mark a fixed triangle.
Since the numbers need not be distinct, we have 6 possibilities for
each vertex, so by the multiplication rule there are 6 · 6 · 6 = 216
triangles.
Does every tile correspond to three marked triangles as before?

Think about it!

No, since a tile with three equal numbers only corresponds to one
marked triangle. There are 6 such tiles. All other tiles correspond to
three marked triangles as before. Therefore we apply the multiple-
counting principle to those triangles whose three numbers are not
all the same. There are 216 − 6 = 210 such triangles, so 210 3 = 70
corresponding tiles. In addition, there are the 6 tiles with all numbers
equal. Altogether we have 76 tiles. !

 Review
The multiple-counting principle can only be applied if the size of the
multiple-counts (m) is always the same. If this is not the case then one
can try to divide the set that is to be counted into subsets (classes) to
which the principle is applicable. So here we first used the addition
rule and then the multiple-counting rule. 
114 5 Counting

The main counting problems

Using the fundamental counting rules we can solve the most impor-
tant general counting problems. Many real-life counting problems
can be translated into one of these basic types.
Let A be a set with n elements: | A| = n. We want to count tuples
and subsets of elements of A. We denote the size of the tuple or
subset by k ≥ 1. We always consider the case k = 2 first since the
principles are most easily understood in this case.
You should work out a special case for each type of counting
problem, for example A = {1, 2, 3, 4} and k = 3. Write out all
possibilities and make sure you understand the counting principle.

Number of pairs: |{( a, b) : a, b ∈ A}| = n2

Proof by the multiplication rule:

number of possibilities for a: n
and for each of them: number of possibilities for b: n
This is the special case of equation (5.1) where A = B.

Number of k-tuples: |{( a1 , . . . , ak ) : a1 , . . . , ak ∈ A}| = nk

Proof by the multiplication rule:

number of possibilities for a1 : n
and for each of them: number of possibilities for a2 : n
..
.
and for each of them: number of possibilities for ak : n
Number of pairs with different components:

|{( a, b) : a, b ∈ A, a = b}| = n(n − 1)

Proof by the multiplication rule:

number of possibilities for a: n
and for each of them: number of possibilities for b: n−1
Number of k-tuples with different components:

|{( a1 , . . . , ak ) : a1 , . . . , ak ∈ A, all ai different}|

= n ( n − 1) . . . ( n − k + 1)
5.1 The basic principles of counting 115

Proof by the multiplication rule:

number of possibilities for a1 : n
and for each of them: number of possibilities for a2 : n−1
..
.
and for each of them: number of possibilities for ak : n−k+1
(This argument is for k ≤ n, but the formula is also true for k > n
since then both sides are equal to zero.)
An important special case is k = n:

Number of orderings:

|{Orderings of n different objects}| = n!

Here n! = 1 · 2 · · · · · n (n factorial). Another word for ordering is

permutation.

Number of 2-element subsets:

n ( n − 1)
|{subsets of A with 2 elements}| =
2

Proof by the multiple-counting principle, compare Problem 5.1.4:

Each subset with 2 elements can be written as { a, b} where a < b.
To each such subset we assign the pairs ( a, b) and (b, a). Clearly we
obtain all pairs with different components, and each such pair arises
from a unique subset. Since there are n(n − 1) pairs with different
components in A we get the formula.
In short: When counting pairs ( a, b) with a = b instead of subsets
then we count each subset twice.

Number of k-element subsets:

n ( n − 1) . . . ( n − k + 1)
|{subsets of A with k elements}| =
k!

Proof by the multiple-counting principle: Each k-element subset of

A can be written { a1 , a2 , . . . , ak } with a1 < a2 < · · · < ak . To each
subset written in this way we assign all k-tuples which can be formed
from the ai in all possible orderings. There are k! such orderings.
116 5 Counting

In this way we obtain all k-tuples with different components, and

each such k-tuple arises from a unique subset. Since there are n(n −
1) . . . (n − k + 1) such k-tuples the formula follows by the multiple-
countingprinciple where m = k!.3
Since the expression in the last problem appears frequently, there is a
short notation for it.

Definition Let n, k ∈ N0 , 0 ≤ k ≤ n. The binomial coefficient

(nk) (read: n choose k) is defined for k ≥ 1 by
 
n n ( n − 1) . . . ( n − k + 1)
= .
k k!

For k = 0 we define (n0 ) = 1.

n ( n −1) n(n−1)(n−2)
Examples: (n1 ) = n, (n2 ) = 2 , (n3 ) = 6 etc.4
Every set has precisely one 0-element subset (the empty set), so
using the last result above we get for all k, n with 0 ≤ k ≤ n:
 
n
Number of k-element subsets of an n-element set:
k

The numbers (nk) appear in Pascal’s triangle, which is introduced in

Exercise E 3.5, and in the general binomial theorem. See Exercises
E 5.14 and E 5.15.

Examples using the main counting problems

Many counting problems can be reduced to one of the basic types.

You only need to recognise it! Here are a few examples.

? Problem 5.4
10 people meet. Each one shakes hands with every other person exactly once.
That’s how many handshakes?
3 Thecorresponding map – see the additional remarks before Problem 5.2 – is
very easy to write down in this case: if (b1 , . . . , bk ) is a k-tuple with all bi different
then f (b1 , . . . , bk ) = {b1 , . . . , bk }.
4 A good way to remember the lengthy expression in the numerator is to notice

that it has precisely k factors, just like the denominator.

5.2 Counting using bijections 117

! Solution
Let A be the set of 10 people. Each handshake corresponds to a 2-
10·9
2 ) = 2 = 45.
element subset of A. So the number of handshakes is (10
!
The same counting problem appears in many disguises, for example:
number of connections between n points; number of pairs ( a, b) with
a, b ∈ {1, . . . , n} and a < b (note that these pairs correspond precisely
to 2-element subsets since any such subset can be written as { a, b}
with a < b; compare Problem 5.1.4).

? Problem 5.5
A group of 10 women and 10 men go dancing. How many ways are there
to make up 10 mixed pairs?

! Solution
We number the women 1, . . . , 10. The first woman can choose among
all 10 men, then the second woman can only choose among 9 men
(no matter which man was chosen by the first woman), the next
one among 8 etc., the last woman has the choice of only one man.
Therefore there are 10 · 9 · · · · · 1 = 10! = 3, 628, 800 possibilities.
Alternatively you could line up the women. Then each pairing
corresponds to one way of lining up the men next to the women. So
you need to count the orderings of the 10 men. !

5.2 Counting using bijections

There is another important counting rule. To introduce it we need the

concept of bijection. A bijection between two sets X and Y is a rule
that associates with each element of X an element of Y, such that each
element of Y appears precisely once.5 Figure 5.2 shows an example.
The following statement is so obvious that it may surprise you that
it is useful at all.

5 See Appendix B for more information on bijections.

118 5 Counting

a 1

b 2

c 3

d 4

X Y

Figure 5.2 A bijection X → Y where X = { a, b, c, d}, Y = {1, 2, 3, 4}

Counting by bijection

If there is a bijection X → Y then | X | = |Y |.

This can be used in two ways:

J For counting: If you want to count X then you can try to find a set
Y whose cardinality you know, and a bijection X → Y.
Even if you already know a formula for | X | such a bijection can
lead to a better understanding of the formula.

J For deriving or proving equations: If you have two sets X, Y

and expressions for their cardinalities, and if you have a bijection
X → Y then the two expressions must give the same result.

Here are some examples for illustration.

Variations on the number of subsets

We know from Chapter 2.2 that the number of subsets of {1, . . . , n}

is 2n . Recall that we denote the set of subsets of {1, . . . , n} by
P ({1, . . . , n}), so
|P ({1, . . . , n})| = 2n .
We conjectured this from a few examples and then proved it using a
recurrence relation. Is there a better way to understand the formula?
5.2 Counting using bijections 119

The number 2n arises among the basic types of counting problems

as the number of n-tuples ( a1 , . . . , an ), where each ai takes one of
two possible values. Can we relate these n-tuples to the subsets of
{1, . . . , n}? Can we find a bijection?
Let us represent the subsets as follows, for example if n = 2:

subset 1 2
∅ − −
{1} + −
{2} − +
{1, 2} + +
For each subset there is one row. In the column marked 1 we write a
+ whenever the subset contains the element 1, otherwise we write a
−, and similarly for the column marked 2. If we read the table row
by row then we see that each subset corresponds to a pair of elements
of A = {−, +}. For subsets of {1, 2, 3} we would get three columns,
hence triples. In general we obtain:
Second proof of the equation |P ({1, . . . , n})| = 2n : The subsets of
{1, . . . , n} correspond to the n-tuples ( a1 , . . . , an ) with ai ∈ {−, +}
for all i, as follows. Each subset S ⊂ {1, . . . , n} is associated with the
n-tuple ( a1 , . . . , an ) defined by

+ if i ∈ S
ai =
− else.

Clearly each n-tuple arises from a unique subset. Therefore the

number of subsets equals the number of n-tuples of pluses and
minuses, hence is 2n . q. e. d.
What did we do? In order to count the subsets we associated them
with objects that we knew how to count. We constructed a bijection

P ({1, . . . , n}) → {( a1 , . . . , an ) : ai ∈ {−, +} for all i }

(the map was defined above).

By interpreting plus and minus signs as decisions we can reformu-
late this as follows:
Third proof of the equation |P ({1, . . . , n})| = 2n : In order to specify
a subset S of {1, . . . , n} we first decide whether 1 is in S or not; then
120 5 Counting

we decide whether 2 is in S or not; and so on up to n. These are n

independent decisions, and each decision has two possible outcomes.
The product rule then implies that there are 2n subsets. q. e. d.
This gives us a clear understanding why the number of subsets is 2n .
The recursive proof given in Problem 2.1 did not have this virtue.
In the following problem we use a bijection for proving a formula,
rather than for counting.

? Problem 5.6
Prove the identity    
n n
=
k n−k
(for 0 ≤ k ≤ n) using a bijection.

! Solution
The left-hand side counts the k-element subsets of {1, . . . , n}, the
right-hand side counts the (n − k )-element subsets. How can we
relate these to each other? Can we find a bijection?

Think about it!

By associating a k-element subset S ⊂ {1, . . . , n} with its complement

{1, . . . , n} \ S we obtain a bijection from the set of k-element subsets
to the set of (n − k)-element subsets. Therefore these two sets have
the same cardinality.
More formally: Let

X = {S ⊂ {1, . . . , n} : |S| = k }, Y = {S ⊂ {1, . . . , n} : |S| = n − k } .

Then F : X → Y, S → {1, . . . , n} \ S is a bijection, so | X | = |Y | and

hence (nk) = (n−
n
k ). !
This means that Pascal’s triangle, see Exercise E 3.5, is symmetric.
5.2 Counting using bijections 121

? Problem 5.7
For n ≥ 1 let en and on be the numbers of subsets of {1, . . . , n} which have
an even or odd number of elements, respectively. Find formulas for en and
on .

 Investigation
G
 Let us look at some examples. We leave out the braces (or “curly
brackets”) for simplicity. Let us call a subset even/odd if it has an
even/odd number of elements.
n even subsets odd subsets en on
1 ∅ 1 1 1
2 ∅, 12 1, 2 2 2
3 ∅, 12, 13, 23 1, 2, 3, 123 4 4

 We observe that en = on for n = 1, 2, 3. Also, these numbers are

powers of 2.
 The second of these statements follows from the first: Since every
subset must be even or odd and there are a total of 2n subsets we
have en + on = 2n . So if en = on then it follows that en = 12 2n =
2n −1 .
 Conjecture: We have en = on for all n ≥ 1. How could we prove
this?
 An equality of two numbers suggests that a bijection should exist
between the two sets being counted. Denote by En , On the set of
even/odd subsets of {1, . . . , n}. We want to find a bijection

F : En → On .

What does this mean? We are looking for a rule that associates
an odd subset with every even subset. The rule must be such that
every odd subset appears precisely once as a result.

Think about it!

122 5 Counting

 How can you produce one subset from another subset? You know
one way to do this from Problem 5.6: taking the complement.
That is, associate S ⊂ {1, . . . , n} with its complement {1, . . . , n} \ S.
Does this rule produce an odd subset from an even one? If n is
odd then this is the case, as you can easily check. But if n is even
then the complement of an even subset is even again.
So for odd n we have solved our problem.
But what about even n? We need a new idea.

Think about it!

 We want to associate odd subsets with even subsets. A similar

problem would be to associate odd numbers to even numbers.
That’s easy: add or subtract 1.
 Can we do anything similar for sets? We could transform an even
subset into an odd one by adding or removing an element. How
do we turn this into a general rule? When do we add, when do
we remove an element? And which element?
Here is a method to do this: If the subset contains the element 1
then take it away, otherwise add it in. That is, our map is defined
by 
S \ {1}, if 1 ∈ S
F (S) = (5.2)
S ∪ {1}, if 1 ∈ S .
Clearly, if S is even then F (S) is odd. We need to check that every
odd subset T appears as F (S) for precisely one S. For this we
simply invert the operation F: Let T be an odd subset of {1, . . . , n}.
If 1 ∈ T then we have T = F (S) for S = T \ {1};
if 1 ∈ T then we have T = F (S) for S = T ∪ {1}.
In both cases this is clearly the only possible choice for S.
Therefore we have proved that F : En → On is bijective.
 Inverting the operation F in the last step is nothing else but finding
the inverse map for F. By coincidence the inverse map is defined
by the same formula as F. That is, the rule for finding S from T is
the same as the rule for finding T = F (S) from S.
5.3 Counting in two ways 123

 This argument works for all n, even and odd. 

G
We write up the argument efficiently.

! Solution of Problem 5.7

For any subset S ⊂ {1, . . . , n} define the subset F (S) ⊂ {1, . . . , n}
by formula (5.2). We have F ( F (S)) = S for all S, for if 1 ∈ S then
1 ∈ F (S), so F ( F (S)) = F (S) ∪ {1} = (S \ {1}) ∪ {1} = S, and
similarly if 1 ∈ S then we get F ( F (S)) = (S ∪ {1}) \ {1} = S.
Therefore F : P ({1, . . . , n}) → P ({1, . . . , n}) is its own inverse
map. A map having an inverse is bijective. Also, F maps En to On
and On to En , therefore it defines a bijection En → On . Therefore we
get en = on , and using en + on = 2n we get en = on = 2n−1 . !

 Review of Problem 5.7

We looked for a bijection in order to prove en = on . The first attempt
(taking complements) only worked for odd n. Then we found another
map (rule) that worked for all n.
Lesson: Finding a bijection requires creativity. If the first attempt
fails there may still be other ways to find one. 

5.3 Counting in two ways

Usually we count in order to determine some number of possibilities.

Now we want to use counting for other purposes. The starting point
is that most problems have many solutions:6

Counting in two ways

By counting the same set in two different ways we can often

derive interesting formulas.

You have already encountered two examples of this in Chapter 4: the

edge-face formula (4.2) and the edge-vertex formula (4.4) for graphs.7

6 Counting in two ways is sometimes called double counting, but this expression
is also used for ‘overcounting by a factor of two’ as in the multiple-counting principle.
7 Problem 5.6 showed another way of deriving a formula by counting.
124 5 Counting

? Problem 5.8
Connect n points pairwise by lines. Count the number of lines in two ways
and derive an identity.

! Solution
J First count: Number the points 1, . . . , n.
– First connect 1 and 2, 1 and 3, . . . , 1 and n, that’s n − 1 lines,
– then connect 2 and 3, 2 and 4, . . . , 2 and n (note that 2 and 1
are already joined), that’s n − 2 lines,
– then connect 3 to the points 4, . . . , n, that’s n − 3 lines,
– and so on, and finally connect n − 1 and n.
Therefore the total number of lines is

( n − 1) + ( n − 2) + · · · + 1 .

J Second count: Since we want to draw a line between any two points,
the lines correspond to the 2-element subsets of the set of points.
There are (n2 ) such subsets. (Compare Problems 5.4 and 5.1.4.)
J Therefore we get
 
n
( n − 1) + ( n − 2) + · · · + 1 = . !
2

If we reverse the order on the left and replace n by n + 1 we obtain

 
n+1 ( n + 1) n
1+2+···+n = = .
2 2
This is a new derivation of the formula for the sum of the first n
natural numbers. In Section 1.3 we derived this formula using the
Gauss trick.
A similar idea can be used to find a formula for the sum of squares:

? Problem 5.9
Count the number of triples ( a, b, c) which satisfy

a, b, c ∈ {1, . . . , n} and a < c, b < c

5.3 Counting in two ways 125

in two ways and derive an identity.

 ! Investigation and solution

G
 First count: c seems to play a special role since it appears in both
conditions a < c, b < c. Therefore it may be useful to sort the
triplets by the value of c. How many triplets are there for a fixed
value of c?

Think about it!

It may help to make a table:

c triples number
1 none 0
2 (1, 1, 2) 1
3 (1, 1, 3), (1, 2, 3), (2, 1, 3), (2, 2, 3) 4

Conjecture?
 How many ways of choosing a, b are there for any fixed c? The
conditions are a < c, b < c, so a, b ∈ {1, . . . , c − 1}. Since a, b can
be chosen independently, there are (c − 1)2 possibilities.
 Since any choice of c ∈ {1, . . . , n} is possible, by the addition rule
we get that there are

02 + 12 + 22 + · · · + (n − 1)2

triplets.
 Second count: How else could we count the triplets? Instead of
sorting them by the value of c, we could first fix the numbers that
appear in the triplet and then count how many triplets can be
formed from these numbers. For example, from the numbers 1, 3, 4
we can form the triplets (1, 3, 4) and (3, 1, 4), but from the numbers
2, 5 we can only form the triplet (2, 2, 5).
 The examples suggest that we should distinguish whether two or
three different numbers appear in a triplet (it is not possible that
126 5 Counting

only one number appears since c is always larger than a and b).
Therefore we divide the triples into two classes and do a case by
case analysis:
1. The triplets having a = b. Each such triplet has two different
numbers a and c, so it determines a 2-element subset S of
{1, . . . , n}. Conversely any 2-element subset S determines a
unique triplet: a = b must be the smaller element of S and c
must be the larger element. Therefore there are (n2 ) such triplets.
2. The triplets having a = b. Each such triplet determines a 3-
element subset S of {1, . . . , n}. Conversely, we saw in the ex-
ample that a 3-element subset S determines two triplets. This
works in general: Write S = { x, y, z} with x < y < z, then S
corresponds to the triplets ( x, y, z) and (y, x, z). Since there are
(n3 ) 3-element subsets S, we get 2(n3 ) such triplets.
By the addition rule we get that the total number of triplets is
   
n n
+2 .
2 3

 We obtain the identity

   
n n
1 + 2 + · · · + ( n − 1) =
2 2 2
+2 . !
G
2 3

If we replace n by n + 1 and calculate a little we get

   
n+1 n+1 n(n + 1)(2n + 1)
1 +2 +···+n =
2 2 2
+2 = .
2 3 6

Remark

We have found a closed formula for ∑nk=1 k2 , a highly non-trivial

result (where the Gauß trick would not work)!
Imagine you had been given the problem “Find a closed formula
for the sum ∑nk=1 k2 ”. Would you have got the idea of using the
counting problem above to find the answer? No? Don’t worry.
With practice you will build up a repertoire of ideas, which then
helps you to get new ones.
5.3 Counting in two ways 127

? Problem 5.10
Prove the formula
n  
n
2 = ∑
n

k =0
k

(n ∈ N) by counting a set in two ways!

! Solution
What is counted by 2n ? The subsets of An = {1, . . . , n}, see Problem
2.1 and Section 5.2. And (nk) counts the subsets with k elements. So
both the left and the right side count the number of subsets of An ,
but on the right we sort them by their number of elements.
More formally: Let An = {1, . . . , n} and X = P ( An ) the power set
of An . For k ∈ {0, 1, . . . , n} let Xk be the set of k-element subsets of
An : Xk = {S ⊂ An : |S| = k }. Then

X = X0 ∪ X1 ∪ · · · ∪ X n
n
and the Xk are pairwise disjoint. By the addition rule, | X | = ∑ | Xk |.
k =0
The formula now follows from | X | = 2n and | Xk | = (nk). !

You can also interpret the formula in terms of Pascal’s triangle,

which was introduced in Exercise E 3.5: The sum of entries in the nth
row is 2n . This can also be proved by induction using the formula in
Exercise E 5.14 (do it!), but the counting proof gives more insight.

Here is a proof by counting in two ways which you may know from
elementary school:

? Problem 5.11
Prove by counting in two ways that

n·m = m·n

for all natural numbers n, m.

128 5 Counting

! Solution
Consider a rectangular array of dots, which is n dots wide and m dots
high. If you count the dots column by column then you get n · m dots
(each column has m dots, there are n columns); if you count them
row by row then you get m · n dots. !
This may look trivial to you – for children who have just learned to
multiply it is not.

5.4 Going further: double sums, integrals and infinities

You can find variations on the idea of counting in two ways in all
areas of mathematics. Here are some examples. Even if you don’t
understand all the words in the explanations that follow, keep reading
anyway. You will remember the ideas when you learn more math-
ematics. Then you should also look out for more examples of this
principle.

Proofs without words

Figure 5.3 shows that (and why) the identity

1 + 3 + · · · + (2n − 1) = n2

holds, no explanations needed. There are many more pretty proofs

without words, just search the internet.

B URNSIDE’s lemma

Burnside’s lemma is a useful tool for counting structures with sym-

metry. The triominoes in Problem 5.3 are a good example. The
symmetries are the rotations of the triangle by 0, 120 and 240 degrees.
Burnside’s lemma gives a new way of counting triominoes: For each
rotation determine the number of its fixed points, i.e. the number
of marked triangles which stay unchanged under this rotation. For
the rotation by 0 degrees (i.e. no rotation) that’s all triangles (216).
For the two other rotations that’s only the triangles marked by three
equal numbers (6 each). Burnside’s lemma tells us that the number
5.4 Going further 129

1 + 3 + 5 + . . +(2n-1)
.

...

... n
.. .. ..
. . .

Figure 5.3 A proof without words

of triominoes is precisely the mean value of these three numbers, that

is 13 (216 + 6 + 6) = 76. That is the result we obtained by a different
argument.
Why does this work in general? You can find the general statement
and a proof of Burnside’s lemma in the book (Aigner, 2007), for
example. The proof uses counting in two ways. There you can also
find Pólya’s enumeration theorem, a more sophisticated version of
Burnside’s lemma, which can be used to count chemical compounds,
for example.

Double sums

A variant of counting in two ways is calculating a double sum in two

different orders: Suppose you want to find the sum of numbers aij ,
i = 1, . . . , n, j = 1, . . . , m which you can arrange in a rectangular array
(matrix) as follows (for n = 2, m = 3 say):
 
a11 a12 a13
.
a21 a22 a23
To find the sum of all the numbers you can proceed in (at least) two
ways:
J first sum over j for each fixed i, then add the sums (i.e. sum over
n m
i): ∑ ∑ aij . This means first summing each row and then adding
i =1 j =1
130 5 Counting

the row sums, in the example ( a11 + a12 + a13 ) + ( a21 + a22 + a23 );
J first sum over i for each fixed j, then add the sums (i.e. sum over j):
m n
∑ ∑ aij . This means first summing each column and then adding
j =1 i =1
the column sums, in the example ( a11 + a21 ) + ( a12 + a22 ) + ( a13 +
a23 ).
Since we are adding all numbers in both cases, the results must be
equal:
n m m n
∑ ∑ aij = ∑ ∑ aij . (5.3)
i =1 j =1 j =1 i =1

This generalizes counting in two ways. For example, if all aij = 1 then
you obtain the counting proof of n · m = m · n. Another example is
the edge-vertex formula (4.4) for graphs: Suppose a given graph has
its vertices numbered by i and its edges numbered by j. Set
⎧
⎪
⎪
⎨1 if edge j has vertex i as one endpoint
aij = 2 if edge j has vertex i as both endpoints
⎪
⎪
⎩0 otherwise.

Then the sum of row i is the degree of vertex i and the sum of column
j is 2 for each j. So (5.3) is just the edge-vertex formula (here n = v,
m = e).
Changing the order of summation as in equation (5.3) also gives
interesting results for infinite sums (so-called series). A famous
example of a series is ∑∞ n=1 n2 = 1 + /4 + /9 + . . . . There is no
1 1 1

simple way to compute the value of this series. A beautiful argument

that is based on interchanging summations is the Herglotz trick, see
for example (Aigner and Ziegler, 2014). By the way, the value of the
series is π2/6.

Double integrals

If you are familiar with integration then you know that integrals may
be considered as the continuous analogue of sums.
This leads to further generalizations of the idea of counting in two
ways: If you have a function of two variables, f ( x, y), then you can
integrate it first in x and then in y, or first in y and then in x – the
fact that the results are the same is known as Fubini’s theorem. If
5.4 Going further 131

you have several (possibly infinitely many) functions of one variable,

f 1 (x ), f 2 ( x ), . . . ,then you can interchange summation and integration:
∑i f i ( x ) dx = (∑i f i ( x )) dx.8
A different, rather spectacular example of counting intwo ways
∞
with integrals occurs in the computation of the integral I = −∞ e− x dx,
2

the area under the bell-shaped curve e− x that appears in probability

2

theory as Gauß’s normal distribution. Because you cannot write

down an antiderivative for the function e− x it is not at all obvious
2

how one should go about computing

∞ theintegral. Here is an inge-
∞
nious trick: You write I = −∞ e− x dx = −∞ e−y dy (renaming the
2 2

integration variable does not change the value of the integral!) and
multiply:
 ∞  ∞  ∞  ∞
e− x dx e−y dy = e−( x
2 2 2 + y2 )
I2 = dx dy
−∞ −∞ −∞ −∞
This is a double integral, which computes the volume of a certain
body B. We use coordinates x, y, z in space. The body B is infinitely
extended in the horizontal x,y-directions; its base is the z = 0 plane
and its top is the graph of the function z = e−( x +y ) . The double
2 2

integral tells us to compute the volume of B ‘slice by slice’: If you fix

y then the points of B having this y-value form a planar ∞ −( slice of B.
The area of this slice is precisely the inner integral −∞ e x +y ) dx.
2 2

Then the volume of B is obtained by integrating these areas over y.

This was the first ‘count’, see the left-hand diagram in Figure 5.4.
Now comes the second ‘count’, see the right-hand diagram in
Figure 5.4: Instead of slicing B into planar slices we slice it into
cylindrical slices. For r > 0 let Cr = {( x, y) : x2 + y2 = r2 } be
the circle of radius r around the origin in the x, y-plane. On Cr the
function e−( x +y ) has the constant value e−r . So the part of B having
2 2 2

x2 + y2 = r2 is a cylindrical surface of radius r and height e−r . The

2

area of this surface is 2πr · e−r . In order to find the volume of B we

2

integrate all these cylinder areas and get

 ∞  ∞ ∞
−r 2
2πre dr = π e−s ds = −πe−s s=0 = π
0 0

(substitute = s). Since the two ways of computing the volume

r2
(planar or cylindrical slices) must give the same result, we get I 2 = π
8 These theorems are only true under certain conditions on f resp. the f i . See
(Pugh, 2010, Section 4.1) or (Apostol, 1967, Vol. I, Section 11.4) for details.
132 5 Counting

Figure 5.4 Two ways to compute the volume under the graph z = e−( x
2 + y2 )

√
and hence I = π. So we have derived the surprising result9
 ∞ √
e− x dx =
2
π.
−∞

As for the series ∑∞n=1 n2 the appearance of π is quite unexpected

1

since there is no sign of anything geometric or ‘circle-like’ in the

integral. Our derivation explains where the π comes from.

Bijections, parties and infinities

The idea of bijections is also ubiquitous in mathematics, for example

when considering the question: How big is infinity?
Imagine you throw a big party, and you want to know whether
there are as many men as women among your guests. There are (at
least) two ways to find out:

1. You count the women and you count the men and compare the
numbers.

2. You ask your guests to pair up (always a man and a woman). If no

one is left then there are as many women as men.

We used implicitly that for any r > s > 0

9 We omitted one detail in the argument:

the circles Cr and Cs have constant distance r − s from each other. If this was not
the case we would get an additional factor in the second ‘count’, stemming from the
change of variables formula for integrals. For details see (Pugh, 2010, Section 5.7) or
(Apostol, 1967, Vol. 2, Section 11.26).
5.5 Toolbox 133

The second method allows you to ascertain the equality of the num-
bers without actually knowing the numbers! All you did is establish
a bijection between the set of women and the set of men.
The same idea is used for classifying infinities (cardinalities). Two
sets are said to be of the same cardinality (or equinumerous) if there
is a bijection between them. This notion makes just as much sense
for infinite as for finite sets since we don’t actually need to count the
‘number of elements’. However, with infinite sets some funny things
happen. For example, the set of natural numbers N has the same
cardinality as the set of even natural numbers, since mapping n to
2n for all n ∈ N gives a bijection. Even more surprisingly, N has the
same cardinality as the set of rational numbers (fractions), but not as
the set of real numbers.10
In short: There are as many natural numbers as there are even
natural numbers or fractions, but less than there are real numbers.

5.5 Toolbox

The fundamental counting rules (addition rule, multiplication rule,

multiple-counting principle) enable us to solve the basic types of
counting problems (counting tuples and subsets). You should know
the solutions of these problems (the formulas and also their deriva-
tions) very well because many counting problems can be reduced to
one of them.
Bijections can give a deeper, immediate understanding of counting
formulas, for example for the formula |P ({1, . . . , n})| = 2n .
Using the ideas of bijections and counting in two ways one can
derive many interesting identities.

Exercises

E 5.1 .What is the number of triples ( a1 , a2 , a3 ) of numbers a1 , a2 , a3 ∈ 1

{1, 2, . . . , 10} satisfying a1 < a2 < a3 ?

E 5.2 . You throw two dice, one black and one white. How many 1-2
possible outcomes are there? An example of an outcome is ‘black 1,

10 For proofs and more explanations see (Pugh, 2010, Section 1.4), for example.
 134 5 Counting

white 3’. How many outcomes show two different numbers? How
many outcomes are there if the dice are indistinguishable?

1-2 E 5.3 .In the game of dominoes each tile has two fields, each showing
0, 1, 2, 3, 4, 5 or 6 spots. How many domino tiles are there? Give at
least two derivations of your answer.

2 E 5.4 .The game of poker is played with a deck of 52 playing cards,

showing 13 ranks (Ace, 2, 3, . . . , 10, Jack, Queen, King) of 4 suits. A
poker hand consists of 5 cards. The value of a hand is determined by
its category. The categories are:
• One pair/Three of a kind/Four of a kind: 2/3/4 cards have
equal ranks, but apart from this no rank appears multiple times
• Two pair: 2 cards of one rank, 2 cards of another rank, and one
card of a third rank
• Full House: 3 cards of one rank and 2 cards of another rank
• Straight: successive ranks
• Flush: all cards have the same suit
• Straight Flush: successive ranks and equal suit
How many poker hands are there? How many of each category are
there?

2 E 5.5 . How many ways are there to go home from campus, see
Figure 5.5? Only direct routes are counted, i.e. those that always go
east or north. How many ways are there if you want to buy bread on
your way home?
n −1
2 E 5.6 . Find a formula for the sum of cubes ∑ k3 by counting the
k =1
elements of the set
 
( a, b, c, d) : a, b, c, d ∈ {1, . . . , n}, a < d, b < d, c < d

in two ways.

2 E 5.7 . Let P1 , . . . , Pn be points on a circle. Assume that no three

chords (i.e. lines Pj Pk ) intersect in a point. Find a bijection between
the set of 4-element subsets of { P1 , . . . , Pn } and the set of intersection
points of chords.
Exercises 135

home

campus bakery

Figure 5.5 Paths in a street grid

How many intersection points are there?

E 5.8 .26 people named A, B, C, . . . , Z write, one after another, their 2

birthday in a calendar of the year 2016, a leap year. Each person
writes her name next to the date of her birthday. How many possible
calendars are there? In how many of them does everyone have a
different birthday? In how many do at least two people have the same
birthday? In how many does at least one person have her birthday
on January 1? In how many exactly one? In how many exactly two?
In how many does exactly one person have her birthday on January 1
and exactly one on March 28?
Also answer these questions in the case where the people write a tick
mark next to the date of their birthday, instead of their name.

E 5.9 . Let f 1 , f 2 , f 3 , . . . be the Fibonacci numbers: f 1 = 1, f 2 = 2 2

and f n = f n−1 + f n−2 for all n ≥ 3. Prove the formula f 2n = f n2 + f n2−1
for all n ≥ 2. Give two proofs: one using induction and one by
counting in two ways. Here you may use that f n is the number of
trains of length n with coaches of length 1 or 2, see Exercise E 2.8.
Find more formulas for the Fibonacci numbers, e.g. a similar one
involving f 2n+1 .

E 5.10 .Prove the equation 4 · (50

4 ) = 50 · ( 3 ) by counting in two ways:
49
2
how many ways can you choose a group of 4, with a leader, from 50
people? Generalize this identity to any number of people and group
 136 5 Counting

sizes. Also give a proof by direct calculation.

2 E 5.11 . Count in two ways to prove the formula ∑nk=0 k(nk) = n2n−1
for n ∈ N.

2 E 5.12 .Give a new proof of the Fibonacci recursion for the numbers
on in Exercise E 2.15 using a bijection, by distinguishing representa-
tions according to whether the last summand is equal to 1 or not.

2-3 E 5.13 .Consider the pairs ( a, b) with a, b ∈ {0, 1, . . . , n}. Let en be the
number of such pairs for which a + b is even and on the number of
pairs for which a + b is odd. What do you observe for n = 0, 1, 2, 3?
State a conjecture and give proofs via bijection, direct calculation and
induction. Can you generalise to more than two numbers?

2 E 5.14 .The binomial coefficients satisfy the recursion

     
n+1 n n
= + .
k k−1 k

Give two proofs: one by direct calculation, the other by counting the
k-element subsets of {1, . . . , n + 1} in two ways. Conclude that (nk) is
the kth entry in the nth row of Pascal’s triangle, see Exercise E 3.5.
Here you always start counting at zero.

2 E 5.15 .The binomial coefficients appear in many places in mathemat-

ics. They get their name from the general binomial theorem:
     
n n −1 n n −2 2 n
( a + b) = a +
n n
a b+ a b +···+ abn−1 + bn
1 2 n−1

for real numbers a, b and n ∈ N0 . Check the formula for n = 2 and

n = 3 by multiplying out the left-hand side. Give two proofs for
the general formula: one by induction using the formula in Exercise
E 5.14, the other by using the meaning of (nk) as number of subsets.

2 E 5.16 .
a) Check by direct calculation that the formula (2.9) for the nth
Fibonacci number always yields a rational number. That is,
check that when multiplying out the powers using the general
binomial theorem (see Exercise E 5.15) all square roots disappear.
Exercises 137

b) We know from their recurrence relation that the Fibonacci num-

bers are integers. This implies something about the divisibility
of certain combinations of binomial coefficients. What can we
conclude?
√
E 5.17 . Let a = 1 + 2. Calculate the decimal representations of 2-3
a, a2 , a3 , a4 , . . . with your calculator. What do you observe? Find a
reason. What is the digit in the 50th decimal place of a200 ?

E 5.18 .Find at least one more proof for the equality of the numbers 2-3
of even and odd subsets of {1, . . . , n}.

E 5.19 .Let n ∈ N. Find a bijection proof for the fact that the number 3
of ways to write n as an ordered sum of natural numbers is 2n−1 (see
Exercise E 1.9).

E 5.20 .Let n ∈ N. How many solutions does the equation a + b + c = 3

n have with a, b, c ∈ N?

E 5.21 .Use the following idea to give a new derivation of the formula 2-3
for the number of regions which result from n lines in the plane in
general position (Problem 1.3):
First turn the plane so that no line is horizontal. Then associate
with every region its lowest point, if it has a lowest point. Do all
intersection points of lines arise as lowest points? How many regions
do not have a lowest point?

E 5.22 .Generalise the idea of Exercise E 5.21 to find the number of 3

regions into which space is divided by n planes in general position
(no two parallel, no three through a line, no four through a point).

E 5.23 .Generalise the idea of Problem 5.8 to the counting of triangles 2-3
that can be formed from the n points, and derive an identity.

E 5.24 .Find the number of regions into which the interior of a circle 3-4
is divided if you choose n points on its circumference and draw all
chords between them. We assume the points are chosen so that no
three of the chords pass through a common point.
Find the number for n = 2, 3, 4, 5 and state a conjecture. Then find
the number for n = 6. Find a general formula.
6 General problem solving strategies: Similar
problems, working forward and backward,
interim goals

General problem-solving strategies are strategies which can be used

in everyday life, not just in mathematics: If I want to solve a problem
then it will help me to recall how I solved a similar problem. If I want
to reach a goal then I can think about which steps I should do first
in order to get there (working forward); or I can think about what
could be the last step, reaching the goal (working backward), and
what interim goals I could set for myself.
We now take a systematic look at these problem-solving strategies.
We also represent them in schematic diagrams, which can help us to
keep track of our progress. Then we investigate two problems, one
from geometry and one about representing integers as sums, and pay
particular attention to the strategies we use to solve them.

6.1 General problem solving strategies

The first step in problem-solving is to understand the problem. We

ask ourselves:
J What is given?
J What are we looking for?
In most mathematical problems either you want to find something,
or you want to give a proof. For example, in the log-cutting problem
1.1 and in most counting problems you want to find a number, and
in Problem 4.1 on Euler’s formula you are asked to give a proof.
In a finding problem you are given data (for example, the length
of the log) and you are looking for certain unknowns (for example the
time to cut it into pieces).
In a proving problem you are given conditions (for example, the
graph G is plane and connected) and you are looking for the proof of
a conclusion (for example, Euler’s formula holds for G).

© Springer International Publishing AG, part of Springer Nature 2018 139

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7_6
140 6 General problem solving strategies

Not every mathematical problem can be classified as a finding

problem or a proving problem. In a finding problem you will also
need to prove that the method you used is correct. However, often
this proof is already implicit in the derivation of the solution. In
addition, there are many mathematical problems where you just want
to understand some situation, and where it is not clear at the outset
what you want to find or prove. You will find an example of this in
Chapter 9.3.
In order to solve the problem we need to find a connection from
the data to the unknown, from the conditions to the conclusion.

data, unknown,
conditions ? conclusion
given goal

Figure 6.1 The problem: Looking for a connection

The most important general strategies

J Have I seen a similar problem before?

This is one of the first questions that you should ask yourself.

J Working forward: How can I use what is given?

Working backward: How can I reach the goal?
You can approach a problem from the front or from the back.
Approaching it from the front means asking how you can use
the data/the conditions. Approaching it from the back means
focusing on the goal and asking how you could get there. See
Figures 6.2 and 6.3.

J Can I formulate useful interim goals?

An interim goal can be anything that helps to bridge the gap
between the given and the goal. See Figure 6.4.

You already know some other general strategies: looking at special

cases and examples, making a sketch or a table, simplifying, intro-
ducing suitable notation, formulating a conjecture. Typically you
use a combination of these (and other) strategies. If one strategy
6.1 General problem solving strategies 141

How can I
given goal
use the given?

Figure 6.2 Working forward

How can I
given goal
reach the goal?

Figure 6.3 Working backward

given interim goal goal

Figure 6.4 Solution scheme with an interim goal

turns out not to help then try another. For example, you first try
working forward, and if you don’t see what to do, you try working
backward. And when this has given you more understanding, you
try working forward again. Sometimes you also have more complex
solution schemes with several interim goals. For example, we could
first formulate one interim goal and then, considering how we could
reach that interim goal, formulate another interim goal.
Let us look back at some of our solutions from previous chapters
from this perspective.

J In the log cutting problem 1.1 we set the interim goal of finding
the number of pieces. This number was easy to find from the data
(the length of the log), and from there it was easy to determine the
time.

J We worked forward when solving the Fibonacci recursion in

Chapter 2.4, and also in most counting problems in Chapter 5.

J In counting problems the search for a recurrence relation can be

a useful interim goal. Once we set ourselves this interim goal we
can look for ways to derive a recurrence relation; this is working
backward (from the interim goal to the data). See the problems in
Chapter 2.

J In Problem 1.2 (zeroes of 100!) we worked backward: we looked

142 6 General problem solving strategies

looking for a recurrence solving the

problem goal
rec. relation relation rec. relation
working backward working forward

Figure 6.5 Solution scheme for solving a counting problem using a recur-
rence relation

for the mechanism that produces the zeroes (the goal) at the end
of n!.

J The indirect proof in Problem 4.6 (five points with all connections)
was a kind of working backward: Suppose the conclusion (the
goal) is wrong, what follows from this?

A simple example of working backward

? Problem 6.1
A merchant has a crate of apples. A customer comes and buys half of the
apples and one extra. Then another customer comes and buys half of the
remaining apples and one extra. Then a third and a fourth and a fifth
customer come and do the same. At the end there is one apple left. How
many apples did the merchant have at the beginning?

! Solution
The simplest way to solve this is to start at the end: At the end the
merchant has one apple, so before giving the extra apple away he had
2, and before giving away the last half he had 4. So he had 4 apples
before the fifth customer came. Similarly he had 5 apples before
the fourth customer got the extra apple and before that 10. So the
merchant had 10 apples before the fourth customer came. Similarly,
22 = (10 + 1) · 2 apples before the third, 46 = (22 + 1) · 2 before the
second and 94 = (46 + 1) · 2 before the first customer. Therefore the
merchant had 94 apples at the beginning. !
This is a classic example of working backward in a finding problem.
Note that doing “first halve, then subtract one” backward means “first
6.2 The diagonal of a cuboid 143

add one, then double”. When turning things around, the order also
gets turned around. See also Exercise E 6.12.

6.2 The diagonal of a cuboid

We now investigate a geometric finding problem.

? Problem 6.2
Find the length d of the diagonal of a cuboid with side lengths a, b, c.

c
d
C
b
A B
a

Figure 6.6 Diagonal of a cuboid

 ! Investigation and solution

G
 We first make a sketch, see Figure 6.6.
 What is the data? The side lengths a, b, c. What is the unknown?
The length d of the diagonal AD.
 Special cases and examples seem to be useless here.
 Have I seen a similar problem before? Can I simplify the prob-
lem?
Solid geometry is harder than plane geometry. Maybe we can solve
a similar problem in plane geometry first? Here is such a problem:
Find the length x of the diagonal of a rectangle with side lengths
a, b. See Figure 6.7.
You know this from school: The triangle ABC is right-angled,
√ so we
have x2 = a2 + b2 by Pythagoras’ theorem, hence x = a2 + b2 .
144 6 General problem solving strategies

x
b

A B
a

Figure 6.7 Diagonal of a rectangle

 How can I relate the given problem to the similar problem?

Think about it!

 Let us draw the diagonal of the ‘bottom rectangle’ of our cuboid,

see Figure 6.8. We know the length of this diagonal, x. Can we
relate this bottom diagonal x to the space diagonal d?

c
d
C
x c
b
A B
a

Figure 6.8 Sketch for the solution

 Observation: The bottom diagonal and the space diagonal are two
sides of the triangle ACD. Let us take a close look at this triangle.
Its third side length is c. Do we know anything else? The angle
at C must be a right angle since the side CD is orthogonal to the
6.3 The problem of trapezoidal numbers 145

bottom plane, hence to any line contained in the bottom plane.

 Therefore we can use Pythagoras’ theorem again, now for the
triangle ACD. We get d2 = x2 + c2 . Using x2 = a2 + b2 we get
d2 = a2 + b2 + c2 , hence

d = a2 + b2 + c2 .

This is the solution! !

G

 Review of Problem 6.2

We were given a, b, c, the unknown was d. The situation is shown
schematically in the left part of Figure 6.9. The question mark sym-
bolizes the connection that we are looking for. The key to the solution
was to look at the bottom diagonal. It acts as a link between the given
and the unknown. The right part of Figure 6.9 shows the solution
scheme.1 

a a
x
b ? d b d
c c

given unknown given link unknown

Figure 6.9 The problem and the solution scheme

6.3 The problem of trapezoidal numbers

Let us try a more difficult problem!

? Problem 6.3
Which natural numbers n can be written as the sum of several consecutive
natural numbers?
1 George Pólya discusses this example at length in his classic ‘How to solve it

– a new aspect of mathematical method’, (Pólya, 2014), in a hypothetical teacher -

student interaction. Delightful!
146 6 General problem solving strategies

For brevity we call such a representation of n a representation as a

trapezoidal number, or a trapezoidal representation. Look at Figure
6.10 to understand why.

Figure 6.10 The trapezoidal numbers 12 = 3 + 4 + 5 and 11 = 5 + 6

 Investigation
G
 What is given?
Only the definition of a trapezoidal number.
What are we looking for?
1. The answer to the question: Which numbers n have trapezoidal
representations?
2. A proof that our answer is correct. For the numbers that we
claim to be trapezoidal the simplest proof would be to write
down a trapezoidal representation. For the others we need to
find a proof that they are not representable in this way.
So we have both a finding problem and a proving problem.
 To get a feel for the problem look at some small n.

Think about it!

You get Table 6.1. The numbers 1, 2, 4, 8 are not trapezoidal. This
may give you an idea:
Vague conjecture: All numbers except the powers of 2 are
trapezoidal.
The conjecture is quite vague because we have only a little data,
and because there does not seem to be a good reason why powers
of 2 should play a role for trapezoidal numbers. If you extend the
table then the conjecture holds up, but you cannot go on forever.
We need a general method.
The table also shows that a number can have several trapezoidal
6.3 The problem of trapezoidal numbers 147

n trapezoidal representations
1 none
2 none
3 1+2
4 none
5 2+3
6 1+2+3
7 3+4
8 none
9 4 + 5 or 2 + 3 + 4

Table 6.1 Trapezoidal representations

representations. But this is not part of the question. See Problem

E 6.5, however.
 How can we approach the conjecture? Looking at the table, you
might recognize a pattern for odd n.

Think about it!

 If n > 1 is odd then it can be written as the sum of two consecutive

integers: Write n as n = 2m + 1 where m ∈ N, then n = m + (m +
1) is the representation we needed.
So odd numbers bigger than one are trapezoidal.
 To prove the general (vague) conjecture we need to do two things:2
1. Find a trapezoidal representation for each n which is not a
power of 2.
2. Prove that powers of 2 do not have such a representation.
We start by looking at the first task. Sometimes it helps to refor-
mulate: What does it mean that n is not a power of 2?
It means that n has an odd divisor (other than 1). For example, 12
2 However, we must always remember that it is only a conjecture. It might

happen that during the investigation we find out that it is wrong. Keep your eyes
open.
148 6 General problem solving strategies

is divisible by 3 and therefore cannot be a power of 2. And any

power of 2 cannot be divisible by any odd number other than 1.3
So apparently odd divisors play a role.
 First attempt: Let us try to understand in an example why odd
divisors could be useful: 50 = 5 · 10. Idea: We could write this
as 50 = 10 + 10 + 10 + 10 + 10, then reduce the first two tens by
2 and 1 and enlarge the last two tens by 1 and 2 – this will not
change the sum, and we get 50 = 8 + 9 + 10 + 11 + 12, a trapezoidal
representation of 50.
 This looks promising! You can see that this works for any odd
number of summands, but not for an even number: The same
method would transform 20 = 10 + 10 into 20 = 9 + 11, but here 9
and 11 are not consecutive integers. So odd numbers play a role
here, we are on the right track. Can we formulate this procedure
in general?
We let the example guide us. We write n = xy with odd x > 1.
In the example we had x = 5, y = 10. We write n = y + · · · + y
where there are x summands. Since x is odd there is a central
summand, and the other x − 1 summands are on the left and on
the right, so that’s x−2 1 summands each. Going to the left from the
center we reduce the summands by 1, 2, etc., and to the right we
increase them by 1,2 etc. So the last summand will be reduced resp.
increased by x−2 1 . We obtain
x−1 x−1
n = (y − ) + · · · + y + · · · + (y + ),
2 2
where each summand is one bigger than the preceding summand.
This is the desired trapezoidal representation of n. Is it really?

Think about it!

Look again! The summands must be not only consecutive but also
positive. This is only the case if y − x−2 1 is positive. For example, if
we start with 14 = 7 · 2 then x = 7, y = 2, so y − x−2 1 = −1 and our
3 See Chapter 8 for more on divisibility and divisors.
6.3 The problem of trapezoidal numbers 149

procedure yields 14 = (−1) + 0 + 1 + 2 + 3 + 4 + 5. This is correct,

but not what we want. So our method only yields a trapezoidal
representation if y > x−2 1 .
There is a nice little trick to turn this forbidden representation
into a permitted one. Do you see it? See Exercise E 6.4. For now
we will start again and try to approach our problem in a more
systematic way. This will also give us a way to find out how many
representations n has, see Exercise E 6.5.
 Second attempt: We focus on our goal (working backward). What
are we looking for? A representation
n = a + ( a + 1) + · · · + b where a < b
and a, b are natural numbers. It would be nice if we could simplify
(‘calculate’) the right hand side. How to do that?
 Have I seen a similar problem before? Yes, when calculating 1 +
· · · + n. Remember the pretty Gauss trick from Section 1.3? Let us
apply it here:
n = a + ... + b
n = b + ... + a
2n = ( a + b) + ... + ( a + b)
How many summands a + b are there? You can read it off from
the first line: there must be b − a + 1 summands (numbers from a
to b, counting both a and b, therefore · · · + 1). We get
2n = ( a + b)(b − a + 1) . (6.1)

 What are we looking for? For given n we want to find a, b solving

Equation (6.1). This equation looks complicated. Can we simplify
it? We could introduce some notation:
c = a + b, d = b − a +1. (6.2)
Thus 2n = cd, which looks much nicer. How can we get a, b from
c, d? We just need to solve the system of equations (6.2) for a and
b: adding the equations yields c + d = 2b + 1, subtracting yields
c − d = 2a − 1. So we get
c−d+1 c+d−1
a= , b= . (6.3)
2 2
150 6 General problem solving strategies

 What have we accomplished? We have replaced the complicated

task of solving Equation (6.1) (given n, find solutions a, b) by two
simpler tasks:
a) Find c, d satisfying 2n = cd.
b) From c, d determine a, b using Equations (6.3).
This is an example of an interim goal: to find c, d.
 Let us look at the new tasks. Part b) is easy, just plug in. Part a) is
also easy. For example, one could take c = 2, d = n or vice versa.
Are we done yet? Something must be missing since this works for
all n, including powers of 2.
Let us take a closer look. What are the conditions on a, b? In order
for n = a + ( a + 1) + · · · + b to be a trapezoidal representation, a, b
must satisfy some conditions:
– a, b ∈ N. The equations (6.3) show that then c − d + 1 and
c + d − 1 must be even, and c > d. We get:

c, d have different parity4 , and c > d. (6.4)

Conversely, it is clear that this guarantees a, b ∈ N.

– a < b. Let us rewrite this:
c−d+1 c+d−1
a < b ⇐⇒ <
2 2
c+d−1 c−d+1
⇐⇒ 0 < −
2 2
⇐⇒ 0 < d − 1 ⇐⇒ 1 < d .

We see that a < b is equivalent to

d > 1. (6.5)

 What did we accomplish? We obtain a trapezoidal representation

of n if and only if we can write 2n = cd with c, d satisfying the
conditions (6.4) and (6.5).
 What does this mean for n? The conditions are: c > d > 1, and
c, d have different parity. So one of the numbers c, d must be odd,
4 Parity
is the property of an integer being even or odd. So one of the numbers
c, d must be even, the other odd.
6.3 The problem of trapezoidal numbers 151

hence n must have an odd divisor, hence it cannot be a power of

2. Conversely, any odd divisor of n yields c and d. So n has a

trapezoidal representation if and only if it is not a power of 2. G

Let us write up our solution cleanly and fill in some details.

! Solution of Problem 6.3

Claim: n has a trapezoidal representation if and only if n is not a
power of 2.
Proof: We prove this in several steps. More precisely, we prove that
the following statements are all equivalent:

(i) n has a trapezoidal representation.

(ii) There are a, b ∈ N, a < b satisfying 2n = ( a + b)(b − a + 1).

(iii) There are c, d ∈ N of different parity satisfying c > d > 1 and

2n = cd.

(iv) n is not a power of 2.

Proof of (i) ⇐⇒ (ii): By definition n has a trapezoidal representation

if and only if there are a, b ∈ N, a < b satisfying n = a + · · · + b. By
the calculation above, a + · · · + b = 12 ( a + b)(b − a + 1). Therefore
n has a trapezoidal representation if and only if there are a, b ∈ N,
a < b satisfying 2n = ( a + b)(b − a + 1).
Proof of (ii) ⇒ (iii): Suppose there are a, b as in (ii). Let c = a + b,
d = b − a + 1. Then 2n = cd. Also, we get (6.3) by the calculation
there, and this implies (6.4) and (6.5) as shown there. These are the
conditions on c, d in (iii).
Proof of (iii) ⇒ (ii): Suppose there are c, d as in (iii). Define a, b by
(6.3). Then c = a + b, d = b − a + 1, so 2n = ( a + b)(b − a + 1), and
a, b ∈ N, a < b follow from (6.4) and (6.5) as proved there.
Proof of (iii) ⇒ (iv): Suppose (iii) holds. Since c, d have different parity,
one of them is odd. Since both are bigger than 1 it follows that 2n
must have an odd divisor. Then n must also have an odd divisor, so
n cannot be a power of 2.
152 6 General problem solving strategies

(ii) repre- (iii) repre-

(i) n is (iv) n is not
sentation sentation
trapezoidal a power of 2
using a, b using c, d

Figure 6.11 Solution scheme for the trapezoidal number problem

Proof of (iv) ⇒ (iii): Suppose (iv) is true. Since n is not a power of 2 it

must have an odd divisor x > 1. Let y = 2n x = 2 · x . Then 2n = xy.
n
n
Since x is a divisor of n the number x is a natural number. Therefore,
y is even, in particular y > 1 and x, y are different. So if we set

c = max{ x, y}, d = min{ x, y},

(that is, c is the bigger and d the smaller of the numbers x, y) then c, d
satisfy the conditions of (iii).

All in all, we obtain that (i) and (iv) are equivalent, as required. !

 Review of Problem 6.3

J The first attempt was based on an observation about examples and
generalizing this. In the second attempt we started by formulating
the goal as an equation; working backward and simplifying we
were able to formulate interim goals that finally led us to the
solution.
J One challenge in complex problems and long investigations is not
to get lost in the details, to keep track of where we are in the
solution process. Therefore it is important to ask yourself again
and again: what is given, what are we looking for, what have we
accomplished so far? It is a good idea to make notes of that during
the investigation.
Writing things up in a tidy, ordered manner is an important check
on whether the argument is complete.
J In the solution you saw a clear, systematic way to write up argu-
ments with several intermediate steps.
J The scheme of our solution is given in Figure 6.11. 
6.4 Going further: sum representations of integers 153

6.4 Going further: sum representations of integers

Trapezoidal numbers, triangular numbers etc.

There are also other kinds of ‘figurate numbers’: maybe you have
heard of triangular numbers. These are numbers which occur as num-
bers of points in certain triangular patterns, see Figure 6.12. The nth
triangular number is 1 + 2 + · · · + n. You already know a formula for
this. Square numbers are well-known. From pentagons you get the
pentagonal numbers (look them up on Wikipedia!). Very surprisingly
they come up in the partition number problem: in how many ways
can you write a natural number n as a sum of natural numbers in
nonincreasing order? For example 4 = 3 + 1 = 2 + 2 = 2 + 1 + 1 =
1 + 1 + 1 + 1, that’s five ways for n = 4; here 3 + 1 and 1 + 3 are
considered to be the same representation. This problem is much
more difficult than the problem where 3 + 1 and 1 + 3 are counted as
different, see Exercises E 1.9 and E 5.19.5

Figure 6.12 The triangular numbers 1, 3, 6, 10

Sum representations of natural numbers

Questions such as “Which natural numbers can be represented in

a certain way?”, “... and in how many ways?” are a rich source of
mathematical problems, which have led to many important devel-
opments. The difficulty of the problems varies drastically with the
kind of representation. Here are some more examples. You can find
solutions for some of them in the books (Hardy and Wright, 2008),
(Silverman, 2012).
J n = p1 + p2 : Representing n as the sum of two primes. Most math-
ematicians believe that every even number n can be represented
in this way. This conjecture was first formulated by Goldbach
5 See(Hardy and Wright, 2008), and also ‘Partition’ and ‘Pentagonal number
theorem’ in Wikipedia for explanations and references.
 154 6 General problem solving strategies

in 1742. The conjecture is known to be true for all even n up to

4 × 1018 , but is it true for all n? This is still unknown, in spite of
intensive efforts by many mathematicians.
J n = x2 − y2 : Representing n as the difference of two squares.
(From now on we always assume x, y ∈ N.) You can solve this, see
Exercise E 6.6. When modified slightly it gets much more difficult:
J n = x2 + y2 : Representing n as the sum of two squares. This is a
very pretty but difficult topic. For example, it can be shown that
all prime numbers that leave the remainder 1 when divided by 4
can be represented in this way (for example, 13 = 22 + 32 ), but
those that leave the remainder 3 cannot (for example, 19). See also
Exercise E 8.10. The answer for general n is known but a little
more complicated.
J n = x2 − 2y2 : Representing n as the difference of a square and
twice a square. This is also not easy, but known. The equation
x2 − 2y2 = 1 is called Pell’s equation. One solution is x = 3, y =
2. There are infinitely√many solutions, and you√obtain them by
multiplying out (3 + 2 2)k and writing it as√x + y 2 with x, y √ ∈ N.
For example, √ if k = 2 then you get ( 3 + 2 2 ) 2 = 9+2·3·2 2+

8 = 17 + 12 2, and indeed 172 − 2 · 122 = 1. It is a nice exercise to

check that ( x, y) is a solution of Pell’s equation for each k. It is a
little harder to show that you get all solutions in this way.
J n = x2 + y2 + z2 + u2 , x, y, z, u ∈ N0 : Representing n as the sum
of at most four squares. This is possible for all n (Lagrange’s
four-square theorem), but again this is not easy to prove.
J n = 2i0 + 2i1 + · · · + 2ik , i0 < · · · < ik , k arbitrary: Representing n
as the sum of different powers of 2. For each n there is a unique
such representation (exercise!). This is a restatement of the fact
that one can represent numbers in the binary system, hence is the
basis of all computer technology.

Exercises

1 E 6.1 .Find perimeter and area of the shape in Figure 6.13 (not by
measuring, it is not drawn to scale). Which strategies do you use?
1-2 E 6.2 .You have two pots, with volumes 9 and 4 litres respectively.
Exercises 155

5 cm 6 cm
2 cm
8 cm

Figure 6.13 Exercise E 6.1

How can you measure precisely 6 litres? You have no tools except
for the pots and an unlimited supply of water. Formulate a suitable
interim goal. Is it useful to work backward?

E 6.3 .Problem 6.2 (diagonal of a cuboid) is symmetric with respect 2

to the side lengths a, b, c: each of them plays the same role. Is the
solution formula symmetric? Is the derivation symmetric? If not,
what can you say about alternative derivations?

E 6.4 .Look back at the investigation of Problem 6.3. Complete the 2

first attempt at proving that a trapezoidal representation of n exists if
n is not a power of two.

E 6.5 . How many representations of the numbers 12 and 750 as 2-3

trapezoidal numbers are there? How many ways are there for an
arbitrary n ∈ N?

E 6.6 .Which numbers n ∈ N can be written as the difference of two 2

square numbers?

E 6.7 .Starting at 1 add the natural numbers in increasing order one 2-3
by one. Is it possible to obtain a 5 digit number with all digits equal?

E 6.8 . Let digits a1 , . . . , an be given. Prove that √

there is a natural 3
number N such that the decimal representation of N has the digits
a1 ...an immediately after the decimal point.

E 6.9 . Let three points A, B, C in the plane be given. Explain a 2-3

constructive procedure to find a line that passes through A and has
the same distance from B and C.

E 6.10 .Write 4 zeroes and 5 ones round a circle in any order. Keep 2
repeating the following step: Consider each pair of neighbouring
numbers. If they are equal, write a new one between them; if different,
 156 6 General problem solving strategies

a new zero. Now delete the original (so that you are left with 9
numbers again). Can it happen that you ever get 9 ones? Can you
ever get 9 zeroes?

3-4 E 6.11 .100 cards lie in a row on the table, face down. First you turn
over the cards at positions 2,4,6,. . . Then you turn over the cards at
positions 3,6,9,. . . (so that now the card at position 6 is face down
again), then the cards at positions 4,8,12,. . . are turned etc. until in the
last round only the card at position 100 is turned over. Which cards
lie face down at the end?

2 E 6.12 .What is the solution of Problem 6.1 if there are n customers

instead of 5?
 √  √
2 E
√ 6.13 .Investigate whether the number x = 4 + 7 − 4 − 7−
2 is positive, negative or zero.
7 Logic and proofs

We formulate mathematical arguments in our natural language. But

our everyday language is often ambiguous, and quite often you hear
logical fallacies. Therefore, if you want to argue reliably then you
should know well both the basic logical structures and the phrases
we use to express them. These are the topic of the first section in this
chapter.
In the course of a mathematical investigation you make observa-
tions, discover patterns, have insights, make conjectures. In order to
be sure that a conjecture is true you need a proof.
Logic is one of the foundations of proving. It helps you to be sure
that your argument is correct. How to find a proof is quite a different
matter. Experience will help, and it is useful to have a good command
of problem solving strategies, and to know the main types and patterns
of proofs.
The general types of proof are direct proof, indirect proof, and
proof by contradiction. Proof patterns are more specific to context.
In the second section of this chapter you will learn about the gen-
eral types of proofs and about some patterns which are common in
proofs of formulas, proofs of existence, and proofs of non-existence
or impossibility. In subsequent chapters you will study three of these
patterns in greater depth.
Of course there are no limits to your creativity in finding your own
ways of proving.

7.1 Logic

When doing mathematics we try to understand, calculate, determine

things. We formulate our conjectures or results as propositions. In
logic we investigate how propositions are combined to form new
propositions and how the truth value (true or false) of the new propo-
sitions depends on that of the old ones. We also study how to draw
correct conclusions from true propositions.

© Springer International Publishing AG, part of Springer Nature 2018 157

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7_7
158 7 Logic and proofs

Propositions and predicates

In logic a proposition is a statement which can be either true or false.

This is called the truth value of the proposition.
Here are some examples of propositions:

J A: ‘1 + 1 = 2’

J B: ‘the square of any even number is even’

J C: ‘1 = 2’

J D: ‘5 is negative’

J E: ‘Every even number n > 2 can be written as the sum of two

primes’

The propositions A, B are true, C, D are false; as for E, we don’t know

whether it is true or false.1 . ‘It is raining’ is also a proposition. Look
out of the window to see whether it is true.
On the other hand, ‘maybe it is raining’, ‘I hope it will rain soon’,
‘is 1+1=2?’ and ‘xy&,%’ are not propositions.
The statement ‘n is even’ is not a proposition because it is not clear
what n is. If you plug in a value for n, say n = 3, then the statement
turns into the proposition ‘3 is even’ (which is false). A statement
containing one or several variables, which becomes a proposition
when we plug in values for the variables, is called a predicate. If we
denote by A(n) the predicate ‘n is even’ then the proposition A(1) is
false and the proposition A(2) is true, for example.
The variables in a predicate need not correspond to numbers. They
can also represent other mathematical objects, for example triangles
(‘the triangle T has two equal angles’) or graphs (‘the graph G is
connected’).
If A is a proposition then the negation of A is another proposition,
which is sometimes denoted ¬ A. The negation of ‘1 + 1 = 2’ is
‘1 + 1 = 2’; the negation of ‘it is raining’ is ‘it is not raining’.

1 Most mathematicians conjecture that it is true. This is called Goldbach’s

conjecture.
7.1 Logic 159

A B A and B A or B A⇒B ¬A
t t t t t f
t f f t f f
f t f t t t
f f f f t t

Table 7.1 Truth tables for and, or, if-then and negation. t = true, f = false

The logical operators ‘and’, ‘or’

From two propositions A, B you can form the new propositions ‘ A

and B’, ‘ A or B’. The truth values of these new propositions are
determined by the truth values of A, B as follows:

J ‘ A and B’ is true if and only if both A and B are true;

J ‘ A or B’ is true if and only if at least one of the propositions A, B is

true.

In the examples above ‘ A and B’ is true, ‘ A and C’ is false, ‘ A or C’ is

true, and ‘C or D’ is false. How about ‘ A or E’, ‘D or E’, ‘ A and E’?

Think about it!

The proposition ‘ A or E’ is true because A is true. Whether the

propositions ‘D or E’, ‘ A and E’ are true we don’t know. This
depends on the truth value of E.
The proposition ‘ A or B’ in the example is also true: In mathematics
‘or’ is always inclusive, that is, it is also true if both A and B are true.
To express the exclusive ‘or’ you say ‘either A or B’ (this would be
false for the example). In everyday language it depends on the context
whether ‘or’ is meant in the inclusive or the exclusive sense, and can
still be ambiguous.
The logical meaning of ‘and’, ‘or’ can be represented in a truth
table, see the third and fourth column of Table 7.1.
160 7 Logic and proofs

The quantifiers ‘for all’, ‘there is’

Many mathematical statements begin ‘For all natural numbers . . . ’ or

‘For all triangles . . . ’ or ‘There is a number for which . . . ’. In general,
suppose A(n) is a predicate and M is a set of values that n can take.
Then we can consider the following propositions:

J For all n ∈ M: A(n) holds (in symbols: ∀n ∈ M : A(n)).

J There is an n ∈ M for which A(n) (in symbols: ∃n ∈ M : A(n)).
holds

The expressions ‘for all’ and ‘there is’ (or the symbols ∀ and ∃) are
called quantifiers. Here are a few examples.

J ‘For all n ∈ N: n + 1 ∈ N’ is true.

J ‘For all n ∈ N: n − 1 ∈ N’ is false since for n = 1 you get

1 − 1 ∈ N, which is false since 0 is not a natural number2 .

J ‘There is a triangle that has two equal angles’ is true.

J ‘There is a triangle whose angles sum to 200 degrees’ is false, since

for all triangles the sum of angles is 180 degrees.

To ‘refute‘ a statement means to show that it is false. We see that a

‘for all’ statement can be refuted by a single counterexample. A ‘there
is’ statement is refuted by a ‘for all’ statement.
‘There is a . . . ’ always means ‘there is at least one . . . ’. If you want
to express that, in addition, there can’t be more than one, then you
say ‘there is precisely one . . . ’ or ‘there is a unique . . . ’.
Often you use different wording, for example: instead of ‘for all n: n
is positive’ we often say ‘all n ∈ N are positive’ or ‘for any natural
numbers n we have that n is positive’ or ‘let n ∈ N; then n is positive’.
Instead of ‘there is a prime n which is even’ we also say ‘there is an
even prime’. Also, often we say ‘there exists’ instead of ‘there is’.
Important: In mathematics a proposition of the form ‘ ∀n ∈ M :
A(n) ’ is always considered true if M is the empty set. This makes
sense since there is nothing that could refute the proposition. In

2 N is defined as {1, 2, 3, . . . }. Some authors denote the set {0, 1, 2, . . . } by N.

In this case the proposition is refuted by n = 0. In this book we denote the set
{0, 1, 2, . . . } by N0 .
7.1 Logic 161

everyday language this is not so: if you say “All of my houses are
built of stone” then people will believe that you own at least one
house. But logically this does not follow – the statement is true even
if you do not own any houses.
You can change the name of a variable inside a ‘for all’ or a ‘for is’
proposition without changing its truth value: instead of ‘for all n ∈ N:
n + 1 ∈ N’ you may just as well say ‘for all m ∈ N: m + 1 ∈ N’. Of
course you have to change the name at all places referred to by the
quantifier. For example, ‘for all m ∈ N: n + 1 ∈ N’ is not the same
proposition – in fact it is not even a proposition since it is unclear
what n is.
This is often useful, even necessary, in arguments where several
quantifiers occur. See the proof of the divisibility rules in Section 8.1
for an example.
In many propositions several quantifiers are combined: the every-
day wisdom ‘every pot has a lid’ can be rephrased ‘for each pot there
is a lid which fits’.
Watch out! Order matters. The proposition ‘there is a lid which fits
on every pot’ would mean something quite different. Formally you
could write the two propositions as follows (we usually leave out the
colon between quantifiers):

‘ ∀ p ∈ {pots} ∃l ∈ {lids} : l fits on p’

‘ ∃l ∈ {lids} ∀ p ∈ {pots} : l fits on p’

Here is a mathematical example:

‘For all n ∈ N there is an m ∈ N with m > n’
‘There is an m ∈ N such that for all n ∈ N: m > n’

The first proposition is true, the second is false.

The implication: ‘if . . . then . . . ’

From two propositions A, B you can form the new proposition ‘if A
then B’, in symbols ‘ A ⇒ B’. We also say ‘ A implies B’ and call A
the premise and B the conclusion.3 The proposition ‘ A ⇒ B’ is false
3 Sometimes different words are used, for example hypothesis, assumption or
antecedent instead of premise, and consequent instead of conclusion.
162 7 Logic and proofs

if A is true and B is false, and is true otherwise. In particular, it is

always true if A is false. See the fifth column in Table 7.1. Remember:
The only way that ‘ A ⇒ B’ can be false is that A is true
and B is false.
Why is this reasonable? As an example, consider the proposition:

For all n ∈ N: if n is divisible by 4 then n is even,

which is clearly true. Then all of the propositions

if 1 is divisible by 4 then 1 is even
if 2 is divisible by 4 then 2 is even
if 3 is divisible by 4 then 3 is even
if 4 is divisible by 4 then 4 is even
etc.
must be true. Let us check the truth values of premise (‘. . . is divisible
by 4’) and conclusion (‘. . . is even’) for each of these propositions:
in the first and third line both premise and conclusion are false; in
the second line the premise is false and the conclusion is true; and
in the fourth line both premise and conclusion are true. These cases
correspond to the three lines of the truth table in which ‘ A ⇒ B’ is
true. And if we had found a number n which is divisible by 4 but not
even, then the implication would have been wrong. This explains all
lines in the truth table for ‘ A ⇒ B’.
Here is another example. A: ‘it is raining’ and B: ‘the street gets
wet’. The proposition ‘ A ⇒ B’ says ‘if it rains then the street gets
wet’. This could also be phrased as ‘every time it rains, the street gets
wet’.4 This is only wrong if there is a time when it rains but the street
remains dry. If it never rains then the statement is still true, no matter
whether the street gets wet or not – since nothing is claimed in that
case.
As in these examples, propositions of the form ‘if . . . then . . . ’ are
almost always part of a ‘for all’ clause, even if this is not stated
explicitly. Therefore, whenever you see ‘if . . . then . . . ’ you should
add something like ‘every time . . . ’.
4 Very precisely: Let A ( t ), B ( t ) be the propositions, for a point in time t: ‘it is

raining at time t’ and ‘the street is getting wet at time t’. Then ‘ A ⇒ B’ is short for
the ‘for all’ proposition ‘ ∀t : A(t) ⇒ B(t) ’.
7.1 Logic 163

We can rephrase implications in various ways: instead of saying

‘if it rains then the street gets wet’ we can say ‘if the street remains
dry then it is not raining’, or ‘it does not happen that it rains and the
street remains dry’. Formally this says that for any two propositions
A, B the following three propositions are equivalent:

A⇒B ¬B ⇒ ¬ A ¬( A and ¬ B) .

This is the basis of indirect proof and of proof by contradiction, see

below. Here, ‘equivalent’ means that for any truth values of A, B
the three propositions above have the same truth values. You can
check this directly using Table 7.1. We also say that they are logically
equivalent.
But note that A ⇒ B is not logically equivalent to ¬ A ⇒ ¬ B. In the
example this would be ‘if it is not raining then the street remains dry’
– this is clearly a different proposition (which is false since the street
could also get wet from a lawn sprinkler). Convince yourself using
the truth table. Similarly, A ⇒ B is not logically equivalent to B ⇒ A.
You hear this sort of fallacy quite often in everyday life, for example
like this: ‘If it rains then the street gets wet. It’s not raining. Therefore
the street remains dry.’ or ‘If it rains then the street gets wet. The
street is wet. Therefore it must be raining.’ Both conclusions are
logically incorrect.

Necessary and sufficient

Instead of ‘ ∀n ∈ M : A(n) ⇒ B(n) ’ we also say ‘ A(n) is sufficient for

B(n), for n ∈ M’ and ‘B(n) is necessary for A(n), for n ∈ M’. Often
we leave out ‘for n ∈ M’ when M is clear from the context.
These phrases emphasize different points of view:

J A(n) is sufficient for B(n): we are looking for conditions on n

which ensure that B(n) is true; if we have proved ‘ ∀n ∈ M :
A(n) ⇒ B(n) ’ then we know that A(n) is such a condition.

J B(n) is necessary for A(n): we are looking for conditions on n

which must necessarily hold for A(n) to be true; if we have proved
‘ ∀n ∈ M : A(n) ⇒ B(n) ’ then we know that B(n) is such a
condition.
164 7 Logic and proofs

As an example consider the propositions A(n): ‘the decimal repre-

sentation of n ends in a zero’ and B(n): ‘n is even’ about natural
numbers n ∈ N. The true proposition ‘ ∀n ∈ N : A(n) ⇒ B(n) ’ can
be restated as follows:

J A(n) is sufficient for B(n): in order to know that n is even, it

suffices to know that n ends in a zero.

J B(n) is necessary for A(n): if n ends in a zero then it must neces-

sarily be even.

But B(n) is not sufficient for A(n): an even number need not end in
a zero. Similarly, A(n) is not necessary for B(n). Pay attention to the
reversal of order when using ‘necessary’.
We can summarize the meaning of ‘necessary’ and ‘sufficient’ as
follows: If A(n), B(n), C (n) are predicates then A(n) ⇒ B(n) ⇒ C (n)
means that A(n) is sufficient for B(n) and that C (n) is necessary for
B ( n ).
We usually use these words when we are investigating a compli-
cated property of certain objects, and trying to find a condition which
is easy to check and which is logically related to this property.
Here is an example: Consider the following proposition A( G ) about
a graph G: ‘G can be drawn in the plane without edge intersections’.
The solution of Problem 4.6 shows that a necessary condition for
A( G ) is: G has no 5 vertices which are all joined by edges. How-
ever, this condition is not sufficient: suppose G contains 6 vertices
E1 , E2 , E3 , F1 , F2 , F3 where each Ei is joined to each Fj by an edge. Then
G does not satisfy A( G ) either, see Exercise E 4.6.5

Equivalence: ‘if and only if’

We call propositions A, B (logically) equivalent and write A ⇔ B if

both A ⇒ B and B ⇒ A are true. We also say ‘ A holds if and only if
B holds’ or ‘ A is necessary and sufficient for B’.
As for the implication, this usually occurs inside a ‘for all’ clause:
∀ n ∈ M : A ( n ) ⇔ B ( n ).

5 Amazingly, in a sense this is all that can happen: using these two special cases

one can formulate a condition which is necessary and sufficient for A( G ). This is
Kuratowski’s theorem, see (Aigner, 1987) for example.
7.1 Logic 165

Negation of composite propositions

When negating a composite proposition you have to be very careful.

It is worth clarifying this in simple examples and then memorising
how it works.
In order to find the negation of a proposition A you ask yourself:
what precisely must be true in order for A to be false? What do I
need to show in order to refute A?
This leads to the inversion rule for negations:
When negating, interchange ‘and’ with ‘or’ and ∀ with ∃,
and pull the negation sign to the end.
This means:6
the proposition is equivalent to the proposition
¬( A and B ) ¬ A or ¬ B
¬( A or B ) ¬ A and ¬ B
¬( ∃n ∈ M : A(n) ) ∀n ∈ M : ¬ A(n)
¬( ∀n ∈ M : A(n) ) ∃n ∈ M : ¬ A(n)

Examples

J The proposition ‘yesterday it rained and snowed’ is false if

and only if it did not rain or it did not snow.

J The proposition ‘yesterday it rained or snowed’ is false if and

only if it did not rain and it did not snow.

J The proposition ‘it rains every day’ is false if and only if there
is a day when it does not rain.

J The proposition ‘this week there was a day when it rained’ is

false if and only if it was dry every day of the week.

The inversion rule is especially useful when negating more complex

combinations of propositions. You apply it several times:
The negation of ‘for all n ∈ N there is an m ∈ N with m > n’
is ‘there is an n ∈ N so that for all m ∈ N: m ≤ n’.
6‘¬ A or ¬ B’ is to be understood as ‘(¬ A) or (¬ B)’
166 7 Logic and proofs

Be sure you understand that this is the logically correct negation.

Here is an another example: the negation of ‘every pot has its lid’ is
‘there is a pot which does not have a lid’. More systematically: the
negation of ‘for every pot there is a lid that fits’ is ‘there is a pot for
which all lids don’t fit’.
You can also check this purely formally by pulling the ¬ sign to
the right step by step, using the inversion rule in each step:

¬(∀n ∈ N ∃m ∈ N : m > n) ⇐⇒
∃n ∈ N ¬(∃m ∈ N : m > n) ⇐⇒
∃n ∈ N ∀m ∈ N : ¬(m > n) ⇐⇒
∃n ∈ N ∀m ∈ N : m≤n

In newspapers and politicians’ speeches you will find many incorrect

negations. For example, the statement ‘For all t: A(t) ’ is often negated
as ‘For all t: ¬ A(t) ’, but this means something quite different.

7.2 Proofs

A proof is a logically complete justification of a proposition. As long

as we have not proven a proposition it is possible that it is false – even
if it is supported by many examples.

Example

In the year 1637 Fermat investigated the numbers Fn = 2(2 ) + 1.

n

He noticed that F0 , F1 , F2 , F3 and F4 are prime numbers. For

example: F0 = 2(2 ) + 1 = 21 + 1 = 3, F1 = 2(2 ) + 1 = 22 + 1 =
0 1

4 + 1 = 5, F2 = 2(2 ) + 1 = 24 + 1 = 16 + 1 = 17. Therefore he

2

conjectured that Fn is prime for all n.

It was almost 100 years until Euler discovered, in 1732, that
F5 = 4294967297 is not prime: it is divisible by 641. Today the
numbers Fn are called Fermat numbers.

This shows that examples can be quite misleading.7 So we need a

proof if we want to be sure that a proposition is true.

7 See the Exercises E 1.10 and E 5.24 for other such problems.
7.2 Proofs 167

In addition, a proof can give you a deeper understanding of a

fact. A good example is the second/third proof of the formula
|P ({1, . . . , n}| = 2n in Section 5.2.

General types of proofs

Often we want to prove propositions of the form A ⇒ B.89 For this

we can proceed in different ways.

Direct proof The proof is composed of steps whose validity is ob-

vious or which have been proved before: instead of proving
‘ A ⇒ B’ directly, we prove ‘ A ⇒ A1 ⇒ A2 ⇒ . . . ⇒ B’.

Indirect proof We suppose that the conclusion B is false and deduce

from this that the premise A must also be false. So instead of
‘ A ⇒ B’ we prove ‘ ¬ B ⇒ ¬ A’, which is logically equivalent.
Of course the latter proof can be composed of several steps.

Proof by contradiction A proof by contradiction is very similar to

an indirect proof. We suppose that A is true and B is false and
derive a contradiction (a false statement) from this.
So we prove ‘(A and ¬ B) ⇒ f’. Since true propositions can
imply only true propositions, it follows that ‘ A and ¬ B’ is false.
This is logically equivalent to ‘ A ⇒ B’.

Sometimes there is no premise: We want to prove a proposition B.

Then a direct proof concludes B from known propositions, and a
proof by contradiction derives a false statement from ¬ B: ¬ B ⇒ f.10
Here are two more general types of proofs.

Refutation by counterexample In order to refute a ‘for all’ proposi-

tion (i.e. in order to prove that it is false) it suffices to give one
counterexample.
8 Instead of ‘to prove’ we also say ‘to show’.
9 More precisely, these are propositions of the form ‘for all X: A( X ) ⇒ B( X ) ’,
for certain mathematical objects X (for example, natural numbers or graphs). For
better readability we use the abbreviated form A ⇒ B in the sequel. Compare the
discussion on implication in the previous section.
10 Strictly speaking, there is no indirect proof in this case. However, the proof by

contradiction is sometimes called indirect proof.

168 7 Logic and proofs

Mathematical induction This can often be used to prove proposi-

tions of the form ‘For all n ∈ N . . . ’. See Chapter 3.
We have already seen a refutation by counterexample: the proposition
‘for all n ∈ N0 : Fn is a prime number’ about the Fermat numbers
was disproved by showing that F5 is not a prime.

Remark

In a proof you can use propositions which have been proved

before or which are considered to be known in a given context.
In a strictly axiomatic build-up of mathematics, which you learn
when you study mathematics at university, you are only allowed
to use axioms, or propositions which have been proved from the
axioms beforehand.11 This assures you that all proven proposi-
tions are true – assuming only that the axioms are true. You have
to begin somewhere, and proceeding in this way allows you to
make this ‘base’ very small.

Here is a simple example of a direct proof.

Theorem The sum of two even numbers is even.

We first reformulate this to show clearly the premise and the conclu-
sion:
Claim:
Let n, m be even numbers. (premise)
Then n + m is even. (conclusion)
Proof.
Since n, m are even, there are j, k ∈ Z so that n = 2j and m = 2k (this
is the definition of ‘even’). Then n + m = 2j + 2k = 2( j + k ). We have
j + k ∈ Z. Therefore, n + m is also of the form n + m = 2q with q ∈ Z
and hence an even number. q. e. d.
11 An axiom is a proposition which is postulated to be true in a given context. For
example, one of the axioms of plane geometry is the parallel postulate: Given any
line g and any point P not lying on g there is a unique line g
which contains P and
does not intersect g. As foundation for a mathematical theory one chooses as few
axioms as possible.
7.2 Proofs 169

Proof analysis: This was a direct proof. We first applied the

definition of ‘even’. Then there was one intermediate step: the
calculation of n + m using the distributive law. Then we applied
the definition again.

We now consider some common types of propositions and the proof

patterns which are typical for them.

Proofs of formulas

Formulas express general relations between numbers or other mathe-

matical objects. In previous chapters you have already encountered
many formulas, for example 2n = ∑nk=0 (nk) or counting formulas.
Sometimes you conjecture a formula when investigating a problem
(for example in Problem 1.3). Then you need to prove it in general.

Typical proof patterns for formulas:

J Direct derivation by combining known formulas

J Mathematical induction (see Chapter 3)
J Counting in two ways (see Chapter 5)

When penetrating deeper into mathematics you will get to know

many more ways to derive or prove formulas. One of them is the
solution of the Fibonacci recursion in Chapter 2.
The point of view of modern mathematics is that formulas are
usually expressions of deep relationships between structures (for
example geometric objects, algebraic structures etc.). An example of
this is counting in two ways.

Proofs of existence

Many mathematical questions are about the existence of certain ob-

jects: does an equation have a solution? Are there numbers, triangles,
graphs, . . . having certain properties?
As an example, we ask whether the following equations have a
solution (in real numbers):

a) x5 + x = 2: we can exhibit a solution: x = 1.

170 7 Logic and proofs

b) x2 + 1 = 0: this equation does not have a solution since the left-

hand side is positive for all real numbers x.

c) x5 + x = 1: this is harder. Try out a few values for x. It is very

unlikely that you will find a solution. But the equation does have a
solution. However, this can only be proved using the intermediate
value theorem from analysis.12
So the simplest way to prove existence is to exhibit the desired object.
But often we want to prove a proposition of the kind ‘for all n ∈ M
there is . . . ’ where M is an infinite set. That is, we need infinitely
many proofs of existence – or one which works for all n. The simplest
way to do this is to give a general construction.
Here are two examples:

Theorem For every odd number n ∈ N there are two square

numbers whose difference equals n.

Try to find a general construction, i.e. a prescription for finding such

a pair of squares from n. Use problem solving strategies (for example,
looking at small examples).

Think about it!

Proof.
Since n is odd we can write it as n = 2m + 1 where m ∈ N0 . Then
n = ( m + 1)2 − m2 . q. e. d.
Proof analysis: We needed to supply a proof of existence for each
of the infinitely many odd numbers n. This was accomplished
by a general construction, in this case a general formula for the
two squares.
12 Idea: for x = 0 we have x5 + x = 0 < 1 and for x = 1 we have x5 + x = 2 > 1,

therefore there must be an x between 0 and 1 satisfying x5 + x = 1. The intermediate

value theorem says that the last conclusion is correct, based on the fact that the
function f ( x ) = x5 + x is continuous.
7.2 Proofs 171

Theorem For any triangle there is a point that is the same

distance from all three vertices.

 Investigation
G
How could we find such a point? Let us first simplify the problem:
instead of requiring the same distance from three points we content
ourselves with the same distance from two points. So we ask: Given
two points A, B in the plane, which points are the same distance
from A and from B? Maybe you remember the answer from school:
these are precisely the points on the perpendicular bisector of the line
segment AB.
How do we get from same distance from two points to same
distance from three points A, B, C?

Think about it!

By first considering the points A, B and then the points B, C: the

point we are looking for must lie on the perpendicular bisector of AB
and on the perpendicular bisector of BC, so it must be the intersection
point of these two lines. Do they intersect? Yes, if they are not parallel.
Why are they not parallel? Since AB, BC are not parallel if ABC is a
triangle. G
We write up the proof in order.
Proof.
Denote the vertices of the triangle by A, B, C. Let g be the perpendic-
ular bisector of the side AB and h the perpendicular bisector of the
side BC. Every point of g is the same distance from A and B, and
every point of h is the same distance from B and C.
Since the sides AB, BC are not parallel, the lines g and h are not
parallel, therefore they have a point of intersection. We call this point
P.
Since P lies on g it is the same distance from A and B. Since P
lies on h it is the same distance to B and C. Therefore P is the same
distance from A, B and C. q. e. d.
172 7 Logic and proofs

Proof analysis: We wanted to show the existence of a special

point for any triangle. For this we made a general construction.
For finding the construction it was useful to first consider a
simpler variant of the problem.

Sometimes we want to prove the existence not only of a single object

of a certain type, but of infinitely many such objects. This turns out to
be similar to proving a proposition of the type ‘for all n ∈ M there is
. . . ’. Here is a famous example.

Theorem There are infinitely many prime numbers.

Proof.
We give a procedure which does the following: for an arbitrary finite
set of primes it produces a prime which is not contained in this set.
Let p1 , p2 , . . . , pm be primes. Consider the number n = p1 · p2 · · · pm +
1. This is clearly bigger than each of the numbers p1 , . . . , pm . So if n
is prime then we are done.
If n is not prime then it is divisible by a prime.13 Call this prime
p. Since n is divisible by p, the number n − 1 is not divisible by p.
Therefore p1 · p2 · · · pm = n − 1 is not divisible by p, so p cannot be
one of the primes p1 , . . . , pm .
We have proved that for any finite set of primes there is another
prime not contained in the set. Therefore there are infinitely many
primes. q. e. d.

Proof analysis: This was a constructive proof of existence. We

could have formulated the same idea as proof by contradiction:
suppose there were only finitely many primes p1 , . . . , pm . Then
construct a prime as above, which does not appear among
them. This contradicts the assumption that p1 , . . . , pm are all
the primes.14

13 Why? Let p be the smallest integer which divides n and which is bigger than
1. Then p must be a prime number, since otherwise we could write p = ab with
1 < a < p, and a would divide n and would be smaller than p, contradicting the
choice of p.
14 You will find this indirect argument in many books. However, it is better to

phrase arguments in a direct way whenever possible. This makes them easier to
understand. Principle: use as few negations as possible.
7.2 Proofs 173

The limits of general constructions: In many problems of the

type “Show that for all n ∈ N there is an X satisfying . . . ” (where X
is some sort of mathematical object) you will experience the following:
You can prove the claim for many special cases (say n = 1, 2, 3, 4, 5) by
finding an X. But you don’t find a general construction which works
for all n, and you don’t find a pattern that could lead you to such
a construction. You will encounter such problems in the following
chapters, and you will get to know two far-reaching ideas for such
situations: the pigeonhole principle and the extremal principle. Let
us summarize:

Typical patterns for proofs of existence:

The simplest way to prove existence of an object is to exhibit it.

In order to prove a proposition of the type ‘for all n ∈ M there is
. . . ’ you can give a general construction which works for all n, or
try one of the following strategies:
J The pigeonhole principle (see Chapter 9)
J The extremal principle (see Chapter 10)
Of course general proof strategies like proof by contradiction
or mathematical induction (here in the form of an inductive
construction) can be useful as well.

In each area of mathematics you will find more specific tools for
proofs of existence. For example, in analysis the intermediate value
theorem and fixed point theorems, and in linear algebra (and other
areas) dimensional arguments (a homogeneous linear system of equa-
tions with more variables than equations must have a non-trivial
solution).
We have already encountered a special type of proof of existence:
refutation by counterexample. This is a case of proof of existence by
exhibiting an object: the negation of ‘for all n ∈ N: A(n) ’ is ‘there
is an n ∈ N for which A(n) is false’, and this can be proved by
exhibiting an n for which A(n) is false.
When disproving Fermat’s conjecture that all Fermat numbers
are prime Euler used the principle ‘proof of existence by exhibiting’
twice: he exhibited the Fermat number F5 which is not prime; and
174 7 Logic and proofs

in order to prove this he exhibited the non-trivial factor 641 of F5 . So

the second proof of existence is nested inside the first.

Proofs of nonexistence and impossibility

To prove that something does not exist or is impossible is a fascinating

kind of problem and can be quite difficult. Here are some examples
of such statements:15
J It is impossible to join five points in the plane pairwise in such a
way that the connecting lines don’t intersect.
J A power of 2 cannot be represented as a trapezoidal number.
J There is no algorithm for calculating the prime factorisation of a
large integer in acceptable time.
J For all natural numbers n > 2 the equation x n + yn = zn has no
solution with x, y, z ∈ N.
The first two examples are Problem 4.6 and Problem 6.3. The third
example needs to be made precise, for example like this: there is no
algorithm which calculates the prime factorisation16 of an arbitrary
1000 digit number (in the decimal system) in less than 1000 years, on
current computer hardware. This could be read as a statement about
known algorithms, and read this way it is true. However, whether
such an algorithm cannot exist at all is a difficult, unsolved problem:
So far nobody has found such an algorithm, but no one has been able
to prove it cannot exist. See the end of Section 11.4 for the practical
relevance of this problem.
The fourth example is the famous Fermat problem. As early as
1637 Fermat claimed that there cannot be a solution, but this was
proved only in 1995.
Two other famous examples are the impossibility of trisecting an
arbitrary angle using only ruler and compass, and Gödel’s incom-
pleteness theorem, which says that it is impossible to prove mathe-
matically that mathematics is free of contradictions. See Wikipedia
for explanations and references.
15 Nonexistence and impossibility are really the same thing. For example, the

second statement can be rephrased as ‘there does not exist a representation of a

power of 2 as . . . ’ or as ‘it is impossible to represent a power of 2 as . . . ’.
16 We will discuss prime factorisation in detail in Chapter 8.
7.2 Proofs 175

Here is a classical nonexistence statement.

√
Theorem The number 2 is irrational.

 Investigation
G
Let us
√ first see that this is a statement of nonexistence: the claim is
that 2 is √not rational, that is, that there are no natural numbers p, q
p
satisfying 2 = q .
How could we prove that? We could try a proof by contradiction:
wesuppose there were such p, q, then we try to derive more and more
consequences from this until we arrive √ at a contradiction.
p
So assume there are p, q ∈ N with 2 = q . What could we do
with this equation? Can we simplify it?

Think about it!

There are two complications: the fraction and the square root. So let’s
multiply by q to get rid of the fraction and then square to get rid of
the root: √ √
p
2 = ⇒ 2 q = p ⇒ 2q2 = p2 .
q
How could we continue, what can we conclude from this?
The equation 2q2 = p2 implies that p2 is even. Then p must also
be even (see below for a complete proof). We make this explicit by
writing p = 2r for some natural number r.
We square and get p2 = 4r2 . We plug this into 2q2 = p2 and obtain
2q2 = 4r2 . Now we can divide by 2 and get

q2 = 2r2 .

By the same argument as before q must be even. So both p, q must be

p
even, so we can cancel the factor 2 in the original fraction q .
√ p
But what if we had started with a fraction 2 = q in lowest terms?
We could certainly choose p, q to do this. Then we would have the
desired contradiction. 
G
 176 7 Logic and proofs

Let us write up the proof in order. First we prove a lemma.17

Lemma Let n ∈ N. If n2 is even then n is even.

Proof.
We prove this indirectly. Suppose n was odd. Then we could write
n = 2m + 1 where m ∈ N0 . Then we would get n2 = (2m + 1)2 =
4m2 + 4m + 1 = 2(2m2 + 2m) + 1, so n2 would be odd. q. e. d.
√
We now prove that 2 is irrational.
Proof. √ p
Suppose 2 = q with p, q ∈ N. We may assume that the fraction is
reduced to lowest terms, so that p and q have no common factor. We
multiply by q and square to obtain 2q2 = p2 . Therefore p2 is even,
so p is even by the lemma. Write p = 2r with r ∈ N. Squaring this
and plugging in we obtain 2q2 = 4r2 , hence q2 = 2r2 . Therefore q2
is even, hence q is even by the lemma. Therefore p and q are both
even, so they have the common factor 2. This is a contradiction√to
p
the assumption that the fraction q was in lowest terms. Therefore 2
cannot be rational. q. e. d.

Proof analysis: This was a proof by contradiction.

Typical patterns for proofs of nonexistence/impossibility:

J Proof by contradiction
J The invariance principle (see Chapter 11)

Exercises

1 E 7.1 .Take a newspaper and find at least three examples of logically

incorrect reasoning.

1-2 E 7.2 . Each of the five cards in Figure 7.1 has a letter on one side
and a number on the other side. How many cards do you need to
17 A lemma (from the Greek) is an auxiliary proposition which is used in the
proof of a theorem.
Exercises 177

turn over in order to find out whether the following proposition is

true: If the card has an even number on one side then the other side shows
a vowel.

B 6 7 E X

Figure 7.1 Exercise E 7.2

E 7.3 .Consider a colouring of the natural numbers using the colours 1-2
red and blue. Formally, this is a map f : N → {red, blue}. Also
consider the following proposition:
For each blue number there is a larger number which is red.
Which of the following propositions follow from this?
a) There is a blue natural number.
b) ∃n ∈ N : f (n) = red
c) For each red number there is a smaller number which is blue.
d) For each red number there is larger number which is red.
e) |{m ∈ N : f (m) = red}| = ∞
f) If f (1) = blue then there is n ∈ N such that f (n) = blue and
f (n + 1) = red.

E 7.4 .Suppose a connected figure consisting of unit squares is given. 1-2

Decide for each of the following propositions which of the other
propositions it implies. Give proofs.

a) The number of squares is three.

b) The number of squares is divisible by three.
c) The figure can be tiled by tiles of the form .
d) The figure can be tiled by tiles of the form .

E 7.5 .A restaurant is good if each guest finds a least three dishes 1

on the menu that he likes. When is a restaurant not good?
 178 7 Logic and proofs

2 E 7.6 . Formulate the negations. (For b) and c) let S1 ⊂ N and

S2 ⊂ N.)

a) All trees in this garden bear at least one apple.

b) For each n ∈ S1 there is m ∈ S1 such that m > n.
c) For each n ∈ S2 : if n is divisible by three then n is odd.

Construct sets S1 ⊂ N and S2 ⊂ N so that the propositions b) and c)

are true. Formulate a shorter proposition equivalent to b).

2 E 7.7 .For natural numbers m, n we write m|n (in words: m divides

n) if n is divisible by m. Formulate the following proposition in as
simple words as possible.

∀n ∈ N : (∃m ∈ N : 1 < m < n and m|n) ⇒

 
∃m ∈ N : 1 < m2 ≤ n and m|n

Is the proposition true? Formulate its negation formally and in words.

1-2 E 7.8 .Formulate the proposition ‘Each person has a secret’ using
quantifiers. Here a secret is a fact that no other person knows about.

1-2 E 7.9 .Let x, y ∈ N. Which of the following equivalences hold?

?
xy odd ⇔ x and y odd
?
xy even ⇔ x and y even
?
xy even ⇔ x or y even

2 E 7.10 .Decide for each proposition on the right whether it is neces-

sary and/or sufficient for the proposition to the left of it.
Here a, b are always real numbers.

x is a man. x is a mammal.
a<b ∃c ∈ R : a < c and c < b
a<b ∃c ∈ N : a < c and c < b
A figure consists of an The figure can be tiled by
even number of identical dominoes.
squares.
Exercises 179

E 7.11 . Which of the following propositions about a triangle T are 2

necessary, which are sufficient and which are both necessary and
sufficient for the truth of the proposition ‘D is equilateral’?
a) ‘T has two equal angles’
b) ‘All sidelengths of T are equal to 1’
c) ‘All angles of T are at least 60 degrees.’

E 7.12 .Let a, b be rational numbers satisfying a < b. Prove that there 1-2
is a rational number r such that a < r < b. Give a direct proof by
construction. (See also Exercise E 10.7 for the case where a, b are real
numbers.)

E 7.13 .Prove that among any seven numbers in {1, 2, . . . , 128} there 1-2
are x, y with x < y ≤ 2x.

E 7.14 .Find a necessary and sufficient condition on a natural number 2

v for there to be a graph having v vertices all of whose degrees are
odd.
Prove that your condition is necessary and sufficient.
√ √
E 7.15 .Because 2 − 1 = 0.41 . . . is less than 1, the powers ( 2 − 1)n 2
approach zero when n gets large. Use this and the general√binomial
theorem (see Exercise E 5.15) to give another proof that 2 is irra-
tional.
√
Also prove that m is irrational whenever m is a natural number
which is not a square.
8 Elementary number theory

Number theory is the mathematics of the integers: it is about divisi-

bility, prime numbers etc. Integers are very concrete objects that you
have played with since you were a small child. Therefore number the-
ory is very suitable for your exploration of mathematics, and you will
find number-theoretic problems in many places in this book. Number
theory has many faces: some of the hardest problems of mathematics,
still unsolved today, are stated in simple number-theoretic terms.
In this chapter we collect some of the basic definitions and theorems
of number theory. Most of the theorems will seem quite natural from
a few examples, and turning your number intuition into rigorous
proofs will be good exercise for you. You will also learn about the
notion of congruence, which magically turns some difficult problems
into easy exercises.

8.1 Divisibility, prime numbers and remainders

The most important concept of number theory is divisibility. What

does divisibility mean? For example, 15 is divisible by 3 because
dividing 15 by 3 leaves no remainder: 15 = 5 · 3. This leads to the
following definition.

Definition Let a, n ∈ Z. We write n| a if there is a number q ∈ Z

satisfying a = qn.

Some ways to express this are: n divides a, or n is a divisor of a, or a

is a multiple of n. For example, 1| a for all a ∈ Z, and 2| a means that
a is even.
Note that negative numbers and zero are permitted. For example,
the divisors of −5 are the numbers 1, −1, 5, −5. Zero is divisible by
every integer n since 0 · n = 0. Here are the most important rules for
divisibility:

© Springer International Publishing AG, part of Springer Nature 2018 181

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7_8
182 8 Elementary number theory

Lemma (Rules for divisibility) Let a, b, n ∈ Z.

a) If n| a and a|b then n|b.

b) If n| a and n|b then n| a + b and n| a − b.

Read n| a + b as n|( a + b), and similarly for n| a − b.

Proof.
This follows directly from the definition. Let us prove a):
n| a means that there is q ∈ Z with a = qn.
a|b means that there is q
∈ Z with b = q
a.
Plugging the first equation into the second we get b = q
(qn) = (q
q)n.
Because q
q is an integer, this implies n|b. The proof of b) is similar
(exercise). q. e. d.

Proof analysis: This was a direct proof. Why did we write q

in
the second line of the proof, not q as in the definition? Because
q was already in use.

Definition A number p ∈ N is called a prime number (or a

prime) if it has precisely two positive divisors.

For example, 2 has the positive divisors 1 and 2, hence is prime, but
4 has the positive divisors 1, 2 and 4, hence is not prime. 1 has only
one positive divisor (namely 1), hence is not prime. To phrase the
definition in this way (which excludes 1 from being prime) is only a
matter of convention. We will see below that it is a useful convention,
because only then is the prime factorisation unique.
Prime numbers have the following important property.

Lemma (Euclid’s lemma) Let a, b ∈ N and p a prime num-

ber. If p divides the product ab then p divides a or b:

p| ab ⇒ p| a or p|b
8.1 Divisibility, prime numbers and remainders 183

This is familiar for p = 2: If ab is even then a or b must be even. The

proof for general p is trickier than you might expect, see Exercise
E 8.12. Be careful: If p is not a prime then the statement is wrong.
For example, 4|2 · 6 but 4  2 and 4  6. Here n  a means that n does
not divide a.
The lemma implies: If p is a prime and if p divides a product
a1 · · · al then p divides one of the factors ai (proof by induction as
exercise).
The following theorem is so important that it is called the funda-
mental theorem of arithmetic.

Theorem (Prime factorisation) Every natural number bigger

than one has a prime factorisation. It is unique up to the order of
the factors.
This means:
Let n ∈ N, n > 1.
(i) There are m ∈ N and prime numbers p1 , . . . , pm with n =
p1 · · · p m .
(ii) If, in addition, q1 , . . . , ql are prime numbers satisfying n =
q1 · · · ql then l = m and the numbers q1 , . . . , qm are the same
as the numbers p1 , . . . , pm , up to reordering.

The same prime can occur multiple times. Then it appears as many
times among the pi as among the q j .
Examples: 2 = 2, 10 = 2 · 5 = 5 · 2, 999 = 3 · 3 · 3 · 37.
We have used the prime factorisation already in the solutions of
Problems 1.2 and 6.3. Only the following proof makes these solutions
complete.
Proof.
Since this is a statement about all natural numbers n > 1 it is natural
to look for an inductive proof. Try it yourself first!

Think about it!

184 8 Elementary number theory

Proof of (i) by induction on n:

Base case: n = 2 is a prime number, so we can take m = 1, p1 = 2.
Inductive hypothesis: Let n ∈ N, n > 2. Assume that (i) holds for all
natural numbers bigger than one and less than n.
Inductive step: If n is prime then we are done. If n is not a prime then
we can write n = ab where 1 < a, b < n. Applying the inductive
hypothesis we know that a, b are primes or products of primes. Then
n = ab is also a product of primes.
Proof of (ii) by induction on n:
Base case: The claim is obvious for n = 2.
Inductive hypothesis: Let n ∈ N, n > 2. Assume that (ii) holds for all
natural numbers bigger than one and less than n.
Inductive step: Let n = p1 · · · pm = q1 · · · ql . Then p1 divides the
product q1 · · · ql , so by Euclid’s lemma (or rather the remark after it)
p1 divides one of the factors qi . By reordering these factors we may
arrange that p1 |q1 . Because q1 is a prime, this implies p1 = q1 . Then
we can divide by p1 and obtain p2 · · · pm = q2 . . . ql . This number is
smaller than n, so we can apply the inductive hypothesis. It says that
m − 1 = l − 1, hence m = l, and that p2 , . . . , pm agree with q2 , . . . , qm
up to reordering. The claim follows. q. e. d.
We now discuss division with remainder.1 You know this from school.
For example: 20 divided by 3 equals 6 with a remainder of 2. This
means 20 = 6 · 3 + 2. Here it is essential that the remainder (2) is
smaller than the divisor (3). In general, it works like this:

Theorem (Division with remainder) Let a ∈ Z, n ∈ N. Then

there are unique numbers q, r ∈ Z satisfying

a = qn + r and 0 ≤ r < n .

Here q is called the quotient and r the remainder.

We also call r the remainder of a modulo n.

In words: a divided by n equals q with a remainder of r. The possible

remainders when dividing by n are
0, 1, . . . , n − 1.
1 This is also known as Euclidean division.
8.1 Divisibility, prime numbers and remainders 185

Proof.
Initial ideas: How do you think about division with remainder?
Maybe like this: You take as many n as you can fit into a; there are q
of them, and what remains is the remainder r. The remainder is less
than n because otherwise another n would fit into a. Therefore q and
r exist. They are unique, for if we used another q then the remainder
would be negative or bigger than n.
Now we just need to express this in mathematical notation. Try it
yourself!

Think about it!

We first prove the existence of q and r. Let q be the largest integer

satisfying qn ≤ a. Let r = a − qn. Then qn ≤ a implies r ≥ 0. Also
r < n since otherwise r ≥ n, hence a − (q + 1)n = a − qn − n =
r − n ≥ 0 and thus (q + 1)n ≤ a, which would contradict the choice
of q as largest integer satisfying qn ≤ a.
Now we prove the uniqueness of q and r. Let q and r be constructed
as above, and suppose that a = q
n + r
is any representation. If q
> q
then q
n > a because of the maximality of q, so r
< 0. On the other
hand, if q
< q then r
= a − q
n = qn + r − q
n = r + (q − q
)n ≥ n.
We see that whenever q
= q then r
does not satisfy the condition
0 ≤ r
< n. Therefore we must have q
= q and therefore r
= r. q. e. d.

Proof analysis: The theorem has two parts: Existence (“there are”)
and uniqueness (“unique”) of q, r. For the proof of existence
we gave a way to construct q, r. To prove the property r < n we
used an indirect proof.
Here it was natural to construct the number q using an extremal
property (the largest integer satisfying qn ≤ a). We will develop
this idea systematically in Chapter 10.
The proof of uniqueness is the direct translation of the initial
ideas into mathematical language. This was also an indirect
proof: We excluded the possibility of another representation
than the one we first constructed.
186 8 Elementary number theory

Note that a may be negative, but n must be positive.2 This will be

useful later on. Pay special attention if a is negative:

Example

What is the remainder when dividing -10 by 3? The answer is in

this3 footnote.

The practical use of remainders

? Problem 8.1

a) Suppose it is 11 o’clock now. What time will it be in 40 hours?

b) Suppose today is January 18, a Wednesday. Which day of the week will
February 18th be?

! Solution
a) After every 24 hours (one day) it will be the same time of day. So
we need to find the remainder of 11 + 40 = 51 modulo 24. That’s
three. The answer is 3 o’clock.
b) January has 31 days, so we have to advance by 31 days to get from
January 18 to February 18. After every 7 days (one week) it will be
the same day of the week. The remainder of 31 modulo 7 is 3, so
February 18 is a (Wednesday+3) = Saturday. !

8.2 Congruences

Dividing the integers into evens and odd numbers is useful for many
mathematical arguments.
√ You saw examples of this in Problem 6.3
and in the proof that 2 is irrational in Section 7.2.
Even numbers leave the remainder 0 modulo 2, odd numbers leave
the remainder 1. Instead of using remainders modulo 2 we could
2 Initially you probably thought about a > 0. Convince yourself that the proof is

also correct for a ≤ 0.

3 The remainder is 2, for −10 = (−4) · 3 + 2. You could also write −10 =

(−3) · 3 − 1, but the remainder must always be ≥ 0.

8.2 Congruences 187

classify integers according to their remainders modulo 3, or 4, or in

general modulo n. This turns out to be so useful that there is a special
notation for it. You will be astonished how a new notation, and a few
simple facts about it, make certain difficult problems easy.
The following simple fact will be useful.

Lemma Let a, b ∈ Z and n ∈ N. The following are equivalent:

(i) a and b leave the same remainder when divided by n.

(ii) n| a − b, that is, a and b differ by a multiple of n.

For example, 8 and 22 leave the same remainder when divided by

7 (namely 1) because their difference 22 − 8 = 14 is divisible by 7.
Conversely, the difference is divisible by 7 since the remainders are
equal.
Try to prove the lemma yourself.

Think about it!

Proof.
Let a, b ∈ Z and n ∈ N. Write (division by n with remainder)

a = qn + r, b = pn + s (8.1)

with quotients p, q ∈ Z and remainders r, s ∈ {0, 1, . . . , n − 1}.

Proof of (i) ⇒ (ii): If r = s then a − b = (qn + r ) − ( pn + r ) =

(q − p)n, so n| a − b since q − p ∈ Z.
Proof of (ii) ⇒ (i): n| a − b means that there is k ∈ Z satisfying
a − b = kn. Then kn = a − b = (q − p)n + (r − s) implies that r − s
is divisible by n. Because two different numbers r, s ∈ {0, . . . , n − 1}
have distance less than n it follows that r = s. q. e. d.
188 8 Elementary number theory

Proof analysis: This was a typical example of a proof of equiv-

alence. In order to prove the equivalence of the statements
(i) and (ii) about the numbers a, b, n, you fix a, b, n and prove
that (i ) ⇒ (ii ) and (ii ) ⇒ (i ). Both of these proofs were direct
proofs.

Equality of remainders is so important that there is a notation for it.

Definition Let a, b ∈ Z and n ∈ N. We write

a≡b mod n

(read: a is congruent to b modulo n) if n| a − b, or equivalently if

a and b leave the same remainder when divided by n.4

Example

22 ≡ 8 mod 7. Practically this means: In 22 days we have the

same day of the week as in 8 days.
That’s which day of the week, supposing today is a Wednesday?
Because 8 ≡ 1 mod 7 and tomorrow (in 1 day) is a Thursday, it
must be Thursday in 8 and in 22 days.
The numbers a, b may well be negative. For example, −6 ≡ 1
mod 7 because −6 − 1 = −7 is divisible by 7: six days ago it was
also a Thursday.

The definition implies that we always have a ≡ r mod n if r is the

remainder of a modulo n. Conversely the remainder of a modulo n is
the only number r ∈ {0, . . . , n − 1} which satisfies a ≡ r mod n.
You can calculate with congruences:

4 This notion of congruence is not related to the geometric notion of congruence

of two figures in the plane or in space, which means that one figure can be turned
into the other by means of a translation, rotation or reflection.
8.2 Congruences 189

Theorem (Calculation rules for congruences) Let n ∈ N.

Congruences mod n can be added, subtracted, multiplied and
exponentiated. That is: If

a≡b mod n
c≡d mod n

then
a+c ≡ b+d mod n
a−c ≡ b−d mod n
ac ≡ bd mod n
ak ≡ bk mod n
for all k ∈ N.

Be careful! You cannot divide congruences. For example, we have

2 ≡ 6 mod 4 and 2 ≡ 2 mod 4, but 22 ≡ 62 mod 4. See also Exercise
E 8.11.
Proof.
By definition a ≡ b, c ≡ d mod n mean that n| a − b and n|c − d. Then

n| a − b, n|c − d ⇒ n|( a − b) + (c − d) = ( a + c) − (b + d)

which gives the first claim, and the second claim follows similarly.
Now try to find a proof of the third claim (multiplication) yourself!

Think about it!

For the third claim write

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

Now n| a − b ⇒ n|( a − b)c and n|c − d ⇒ n|b(c − d), and adding

implies n|( a − b)c + b(c − d) = ac − bd, so ac ≡ bd mod n.
If you use the third claim with a = c, b = d then you get a2 ≡ b2
mod n. Applying the same idea repeatedly you get ak ≡ bk mod n
for all k ∈ N. q. e. d.
190 8 Elementary number theory

n divides
n dvides
( a − b)c n divides
a−b
and ac − bd
and c − d
b(c − d) goal
given
connector

Figure 8.1 Solution scheme for the proof analysis

Proof analysis: This was a direct proof, and the main point was
skilful calculations.
The transformations used in the proof for multiplication seem
difficult to find. How on earth can you get this idea? This is
a typical problem-solving situation. What is our goal? That
ac − bd is divisible by n. What is given? That a − b and c − d
are divisible by n. In short: Goal: ac − bd. Data: a − b, c − d.
How can we connect the goal with the data? The expression
ac − bd looks complicated, so let’s begin with one part of it, the
term ac. How can we connect this with the data? One way
to do this is through the expression ( a − b)c = ac − bc since
it has both a − b and ac in it. What do we need to get from
here to our goal? To get from ac − bc to ac − bd we need to
add bc − bd, which equals b(c − d). Summarizing, we have
ac − bd = ( a − b)c + b(c − d). Figure 8.1 shows the solution
scheme; it is an example of the general scheme in Figure 6.4.5

5 This
type of calculation, which is also called “clever addition of zero” (because
we added 0 = −bc + bc between ac and bd) can be found in various places in
mathematics. For example, in analysis, we add a clever form of zero when proving
that the product of two convergent sequences converges to the product of their
limits.
Exercises 191

Examples

J What is the remainder of 100 modulo 7? You could calculate

it directly, but this way is easier: From 10 ≡ 3 mod 7 we get
102 ≡ 32 mod 7, so

100 = 102 ≡ 32 = 9 ≡ 2 mod 7.

So the remainder of 100 modulo 7 is 2. Cross-check: 100 =

14 · 7 + 2.

J What is the remainder of 2100 modulo 3? This would be very

tedious to calculate directly. Here is a simple method: From
22 = 4 ≡ 1 mod 3 we get

2100 = 22·50 = (22 )50 ≡ 150 = 1 mod 3 .

So the remainder is 1.

J What is the last digit of the number 777 ? The last digit of
a natural number (in the decimal system) is its remainder
modulo 10. Let us try smaller exponents first: 72 = 49 ≡
9 ≡ −1 mod 10. This is easy to exponentiate, for example
776 = 72·38 = (72 )38 ≡ (−1)38 = 1 mod 7, hence

777 = 776 · 7 ≡ 1 · 7 = 7 mod 10 .

So the last digit of 777 is 7.

Exercises

E 8.1 .If January 18, 2012 is a Wednesday, which day of the week is 1
January 18, 2013? Formulate your solution using congruences. (Note
that 2012 is a leap year.)

E 8.2 .Explain how the well-known rule “even + even = even” can 1
be reduced to the equation 0 + 0 = 0 using congruences. Which
equations correspond to the rules “even + odd = odd”, “odd·odd =
odd” etc.?
 192 8 Elementary number theory

1-2 E 8.3 .Prove 7|3105 + 4105 .

2 E 8.4 .We write lg 2 for the logarithm of 2 to the base 10, that is, the
number which satisfies 10lg 2 = 2. Prove that lg 2 is irrational.

1-2 E 8.5 .Prove that 6  n3 + 5n + 1 for all n ∈ Z.

3 E 8.6 .Prove the following divisibility rules using congruences. All

numbers are represented in the decimal system.
a) A natural number is divisible by 3 (resp. 9) if and only if the sum
of its digits is divisible by 3 (resp. 9).
b) A natural number n is divisible by 7 if and only if the number n

obtained as follows is divisible by 7: Let a0 be the last digit of n,

so n = 10a + a0 where a0 ∈ {0, . . . , 9}. Then let n
= a − 2a0 .

3 E 8.7 .Find and prove a divisibility rule for division by 11.

3-4 E 8.8 .Find more divisibility rules in the decimal system, for example
for division by 13 or 17. Find a divisibility rule for division by 7 in
the octal number system (number system with base 8).

2 E 8.9 .Let n ∈ N. Imagine writing down the remainders modulo n

of 12 , 22 , . . . , (n − 1)2 . Are they all different? Why not?

2-3 E 8.10 .(Quadratic residues)

a) Produce for each number n = 3, 4, 5 a table in which each number
a = 0, 1, . . . , n − 1 is followed by the remainder of a2 modulo n.
(These remainders are called quadratic residues modulo n.)
b) Conclude that every square number has remainder 0 or 1 when
divided by 4. What are the possible remainders modulo 3 and
modulo 5?
c) Which remainders can the sum of two square numbers have mod-
ulo 4?
d) Are there square numbers of the form 3n − 1, 5n + 4, 7n + 3 (with
n ∈ N)?
e) Can 999,999 be written as the sum of two squares?
f) Prove: if 5| a2 + b2 + c2 with a, b, c ∈ N then 5| a or 5|b or 5|c.
Exercises 193

E 8.11 .Two numbers n, c ∈ N are called coprime (or relatively prime) 2

if 1 is the only natural number which divides both n and c.
a) How can you read off from the prime factorisations of n and c
whether they are coprime?
b) Prove: if n, a, c ∈ N and n, c are coprime then n| ac implies n| a.
c) Prove for n, a, b, c ∈ N satisfying c| a, c|b:

a b
a≡b mod n ⇒ ≡ mod n, if n, c are coprime.
c c

E 8.12 . The aim of this exercise is to prove Euclid’s lemma. Be 3-4

careful: in the proof we cannot use the uniqueness of the prime
factorisation, since we used Euclid’s lemma to prove it.
a) (The case p = 2) Prove: if ab is even and b is odd then a is even.
Argue carefully.
b) (The case a < p) Prove: if p is a prime number and if a, b ∈ N are
such that p  b (p does not divide b) and a < p then p  ab.
c) Prove Euclid’s lemma using b).

E 8.13 . Consider the following proposition and the given (wrong) 2

“proof”.
Lemma. Every power 6k with k ∈ N ends in the digit 6 when
written in the decimal system.

Proof. “Proof”
In order to prove the claim, we show

∀k ∈ N : 6k ≡ 6 (mod 10).

We will show this by contradiction.

Assumption: 6k ≡ 6 (mod 10) for all k ∈ N.
If 6k ≡ 6 (mod 10) for each k ∈ N then this is true for k = 2 in
particular. However,

62 = 36 ≡ 6 (mod 10).

This is a contradiction to the assumption, so the assumption must

have been false. This proves the lemma.
 194 8 Elementary number theory

Why is the proof wrong? Give a correct proof.

1-2 E 8.14 . (Inverse of Euclid’s lemma) Let p ∈ N, p ≥ 2 be such that

for all a, b ∈ N the following is true: If p| ab then p| a or p|b.
Show that then p is a prime number.

2 E 8.15 .Prove: the number of positive divisors of a natural number n

is odd if and only if n is a square number.

3 E 8.16 .Let m, n ∈ N be coprime. Calculate

m m m
  + 2  + · · · + (n − 1) 
n n n
where  x  is x ‘rounded up’, for example 2.4 = 3, 5 = 5.

2 E 8.17 .Let a be a natural number. Decide for each of the propositions

a) to d) below whether it implies the proposition

a≡1 (mod 4) (8.2)

or whether it is implied by this proposition.

a) a is odd.
b) a is odd and can be represented as the sum of two positive
square numbers.
c) a2 ≡ 1 (mod 8)
d) 2a ≡ 2 (mod 4)
Give a proof or a counterexample for each answer.
9 The pigeonhole principle

If many pigeons roost in few pigeonholes then there must be a hole

with many pigeons.1 This is so obvious that you may be surprised
how many mathematical arguments are based on such reasoning, and
what surprising consequences it has. It is an important tool for proofs
of existence. The art is in recognising when and how it can be used.
This is sometimes quite obvious, but sometimes hard to discern.
In this chapter we start with a precise statement of the pigeonhole
principle and give a few examples from everyday life. Turning to
mathematics, you will see how you can deduce some interesting
facts about integers by considering remainders as pigeonholes. Then
we will investigate the question how well any real number can be
approximated by fractions, and see how the pigeonhole principle can
be useful in a quite non-obvious way. Finally we apply the principle
to a problem from graph theory and find that even in the greatest
chaos there must be some order.

9.1 The pigeonhole principle, first examples

We start with the simplest version of the pigeonhole principle:

Pigeonhole principle

Let n ∈ N. If n + 1 pigeons roost in n holes then there must be a

hole containing more than one pigeon.

This statement is optimal in the following sense: if there are only n

pigeons in n holes then it is possible that there is only one pigeon in
each hole.

1 Nowadays when speaking of pigeonholes you usually think of open mail-boxes.

So instead of pigeons you may think of envelopes if you wish. But it’s more fun to
think of birds, so we will stick to the original meaning.

© Springer International Publishing AG, part of Springer Nature 2018 195

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7_9
196 9 The pigeonhole principle

Proof.
If each hole contained at most one pigeon then there would be at
most n pigeons in total. q. e. d.

Proof analysis: This is an indirect proof.

If you replace pigeons and holes suitably then you get interesting
facts.

Examples

J Among any 13 people there are two whose birthdays are in

the same month.
(pigeons: people; holes: months)
Important: ‘there are two’ always means: ‘there are at least
two’; there could also be three or more people having their
birthdays in the same month.

J Among any three people there are two of the same sex.
(pigeons: people; holes: sexes)

J Among the inhabitants of Liverpool there are two with the

same number of hairs on their head.2
(pigeons: inhabitants of Liverpool; holes: possible numbers of
hairs, i.e. the numbers 0, 1, 2, . . . , 300, 000. Put an inhabitant
in hole i if he/she has i hairs on his/her head.)

The pigeonhole principle is very effective for proofs of existence: in

the last example you know readily that there are two inhabitants with
the same number of hairs. If you wanted to know who they are, you
would need to count the hairs of many people – an almost impossible
task. In this sense the pigeonhole principle is non-constructive.
Often we need a more general version of the pigeonhole principle.

2 For this you need to know: 1. Every human has at most 300,000 hairs on

his head (usually it’s about 150,000). 2. the city of Liverpool has about 466,000
inhabitants (data from 2013).
9.1 The pigeonhole principle, first examples 197

General pigeonhole principle

Let a, n ∈ N. If an + 1 pigeons roost in n holes then there must

be a hole containing more than a pigeons.

Again this is optimal: it would not be true with an or fewer pigeons.

Proof.
If each hole contained at most a pigeons then there would be at most
an pigeons in total. q. e. d.

Examples

J There are 13 gifts under the Christmas tree, for 3 children.

Then at least one child gets at least 5 gifts.
(pigeons: gifts; holes: children. The argument is: if every child
got at most 4 gifts then there would be at most 3 · 4 = 12 gifts
in total. Here n = 3 and a = 4.)

J Among the inhabitants of England there are at least 170 with

the same number of hairs on their heads.3
(There are more than 51, 000, 170 = 170 · 300, 001 inhabitants,
so it cannot be that each number of hairs occurs only at most
170 times.)

In these examples it was easy to see how to use the pigeonhole

principle. In the following problems you need to think carefully
about what you could take as pigeons and what as holes.
A useful rule of thumb is: the objects whose (multiple) existence
you want to prove are the pigeons, their properties are the holes.

? Problem 9.1
Some people are in a room. Show that there are two among them who are
acquainted with the same number of people in the room.4
3 By the 2011 census England has about 53,000,000 inhabitants.
4 We assume that the ‘acquaintance’ relation is symmetric, that is: if A is ac-
quainted with B then B is acquainted with A.
198 9 The pigeonhole principle

 Investigation
G
 Let n ≥ 2 be the number of people in the room. The statement
‘there are two . . . ’ suggests that we use the pigeonhole principle.
What could be the pigeons, what the holes? We want to prove
the existence of two people, the relevant property is the number of
acquaintances. So the pigeons should be the people, the holes the
possible numbers.
 Which holes are there? A person in the room can have 0, 1, . . . , or
n − 1 acquaintances in the room. These numbers are the holes.
 So we have n pigeons and n holes. The pigeonhole principle is not
applicable. What now?
 Look again: if one person knows everyone else (thus has n − 1
acquaintances) then there can be no person having 0 acquaintances.
So either hole 0 or hole n − 1 is empty.
 Therefore only n − 1 holes can be used, so one of them must contain
2 pigeons. G
You could write this up as follows:

! Solution of Problem 9.1

Let n ≥ 2 be the number of people in the room. We number them
1, . . . , n. Let bi be the number of acquaintances of person i. Then
bi ∈ {0, 1, . . . , n − 1} for each i. If there is a person i with bi = n − 1
(a ‘socialite’) then she knows everyone else, so there can be no person
j having b j = 0. Therefore at most n − 1 different numbers appear
among b1 , . . . , bn , so there must be i = j with bi = b j . !

Remark

If you represent the people as vertices of a graph and the acquain-

tanceships as edges then you can state this result in the language
of graphs:
In any graph5 there are two vertices having the same degree.
9.2 Remainders as pigeonholes 199

9.2 Remainders as pigeonholes

Sometimes it is useful to use remainders modulo n as pigeonholes.

Here is a simple example.

? Problem 9.2
Given 10 natural numbers, show that there are two among them whose
difference is divisible by 9.
The goal is divisibility by 9, this suggests that we use remainders
modulo 9 as holes.

! Solution
There are 9 possible remainders when dividing by 9: 0, 1, . . . , 8. There-
fore two of the 10 given numbers have the same remainder modulo 9.
Their difference is divisible by 9. !
The analogous statement with 10 replacing 9 would be: among 11
natural numbers there are two that end in the same digit. This is
obvious: there are only 10 possible digits. Note that the last digit is
the remainder modulo 10.
The pigeonhole principle can also be useful for proving the exis-
tence of a single object with certain properties:

? Problem 9.3
Prove that among the powers 7, 72 , 73 , . . . there is one ending in 001.

 Investigation
G
What are we looking for? Three final digits. That’s the same as
the remainder modulo 1000. So we could try to use the pigeonhole
principle where the holes are the remainders modulo 1000. What
should be the pigeons? The objects whose existence we want to
prove: the powers of 7. So we get two powers of 7 having the same
remainder modulo 1000. How can we use this for the problem?

5 with at least two vertices, and without double edges or loops

200 9 The pigeonhole principle

Think about it!

We proceed step by step. Let us write it down: 7k ≡ 7l mod 1000, or

equivalently 1000|7k − 7l for two different natural numbers k, l. How
can we use this? We could take out the smaller power: let’s say k > l,
then 7k = 7l · 7k−l , so 7k − 7l = 7l · (7k−l − 1). If this is divisible by
1000 then 7k−l − 1 must also be divisible by 1000, so 7k−l ends in 001,
done! G
This was almost perfect, but let’s write it down with all details.

! Solution of Problem 9.3

A number ends in 001 if and only if it has the remainder 1 modulo
1000. Therefore we consider the remainders modulo 1000 as holes.
There are 1000 remainders: 0, 1, . . . , 999. Therefore, among the 1001
numbers
71 , 72 , 73 , . . . , 71001
there must be two, say 7k and 7l where k > l, having the same
remainder. Then

1000|7k − 7l = 7l (7k−l − 1).

Now 1000|7l (7k−l − 1) implies 1000|7k−l − 1 since the prime factorisa-

tion of 7l (7k−l − 1) must contain 1000 = 23 · 53 , and since 7 is prime,
this can only be contained in the prime factorisation of 7k−l − 1. There-
fore, 7k−l leaves the remainder 1 modulo 1000, so it ends in 001. !

 Review of Problem 9.3

The problem suggested using remainders modulo 1000 as holes and
powers of 7 as pigeons. It was less clear what use it would be to have
two powers with the same remainder. What saved us was the fact
that 7k = 7l · 7k−l . 
9.3 An exploration: approximation by fractions 201

The proof shows that among the numbers 7, 72 , . . . , 71000 there must
already be one ending in 001.
An analogous statement holds for any number other than 7 which
is not divisible by either 2 or 5. And instead of 001 we can require n
zeroes followed by 1, for any n. The proof is the same.

9.3 An exploration: approximation by fractions

We want to explore the following question.

Approximation problem: How well can an arbitrary real number

be approximated by fractions?

Your first reaction might be: what a strange question, of course any
real number a can be approximated arbitrarily well by fractions:6 Just
cut off the decimal representation of a after a few digits. For example,
π = 3.1415 . . . can be approximated by 3.141 = 3141/1000 with an error
of 0.0005 . . . . If you want a better approximation then simply use
more digits.7
We could stop here: problem solved, exploration finished. Or
we could try to discover more, taking our vague question only as a
starting point.
Let us stick to the example π = 3.1415926 . . . . If the fraction m/n is
to be close to π then nπ should be close to an integer. So let us look
at some multiples of π and their distance to the nearest integer, see
Table 9.1:
Surprise: 7π is much closer to an integer (22) than all the other
listed multiples of π. So 22/7 will be a very good approximation of
π, considering that the denominator is quite small. We calculate
22/7 = 3.142 . . . and see that the error is about 0.001. Compare this

with our first idea of cutting off the decimal expansion. Table 9.2
shows some of these fractions and their approximation errors. It also
shows another very good approximation, 355/133. What is remarkable
about this table? The fractions 22/7 and 314/100 approximate π about
equally well, but 22/7 has a much smaller numerator and denominator.

6 By fraction we always mean a fraction of integers; numbers that can be written

as fraction of integers are called rational. In what follows we do not distinguish
between a fraction and the rational number represented by it.
7 Maybe you are also wondering: Why is this question in the chapter on the

pigeonhole principle? Just wait. Let the question, not the method, be our guide.
202 9 The pigeonhole principle

n nπ (rounded) distance
1 3.14 0.14
2 6.28 0.28
3 9.42 0.42
4 12.57 0.43
5 15.71 0.29
6 18.85 0.15
7 21.99 0.01
8 23.13 0.13
9 28.27 0.27
10 31.42 0.42

Table 9.1 Distance of nπ to the nearest integer, rounded to 2 digits

fraction error
31/10 = 3.1 0.04 . . .
314/100 = 3.14 0.001 . . .
3141/1000 = 3.141 0.0005 . . .
22/7 = 3.1428 . . . 0.001 . . .
355/113 = 3.1415929 . . . 0.0000003 . . .

Table 9.2 Approximations of π by fractions

This is even more extreme with 355/113. Although the denominator of

this fraction is only a little bigger than that of 314/100, it gives a much
better approximation.
We see that interesting things happen if we take the size of the
denominator seriously in our investigation.8 So we make our problem
more specific:

Approximation problem, version 2: Investigate how the approx-

imation error is related to the size of the denominator when we
approximate a real number a by fractions.

So far we have only considered the example a = π. A special property

of this number is that it is irrational, that is, it cannot be written
8 Instead
we could take the size of the numerator. This would not make much of
a difference since a ≈ m/n implies m ≈ an.
9.3 An exploration: approximation by fractions 203

precisely as a fraction.9 But even for rational numbers a the problem

is interesting. For example, consider a = 0.51: Of course this is
precisely 51/100, but you can approximate it by the much simpler
fraction 1/2, up to the quite small error 0.01.
Our question also has practical relevance: suppose you want to
construct a transmission – for example, on a bicycle – with two gears
which has a transmission ratio close to a given value a. You want
your gears to have few teeth, in order to save money. If the gears have
m and n teeth then the transmission ratio will be m/n. So you will
need to balance the cost (which depends on m + n ≈ ( a + 1)n, so on
n) against the error | a − m/n|, and this is precisely our problem above.
Version 2 of the problem is still a little vague. Let us state a few
concrete questions. First we fix some notation:

J a: a real number, assumed positive for simplicity

J n ∈ N: the denominator in an approximating fraction
J ε > 0: the approximation error10

We will denote the following proposition by Approx( a, n, ε).

‘ a can be approximated by a fraction of denominator n

with an error of at most ε ’
 
That is: ‘There is an m ∈ N0 satisfying  a − mn  ≤ ε ’
Here are a few questions that we could ask.

Approximation problem, version 3:

1. Given a and n, what is the minimal ε for which Approx( a, n, ε)
is true?
2. Given a, are there certain values of n for which Approx( a, n, ε) is
true with very small ε?
3. Given n, what is the minimal ε for which Approx( a, n, ε) holds
for all a?

9 This is not so easy to prove. See (Hardy and Wright, 2008) for example.
10 The Greek letters ε, δ (epsilon and delta) are frequently used in mathematics
for small positive numbers.
204 9 The pigeonhole principle

Maybe you can think of more questions?

Note the difference between the third question and the other two:
In 1. and 2. the number a is fixed at the beginning, and ε may depend
on a. In 3. the integer n is fixed, and we are asking for an ε which
works for all a. In both cases ε will depend on n.
To understand what happens we make a sketch, see Figure 9.1. We
draw the points 0, n1 , n2 , . . . on the real line. Then a will lie between
two such points, or coincide with one of them.

a
0 1 2 m m +1
n n n n

m
Figure 9.1 Approximation of a by the fraction n

We can read off that ε = 12 · n1 = 2n 1

always works, but also that
smaller values, even ε = 0, might work, depending on a. This is all
we can really say on Question 1.
Now let us check Question 3: which ε will work for all a? As we
saw, ε = 2n 1
works for all a; but if a lies half way between two of the
marked points, for example if a = 2n 1
, then no smaller value of ε will
work for it. So the answer to Question 3 is: the smallest such ε is
ε = 2n1
.
Question 2 is much more interesting, as we saw in the case a = π.
It is still not a very precise question. What does “very small ε” mean?
Of course very large values of n will allow very good approximations,
so ε will depend on n. We saw that ε = 2n 1
works for all n. But in the
case a = π the approximation errors of /7 and of 355/113 are much
22

smaller than 1/14 and 1/226, respectively. Can we find such especially
good approximations for any a? How good can they possibly be?
We found the good approximation 22/7 of π by considering the dis-
tance of nπ to the nearest integer. So let us investigate this in general:
how small can we make this distance when we allow numbers n up
to a certain size?

? Problem 9.4
Let N ∈ N. Find the smallest number δ, depending on N, having the
following property: for any real number a > 0 at least one of the numbers
a, 2a, . . . , Na lies within distance δ of an integer.
9.3 An exploration: approximation by fractions 205

This problem seems to be closer to Question 3 than to Question 2

since the same δ is to work for all a. However, we will see that the
solution will help us in investigating Question 2. Clarify for yourself
the difference between this problem and Question 3.

 Investigation
G
 We need to understand the problem well. The statement is quite
complicated: . . . smallest . . . any . . . at least one . . . . Look at each
of these words to understand what we are looking for. Consider
some examples of a and N and find the best possible δ.
 We look at a special case: what is the answer for N = 1? For which
δ is it true that any a lies within δ of an integer?
Clearly this is true for δ = 1/2 but not for smaller values of δ (take
a = 1/2). So the answer for N = 1 is δ = 1/2.
 How about N = 2? Now we need to consider both a and 2a. Play
with this, try out different a, try to construct the worst case.

Think about it!

 It is time to introduce some new notation. For x ∈ R, x ≥ 0

denote by frac( x ) the fractional part of x.11 Here is an exact defini-
tion.

Definition For x ∈ R, x ≥ 0 let  x  (Gauss bracket of x) be

the largest integer which is less than or equal to x, and let
frac( x ) = x −  x .

Examples: 3.1 = 3 and frac(3.1) = 0.1. 5.9 = 5 and frac(5.9) =

0.9. 3 = 3 and frac(3) = 0.
We always have 0 ≤ frac( x ) < 1.

11 Insteadof frac( x ) the notation { x } is often used. We use frac( x ) in order to

avoid confusion with the set having x as its only element.
206 9 The pigeonhole principle

 The fractional part of a real number determines its distance to the

nearest integer. Therefore we consider frac( a), frac(2a). Maybe you
noticed that δ cannot be smaller than 1/3 (for a = 1/3 or a = 2/3).
So we conjecture that for N = 2 the optimal δ is δ = 1/3. Let us try
to show this.
 We want to show: for any a at least one of the numbers frac( a),
frac(2a) lies in one of the outer intervals, shown thicker, in Figure
9.2. Here we include the endpoints 1/3, 2/3 in the intervals.

0 1 2 1
3 3

Figure 9.2 Three thirds for frac( a), frac(2a)

Think about it!

 What could we conclude if this was wrong? (We try a proof by

contradiction.) Then both frac( a), frac(2a) would have to lie in the
middle third. Then the distance between them would be at most
1/3.12

 But if the distance between frac(2a) and frac( a) is at most 1/3 then
2a − a = a must lie within 1/3 of an integer, so frac( a) would lie in
one of the outer thirds. Contradiction!
 We need to check something here: is it true that the distance from
b − a to the nearest integer is at most | frac(b) − frac( a)|? (In our
case b = 2a.) This looks reasonable, but let’s postpone the details
and first consider general N.
Try to generalize the argument to general N. What could be the
optimal δ? How could you prove that? If you are not ready for the
general case yet, try N = 3 first.

12 In fact it is less than 1/3, but this does not matter for the proof.
9.3 An exploration: approximation by fractions 207

Think about it!

 For N = 3 we divide the interval [0, 1] into four quarters. If none

of the three numbers frac( a), frac(2a), frac(3a) lies in the leftmost
or rightmost quarter then they must all lie in the two remaining
quarters, so by the pigeonhole principle two of them must lie in the
same quarter. Then their distance would be less than 1/4, and we
get a contradiction as before. Therefore δ = 1/4 works. A smaller δ
would not work since for a = 1/4 each of the numbers 1/4, 2/4, 3/4
has distance ≥ 1/4 to the nearest integer.
 For N = 1 we got δ = 1/2, for N = 2 we got δ = 1/3 and for
N = 3 we obtained δ = 1/4. This cries out for a conjecture: the best
possible δ is 1/( N +1). G

! Solution of Problem 9.4

Claim: The smallest δ in Problem 9.4 is δ = 1
N +1 .

For the proof we need a lemma.

Lemma Let x, y ∈ R, x, y ≥ 0. Then |frac( x ) − frac(y)| ≤ δ

implies that x − y lies within δ of an integer.

Proof.
Write x =  x  + frac( x ), y = y + frac(y). Then

x − y = ( x  − y) + (frac( x ) − frac(y)) .

The first parenthesis on the right is an integer, the second one has
absolute value less than or equal to δ by assumption. The lemma
follows. q. e. d.

We now prove the claim above, which solves Problem 9.4.

208 9 The pigeonhole principle

Proof.
For a = N1+1 we get the numbers N1+1 , . . . , NN+1 . Each of them differs
by at least N1+1 from the nearest integer. Therefore we must have
δ ≥ N1+1 .
Now let δ = N1+1 . Let a > 0 be arbitrary. We show that at least one
of the numbers a, 2a, . . . , Na lies within δ of an integer. Consider the
N + 1 intervals13

I0 = [0, δ], I1 = [δ, 2δ], . . . , IN = [ Nδ, 1].

Each interval has length δ, including the last one since 1 = ( N + 1)δ.
Suppose none of the numbers na, n = 1, 2, . . . , N lies within δ of
an integer. Then none of the numbers frac(na) lies in I0 or in IN .
Therefore, all of these N numbers lie in the remaining N − 1 inter-
vals. By the pigeonhole principle at least two of them, say frac(ka)
and frac(la) with k > l, must lie in the same interval. This im-
plies | frac(ka) − frac(la)| ≤ δ. Then the lemma implies that ka − la
lies within δ of an integer. Now k, l ∈ {1, . . . , N }, k > l imply
k − l ∈ {1, . . . , N − 1}, so ka − la = (k − l ) a is one of the numbers
a, 2a, . . . , Na. This contradicts our assumption. Therefore the assump-
tion was wrong and the claim is proven. q. e. d.
!
Proof analysis: The proof that δ = N1+1 works was a proof
by contradiction: the assumption that the claim is wrong for
some a led to a contradiction. The pigeonhole principle could
be applied since from two numbers in the same ‘hole’ a new
number (their difference) could be constructed, which provided
the contradiction.
In order to show that smaller values of δ don’t work it sufficed
to give an example a. (Recall that the negation of a ‘for all’
proposition is a ‘there exists’ proposition.)

Let us return to Question 2 in the third version of the approximation

problem: given a, are there
 certain special values of n and correspond-
ing m for which  a − mn  < ε holds with very small ε? We already
13 It is irrelevant that some of the intervals intersect at their endpoints. What

matters is that their union contains the interval [0, 1), the set of possible values of
frac( a).
9.3 An exploration: approximation by fractions 209

know that ε = 2n 1
always works, for any n; but we are aiming for
much smaller ε, generalizing the example of a = π and mn = 22 7 . We
begin by restating the solution of Problem 9.4 as follows: Let a > 0.
14

Given N ∈ N there are n ∈ N, n ≤ N, and m ∈ N0 with

1 (9.1)
|na − m| ≤ .
N+1
 
Dividing by n we get  a − mn  ≤ n( N1+1) . For N ≥ 2 the error 1
n ( N +1)
1
is already better than the error 2n . Still this is not quite satisfactory
since we would like to express (or estimate) the error in terms of the
denominator n. That’s easy: From n ≤ N we get N1+1 ≤ n+ 1
1 and then
 m 
 1
 a − ≤ . (9.2)
n n ( n + 1)

But this isn’t quite what we wanted either, since all we know is that
there is some n (and then a corresponding m) for which this inequality
is true. If n was, say, 2 then that’s not a very good approximation!
However, there is a better way to use (9.1), which yields:

Theorem Let a > 0 be irrational. Then there are infinitely

m
many fractions n for which inequality (9.2) holds.

Proof.
We show that given any finite set of fractions there is another fraction
which is not in the set and which satisfies inequality (9.2). Starting
with the empty set we get arbitrarily many fractions satisfying the
inequality. Therefore there cannot be only a finite number of them.
Thus, let M be a finite set of fractions. For each of these fractions
n consider the number | na − m |. It is positive, not zero, since a is
m

irrational. So we get finitely many positive numbers. Let δ0 be the

smallest among them. (If M is the empty set then choose δ0 to be
any positive number.) Now we choose a natural number N satisfy-
ing N1+1 < δ0 and then, using (9.1), find n
≤ N and m
satisfying
|n
a − m
| ≤ N1+1 . We use this inequality in two ways: First, it implies
14 This is known as Dirichlet’s approximation theorem.
210 9 The pigeonhole principle

|n
a − m
| < δ0 , hence ∈ M by definition
n
 of δ0 . Second, we divide
 m

by n
,
and using n
≤ N we get  a − n
 ≤ n
(n1
+1) , which was to be
shown. q. e. d.

Proof analysis: This was a direct proof. When constructing

another fraction the extremal principle was useful (‘Let δ0 be
the smallest . . . ’). We will introduce this principle systematically
in Chapter 10.

What does the theorem buy us? First, note that for each n ≥ 2
there can be at most one m satisfying (9.2) (why?). Therefore, the
theorem implies that the estimate is true for infinitely many n, hence
for any given ε > 0 we can find such n with n(n1+1) < ε, and then
Approx( a, n, ε) holds.
But there is much more to it: the estimate (9.2) is much better than
the approximation obtained by cutting off the decimal expansion.
If you cut off after the kth digit after the decimal point, then the
approximating fraction has the form 10mk , and the error is at least 101k+1
unless the (k + 1)st digit happens to be zero. So here we have n = 10k ,
1
and the error is of the order of 10n . For k > 1 this is much bigger than
1
the error n(n+1) in (9.2): The former has denominator linear in n and
the latter quadratic.

More remarks on the approximation problem

J The statement of our last theorem also holds for rational a but
p
is uninteresting then: write a = q and choose m = kp, n = kq
for k = 1, 2, . . . . However, the question of good approximation by
fractions with small denominators remains interesting for rational
numbers, as we saw. The theory of continued fractions gives
an effective algorithm for finding such fractions (in the case of
rational or irrational a). See for example (Hardy and Wright, 2008)
or (Silverman, 2012).

J The approximation (9.1) and the subsequent theorem hold for all
a > 0. However, for certain numbers a there are much better ap-
proximations. For example, for some a the error n(n1+1) in (9.2) can
be replaced by n13 , which means a much better approximation for
large n. On the other hand it is known that such an improvement
9.3 An exploration: approximation by fractions 211

√
1+ 5
is not possible for the golden ratio, a = 2 . So this number is
only badly approximable.
This circle of questions is called diophantine approximation. It
plays an important role in the theory of dynamical systems. For
example, around 1900 Poincaré was led to such questions when
investigating the motion of the planets.

J The gap problem: The fact stated in Problem 9.4 has other inter-
esting consequences. Consider Figure 9.3. A goblin walks on a
circular path of length 1 with constant step size a. There is a gap
in the path of length ε > 0. Then the following is true: if a is
irrational then the goblin will fall into the gap at some point – no
matter how small ε is and where he starts.15

ε
a

Figure 9.3 If a is irrational then the goblin will fall into the gap eventually

Proof.
Mark the starting point by 0 and measure lengths in the walking di-
rection. Since the path has length 1 the steps are at frac( a), frac(2a),
frac(3a), . . . . Choose N ∈ N satisfying N1 < ε. By Problem 9.4
there is an n ≤ N such that frac(na) < N1 or frac(na) > 1 − N1 . We
consider the first case, the argument in the second case is similar.
Let b = frac(na). Since a is irrational we have b > 0. The first n
steps move the goblin from 0 to b. After another n steps it will
have moved to 2b, after the next n steps to 3b etc. Now |b| < N1 < ε,
so one these numbers b, 2b, . . . falls into the gap. q. e. d.

In mathematical language we have proved: Let a be irrational.

Then for each interval [ x, y] ⊂ [0, 1], where x < y, there is a k ∈ N
so that frac(ka) ∈ [ x, y]. This is usually expressed as follows:

15 We assume that the goblin’s feet are infinitely thin.

212 9 The pigeonhole principle

Theorem Assume a is irrational. Then the set of numbers

frac(ka), k ∈ N, is dense in [0, 1].

Convince yourself that this is not true if a is a rational number.

The sequence frac(ka), k = 1, 2, 3, . . . , is a simple example of an
ergodic dynamical system.
In Exercise E 9.15 you find a pretty application to powers of inte-
gers.

9.4 Order in chaos: the pigeonhole principle in

graph theory

? Problem 9.5
Prove that among any 6 people there are 3 who know each other or 3 who
don’t know each other.

It is useful to represent this by a graph with coloured edges. The

vertices of the graph correspond to the people. Any two vertices are
connected by an edge, which is coloured red (continuous line) or
green (dashed line). Red means that the two people know each other,
green means they don’t. See Figure 9.4 for an example.
We want to show that there is a monochromatic triangle (i.e. three
vertices such that all three edges between them have the same colour),
no matter how the edges are coloured.

Figure 9.4 A graph with edges coloured by two colours

Order in chaos: the pigeonhole principle in graph theory 213

Play with this to see it is actually true. Try to prove it. Hint: Look
at all the edges emanating from one vertex. What can you say about
them?

Think about it!

! Solution of Problem 9.5

We solve the problem translated into graph language as explained.
Let V be one of the vertices. There are 5 edges emanating from it. By
the pigeonhole principle three of them must have the same colour,
say red. Say the edges from V to V1 , to V2 and to V3 are red. Now
look at the edges between V1 , V2 and V3 . If one of them is red then its
two endpoints and V form a red triangle. Otherwise all of them are
green, so they form a green triangle. !

Does this also work with 5 vertices (people)? No: draw all outer
edges of a pentagon red and all diagonals green. Then there is no
monochromatic triangle.
Let us generalize the question: is it true that among sufficiently
many people there must be 4 who know each other, or 4 who don’t
know each other (i.e. none of the 4 knows any other)? In the lan-
guage of graphs this would correspond to 4 vertices so that all 6
edges between them have the same colour. We call this a complete
monochromatic quadrilateral.
It is a little simpler to first consider an intermediate case: colour all
edges between 10 vertices red or green. Then there is a red triangle
or a complete green quadrilateral. Prove this!
Is 10 the smallest number for which this is true? No. It turns out
that 9 also works, but this is a little harder to prove. For 8 vertices
this is not true. We symbolise this fact by the equation R(3, 4) = 9
and say that 9 is the 3,4-Ramsey number. Similarly, we showed that
R(3, 3) = 6. The higher Ramsey numbers are difficult to find precisely,
for example, R(5, 5) is not known.
 214 9 The pigeonhole principle

We can generalize the question further by using more than two

colours. One can show: Let k, a1 , . . . , ak ∈ N. Then there is a number
R so that every colouring of all edges between R points with the
colours 1, 2, . . . , k has a complete 1-coloured a1 -gon or a complete
2-coloured a2 -gon or . . . or a complete k-coloured ak -gon. That is:

If there is enough chaos then you will find order in it.

9.5 Toolbox

The pigeonhole principle is a tool which you should always have

in mind when you want to prove existence of something. It is not
always obvious how to use the pigeonhole principle. You need to
think about what could be the pigeons and the holes. If you are
looking for objects with certain properties then the objects will usually
correspond to the pigeons and the properties to the holes. In problems
about integers remainders are often good candidates for the holes.
The pigeonhole principle yields the existence of several objects with
certain properties. Sometimes it is also useful if you only need one
object, for example when you can use two objects to create another
one using algebraic operations, e.g. difference or quotient (Problems
9.3 and 9.4).

Exercises

1 E 9.1 .At least how many children must be in a room so you can be
sure that three of them have their birthday in the same month?

1 E 9.2 .
a) There are five pairs of green socks and five pairs of red socks in a
drawer. You cannot tell left socks from right socks. At least how
many socks do you need to take blindly out of the drawer (without
putting them back) if you want to be sure to have a matching pair?
b) In a shoe cabinet there are 10 different pairs of shoes. At least how
many shoes do you need to take blindly out of the cabinet (without
putting them back) if you want to be sure to have a matching pair?
How many do you need to take if you have five identical brown
pairs and five identical black pairs?
Exercises 215

E 9.3 . Suppose you walk around on a field of snow. Prove that 1-2
at some point you have to step into your own footsteps (at least
partially). How soon will this happen? Which quantities do you need,
to find a number of steps by which you will certainly have done this?

E 9.4 .Prove using the pigeonhole principle: Let a triangle and a line 1-2
be given. Then the line intersects at most two sides of the triangle.

E 9.5 .Show that you cannot cover an equilateral triangle completely 2

by two smaller equilateral triangles.

E 9.6 . In Figure 9.4 there are three monochromatic triangles, not 2-3
just one. Is this always the case? Or else are there always at least two
such triangles?

E 9.7 .Let a ∈ N and let x ∈ Z be a number which is not divisible 2

by 7. Consider the numbers a, a + x, a + 2x, . . . , a + 6x. Does there
have to be one among them which is divisible by 7?

E 9.8 .Let n ∈ N and a1 , . . . , an ∈ N. Prove that there are indices j 2-3

and k with j ≤ k so that a j + . . . + ak is divisible by n.

E 9.9 . Prove that among any 52 different numbers from the set 2-3
{0, . . . , 99} there must be two whose sum is 100.

E 9.10 . Prove that among any 52 natural numbers there are two 3
whose sum or difference is a multiple of 100. (Note that 0 is also a
multiple of 100.)

E 9.11 . A lattice point in the plane is a point whose coordinates 2-3

are integers. Imagine a tree standing at every lattice point except
the origin, see Figure 9.5. Standing at the origin, we look in some
direction. Prove:
a) If all ‘trees’ have thickness zero then you will see a tree in all
directions of rational slope, and you will see no tree in directions
of irrational slope.16

16 The slope of a ray starting at zero is defined as b/a if ( a, b ) is any point on the

ray except the origin. For vertical rays the slope is defined to be ∞. For the purposes
of this problem we count ∞ as rational.
 216 9 The pigeonhole principle

b) If each tree has thickness ε > 0 then we will see a tree in every
direction. That is, every ray starting at the origin hits a tree, no
matter how small ε is.
c) Find a value dε , depending on ε, for which you can be sure that in
b) you will see a tree in every direction at distance at most dε .

Figure 9.5 Trees obstruct the view

2 E 9.12 . What is the largest number of lattice points which you can
choose to satisfy the following condition: the midpoint between any
two of the chosen points is not a lattice point.

3 E 9.13 . Let a, b ∈ N be coprime, that is, the only natural number

c dividing both a and b is c = 1. Prove that there are n, m ∈ N
satisfying an − bm = 1.

3 E 9.14 .Prove that there is a Fibonacci number which is divisible by

1000.

3 E 9.15 .Prove that there is a power of 2 whose decimal representation

starts with the digits 999999. Also show that the same is true if 2 is
replaced by any natural number n which is not a power of 10.

3 E 9.16 .Let n ∈ N. At least how many numbers do you need to pick

blindly from the set {1, 2, . . . , 2n} if you want to be sure that one of
the chosen numbers divides another chosen number?
Exercises 217

E 9.17 .Suppose 101 distinct numbers are written in a row (they do 3-4
not need to be ordered by size). Prove that you can delete 90 of these
numbers so that the remaining 11 numbers form either an increasing
or a decreasing sequence.

E 9.18 .Investigate the following number-theoretic variant of Problem 2-3

9.4: Let n, N ∈ N. Find the smallest number d ∈ N0 , depending on n
and N, having the following property: for any natural number a at
least one of the numbers a, 2a, ..., Na lies within d of a multiple of n.
10 The extremal principle

Since the Form of the whole Universe

is most perfect, designed by a most wise Creator,
nothing at all takes place
that does not manifest, in some respect,
a Rule of Maximum or Minimum. 1
(Leonhard Euler)

Extremes are fascinating. Who is the smallest, tallest, fastest, strongest?

Everyday metaphors (the path of least resistance, to carry something
to extremes etc.) show how deeply ingrained the idea of the extreme
is in us. Moreover, the scientific view of the world reveals extremes
everywhere: the soap bubble tries to minimise its surface area and
is therefore spherical, chemical reactions strive towards a state of
minimal energy, and so on.
The soap bubble shows us more: extreme forms often have particu-
lar properties, for example they may be very regular or symmetric.
Reading this backwards we get a problem-solving strategy: If you are
looking for an object with particular properties, try to characterise it
by an extremal property. In this way extremes can be used to find
special objects or at least to prove their existence. Looking out for
extremes when you are solving problems can also be useful in many
other ways. For example, it may help you find a starting point or
structure your ideas.
In this chapter you will find examples for all of these different
aspects of the extremal principle. Along the way you will learn about
two fundamental inequalities and discover interesting facts about
mirrors and billiards.

1 Da aber die Gestalt des ganzen Universums höchst vollkommen ist, entworfen

vom weisesten Schöpfer, so geschieht in der Welt nichts, ohne dass sich irgendwie
eine Maximums- oder Minimumsregel zeigt.

© Springer International Publishing AG, part of Springer Nature 2018 219

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7_10
220 10 The extremal principle

The extremal principle is an idea with far-reaching consequences.

You will encounter it over and over again when you do mathematics,
although it is rarely mentioned explicitly. Keep your eyes open, look
out for it!

10.1 The general extremal principle

General extremal principle

Whenever something is extremal (largest, smallest etc.) particular

structures will emerge.

For example, the shortest path between two points is a straight line
(structure: no curves), the body of agiven volume with the smallest
surface area is the ball (structure: perfect symmetry), and beautiful
regular crystals appear when molecular configurations attain a state
of minimal energy.
We now look at some examples in more detail.

Rectangles, squares of distances and an important inequality

? Problem 10.1
Among all rectangles with perimeter 20 cm, which one has the largest area?

Think about it!

 ! Investigation and solution

G
 Denote the side lengths by a and b. Then the perimeter is 2a + 2b.
From 2( a + b) = 20 we get a + b = 10. The area is ab. So we are
looking for a, b satisfying a + b = 10 with largest ab. Let us check a
few examples. Since a, b play exactly the same role in the problem,
we may assume a ≤ b. See Table 10.1.
10.1 The general extremal principle 221

a b ab
1 9 9
2 8 16
3 7 21
4 6 24
5 5 25

Table 10.1 Some rectangle areas

The table suggests that the area is largest if a = b = 5. However,

the table is not a proof since a, b need not be integers. How can we
prove it?

Think about it!

 We focus on our goal: the number 5 plays a special role. If a is

less than 5 then b must be bigger by the same amount. So for any
a, b with a + b = 10 we can write a = 5 − x, b = 5 + x for some x.
Then
ab = (5 − x )(5 + x ) = 52 − x2 = 25 − x2 .
Clearly this is less than or equal to 25 for all x, and equal to 25 if
and only if x = 0, i.e. a = b = 5.
Therefore the rectangle of largest area is a square with sidelength
5 cm. !
G

 Review
The first thing you may have tried is eliminating b by writing b =
10 − a. Then you need to minimize a(10 − a) = 10a − a2 . It is not
obvious how to proceed. One way would be to complete the square:

10a − a2 = −( a2 − 10a) = −( a2 − 10a + 25 − 25) = −( a − 5)2 + 25

and this yields the solution a = 5 as above. Another way would be to

use calculus (if you know calculus).
222 10 The extremal principle

Focussing on the expected answer 5 as in the solution above led us

to the path of least resistance. 
This is an example of the extremal principle: the extremal rectangle is
a square, which is the most regular kind of rectangle.

? Problem 10.2
Which number between 0 and 2 minimizes the sum of the squares of its
distances to 0 and to 2?

! Solution
If we call the number a then the two distances are a and 2 − a, so we
want to minimize a2 + (1 − a)2 . Recalling the previous solution we
look at the deviation from the center, i.e. we write a = 1 − x. Then
2 − a = 1 + x, so

a2 + (2 − a )2 = (1 − x )2 + (1 + x )2 =
1 − 2x + x2 + 1 + 2x + x2 = 2 + 2x2 .

Clearly this is minimal for x = 0, so for a = 1. !

Again the extremal principle is at work: the smallest sum of distances
squared is attained in the symmetric case where a is halfway between
0 and 2.
A similar result holds for more than one point in the interval, see
Exercise E 10.22. This is a simple one-dimensional model for a crystal:
the sum of distances squared corresponds to the energy, and the state
of minimal energy is the regular configuration.
The core of these two problems can be written as an inequality.
10.1 The general extremal principle 223

Theorem (Inequality of the geometric, arithmetic and

quadratic mean)
Let a, b ≥ 0. Then

√ a+b a2 + b2
ab ≤ ≤ .
2 2
Each inequality becomes an equality if and only if a = b.

√ a+b
We call ab the geometric mean, the arithmetic mean and
 2
a2 + b2
2 the quadratic mean of a and b. See Exercise E 10.4 for the
significance of these means. There is a similar inequality for more
than two numbers, see Exercise E 10.20.
Proof.
a+b b− a
Let c = 2 and x = 2 . Then a = c − x, b = c + x, so ab = c2 − x2
hence ab ≤ c2 ≤ a +2 b . Now take square
2 2
and a2 + b2 = 2c2 + 2x2 ,
roots. Equality holds if and only if x = 0, i.e. a = b. q. e. d.

Make sure you understand why this inequality immediately yields

the solutions of Problems 10.1 and 10.2.

Shortest paths

Everyone knows that the shortest path between two points is the
straight line. Actually this is not easy to prove rigorously. But here
we will take it for granted. A special case of this is:2

Theorem (Triangle inequality) In a triangle each side is

shorter than the sum of the two other side lengths.

? Problem 10.3
Let a line g and two points A, B lying on the same side of g be given. How
do you determine a point C on g for which the sum of distances from A to
C and from C to B is minimal? See Figure 10.1.
224 10 The extremal principle

B
A

g
C

Figure 10.1 The problem of the shortest path touching the line g

Think about it. Then look at the following solution, which is beautiful
and ingenious. Put it in your bag of tricks.

! Solution
Reflect B across the line g. Let B
be the mirror point. Draw the
straight line from A to B
. It intersects g at a point C.
How do you proceed from here? Make a sketch.

Think about it!.

Let D be any other point on g. See Figure 10.2. on page 226

Since B and B
are mirror points they have the same distance to C.
Therefore the distance from A to B via C is the same as the distance
from A to B
via C.
Similarly, the distance from A to B via D is the same as the distance
from A to B
via D.
By construction of C the path from A to B
via C is a straight line,
and by the triangle inequality it is shorter than the path via D.
Therefore the path from A to B via C is shorter than the path via
D. !
The solution has an interesting consequence:

2 See Exercise E 10.8 for a direct proof of the triangle inequality.

10.1 The general extremal principle 225

Theorem Suppose the line g and the points A, B are as in

Problem 10.3, and let C be the optimal point on g.
Then the angles that g makes with the lines CA and CB are
equal: α = β in Figure 10.3.

Again this is an example of the general extremal principle.

Proof.
This follows immediately from the construction of C: The angles α and
β
are opposite angles at the intersection of two lines and therefore
equal, and β
= β because B
is the reflection of B. Therefore α = β.
q. e. d.

Remark

We call α the angle of incidence and β the angle of reflection.

We have shown that the shortest path obeys the law of reflection:
angle of incidence = angle of reflection. This is interesting for
players of billiards: If you want to play a ball from A to B off the
cushion g, try to imagine the reflection B
of B in g and aim at B
.
3

In Section 10.4 you will find more on this topic.

Other extremes

Extremal configurations need not always be symmetric: the rectangle

of smallest area and perimeter 20 cm is a line segment 10 cm long
(a ‘degenerate’ rectangle whose area is zero); the sum of squares of
the distances from a to 0 and 2 is largest when a = 0 or a = 2. In
both cases the extremum is attained at the boundary of the set of
possibilities. This is also a type of particular property. So the general
extremal principle still holds.

10.2 The extremal principle as problem solving strategy, I

By reading the general extremal principle backwards we obtain a
problem-solving strategy:

3 With real billiards you need to take into account that the balls are not points.
226 10 The extremal principle

B
A

D g
C

Figure 10.2 The solution

B
A

α β g
C β

Figure 10.3 Equal angles

Extremal principle as problem-solving strategy

When looking for an object with particular properties, try to

characterise it by an extremal property: largest, smallest, simplest,
closest to having the desired property, etc.

Therefore you should reflect along a line which is on the table parallel to the cushion
at distance one ball radius. Also you need to take into account that the ball will
slightly dent the cushion, which has the effect that the angle of reflection is slightly
bigger than the angle of incidence. This effect increases with ball speed.
10.2 The extremal principle as problem solving strategy, I 227

So extrema can be useful for proofs of existence. In contrast to

the pigeonhole principle (see Chapter 9) these proofs are usually
constructive, as we will see.
We now consider an example, then read off the typical structure of
such proofs, and then solve a more difficult problem using this idea.

A problem about tournaments

? Problem 10.4
In a tournament every player plays against every other player once. There
are no draws. At the end every player makes a list containing the names of
all players he defeated, and also of all the players defeated by a player whom
he defeated.
Show that there is a player whose list contains the names of all other
players.

 ! Investigation and solution

G
 Look at a few examples to understand the problem. It is useful to
represent the outcome of the games in the tournament by a graph:
vertices are players, and edges correspond to games. Every edge is
marked by an arrow, which points from the winner to the loser.4
See Figure 10.4 for an example. Player a defeated b and d, and
b defeated c. Therefore a’s list contains b, c and d. So player a
satisfies the condition of the problem.
Player d’s list only contains c and a, so d does not satisfy the
condition. This is irrelevant since we only ask for at least one player
with a complete list. Player b also has a complete list.
 How could we prove that there is a player with a complete list,
no matter how each game ended? The difficulty is that we know
almost nothing. In addition, the problem seems unapproachable
because of the complicated description of the list.
 Could we replace the condition by a simpler one?
A player who has won many games seems likely to have a long list.
So let us try this:
4A graph all of whose edges are marked by an arrow is called a directed graph.
228 10 The extremal principle

b
c

Figure 10.4 A tournament with 4 players

 An attempt: Let A be a player who has won the most games.5

Can we conclude that A’s list contains the names of all other
players?

Think about it!

Suppose B is not on A’s list. Then B must have defeated A. Also, B

must have defeated all players that A has defeated, for if B had lost
against one of them then B would be on A’s list. So B has defeated
all players that A has defeated, and also A. Therefore B has won
more games than A. This is a contradiction to the assumption that
A has won the most games.
Therefore the assumption that B is not an A’s list must be wrong,
so A’s list contains the names of all other players. G!

How to use the extremal principle

The solution of Problem 10.4 shows how the extremal principle is

typically used:

Problem: We are given a set M of objects and the specification of

some property E that these objects can have.
5 Why do we say ‘a player’, not ‘the player’? Because ‘the’ would suggest that

there is only one such player. But several players could have the same maximal
number of victories, as in Figure 10.4, where both a and b have won twice.
10.2 The extremal principle as problem solving strategy, I 229

We want a proof that there is an object in M having property E .

Strategy: We look for a quantity whose maximisation allows us to

infer property E . Here a quantity is an assignment of a number to
each object in M, that is, a function q : M → R.

Proof of existence: Suppose we have chosen a quantity q which

we believe to serve our purpose.

1. Let A ∈ M be an object for which q( A) is maximal: q( A) ≥ q( B)

for all B ∈ M.

2. We claim that A has property E .

3. To prove this we assume that A does not have property E . Using

this information we construct an object B satisfying q( B) > q( A). This
is a contraction to the maximality of q( A). Therefore A must have
property E .

Instead of maximising a quantity it may be more natural to minimise,

depending on the problem.

Example

In Problem 10.4 M was the set of players, E was the property of

having a complete list, and q( A) was the number of games that
A has won.

We are free to choose the quantity q: If it works it is permitted. The

criterion is that step 3 can be done, in particular the construction of
B. While we work on the problem the following may happen: we
find a quantity q which we believe will work, but then we cannot do
step 3. In this case we should look for another quantity q
– or try out
different proof strategies, like induction or other ideas.
Looking for a suitable quantity q is a creative process and may
require several attempts.
Once you have found a quantity q which works you get a con-
structive procedure to find an object with property E : Start with any
object A. If it has property E then we are done. If not, use step 3 to
230 10 The extremal principle

construct an object B with q( B) > q( A). If B has property E then we

are done, otherwise construct C with q(C ) > q( B) etc.
In the argument we use the following simple facts:

1. Every finite set of real numbers has a largest and a smallest element.

2. Every set of natural numbers has a smallest element.

A smallest element is also called a minimum, a largest element a

maximum.

Remark

The second fact is also true for infinite sets of natural numbers,
but not necessarily for infinite sets of positive real numbers. For
example, the set { x ∈ R : x > 0} does not have a smallest
element.
See Section 10.4 (after Problem 10.8) for more on the question of
existence of extrema.

The problem of farms and wells

We now consider a problem where it is harder to find a quantity

which we can maximise or minimise to solve the problem.

? Problem 10.5
We are given 2n points in the plane: n farms and n wells. We assume that
no three of the points lie on a line. Is it possible to assign a well to each
farm in such a way that each well belongs to only one farm and the straight
paths from each farm to its well do not intersect?

Figure 10.5 The first and third assignment of wells to farms are admissible,
the others are not. Farms are squares and wells are circles. Assignments are
indicated using line segments.
10.2 The extremal principle as problem solving strategy, I 231

 Investigation
G
 Let us call an assignment of wells to farms admissible if it satisfies
the conditions, i.e. if it is bijective and intersection-free. Figure 10.5
shows a few examples with n = 2. These are deceptively simple.
For large n things could look quite complicated. If you try a
few more examples you will notice that an admissible assignment
seems to always exist. So we formulate:
Conjecture: There is always an admissible assignment.
 How could we prove this? The difficulty is that there are so many
possibilities for the positions of farms and wells. As a first idea we
could put ourselves in the position of a farmer:
First attempt: To each farm assign the well which is closest.
If the farms lie as in the first picture in Figure 10.5 then both
farms get the same well (second picture), so the assignment is not
bijective.
So that didn’t work: it was too short-sighted to take the individual
points of view of each farmer as a basis. We should keep in mind
the total configuration.
 Let us try to use the extremal principle. Let us make a plan. What
are we looking for? Among all bijective assignments we are looking
for one without intersections. It is natural to try the following.
Second attempt: Consider a bijective assignment with the minimal
number of intersections.6
This seems promising since for a solution of the problem this
number must be zero, hence minimal.
 We want to prove: If A is a bijective assignment with minimal
number of intersections, then A has no intersection. To prove this,
we assume that A has an intersection. Then there would be two
farms and wells with assignments as in Figure 10.6 (solid lines;
there could be more farms and wells which are not drawn). We
want to change this to a different bijective assignment B having
fewer intersections than A. How could we do this?

6 So in the general scheme of the previous section M is the set of bijective

assignments, E is the property of having no intersections, and q( A) = number of

intersections in A ∈ M.
232 10 The extremal principle

f2

f1

w2 w1

Figure 10.6 A simple idea for reducing the number of intersections. Here
farms and wells are represented by dots.

 We want to remove the intersection S. That’s easy: replace the

solid lines by the dashed lines and leave all other lines (i.e. pairs
farm – well) unchanged. Let us call the new assignment B. Does B
necessarily have fewer intersections than A?

Think about it!

 The left-hand picture in Figure 10.7 shows that this is not necessar-
ily true. In this example, while removing the intersection S (using
the dashed lines) we produce a new intersection T. So B has the
same number of intersections as A has. Even worse, you can easily
modify the example so that B has more intersections than A.

f2 f2

f1 f1

S f3 S f3
w3 w3
T

w2 w1 w2 w1

Figure 10.7 A problem with the second attempt and a solution for this
example
10.2 The extremal principle as problem solving strategy, I 233

 We see that our way to construct B from A was too simple-minded.

The right-hand picture in Figure 10.7 shows a way to get an
intersection-free assignment in this case (dashed lines). Can we
find a general construction, which in this example replaces the
solid lines (assignment A) by the dashed lines as in the right-
hand picture (assignment B), and which reduces the number of
intersections for any A?
For example, the construction must also work if there are more
farms in the triangle Sw1 f 2 (or in the triangle Sw2 f 1 ) and A assigns
them to wells outside the triangle.
 The complexity of the situation increases dramatically. It is not at
all clear how to proceed.7
So let us try something completely different. In the second attempt
we minimised the number of intersections; that didn’t work. Are
there other quantities whose minimisation might work?
We could try to combine elements of the first and second attempt:
we consider the distances from the farms to the wells – however,
this time not individually, but with a view to the entire configu-
ration. So we should cook up a single overall number from all of
these numbers.
The simplest thing is to add all these numbers. Surprisingly this
works. G

Think about it!

! Solution of Problem 10.5

There always is such an assignment.
Proof: Among all bijective assignments of wells to farms consider one
which has the smallest sum of all distances from the farms to their wells.
Call this assignment A. We show that A has no intersections and thus
solves the problem.
7 If you do find a way to finish this attempt successfully, please let me know –
D.G.
234 10 The extremal principle

Suppose this was not the case. Then there would be wells w1 , w2
assigned to farms f 1 , f 2 , respectively, so that the line segment f 1 w1
intersects the line segment f 2 w2 . Let S be the point of intersection. So
we have a situation as in Figure 10.6, since no three points lie on a
line. We didn’t draw the other farms and wells.
Claim: Let B assign the well w2 to the farm f 1 and the well w1
to the farm f 2 (dashed lines) and leave the assignments of all
other wells to their farms as in A. Then B has a smaller sum of
distances than A.
This would be a contradiction to the minimality of A. Therefore A is
admissible.
It remains to prove the claim. Since all other assignments stay the
same, it suffices to show that the two dashed lines together are shorter
than the two solid lines together:
!
( f 1 w2 ) + ( f 2 w1 ) < ( f 1 w1 ) + ( f 2 w2 ) (∗)
where ( PQ) denotes the length of the line segment from a point P
to a point Q. To prove (∗) we use the triangle inequality, applied to
the triangles w2 S f 1 and w1 f 2 S:
( f 1 w2 ) < ( f 1 S) + (Sw2 ), ( f 2 w1 ) < ( f 2 S) + (Sw1 ).
Adding these inequalities we obtain the left hand side of (∗) on the
left, and on the right
( f 1 S) + (Sw2 ) + ( f 2 S) + (Sw1 ) = ( f 1 w1 ) + ( f 2 w2 ),
that is, the right hand side of (∗), where we used ( f 1 S) + (Sw1 ) =
( f 1 w1 ) and ( f 2 S) + (Sw2 ) = ( f 2 w2 ). !
 Review of Problem 10.5
J The idea of minimising the natural quantity q( A) = (number of
intersections for the assignment A) did not lead to a proof. This
may seem paradoxical since we know that this number is minimal
for the assignment we are looking for. However, at this point in
the argument we do not know yet that there is an intersection-free
assignment; this is only a conjecture. It could have turned out that
the conjecture was wrong, i.e. that there is a configuration of farms
and wells for which every bijective assignment has intersections.
10.2 The extremal principle as problem solving strategy, I 235

J It is essential that the construction of B in the second step of the

proof works for all A that have intersections. Only in this way we
get a complete proof. Examples can help us to get an idea, but
they are not sufficient.
For q( A) = (number of intersections for the assignment A) we did
not succeed in finding such a construction.
J However, the choice q( A) = (sum of distances for A) quickly
led to success. Note that in the construction of B with smaller
distance sum we could have created new intersections; but this is
irrelevant since only the sum of distances matters for the argument.
Make sure that you understand this important point. See also the
algorithmic point of view below.
J Quite generally in geometric problems it makes sense to look at
maxima or minima of geometric quantities, i.e. those related to
lengths, distances, areas etc. Of course this is only a heuristic and
not a universal recipe. 

Let us reconsider the problem of farms and wells from an algo-

rithmic point of view: we want to give an algorithm for finding an
admissible assignment. We try the following simple algorithm:

1. Choose any bijective assignment. If it has no intersections we

are done. Otherwise proceed to step 2.

2. Choose an intersection and remove it by changing the assign-

ments for the two farms involved as in Figure 10.6.

3. Repeat step 2 until you get an assignment without intersections.

Does this algorithm terminate at some point? This is not obvious at

all. We saw that when removing one intersection we might create
several others, so it seems possible that we might get into a loop,
arriving at an assignment that we have already considered some steps
before. Then we could repeat the loop forever.
However, our solution of Problem 10.5 implies that the algorithm
does terminate. How?
236 10 The extremal principle

Think about it!

We showed that the sum of distances of farms to their assigned

wells decreases every time that we do step 2 of the algorithm. This
already shows that no loops are possible. More to the point, there
are only finitely many numbers that appear as sums of distances
since there are only finitely many assignments. Therefore we must
arrive at the smallest of these numbers after a finite number of steps
(unless the algorithm stopped before that). Then our proof shows that
the corresponding assignment has no intersections, so the algorithm
stops.
Summarizing, we have seen that although the number of intersec-
tions may increase while we run the algorithm, at the end there will
be no intersections.

10.3 The extremal principle as problem solving strategy, II

We saw that the extremal principle can be a tool for proofs of existence.
But considering extremes can also be useful as a more general idea.

Extremal principle as problem solving strategy, II

In a complex situation look at objects with an extremal property.

In the simplest case, looking at extremes may help you to formulate

an idea: we saw examples of this in the proofs of the theorems on
division with remainder in Section 8.1 and on rational approximation
in Section 9.3.
We now consider some examples where the general idea of ex-
tremes helps to solve problems.

A problem about averages

? Problem 10.6
1000 numbers are written along the perimeter of a circle. Each number is
the average of its two neighbours. Show that all numbers are equal.
10.3 The extremal principle as problem solving strategy, II 237

Think about it!

 ! Investigation and solution

G
 First solution: Look at one of the numbers. Since it is the average
of its neighbours, there are two possibilities:
1. It is equal to its neighbours, or
2. one neighbour is larger, the other smaller.
In the second case consider the larger neighbour. By the same
reasoning its other neighbour must be still larger. Continuing,
we see that the numbers must get larger and larger while we go
around the circle in one direction. This leads to a contradiction
once we arrive at the starting number since it cannot be bigger than
itself.
Therefore the first case must apply. Since the number we started
with was arbitrary, we have proved that each number is equal to
its neighbours. Therefore all numbers must be equal.
Here is another argument which is shorter:
 Second solution: Let x be the smallest of the 1000 numbers. Since
it is the average of its neighbours, these neighbours must equal x
since otherwise one of them would be smaller than x, contradicting
minimality of x. So both neighbours are equal to x, hence also
minimal. By the same argument, their neighbours must also equal
x, then the next neighbours etc. Since in this way you reach all
numbers, they must all equal x. G!

The two solutions are very similar; it almost seems as if the second is
a rephrasing of the first. This is not so: the second solution, using the
extremal principle, can be generalised to a two-dimensional version
of the problem, but the first can’t. See Exercise E 10.14.
238 10 The extremal principle

Infinite descent

Infinite descent is a type of argument which is closely related to the

extremal principle. Here is an example.

? Problem 10.7
Can three times a square number equal the sum of two square numbers?

 Investigation
G
 The question is: Are there natural numbers a, b, c satisfying

3a2 = b2 + c2 ?

Try to find a solution by testing a few numbers.

 Have you seen a similar problem before?

Think about it!

√
 Maybe the problem reminds you of the proof that 2 is irrational,
see Section 7.2. There we had to prove that the equation 2q2 = p2
has no solution in natural numbers.
We don’t know yet whether √ our problem has a solution. If we
could argue as we did for 2 then we might be able to prove that
there is no solution. Try it!
 How did we proceed with the equation 2q2 = p2 ? We first showed
that p is even, using that 2q2 and hence p2 is even.
In our problem we have a 3, not a 2, so we might try to argue with
divisibility by 3, not 2.
 So let us suppose we had a solution of 3a2 = b2 + c2 . Then b2 + c2 is
divisible by 3. Unfortunately we cannot conclude from this directly
that b or c are divisible by 3. How to go on?
 The generalisation of even/odd is the remainders modulo 3, see
Chapter 8. Let us check the possible remainders modulo 3 of b and
10.3 The extremal principle as problem solving strategy, II 239

b2 . There are three cases:

b≡0 mod 3 ⇒ b2 ≡ 02 = 0 mod 3

b≡1 mod 3 ⇒ b ≡ 1 = 1
2 2
mod 3
b≡2 mod 3 ⇒ b ≡ 2 = 4 ≡ 1
2 2
mod 3

We see that b2 can only have the remainder 0 or 1 (compare Exercise

E 8.10). Can you go on from here?

Think about it!

 For the same reason c2 can only have the remainder 0 or 1, so

b2 + c2 can only have the remainder 0, 1 or 2, and the remainder 0
can only occur if both b2 and c2 have remainder 0.
 We summarise: From 3a2 = b2 + c2 we can conclude that b2 + c2 ≡
0 mod 3 and from this that b ≡ c ≡ 0 mod 3. Therefore we can
write b = 3B, c = 3C with integers B, C.
 When we plug these in, we get 3a2 = 9B2 + 9C2 , so a2 = 3B2 +
3C2 = 3( B2 + C2 ). Therefore, a2 ≡ 0 mod 3, and by the cases
above we conclude that a ≡ 0 mod 3. Then we can write a = 3A,
and plugging this in yields 9A2 = 3( B2 + C2 ), hence

3A2 = B2 + C2 .

 How can we use this? This is the same equation as the one we
started with. But A is smaller than a (since a = 3A), and similarly
for B, C. This easily leads to a contradiction, see below. G

! Solution of Problem 10.7

There are no such square numbers.
Proof: Suppose there were natural numbers a, b, c satisfying 3a2 =
b2 + c2 . Then b2 + c2 is divisible by 3, and the argument above shows
that then b and c are divisible by 3, so b = 3B, c = 3C with B, C ∈ N.
240 10 The extremal principle

Again by the argument above it follows that a is divisible by 3, so

a = 3A with A ∈ N. We obtain 3A2 = B2 + C2 .
We have shown: If 3a2 = b2 + c2 then a, b, c are divisible by 3, and
A = 3a , B = 3b , C = 3c are solutions of the same equation. Then by
the same argument A, B, C must be divisible by 3 etc. In this way
we would obtain an infinite sequence of ever smaller and smaller
solutions, all natural numbers. This is clearly impossible. !
Proof analysis: This was an indirect proof. The core of the
argument was to consider the remainders modulo 3 of square
numbers. The method by which we arrived at a contradiction
is called infinite descent.
Infinite descent is really an extremal principle argument in
disguise; the proof above could be reformulated as follows:
Suppose there is a solution a, b, c. Let a0 be the smallest value
of a among all solutions. Construct A, B, C as before. This is a
solution again, and because A = a30 < a0 we get a contradiction
to the fact that a0 was minimal. Therefore a solution cannot
exist.

10.4 Going further: optimisation, mirrors and billiards

We encounter extrema (that is, maxima and minima) in mathematics

in two types of contexts:

1. Determining extrema as a goal in itself: often motivated by ap-

plications, we want to determine how to make something fastest,
cheapest, biggest, smallest etc. This is called optimisation.

2. Using extrema as a means for other purposes, for example for

proofs of existence. You saw some examples in this chapter.

In this section you will find more examples and a more extensive
discussion of shortest paths. Even if you don’t understand all of the
words in the explanations that follow, keep reading anyway. You will
be astonished how universal the extremal principle is.
In the books (Nahin, 2004) and (Hildebrandt and Tromba, 1996)
you will find more information about many of these topics and their
history.
10.4 Going further 241

Optimisation problems

Here are some examples of optimisation problems from different

strands of life and mathematics.

J An engineer who constructs a bridge wants it to be as stable as

possible, while at the same time it should be light-weight and
cheap.8

J A company wants to use its resources (personnel, means of pro-

duction) in such way that the profit is maximised.

J Given several points in the plane, find a curve which is ‘simple’

and comes as close as possible to the points. A typical version of
this problem is the least squares problem of linear regression in
statistics. Here real numbers x1 < x2 < · · · < xn and y1 , y2 , . . . , yn
are given, and we look for those numbers a, b for which the linear
function f ( x ) = ax + b produces values at the xi which are closest
in mean square to the yi , i.e. for which (y1 − f ( x1 ))2 + · · · + (yn −
f ( xn ))2 is minimal.

J The isoperimetric problem is the question which planar figure

with given area has the smallest perimeter, or which body with
given volume has the smallest surface area. The answers are the
circle and the ball.9 It is not easy to prove this rigorously. A related
problem is to find the surface of smallest area whose boundary is
a given closed curve in space.10

J Suppose A, B are points in space, where A lies higher than B and

not vertically above it. We want to build a track on which a ball
can roll from A to B. What shape must the track have so that the
time needed is as short as possible? Maybe your first guess is
that the track should be as short as possible, hence a straight line.
However, it turns out that the total time will be shorter if the track
is very steep at the beginning and then gets flatter – and maybe
even turns upwards near the end, for certain positions of A and B!
8 Not to mention aesthetic considerations.
9 An equivalent problem is to find the planar figure with given perimeter and
largest area, and similarly for bodies. Compare Problem 10.1.
10 Imagine the curve is a bent wire. Put the wire into soapy water. When you pull

it out carefully you will see a soap film, and this will be a surface of minimal area.
242 10 The extremal principle

In this way the ball will pick up more speed in the beginning and
arrive sooner than for the straight track, although its path is longer.
What is the precise shape of the optimal track? To answer this you
need to know differential equations. The best curve is called the
brachistochrone. This sort of optimisation problem, where we are
looking for a curve, not a number, is called a variational problem.

J The travelling salesman problem: You want to visit n cities. The

cities are fixed in advance, but you can choose the order in which
you visit them. How should you travel in order to minimise the
length of the journey?11

J The problem of optimal mass transport: You have a pile of sand

and a hole, and you want to put the sand in the hole. The hole is big
enough for all the sand. Moving the sand takes work. The problem
is: where should each grain of sand go in order to minimise the
total work? What makes this problem non-trivial is that the hole
can have a different shape from the pile.

J The problem of densest sphere packing: how do you pack a large

number of spheres of equal size so that the gaps are as small (in
total) as possible? If you try stacking oranges you will quickly
arrive at a guess for the best packing. Kepler conjectured in 1611
that this packing is optimal. However, to prove that this is the
optimal packing is extremely difficult and was done only in 1998.

Examples of the extremal principle in different areas of

mathematics

The extremal principle as a method of proof is used in all areas of

mathematics. You have already encountered a few instances, for
example from number theory. Here are a few more examples. Only a
few keywords are given, and the areas of mathematics are indicated in
parentheses. You will learn about some of these as an undergraduate
student of mathematics; for some of them you may have to wait until
postgraduate studies.

J The supremum axiom for the real numbers (analysis).

11 There is no known algorithm which solves this problem generally for n = 100
in an acceptable time (the age of the universe, say)!
10.4 Going further 243

J Proving the mean value theorem using the extreme value theorem
(analysis).

J Proving that symmetric matrices A have eigenvalues, by minimis-

ing or maximising the associated quadratic form f ( x ) =  x, Ax 
over the unit sphere (linear algebra, functional analysis, partial
differential equations).

J Zorn’s lemma, which can be used to prove that every vector space
has a basis, or to prove the Hahn-Banach theorem and many
other important theorems (linear algebra, functional analysis, etc.).

J The first complete proof of the fundamental theorem of algebra

(every polynomial has a zero in the complex numbers), published
by Argand in 1814, uses the extremal principle (complex analysis).

J The maximum principle for harmonic functions, which is useful for

analysing these functions and also for proving that the Dirichlet
problem can have at most one solution (analysis, partial differential
equations).

J Proving that in a Hilbert space the projection to a given closed

subspace exists, via minimising distances from points to the sub-
space; this is the core for the existence theory for solutions of
partial differential equations using the Hilbert space method
(functional analysis, partial differential equations).

J In the variational calculus the extremal principle is used systemati-

cally for solving differential equations.

J Investigation of the topology of a space via Morse theory (topol-

ogy).

Keep your eyes open; you will find many more examples!
Another beautiful example is the rainbow: Why are there rainbows,
why are they circular arcs, and how can you calculate their position?
This also amounts to solving an extremal problem (see (Nahin, 2004)).
244 10 The extremal principle

Derivatives: a new look at shortest paths, light rays, billiards

etc.

Calculus provides an important technique for solving many optimisa-

tion problems: to find the extrema of a function you find the zeroes
of its derivative.12 This is the general idea, but there are some details
to watch out for: an extremum must be a stationary point only if it
is attained at an interior point of the domain of definition; an interior,
stationary point need not be a (local) extremum – it could also be a
point of inflection. So you need to check boundary points separately,
and for each interior, stationary point you need to check whether it
is a local extremum. Finally, if you are only interested in the global
extrema, then you have to check for that. But in most cases these
details are easy to deal with.
Using this technique you can solve Problems 10.1 and 10.2 and give
a proof for the inequality of the geometric, arithmetic and quadratic
mean. However, the elementary solutions given in this book are more
appropriate considering the simplicity of these problems.
The relationship of extrema and derivatives gives another hint why
the general extremal principle holds: if the function x → f ( x ) has
an extremum xextr in the interior of its domain of definition then
f
( xextr ) = 0. So the point xextr is in the solution set of this equation,
and the solution sets of equations often have particular structure.
As an example we give a calculus solution of Problem 10.3: with
notation as in Figure 10.8 (where d is the distance from P to Q)
we want to find the point C, i.e. the number x, so that f ( x ) =
( AC ) +√(CB) is minimal.

13 By Pythagoras’ theorem we have

f ( x ) = a2 + x2 + b2 + (d − x )2 , and after a short calculation you

get the derivative
f
( x ) = cos α − cos β. (10.1)
At a minimum xmin of f the equation f
( xmin ) = 0 must hold (since
the domain of definition is R, so all points are interior), hence cos α =
cos β, which implies α = β since α, β ∈ (0, π ).
This is a new proof of the theorem, stated after Problem 10.3, that
the shortest path from A to B obeys the reflection law. This new
12 These are also called the stationary points of the function. Here we always

assume that the function is differentiable at all relevant places.

13 ( AC ) denotes the distance between the points A and C.
10.4 Going further 245

B
A
b
a
α β g
P x C d−x Q

Figure 10.8 Notation for the calculus solution of Problem 10.3; C is not the
minimum in the picture

proof does not use the mirror point B

(compare Figure 10.3), and
therefore it has the advantage of being applicable in situations where
the mirror construction doesn’t work. For example, we can replace
the line g by a curve. Then one can prove in a similar way:

Theorem Let c be a curve and let A, B be two points lying on

the same side of c, see Figure 10.9. Let C be an interior point on
c for which the sum of the distances ( AC ) + ( BC ) is extremal.
Then α = β, that is, angle of incidence = angle of reflection at
C.

Conversely, at any point C where α = β the function f (C ) = ( AC ) +

( BC ) is stationary at C, i.e. it has a local maximum or minimum or
an inflection point.

c
B
A

β
α
C

Figure 10.9 Reflection in a curved mirror; α, β are the angles with the
tangent line of the curve c in the point C

If we think of c as a mirror then α = β means that a light ray

coming from A gets reflected to B. So we can rephrase the theorem
246 10 The extremal principle

as follows:

Longest and shortest paths are light paths.

The converse of this is known as Fermat’s principle:14 Light will

travel along those paths between two points whose length is stationary
under small variations.15
Looking at the previous theorem from a different angle we get
another interesting consequence. Consider the following problem.

? Problem 10.8
Let c be a curve and let A, B be two points lying on the same side of c. Does
it follow that there is a point C for which the angles α and β are equal?
So we want to send a billiard ball from A to B via the cushion c. Is
that possible?

! Solution

(incomplete) Choose an interior point C on c so that ( AC ) + (CB)

is minimal. Then the theorem above implies α = β. So a point C as
desired exists. !
This is a beautiful example of the extremal principle as a tool for
proofs of existence. See Exercise E 10.25 for an interesting variant of
this problem.
Why is the solution incomplete? We need to prove the existence of
C, i.e. the minimum of f ! For the set of possible function values f (C )
is neither finite nor a set of natural numbers, so it is not obvious that
it has a minimum. How can we complete the proof?

14 Pierre de Fermat, 17th century. The principle was generalised in the 18th and
19th century by Euler, Lagrange, Maupertuis and Hamilton to the ‘principle of
least action’, which aims to describe all physical processes in a similar way.
15 This is the version for homogeneous media, where the speed of light is constant.

If you replace the length of the path by the time of travel then this still holds for
inhomogeneous media. If you apply this to the transition between two media then
you get the law of refraction (Snell’s law).
10.5 Toolbox 247

1. The extreme value theorem of analysis says that any continuous

function on a bounded closed interval has a minimum and a
maximum. By parametrising the curve we can think of f as a
function on an interval. It is bounded and closed if we include the
endpoints of c. Continuity of f is easy to show using the triangle
inequality. So we know that f does have a minimum.

2. In order to use the theorem above we need to show that the

minimum is attained in the interior of the curve. This is not clear,
and indeed need not be true: if c is an interval on the x axis and A,
B both lie above and to the left of c then the minimum is attained
at a boundary point, and clearly there is no reflection point C on c.
So the answer to the problem is yes only if it can be guaranteed
that the minimum is attained at an interior point.
This is the case, for example, if c is a closed curve, such as an
ellipse.
In many situations where we would like to use the extremal princi-
ple the function f depends on several variables. The extreme value
theorem is then still true if each variable can vary independently on
a bounded closed interval. However, this is an unnatural restriction
on the domain of definition of f . Here it is more natural to use the
notion of a compact set. You can learn more about this central notion
of mathematics in books on analysis or topology.

10.5 Toolbox

Starting from the general extremal principle, a general scientific idea,

we arrived at the extremal principle as a problem-solving strategy.
We first used this strategy as a means for proofs of existence: try
to characterise the desired object by an extremal property. Often it
is not obvious which property will work for this, and a dedicated
search and several attempts may be necessary. There are no limits
to your creativity (as long as you argue correctly, of course). In a
more general sense, the extremal principle as a strategy tells you that
in any kind of complex situation it is worth looking for extremes.
One variant of this idea is the method of infinite descent, which
sometimes helps to prove that an equation has no solution in natural
numbers.
 248 10 The extremal principle

Exercises

1 E 10.1 .Choose a natural number bigger than two. Take its square
root and round it down, then take the square root again and round
it down etc. Show that at some point you will obtain 2 or 3. How is
this related to the extremal principle?

1-2 E 10.2 . Suppose the product of two positive real numbers is 20.
What is the smallest possible value for their sum? What is the largest?

1-2 E 10.3 .In Problem 10.2 replace the squares of the distances by the
reciprocals of the distances and solve the problem.

2 E 10.4 .Let 0 < a ≤ b be given.

a) A car moves at speed a (kilometers per hour) for an hour and then
at speed b for an hour. What is its average speed? (average speed =
total distance / total time)
b) A car moves at speed a for 100 km and then at speed b for another
100 km. What is its average speed?
c) Prove √
the inequality of the harmonic and the geometric mean,
1
2
+1
≤ ab. Deduce the inequality 1 +2 1 ≤ a+2 b . Also argue without
a b a b
calculation why the latter inequality holds, using a) and b).
√
d) Find a real-life context where the geometric mean ab appears.

2 E 10.5 .In Problem 10.4 we looked at a player who has won the most
games, and this led to a solution. Instead we could have looked at a
player with the longest list. Can you find a solution of the problem
from here?

1-2 E 10.6 .Does the argument in the solution of Problem 10.5 also work
if you use q( A) = (the biggest distance from a farm to a well) instead
of the sum of distances?

2 E 10.7 .One of the fundamental facts of analysis is that Q is a dense

subset of R. This means: If a < b are real numbers then there is a
rational number r satisfying a ≤ r ≤ b. Prove this.

2 E 10.8 . Prove the triangle inequality (stated before Problem 10.3)

using coordinates in the plane, for example as follows: We may
Exercises 249

assume that one vertex of the triangle is the origin and one vertex lies
on the positive x axis, so A = (0, 0), B = (b, 0) where b > 0. Let the
third vertexbe C = ( x, y), y = 0. Then
 the side lengths are ( AB) = b,
( AC ) = x2 + y2 and ( BC )= ( x − b)2 + y
2 by Pythagoras’

Theorem. Prove the inequality ( x − b)2 + y2 < x2 + y2 + b.

E 10.9 .Prove that every triangulation of a convex n-gon (see Prob- 2-3
lem 2.4) has a boundary triangle, i.e. a triangle whose vertices are
successive vertices of the n-gon.

E 10.10 . Can five times a square be the sum of two squares? How 2
about seven times a square?

E 10.11 . Prove that one cannot find four natural numbers a, b, α, β 2

satisfying the equation a2 + b2 = 3(α2 + β2 ).

E 10.12 .Show that the equation 4x4 + 2y4 = z4 has no integer solution 2-3
except x = y = z = 0.

E 10.13 .Is there a square number which lies in the middle between 2
two other square numbers?

E 10.14 . 2

a) Suppose at each integral point n ∈ Z of the number line a natural

number is written. Also suppose that each number is the arithmetic
mean of its two neighbours. What can you conclude about the
numbers?
b) Suppose at each lattice point of the plane (i.e. at each point with
integer coordinates) a natural number is written. Also suppose
each number is the arithmetic mean of its four neighbours. What
can you conclude about the numbers?
c) What can you conclude if instead of a natural number you write a
positive real number at each n in a)?

E 10.15 . Let a graph be given. A black and white colouring of its 2-3
vertices is called desegregated if each white vertex has at least as many
black neighbours as white neighbours, and vice versa.
Prove that there is a desegregated colouring of the vertices.
 250 10 The extremal principle

3 E 10.16 .Suppose in a group of people everyone knows at most three

other people. Prove that it is possible to divide the group into two
smaller groups such that in each of the small groups every person
knows at most one other person in the same small group.
2-3 E 10.17 .Consider a convex polygon and an arbitrary point P in its
interior. Each side can be extended to an infinite line, and we can
drop a perpendicular from P to any of these extended sides. Show
that there must be at least one case where the perpendicular hits the
side itself in its interior. See Figure 10.10.

P
Perpendicular does not hit interior of side

Perpendicular hits interior of side

Figure 10.10 Exercise E 10.17

3-4 E 10.18 .Several barrels of petrol are arranged along a circular track.
The barrels may be filled to different levels, and may be positioned at
irregular intervals. Suppose the total amount of petrol in the barrels
suffices for a car to go around the track once.
You have a car but no petrol to start with. However, you may choose
the barrel at which you start your journey. Can you choose the place
in such a way that you can go full circle, filling up at each barrel
when you reach it?
3-4 E 10.19 .Suppose in a group of 100 people each person knows at least
50 other people. Prove that it is possible to seat the people around
a circular table in such a way that everyone knows his or her two
neighbours.
3-4 E 10.20 . The general inequality of the geometric, arithmetic and
quadratic mean says: Let n ∈ N and a1 , . . . , an ≥ 0. Then

√ a1 + · · · + a n a21 + · · · + a2n
n
a1 · · · · · a n ≤ ≤ ,
n n
Exercises 251

and each inequality becomes an equality if and only if a1 = · · · = an .

Prove this.

E 10.21 . Among all trapezoids inscribed in a circle, which one has 2

the largest area?

E 10.22 .Let n ∈ N. We consider numbers 0 = x0 < x1 < · · · < xn = 3

1. For which values of the xi does the sum ( x1 − x0 )2 + ( x2 − x1 )2 +
· · · + ( xn − xn−1 )2 take on its smallest possible value?

E 10.23 .Prove that the equilateral triangle has the largest area among 3-4
all triangles of a given perimeter.

E 10.24 .Suppose you are given a finite number of points in the plane, 4
not all on a line. Is there always a line which contains precisely two
of the points?

E 10.25 .Let n ∈ N, n ≥ 2. Show that on every convex billiard table 4

without corners you can find n boundary points with the following
property: a billiard ball16 shot from one of the points to the next
will hit the boundary at precisely the n points in the given order,
come back to the initial point and then keep repeating the same track.
See Figure 10.11 for an example of an eternal triangular track on an
elliptical billiard table.

B C

Figure 10.11 An eternal billiard ball track on an elliptical table, n = 3 (B

and C lie a little below the middle)

16 We assume that the billiard ball is a point. The claim remains true for ‘real’

billiard balls, if their radius is smaller than the smallest radius of curvature of the
boundary. Also we assume there is no friction.
11 The invariance principle

“When things change, pay attention to what remains constant.” While

this may sound like ideology, it is really one of the fundamental
ideas pervading all of mathematics, and a powerful problem-solving
strategy.
Many complex processes are composed of simple steps: from a
small number of legal moves the most complex chess games can arise.
If you go for a walk in a city then each step is simple, but after a
while you may get lost. A physical system changes only slightly in a
moment, but after some time it may have changed its state completely,
and almost unpredictably.
The invariance principle helps you gain information about such
processes in spite of their complexity. The problems in this chapter
will show you how and where to use it for solving mathematical
problems. Along the way you will find out surprising facts about the
15-puzzle and peg solitaire, and learn about permutations and their
signature, fundamental concepts of mathematics.

11.1 The invariance principle, first examples

We start with a few simple examples to show what the invariance

principle1 is and how to use it.

? Problem 11.1
Two joggers A and B run on a straight track. B starts one metre ahead of
A. They run at the same speed.
Show that A can never catch up with B.

! Solution
Since A and B have the same speed their distance is constant. Since it
is positive at the beginning it will always be positive. !
1 invariant = not changing

© Springer International Publishing AG, part of Springer Nature 2018 253

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7_11
254 11 The invariance principle

A B A B
initial position later
position

? Problem 11.2
You have a board composed of 6 × 6 squares. A coin lies on the lower left
corner square. In one move the coin can be moved from its present square
to one of the squares diagonally adjacent to it.
Show that the coin can never get into the lower right corner square, no
matter how long you play.

Think about it!

! Solution
Colour the squares black and white as on a chessboard. The lower
left square is black, say. In each move the coin will remain on the
same colour, so it will always remain on black squares. Since the
lower right square is white the coin can never get there. The coin is
like a (lame) bishop in chess. !
What do the two problems have in common? Something remains
constant (in each moment or in each move), therefore it must remain
constant over an arbitrary time span or after arbitrarily many moves.
This is the invariance principle. In short:

Invariance principle

When things change look out for what remains constant!

The examples show that the invariance principle helps with proofs
of impossibility: B cannot catch up with A, the coin can never reach
the lower right square.
11.1 The invariance principle, first examples 255

A
B C
Figure 11.1 Boards for domino tilings

? Problem 11.3
Is it possible to tile a 5 × 5 board with dominoes? How about a 6 × 6 board?

The boards are shown in Figure 11.1 A and B. A domino is a 2 × 1

rectangle, and to tile means to cover without gaps and without
overlaps. Each domino covers two adjacent 1 × 1 squares.

Think about it!

! Solution
A domino covers two squares. Therefore any board tiled with domi-
noes must have an even number of squares. The 5 × 5 board has 25
squares, so it cannot be tiled.
The easiest way to show that a 6 × 6 board can be tiled with
dominoes is to give an example of a tiling, see Figure 11.2. !
Be careful: For proving that the 6 × 6 board can be tiled it is
not enough to argue that it has an even number of squares. For
example, the board has an even number of squares, but it
obviously cannot be tiled with dominoes.

The proof of impossibility for the 5 × 5 board can be viewed as an

instance of the invariance principle: think of putting dominoes on the
256 11 The invariance principle

Figure 11.2 Example of a domino tiling

board one by one. The number of covered squares increases by 2 in

each step, so its parity stays the same. Since this number starts out at
0 it can never become 25.
Recall that the parity of an integer is its property of being even or
odd. Parity is often a good candidate for an invariant, but sometimes
it does not help, as the next example shows.

? Problem 11.4
Is it possible to tile the board in Figure 11.1 C with dominoes?

 ! Investigation and solution

G
 The board has 36 − 2 = 34 squares, an even number. So it might
be possible to tile it, but we cannot be sure. In any case we cannot
use the parity argument to prove that it is impossible. Try to find a
tiling!
 You didn’t manage to tile the board? Then we formulate the
conjecture: It is impossible.
 How could we prove the conjecture?

Think about it!

We could try out all possibilities – but there is no hope of doing

that, there are just too many, and there is no obvious way to check
them systematically.
11.1 The invariance principle, first examples 257

 Therefore we should look for a new invariant: some quantity that

remains the same when we add a domino to the board. In addition,
this quantity should have different values for the empty board and
the (hypothetically) completely tiled board, otherwise it won’t be
useful for our proof of impossibility.
The parity of the number of covered squares is an invariant, but it
does not satisfy the additional condition since 0 and 34 are both
even.
 Idea: We use the chessboard colouring from above. A domino will
always cover a white square and a black square, no matter where
you place it. Therefore any board tiled with dominoes must have
the same number of white and black squares.
 The two corner squares which we removed have the same colour
(say white). Therefore our board has more black squares than
white squares, so it cannot be tiled with dominoes. G!

 Review of Problem 11.4

The essential idea was to introduce the chessboard pattern. This was
not present in the original problem. Often it is useful to introduce
additional structure. This has already helped us to solve Problem
11.2. 
Summary on tilings: In order to decide whether the boards A, B,
C in Figure 11.1 can be tiled with dominoes we argued as follows:

J Using the invariant ‘parity’ we proved that board A cannot be

tiled.

J For boards B and C the parity argument is useless. It does not

help us decide whether a tiling is possible.

J Board B can be tiled. We proved this by giving an example of a

tiling.

J Board C has no tiling. We proved this using another invariant: the

difference of the numbers of covered white and black squares.
258 11 The invariance principle

11.2 How to use the invariance principle

Now we can describe more precisly how to use the invariance princi-
ple.

Problem: Suppose we have a problem which involves a process

which advances by steps through various states. See Table 11.1.
Our goal is to understand which states can be reached from a given
initial state. It is known which steps are possible from any state.
Usually a single step is very simple, but after several steps things get
complicated.2
An invariant of the process is a rule assigning to each possible
state a number (or a symbol, a colour, a word etc.), which remains
constant during each step.

jogger coin on square domino tilings

process running moving the coin tiling the board
step – move the coin to ad- add a domino to the
jacent square board
state positions of position of the coin tiling of part of the board
the joggers
invariant distance of colour of the square parity of the number
the joggers that the coin is on of covered squares; dif-
ference white/black cov-
ered squares

Table 11.1 Problems 11.1–11.4

Preparation: We familiarise ourselves with the problem (using ex-

amples etc.) and arrive at a conjecture about which states can be
reached and which cannot.

Proof of existence for reachable states: The simplest way to

show that a state can be reached is to establish a concrete sequence of
steps leading to this state.

2 The
process in the jogger problem 11.1 is continuous: it does not have single
steps. However, the question and the solution are analogous. See also Section 11.4.
11.2 How to use the invariance principle 259

Proof of nonexistence for non-reachable states In order to show

that a state a cannot be reached we look for an invariant of the process
which takes different values at the initial state and the state a.
As usual it may happen that we need to correct our initial conjec-
ture, or that an invariant which we are trying to use does not help.
And as usual the name of the game is: try again, don’t give up!
As for the extremal principle, you will need to use your creativity
to find a suitable invariant. There are no limits: if it works then it is
permitted.
We can formalise the problem and the invariance principle as
follows:3
We are given a set M, the set of states, and a subset S ⊂ M × M, the set
of steps.
(Interpretation: ( a, b) ∈ S means that one can get from state a to state b in
one step.)
We say that a state b ∈ M is reachable from a state a ∈ M if there is
a run of the process starting with a and ending with b. Here a run is
a sequence a, x1 , x2 , . . . , xn , b of elements of M for which ( a, x1 ), ( x1 , x2 ),
. . . , ( xn , b) ∈ S, for some n ∈ N.
An invariant is a map I : M → T into an arbitrary set T which satisfies:
for all steps ( a, b) ∈ S we have I ( a) = I (b).
Invariance principle: If I is an invariant and a, b ∈ M satisfy I ( a) = I (b)
then b is not reachable from a.

Example

In Problem 11.2 M is the set of squares of the board, S is the set

of pairs ( a, b) of squares a, b which are diagonally adjacent to
each other, T = {black, white} and I ( a) is the color of square a
in the chessboard colouring, for each a ∈ M.

3 This formalisation is not needed for understanding the rest of the chapter. It
serves to make the concepts precise, and illustrates the way many mathematical
texts are written. However, it will not usually help you to get good ideas for finding
invariants. For that it is more important to have an intuitive feeling for the invariance
principle, and you get this by doing examples, solving problems. Conversely you
may conclude that you should always look for the idea behind the formal definitions
that you find in many textbooks on mathematics.
260 11 The invariance principle

11.3 Further examples: number games, permutations,

puzzles and peg solitaire

Often we describe processes as games. Then the states are the possible
positions of the game and the steps are the moves that are possible in
a given position.

? Problem 11.5
Let n be an odd natural number. We play the following game: We write
down the numbers 1, 2, . . . , 2n. A move consists of choosing two numbers
and replacing them by their difference. We play until only one number is
left.
Can the remaining number be zero?

 Investigation
G
 Let us consider an example with n = 3. We mark the two numbers
chosen for the next move by a dot:

1̇ 2 3 4 5 6̇
2 3 4 5̇ 5̇
2̇ 3̇ 4 0
1̇ 4̇ 0
3̇ 0̇
3̇

In this case the number 3 remains. We could have proceeded

differently; for example we could have chosen 1 and 2 in the first
step. Play a few different runs of the game. You will notice that
you never get zero at the end. So we formulate:
Conjecture: For n = 3 the remaining number cannot be zero.
 How could we prove that? Certainly we should not try out all
possibilities. There are too many (and in any case we want to
understand the problem for any odd n).
 The form of this problem matches exactly the scheme described
above: we have a process (the game) consisting of single steps (the
11.3 Further examples 261

moves), and we want to show that a certain final state is impossible.

So we should look for an invariant.
 What are the states? They are finite sequences of numbers: the
lines in the example above. So we look for a rule assigning to each
such sequence a number (or colour or ...), which remains constant
during each move, no matter what move we make.

Think about it!

 One of the simplest things you can do with a finite sequence of

numbers is to add all the numbers. Let’s try this in the example:
sequence sum
1̇ 2 3 4 5 6̇ 21
2 3 4 5̇ 5̇ 19
2̇ 3̇ 4 0 9
1̇ 4̇ 0 5
3̇ 0̇ 3
3̇ 3
Do you notice something? The sums are always odd. We get an
idea:
Conjecture: The parity of the sum of the numbers is an invariant.
If we could prove this then we would be done for n = 3 since the
sum starts out as 21, so it can never become zero. Then for the
general case we only need to check whether the sum 1 + 2 + · · · +
2n is always odd if n is odd. Try it yourself! G 

Think about it!

! Solution of Problem 11.5

The number remaining at the end cannot be zero. In order to show
this we prove two propositions:
262 11 The invariance principle

1. The sum 1 + 2 + · · · + 2n is odd.

2. The parity of the sum of the remaining numbers stays the same
during each move.

1. and 2. together imply that the single number remaining at the end
must be odd, so it cannot be zero.
Proof of 1.: We have

2n(2n + 1)
1 + 2 + · · · + 2n = = n(2n + 1).
2
By assumption n is odd. Since 2n + 1 is also odd, it follows that
n(2n + 1) is odd.
Proof of 2.: Suppose at some stage in the game the numbers a1 , . . . , ak , x, y
remain, where x, y are the numbers chosen for the next move. Here
k ∈ N0 could be zero. Without loss of generality we may assume
x > y. Then after the next move the numbers a1 , . . . , ak , x − y remain.
So we need to show

a1 + · · · + a k + x + y ≡ a1 + · · · + a k + x − y mod 2

(for the congruence notation see Chapter 8; this just means the two
numbers have the same parity). The congruence is true since the
difference of the left- and right-hand sides is 2y, hence even. !

 Review of Problem 11.5

The setup (moves etc.) and our conjecture of impossibility suggested
that we use the invariance principle. But it was not obvious what
to use as an invariant. The idea to use the sum of the numbers at a
given stage of the game yielded a solution. Parity helped, as is often
the case. 
The signature of a permutation and an 8-puzzle problem

A permutation is an ordering of the numbers 1, 2, . . . , n. For n = 3

there are 6 permutations: 123, 132, 213, 231, 312, 321. We have already
encountered permutations in Chapter 5, where we determined that
there are n! of them.
11.3 Further examples 263

The following problem will guide you to the notion of the signature
of a permutation, one of the fundamental invariants of mathematics.

? Problem 11.6
Write down the numbers 1, 2, . . . , n in any order. A move swaps any two
adjacent numbers. Is it possible to arrive at the initial order after an odd
number of moves?

 Investigation
G
 Again we are talking about a process: the states are the permuta-
tions, the moves are the neighbour swaps.
 Try out a few examples. For example, starting with 213 we could
swap 1 and 3 and arrive at 231; then we could continue as 321, 312,
132, 123, 213, which is the order we started with. We have made
six moves. You will notice that whenever you arrive back at the
initial order you will have made an even number of moves. We
conjecture that this is always the case. How could we prove that
in general?
 It is not clear how to use the invariance principle here. But if we
could assign a number to each permutation in such a way that this
number changes its parity with each move then our conjecture would
follow immediately. This is our plan.
 What happens when we swap two adjacent numbers? Two num-
bers that were in natural order (i.e. the smaller number first) will
now be out of order, and conversely.
 Idea: this suggests that we consider the number of inversions of
a permutation. An inversion is a pair of places i < j where the
number at place i is bigger than the number at place j.
Example: How many inversions does the permutation 3412 have?
We consider all pairs of places systematically: 34 (at places 1 and
2) is not inverted, 31 inverted, 32 inverted, 41 inverted, 42 inverted,
12 not inverted. So there are 4 inversions.
 How does the number of inversions change when we swap two
adjacent numbers? Clearly, if these numbers were not inverted
264 11 The invariance principle

before the swap then they will be afterwards, and if they were
inverted then they won’t be afterwards. The position of these two
numbers relative to all other numbers remains unchanged, as do
the positions of any pair of the other numbers.
So the number of inversions changes by one when we swap two
adjacent numbers. In particular the number of inversions will
change its parity, which is what we wanted. Here is an example.
The signature is the parity of the number of inversions.

number of
permutation signature
inversions
213 1 o
231 2 e
321 3 o
312 2 e
132 1 o
123 0 e
213 1 o

Figure 11.3 Example for Problem 11.6; o = odd, e = even

 At the beginning and the end we have the same permutation, so

the number of inversions is the same. Therefore the parity of the
number of inversions must have changed an even number of times;
that is, we must have made an even number of moves. 
G

We summarise the definitions and write up the solution neatly.

Definition An ordering of the numbers 1, 2, . . . , n is called a per-

mutation of 1, . . . , n. We write it as4 ( a1 , a2 , . . . , an ). An inversion
is a pair of numbers i, j ∈ {1, . . . , n} satisfying

i<j and ai > a j .

The signature of the permutation ( a1 , . . . , an ) is the parity of the

number of its inversions. We call the permutation even or odd if
its signature is even or odd, respectively.
11.3 Further examples 265

Often we denote permutations by small Greek letters, for example π

(pi) or σ (sigma).

! Solution of Problem 11.6

We first prove a lemma.

Lemma Swapping two adjacent numbers in a permutation

changes its signature.

Proof.
Let π = ( a1 , . . . , an ) be a permutation. Let 1 ≤ k < n and σ =
( a1 , . . . , ak+1 , ak , . . . , an ) be the permutation obtained after swapping
ak and ak+1 . We show that the number of inversions of σ is one bigger
or one smaller than the number of inversions of π. This implies the
lemma. For this we consider the possible inversions i < j of π and σ:

J If neither i nor j lie in {k, k + 1} then σ has the same numbers at

positions i, j as π, so i, j is an inversion of σ if and only if it is an
inversion of π.

J Any inversion of π with i = k and j > k + 1 will lead to an

inversion of σ with i = k + 1 and the same j > k + 1 since ak moves
to position k + 1 in σ. Similarly, any inversion of π with i = k + 1
and j > k + 1 leads to an inversion of σ with i = k and j > k + 1.
Therefore σ and π have the same number of inversions with i ∈
{k, k + 1} and j > k + 1.

J Similarly, σ and π have the same number of inversions with i < k

and j ∈ {k, k + 1}.

J It remains to consider the case i = k, j = k + 1. If π has an

inversion at the position pair k, k + 1, i.e. if ak > ak+1 , then σ does
not have an inversion at this position pair. If π does not have an
inversion there then σ does.

4 You can also view a permutation ( a , . . . , a ) as a bijective map {1, . . . , n } →

1 n
{1, . . . , n}, n → an . For example, the permutation (2, 1, 3) corresponds to the map
1 → 2, 2 → 1, 3 → 3.
266 11 The invariance principle

All in all, we see that the number of inversions of σ differs from the
number of inversions of π by 1. q. e. d.
We can now solve Problem 11.6: Since the signature changes at each
move, and is the same at the beginning and the end, the number of
moves must be even. !
Remark

The signature is not an invariant but a semi-invariant for this

problem: while it does not remain constant, it changes in a simple
way with each move, and the change remains predictable even
after many moves. This was sufficient for solving the problem.

We note two other important properties of the signature. You should

prove them as Exercise E 11.6.

Theorem (Properties of the signature)

a) When you move a number in a permutation to the left or right,

skipping k other numbers, then the signature changes if k is
odd and remains the same if k is even.
b) When you swap any two numbers in a permutation then the
signature changes.

Example for a): 31254 is odd, and if you move 1 to the right, skipping
2 and 5, you obtain 32514, which is also odd.
Part b) generalizes the previous lemma on swapping adjacent num-
bers.
Here is a beautiful application of the signature:

? Problem 11.7
Consider a sliding puzzle in the shape of a 3 × 3 square, filled with 8 square
tiles numbered 1 to 8 as in Figure 11.4, so one tile is missing. A move slides
an adjacent tile into the empty space.
Somebody takes out the tiles 7 and 8 and puts them back in the oppo-
site order. Can you get back from this arrangement to the original arrange-
ment by sliding moves?
11.3 Further examples 267

This puzzle is called the 8-puzzle. Try it!

1 2 3 1 2 3

4 5 6 4 5 6

7 8 8 7

Figure 11.4 Left: original position of the 8-puzzle, with two possible moves;
right: tiles 7 and 8 are interchanged

Think about it!

! Solution

The states are all possible arrangements of the tiles. By reading the
numbers row by row and ignoring the empty space we can associate
a permutation of the numbers 1, 2, . . . , 8 with each arrangement. The
original arrangement corresponds to the permutation 12345678; the ar-
rangement on the right in Figure 11.4 corresponds to the permutation
12345687.
When we move a tile horizontally then the corresponding permu-
tation does not change. Let us see what happens when we move a
tile vertically: Let m be the number on the tile. If we slide m down
then in the corresponding permutation it moves right, skipping two
numbers. If we slide m up then it moves left, skipping two numbers.
By the theorem above the signature does not change, whichever way
we move m.
Therefore the signature of the associated permutation is an invari-
ant of the game. Since the permutations 12345678 and 12345687 have
different signatures it is impossible to get from one arrangement to
the other, no matter how many moves we make. !
The analogous problem for a 4 × 4 square (the so-called 15-puzzle,
see Exercise E 11.9) has an interesting history. It is reported that in
268 11 The invariance principle

the year 1880 the puzzle created a craze first in the US and then in
Europe, after someone had promised a 1000 Dollar award for finding
a sequence of moves from the arrangement with 14, 15 swapped to
the original arrangement.

Three-coloured balls and peg solitaire

? Problem 11.8
You have a box containing 4 red balls, 5 green balls and 6 blue balls. In
each move you take out any two balls of different colours and replace them
by a ball of the third colour.5 You play until you cannot move any more.
Suppose a single ball remains at the end. Can you predict which colour
it is?
If you start with 4 red balls, 6 green balls and 8 blue balls, is it possible
that a single balls remains at the end?
Try it. Observe carefully the numbers of red, green and blue balls in
the course of the game. Do you notice anything? Can you prove it?
What can you conclude from it?

Think about it!

 ! Investigation and solution

G
 Denote the numbers of red, green and blue balls by r, g, b, respec-
tively. Let us observe what happens to r, g during a move: if you
take out red - green then they change to r − 1, g − 1. If you take
out red - blue then they change to r − 1, g + 1. And if you take
out green - blue then they change to r + 1, g − 1. Can you name
something that does not change in each case?

5 It is assumed that you have an extra supply of balls of each colour outside the
box.
11.3 Further examples 269

Think about it!

 If r, g have the same parity before the move then they have the
same parity after the move. And if they have different parity before
the move they have different parity after the move. Put differently,
the parity of their difference does not change, no matter which move
we make.
 By symmetry the same is true for r, b and for b, g: what remains
constant is whether they have the same or opposite parity.
 Suppose a single red ball remains at the end. Then r, g, b are 1, 0, 0
at the end, so r has different parity from g and b. Therefore, r must
have had different parity from g and b at the beginning.
Similarly, if a green or blue ball remains at the end then g or b
respectively must have had a different parity from the other two
numbers at the beginning.
 Since the starting values of r, g, b are 4, 5, 6 and only 5 has a different
parity from the other two numbers, it follows that the remaining
ball is green.
 In the case 4, 6, 8 all parities are equal at the start, so they must
always be equal. Therefore it is not possible that a single ball
remains at the end. !
G
Remarks

J We did not show that in the case 4, 5, 6 it is actually possible

to choose the moves so that a single ball remains at the end
(which we proved would have to be green). This was not part
of the question. On the other hand, you will easily find a way
to achieve this. However, this is not completely trivial: if you
start badly then you can manoeuvre yourself into a dead end.
For example if you first take out green - blue you get 5, 4, 5; if
you now take red - blue five times then you get 0, 9, 0, and the
game is over.
270 11 The invariance principle

J Our solution yields a necessary condition for the possibility of

leaving one ball only at the end: the initial numbers must be
not all odd, and not all even. Is this condition also sufficient?
(No, see the previous remark and Exercise E 11.19).

J This pleasing problem has more to offer: it can be used to give

surprising insights into the famous game of peg solitaire. For
example: if you manage to leave a single peg anywhere, then
you could easily have left a single peg in the centre instead.
See Exercise E 11.23.

11.4 Going further: knots, conserved quantities, and why

you should care about proofs of impossibility

Invariants of continuous processes

The process in the jogger problem 11.1 is not composed of discrete

steps. Its state changes continuously. This is best modelled using
notions of mathematical analysis. In this context the invariance
principle is a theorem of analysis:

If the derivative of a function is zero everywhere then the func-

tion is constant.

More precisely: Suppose the function f : [ a, b] → R is differentiable

and f
( x ) = 0 for all x ∈ [ a, b]. Then f is constant, in particular
f ( a) = f (b). Loosely speaking: if f doesn’t change “in the small”
anywhere – that is, its graph has only horizontal tangents – then it
does not change “in the large”, i.e. from a to b.
In the jogger problem we take f ( x ) = B( x ) − A( x ) to be the
distance between the joggers, where x is time and A( x ), B( x ) are
the locations of the joggers, measured along the number line, at
time x. The speeds of the joggers at time x are the derivatives
A
( x ), B
( x ), and since these are assumed to be the same we have
f
( x ) = B
( x ) − A
( x ) = 0 for all x, so f is constant.
This continuous invariance principle is the basis for the proof that
solutions of many differential equations are unique.
11.4 Going further 271

Examples of the invariance principle in various areas of

mathematics

The idea of invariants pervades all areas of mathematics. Here are

some examples.
As in the case of the extremal principle in Section 10.4, you may
encounter here some words of higher mathematics that you don’t
know yet. But keep reading anyway; you will meet them again when
you learn more maths, and then remember that they fit into this grand
scheme.

J In Section 4.5 you learned about the Euler characteristic. This is

an invariant of topological spaces – these are for example surfaces
in space, like a sphere or a torus. The Euler characteristic is
invariant are deformations of these spaces. These are continuous
processes.

J For a plane curve which is closed (i.e. starting point and endpoint
are the same) and which is allowed to intersect itself, and for
a point P not lying on the curve, one can consider the winding
number: the number of times the curve “winds around” P. The
winding number is an invariant of such curves under deformations,
assuming that during the deformation the curve always avoids
P. It is not easy to turn this intuitive description of the winding
number into an exact definition. This can be done using tools from
topology or from complex function theory.
The winding number has many applications. For example it can
be used for a very elegant proof of the fundamental theorem of
algebra (any polynomial has a zero in the complex numbers).6

J A very appealing and important application of invariants is the

investigation of knots. The rope on the left in Figure 11.5 can
easily be unknotted, while the second one – the so-called trefoil
knot – cannot; at least that’s what our experience tells us. But can
we be sure, have we tried all possibilities?7 And how about the
third and fourth rope? How can we decide generally whether a

6 See the Wikipedia entry for ‘Fundamental theorem of algebra’.

7 Remember Problem 4.5!
272 11 The invariance principle

Figure 11.5 Four ‘knots’

given closed rope can be unknotted or not?8 Can one prove that
the three knots on the right can not be unknotted?
These questions are investigated in knot theory, a fascinating mod-
ern mathematical theory which has applications for example in the
understanding of DNA, which encodes our genetic information. In
knot theory one constructs invariants which are easy to compute
and which can be used to prove, for example, that the second and
third knot in Figure 11.5 cannot be unknotted.
However, up to this day no one has succeeded in proving that
these invariants are sufficient for deciding whether any given knot
can be unknotted. We don’t know whether they are.
An even more ambitious goal would be to find a complete set of
invariants, which would enable us to decide for any two knots
whether they can be deformed into each other.9
An extension cable is useful for playfully investigating knots (knot
it, then plug the ends together).
By the way, the knot on the right in Figure 11.5 can be unknotted.

J Processes and their long-term behaviour are the subject of the the-
ory of dynamical systems. Important tools in this theory are invari-
ant manifolds and Lyapunov functions (these are semi-invariants:
they are monotone increasing or decreasing during a run of the
process).
8 Wemake this more precise: A knot is a closed curve in space which does not
intersect itself. Two knots are called equivalent if one can be deformed into the other,
where also during the deformation no self-intersections are allowed. A knot can be
unknotted (alternatively, it is called trivial) if it can be deformed to the unknot, i.e.
a circle.
9 The book (Adams, Colin C., 2004) gives an informal introduction to knot

theory, with applications in physics, chemistry and biology; on the web page
http://katlas.org/wiki/Main_Page you find pictures of many knots.
11.4 Going further 273

J The invariance principle is also of fundamental importance in

physics. For example, the law of conservation of energy says: the
energy of any closed system remains constant, no matter how
the system changes. Such physical invariants are usually called
conserved quantities. Other conserved quantities are momentum
and angular momentum. Using these invariants skilfully will
enable you to find simple solutions for some apparently complex
problems.10 Albert Einstein discovered (special) relativity by
searching for a theory in which the laws of nature are described
in a Lorentz invariant way and in which the speed of light is
constant.

Invariance is often understood in a broader sense, not just referring

to constancy in a process. Here are some examples which you meet
quite early in university mathematics.

J Invariance under maps: Modern mathematics is usually formu-

lated in the language of sets and maps, since this gives it a unified
basis. Sets may carry additional structure. For example, there are
vector spaces, groups or topological spaces. Then the invertible
maps which preserve the extra structure, so-called isomorphisms,
are of special importance. In the examples they are linear isomor-
phisms, group isomorphisms and homeomorphisms, respectively.
An invariant of a structure is a quantity (or some other mathe-
matical structure, for example a group) which can be associated
with any instance of the structure, and which is the same for two
instances if there is an isomorphism between them.
For example the dimension is an invariant of vector spaces, and
the Euler characteristic is an invariant of topological spaces.11

J Invariance as independence of a result (of a calculation) from the

details of its derivation. For example, we can find the dimension
10 Example: Suppose an object falls down, starting at height h. What is its velocity
v when it hits the ground? The easiest way to compute v is by conservation of energy:

potential energy lost = kinetic energy gained, so mgh = 12 mv2 , hence v = 2gh,
where g is the gravitational acceleration of the earth. This method also works if the
object slides down a slope (without friction) – no matter how steep the slope is, and
even if it has variable slope, the result is always the same!
11 This implies in particular the invariance of the Euler characteristic under

deformations that we talked about in Section 4.5.

274 11 The invariance principle

of a vector space by choosing a basis and counting its vectors – any

other basis would yield the same result. The Euler characteristic of
a surface can be computed by drawing a graph on the surface and
counting its vertices, edges and faces (see Section 4.5 for details) –
the result is independent of the choice of graph. Determinant and
trace of a linear map from a vector space to itself are computed
by taking the determinant or trace of a matrix representing the
map – the result is independent of the choice of basis used to
compute that matrix. In differential geometry many quantities,
for example curvature, can be defined and computed using local
coordinates. Such quantities will only be geometrically meaningful
if their calculation in a different coordinate system yields the same
result.
We could go on and on. Keep your eyes open!

Summarising, we can say that invariance is one of the fundamen-

tal ordering principles of mathematics – in addition to its use as a
problem solving strategy which you learned about in this chapter.

Why should you care about proofs of impossibility?

If we prove that something is impossible then we can stop spending

energy on trying to do it. Such knowledge improves our understand-
ing of the subject, and we may go on to modify the question or take a
different point of view.
There is also a very practical side to proofs of impossibility. Cryp-
tography provides a good example: Many a convenience of everyday
life relies on the possibility of encrypting information: using cash
cards, shopping on the internet etc. A common algorithm for encryp-
tion is the RSA algorithm. It is safe as long as no one has found a
fast way to decompose very large numbers (several hundred decimal
digits) into their prime factors.12
But up to now nobody has succeeded in proving that no fast way
to do this exists. Without such a proof we have to live with the
residual uncertainty that someone may find a fast way – or has found
it already, but just hasn’t told us yet. A proof of impossibility would

12 ‘Fast’is not a mathematical but rather a practical notion. You may think of ‘in
less than a thousand years on state-of-the-art computers’.
11.5 Toolbox 275

be very welcome for us all here, besides being a big breakthrough in

research.

11.5 Toolbox

The invariance principle is a fundamental tool for proving that some-

thing is impossible or non-existent: If you can reformulate the prob-
lem in terms of a process then you should look for invariants. To
find an invariant that yields a desired proof of impossibility or non-
existence may be difficult, and may require creativity and many
attempts. You also learned about the signature of a permutation
which can be useful as an invariant or semi-invariant. In Problem
11.4 you learned about another important strategy: introducing ad-
ditional structure (like a chessboard pattern), that is, a structure that
is not present in the problem but helps with its solution.

Exercises

E 11.1 .We shred a sheet of paper. In each step we take one piece 1
and tear it into 3 or 5 parts. Is it possible to obtain precisely 100
pieces?

E 11.2 .To show that there is no tiling of a 5 × 5 square with domi- 1

noes we used a parity argument. Can you show this using a chess-
board colouring instead?

E 11.3 .We colour the board in Figure 11.1 C as follows: the lower 1
left 3 × 3 square and the upper right 3 × 3 square are black, the rest is
white. Can you use this colouring instead of the chessboard colouring
to prove that the board cannot be tiled with dominoes?

E 11.4 .Let n ∈ N be odd. Write down the numbers 1, 2, . . . , n. In 1-2

each step delete any three numbers and replace them by their sum.
Show that after some time only one number is left, and determine
this number.

E 11.5 .Consider Problem 11.6 about permutations again. Since only 2

adjacent numbers are swapped in each step you might find it more
natural to count inversions of adjacent numbers only, that is, pairs
 276 11 The invariance principle

i, i + 1 having ai > ai+1 . Can you solve the problem using this number
instead of the total number of inversions?

2 E 11.6 . Prove the theorem about the properties of the signature,

stated before Problem 11.7.

2 E 11.7 .Find a bijection between the set of even permutations and

the set of odd permutations of 1, . . . , n. How many even permutations
are there?

2 E 11.8 . A chocolate bar consists of four by six squares. You want

to break it into single squares. You are only allowed to break along
the lines between the squares, and in each step you can only break
up one of the pieces you have made so far. How many steps do you
need at least? How many at most?

3 E 11.9 . Solve the analogue of Problem 11.7 for the 4 × 4 square

where initially the tiles 14 and 15 are swapped.

2-3 E 11.10 .From a 12 × 12 board you remove one little square from each
of three corners. Is it possible to tile the remaining board using tiles
of the form ?

3 E 11.11 .A bathroom floor is tiled with 1 × 4 and 2 × 2 tiles. One of

the 1 × 4 tiles breaks, but we only have a 2 × 2 tile to replace it. Is
it possible to rearrange the tiles so that they still completely tile the
bathroom floor?

2 E 11.12 .Consider the boards in Figure 11.6.

a) Show that it is impossible to tile the right-hand board (where three
squares have been removed) with tiles of the form .
b) Prove that you can tile the left-hand board with tiles of the form
.
c) Is it possible to tile the left-hand board with tiles of the form
?

2-3 E 11.13 .Consider a strip of 2n unit squares. The squares are coloured
alternately black and white. In each move choose a contiguous group
of squares (for example the third square, or squares two to four) and
swap all colours in this group. Clearly you can make all squares
white using n moves. Is this also possible with fewer moves?
Exercises 277

complete board board with three squares removed

Figure 11.6 Boards for Exercise E 11.12

E 11.14 .A heavy square armchair can only be moved by turning it 2-3

around one of its corners by 90 degrees. Can you move it multiple
times so that in the end it is next to its original position, and the back
of the chair faces in the same direction as at the beginning?

E 11.15 . The numbers 1,0,1,0,0,0 are written round a circle in this 3

order. In each move you are allowed to add 1 to two neighbours or
to subtract 1 from two neighbours. Can you ever make all numbers
equal zero? For which initial arrangements of integers round a circle
is it possible to do this?

E 11.16 . Three frogs sit at the points of the plane with coordinates 3
(0, 0), (0, 1) and (1, 0). Now they start leaping. In each leap one
frog leaps across another frog so that the leaped-over frog sits in the
middle between the starting and landing positions of the leaping frog.
Can a frog ever land at position (1, 1)? (These are super-frogs: they
can leap a long way.)

E 11.17 . Starting with the numbers 1, 2, . . . , 4n − 1 replace any two 2

numbers by their difference. Repeat until only one number is left.
What can you say about its parity?

E 11.18 .Let P be a convex n-gon where n is even. Let p be a point 2

in its interior which does not lie on any diagonal. We consider the
number of triangles whose vertices are vertices of P and which contain
p. Find this number for some examples of P and p. What do you
observe? Proof?

E 11.19 . Consider Problem 11.8 with initially r red, g green and b 2-3
 278 11 The invariance principle

blue balls, where r, g, b ∈ N0 . For which numbers r, g, b is it possible

to play in such a way that only one ball is left at the end?

2-3 E 11.20 .Consider the following game. You start with g ∈ N green
and r ∈ N red balls in a box. In each move you take out two balls.
If they have the same colour then you put a green ball back into the
box, otherwise a red ball. We assume you have a supply of extra balls
as needed.
You play until only one ball is left in the box. Can you predict from r
and g what colour this ball is?

2 E 11.21 .Imagine a museum with one hundred identical square rooms,

with a floor plan as in Figure 11.7, where the heavy lines are doors.
The entry is at the bottom left, the exit is at the top right.
Is it possible to enter the museum, visit each room exactly once and
then leave?

exit

entrance

Figure 11.7 The museum in Exercise E 11.21

3 E 11.22 . An island is inhabited by a kind of chameleon which can

take on the colours red, green and blue. Whenever two chameleons
of different colour meet then they will both change their colour to the
third, remaining colour. Suppose the numbers of red, green and blue
chameleons are r, g, b ∈ N0 initially. For which values of r, g, b can it
happen that after a while all chameleons have the same colour? For
which values is this colour red?
Exercises 279

E 11.23 . Peg solitaire is a game played on the board shown in 3-4

Figure 11.8. The circles are holes in the board. They are occupied
by a peg or are empty . In each move you take a peg and jump
horizontally or vertically over an adjacent peg, landing on the empty
hole directly behind it. Then you remove the jumped peg. So in
the initial position shown on the left in the figure, four moves are
possible, and after any of them there will be two empty holes. The
game ends when you cannot move any more.
The aim is to play in such a way that only one peg is left at the end.
a) Show that in a successful game the remaining peg can only be
at one of the five positions shown in the middle figure. Anal-
yse the statement “I managed to leave one peg at the end, but
unfortunately it was not in the center.”
b) Show that no successful game is possible if you start with the
initial position shown in the right-hand figure.

starting position possible positions of the a different starting

last peg position

Figure 11.8 Peg solitaire, exercise E 11.23

E 11.24 . Infinite solitaire is played like solitaire but on an infinite 4

board: there is a hole at each lattice point (n, m), n, m ∈ Z, of the
plane, and initially all holes in the lower half plane are occupied by
pegs, that is all holes at points where m ≤ 0. The aim is to advance
as far as possible into the upper half plane. So you want to get a peg
to a position (n, m) with big m. What is the largest possible m?

E 11.25 .In this problem it is assumed that you know Rubik’s cube. 4
From the outside you can see 26 little coloured cubes called cubelets.
When you play with the cube you will notice that from the starting
280 11 The invariance principle

configuration, with each face just one colour, you can never get to a
configuration where
a) all except two cubelets are in the correct position, or
b) all corner cubelets are in the correct position and all except one
are correctly oriented, or
c) all edge cubelets are in the correct position and all except one
are correctly oriented.
Prove that it is really impossible to reach any of these configurations
(unless you disassemble the cube).
A A survey of problem-solving strategies

This is a survey of the problem-solving strategies that you have

learned about in this book.
The main steps in problem-solving are:
1. Understand the problem
2. Investigate the problem
3. Write up the solution properly
4. Review your solution
Usually the hardest step is the investigation. But the other steps are
just as important: obviously, understanding the problem is essential
and will help in the investigation. By writing up the solution properly
you communicate it to others and also check that your solution is
complete and correct. The review provides another check, and also
helps you to transfer your insights to problems that you will solve in
the future.

1. Understand the problem

Read the problem carefully.

J What is given?
J What are we looking for?
J What are the premises?

2. Investigate the problem

Suppose you are feeling ‘I have no idea what to do!’ What should
you do?
Then you take out your toolbox. It contains general strategies,
special strategies, techniques, and concepts.
Keep going, be stubborn but also flexible: if one strategy does not
work then try another. Work step by step.

© Springer International Publishing AG, part of Springer Nature 2018 281

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7
282 A A survey of problem-solving strategies

General strategies

J Get a feel for the problem, get well acquainted with it:
Consider special cases or examples, make sketches and tables

J Have I seen a similar problem before?

J Is there a simpler problem that I could consider first?

J How can I use what is given? – Working forward

J How can I reach the goal? – Working backward

J Can I formulate useful interim goals?

J Introduce notation

J Formulate conjectures

Here are some other questions that you may ask yourself:

J Is there a pattern?

J What is essential, what matters (and what doesn’t)?

Special strategies

Special strategies are useful for particular classes of problems.

J Recursion, induction

J Counting by bijection, counting in two ways

J Pigeonhole principle

J Extremal principle

J Invariance principle

If you think that one of these principles could be useful, plan how
to use it: what would I need to do for the inductive step, what
could be the pigeons and the holes, which quantity could we max-
imise/minimise to solve our existence problem, what could be an
invariant that helps to prove non-existence?
A A survey of problem-solving strategies 283

Techniques, concepts

There are many techniques and concepts that can be useful for solving
problems. Some that you learned about in this book are:

J Solving recursions using the ansatz that an = αn and the superpo-

sition principle

J Graphs: using them to represent complex webs of relationships;

even/odd arguments, Euler’s formula

J Counting principles

J Congruences

J Signature of a permutation

There are many more techniques and concepts that you will learn
about when studying more mathematics.
Of course with experience you will get better at solving problems.
You acquire it by solving problems yourself and also by understand-
ing given solutions. After a while you will build up a stock of
problems and solutions, which you can use with the strategy “simi-
lar problems”. Or you may follow a middle course: first think about
the problem, then read a little, then try to make progress yourself,
then get another idea from the book, etc. This can be an efficient way
to build up your stock and gain experience.
Many solutions to problems have solidified into mathematical theo-
ries (for instance number theory, or graph theory) which have been
refined more and more over the centuries. When you study such a
theory you will learn more techniques, concepts and ideas. Always
ask yourself: Which problems does the theory help me to solve?
Which problems can I solve now that I could not solve before?
And if none of the tools help – build your own. This is the high art
of problem-solving and of mathematical research.

3. Write up the solution properly

Be very careful to:

J Argue conclusively, use correct logic

284 A A survey of problem-solving strategies

J Structure sensibly

J Write understandably (mention ideas and motivations, make a

sketch etc.)

Mathematical terminology often helps us to say things precisely. But

it should be accompanied by explanations: a sequence of equations
alone is usually hard to understand. Words are important!

4. Review your solution

When you have solved the problem you should ask yourself more
questions, for example:

J What have I learned?

J Is the solution reasonable?

J Is there a different or even a better way to do it?

J Did I use all the premises? How did they matter?

J Can I find interesting problems by modifying the question, the

premises?
EUREKA!
B Basics on sets and maps

Modern mathematics is formulated in the language of sets and maps

(also called mappings). This language is precise and gives a common
basis to the enormous breadth of the subject. Therefore we also
use it at many places in this book. Here you will find the essential
definitions and facts.

Sets and elements

A set is a collection of objects, its elements, where it only matters

whether an element belongs to the set or not. That is, the same
element cannot occur several times in a set, and order does not matter.
The elements are listed between braces { and }.
For example, {1, 2} and {5, 8, 9} are sets. The sets {1, 2}, {2, 1} and
{1, 1, 2} are all the same. (Switching order and multiple entries are
allowed but have no effect.)
If an element belongs to a set then we write ∈, otherwise ∈. For
example 1 ∈ {1, 2} but 3 ∈ {1, 2}. (In words: 1 is an element of {1, 2},
3 is not an element of {1, 2}.)
A subset of a set A is a set B all of whose elements are also in A.
Then we write B ⊂ A. For example, {1, 2} ⊂ {1, 2, 5}.
Given a set, you can select from it elements having special proper-
ties. This is indicated by a colon (sometimes a vertical line | is used).

Example:
Let A = {1, 2, 3, 4, 5, 6}. Then {n ∈ A : n is even} = {2, 4, 6}.
(In words: the set of elements n of A which are even)
Number systems are sets:

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

Z = {. . . , −2, −1, 0, 1, 2, . . . } the set of integers
p
Q = { q : p ∈ Z, q ∈ N } the set of rational numbers
R the set of real numbers

© Springer International Publishing AG, part of Springer Nature 2018 287

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7
288 B Basics on sets and maps

In addition we write N0 = {0, 1, 2, . . . }. In this book we do not need

the precise definition of R.
It is useful to have a symbol for the set that has no elements, called
the empty set. This set is denoted ∅. If A is any set then ∅ ⊂ A.
The elements of a set need not be numbers, they can be any objects,
for example graphs. Since sets are objects, they can also be elements
of another set. For example, the set {{1, 2}, {1, 3, 6}} has the two
elements {1, 2} and {1, 3, 6}, and the set {∅} has the element ∅. If A
is a set then you can consider all subsets of A as elements of a new
set, called the power set of A and denoted P ( A). For example

P ({1, 2}) = {∅, {1}, {2}, {1, 2}}.

Note that {1} is an element of P ({1, 2}). However, 1 is not an

element of P ({1, 2}), but it is an element of {1}, and of {1, 2}. This
may appear pedantic at first, but it is consistent, and in the long run
this consistency is very useful! So the power set P ({1, 2}) has four
elements: ∅, {1}, {2} and {1, 2}.
If A is a set then | A| denotes the number of elements of A. It
can be infinite. For example |{3, 5}| = 2, |∅| = 0, |N | = ∞ and
|P ({1, 2})| = 4.

Operations with sets

From two sets you can create new sets using various operations: For
sets A and B we define:
Name Symbol Definition
union of A, B A∪B { x : x ∈ A or x ∈ B}
intersection of A, B A∩B { x : x ∈ A and x ∈ B}
difference of A, B A\B { x : x ∈ A, but x ∈ B}
product of A, B A×B {( x, y) : x ∈ A and y ∈ B}
(In words: A union B, A intersected with B, A minus B, A times B.)
For example, {1, 2} ∪ {2, 4, 6} = {1, 2, 4, 6}, {1, 2} ∩ {2, 4, 6} = {2}
and {1, 2} \ {2, 4, 6} = {1}. For the product (also called product set
or cartesian product) see the next section.
Two sets are called disjoint if they have no common elements, i.e.
if A ∩ B = ∅. If A, B are disjoint then A ∪ B is sometimes called their
disjoint union.
B Basics on sets and maps 289

Several sets A1 , A2 , . . . , An are disjoint if A1 ∩ A2 ∩ · · · ∩ An = ∅,

and are called pairwise disjoint if every two of them are disjoint:
Ai ∩ A j = ∅ for all i = j. For example, A1 = {1, 2}, A2 = {2, 3} and
A3 = {1, 3} are disjoint (there is no element in all of them) but not
pairwise disjoint.

Tuples

Tuples are another way of grouping several objects (numbers, graphs,

sets etc.) into new objects. After sets, they are the next simplest and
most useful way to do this.
From two objects a, b you can form the pair ( a, b). Here order
matters, that is, ( a, b) and (b, a) are different if a = b. If the objects
a, b are equal then we still consider this as a pair: ( a, a). If A and B
are sets then A × B denotes the set of all pairs ( a, b) with a ∈ A and
b ∈ B. For A = B we often write A2 instead of A × A.
For example, {1, 2} × {1, 2, 3} = {(1, 1), (1, 2), (1, 3), (2, 1), (2, 2), (2, 3)};
you can arrange the elements of A × B in a rectangular array:

1 2 3
1 (1, 1) (1, 2) (1, 3)
2 (2, 1) (2, 2) (2, 3)

The rows correspond to the elements of A, the columns to the ele-

ments of B. (One could also do it the other way around, but this
representation is most common, for example in the context of matri-
ces.)
From three objects a, b, c you can form the triple ( a, b, c). Again
order matters, and multiple entries are allowed. For example, ( a, a, b)
is a triple that must not be confused with the pair ( a, b). If A, B, C are
sets then A × B × C denotes the set of all triples ( a, b, c) with a ∈ A,
b ∈ B, c ∈ C. You could arrange these in a cuboid (box). If A = B = C
then we often write A3 instead of A × A × A.
From four objects we can form quadruples, from five quintuples,
from k objects k-tuples ( a1 , . . . , ak ), where k ∈ N. So a 2-tuple is
a pair, a 3-tuple a triple etc. The objects a1 , . . . , ak are called the
components of the tuple.
The notion of a tuple allows us to talk about higher-dimensional
space: Real numbers can be thought of as points on the number
290 B Basics on sets and maps

line. Pairs of real numbers correspond to points in the plane (the

two numbers indicating the coordinates with respect to a chosen
coordinate system). Triples of real numbers correspond to points in
space. Therefore, for any n ∈ N we define by analogy n-dimensional
space as the set of n-tuples of real numbers. This is an important idea,
but we do not use it in this book.

Maps

A map or mapping is a rule that assigns to each element of a set A

an element of a set B. We write f : A → B, and f ( a) for the element
of B assigned to a ∈ A. Here f is a name for the map.
For example the rule f (n) = 2n defines a map f : N → N. Each
natural number is mapped to its double.
Instead of f : N → N, f (n) = 2n we may write f : N → N, n →
2n (notice the two types of arrows). You may change the name of the
variable n: the same map is defined by f : N → N, m → 2m or by
f : N → N, γ → 2γ.1 If you don’t need the name f then you may
write the map simply as N → N, n → 2n.
As another example consider f : R2 → R2 , ( x, y) → ( x, −y). The
map f describes a reflection in the x axis. That is, f ( x, y) is the point
in the plane that you obtain by reflecting the point ( x, y) across the x
axis.
A map taking values in numbers (i.e. B = R or another number
system) is called a function. A map f : A → B is called
injective if for each b ∈ B there is at most one a ∈ A with f ( a) = b
surjective if for each b ∈ B there is at least one a ∈ A with f ( a) = b
bijective if for each b ∈ B there is exactly one a ∈ A with f ( a) = b

Put differently, f is injective if it maps any two different elements

of A to different elements of B, it is surjective if each element of B
occurs as a value f ( a), and it is bijective if it is both injective and
surjective.
For example the map f : N → N, f (n) = 2n is injective but not
surjective since for each b ∈ N there is at most one n with 2n = b,
but for odd b there is no such n.
1 However, there are some useful conventions: natural numbers are usually de-
noted n, m, k, l, i, j; real numbers x, y, z. But there is no strict rule about this.
B Basics on sets and maps 291

If E denotes the set of even natural numbers then the map f : N →

E, f (n) = 2n is bijective, because for each even number b ∈ E there is
exactly one n ∈ N with 2n = b. This shows that the properties of a
map f : A → B depend on the sets A and B, called the domain and
codomain of f respectively.
A bijective map is also called a bijection. It sets up a one-to-one
correspondence between the elements of A and B, i.e. each element
of A corresponds to exactly one element of B and each element of B
corresponds to exactly one element of A.
A map g : B → A is called the inverse map of a map f : A → B if
“the rules for f and g cancel each other out”. More precisely, for all
a ∈ A we have g( f ( a)) = a (first applying f to a and then applying g
to the result will give back a), and for all b ∈ B we have f ( g(b)) = b.
Since bijective maps are used repeatedly in this book we note:

Theorem Let f : A → B be a map. Then the following state-

ments are equivalent (i.e. each implies the other):
a) f is bijective
b) f is surjective and injective.
c) There exists an inverse map for f .

Proof.
The equivalence of a) and b) follows directly from the definition.
Suppose c) holds, i.e. f has an inverse map g : B → A. Then f is
surjective since to any b ∈ B we can take a = g(b), this satisfies the
equation f ( a) = b since f ( g(b)) = b.
f is also injective: if f ( a) = b then applying g to both sides yields
g( f ( a)) = g(b). By assumption g( f ( a)) = a, so we get a = g(b). We
have shown that there can be only one a ∈ A satisfying f ( a) = b.
Thus we have shown that c) implies b) and hence a).
It remains to prove that a) implies c). Assume f is bijective. Then we
can write down g: Let b ∈ B. Since f is bijective there is exactly one
a ∈ A with f ( a) = b. We define g(b) = a. Then g is an inverse map
of f . q. e. d.
292 B Basics on sets and maps

Examples

Let E be the set of even natural numbers. The inverse map of

f : N → E, n → 2n is g : E → N, m → m2 .2
Let f : R → R, x → x3 . This map is bijective with inverse map
√
g : y → 3 y.
Let f : R → R, x → x5 + x. Is f bijective? This would mean that
for each y ∈ R there is exactly one x ∈ R satisfying x5 + x = y.
So we are talking about solving an equation. Using tools from
analysis one can show that indeed for each y there is exactly
one solution x (compare the discussion on proofs of existence in
Section 7.2).3 So we know that f is bijective. The inverse map
is g : R → R, g(y) = the unique solution x of the equation
x5 + x = y. This defines g, even though we did not write down
an expression for g(y). In fact, it can be proved that there is no
such explicit expression (for example using roots and arithmetic
operations).
The map f : {1, 2, 3} → {1, 2, 3}, 1 → 2, 2 → 3, 3 → 1 is bijective
with inverse map g : {1, 2, 3} → {1, 2, 3}, 2 → 1, 3 → 2, 1 → 3.

The examples show the most important ways of defining a map

f : A → B: by a formula, by a prescription (g in the third example)
or by listing the function values for each argument (if A is finite, as
in the last example). They also show that finding the inverse map
amounts to solving an equation: in order to find the inverse map
of f : R → R, x → x3 we need to solve the equation y = x3 for x.
√ √
Taking the cube root yields x = 3 y, so g : y → 3 y is the inverse
map.

2 Itis useful and helps to avoid confusion and mistakes to use a different variable
name for the inverse map, for example m instead of n.
3 This follows from the intermediate value theorem and the following properties

of f : the map f is continuous and strictly monotone, and f ( x ) → ∞ for x → ∞,

f ( x ) → −∞ for x → −∞.
List of symbols

∀, ∃ ‘For all’, ‘There is’ 160

¬ negation: if A is a proposition then ¬ A is the proposition 158
that A is false
⇒, ⇔ implication and equivalence 161
:= is defined as 66
⊂ is a subset of 285
∈ is an element of 37
∪, ∩, \, × the set operations union, intersection, set difference, prod- 108, 286
uct set
∅ the empty set 286
|X| number of elements (cardinality) of the set X 107
( a, b), ( a, b, c) a pair or triple 113, 287
N, Z, Q, R, N0 natural numbers, integers, rational numbers, real numbers, 285
natural numbers including zero
→, → mapping arrows: f : A → B denotes a map from a set A to 288
a set B. Instead of f ( a) = b one may write f : a → b
∑ short notation for sums. Instead of f (1) + f (2) + · · · + f (n) 66
n
one writes ∑ f (k), or ∑nk=1 f (k) to save printing space,
k =1
where f is any expression (i.e. a function of k). For example
n n
∑ k = 1 + 2 + · · · + n and ∑ k2 = 12 + 22 + · · · + n2
k =1 k =1
x Gauß bracket of the real number x: the value of  x  is the 205
largest integer which is less than or equal to x
frac( x ) The part after the decimal point of a positive real number 205
x; that is, frac( x ) = x −  x 
n used throughout as notation for an arbitrary natural num- 25
ber
n! 
 factorial: n! = the product
  of the numbers 1, 2, . . . , n 25, 115
n n n ( n − 1) . . . ( n − k + 1)
binomial coefficient: = 116
k k k!
| divides: n| a means that n is a divisor of a 181
≡, mod congruent, modulo: a ≡ b mod n means that b − a is 188
divisible by n
dV degree of the vertex V in a graph (number of edges con- 93
taining V, with loops counted twice)
bF number of border edges of a face F of a plane graph 91
( AB) length of the straight line from A to B 232

© Springer International Publishing AG, part of Springer Nature 2018 293

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7
Glossary
addition rule
Basic counting rule. 108
axiom
A proposition which is taken to be true in a given context. 168

bijection, bijective map

A map f : A → B is bijective if for every b ∈ B there is exactly one a ∈ A
satisfying f ( a) = b. 117, 288
binomial coefficient  
n n ( n − 1) . . . ( n − k + 1)
The binomial coefficient = is the number of
k k!
k-element subsets of an n-element set. 116
binomial theorem
The formula ( a + b)n = an + (n1 ) an−1 b + (n2 ) an−2 b2 + · · · + (n−
n
1) ab
n −1 + b n

for real numbers a, b and n ∈ N0 . It generalises the well-known formula

( a + b)2 = a2 + 2ab + b2 .. 136

cardinality
The cardinality of a finite set X is the number of elements of X. Notation: | X |.
Two (possibly infinite) sets are said to have the same cardinality if there is a
bijection between them. 107, 118, 133
C ATALAN numbers
Sequence of numbers which occurs when counting triangulations, for exam-
ple. 67
colouring problems
An interesting class of problems for graphs. 101
conclusion
The proposition B in an implication A ⇒ B. 161
congruent
For integers: For n ∈ N and a, b ∈ Z we say that a, b are congruent modulo
n if a and b leave the same remainder when divided by n, or equivalently if
b − a is divisible by n. In geometry: Two figures (i.e. subsets of the plane or
of space) are congruent if one can be transformed into the other by a rigid
motion, i.e. by a translation, rotation or reflection. 188
connected
A set is connected if you can get from any point (i.e. element) of the set to
any other point of the set along a path which runs inside the set. Here the set
can be a graph or a subset of the plane or of space. 84

© Springer International Publishing AG, part of Springer Nature 2018 295

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7
296 Glossary

counting in two ways

Method of deriving identities by counting the elements of a set in two different
ways. 123, 169

degree of a vertex
In a graph the degree of a vertex V is the number of edges containing V
(loops are counted twice). Notation: dV . 93, 198
dense
For subsets A ⊂ B ⊂ R we say that A is dense in B if for any two numbers
x, y ∈ B with x < y there is an element r ∈ A satisfying x ≤ r ≤ y. 211, 246
disjoint
Two sets A, B are called disjoint if they have no common elements, i.e. if
A ∩ B = ∅; several sets are called pairwise disjoint if any two of them are
disjoint. 286
division with remainder
For a ∈ Z, n ∈ N the representation of a as a = qn + r with q, r ∈ Z and
0 ≤ r < n. The remainder is r. 184
divisor
If a, n ∈ Z then n is a divisor of a if a is divisible by n, that is if there is q ∈ Z
satisfying a = qn. 181
domino effect
Intuition for mathematical induction. 71

edge-face formula
The formula 2e = ∑ bF for plane graphs G, where e is the number of
F face of G
edges and bF is the number of boundary edges of the face F. 91
edge-vertex formula
The formula 2e = ∑ dV for graphs, where e is the number of edges
V vertex of G
and dV is the degree of the vertex V. 92
(logically) equivalent
Two propositions A, B are (logically) equivalent, in symbols A ⇐⇒ B, if
both ‘ A implies B’ and ‘B implies A’ are true. For example, the propositions
n + 3 = 8 and n = 5 about natural numbers n are equivalent. 163, 164
E ULER’s formula
The formula v − e + f = 2 for planar graphs or for polyhedra, where v is the
number of vertices, e is the number of edges and f is the number of
faces. 84, 98
extremal principle
Important idea for proofs of existence. Fundamental principle of science that
extremal configurations have special significance. 173, 217
Glossary 297

face of a plane graph

One of the regions into which a plane graph subdivides the plane. 84
factorial
n factorial = n! = the product of the numbers 1, 2, . . . , n, e.g. 4! = 1 · 2 · 3 · 4 =
24. 115
F ERMAT numbers
The numbers Fn = 2(2 ) + 1 where n ∈ N0 .
n
166, 168, 173
F IBONACCI numbers
The sequence of numbers 1, 1, 2, 3, 5, 8, . . . defined by a0 = a1 = 1 and an =
an−1 + an−2 for n ≥ 2. 54

G AUSS trick
Method for calculating 1 + · · · + n quickly. 33, 124
graph
A finite structure given by a set of objects, called vertices, and a set of edges,
each joining two vertices. Examples: 1. a plane graph; 2. vertices are the
people in a room, and two people are joined by an edge if they know each
other. 93
graph, plane
A graph which is drawn in the plane in such a way that vertices are points,
edges are lines (not necessarily straight) joining the vertices, and the lines
don’t intersect (except at vertices). 83

implication
Proposition of the form ‘ A implies B’ (A ⇒ B). 161
in general position
A set of lines is in general position if no three go through a point and no two
are parallel. 28
infinite descent
Method which is sometimes useful for proving that an equation has no
solution in natural numbers: you prove that if there was a solution then there
would have to be another solution which is smaller. 235
injective, one-to-one, 1-1
A map f : A → B is injective (or one-to-one) if for every b ∈ B there is at most
one a ∈ A satisfying f ( a) = b. 288
invariance principle
Important idea for proofs of nonexistence. Fundamental principle of science
that in complex processes the quantities that don’t change have special signif-
icance. 176, 251

invariant
Something that does not change during a certain process. 256
298 Glossary

inverse map
g : B → A is called the inverse map of the map f : A → B if g( f ( a)) = a and
f ( g(b)) = b hold for all a ∈ A, b ∈ B. A map f has an inverse map if and
only f is bijective. 289
irrational number
A number which cannot be written as the quotient of two integers. 175, 203

lattice point
A point of the plane both of whose coordinates are integers. 215
lemma
An auxiliary statement which is used in the proof of a theorem. 182
loop in a graph
An edge of a graph whose two endpoints coincide. 93

map, mapping
A prescription which assigns to each element of a set A an element of a set B.
We write f : A → B where f is a name for the map. 288
mean: harmonic, geometric, arithmetic or quadratic
√
For a, b > 0 we call 1 2 1 the harmonic mean, ab the geometric mean, a+ b
a + b 
2

the arithmetic mean, and a +

2 b2
2 the quadratic mean of a and b. 221, 234, 246

modulo, mod
The remainder of a modulo n is the remainder when dividing a by n. 184
multiple
If a, n ∈ Z then a is a multiple n if a is divisible by n, that is if there is q ∈ Z
satisfying a = qn. 181
multiple counting principle
A useful idea for many counting problems. 110
multiplication rule
Basic counting rule. 108

necessary
Proposition A is necessary for proposition B if B implies A. 163

ordering
An order in which you can line up some objects, for example the numbers
1, . . . , n. 115, 262

parity
Property of an integer being even or odd. 150, 254
Glossary 299

PASCAL’s triangle
Arrangement of the binomial coefficients in a triangular array. 79, 120, 127
path in a graph
A sequence of edges in a graph where each edge has a common vertex with
the previous edge, so that the edges can be traversed one after another. The
path is called closed if the first and last vertex are the same. 84
permutation
A way to order the elements of a set M, e.g. M = {1, . . . , n}; may also be
interpreted as bijection M → M. 115, 262
pigeonhole principle
Fundamental idea for proofs of existence: If n + 1 pigeons sit in n pigeonholes
then at least one hole must be occupied by more than one pigeon. 173, 195
polygon
A plane figure bounded by finitely many straight lines, for example a triangle
or hexagon. A polygon is called convex if the straight line connecting any
two of its vertices lies in the polygon. 61
power set P ( A)
Set of subsets of the set A. 44, 118, 286
predicate
A statement containing variables, which becomes a proposition when values
are plugged in for the variables. 158
premise
The proposition A in an implication A ⇒ B. 161
prime factorisation
The representation of a natural number n > 1 as product of prime num-
bers. 183
prime number
A natural number which has precisely two divisors. 182
proof by contradiction
In order to show A ⇒ B you prove that ‘ A and ¬ B’ implies a false statement
(a contradiction). 163, 167
proof by mathematical induction
A type of proof which is often useful for proving statements about all natural
numbers. 71, 169
proof of existence
A proof which shows the existence of an object with certain proper-
ties. 169, 256
proof of impossiblity
Proof that something is impossible. 98, 174, 252, 272
300 Glossary

proof of nonexistence
A proof which shows that an object with certain properties cannot exist. 174, 257

proof without words

Sketch or diagram which suggests a proof of an arithmetic or geometric
proposition. 128
proof, direct
In order to show A ⇒ B you prove a sequence of implications A ⇒ A1 ⇒
A2 ⇒ · · · ⇒ B. 167
proof, indirect
In order to show A ⇒ B you prove ¬ B ⇒ ¬ A. 163, 167
proposition
A statement which can be true or false. 157

rational number
A number that can be written as the quotient of two integers. 201, 285
recurrence relation
An equation giving each term in a sequence as a function of previous terms.
It expresses the idea of recursion in a counting problem. 41
recursion
A problem solving technique: reduce the problem to a smaller problem of the
same kind. 141
remainder
The remainder of the division of a ∈ Z by n ∈ N is the number r in the
representation a = qn + r where q, r ∈ Z und 0 ≤ r < n. 184, 199

semi-invariant
Something that changes in a controlled way during a certain process. 264
shift by one
Typical phenomenon in counting problems, frequent source of errors. For
example, a train with 10 cars has 9 couplings, and the number of integers
from 10 to 100 is 91, not 100 − 10 = 90, since one needs to count both the first
and last. 23, 109, 149
signature of a permutation
The parity of the number of inversions of a permutation. 262
sufficient
Proposition A is sufficient for proposition B if A implies B. 163
superposition principle
In linear problems (equations) one may obtain new solutions by adding two
solutions or multiplying a solution by a constant. 57
Glossary 301

surjective, onto
A map f : A → B is surjective (or onto) if for every b ∈ B there is at least one
a ∈ A satisfying f ( a) = b. 288

to tile
To cover completely without overlaps, and without pieces (tiles) extending
over the boundary. 50, 253
topology
Area of mathematics which studies those properties of shapes (or other
mathematical structures, like maps) which remain the same when the shapes
(or maps etc.) are changed in a continuous way. 99
torus
A surface which looks essentially like a bike tyre inner tube (without valve). 99
trapezoidal number
Natural number that can be represented as the sum of several consecutive
natural numbers. 146
triangle inequality
In a triangle each side is shorter than the sum of the lengths of the two other
sides. 221
triangulation
subdivision of a polygon into triangles by non-intersecting diagonals. 60
truth value
true (t) or false (f). 158
Lists of problems, theorems and methods

List of problems
Cutting up a log . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Trailing zeroes of 100! . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
n lines in the plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Number of subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Number of tilings by dominoes . . . . . . . . . . . . . . . . . . . . . . 50
Solving the Fibonacci recurrence relation . . . . . . . . . . . . . . . 54
Number of triangulations of a convex n-gon . . . . . . . . . . . . . . 61
Colouring regions in the plane . . . . . . . . . . . . . . . . . . . . . . 74
Euler’s formula for graphs . . . . . . . . . . . . . . . . . . . . . . . . 84
Counting in two ways for graphs: edge-face formula . . . . . . . . . 91
Counting in two ways for graphs: edge-vertex formula . . . . . . . . 92
Handshakes: even and odd . . . . . . . . . . . . . . . . . . . . . . . . 94
Three connecting lines without intersections . . . . . . . . . . . . . . 95
Five points in the plane, all joined by edges . . . . . . . . . . . . . . . 96
Colourings of plane graphs . . . . . . . . . . . . . . . . . . . . . . . . 101
Simple counting problems . . . . . . . . . . . . . . . . . . . . . . . . . 109
Counting triominoes, I . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Counting triominoes, II . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Number of handshakes . . . . . . . . . . . . . . . . . . . . . . . . . . 116
Number of ways to form pairs of dancers . . . . . . . . . . . . . . . . 117
Symmetry of the binomial coefficients . . . . . . . . . . . . . . . . . . 120
Number of subsets with even/odd number of elements . . . . . . . 121
Counting in two ways all connecting lines of n points . . . . . . . . . 124
Counting in two ways and sum of squares . . . . . . . . . . . . . . . 125
Counting the number of subsets in two ways . . . . . . . . . . . . . . 127
Counting in two ways for multiplication . . . . . . . . . . . . . . . . 127
Working backward . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
Diagonal in a cuboid . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
Trapezoidal numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
Practical use of remainders . . . . . . . . . . . . . . . . . . . . . . . . 186
Same number of acquaintances . . . . . . . . . . . . . . . . . . . . . . 197
Two equal remainders . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Powers ending in 001 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Approximation by fractions . . . . . . . . . . . . . . . . . . . . . . . . 201
How close can multiples be to an integer? . . . . . . . . . . . . . . . 204
(Non-)Acquaintances among 6 people . . . . . . . . . . . . . . . . . . 212
Rectangle of largest area with given perimeter . . . . . . . . . . . . . 220
Smallest sum of squares of distances . . . . . . . . . . . . . . . . . . . 222

© Springer International Publishing AG, part of Springer Nature 2018 303

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7
304 Lists of problems, theorems and methods

Shortest path touching a given line . . . . . . . . . . . . . . . . . . . . 223

In a tournament there is a player with a complete list . . . . . . . . . 227
Farms and wells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
Numbers on a circle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
3a2 = b2 + c2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
Existence of a reflection point on a curve . . . . . . . . . . . . . . . . 246
Two joggers running at the same speed . . . . . . . . . . . . . . . . . 253
A coin on a 6 × 6 board . . . . . . . . . . . . . . . . . . . . . . . . . . 254
Tiling square boards with dominoes . . . . . . . . . . . . . . . . . . . 255
Tiling a 6 × 6 board with cut-off corners with dominoes . . . . . . . 256
Game: replace pairs of numbers by their difference . . . . . . . . . . 260
Multiple neighbour swapping . . . . . . . . . . . . . . . . . . . . . . . 263
The 8-puzzle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
Balls of three colours . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268

List of theorems and methods

Method for solving the Fibonacci recurrence relation . . . . . . . . 58
Graph formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Euler’s formula for polyhedra . . . . . . . . . . . . . . . . . . . . . . 99
Fundamental counting rules . . . . . . . . . . . . . . . . . . . . . . . . 108
The main counting problems . . . . . . . . . . . . . . . . . . . . . . . 114
even + even = even . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
Difference of two squares . . . . . . . . . . . . . . . . . . . . . . . . . 170
Same distances to three points . . . . . . . . . . . . . . . . . . . . . . 171
Infinitely many primes . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
The square root of 2 is irrational . . . . . . . . . . . . . . . . . . . . . 175
Euclid’s lemma on prime numbers . . . . . . . . . . . . . . . . . . . 182
Prime factorisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Division with remainder . . . . . . . . . . . . . . . . . . . . . . . . . . 184
Calculation rules for congruences . . . . . . . . . . . . . . . . . . . . 189
Approximation by fractions . . . . . . . . . . . . . . . . . . . . . . . . 209
The fractional parts of multiples of an irrational number are dense . 212
Inequality of the geometric, arithmetic and quadratic mean . . . . . 223
Triangle inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
Shortest path touching a line satisfies the law of reflection . . . . . . 225
Shortest path via a point on a given curve satisfies the law of reflection245
Properties of the signature . . . . . . . . . . . . . . . . . . . . . . . . . 266
Bijective maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
Hints for selected exercises

Exercises in Chapter 1

E 1.5 First consider the powers of two: 4, 8, 16 etc. The numbers in

between can be dealt with similarly to the next larger power of two.
To argue that your answer is best possible consider this question: how
much can the number of pieces grow with each cut? The smallest
number of cuts is log2 n where log2 is the logarithm to base 2 and
 x  is x rounded up to the next integer.

E 1.6 Here the Gauss bracket is useful, the symbol for rounding
down to the previous integer. By definition  x  = the largest integer
which is ≤ x. For example  72  = 3 is the number of even natural
numbers which are less than or equal to 7.

E 1.7 To recognize the patterns it is useful to know the values of n!

for n = 1, . . . , 5 by heart. To prove the formulas which you might
conjecture from this the telescope trick is useful: for example in a)
write 1 · 1! = (2 − 1)1! = 2! − 1!, 2 · 2! = (3 − 1)2! = 3! − 2!, and in
general n · n! = · · · = (n + 1)! − n!. Then

1 · 1! + 2 · 2! + · · · + n · n! = (2! − 1!) + (3! − 2!) + · · · + ((n + 1)! − n!)

= (n + 1)! − 1! = (n + 1)! − 1 ,

since 2! cancels with −2!, similarly 3! with −3! (in the next term), up
to n! with −n!.
The telescope trick also helps with b) and c).
The sum in c) is n + 1 − 1
n +1 .

E 1.8 a) an = n + 2 b) an = 32 · (n + 1)! c) (−1)n−1

d) an = an−1 + an−2 + an−3 where a1 = a2 = a3 = 1
e) an = n2n

E 1.9 For example, you could try to prove the recursion sn = 2sn−1 .

© Springer International Publishing AG, part of Springer Nature 2018 305

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7
306 Hints for selected exercises

E 1.10 Consider n = 1, 2, . . . , but also n = 40. Can you do this

without a calculator?

E 1.11 Consider the number of lines in each group of parallels. For

example, if the lines 1 , 2 are parallel and the lines 3 , 4 , 5 are
parallel to each other but not to 1 , 2 then these numbers are 2 and 3.
If no two lines are parallel then these numbers are all equal to 1.

Exercises in Chapter 2

E 2.1 Use |γ| < 1.

E 2.4 For an = 2an−1 − an−2 a second solution is an = n. In general:

nαn .

E 2.10 Distinguish two kinds of subsets: those that contain the ele-
ment n and those that don’t.

E 2.12 When searching for a recursion you may find it difficult to

reduce the problem to a smaller problem of the same kind. One way
out is to consider a second problem in parallel: Let bn be the number
of tilings of a 3 × n rectangle from which the squares at positions
(n, 2) and (n, 3) are deleted. Find formulas that express an and bn in
terms of an−1 , bn−1 , an−2 , bn−2 . Eliminate bn .

E 2.13 See the hint for Exercise E 2.12.

E 2.14 First check for n = 1, 2, 3, 4, 5 whether the first player can force
victory. How can you reduce the answer for a game with n matches
to the answer for a smaller number of matches?

E 2.15 The table suggests the Fibonacci recursion on = on−1 + on−2 .

Directly from the problem you can deduce the recursion on = on−1 +
on−3 + on−5 + . . . (where we set oi = 0 if i ≤ 0). How? How can you
derive the Fibonacci recursion from this?

Exercises in Chapter 3

E 3.1 Prove by induction that for each n ∈ N you can put n pins into
the suitcase :-).
Hints for selected exercises 307

E 3.4 What could be the inductive step? First try to understand the
step from n = 2 to n = 3.

E 3.7 Compare Exercise E 2.10.

E 3.9 The main task is to show that this rule makes sense: there could
be several paths from A to B, and you need to show that each path
yields the same resulting colour for B. Here is an idea how to prove
this: if you have two such paths then you can deform or ‘slide’ one
path so it becomes the other path. What happens to the number of
intersection points when you slide across a vertex in the process?

Exercises in Chapter 4

E 4.6 What can you say about the numbers bF in a hypothetical

drawing without intersections?

E 4.7 Use the graph formulas.

E 4.8 Cut the torus open so you can flatten it into the plane and
consider the resulting plane graph.

E 4.9 Your argument could begin as follows: the points D, E must

either both lie in the interior region or both lie in the exterior region.
Suppose they lie in the interior region (the other case is similar). Now
consider the connections from D to A, B, C and think about where E
could lie.

E 4.10 Have you seen a similar problem before?

E 4.11 Use the graph formulas.

E 4.12 Try induction on f . Let F be a face with at most 5 borders.

How can you remove this face and produce a smaller graph, so that
any proper 6-colouring of the smaller graph can be modified into a
proper 6-colouring of the original graph?

E 4.13 Use the representation of the torus as a square with opposite

sides glued together as in Figure 4.12.

E 4.14 Proceed similarly to the solution of Exercise E 4.12.

308 Hints for selected exercises

E 4.15 Pay attention to the vertex degrees. Why can’t you start at the
top of the roof?

E 4.16 Use the graph formulas to derive restrictions on the possible

numbers of vertices per face and edges per vertex.

E 4.17 Choose a point in the interior of each small triangle. Construct

a graph as follows: the vertices are these chosen points. Connect two
vertices by an edge if the corresponding small triangles have an edge
in common and the labels at the endpoints of this edge are different.

Exercises in Chapter 5

E 5.4 Number of hands: (52 5 ).

One pair: choose the rank of the pair, then the two suits of the pair,
then the ranks of the three other cards, then their suits. This results
in 13 · (42) · (12
3 ) · 4 hands.
3

Therefore, about 42% of all hands have one pair, so the probability of
getting one pair is about 0.42.

E 5.6 First count: find the number of quadruples for each fixed d.
Second count: find the number of quadruples in each of these three
cases: a, b, c are all distinct; two of the three numbers a, b, c are equal;
and a = b = c.

E 5.11 Analogous to Exercise E 5.10.

E 5.13 For n odd we have en = on , for n even we have en = on + 1.

Bijection proof for n odd: if a < n then map ( a, b) to ( a + 1, b),
otherwise to (0, b).
A possible generalisation: Let k ∈ N and n1 , . . . , nk ∈ N0 . Consider
k-tuples ( a1 , . . . , ak ) with ai ∈ {0, 1, . . . , ni } for each i. Let e be the
number of such k-tuples with a1 + · · · + ak even and o the number
with a1 + · · · + ak odd. Then e = o + 1 if all ni are even, and e = o
otherwise.

E 5.14 Distinguish two cases: Either the subset contains the element
n + 1 or it does not.

E 5.15 When multiplying out ( a + b)n = ( a + b)( a + b) · · · ( a + b), in

which ways can the term an−k bk arise?
Hints for selected exercises 309

E 5.16 b) (n+ n +1 2 n +1
1 ) + 5( 3 ) + 5 ( 5 ) + . . . is divisible by 2 . Question
1 n

for further investigation: Can you find a direct proof for the formula
2n a n = ( n + n +1 2 n +1
1 ) + 5( 3 ) + 5 ( 5 ) + . . . ? For example, by counting a
1

certain set in two ways?1

√ n
√ (1 − 2) to a . Use the idea of Exercise E 5.16 a). How
E 5.17 Add n

big is 1 − 2 approximately?

E 5.18 You could use the identity of Exercise E 5.14.

E 5.19 Imagine n balls lying in a row. Decompositions of n corre-

sponds to decompositions of the set of balls into connected groups.
In what other way could you fix such a decomposition into groups?
If needed, draw all decompositions for n = 3.

E 5.20 Find a bijection from the set {( a, b, c) ∈ N3 : a + b + c = n} to

the set {( x, y) ∈ N2 : x < y < n}.

E 5.23 The identity is (n3 ) = (n− n −2

2 ) + ( 2 ) + · · · + (2) for n ≥ 3.
1 2

E 5.24 For n = 2, 3, 4, 5 the numbers are 2, 4, 8, 16, so you might

conjecture the formula 2n−1 . However, for n = 6 you only get 31
regions. To find a general formula it may be helpful to look at
Exercise E 5.7.

Exercises in Chapter 6

E 6.3 Problem and solution formula are symmetric; the derivation is

not as is clear from the solution scheme in Figure 6.9. This asymmetry
has the effect that there are different paths to the result: instead of
starting with the bottom rectangle (the a, b rectangle) and its diago-
nal we could have argued starting with one of the other boundary
rectangles (a, c rectangle or b, c rectangle).

E 6.4 One idea: delete a suitable part of 14 = (−1) + 0 + 1 + 2 + 3 +

4 + 5.
Another idea: for example n = 14 = 2 · 7 could be written using
7 = 3 + 4 as 14 = (3 + 4) + (3 + 4), then decrease one 3 and increase
one 4 by one to get 14 = 2 + 3 + 4 + 5.
1I would be interested to hear about it if you find one.
310 Hints for selected exercises

E 6.5 Build on the solution of Problem 6.3. It is useful to represent n

as a product of prime powers (prime factorisation, discussed in detail
in Chapter 8).

E 6.6 Use the formula ( a + b)( a − b) = a2 − b2 .

E 6.7 Work backward. Write 11111 as a product of prime numbers.

√6.8 Work backward.

E The goal can be formulated as a.a1 . . . an <
N < a.a1 . . . an + 10−n where a ∈ N.

E 6.9 Work forward. How can I use the data? For example you
could connect B, C. There is more than one solution.

E 6.10 What could be the last step before such a configuration is

reached?

E 6.11 What does it mean for a number n that the card at position n
is face down at the end?

E 6.12 Either write down examples and guess (‘see’) a formula (and
then prove it!), or else work systematically: number the customers
1, . . . , n where 1 is the last customer etc. Let xi be the number of apples
left after customer i + 1 has gone, then x0 = 1 and xi = 2xi−1 + 2 for
i = 1, . . . , n. A clever way to solve this recursion is to add 2 on both
sides and to write the result as xi + 2 = 2( xi−1 + 2).
?
E 6.13√Work backward. Write x > 0 and try to eliminate the roots
(add 2, square etc.) Be careful with signs.

Exercises in Chapter 7

E 7.3 b), d), e) and f) only.

E 7.4 All propositions imply b). There are no other implications. To

show this give a counterexample for each of the other implications.

E 7.6 b) says that the set S1 is infinite.

E 7.7 If n is not a prime number then it has a divisor which is bigger

√
than 1 and at most n. This is true (and useful if you want to check
Hints for selected exercises 311

in your head whether 113 is a prime number). Negation (which is

false):

∃n ∈ N : (∃m ∈ N : 1 < m < n and m|n) and

 
∀m ∈ N : m = 1 or m2 > n or ¬(m|n)

E 7.8 For example: for each person there is a fact which she knows
and for which there is no other person knowing it.

E 7.14 First find a necessary condition, then prove that it is sufficient

also.
√
E 7.15 If 2 were rational,
√ what could you say about the denomina-
tors of the numbers ( 2 − 1)n ? Multiply out and collect terms. What
happens for large n?

Exercises in Chapter 8

E 8.3 Use 4 ≡ −3 mod 7.

E 8.4 You need to show that there are no natural numbers p, q

satisfying 2q = 10 p .

E 8.6 If n has the decimal representation am . . . a1 a0 then n = 10m am +

· · · + 10a1 + a0 . To get a feeling for this, first consider the case m = 1,
i.e. two-digit numbers.

E 8.7 Use 10a ≡ − a mod 11.

E 8.8 Try a similar procedure as for 7 in Exercise E 8.6.

E 8.10 e) Investigate this modulo 4.

E 8.11 Use the uniqueness of prime factorisation.

E 8.12 b) For example in the case p = 3: If 3|2b then using 3|3b we

would get 3|(3b − 2b), hence 3|b. Try to generalise this. For example
for p = 5: How can you conclude from p|2b that p|b, using p|5b?
How can you then argue for p|3b? When you have understood this,
try induction on a for the general case.
c) divide a by p with remainder in order to reduce to b).
312 Hints for selected exercises

Remark on Exercise E 8.12: in most books on number theory these

facts are proved using the Euclidean algorithm. The proof suggested
here is attributed to Gauss.

E 8.13 The claim is not negated correctly. Instead of “6k ≡ 6 (mod 10)
for all k ∈ N” the correct negation is “There is k ∈ N such that 6k ≡ 6
(mod 10) ”.
For a correct proof use induction.

E 8.15 Find a suitable bijection from the set of divisors to itself.

E 8.16 Write k mn  using the remainder rk of km modulo n. What can

you say about r1 , . . . , rn−1 ?
(n+1)(m+1)
Result: 2 − 1.

Exercises in Chapter 9

E 9.4 Of course you could argue geometrically (enter the triangle,

exit the triangle). But you could also use the pigeonhole principle:
take the vertices as pigeons. What are the holes?

E 9.5 Consider the vertices of the triangle.

E 9.8 Consider bk := a1 + · · · + ak for k ∈ {1, . . . , n}.

E 9.10 First solve Exercise E 9.9 and use its result.

E 9.11 Use the result of Problem 9.4.

E 9.12 Let ( a, b) and (c, d) be lattice points. What does it mean for
a, b, c, d that the midpoint between ( a, b) and (c, d) is a lattice point?

E 9.13 Consider the remainders of a, 2a, . . . , (b − 1) a modulo b and

use Exercise E 8.11.

E 9.14 Consider the pairs (remainder of f n modulo 1000, remainder

of f n−1 modulo 1000).

E 9.15 Think about what it means that there are m, k satisfying 999999 ·
10k−6 ≤ 2m < 10k . Take the logarithm to base 10 and relate this to the
problem of the goblin falling into the gap, see Figure 9.3.
Hints for selected exercises 313

E 9.16 The prime factor 2 matters.

E 9.17 First investigate the same problem with 11 replaced by 2 or 3.

How many numbers do you need to write down so that the analogous
statement is true? Do you see a pattern?

E 9.18 Either imitate the solution of Problem 9.4 (that’s a good exer-
cise), or use the result of Problem 9.4.

Exercises in Chapter 10

E 10.2 Is there a largest value for the sum?

E 10.5 This is possible, but a little more complicated than the given
solution.
1
E 10.7 Consider multiples of n for sufficiently large n.

E 10.9 What extremal property does a boundary triangle have? The

vertices lie ‘as close together as possible’. Quantify this.

E 10.13 First find conditions modulo 2 and modulo 3 that a minimal

solution must satisfy. Then try to find a solution.

E 10.14 The numbers must all be equal. In a), b) this can be proved
using the extremal principle, but in c) you need to find a different
argument since a set of positive real numbers need not have a smallest
element. (Of course c) implies a).)
Remark: The same conclusion also holds if you write positive real
numbers at lattice points in the plane, but this is harder to prove. Try
it!

E 10.15 For each colouring consider the number of edges whose

endpoints have different colours.

E 10.16 For each splitting into two groups G1 , G2 consider the number
of pairs of people in G1 who know each other, plus the corresponding
number for G2 .

E 10.18 First imagine you started with enough petrol to go around

the track once. Start anywhere, take all the petrol from each barrel
314 Hints for selected exercises

you pass and note at each point in time how much petrol you have.
Consider the minimum.

E 10.19 First show: if person a is the left-hand neighbour of person b

and a doesn’t know b, then there are persons c and d such that c is
the left-hand neighbour of d, a knows c, and b knows d.
Then consider the seating with the least number of ‘mistakes’.

E 10.20 One approach works as follows: 1. Prove this for n = 4 by

setting b1 = a1 +2 a2 , b2 = a3 +2 a4 and then using Theorem 10.2 several
times. 2. Prove it for n = 3 and numbers a1 , a2 , a3 by applying the
n = 4 case to a1 , a2 , a3 and a4 = a1 +a32 +a3 . 3. Prove in general that
knowing the assertion for n implies its validity for 2n and for n − 1.
Conclude using induction.

E 10.22 Set ai = xi − xi−1 and use the inequality of Exercise E 10.20.

E 10.23 Via calculation: Use the inequality of the

 geometric and arith-
metic mean and the Heron formula: area = s(s − a)(s − b)(s − c)
where s = a+2b+c and a, b, c are the side lengths of the triangle.
Via geometry: Use the solution of Problem 10.3 to prove the equivalent
assertion that among all triangles of given area the equilateral has the
smallest perimeter.

E 10.24 Let S be the set of given points. Consider all pairs (, P)
where  is a line containing at least two points of S and P is a point
of S not lying on . Prove that there is at least one such pair. For each
such pair consider the distance of P from .

E 10.25 Choose the n points in such a way that the n-gon they form
has maximal perimeter.

Exercises in Chapter 11

E 11.1 No. Why not?

E 11.2 Yes, but this is unnecessarily complicated.

E 11.3 No. The number of white and black squares covered by a

domino depends on the position of the domino. With the chessboard
Hints for selected exercises 315

colouring it does not depend on the position, and this is essential for
the argument.

E 11.5 No, for example this number does not change in the move
1324 → 1342.

E 11.6 Use the lemma stated before the theorem.

E 11.8 You always need the same number of steps, no matter how
you go about it. Why? What changes in a controlled way at each
step?

E 11.9 Read off the numbers in a different order, for example in a

wavy line.

E 11.10 Consider a suitable colouring with 3 colours.

E 11.11 Use a suitable colouring with the ‘colours’ 1,2,3,4.

E 11.13 Think about how to quantify the difference between the initial
and the final position.

E 11.14 The armchair’s position is described by its place and its

orientation (how much it is turned relative to the initial position).
How do these interact?

E 11.15 The moves might remind you of adding or subtracting 11. So

the divisibility rule for 11 could give you an idea, compare Exercise
E 8.7.

E 11.16 Parities help.

E 11.18 The number is always even. What happens when you move
p across a diagonal?

E 11.19 The remarks after Problem 11.8 describe problems which

might arise. There are no others. Formulate this as a condition on
r, g, b and prove that the condition is sufficient, by showing how to
get from one position satisfying it (and having at least two balls) to
another position satisfying it. Why is this sufficient for the proof?

E 11.20 Play some games with g = 3, r = 4 and with g = 3, r = 3 to

get an idea what happens. Starting with the numbers (r, g), which
316 Hints for selected exercises

numbers can result after a move? What remains constant in each

move?

E 11.22 Look at some examples. Starting with the numbers (r, g, b),
which numbers can result after a move? What remains constant in
each move?

E 11.23 Colour the holes suitably with three colours and use the
result of Problem 11.8.2

E 11.24 You can only get to the fourth row, so the biggest possible m
is 4. Why? It is easy to find a sequence of moves which gets you to
the fourth row. It is harder to prove that m = 5 is impossible. Here is
an idea: Suppose you could get a peg to position (0, 5). Now assign
numbers to all lattice points in such a way that the sum of numbers
at occupied points does not increase during any move which might
be done at any time during a game. If you choose the numbers well
then the sum of numbers of pegs involved in a putative solution is
smaller than the number at position (0, 5).
How do you choose the numbers for this to work? Here the equation
x2 + x = 1 will play a role. Also, the geometric series will be useful:
if | x | < 1 then 1 + x + x2 + · · · = 1−1 x .

E 11.25 Find quantities which remain constant when rotating any

face by 90 degrees. For example, in a) consider the permutations of
the cubelets and show that the parity remains constant.

2 The statement a) is proved in (A. Bialostocki, An Application of Elementary

Group Theory to Central Solitaire, The College Mathematics Journal, v 29, n 3, Mai
1998, 208-212) with the help of group theory. The proof proposed here is simpler.
References

General literature

Engel, A. (1998). Problem-solving strategies. Problem Books in Mathe-

matics. Springer, New York, NY.
Extensive collection of mathematical problems at all levels of difficulty,
with solutions. Also includes short introductions to problem-solving
strategies, from the veteran coach of the German team of the Interna-
tional Mathematical Olympiad.

Zeitz, P. (1999). The art and craft of problem solving. Wiley, New York,
NY.
Very well written textbook on problem-solving.

Tao, T. C. S. (2006). Solving mathematical problems. A personal perspective.

Oxford University Press.
Practical guide to problem-solving, with many problems, some of them
quite difficult, written by one of the most successful mathematicians of
our time.

Mason, J., Burton, L. and Stacey, K. (2010). Thinking mathematically.

Pearson, 2nd edition.
Practical guide to problem-solving, very detailed, with problems of low
to moderate difficulty.

Pólya, G. (2014). How to solve it. A new aspect of mathematical method.

Princeton University Press
THE classic on problem-solving, first published in 1945, written by
one of the grand masters. Also includes theoretical considerations on
problem-solving from the perspective of teachers. Worth reading!

Pólya, G. (1981). Mathematical discovery. On understanding, learning,

and teaching problem solving. John Wiley & Sons.
Similar to (Pólya, 2014). Some examples are discussed at great length.

© Springer International Publishing AG, part of Springer Nature 2018 317

D. Grieser, Exploring Mathematics, Springer Undergraduate
Mathematics Series, https://doi.org/10.1007/978-3-319-90321-7
318 References

Aigner, M. (2007). Discrete Mathematics. American Mathematical

Society.
A comprehensive introduction to discrete mathematics, for example
counting problems, graph theory and much more.

Aigner, M. and Ziegler, G. M. (2014). Proofs from the book. Springer,

5th edition.
A collection of beautiful mathematical proofs.

Hathout, D. (2013). Wearing Gauss’s jersey. A K Peters/CRC Press.

Nice collection of problems from algebra, combinatorics, trigonometry
and number theory, with solutions, which usually involve a clever idea.
Some mathematical tools are developed along the way.

Berlekamp, E.R., Conway, J.H. and Guy, R.K. (2001-2004). Winning

Ways for your Mathematical Plays, vol. 1-4. A K Peters, 2nd edition.
Extensive collection of mathematical investigations of games, including
nim, peg solitaire, Rubik’s cube and many more.

For Chapter 2

Stanley, R. (2015). Catalan numbers. Cambridge University Press.

214 ways in which the Catalan numbers occur; exercises with solutions.

For Chapter 4

Aigner, M. (1987). Graph theory. A development from the 4-color problem.

BCS Associates.
A general introduction to graph theory, starting with colouring prob-
lems, with many historical remarks.

Lakatos, I. (1976). Proofs and refutations. The logic of mathematical discov-

ery. Edited by John Worrall and Elie Zahar. Cambridge University
Press.
Socratic dialogues on the topic of the title, discussed in the context of
the Euler characteristic.

Prasolov, V. V. (1995) Intuitive topology. American Mathematical

Society.
Very good introduction to the ideas of topology, with many pictures.
References 319

Jordan, D. and Huggett, S. (2009). A topological aperitif. Springer.

Elementary yet rigorous introduction to the basic ideas of topology,
with many pictures and many examples worked out in detail.

Flapan, E. (2016). Knots, Molecules, and the Universe: An Introduction

to Topology. American Mathematical Society.
Elementary introduction to topology, emphasizing intuition, with ap-
plications.

Armstrong, M. A. (1983). Basic topology. John Wiley and Sons.

A rigorous introduction to topology.

Conway, J.H., Burgiel, H. and Goodman-Strauss, H. (2008). The Sym-

metries of Things. Taylor & Francis.
Many beautiful pictures and good explanations on the topic of symme-
try. Includes a chapter on the Euler characterstic.

For Chapter 5

Pugh, C.C. (2010). Real Mathematical Analysis. Springer, 2nd edition.

Introduction to analysis in one and several variables, rigorous and with
many good illustrations.

Apostol, T.M. (1967). Calculus. Vol. I and Vol. II. Blaisdell Publishing
Co., 2nd edition.
A comprehensive introduction to one-variable (vol. I) and multi-variable
(vol. 2) calculus, rigorous, with applications and many examples. A
classic.

For Chapter 8

Hardy, G.H. and Wright, E.M. (2008). An Introduction to the Theory of

Numbers. Oxford University Press, 6th edition.
THE classic text on elementary number theory.

Silverman, J. (2012). A Friendly Introduction to Number Theory. Prentice-

Hall, 4th edition.
Very well written introduction to many aspects of elementary number
theory.
320 References

For Chapter 10

Nahin, P. J. (2004). When least is best. How mathematicians discovered

many clever ways to make things as small (or as large) as possible.
Princeton University Press, Princeton, NJ.
A diverse collection on the topic of the extremal principle, presented
in an entertaining way. Historical account interspersed with modern
examples.

Hildebrandt, S. and Tromba, A. T. (1996). The parsimonious universe.

Shape and form in the natural world. Copernicus, New York.
A diverse collection of geometric extremal problems, with many illus-
trations and including history and applications.

For Chapter 11

Adams, Colin C. (2004). The knot book. An elementary introduction to

the mathematical theory of knots. Providence, RI: American Mathe-
matical Society.
Informal and accessible introduction to knot theory, with applications
in physics, chemistry and biology.