problem solving tactics

Angelo Di Pasquale,

problem solving tactics

was twice a
contestant at
the International
Mathematical
Olympiad. He
completed a PhD M

in Mathematics
at The University A Di Pasquale, N Do & D Mathews A
t

of Melbourne
studying alge-
braic curves. He is currently Director of Training
for the Australian Mathematical Olympiad
Committee (AMOC), and Australian Team
Leader at the International Mathematical
Olympiad. He enjoys composing Olympiad
problems for mathematics contests.

Norman Do
represented
A u s t r a l i a
at the 1997
International
Mathematical
Olympiad.
In 2010 he
obtained a PhD
in Mathematics
from The Uni-
versity of Melbourne. He is now a lecturer
at Monash University, where he researches
problems that combine geometry, topology,
combinatorics, and mathematical physics.
He is currently the Chair of the AMOC

A Di Pasquale, N Do & D Mathews

Senior Problems Committee.
D aniel M athews
was twice a
contestant at
the International
Mathematical
Olympiad. He
obtained a PhD
in Mathematics
from Stanford
University, Cal-
ifornia. He is
currently a lecturer at Monash University
and his research studies relationships
between geometry, topology, and
mathematical physics. He is a member of
the AMOC Senior Problems Committee.

A u s t r a l i a n M a t h e ma t i c s T r u s t
M
33
978-1-876420-75-8

A
E n r i c h m e n t S e r i e s t
 Published by

AMT P u b lishing

Australian Mathematics Trust

University of Canberra Locked Bag 1
Canberra GPO ACT 2601
AUSTRALIA

Copyright ©2014 AMT Publishing

Telephone: +61 2 6201 5137

www.amt.edu.au
AMTT Limited ACN 083 950 341
National Library of Australia Card Number and ISSN
Australian Mathematics Trust Enrichment Series ISSN 1326-0170
Problem Solving Tactics
ISBN 978-1-876420-75-8
The Australian Mathematics Trust

En r i c h m e n t Se r i e s

E d i t o r i a l C o m m i t t ee

• Editor Mike Clapper, Canberra Australia

Warren J Atkins, Newcastle Australia

Ed J Barbeau, Toronto Canada
George Berzsenyi, Pine USA
Ron Dunkley, Waterloo Canada
Walter E Mientka, Lincoln USA
Nikolay Konstantinov, Moscow Russia
Andy Liu, Edmonton Canada
Andrei Storozhev, Canberra Australia
Jordan B Tabov, Sofia Bulgaria
Peter J Taylor, Canberra Australia
John Webb, Cape Town South Africa
The books in this series are selected for their motivating, interesting
and stimulating sets of quality problems, with a lucid expository
style in their solutions. Typically, the problems have occurred in either
national or international contests at the secondary school level.
They are intended to be sufficiently detailed at an elementary level
for the mathematically inclined or interested to understand but, at
the same time, be interesting and sometimes challenging to the
undergraduate and the more advanced mathematician. It is believed
that these mathematics competition problems are a positive influence
on the learning and enrichment of mathematics.
T h e A u s t r a l i a n M at h e mat i c s T r u s t

E n r i c h m e n t S e r i e s

B oo k s in the Series

1 Australian Mathematics Competition 1978–1984 Book 1

WJ Atkins, JD Edwards, DJ King, PJ O’Halloran & PJ Taylor
2 Mathematical Toolchest
AW Plank & NH Williams
3 International Mathematics Tournament of Towns 1984–1989 Book 2
PJ Taylor
4 Australian Mathematics Competition 1985–1991 Book 2
PJ O’Halloran, G Pollard & PJ Taylor
5 Problem Solving Via the AMC
WJ Atkins
6 International Mathematics Tournament of Towns 1980–1984 Book 1
PJ Taylor
7 International Mathematics Tournament of Towns 1989–1993 Book 3
PJ Taylor
8 Asian Pacific Mathematics Olympiad 1989–2000
H Lausch & C Bosch Giral
9 Methods of Problem Solving Book 1
JB Tabov & PJ Taylor
10 Challenge! 1991–1998 Book 1
JB Henry, J Dowsey, AR Edwards, LJ Mottershead,
A Nakos, G Vardaro & PJ Taylor
11 USSR Mathematical Olympiads 1989–1992
AM Slinko
12 Australian Mathematical Olympiads 1979–1995 Book 1
H Lausch & PJ Taylor
13 Chinese Mathematics Competitions and Olympiads 1981–1993
A Liu
14 Polish & Austrian Mathematical Olympiads 1981–1995
ME Kuczma & E Windischbacher
15 International Mathematics Tournament of Towns 1993–1997 Book 4
PJ Taylor & AM Storozhev
16 Australian Mathematics Competition 1992–1998 Book 3
WJ Atkins, JE Munro & PJ Taylor
17 Seeking Solutions
JC Burns
18 101 Problems in Algebra
T Andreescu & Z Feng
19 Methods of Problem Solving Book 2
JB Tabov & PJ Taylor
20 Hungary-Israel Mathematics Competition: The First Twelve Years
S Gueron
21 Bulgarian Mathematics Competition 1992–2001
BJ Lazarov, JB Tabov, PJ Taylor & A Storozhev
22 Chinese Mathematics Competitions and Olympiads 1993–2001 Book 2
A Liu
23 International Mathematics Tournament of Towns 1997–2002 Book 5
AM Storozhev
24 Australian Mathematics Competition 1999–2005 Book 4
WJ Atkins & PJ Taylor
25 Challenge! 1999–2006 Book 2
JB Henry & PJ Taylor
26 International Mathematics Tournament of Towns 2002–2007 Book 6
A Liu & PJ Taylor
27 International Mathematical Talent Search Part 1
G Berzsenyi
28 International Mathematical Talent Search Part 2
G Berzsenyi
29 Australian Mathematical Olympiads 1996–2011 Book 2
H Lausch, A Di Pasquale, DC Hunt & PJ Taylor
30 Methods of Problem Solving Book 3
JB Tabov, EM Kolev & PJ Taylor
31 Australian Mathematics Competition 2006–2012 Book 5
WJ Atkins & PJ Taylor
32 Australian Intermediate Mathematics Olympiads 1999–2013
JB Henry & KL McAvaney
33 Problem Solving Tactics
A Di Pasquale, N Do & D Mathews
 i

About this book

What is this book about?

Each year the Australian Mathematical Olympiad Committee (AMOC) runs two training
schools. These are designed to extend and challenge the mathematical skills of the 25
secondary school students who are invited to attend. Particular emphasis is given to honing
the skill of problem solving.
This book is based on past and present lectures given at the two annual AMOC training
schools. As such it is suitable for

anyone who wishes to qualify for an Olympiad training school in mathematics, either in
Australia or overseas

anyone who has attended an Olympiad training school in mathematics and who would
like to be better prepared should they qualify again for an invitation

interested students, teachers and parents, as it will give an idea of the sorts of mathe-
matics considered there

any mathematically able students, hobbyists or problem solvers, whether local or abroad,
who would find this publication enriching.

What is in this book?

The authors have gone to considerable care to showcase many of the tricks and problem-solving
tactics they consider to be important for Olympiad mathematics and problem solving in
general. Apart from the first chapter the topics are grouped into the four broad traditional
Olympiad divisions of number theory, geometry, algebra and combinatorics.
Most of the sections within each main chapter highlight a particular idea important for
problem solving, thus providing over 150 such ideas in total. Each idea is illustrated with one
or two problems along with solutions. For extra practice, most chapters begin with a list of
problems. Although they are not necessarily in order of difficulty, we have tried to arrange
them so that the first few problems tend to be easier than the later ones.
Mathematics can be quite hard to read and digest and so the style has purposely been kept
rather informal and conversational. The book often gives the impression that it is conversing
with the reader.

How do you use this book?

Most chapters do not depend much on other chapters. Therefore, apart from the first chapter,
most can be studied almost independently of each other. However, where dependencies arise
there are cross references.
It is the opinion of the authors and many others involved in AMOC training schools that the
chief way to improve one’s problem-solving ability is to go through the struggle of trying to
solve problems oneself. So we recommend that:

The focus of the user of this book should not be on reading solutions but on trying
to solve problems.
ii

That is why solutions are not provided to the problems at the beginning of each chapter. It
is also why we recommend that a problem be tried thoroughly with the showcased idea of the
section in mind, before the solution is studied.
We recognise that some problems are relatively easy exercises while others are of the difficulty
of the International Mathematical Olympiad—the pinnacle of problem-solving mathematics
for high school students the world over. So the reader definitely should not expect to be able
to solve all of the problems straight away.

Acknowledgments
Some problems are the inventions of staff members at AMOC training schools. However, many
of the problems have come from contests such as the Australian Mathematical Olympiad
(AMO), national mathematical Olympiads of some other countries, the Asian Pacific Mathe-
matics Olympiad (APMO), the International Mathematical Olympiad (IMO) and problems
shortlisted for the IMO. Since many of these problems have appeared in multiple contests,
in many cases it has been hard to identify their true origin and so we would simply like to
acknowledge all of the above sources.
Although this book has three listed authors, the ideas it contains are the product of many
AMOC staff interacting with each other and with students over many years.
We also express our appreciation to Ross Atkins, Andrew Elvey Price, Ivan Guo, Konrad
Pilch, Chaitanya Rao, Sally Tsang, Graham White and Sampson Wong, who assisted in
proofreading for mathematical content and accuracy, and provided other feedback.
The document was typeset using LATEX. The drawings were done with the help of GeoGebra
and TikZ.

About the authors

Angelo Di Pasquale was twice a contestant at the International Mathematical Olympiad.
He completed a PhD in mathematics at The University of Melbourne in 1999. His research
combined topology, combinatorics and algebra to better understand the relationship between
algebraic curves and their complements. Since 2000 he has been Director of Training for
the Australian Mathematical Olympiad Committee, and since 2002 he has been Australian
Team Leader at the International Mathematical Olympiad. He enjoys composing Olympiad
problems for Australian and international mathematics contests.
Norman Do represented Australia at the 1997 International Mathematical Olympiad and was
Deputy Leader for the Australian team on four occasions. He obtained a PhD in mathematics
from The University of Melbourne in 2010. He has worked at McGill University, Quebec,
and is now a lecturer at Monash University, where he researches problems that combine
geometry, topology, combinatorics, and mathematical physics. He is currently the Chair of
the Australian Mathematical Olympiad Committee’s Senior Problems Committee, which sets
national mathematics Olympiad papers and proposes problems for international competitions.
Daniel Mathews was twice a contestant at the International Mathematical Olympiad, and three
times Deputy Leader of the Australian team. He studied mathematics, law and languages at
The University of Melbourne before obtaining a PhD in mathematics from Stanford University,
California, in 2009. He has worked as a mathematician in France and the US and is currently
a lecturer in the School of Mathematical Sciences at Monash University. His mathematical
research studies the relationship between geometry, topology, and mathematical physics.
 Contents

About this book i

1 Methods of proof 1
1.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Logic and deduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Converse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 If and only if . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Contrapositive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5 Proof by contradiction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.6 Proof by induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.7 Strong induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.8 Proof by exhaustion (case bashing) . . . . . . . . . . . . . . . . . . . . . . . 12
1.9 Pigeonhole principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.10 Advanced pigeonhole principle . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.11 Extremal principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.12 Telescoping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2 Number theory 19
2.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1 Fundamental theorem of arithmetic . . . . . . . . . . . . . . . . . . . . . . . 24
2.2 Pigeonhole principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Dealing with digits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4 Floor function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.5 Square roots and conjugates . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.6 Powers of two . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.7 Euclid’s algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.8 Integers base-n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.9 Construction problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

iii
iv

2.10 Modular arithmetic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.11 Chinese remainder theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.12 From Fermat to Euler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.13 The gcd trick . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.14 Existence of a generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3 Diophantine equations 39
3.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.1 Factorisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2 Monotonicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3 Bounding arguments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4 Polynomial modulus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.5 Quadratic discriminants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.6 Modular arithmetic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.7 Divisibility and gcds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.8 Reduction of variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.9 Infinite descent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.10 Vieta jumping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.11 Cyclotomic recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4 Plane geometry 53
4.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.1 Angle chasing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.2 Cyclic quadrilaterals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.3 One step at a time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.4 Triangle centres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.5 Constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.6 Exploit symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.7 Extend to the circumcircle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.8 Reverse reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.9 Trigonometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.10 Areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.11 Relate to known diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.12 Create beautiful pictures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

5 Important configurations in geometry 75

5.1 A-List: extremely useful . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
A1 Angle bisector and perpendicular bisector . . . . . . . . . . . . . . . 76
A2 Pivot theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
A3 Radical axis theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
A4 Similar switch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2 B-List: very useful . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
B1 Radical axis bisects common tangent . . . . . . . . . . . . . . . . . . 80
B2 Perpendicularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
 v

B3 Alternate segment switch . . . . . . . . . . . . . . . . . . . . . . . . . 82

B4 Ratios for collinearity . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
B5 Points of contact of incircle and excircle . . . . . . . . . . . . . . . . 84
B6 Circumcircle, incentre and excentre . . . . . . . . . . . . . . . . . . . 85
B7 Simson line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
B8 Pascal’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
B9 Desargues’ theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
B10 Quadrilateral and incircle . . . . . . . . . . . . . . . . . . . . . . . . 89
5.3 C-List: useful . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
C1 Nine-point circle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
C2 Euler line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
C3 Four lines and four circles . . . . . . . . . . . . . . . . . . . . . . . . 92
C4 Newton–Gauss line . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
C5 Alternative characterisation of symmedian . . . . . . . . . . . . . . . 94
C6 Convex cyclic hexagon and diagonals . . . . . . . . . . . . . . . . . . 95
C7 Quadrilateral, triangles and incircles . . . . . . . . . . . . . . . . . . 96
C8 Median, inradius and chord of incircle . . . . . . . . . . . . . . . . . 97
C9 Incentre and midpoints . . . . . . . . . . . . . . . . . . . . . . . . . . 98
C10 Incentre, excentre, midpoint and contact points . . . . . . . . . . . . 99
C11 Incentre and chord of incircle . . . . . . . . . . . . . . . . . . . . . . 100
C12 Harmonic quadrilateral . . . . . . . . . . . . . . . . . . . . . . . . . . 101
C13 Incentre and mixtilinear incircle . . . . . . . . . . . . . . . . . . . . . 102
C14 Butterfly theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

6 Incidence geometry 105

6.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.1 Collinear points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.2 Menelaus’ theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.3 Concurrent lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.4 Ceva’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.5 Concyclic points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.6 Power of a point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.7 Radical axes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.8 Ellipses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
6.9 Pascal’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

7 Transformation geometry 125

7.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
7.1 Translations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
7.2 Rotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
7.3 Dilations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
7.4 Spiral symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
7.5 Affine transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
vi

8 Complex numbers 137

8.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
8.1 Addition ideas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
8.2 Angles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
8.3 Multiplication ideas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
8.4 Similarity ideas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
8.5 Roots of unity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

9 Polynomials 149
9.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
9.1 Identity theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
9.2 Division algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
9.3 Fundamental theorem of algebra . . . . . . . . . . . . . . . . . . . . . . . . . 155
9.4 Vieta’s formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
9.5 Integer polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
9.6 Complex numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
9.7 Algebraic trickery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
9.8 Irreducibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
9.9 Factorisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
9.10 Polynomials modulo p (upstairs–downstairs) . . . . . . . . . . . . . . . . . . 163
9.11 Polynomials modulo P (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
9.12 Lagrange interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
9.13 Root focus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

10 Functional equations 169

10.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
10.1 Cauchy’s functional equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
10.2 Guess and hope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
10.3 Substitutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
10.4 Injective, surjective and bijective . . . . . . . . . . . . . . . . . . . . . . . . 176
10.5 The associative trick . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
10.6 Exploit symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
10.7 Involutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
10.8 Fixed points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
10.9 Somewhere versus everywhere . . . . . . . . . . . . . . . . . . . . . . . . . . 182
10.10 Completely multiplicative functions . . . . . . . . . . . . . . . . . . . . . . . 183
10.11 Well-ordering of N+ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

11 Inequalities 187
11.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
11.1 Squares are non-negative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
11.2 AM–GM inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
11.3 Rearrangement inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
11.4 Cauchy–Schwarz inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
 vii

11.5 Power means inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

11.6 Jensen’s inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
11.7 Substitutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
11.8 Addition and multiplication of inequalities . . . . . . . . . . . . . . . . . . . 198
11.9 Expand and conquer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
11.10 Homogeneous inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
11.11 Muirhead’s inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
11.12 Weighted inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

12 Geometric inequalities 207

12.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
12.1 Triangle inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
12.2 Reflection principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
12.3 Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
12.4 Trigonometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
12.5 Parametrisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
12.6 Ptolemy’s inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
12.7 Locus and tangency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
12.8 Isoperimetric inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
12.9 Incircle substitution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
12.10 Triangle formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222

13 Combinatorics 225
13.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
13.1 Addition and multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
13.2 Subtraction and division . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
13.3 Binomial identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
13.4 Bijections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
13.5 The supermarket principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
13.6 Pigeonhole principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
13.7 Principle of inclusion–exclusion . . . . . . . . . . . . . . . . . . . . . . . . . 234
13.8 Double counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
13.9 Injections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
13.10 Recursion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
13.11 Double counting via tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
13.12 Combinatorial reciprocal principle . . . . . . . . . . . . . . . . . . . . . . . . 241

14 Graph theory 243

14.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
14.1 Degree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
14.2 Directed graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
14.3 Connected graphs, cycles and trees . . . . . . . . . . . . . . . . . . . . . . . 248
14.4 Complete graphs and bipartite graphs . . . . . . . . . . . . . . . . . . . . . . 250
14.5 Pigeonhole principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
viii

14.6 Euler trails . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251

14.7 Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
14.8 Extremal principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
14.9 Count and count again . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
14.10 Planar graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
14.11 Polyhedra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
14.12 Graph theory and inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . 258

15 Games and invariants 261

15.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
15.1 Number invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
15.2 Parity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
15.3 Modular arithmetic invariants . . . . . . . . . . . . . . . . . . . . . . . . . . 269
15.4 Colouring invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
15.5 Monovariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
15.6 Invariants as cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
15.7 Permutation parity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
15.8 Combinatorial games . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
15.9 Position analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
15.10 The copycat strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
15.11 Pairing strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
15.12 Strategy stealing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278

16 Combinatorial geometry 281

16.0 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
16.1 Proof by contradiction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
16.2 Extremal principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
16.3 Perturbation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
16.4 Induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
16.5 Discrete intermediate value theorem . . . . . . . . . . . . . . . . . . . . . . . 288
16.6 Convex hull . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
16.7 Euler’s formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
16.8 Pigeonhole principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
16.9 Colouring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292

17 Appendices 293
17.1 How do complex numbers work? . . . . . . . . . . . . . . . . . . . . . . . . . 293
17.2 Function notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
17.3 Directed angles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
17.4 Some useful triangle formulas . . . . . . . . . . . . . . . . . . . . . . . . . . 298

Index 299
 Methods of proof
1
Proofs are the essence of mathematics. They are nothing more than logical arguments which
present the solution to a mathematical problem beyond all possible doubt. There is no set
format for a proof, no particular way in which it must be presented on a page. Some people
may try to tell you that proofs must appear in two columns, with statements on the left and
explanations on the right—but this is complete nonsense! As long as you have sufficiently
clear assumptions, a logical argument that is fully explained, and the correct conclusion, then
you have a proof.1
So how do you know whether or not you have a rigorous proof to a mathematical problem?
Well, this is a difficult question to answer. Being able to write correct proofs without leaving
out important details is a skill which can only be learned through experience—that is, by
reading and writing them yourself. However, the following little test might help you on your
way. Imagine that you have an inquisitive younger sibling who is looking over your shoulder as
you write your proof and constantly interrogating you. ‘What does this mean?’ she might ask,
or ‘Why are you doing that?’ or ‘How does that make sense?’ If you claim that something
is obvious, she will ask why and if you are tempted to be vague, she just won’t understand.
Your goal, of course, is to make sure that she understands and to answer all of her questions
before she has even asked them. In order to accomplish this, you should provide a clear
explanation for every single fact you write down that is not blatantly obvious.
Apart from the logical aspect of your proof, there are a few other points to keep in mind.
Proofs should always be clear, concise, and most definitely legible. Keep in mind that a nice
proof to a problem is usually shorter than a messy one. This is often achieved by using good
notation and providing the right definitions. A common problem is for people to provide
extraneous material, which does not contribute to the main argument, or to write long essays,
which are not enlightening at all. One way to avoid these pitfalls is to break a problem
into smaller chunks, which can be individually proved and then reassembled to provide the
complete proof. In general, a good proof should take the reader on a pleasant mathematical
journey ending at the desired result.

1.0 Problems
1. Prove that if a + b is an irrational number, then at least one of a or b is irrational.
1 Thereis a rigorous notion of what constitutes a mathematical proof, but for our purposes and for the
purposes of most modern mathematics, our informal explanation is sufficient.
2 1 Methods of proof

2. Show that at any party, there are always at least two people with exactly the same
number of friends at the party.
19
3. The equal temperament tuning2 of musical instruments is based on the fact that 2 12 is
very close to 3.
Show that there can be no perfect tuning3 by proving that if 2x = 3, then x must be
irrational.
√
4. If m and n are positive integers, prove that m n is either a positive integer or irrational.
5. Prove that there are infinitely many prime numbers of the form 6n + 5, where n is a
positive integer.
6. For every positive integer n, prove that
1 1 1 1 n−1
+ + + ··· + = .
1×2 2×3 3×4 (n − 1) × n n

7. Prove that n2 < 2n for every integer n ≥ 5.

8. For every positive integer n, prove that

n(n + 1)(2n + 1)
1 2 + 2 2 + 3 2 + · · · + n2 = .
6

9. For every positive integer n, prove that

n2 (n + 1)2
13 + 23 + 33 + · · · + n3 = ,
4
and go on to conclude that

13 + 23 + 33 + · · · + n3 = (1 + 2 + 3 + · · · + n)2 .

10. Recall that the Fibonacci sequence is defined recursively by F1 = 1, F2 = 1 and

Fn+1 = Fn + Fn−1 , for n ≥ 2.
Prove the following identity for Fibonacci numbers: for all n ≥ 1,

F12 + F22 + · · · + Fn2 = Fn Fn+1 .

11. Prove that every positive integer can be uniquely expressed as a sum of different numbers,
where each number is of the form 2n for some non-negative integer n.
1
12. Suppose x is a real number such that x + x is an integer.
n 1
Prove that x + xn is also an integer for any positive integer n.
13. Any finite collection of lines in the plane divides the plane up into regions.
Prove that it is possible to colour each of these regions either black or white in such a
way that no two regions which share a common edge have the same colour.
2 This tuning has perfect octaves and almost perfect fifths. Octaves are tuned so that the ratio of the frequency
of the pitch of the higher note to that of the note an octave lower is 2 : 1. Perfect fifths are tuned so that
the ratio of the frequency of the pitch of the higher note to the note a fifth below is 3 : 2. Standard equal
temperament tuning says that rising by an almost perfect fifth 12 times should be the same as rising by a

12
perfect octave seven times. Thus 23 ≈ 27 .
3 By perfect tuning, we mean that all octaves and fifths are perfect.
1.0 Problems 3

14. Show that if there are five points in a square with side length 1 metre, then there exist
two of them which are less than 75 centimetres apart.

15. Four points are given inside a square with side length 8 metres.
√
(a) Prove that two of them are less than 65 metres apart.
(b) Can you prove, beyond a shadow of a doubt, that two of them are less than 8
metres apart?

16. (a) Prove that if x and y can each be written as the sum of the squares of two integers,
then so can xy.
(b) Prove that if x and y are both of the form a2 + 2b2 (a, b ∈ Z), then so is xy.
(c) Let k be a fixed integer. Prove that if x and y are both of the form a2 + kb2
(a, b ∈ Z), then so is xy.

17. Consider the non-empty subsets of {1, 2, . . . , n}. For each of these subsets, consider the
reciprocal of the product of its elements.
Determine the sum of all of these numbers.

18. A finite set of chords is drawn in a circle such that each of them passes through the
midpoint of another chord.
Prove that all of the chords must be diameters.

19. Show that if we take n + 1 numbers from the set {1, 2, 3, . . . , 2n}, there must exist two
which have no common factor greater than 1.
Does this remain true if we take n numbers?

20. A circular island is divided into states by a number of chords of the circle. Consider a
tour that starts and ends in the same state without passing through the intersection of
any two borders.
Prove that the tour must involve an even number of border crossings.

21. Suppose you are given a balance scale and a collection of weights whose masses are
1, 3, 32 , 33 , . . ..

(a) Prove that using these masses you can determine the weight of any object whose
mass is a positive integer.
(b) Prove that apart from interchanging the contents of the left and right pans of the
scale, the configuration of masses on the pans that correctly determines the weight
is unique.
(c) Prove that the weights 1, 3, 32 , 33 , . . . are the only integral weights that uniquely
determine the weight of every integral mass.

22. A group of people played in a tennis tournament where each person played exactly one
match against every other person.
Prove that it is always possible to put the players in a line so that the first player beat
the second, the second player beat the third, all the way down to the last player.

23. A polygon is divided into triangles by diagonals whose endpoints are the vertices of the
polygon in such a way that no two of the diagonals intersect inside the polygon.
Prove that it is possible to colour the vertices of the polygon with three colours so that
the three vertices of each triangle have different colours.
4 1 Methods of proof

24. The Tower of Hanoi is a mathematical puzzle consisting of three rods and n discs of
distinct sizes, which can slide onto any of the three rods. The puzzle starts with the
discs neatly stacked in order of size on one rod, the smallest at the top, as shown in
the diagram below. The aim is to transfer the entire stack to another rod by moving a
disc from the top of one stack to the top of another stack in such a way that no disc is
placed on top of a smaller disc.
Prove that the task can be accomplished in 2n − 1 moves but not in fewer moves.

25. Thirty coins lie on a table, with 17 of them showing heads. Your task, should you
choose to accept it, is to separate the coins into two piles, not necessarily of the same
size, each of which has the same number of heads showing. Unfortunately, you happen
to be blindfolded and cannot feel the difference between the two sides of a coin.
How can you perform the task?
26. Each of the numbers 1, 2, 3, . . . , n2 is written in one of the squares of an n×n chessboard.
Show that there exist two squares which share a vertex or an edge whose entries differ
by at least n + 1.
27. Each square of an 8 × 8 chessboard has a real number written in it in such a way that
each number is equal to the geometric mean4 of all the numbers a knight’s move away
from it.
Is it true that all of the numbers must be equal?

28. There are 1000 positive numbers written at different points on the circumference of
a circle. If the numbers x, y, z appear in a row in that order, then it is known that
xz = y 2 .
Prove that all of the numbers are equal.

29. There are m horizontal lines and n vertical lines drawn in the plane. Each point of
intersection between a pair of lines is coloured in one of 100 colours.
Find values of m and n such that, no matter how the colouring is performed, there
always exists a rectangle whose vertices are the same colour.
30. Prove that from any set of 10 distinct two-digit numbers, it is possible to select two
disjoint subsets whose members have the same sum.
31. Does there exist a convex polyhedron such that no two of its faces have the same number
of edges?

4 See section 11.2 if you don’t know what this is.

1.1 Logic and deduction 5

1.1 Logic and deduction

Mathematics is brimming with statements of the form

If X, then Y .

For example, we could take X to be the statement ‘Rex is a dog’ and Y to be the statement
‘Rex is an animal’. Being such lazy creatures, we mathematicians have invented the following
shorthand for such statements, which is often read as ‘X implies Y ’.

X⇒Y

Now if X ⇒ Y and Y ⇒ Z, then you can automatically deduce that X ⇒ Z. One of the
easiest ways to write a proof is to string together a chain of deductions in this manner, starting
with the assumptions of the problem and ending with the conclusion of the problem. This is
often called a direct proof, an example of which follows.

Problem Prove the quadratic formula, which states that if ax2 + bx + c = 0 and a =
6 0, then
√ √
−b + b2 − 4ac −b − b2 − 4ac
x= or x = .
2a 2a

Solution Since a 6= 0, we can divide both sides by a to obtain

b c
ax2 + bx + c = 0 ⇒ x2 + x + = 0.
a a
Now we use an algebraic trick, known as completing the square, to write the left-hand side as
2
b2 − 4ac
  
b c b
x2 + x + = x+ − .
a a 2a 4a2

This is a good time to mention that you should never ever take equations like this for granted.
So grab a pen and some paper and check that it’s true for yourself! Once you’ve done that,
you should be convinced that the quadratic equation now takes the following form.
2 √
b2 − 4ac ± b2 − 4ac

b b
x+ = ⇒ x+ =
2a 4a2 2a 2a
√
2
−b ± b − 4ac
⇒ x=
2a
Of course, we were careful to consider both the positive and the negative square roots, as one
should always do, and this completes the proof.

1.2 Converse
For every statement of the form X ⇒ Y , there is something known as the converse, which is
the statement Y ⇒ X. You may be tempted to think that these two statements mean the
same thing, that is, if X ⇒ Y is true, then Y ⇒ X is true. But this is most definitely not
the case. For example, using our two example statements from earlier, it is clear that

Rex is a dog ⇒ Rex is an animal

6 1 Methods of proof

is a true statement, whereas the statement

Rex is an animal ⇒ Rex is a dog

is false, since there is the possibility that Rex could be a lizard or some other animal. So if
we know that a statement is true, there is no guarantee whatsoever that its converse is also
true. But sometimes it is, as in the following well-known example.

Problem Pythagoras’ theorem states that, if a right-angled triangle has side lengths a, b, c,
where c is the length of the hypotenuse, then a2 + b2 = c2 .
Assuming that Pythagoras’ theorem is true, prove the converse of Pythagoras’ theorem.

Solution Of course, the first thing we must do is write down what the converse actually is.

If a triangle has side lengths a, b, c, where a2 + b2 = c2 , then the triangle is

right-angled and c is the length of the hypotenuse.

Let’s construct a right-angled triangle whose legs have lengths a and b, and let the hypotenuse
have length d. The reason for this is because we can now invoke Pythagoras’ theorem, which
we already know to be true. It tells us that a2 + b2 = d2 . Using this in conjunction with our
assumption that a2 + b2 = c2 , we deduce that c2 = d2 , which implies that c = d.
Therefore, the triangle with side lengths a, b, c that we were given possesses exactly the same
side lengths as the right-angled triangle that we have constructed. This means that the two
triangles are, in fact, congruent. So the given triangle was indeed right-angled, as we intended
to prove. Furthermore, the equation a2 + b2 = c2 implies that c is the longest side length in
the triangle, and hence is the length of the hypotenuse.

In the previous solution, we relied on Pythagoras’ theorem to prove its converse. This is a
rather general strategy, so keep the following point in mind. If you are given a true statement
and asked to prove its converse, then it is often advantageous, sometimes crucial, to use the
original statement itself.

1.3 If and only if

Is there some way to combine Pythagoras’ theorem and its converse into one super-duper
Pythagorean statement? Yes, there most certainly is!

Pythagoras’ theorem and its converse Suppose that a triangle has side lengths a, b, c,
where c is the longest side. Then the triangle is right-angled if and only if a2 + b2 = c2 .

In general, the statements ‘If X, then Y ’ and ‘If Y , then X’ can be combined to create the
single statement ‘X if and only if Y ’. You can probably guess that the mathematical notation
for this is simply
X ⇔ Y.
Now if someone actually asks you to prove a statement of the form X ⇔ Y , then what do
you do? The simplest approach is to split the problem into two parts. First prove X ⇒ Y ,
then prove Y ⇒ X. The next problem not only demonstrates this point but also provides us
with a useful way to test whether or not a number is divisible by 7. You should think about
why this is so.
1.4 Contrapositive 7

Problem If a and b are integers, prove that 10a + b is divisible by 7 if and only if a − 2b is
divisible by 7.

Solution As with most ‘if and only if’ statements, the proof naturally divides into two parts.

If 10a + b is divisible by 7, then a − 2b is divisible by 7.

If 10a + b is divisible by 7, then certainly 5(10a + b) = 50a + 5b is divisible by 7. And if
50a + 5b is divisible by 7, then certainly (50a + 5b) − 7(7a + b) = a − 2b is divisible by 7.
This proves the statement in one direction.

If a − 2b is divisible by 7, then 10a + b is divisible by 7.

If a − 2b is divisible by 7, then certainly (a − 2b) + 7(7a + b) = 50a + 5b is divisible by 7.
And if 50a + 5b is divisible by 7, then certainly (50a + 5b) ÷ 5 = 10a + b is divisible
by 7. This proves the statement in the opposite direction and completes the proof.

Hopefully, you will have noticed that the two parts are very similar in nature. Although
this is reasonably common, there will be times when one direction is significantly easier to
prove than the other. And, as we mentioned in the previous section, once you’ve proved the
statement in one direction, you can often use it to your advantage to prove the statement in
the other direction.

1.4 Contrapositive
The contrapositive is a way of turning a logical statement on its head to give an equivalent
logical statement. For example, instead of saying

If Rex is a dog, then Rex is an animal,

we could say the equivalent statement

If Rex is not an animal, then Rex is not a dog.

In general, the contrapositive of the statement X ⇒ Y is the equivalent statement

‘not Y ’ ⇒ ‘not X’,

where ‘not X’ is the opposite of X and ‘not Y ’ is the opposite of Y . By calling a statement
and its contrapositive equivalent, we mean that if the statement is true, then its contrapositive
is true, while if the statement is false, then its contrapositive is false. In other words, proving
either one will automatically prove the other. Note that if you take the contrapositive of
the contrapositive, then you actually end up with the statement you started with. As an
example, consider the following two statements, which are the contrapositives of each other.

If a shape is a rectangle, then it has four sides.

If a shape does not have four sides, then it is not a rectangle.

Problem If a and b are real numbers such that ab is irrational, then at least one of a and b
must be irrational.
8 1 Methods of proof

Solution The contrapositive of this statement is the following.

If a and b are both rational, then ab is rational.
This is a fact so simple that you probably take it for granted all the time! Let’s prove it
by writing a = aa12 and b = bb12 , where a1 , a2 , b1 , b2 are integers with a2 and b2 non-zero. The
product is simply ab = aa12 bb12 , which is clearly rational since a1 b1 and a2 b2 are integers with
a2 b2 non-zero.

1.5 Proof by contradiction

Proof by contradiction is a most useful method of proof, one that you have almost definitely
used before without even realising it. For example, how can you prove that the Sun is far
away from the Earth? This is easy, because if the Sun were close to the Earth, then it would
be too hot for you to be alive reading this sentence!5
The idea is that if you assume the opposite of what you are trying to prove, use only correct
logical deductions, and arrive at a nonsensical conclusion, then something must be wrong
somewhere. And since all of your logical deductions were correct, what must be wrong is your
initial assumption. So the opposite of what you are trying to prove is false. In other words,
what you are trying to prove is true.
The two examples we’ll look at are mathematical gems which were known to the ancient
Greeks over two thousand years ago. The first appears in a work by the mathematician Euclid
while the second is usually attributed to Hippasus, a philosopher and disciple of Pythagoras.

Problem Prove that there are infinitely many primes.

Solution Let’s suppose that there are only finitely many primes, so that we can list them
all as p1 , p2 , . . . , pn . Euclid asks us to consider the number N = p1 × p2 × · · · × pn + 1. The
main fact that we’ll use is that two consecutive integers cannot be divisible by the same prime.
Since N − 1 is divisible by p1 , it’s impossible for N to be divisible by p1 as well. Similarly, it’s
impossible for N to be divisible by p2 or p3 or any of the primes in our supposedly complete
list. So what does this all mean? It means that if we look at the prime factors of N , they
must come from outside our list. This contradicts the fact that we started with a complete
list of the primes. Since we’ve shown that no finite list of primes can be complete, it follows
that there are infinitely many primes.

If you’re seeing proof by contradiction for the first time, then it can be a little baffling. Make
sure you understand the previous argument completely before looking at the next example.
Eventually, you should not only be able to understand proofs by contradiction, but also come
up with your own.
√
Problem Prove that 2 is irrational.
√
Solution Let us assume√on the contrary that 2 is actually a rational number and hope to
find a contradiction. If 2 were rational, then we could express it as pq , where p and q are
√
integers which have no common factors greater than 1. Start with the equation pq = 2 and
remove square roots and fractions to obtain
p √ p2
= 2 ⇒ =2 ⇒ p2 = 2q 2 .
q q2
5 Of course from a different point of view, the Sun is close to the Earth, relative to other astronomical objects!
1.6 Proof by induction 9

Since the right-hand side is even, the left-hand side is even. So p2 , and hence p, is even.
Therefore, we can write p = 2m for some integer m and substitute this back into the previous
equation.
(2m)2 = 2q 2 ⇒ 2m2 = q 2
Now it’s time for the left-hand side to be even, which forces the right-hand side to be even.
So q 2 , and hence q, is even. But hold on a second! We’ve assumed that p and q have no
common factors greater than 1 and also proved
√ that both p and q are even! This contradiction
means that our√original assumption, that 2 is rational, was incorrect. Therefore, we must
conclude that 2 is irrational, which was what we set out to do.

Already, we’ve learnt about the converse, the contrapositive and contradiction, which may
create some confusion. So take some time to learn the differences between these three concepts!

1.6 Proof by induction

Suppose that you have an infinite supply of dominoes, numbered 1, 2, 3, and so on. They
are all standing upright on a table and you want them all to fall over. One way of achieving
this goal is to knock over the domino labelled 1, then knock over the domino labelled 2, then
knock over the domino labelled 3, and so on. This would take infinitely long and is simply far
too energy consuming for your average lazy mathematician. A much smarter way to do this
would be to

knock over the domino labelled 1

make sure that if the domino labelled k falls, then it knocks over the domino labelled
k + 1.

This is, in essence, what proof by induction is all about.

1 2 3 . . . k k+1 k+2

Analogously, suppose that you want to prove that some statement is true for all positive
integers n. An incredibly silly way to do this would be to prove it for n = 1, then for n = 2,
then for n = 3, and so on, but you would never get to the end of it. Instead, all you need to
do is

prove the base case, that is, prove the statement for n = 1

prove the inductive step, i.e. prove that whenever the statement is true for n = k (the
inductive hypothesis), then it is also true for n = k + 1.

Using this idea to prove a statement for all positive integers n is called proof by induction.
This may all seem pretty confusing to you at the moment, but the following example might
help to clarify the situation.
10 1 Methods of proof

Problem For every positive integer n, prove that

n(n + 1)
1 + 2 + 3 + ··· + n = .
2

Solution For the base case, it is a simple matter to verify that the statement is true for
n = 1. Indeed, the left-hand side is simply 1 while the right-hand side is 1(1+1)
2 . These two
being equal takes care of the base case.
Next, for the inductive step, we can make use of the inductive hypothesis

k(k + 1)
1 + 2 + 3 + ··· + k =
2
in order to prove that the statement is true for n = k + 1. But if we know the value of
1 + 2 + 3 + · · · + k, then surely it must be an easy matter for us to determine the value of
1 + 2 + 3 + · · · + k + (k + 1). In fact, we have

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

And this is exactly the statement that we are trying to prove, with n = k + 1. This takes
care of the inductive step.
But wait a moment! Do we actually know that the statement is true for n = k? Well, no
we don’t. But what we have shown is that if the statement is true for n = k, then it must
also be true for n = k + 1. And since we already know it’s true for n = 1, then it must also
be true for n = 2. And since it’s true for n = 2, then it must also be true for n = 3. And
since it’s true for n = 3, then it must also be true for n = 4, and so on. In this way we have
managed to prove by induction that, for every positive integer n,

n(n + 1)
1 + 2 + 3 + ··· + n = .
2

Problem An L-tromino is a shape formed by joining three unit squares along their edges to
form an L shape.
Prove that, if any square of a 2100 × 2100 chessboard is removed, then the part of the board
which remains can be tiled by L-trominoes.

Solution Hopefully, one of the first things you realise when reading this problem is the fact
that the number 100 is not important at all! In fact, we will prove that the statement is true
for a 2n × 2n chessboard, where n is any positive integer. Needless to say, we will proceed by
induction.
The base case n = 1 is, as usual, extremely simple though entirely necessary to take care of.
We simply note that removing a square from a 2 × 2 chessboard always leaves an L-tromino.
Next, for the inductive step, we assume the inductive hypothesis. In other words, that it is
possible to tile a 2k × 2k chessboard with any square removed. Our goal now is to prove that
it is possible to tile a 2k+1 × 2k+1 chessboard with any square removed. How can we relate
these two facts? The idea is to divide the larger 2k+1 × 2k+1 board into four blocks, each
one a copy of the smaller 2k × 2k board. After rotating the board, we may assume that the
removed square lies in the top-left block. The inductive hypothesis guarantees that we can
tile the remaining part of the top-left block, but what to do with the remaining three blocks?
1.7 Strong induction 11

If we want to rely on the inductive hypothesis again, it would be nice to remove one square
from each of the remaining three blocks. We can do this by placing an L-tromino as shown
in blue in the diagram above. Once again, the inductive hypothesis guarantees that we can
tile the remaining parts of the remaining three blocks. This completes the inductive step.
Therefore we have proved by induction that, for every positive integer n, a 2n × 2n chessboard
with any square removed can be tiled by L-trominoes.

Proof by induction is such an extremely important concept that we will finish this section
with one further example.

Problem For every integer n ≥ 4, prove that

2n < n!.

(Recall n! = n × (n − 1) × · · · × 3 × 2 × 1 for any positive integer n, and 0! = 1 by convention.)

Solution This time, the statement we are trying to prove is only true for integers n ≥ 4.
This causes no problem since we can simply start from the base case n = 4 instead.
This is easy enough to verify, because 16 = 24 < 4! = 24.
Next, for the inductive step, we assume the inductive hypothesis. In other words, that 2k < k!
for some value of k ≥ 4. From this assumption, we would like to obtain the fact that the
statement is true for n = k + 1, that is, 2k+1 < (k + 1)!. This is easily done, by starting with
the inductive hypothesis and multiplying both sides by 2.

2k < k! ⇒ 2k+1 < 2 × k! < (k + 1) × k! = (k + 1)!

And there we have it! Knowing that the statement is true for n = 4 tells us that it is true for
n = 5. And knowing that the statement is true for n = 5 tells us that it is true for n = 6, and
so on. So we have managed to prove by induction that 2n < n!, for every integer n ≥ 4.

1.7 Strong induction

One aspect of induction that makes it so powerful is that you are allowed to assume the
statement is true for some positive integer k in order to prove it for the positive integer k + 1.
However, sometimes it helps to know that the statement is true for 1, 2, . . . , k in order to
12 1 Methods of proof

prove it for k + 1. This type of induction is certainly legitimate and is often called strong
induction. This is a more advanced method of proof which can be extremely useful, such as
in the solution to the following problem.

Problem You might already know that the Fibonacci sequence is defined by F1 = 1, F2 = 1
and Fm+1 = Fm + Fm−1 for m ≥ 2.
Prove that every positive integer can be expressed as a sum of terms from the Fibonacci
sequence, no two of which appear consecutively.

Solution The Fibonacci sequence begins

1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, . . ..

Let us call a positive integer representable if it can be expressed as a sum of non-consecutive

Fibonacci numbers.
It is easy enough to show that the first few positive integers are representable: 1 = 1, 2 = 2,
3 = 3, 4 = 3 + 1, 5 = 5, 6 = 5 + 1, 7 = 5 + 2, 8 = 8, 9 = 8 + 1, and 10 = 8 + 2. In order to
take care of the base case, we don’t need to prove the statement for all of these values, but it
certainly doesn’t hurt.
Next, assume that every number from 1 up to k is representable. This is the strong inductive
hypothesis. We are now going to show that k + 1 is representable. If k + 1 is actually a
Fibonacci number, then it is, of course, representable. Otherwise, let Fm be the largest
Fibonacci number less than it. Then k + 1 − Fm is a positive integer which must be smaller6
than Fm−1 and, by the strong induction hypothesis, must be representable. Appending Fm
onto the representation of k + 1 − Fm gives a representation of k + 1. So we have managed to
prove by strong induction that every positive integer is representable.

This result is often called Zeckendorf ’s theorem. It gives us a pretty interesting way to
represent positive integers, certainly more interesting than the customary base 10!

1.8 Proof by exhaustion (case bashing)

It is often a good strategy to divide and conquer. By this, we mean splitting a large problem
into several smaller, simpler problems. Handling all of these smaller cases can be tough work,
and that’s why we refer to such a proof as a proof by exhaustion or as a case bash. Whenever
you opt for this sort of proof, you should make absolutely sure that your cases are listed
clearly and that they are indeed exhaustive.

Problem Prove that if n is an integer, then n7 − n is divisible by 7.

Solution First, we notice the factorisation

n7 − n = (n − 1)n(n + 1)(n2 − n + 1)(n2 + n + 1).

We can divide the problem into seven cases according to the remainder that n leaves after
division by 7.

Case 1: n = 7q + 1
In this case, the factor n − 1 = 7q is divisible by 7.
6 Your inquisitive younger sibling is asking why k + 1 − Fm is smaller than Fm−1 . Please provide a good
reason.
1.9 Pigeonhole principle 13

Case 2: n = 7q + 2
In this case, the factor n2 + n + 1 = 49q 2 + 35q + 7 is divisible by 7.

Case 3: n = 7q + 3
In this case, the factor n2 − n + 1 = 49q 2 + 35q + 7 is divisible by 7.

Case 4: n = 7q + 4
In this case, the factor n2 + n + 1 = 49q 2 + 63q + 21 is divisible by 7.

Case 5: n = 7q + 5
In this case, the factor n2 − n + 1 = 49q 2 + 63q + 21 is divisible by 7.

Case 6: n = 7q + 6
In this case, the factor n + 1 = (7q + 6) + 1 = 7q + 7 is divisible by 7.

Case 7: n = 7q
In this case, the factor n = 7q is divisible by 7.

Therefore, in all seven cases, at least one of the factors of n7 − n is divisible by 7. Since these
cases account for every possible situation, we can conclude that if n is an integer, then n7 − n
is divisible by 7, as desired.7

1.9 Pigeonhole principle

The pigeonhole principle is a hugely powerful technique in the hands of an experienced problem
solver. In its simplest form, it states that if you place n + 1 pigeons into n pigeonholes, then
at least one pigeonhole contains at least 2 pigeons. You might think that this is completely
self-evident, although the more astute and sadistic reader might have noticed that it isn’t
even correct. Take, for example, 1 15 pigeons placed in each of 5 pigeonholes! Anyway, it’s a
simple matter to amend the problem and the result is the following.

Pigeonhole principle If you place n + 1 pigeons into n pigeonholes, then at least one
pigeonhole will contain at least 2 pigeons . . . as long as you don’t cut them up!

This is now correct and, despite being blatantly obvious, is surprisingly useful. As with many
of the techniques that we will learn, the power of the pigeonhole principle arises from the
ingenious ways in which it can be applied, and what better way to demonstrate this than
with an example.
The pigeonhole principle is particularly useful for problems where you have to show the
existence of something without being able to construct it explicitly.
For example, suppose that you were asked to prove that there are two people in Australia
with the same number of hairs on their head. Your first instinct might be to go and find two
completely bald people! But if that seems too difficult for you, then you can also solve the
problem with a little thought and an application of the pigeonhole principle. For there are
over twenty million people in Australia, and each one of them has fewer than one million hairs
on their head. In fact, the average head full of hair will have approximately one hundred
thousand hairs. So the population of Australia will form our pigeons and we will put them
into pigeonholes numbered from 0 up to 999999 according to the number of hairs on their
head. The pigeonhole principle then guarantees that there will be at least two who end up in
the same pigeonhole.
7A much quicker way to deal with this question is provided by Fermat’s little theorem, found in section 2.12.
14 1 Methods of proof

The next couple of problems illustrate how the pigeonhole principle can be used in number
theory and combinatorics.

Problem Prove that any set of n integers has a non-empty subset whose sum is divisible
by n.

Solution Let the integers be a1 , a2 , . . . , an and consider the following numbers.

s1 = a1
s2 = a1 + a2
..
.
sn = a1 + a2 + · · · + an

We treat these n numbers as our pigeons and place them into pigeonholes according to their
value modulo n. If the numbers s1 , s2 , . . . , sn are all distinct modulo n, then there must be
one of them, say sk , which is divisible by n. In other words, a1 + a2 + · · · + ak is divisible
by n.
Otherwise, the pigeonhole principle guarantees that two of the numbers, say si and sj , are
the same modulo n. This means that sj − si is divisible by n, where we may assume without
loss of generality that i < j. In other words, ai+1 + ai+2 + · · · + aj is divisible by n.

Problem Suppose that each square in a 3 × 7 grid is coloured blue or white.

Prove that there exist two rows and two columns such that the four squares contained in
their intersection are the same colour.

Solution One rather foolish way to solve this problem would be to consider all of the possible
colourings of the grid, of which there are 23×7 = 2097152. Of course, this is a rather inelegant
solution, not to mention the fact that it could take months! We’ll take a shortcut to the
solution by making use of the pigeonhole principle. One feature of the problem which suggests
that we should do so is the fact that we are asked to prove the existence of something, without
demonstrating precisely what that something is. But even though we know that we would
like to invoke the pigeonhole principle, some ingenuity is still required to know exactly how
to proceed.
Let us call four squares formed by the intersection of two rows and two columns a quartet
and refer to such a quartet as monochromatic if all four squares are the same colour. Our
first observation is that there are eight possibilities for each column, as shown in the diagram
below.

1 2 3 4 5 6 7 8

Our second observation is that if one of these occurs twice, then we must necessarily have a
monochromatic quartet. Our third observation is that if you use column 7 in conjunction
with columns 3, 5 or 6, then you must have a monochromatic quartet and similarly, if you
use column 8 in conjunction with columns 1, 2 or 4, then you must have a monochromatic
quartet. The problem now divides naturally into three cases.
1.10 Advanced pigeonhole principle 15

Case 1: Column 7 is used.

If any of the remaining six columns are of type 3, 5, 6 or 7, we have a monochromatic
quartet. If, on the other hand, each of the remaining six columns are of type 1, 2, 4 or 8,
then by the pigeonhole principle, two are the same, thereby forcing a monochromatic
quartet.
Case 2: Column 8 is used.
A similar argument to case 1 shows that a monochromatic quartet occurs.
Case 3: Columns 7 and 8 are not used.
Therefore, the seven columns must be of type 1, 2, 3, 4, 5 or 6. So, by the pigeonhole
principle, two are the same, thereby forcing a monochromatic quartet.

1.10 Advanced pigeonhole principle

The pigeonhole principle can be applied in quite an advanced manner. For example, the
pigeonholes don’t have to appear to be the same size, as the next problem demonstrates.

Problem Show that if we take n + 1 numbers from the set {1, 2, 3, . . . , 2n}, there must exist
one which is divisible by another.

Solution Since this problem asks for the existence of two numbers satisfying certain con-
ditions, it seems like a prime candidate for the pigeonhole principle. We are given n + 1
numbers, which will most likely be our pigeons, so it seems sensible to look for n pigeonholes.
Furthermore, these should be constructed so that given any two numbers from the same
pigeonhole, one is divisible by the other. If we can construct n pigeonholes from the set
{1, 2, 3, . . . , 2n} which satisfy these conditions, then we will be done.
Let’s play around with some small numbers then. Clearly, 1 and 2 can be in the same
pigeonhole, although 3 cannot lodge with them. The next number which can join them is 4,
and after that 8. In fact, we can put all numbers of the form 2a into one pigeonhole. If we
start with 3 in a pigeonhole, then the next number which can join it is 6, then 12, then 24,
and so on. So the numbers which are of the form 3 × 2a can all be put into one pigeonhole.
Similarly, the numbers which are of the form 5 × 2a can all be put into one pigeonhole.
Continuing in this way, each positive integer is in exactly one pigeonhole.8

X1 = {1, 2, 4, 8, 16, . . .}
X2 = {3, 6, 12, 24, 48, . . .}
X3 = {5, 10, 20, 40, 80, . . .}
..
.
Xk = {2k − 1, 2(2k − 1), 4(2k − 1), 8(2k − 1), 16(2k − 1), . . .}
..
.

Now that we’ve constructed the pigeonholes, the rest is easy. Given n + 1 numbers from the
set {1, 2, 3, . . . , 2n}, they all belong to one of the n pigeonholes X1 , X2 , . . . , Xn . But by the
pigeonhole principle, two of them must end up in the same pigeonhole. These two numbers
are of the form m × 2a and m × 2b , for some positive integer m. Thus the larger one is
divisible by the smaller one.
8 Your younger sibling wants to know why this is the case.
16 1 Methods of proof

In the previous section, we proved that there are at least two people in Australia with the
same number of hairs on their heads. But it seems likely that, with so many Australians
around, we should be able to find a larger group of people, all with the same number of
hairs on their heads. This idea is the basis for the following more advanced version of the
pigeonhole principle.

Pigeonhole principle If you place kn + 1 pigeons into n pigeonholes, then at least one
pigeonhole will contain at least k + 1 pigeons . . . as long as you don’t cut them up!

Problem Seventeen points are given inside a cube of side length 2.

√
3
Show that there exist three of them which form a triangle of area at most 2 .

Solution We have 17 objects and need to prove the existence of three of them satisfying
certain conditions. Given this information alone, it seems that the problem is begging for the
pigeonhole principle to be used on it! Not only that, the numbers 17 and 3 imply that we
should take the given points to be our pigeons and look for eight pigeonholes. Fortunately,
there is a particularly obvious way to divide a cube of side length 2 into eight pieces, that is,
by slicing it into eight unit cubes.
From here, the solution to the problem is evident, since the pigeonhole principle tells us
that three of our points must lie in the same unit cube. These three points would form a
triangle with the largest area if they were at non-adjacent vertices of the unit cube.9√Note
√
that three such points form an equilateral triangle with side length 2 which has area 23 , as
required.

There is yet another version of the pigeonhole principle which you might as well know about,
even if we won’t require it to solve any of the problems in this particular chapter.

Infinite pigeonhole principle If you place infinitely many pigeons into finitely many
pigeonholes, then at least one pigeonhole will contain infinitely many pigeons.

1.11 Extremal principle

When solving mathematics problems, it is often useful to look at the maximum or minimum
of some value. For example, you might consider the largest, the smallest, the longest, the
shortest, the coolest, the most colourful, and so on. This is precisely the extremal principle.
It isn’t so much a mathematical result, but a useful way of thinking. And thinking in this
way can often make difficult problems seem easy and can make long, complicated proofs turn
into simple, concise ones, as you will soon see.

Problem Consider an m × n rectangular grid, where each of the mn unit squares is labelled
with a real number. Suppose that the number in each square is the average of the numbers in
all neighbouring squares.
Prove that all of the labels must be the same.

Solution We start with the simple fact that every finite set of real numbers has a largest
element, a fact which is not true for infinite sets. So, we can consider the largest number
appearing in the grid and call it M . Consider the numbers in the adjacent squares. They are
9 Yourinquisitive younger sibling is asking why this produces a triangle of maximal area. Can you provide an
explanation?
1.11 Extremal principle 17

all less than or equal to M and yet their average is M . The only way that this can happen is
if all of the numbers in the adjacent squares are also M . So what we’ve proved is that every
square adjacent to one containing the number M also contains the number M . It now follows
that every square in the grid is labelled with the number M .

The next problem illustrates the fact that the extremal principle is particularly valuable when
used in conjunction with proof by contradiction.

Problem Consider 10 black points and 10 white points in the plane, no three of which lie
on a line.
Prove that it is possible to connect each black point with a white point by a line segment
such that no two line segments intersect.

Solution Let us call a set of 10 line segments which connect each black point with a white
point a matching. If we don’t care whether or not the line segments intersect, there are only
finitely many (actually 10!) matchings possible. Let us call the sum of the lengths of the 10
line segments the length of the matching. The idea now is that if you don’t want your lines
to cross, then you probably don’t want to connect points faraway from each other. With
this in mind, we consider a matching which has the shortest length and prove that it avoids
intersecting line segments.
In order to obtain a contradiction, suppose that the line segments AB and CD intersect at
a point X, where A and C are black points while B and D are white points. Then we can
replace these with the line segments AD and BC. By the triangle inequality, which essentially
states that the shortest distance between two points is a straight line segment, we have

AD < AX + XD and BC < BX + XC.

C
A

D B

Adding these two inequalities, we obtain

AD + BC < AB + CD.

So we have constructed a new matching which is shorter than the one which was supposed to
be the shortest. This contradiction implies that the shortest matching cannot contain two
line segments which intersect, so we are done.

There are two important points we should make about the previous proof. First, it was
necessary to mention that there are only finitely many matchings. For if there were infinitely
many, then it might not have been possible to choose one with the shortest length. Second,
we were careful to choose a shortest matching rather than the shortest matching, since it is
certainly possible that there could be many of them. Our proof works no matter which of the
shortest matchings we decide to consider.
18 1 Methods of proof

1.12 Telescoping
This is a useful method of evaluating sums. The idea is simply to rewrite things so that
cancellation systematically occurs. Specifically, if we wish to find a closed form for the sum
a1 + a2 + · · · + an , then we seek a function F which satisfies

ai = F (i + 1) − F (i) for i = 1, 2, . . . , n.

The sum telescopes because all the middle terms cancel out as follows.

an + an−1 + · · · + a1 = F (n + 1) − F (n) + F (n) − F (n − 1) + · · · + F (2) − F (1)

= F (n + 1) − F (1)

Problem Find a formula for

n
X
(2i − 1) = 1 + 3 + · · · + (2n − 1).
i=1

Solution This is quite easy once it is discovered that

2i − 1 = i2 − (i − 1)2 .

This corresponds to F (i) = (i − 1)2 . It follows that the given sum telescopes to

F (n + 1) − F (1) = n2 − 02 = n2 . 

As a general principle, if ai is a polynomial function of i of degree m, try F (i) to be a

polynomial function of i of degree m + 1. If ai is of the form p(i)ci , where p is a polynomial of
degree m and c is a constant, try F (i) = q(i)ci where q is a polynomial of degree at least m.
If ai involves binomial coefficients or factorials, try F (i) to be something similar.
 Number theory
2
The great mathematician Carl Friedrich Gauss once stated that ‘Mathematics is the queen of
the sciences and number theory is the queen of mathematics’. You get the feeling that he
considered number theory to be a pretty important area. Indeed, the study of such simple
objects as the integers leads to incredibly deep problems for which a whole assortment of
ideas and techniques have been developed. This chapter comprises a collection of some of the
more simple, though still very useful, ideas and techniques that are used in number theory.

2.0 Problems
1. For any right-angled triangle whose side lengths are positive integers, we define the
tri-product of the triangle to be the product of its three side lengths.
Find with proof the greatest common divisor of all tri-products of all such right-angled
triangles.

2. Find all solutions in positive integers to the equation

m! + 5 = n3 .

3. The numbers from 1 to 81 are written in a 9 × 9 square array.

Prove that there exists a positive integer k such that the product of the numbers in row
k does not equal the product of the numbers in column k.

4. Find all positive integers p such that

p, 4p2 + 1 and 6p2 + 1

are all primes.

5. Let A be the sum of the digits of 44444444 , written in decimal notation.

If B is the sum of the digits of A, find the sum of the digits of B.

6. Let n be a positive integer such that 2n − 1 is a prime number.

Prove that n is a prime number.
20 2 Number theory

7. Let n be a positive integer such that 2n + 1 is a prime number.

Prove that n = 2k for some integer k.
8. Find all 5-digit natural numbers such that after deleting any one digit, the remaining
number is a 4-digit number which is divisible by 7.
(Note that leading zeros do not count as digits, e.g. 00123 counts as a 3-digit number.)
9. Show that there do not exist prime numbers p and q which differ by 2 and have the
property that
pq + 10k
is also a prime number for some non-negative integer k.
10. Determine all positive integers x1 > x2 > x3 > x4 > x5 such that
 2  2  2  2
x1 + x2 x2 + x3 x3 + x4 x4 + x5
+ + + = 38.
3 3 3 3

(Note that bxc is the largest integer not exceeding x.)

11. Let ϕ be the Euler phi function.1
(a) Prove that ϕ(n) is odd if and only if n = 1 or n = 2.
(b) Find all solutions to the equation ϕ(n) = 4.
12. The numbers 1!, 2!, 3!, . . . , 100! are written on a blackboard.
Is it possible to erase one of the numbers so that the product of the remaining 99
numbers is a perfect square?
13. Can you find a set of 2000 distinct positive integers such that the sum of the members
of every subset is not a perfect square?
14. For each positive integer k let
k
ak = 22 + 1.
(a) Prove that if i 6= j, then gcd(ai , aj ) = 1.
(b) Use the result of part (a) to prove that there are infinitely many primes.
15. Show that for every positive integer n,

121n − 25n + 1900n − (−4)n

is divisible by 2000.
16. Show that there are infinitely many positive integers m for which

18m + 45m + 50m + 125m

is divisible by 2006.
17. Let n be a given positive integer. Prove that the sequence
a aa
a, aa , aa , aa , . . .

is eventually constant modulo n for any positive integer a.

1 The Euler phi function is defined by the property that ϕ(n) is equal to the number of integers from the set
{1, 2, . . . , n} which have greatest common divisor 1 with n.
2.0 Problems 21

18. Let V = {1, 2, . . . , 24, 25}. Alexander is looking for a subset W of V with the property
that no two different members of W have a product which is a perfect square.
(a) Find the maximal possible size for W .
(b) Find the number of such sets W with this maximal size.
19. Prove that for each positive integer n, there exists a unique number divisible by 2n
whose decimal representation consists of n digits, each of them equal to 1 or 2.
20. Prove the following useful lemma. Suppose that positive integers a, b, x, y satisfy

ax = by .

Prove there exist positive integers r, m, n such that

a = rm and b = rn .

(Note that in particular this implies that either a | b or b | a.)

21. Call a set of positive integers funky if every pair of elements has greatest common divisor
not equal to 1, while every triple of elements has greatest common divisor equal to 1.
Show that there does not exist a funky set with infinitely many elements.
22. Prove that the sequence a1 , a2 , a3 , . . . defined by
√
 
1
an = n + n +
2
contains every positive integer exactly once apart from the perfect squares.
k
23. Prove that every prime factor of 22 + 1 is of the form 2k+1 x + 1, for some positive
integer x, and hence prove that there are infinitely many primes of the form 2w x + 1
for each positive integer w.
24. Let n be a positive integer such that 4n + 2n + 1 is a prime number.
Prove that n = 3k for some integer k.
25. The squares of an infinite grid are numbered as illustrated. The number 0 is placed in
the top-left corner. Each remaining square is numbered with the smallest non-negative
integer that does not already appear to the left of it in the same row or above it in the
same column.
Which number will appear in the 1003rd row and 1980th column?

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

26. Determine the value of the following sum, where p and q are relatively prime positive
integers.      
p 2p (q − 1)p
+ + ··· +
q q q
22 2 Number theory

27. For any positive integer n, let d(n) denote the number of positive divisors of n.
For which n does the sequence

n, d(n), d(d(n)), . . .

not contain any perfect squares?

28. Let S = {2!, 3!, 4!, . . .}. Some members of S may be written as a product of smaller
members of S, such as 4! = 3! × 2! × 2!. Let’s call such a number a factorial-composite.
If a member of S cannot be written as a product of smaller members of S, such as 3!,
let’s call such a number a factorial-prime.

(a) Prove that S contains infinitely many factorial-composites.

(b) Prove that S contains infinitely many factorial-primes.
(c) Prove that S contains infinitely many factorial-composites which have at least two
different ways of factoring them into factorial-primes.2

29. Prove that the sequence

j√ k j√ k j√ k
2 + 1 , ( 2 + 1)2 , ( 2 + 1)3 , . . .

alternates between even and odd integers.

30. If a is an integer greater than 1, prove that

gcd(am − 1, an − 1) = agcd(m,n) − 1

for all positive integers m and n.

31. Show that for every positive integer n, there are infinitely many terms of the Fibonacci
sequence which are divisible by n.

32. The geometric mean of m non-negative real numbers is the mth root of their product.

(a) Prove that, for every positive integer n, there exists a set of n distinct positive
integers such that the geometric mean of every subset is an integer.
(b) Does there exist an infinite set of distinct positive integers such that the geometric
mean of every finite subset is an integer?

33. (a) For an irrational number x, consider the sequence {x}, {2x}, {3x}, . . . , where as
usual {x} = x − bxc denotes the fractional part of x.
For every positive integer n, show that there exists a term of the sequence which
lies between 0 and n1 .
(b) Hence, for any real numbers a and b satisfying 0 ≤ a < b ≤ 1, prove that there
exists a term of this sequence which lies between a and b.
(c) Use this fact to show that there exists a power of two whose decimal representation
starts with the digits 123456789.

34. Find all powers of two with the property that, after deleting the first digit of its decimal
representation, one again obtains a power of two.
2 Thus the analogue of the fundamental theorem of arithmetic (factorisation into primes is unique up to the
order of factors) is not true in this setting.
2.0 Problems 23

35. Find all solutions in positive integers to the equation

2
ab = ba .

36. Determine all positive integers which are relatively prime to all terms of the infinite
sequence
an = 2n + 3n + 6n − 1, n ≥ 1.

37. (a) Prove the following useful lemma. Let p be an odd prime and a a positive integer
such that p | a − 1.
If pα k a − 1, prove that

pα+β k an − 1 if and only if pβ k n.

(The notation pj k k means that pj | k but pj+1 - k.)

(b) What is the corresponding result if p = 2?
38. A positive integer n is called groovy if, for every positive integer a, n2 divides an − 1
whenever n divides an − 1.
Show that there are infinitely many composite numbers which are groovy.
39. Find all Carmichael numbers3 of the form 3pq, where 3, p and q are distinct primes.
40. Find all pairs of positive integers (m, n) such that
n n
·2
n | 1 + m3 + m3 .

41. Find all positive integers n such that

2n
 

n
is a power of 2.
42. Prove there are infinitely many positive integers n which can be expressed as the sum
of the squares of k positive integers for every integer 1 ≤ k ≤ n − 14.
43. Show that for every positive integer n there exist n distinct positive integers such that
their sum is a perfect 2009th power, and their product is a perfect 2010th power.
44. Let b, n > 1 be integers. Suppose that for each k > 1 there exists an integer ak such
that
k | b − ank .

Prove that b = An for some integer A.

45. A wobbly number is a positive integer whose digits are alternately non-zero and zero
with the units digit non-zero.
Determine all positive integers which do not divide any wobbly number.
46. Prove that, for every positive integer n, there exists a positive integer X such that

X, 2X, 3X, . . . , nX

are all nontrivial perfect powers.

3 See the problem in section 2.14 for the definition of a Carmichael number.
24 2 Number theory

2.1 Fundamental theorem of arithmetic

Let’s start with one of the most basic, though one of the most important, results in number
theory.

Fundamental theorem of arithmetic Every positive integer has a unique prime factori-
sation.
Perhaps the most important word appearing in the above statement is the word unique. It
tells us that a prime factorisation acts like a fingerprint. In other words, it provides a way
to identify positive integers. And it’s quite a useful one, because if you’re given numbers in
prime factorised form, then it’s a simple matter to test for divisibility, count the number of
divisors, calculate greatest common divisors, determine lowest common multiples, and so on.

Problem Call a set of positive integers funky if every pair of elements has greatest common
divisor not equal to 1, while every triple of elements has greatest common divisor equal to 1.
Show that there exists a funky set with n elements for each positive integer n.

Solution The easiest way to show the existence of a funky set with n elements is simply to
construct a funky set with n elements. This is exactly what we are going to do, and our final
set of numbers will be called {a1 , a2 , . . . , an }. Since the problem involves greatest common
divisors, it makes sense to consider prime factorisations. In fact, let’s create a table with n
rows, one for each of the numbers a1 , a2 , . . . , an , and infinitely many columns, one for each of
the primes p1 , p2 , p3 , . . .. We could then place a tick in the row corresponding to a and the
column corresponding to p if and only if p appears in the prime factorisation of a. Your job
now, if you choose to accept it, is to place ticks in the table in order to guarantee that we
end up with a funky set.
For each pair of elements to have greatest common divisor larger than 1 simply means that
for 1 ≤ i < j ≤ n, there exists a column which contains ticks in the rows corresponding to ai
and aj . For each triple of elements to have greatest common divisor equal to 1 simply means
that no column can contain three or more ticks. However, this is easy to do! For each pair
1 ≤ i < j ≤ n, just find a column—any single one of the infinitely many columns—in which to
put exactly two ticks in the rows corresponding to ai and aj . For example, the construction
for n = 5 looks something like the following.

p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 ···
a1 X X X X
a2 X X X X
a3 X X X X
a4 X X X X
a5 X X X X

This would give us the funky set consisting of a1 = p1 p2 p3 p4 , a2 = p1 p5 p6 p7 , a3 = p2 p5 p8 p9 ,

a4 = p3 p6 p8 p10 and a5 = p4 p7 p9 p10 . Since this construction can be carried out for any number
of rows, the problem is solved.

It’s worth noting that we could have recorded in the table the highest power of p which
divides a, although this particular problem didn’t require us to do so. Another lesson we
can learn from this problem is that thinking in terms of tables can often help to simplify a
problem immensely.
2.2 Pigeonhole principle 25

2.2 Pigeonhole principle

You just can’t get enough of the pigeonhole principle! It was introduced in chapter 1 and
will rear its head again in chapter 13. For now, we’ll use it to attack an interesting number
theory problem.

Problem Show that for every positive integer n not divisible by 2 or 5, there exists a
multiple of n all of whose digits are ones.

Solution Our pigeons will be the first n + 1 numbers from the sequence 1, 11, 111, 1111, . . . ,
and our pigeonholes will be the n possible remainders modulo n. So the pigeonhole principle
asserts that two of these numbers must be congruent to each other modulo n. Suppose that
they consist of i ones and j ones, where we can assume without loss of generality that i > j.
Then
111
| {z. . . 1} − 111
| {z. . . 1} = |111{z
. . . 1} 000
| {z. . . 0} ≡ 0 (mod n).
i j i−j j

However, since n is not divisible by 2 or 5, we are allowed to divide both sides of this
congruence equation by 10j . This leaves us with

111
| {z. . . 1} ≡ 0 (mod n),
i−j

so there certainly does exist a multiple of n all of whose digits are ones.

2.3 Dealing with digits

It’s reasonably common to write an an−1 . . . a0 for the number whose decimal digits, from left
to right, are an , an−1 , . . . , a0 . One of the first tricks in dealing with digits is to express this
number as
an an−1 . . . a0 = an × 10n + an−1 × 10n−1 + · · · + a0 × 100 .
Another trick is to use the following facts from modular arithmetic. They’re both easy to
prove and provide simple tests for divisibility by 9 and 11.

an an−1 . . . a1 a0 ≡ a0 + a1 + a2 + · · · + an (mod 9)
n
an an−1 . . . a1 a0 ≡ a0 − a1 + a2 − · · · + (−1) an (mod 11)

Problem Find all integers n > 1 for which there exist distinct positive integers a and b such
that na + 1 can be obtained from nb + 1 by reversing the order of its decimal digits, and vice
versa.

Solution Let’s assume without loss of generality that a < b. The first observation that we
can make about the two numbers na + 1 and nb + 1 is that they must have the same number
of digits. It seems that we should be able to use this fact to show that n can’t be all that
large. Indeed, if we consider that the larger of these two numbers is less than 10 times the
smaller, we obtain the inequality

nb + 1 < 10(na + 1)
⇒ na (nb−a − 10) < 9.
26 2 Number theory

By inspection, the latter inequality simply cannot hold if n ≥ 11. Furthermore, if n = 10,
then na + 1 takes the form 100 . . . 01, so reversing the order of its digits returns exactly the
same number.
Thus, we are left to consider the case 2 ≤ n ≤ 9. Since na + 1 and nb + 1 have exactly the
same digits, they must leave the same remainders modulo 9. Therefore,

na + 1 ≡ nb + 1 (mod 9)
a b−a
⇒ n (n − 1) ≡ 0 (mod 9).

At this stage, it makes sense to divide the problem into cases.

Case 1: Suppose that n is not divisible by 3.

Then it must be the case that nb−a − 1 ≡ 0 (mod 9) or equivalently, nb−a ≡ 1 (mod 9).
Earlier we found that na (nb−a − 10) < 9, from which it follows that nb−a − 10 < n9a < 92 .
Therefore, nb−a < 14 12 . Since nb−a ≡ 1 (mod 9) we deduce that nb−a = 10. This is
clearly impossible, since 10 is certainly not the power of any integer from 2 to 9.

Case 2: Suppose that n is equal to 3.

This provides us with one solution to the problem since 33 + 1 = 28 and 34 + 1 = 82.

Case 3: Suppose that n is equal to 6.

Note that 6a + 1 and 6b + 1 both end in the digit 7, so they must both also start with
the digit 7. However, it’s easy to show that 6b + 1 is at least five times larger than
6a + 1, which then prevents them from having the same number of digits.

Case 4: Suppose that n is equal to 9.

Note that 9a + 1 and 9b + 1 must end in the digit 0 or 2, so they must both also start
with the digit 0 or 2. Since the former doesn’t yield any solutions, we may assume that
both 9a + 1 and 9b + 1 start and end in the digit 2. However, it’s easy to show that
9b + 1 is at least eight times larger than 9a + 1, which then prevents them from having
the same number of digits.

In conclusion, the one and only integer which satisfies the conditions of the problem is 3.

2.4 Floor function

Recall that bxc = n simply means that n ≤ x < n + 1, where n is an integer. So whenever you
are confronted with the floor function, you can immediately write down inequalities which
capture the same information. The floor function pops up in all sorts of places, including the
following amazing result known as Beatty’s theorem.

Problem Let α and β be positive irrational numbers satisfying

1 1
+ = 1.
α β

Prove that the sequences

bαc, b2αc, b3αc, . . . and bβc, b2βc, b3βc, . . .

2.5 Square roots and conjugates 27

together include every positive integer exactly once.4

Solution First, we note that α > 1 and β > 1, which implies that the two sequences must
be strictly increasing. Next, we’ll show that no integer can occur in both sequences. To do
this, suppose that biαc = bjβc = n. Since α and β are irrational, it cannot be the case that
iα = n or jβ = n, so we have the strict inequalities
n < iα < n + 1 and n < jβ < n + 1.
Rearranging these inequalities yields
n n+1 n n+1
<i< and <j< ,
α α β β
which add to give
n n n+1 n+1
n=
+ <i+j < + = n + 1.
α β α β
However, this is a contradiction because i + j is an integer, so it can’t lie between the
consecutive integers n and n + 1. Therefore, no integer can occur in both sequences.
Finally, we’ll show that every integer occurs in one of the sequences. To do this, suppose that
the integer n doesn’t occur in either sequence. Then there must exist integers i and j such
that
iα < n < n + 1 < (i + 1)α and jβ < n < n + 1 < (j + 1)β.
Rearranging these inequalities yields
n n+1 n n+1
i< , i+1> , j< and j+1> .
α α β β
These add to give
n n n+1 n+1
i+j < + =n and i+j+2> + = n + 1.
α β α β
However, this is a contradiction because it implies that the integer i + j must lie between
the consecutive integers n − 1 and n. Therefore, no integer fails to occur in both sequences.
Piecing all the information together, we find that the sequences together include every positive
integer exactly once.

2.5 Square roots and conjugates

√
In this section, we consider numbers of the form a + b d, where a and b are rational numbers,
√
while d is an integer which √ is not a perfect
√ square. You should be able to prove that d is
irrational and that a1√+ b1 d = a2 + b2 d implies that a1 = a2 and b1 = b2 . If you find
the number r = a +√b d appearing in a problem, it’s almost always useful to consider its
conjugate r = a − b d. The beauty of conjugates is that their sum, difference and product
are all nice expressions.
√ √
(a + b d) + (a − b d) = 2a
√ √ √
(a + b d) − (a − b d) = 2b d
√ √
(a + b d) × (a − b d) = a2 − db2 .
4 The converse is also true! If the two sequences bαc, b2αc, b3αc, . . . and bβc, b2βc, b3βc, . . . together include
1
every positive integer exactly once, then α and β are positive irrational numbers satisfying α + β1 = 1. See if
you can prove it!
28 2 Number theory

Conjugates also behave nicely under addition, subtraction,

√ multiplication and division. More
explicitly, if r1 and r2 are numbers of the form a + b d, then we have the following facts.

r1 + r2 = r1 + r2 r1 − r2 = r1 − r2
r1 × r2 = r1 × r2 r1 ÷ r2 = r1 ÷ r2

One particularly useful corollary is the fact that

rn = rn ,

where on the left, the exponent appears inside the conjugate, while on the right, the exponent
appears outside the conjugate.

√
Problem Show
√ that
√ if we raise 2 − 1 to the power of a positive integer, then the result is
of the form m − m − 1 for some positive integer m.

√
Solution If we expand ( 2 − 1)n by using the binomial theorem5 , we obtain
√ √
( 2 − 1)n = a + b 2

for some integers a and b. Now consider the conjugate expression to obtain
√ √
(− 2 − 1)n = a − b 2.

Multiplying these two equations together gives

√ √ √ √
(a + b 2)(a − b 2) = ( 2 − 1)n (− 2 − 1)n
⇒ a2 − 2b2 = (−1)n .

So for n even, we have a2 − 2b2 = 1 and we can take m = a2 to obtain

√ √
( 2 − 1)n = a + b 2
√ √
= ± m ± m − 1.

√ √
However, since 2 − 1 is between 0 √ and 1, it follows
√ that
√ ( 2 − 1)n must also be between 0
and 1. So it must be the case that ( 2 − 1)n = m − m − 1.
The case for n odd can be handled in a similar way.

2.6 Powers of two

The numbers 1, 2, 4, 8, 16, . . . seem to appear in a seeming infinitude of number theory problems.
So in this section, let’s pay our respects to the powers of two by considering the following
problem.

n
Problem For each positive integer n, determine the remainder when 32 − 1 is divided
by 2n+3 .
5 See section 13.3 if you don’t know what this is.
2.7 Euclid’s algorithm 29

n
Solution Given an expression like 32 − 1, you can’t help but want to use the difference of
perfect squares factorisation, otherwise known by the acronym DOPS. This gives
n n−1 n−1
32 − 1 = (32 + 1)(32 − 1)

and once again, DOPS can be happily applied to the second factor. This gives
n n−1 n−2 n−2
32 − 1 = (32 + 1)(32 + 1)(32 − 1).

So if we continue with this veritable feast of DOPS, we will finally be left with
n n−1 n−2
32 − 1 = (32 + 1)(32 + 1) · · · (32 + 1)(31 + 1)(31 − 1).

We’d like to know the remainder after this number is divided by 2n+3 . But with so many
brackets, all being even, it seems that our number might actually be divisible by 2n+3 . Let’s
see if this is the case. A keen observation will reveal that for integers m,

32m + 1 ≡ (−1)2m + 1 ≡ 2 (mod 4).

So all of the brackets, apart from the final two, are divisible by 2, but not 4. Since there are
n − 1 such brackets, and the final two brackets contribute a factor of 8, our number must be
n
divisible by 2n−1 × 8 = 2n+2 , but not by 2n+3 . It follows that 32 − 1 leaves a remainder of
2n+2 when divided by 2n+3 .

2.7 Euclid’s algorithm

One of the gems of number theory is Euclid’s algorithm. Ostensibly, it provides a recipe for
calculating the greatest common divisor of two positive integers, but it is so much more than
that! The basis for it is the division algorithm, which we now state.

Division algorithm For any two integers a and b 6= 0, there is a unique way to write
a = qb + r, where 0 ≤ r < |b|.

A particular corollary of the division algorithm is that, once we write a = qb + r as specified,

then we have gcd(a, b) = gcd(b, r). Iterating this process until we are left with a greatest
common divisor which we can calculate by inspection is what Euclid’s algorithm is all about.
As with many algorithms, the best way to demonstrate Euclid’s algorithm is through an
example, which we do in the following problem.

Problem Find a number N which is divisible by 39 and such that N + 1 is divisible by 106.

Solution We’d like to solve the equations N = 39a and N + 1 = 106b for integers a and b.
However, these imply that 106b − 39a = 1, which is known as a linear Diophantine equation
and can be solved with the help of Euclid’s algorithm. In the left column below, we apply
Euclid’s algorithm, by repeatedly applying the division algorithm as shown.

106 = 2 × 39 + 28 1=1×6−1×5
39 = 1 × 28 + 11 = 2 × 6 − 1 × 11
28 = 2 × 11 + 6 = 2 × 28 − 5 × 11
11 = 1 × 6 + 5 = 7 × 28 − 5 × 39
6=1×5+1 = 7 × 106 − 19 × 39
30 2 Number theory

On the left, we have applied Euclid’s algorithm to the numbers 106 and 39, thereby showing
that gcd(106, 39) = 1. On the right, we have sneakily reversed Euclid’s algorithm. The last
line of Euclid’s algorithm allows us to write the number 1 as a multiple of 5 (namely −5),
plus a multiple of 6 (namely 6). We can express this idea by saying that 1 is a combination of
5 and 6. The second last line of Euclid’s algorithm allows us to exchange the number 5 for a
combination of 6 and 11. Therefore, we can write the number 1 as a combination of 6 and 11.
The third last line of Euclid’s algorithm allows us to exchange the number 6 for a combination
of 11 and 28. Therefore, we can write the number 1 as a combination of 11 and 28. We can
continue in this way until we have finally expressed the number 1 as a combination of 39 and
106, which is what we set out to do. Given that 7 × 106 − 19 × 39 = 1, it is clear that we
should take N = 19 × 39 = 741.

Next, we’ll look at a problem which mixes Euclid’s algorithm with some Fibonacci fun! The
result is a rather amazing fact concerning the greatest common divisor of two Fibonacci
numbers.

Problem Recall that the Fibonacci sequence is defined by

F0 = 0, F1 = 1 and Fm+1 = Fm + Fm−1 for m ≥ 1.

Prove that gcd(Fm , Fn ) = Fgcd(m,n) .

Solution Let’s break the problem into bite-sized pieces which we’ll finally put together to
produce a complete proof.

For every positive integer n, gcd(Fn , Fn−1 ) = 1.

We have gcd(Fn , Fn−1 ) = gcd(Fn−1 + Fn−2 , Fn−1 ) = gcd(Fn−1 , Fn−2 ), by the division
algorithm. Repeated use of this fact yields
gcd(Fn , Fn−1 ) = gcd(Fn−1 , Fn−2 ) = gcd(Fn−2 , Fn−3 ) = · · ·
· · · = gcd(F3 , F2 ) = gcd(F2 , F1 ) = gcd(F1 , F0 ) = 1.

For any positive integers m and n, Fm+n = Fm+1 Fn + Fm Fn−1 .

This statement will follow if we can prove by induction on k that
FN = Fk+1 FN −k + Fk FN −k−1
for all positive integers N and all k = 0, 1, 2, . . . , N . The statement is easily seen to
be true for k = 0 and the induction will be complete once we have verified that
Fk+1 FN −k + Fk FN −k−1 = Fk+2 FN −k−1 + Fk+1 FN −k−2 . However, this is equiva-
lent to Fk+1 (FN −k − FN −k−2 ) = FN −k−1 (Fk+2 − Fk ), which in turn is equivalent
to Fk+1 FN −k−1 = FN −k−1 Fk+1 .
If m | n, then Fm | Fn .
We will prove by induction on k that Fm | Fkm for every positive integer k. The
statement is easily seen to be true when k = 1. Furthermore, Fm | Fkm implies that
Fm | Fkm+1 Fm + Fkm Fm−1 = F(k+1)m , where we have used the previous result. This
completes the induction.
If n = qm + r, then gcd(Fm , Fn ) = gcd(Fm , Fr ).
We simply write Fn = Fqm+r = Fqm+1 Fr + Fqm Fr−1 , which implies that
gcd(Fm , Fn ) = gcd(Fm , Fqm+1 Fr + Fqm Fr−1 )
= gcd(Fm , Fqm+1 Fr )
= gcd(Fm , Fr ).
2.8 Integers base-n 31

To pass from the first line to the second line, we simply use the division algorithm,
while to pass from the second line to the third line, we use the fact that Fm | Fqm and
gcd(Fqm+1 , Fqm ) = 1.

If we write n = qm + r, then the statement gcd(Fm , Fn ) = gcd(Fm , Fr ) is remarkably similar

to the statement gcd(m, n) = gcd(m, r), which forms the basis of Euclid’s algorithm. In fact,
the only difference is that the letter F has been added into the picture. Therefore, applying
Euclid’s algorithm to two Fibonacci numbers gives the result gcd(Fm , Fn ) = Fgcd(m,n) .

If you’re not quite sure about how the proof works, then the following example may be
enlightening.

gcd(106, 39) gcd(F106 , F39 )

106 = 2 × 39 + 28 = gcd(39, 28) = gcd(F39 , F28 )
39 = 1 × 28 + 11 = gcd(28, 11) = gcd(F28 , F11 )
28 = 2 × 11 + 6 = gcd(11, 6) = gcd(F11 , F6 )
11 = 1 × 6 + 5 = gcd(6, 5) = gcd(F6 , F5 )
6=1×5+1 = gcd(5, 1) = gcd(F5 , F1 )
5=5×1+0 = gcd(1, 0) = gcd(F1 , F0 )

2.8 Integers base-n

Just because most of us have 10 digits on our hands, why should we represent positive integers
using 10 digits? Indeed in computer science, it’s useful to work in base-2, also known as
binary, with only two symbols. Some computer science settings work in base-16, also known
as hexadecimal, with 16 symbols. And, in mathematics, it sometimes pays to work in base-n
for other positive integers n.

Problem Is it possible to choose 2000 distinct non-negative integers, all less than 100000,
no three of which are consecutive terms of an arithmetic progression?

Solution Since we want to keep our numbers small, let’s use the following greedy algorithm.6
Start with the number a0 = 0 and let an+1 be the next largest integer which doesn’t form
a three term arithmetic progression with any of a0 , a1 , a2 , . . . , an . So it’s clear that we can
take a1 = 1, but then we have to skip 2 to avoid the arithmetic progression (0, 1, 2). However,
we can take a2 = 3 and continue in this fashion to obtain an increasing sequence of positive
integers. Before you read on, try to write down the first 20 terms of the sequence and see if
you can spot a pattern.
If you can’t spot the pattern, then at least you should have noticed that there are large jumps
at certain parts of the sequence. The first jump is at a2 = 3 and the next largest jump occurs
at a4 = 9. There are two more large jumps at a8 = 27 and a16 = 81. The numbers 3, 9, 27
and 81, are all powers of 3, which suggests that something interesting might be happening in
base-3. So let’s write out a table which displays the numbers of our sequence in base-3.
6 Informally,
a greedy algorithm makes an optimal choice at each step. Such a tactic is not generally guaranteed
to produce an optimal set overall. However, in many questions, including this one, it is sufficient to solve the
problem.
32 2 Number theory

n 0 1 2 3 4 5 6 7 8
an 0 1 3 4 9 10 12 13 27
base-3 0 1 10 11 100 101 110 111 1000

n 9 10 11 12 13 14 15 16 17
an 28 30 31 36 37 39 40 81 82
base-3 1001 1010 1011 1100 1101 1110 1111 10000 10001

Rather amazingly, the base-3 representation of an only contains the digits 0 and 1, just like
numbers written in binary. And even more amazingly, it seems that the ternary representation
of an is simply the binary representation of n. Of course, everything we’ve accomplished so
far has merely been pattern spotting, so let’s try to prove the following conjecture.

Let an be the positive integer whose base-3 representation is the binary represen-
tation of n. Then no three terms of the sequence a0 , a1 , a2 , . . . form an arithmetic
progression.

In order to prove this, we assume that the numbers x, y, z are distinct terms of the sequence
which form an arithmetic progression, that is, x + z = 2y. So x and z both have ternary7
representations containing only the digits 0 and 1, while 2y has a ternary representation
containing only the digits 0 and 2. It’s clear that if we add x and z in ternary, then there can
be no carries. So in order for their sum to contain only the digits 0 and 2, each occurrence of
the digit 1 in x must match up with a corresponding digit 1 in z and vice versa. In other
words, x and z must actually be the same number, contradicting the fact that x, y, z are
distinct. Therefore, we can be sure that no three terms of the sequence a0 , a1 , a2 , . . . form an
arithmetic progression.
Now it just suffices to show that the 2000 numbers a0 , a1 , a2 , . . . , a1999 of our sequence are all
less than 100000. Since the sequence is increasing, we need only consider the number a1999 .
However, 1999 in binary is 11111001111 and 11111001111 in ternary is only 88249, much less
than 100000.

2.9 Construction problems

A number theory problem might ask you to come up with a construction which satisfies
certain conditions. Such problems can sometimes require a great deal of ingenuity, but will
certainly be easier to solve if you are guided by intuition and experience.

Problem Does there exist an infinite increasing sequence t1 < t2 < t3 < · · · of positive
integers such that, for any integer c, the sequence

t1 + c, t2 + c, t3 + c, . . .

contains only a finite number of primes?

Solution Consider a particular value of c, such as 73. How can we easily guarantee that the
sequence t1 + 73, t2 + 73, t3 + 73, . . . contains only a finite number of primes? One way is to
make sure that, eventually, the numbers in the sequence t1 , t2 , t3 , . . . are divisible by 73. In
fact, we would like this to be true for almost any value of c, which means that the numbers
7 This is another term for base-3.
2.9 Construction problems 33

t1 , t2 , t3 , . . . should probably have lots of factors. A good candidate for such a sequence is
tn = n!. In fact, for any c with |c| ≥ 2, the number n! + c is divisible by |c| and larger than it
whenever n > |c|. So it’s certainly true that there are finitely many primes in the sequence
1! + c, 2! + c, 3! + c, . . . as long as |c| ≥ 2. Furthermore, when c = 0, there are clearly no primes
in the sequence apart from 2.
So we’re left to deal with the cases c = −1 and c = +1. Unfortunately, there is no easy way
to see that n! ± 1 is composite, so our initial choice of sequence may have to be tweaked. We
would like to change the sequence somehow, maintaining the abundance of factors in our
numbers, but using numbers which are obviously composite when 1 is added or subtracted.
So what sort of numbers are obviously composite when 1 is added or subtracted? One answer
to this question is perfect cubes, since we have the factorisations
X 3 − 1 = (X − 1)(X 2 + X + 1) and X 3 + 1 = (X + 1)(X 2 − X + 1).

So the sequence tn = (n!)3 certainly satisfies the conditions of the question.

Sometimes, you may be asked to perform a particular construction for all positive integers n.
In that case, it should be easier to find a construction in the case n + 1 if you use the fact
that you have already found a construction in the case n. This means that your construction
will involve induction, quite a common phenomenon in number theory problems.

Problem Let n be a positive integer.

Prove that there are infinitely many perfect squares of the form 2n a − 7, where a is a positive
integer.

Solution In the language of modular arithmetic, the problem is asking us to prove that
there are infinitely many integers m such that m2 ≡ −7 (mod 2n ). Our first observation is
that once you’ve found one such m, then infinitely many such m are easy to find. This is
because
m2 ≡ −7 (mod 2n ) ⇒ (m + 2n )2 ≡ −7 (mod 2n ).
So the problem is really asking us to prove that there exists at least one integer m such that
m2 ≡ −7 (mod 2n ). For the first several values of n, we can simply list such values of m.
12 ≡ −7 (mod 2) 12 ≡ −7 (mod 4) 12 ≡ −7 (mod 8)
2 2 2
3 ≡ −7 (mod 16) 5 ≡ −7 (mod 32) 11 ≡ −7 (mod 64)

The idea behind our construction is as follows: if m2 ≡ −7 (mod 2n ), then it must be true
that either
m2 ≡ −7 (mod 2n+1 ) or m2 ≡ −7 + 2n (mod 2n+1 ).
In the former case, we can simply use m again. But in the latter case, we can use m + 2n−1
and in this case we have
(m + 2n−1 )2 ≡ m2 + m2n + 22n−2 (mod 2n+1 )
≡ −7 + 2n + m2n + 22n−2 (mod 2n+1 )
≡ −7 + 2n (m + 1 + 2n−2 ) (mod 2n+1 )
≡ −7 (mod 2n+1 ).

This is certainly true for n ≥ 3, since we know that m is necessarily odd, which forces
(m + 1 + 2n−2 ) to be even. So given the fact that m2 ≡ −7 (mod 2n ), we’ve shown that
either m2 ≡ −7 (mod 2n+1 ) or (m + 2n−1 )2 ≡ −7 (mod 2n+1 ). Therefore, by induction, for
every positive integer n, we have a value of m for which m2 ≡ −7 (mod 2n ).
34 2 Number theory

2.10 Modular arithmetic

Many questions in number theory involve modular arithmetic in one way or another. One
of the advantages of modular arithmetic is that we can add, subtract and multiply residues.
Better still, if we are working modulo n, then we can also divide residues, as long as they are
relatively prime to n. The explicit statement of this result follows.

Division modulo n For an integer a satisfying gcd(a, n) = 1,

ax ≡ ay (mod n) if and only if x ≡ y (mod n).

For an integer a satisfying gcd(a, n) = d > 1,

a a  n
ax ≡ ay (mod n) if and only if x≡ y mod .
d d d

2p


Problem If p is a prime, show that p − 2 is a multiple of p.

Solution We start by writing

 
2p (2p)(2p − 1)(2p − 2) · · · (p + 1) (2p − 1)(2p − 2) · · · (p + 1)
= =2 .
p p(p − 1)(p − 2) · · · 1 (p − 1)(p − 2) · · · 1

So what we are required to prove is

(2p − 1)(2p − 2) · · · (p + 1)
2 ≡2 (mod p).
(p − 1)(p − 2) · · · 1

If p = 2, then the statement is obvious, since both sides are congruent to 0. On the other
hand, for p > 2 we can divide both sides by 2 and then multiply by the denominator to obtain

(2p − 1)(2p − 2) · · · (p + 1) ≡ (p − 1)(p − 2) · · · 1 (mod p).

This statement is equivalent to the previous one precisely because we are working modulo a
prime, so we may divide by any number which is not 0 modulo p. The last equation is clearly
true, because

2p − 1 ≡ p − 1, 2p − 2 ≡ p − 2, ..., and p + 1 ≡ 1 (mod p).

2p


In fact, more is true. If p > 3 is prime, then p − 2 is a multiple of p3 . See if you can prove
it!

2.11 Chinese remainder theorem

Suppose we are looking for solutions to the simultaneous congruences

x≡1 (mod 4), x≡2 (mod 3), x≡3 (mod 5).

You can verify that x ≡ 53 (mod 60) is a solution. In fact it represents all of the solutions.
This is a special case of the following.
2.12 From Fermat to Euler 35

Chinese remainder theorem Consider the n congruences

x ≡ a1 (mod m1 ), x ≡ a2 (mod m2 ), ..., x ≡ an (mod mn ).

A solution for x definitely exists if m1 , m2 , . . . , mn are pairwise relatively prime. Furthermore,

the solution is unique modulo m1 m2 · · · mn .

This theorem is particularly useful if you wish to show the existence of numbers satisfying
certain congruence conditions.

Problem Prove that for each positive integer n, there exist n consecutive positive integers,
none of which is an integral power of a prime number.

Solution Consider the consecutive integers x + 1, x + 2, . . . , x + n. To guarantee that x + 1

is not an integral power of a prime, we simply need to make sure that it is divisible by two
distinct primes, say p1 and q1 . Similarly, we would like x + 2 to be divisible by two distinct
primes p2 and q2 and, in general, x + k should be divisible by two distinct primes pk and qk .
So what we have to do now is show the existence of a positive integer x which satisfies the
following n simultaneous congruences.
x+1≡0 (mod p1 q1 )
x+2≡0 (mod p2 q2 )
..
.
x+n≡0 (mod pn qn )

And this is where the Chinese remainder theorem comes to our rescue! This is because
it guarantees that there exists a solution as long as the numbers p1 q1 , p2 q2 , . . . , pn qn are
relatively prime. Therefore, if we simply pick the primes p1 , p2 , . . . , pn , q1 , q2 , . . . , qn to be
distinct, then we can invoke the Chinese remainder theorem to find a solution for x modulo m,
where m = p1 q1 p2 q2 · · · pn qn .

Note that x is a solution to the problem if and only if x + m is. This shows that for a given
n there are infinitely many solutions.

2.12 From Fermat to Euler

In this section, we consider results which involve perfect powers in modular arithmetic. The
first is due to Fermat and, to distinguish it from his many others, is referred to as his little
theorem. However, this is by no means an accurate indication of its importance!

Fermat’s little theorem If p is a prime and gcd(a, p) = 1, then

ap−1 ≡ 1 (mod p).

A convenient equivalent form of Fermat’s little theorem is: If p is a prime and a is any
positive integer, then
ap ≡ a (mod p).

Problem A positive integer n is called groovy if, for every positive integer a, n2 divides
an − 1 whenever n divides an − 1.
Show that all primes are groovy.
36 2 Number theory

Solution Let p be a prime and suppose that p divides ap − 1. So we have

ap ≡ 1 (mod p),

whereas Fermat’s little theorem asserts that

ap ≡ a (mod p).

Putting together these two pieces of information, we find that

a≡1 (mod p)

and we can write a = mp + 1 for some integer m. Therefore, by the binomial theorem, we
have
p  
X p
ap − 1 = (mp + 1)p − 1 = mk pk − 1.
k
k=0
2
If we consider this expression modulo p , then each term with k ≥ 2 disappears, due to the
appearance of pk . Therefore, we conclude that
   
p p
ap − 1 ≡ m0 p0 + m1 p1 − 1 ≡ 1 + mp2 − 1 ≡ 0 (mod p2 ),
0 1

and that every prime is indeed groovy.

Sometimes you’re not working modulo a prime, in which case Fermat’s little theorem just
isn’t good enough. Luckily Euler’s theorem, which generalises Fermat’s little theorem, comes
to the rescue!
Euler’s theorem If n is a positive integer and gcd(a, n) = 1, then

aϕ(n) ≡ 1 (mod n).

Here, ϕ(n) denotes the number of integers 1 ≤ k ≤ n such that gcd(k, n) = 1 and is usually
referred to as the Euler phi function.8 For example, ϕ(12) = 4, since the four numbers 1, 5, 7
and 11 are relatively prime to 12. We’ll use Euler’s theorem to provide a quick proof to a
problem we previously encountered in section 2.2.

Problem Show that for every positive integer n not divisible by 2 or 5, there exists a
multiple of n all of whose digits are ones.

Solution Consider the number X = 19 (10ϕ(9n) − 1). It’s an integer whose decimal repre-
sentation consists entirely of ones, specifically ϕ(9n) ones, to be precise. However, since
gcd(10, 9n) = gcd(10, n) = 1 by assumption, we can invoke Euler’s theorem to show that

1
10ϕ(9n) ≡ 1 (mod 9n) ⇒ (10ϕ(9n) − 1) ≡ 0 (mod n).
9
Therefore, X is divisible by n, as desired.

Fermat’s little theorem and Euler’s theorem are most successfully applied in conjunction with
the following useful result, which is worth a section on its own.
8 There is a formula for ϕ(n) in terms of the prime factorisation of n. See if you can find and prove it!
2.13 The gcd trick 37

2.13 The gcd trick

The gcd trick If ax ≡ 1 (mod n) and ay ≡ 1 (mod n), then

agcd(x,y) ≡ 1 (mod n).

This useful result can be proven using the representation of gcd(x, y) as a linear combination
of x and y using Euclid’s algorithm. See section 2.7. The proof is not hard and you should
try to work it out for yourself.

Problem Prove that if p and q are primes satisfying q | 2p − 1, then p < q.

Solution We can write q | 2p − 1 equivalently as 2p ≡ 1 (mod q). But Fermat’s little

theorem tells us that 2q−1 ≡ 1 (mod q). So we can apply the gcd trick to obtain

2gcd(p,q−1) ≡ 1 (mod q).

Since p is prime, we know that gcd(p, q − 1) must be equal to 1 or p. The former case is
absurd, since it would imply that 1 ≡ 2gcd(p,q−1) ≡ 2 (mod q).
We are left to ponder the latter case, when gcd(p, q − 1) = p. But this simply means that
p | q − 1, which is only possible if p < q.

The previous problem provides us with a beautiful proof that there are infinitely many primes.
This is because if there were only finitely many, we could take p to be the largest one. Then
2p − 1 would be divisible by some prime q, which would have to be larger than p. Since this
contradicts the fact that p was the largest prime, we conclude that there must be infinitely
many primes.

2.14 Existence of a generator

Suppose we consider a modulus, say modulo 7, and a number, say 2, and compute powers
of 2 modulo 7,
20 , 21 , 22 , 23 , . . ..
We end up in a cycle
1, 2, 4, 1, 2, 4, . . .
of three elements.
Yet if we chose to compute powers of 3 modulo 7 instead, we would end up with

1, 3, 2, 6, 4, 5, 1, 3, 2, 6, 4, 5, . . . ,

which is a 6-cycle.
The number 3 is called a generator modulo 7 because its successive powers generate everything
which is relatively prime to 7 modulo 7.
More generally, a number a is called a generator9 modulo n if the powers of a, namely,

1, a, a2 , a3 , . . . ,
9 The term primitive root modulo n is used in some literature. It means the same thing.
38 2 Number theory

generate the complete set of elements modulo n which are relatively prime to n.

Existence of a generator There exists a generator modulo n if and only if n = 1, 2, 4,

pk , or 2pk , where p is an odd prime and k is a positive integer.

This theorem is most often used in the case where n is a prime. It can be used very powerfully
in conjunction with the gcd trick. The important point is that if g is a generator modulo n,
then d = ϕ(n) is the smallest positive integer for which g d ≡ 1 (mod n).

Problem A number n is called a Carmichael number10 if it is composite and

an ≡ a (mod n)

for all positive integers a.

Prove that there are no Carmichael numbers which are the product of two distinct primes.

Solution Suppose n = pq is a Carmichael number, where p and q are distinct primes. If a

is any number coprime to n, then we know apq−1 ≡ 1 (mod pq), from which it follows that

apq−1 ≡ 1 (mod p).

By Fermat we know that

ap−1 ≡ 1 (mod p).
Using the gcd trick we find that
ad ≡ 1 (mod p),
where d = gcd(pq − 1, p − 1) = gcd(q − 1, p − 1).
This is not particularly helpful until we realise that we may choose a = g to be a generator
modulo p. Since p − 1 is the smallest positive exponent e for which g e ≡ 1 (mod p), this
means that gcd(q − 1, p − 1) = p − 1. Thus p − 1 | q − 1.
A similar argument modulo q shows that q − 1 | p − 1. Thus p = q, in contradiction to
p 6= q.

For a long time it was suspected that there are infinitely many Carmichael numbers. A proof
of this fact was finally found in 1994.

10 These numbers are of interest because in the light of Fermat’s little theorem, they look prime, but are not!
 Diophantine equations
3
A Diophantine equation is an equation that asks for integer solutions. The equation is usually
posed as, or at least can be reduced to, a polynomial expression in several variables. For
example, a2 + b3 c = d4 is a Diophantine equation.

3.0 Problems
1
1. Show that for every integer n > 1 it is possible to write n as a sum of two reciprocals
of distinct positive integers.

2. Find all pairs of integers (x, y) such that

3x + 4y = 2xy.

3. Find all integer solutions to the equation

(x − y)2 = x + y.

4. Find all ordered triples of positive integers such that each of them is a factor of the
sum of the other two.

5. Show that if u2 + uv + v 2 is divisible by 9, where u and v are integers, then so is uv.

6. Find all integers x and y such that

x3 y + x + y = xy + 2xy 2 .

7. Let a and b be integers.

(a) Prove that 13 | 2a + 3b if and only if 13 | 3a − 2b.

(b) Prove that 13 | a2 + b2 if and only if 13 | 2a + 3b or 13 | 3a + 2b.

8. Suppose that u, v are integers such that uv is not divisible by 7.

(a) Prove that u2 + uv + 2v 2 is not divisible by 49.

(b) Prove that if u2 + uv + 2v 2 is divisible by 7, then so is u2 + 2uv − v 2 .
40 3 Diophantine equations

(c) Find all pairs of integers (u, v) such that u2 + 2uv − v 2 is divisible by 7 but
u2 + uv + 2v 2 is not divisible by 7.
9. Suppose that a and b are positive integers such that

112011 | a2 + b2 .

Prove that
112012 | ab.

10. Find all positive integers of the form

n2 − 29
k= ,
3n + 11
where n is also an integer.
11. Determine all positive integers k such that for all positive integers a and b the following
statement is true.

7a + 8b is divisible by k if and only if 8a + 7b is divisible by k.

12. Find all positive integers a, b and c such that

ab + bc + ca = 2 + abc.

What if we drop the restriction that a, b and c are all positive?

13. Can the product of four consecutive positive integers be a perfect square?
14. Show that
x4 + 131y 4 = 3z 4
has no solutions in positive integers.
15. Find all triples of positive integers (a, b, c) such that

a | bc − 1, b | ac − 1 and c | ab − 1.

16. Determine all pairs (x, y) of positive integers such that

y 2 (x − 1) = x5 − 1.

17. Find all integers a, b and c with 1 < a < b < c such that

(a − 1)(b − 1)(c − 1)

is a divisor of abc − 1.
18. Suppose that a, b, c and d are positive integers satisfying ab = cd.
Prove that neither a + b + c + d nor a2 + b2 + c2 + d2 can be prime numbers.
19. Find all integer solutions to the equation

8x3 − 4 = y(6x − y 2 ).

20. Can the product of five consecutive positive integers be a perfect square?
3.0 Problems 41

21. Prove that the equation

6(6a2 + 3b2 + c2 ) = 5n2
has no solutions in integers apart from a = b = c = n = 0.
22. Prove that if p is prime and x is an integer, then every factor of the expression

xp−1 + xp−2 + · · · + x + 1

is congruent to 0 or 1 modulo p.
23. Find all ordered triples of positive integers (a, b, c) such that for all positive integers t
(at + 1)(bt + 1)(ct + 1) − 1
lcm(at, bt, ct)
is also a positive integer.
24. Given any set A = {a1 , a2 , a3 , a4 } of four distinct positive integers, we denote the sum
a1 + a2 + a3 + a4 by sA . Let nA denote the number of pairs (i, j) with 1 ≤ i < j ≤ 4
for which ai + aj divides sA .
Find all sets A of four distinct positive integers which achieve the largest possible value
of nA .
25. Find all pairs (a, b) of positive integers such that
a2 b + a + b
ab2 + b + 7
is also a positive integer.
26. The positive integers a and b are such that

15a + 16b and 16a − 15b

are both squares of positive integers.

What is the smallest possible value of the smaller of the two squares?
27. Find all pairs of integers (a, b) such that

a5 + a3 b + b4 = 0.

28. Determine all pairs of positive integers (a, b) such that

a2
2ab2 − b3 + 1
is a positive integer.
29. Find the smallest integer n > 1 such that
1 2
(1 + 22 + · · · + n2 )
n
is a perfect square.
30. Find all pairs of integers (a, b) such that

a2 b2 − 4a − 4b

is a perfect square.
42 3 Diophantine equations

31. Find all pairs of integers (a, b) such that

a2 + 4b and b2 + 4a

are both perfect squares.

32. Determine the maximum value of m2 + n2 , where m and n are integers 1 ≤ m, n ≤ 2015
and
(n2 − mn − m2 )2 = 1.

33. Let a and b be positive integers. Show that if

a2 + b2
ab + 1
is a positive integer, then it is a perfect square.
34. Find all positive integers k of the form
x+1 y+1
k= + ,
y x
where x and y are also positive integers.
35. Let a and b be positive integers such that

(4a2 − 1)2
4ab − 1
is also a positive integer.
Prove that a = b.

36. Find all integers x, y such that

y 2 − x3 = 7.

37. Find all integer solutions of the equation

x7 − 1
= y 5 − 1.
x−1
3.1 Factorisation 43

3.1 Factorisation
If someone asked you to solve the equation xy = 100, where x and y are integers, then
hopefully you’d find the problem easy. That’s because x must be a divisor of 100. After
listing out all of the divisors of 100—remembering, of course, that they may be positive or
negative—we find the values for x and then solve to find the corresponding values for y = 100
x .
It’s often possible to rearrange a Diophantine equation so that we have an integer on one side
and a factored expression on the other side. It is then a simple matter of examining every
possible factorisation of the integer and lining it up with the factored expression.

Problem Find all pairs of integers (b, c) such that

2b + 3c = 5bc.

Solution Perhaps you hoped to write 5bc − 2b − 3c in the form

5bc − 2b − 3c = (5b + ∗)(c + ∗) + ∗,
or something similar, where each asterisk is an integer. Unfortunately this is not possible.
The ‘5’ gets in the way. We can remove this difficulty by multiplying the whole equation by 5.
In this case we quickly find that
(5b − 3)(5c − 2) = 6.
It only remains to check the pairs of integers which multiply to give 6. We check each of
(5b − 3, 5c − 2) = (−6, −1), (−3, −2), (−2, −3), (−1, −6), (1, 6), (2, 3), (3, 2) and (6, 1). Of
these only (−3, −2) and (2, 3) result in b and c being integers. These pairs correspond to the
solutions (b, c) = (0, 0) or (b, c) = (1, 1).

3.2 Monotonicity
Suppose, for example, we know that x is a perfect square. Suppose also we can show that
a2 < x < (a + 3)2 by other means. Then we may conclude that the only possibilities we need
pursue further are x = (a + 1)2 and x = (a + 2)2 . This is because the function f (t) = t2 is an
increasing function on the positive integers. This sort of technique is valid for any monotonic
(i.e. increasing or decreasing) function, as in the following example.

Problem Find all pairs of integers (x, y) such that

(x + 2)4 − x4 = y 3 .

Solution Rewrite the equation as

y 3 = 8x3 + 24x2 + 32x + 16.
Thus y is even and we may write y = 2z for some integer z. The equation now becomes
z 3 = x3 + 3x2 + 4x + 2.
Note that x3 + 3x2 + 4x + 2 is very close to (x + 1)3 .
It can be shown (and you should do this!) that
x3 < x3 + 3x2 + 4x + 2 < (x + 2)3
for all real numbers x. Thus if there is a solution, it must be z = x + 1. Substituting this in
for z yields x = −1, and then y = 0.
44 3 Diophantine equations

3.3 Bounding arguments

It often occurs that by examining the orderings of the variables you can bound one of them.
What is left is usually a case bash. If the equation is symmetric, you can use a WLOG
(‘without loss of generality’) argument.

Problem Find all ways of writing the number 1 as a sum of the reciprocals of three positive
integers.

1 1 1
Solution The equation a + b + c = 1 is equivalent to the equation

abc = ab + ac + bc.

Note that none of a, b, c equal 1. Due to symmetry, we may order the variables and say
WLOG that 2 ≤ a ≤ b ≤ c. Using this we note that the RHS ≤ 3bc and so abc ≤ 3bc. Thus
a ≤ 3. We may now take the two cases a = 2 and a = 3.

Case 1: a = 3
Then we have 3b + 3c = 2bc. This may be rewritten as

(2b − 3)(2c − 3) = 9.

As c ≥ b ≥ 2, both factors are positive and thus 2b − 3 = 1, 3 or 9. Remembering that

we had 2 ≤ a ≤ b ≤ c, we quickly find that only b = 3 is possible, which leads to the
solution (a, b, c) = (3, 3, 3).

Case 2: a = 2
In a similar way to case 1 we arrive at

(b − 2)(c − 2) = 4.

Remembering that 2 ≤ b ≤ c we find that only b = 3 and b = 4 are possible, which lead
quickly to the solutions (a, b, c) = (2, 3, 6) and (a, b, c) = (2, 4, 4).

In summary we have solutions (3, 3, 3), (2, 3, 6) and (2, 4, 4). However, remember that we
used a WLOG argument at the beginning due to symmetry. So we must remember to include
all permutations of our solutions too, giving us 10 distinct solutions in all.

Problem Find all triples (a, b, c) of positive integers such that

a2 + b + c = abc.

Solution Rearrange to get

b + c = a(bc − a).
We suspect that the RHS should dominate as the variables get larger, especially as b and c
get larger. Note that if a ≥ bc, then there are no solutions because the RHS is not positive
but the LHS is.
Consider the graph of the parabola f (x) = x(bc − x). It is an upside-down parabola that
crosses the x-axis at x = 0 and x = bc. Thus if f is restricted to the interval 1 ≤ x ≤ bc − 1, we
see that f achieves its minimum value simultaneously at x = 1 and at x = bc − 1, the common
3.4 Polynomial modulus 45

minimum value being f (1) = bc − 1. Since 1 ≤ a ≤ bc − 1, we have b + c = f (a) ≥ f (1). Thus

we must have
bc − 1 ≤ b + c.

This certainly is a candidate for bounding! WLOG c ≥ b. If b ≥ 3, then bc ≥ c+c+c > b+c+1.
So we must have either b = 1 or b = 2.

Case 1: b = 1
The original equation can be rearranged to

a2 + 1 2
c= =a+1+ .
a−1 a−1
2
Since a−1 must be an integer, this leads to a = 2 or a = 3. The solutions corresponding
to this case are then (a, b, c) = (2, 1, 5) and (3, 1, 5).

Case 2: b = 2
The original equation can be rearranged to

a2 + 2
c= .
2a − 1

Thus
(2a)2 + 8 9
4c = = 2a + 1 + .
2a − 1 2a − 1
Hence 2a − 1 | 9, which leads to a = 1, 2, 5. The solutions corresponding to this case
are (a, b, c) = (1, 2, 3), (2, 2, 2) and (5, 2, 3).

a2 +2
If you are wondering how we mysteriously came up with the idea of going from c = 2a−1 to
(2a)2 +8 9
4c = 2a−1 = 2a + 1 + 2a−1 , please read on because section 3.4 will provide illumination!

3.4 Polynomial modulus

If we discover that one polynomial expression must be divisible by another, the use of modular
arithmetic for polynomials can be very helpful.

Problem Find all integers a such that a2 + 2 is divisible by 2a − 1.

Solution Using a polynomial modulus we wish to solve a2 + 2 ≡ 0 (mod 2a − 1).

a2 + 2 ≡ 0 (mod 2a − 1)
2
⇔ (2a) + 8 ≡ 0 (mod 2a − 1)
2
⇔ (1) + 8 ≡ 0 (mod 2a − 1)

Thus 2a − 1 | 9. Checking the six factors of 9, including the negative ones, gives rise to six
solutions, namely, a = −4, −1, 0, 1, 2 or 5.
46 3 Diophantine equations

3.5 Quadratic discriminants

Many Diophantine equations can be seen through quadratic eyes if looked at in the right way.
If such an equation can be written as a quadratic in one of its variables, then we know that
for an integer solution to exist, the discriminant of the quadratic must be a perfect square.

Problem Find all integers a such that a2 + 2 is divisible by 2a + 3.

You would be right in thinking that this would be a prime candidate for trying the method
of using a polynomial modulus. And yes, it works just fine. However, study the following
solution which showcases the quadratic discriminant method.

Solution We want the equation

a2 + 2 = 2ak + 3k

to have integer solutions. Viewing this as a quadratic in a, this is only possible if the
discriminant 4k 2 + 12k − 8 is a perfect square. Thus

4k 2 + 12k − 8 = u2 ,

for some non-negative integer u. Note that u2 is even and so u is also. Write u = 2v.
Rearranging the above relation yields

k 2 + 3k − 2 − v 2 = 0.

This is a quadratic in k. Once again its discriminant 9 + 4(2 + v 2 ) is a perfect square. Thus

4v 2 + 17 = w2 ,

for some non-negative integer w. This may be rewritten as

(w − 2v)(w + 2v) = 17.

Since w + 2v ≥ 0, only the positive factors of 17 need to be checked. The only possibility is
w − 2v = 1 and w + 2v = 17. This yields w = 9 and v = 4. Thus u = 8. Solving for k yields
k = 3 or −6. Finally solving for a yields the four solutions a = −10, −2, −1 or 7.

3.6 Modular arithmetic

Suppose you want to show that a certain number divides a variable. Or you’d like to show
that there are no integer solutions to an equation. Modular arithmetic might be just the tool
you need! What is a good modulus to try? Anything that significantly simplifies the situation.
For instance, if a nasty coefficient is divisible by p, then it may be useful to try modulo p. If
there are mth powers involved, then since aϕ(n) ≡ 1 (mod n) whenever gcd(a, n) = 1 from
Euler’s theorem, it is good to try working modulo n, where ϕ(n) = m. If this does not work,
try to choose n such that m | ϕ(n) or at least so that m and ϕ(n) have factors in common.
Squares work well modulo 3, 4, 5, 8 and 16. Cubes work well modulo 7, 9 and 13.

Problem Show that the equation

x4 + 131y 4 = 3z 4 + 2000

has no integer solutions.

3.7 Divisibility and gcds 47

Solution Since ϕ(5) = 4, we try considering the equation modulo 5. The equation simplifies
as
x4 + y 4 ≡ 3z 4 (mod 5).
Observe that a4 ≡ 0, 1 (mod 5) for any integer a. It follows that LHS ≡ 0, 1 or 2 (mod 5)
and RHS ≡ 0 or 3 (mod 5). So we must have LHS ≡ RHS ≡ 0 (mod 5). But this is only
possible if x ≡ y ≡ z ≡ 0 (mod 5). However, this in turn implies that 54 | 2000, which is a
contradiction.

Problem Show that the equation

5m2 − 6mn + 7n2 = 1985

has no integer solutions.

Solution Viewing the equation as a quadratic in m, if it did have solutions, then the
discriminant
36n2 − 20(7n2 − 1985) = 4(9925 − 26n2 )
should be a perfect square. Thus
9925 − 26n2
is a perfect square.
Squares can only be congruent to 0, 1 or 4 modulo 8. Hence considering this expression
modulo 8 yields

5 − 2n2 ≡ 0, 1, 4 (mod 8)
≡1 (mod 8) (since LHS is odd)
2
⇒ 2n ≡ 4 (mod 8)
2
⇒ n ≡2 (mod 4).

This is impossible because squares are congruent to 0 or 1 modulo 4.

3.7 Divisibility and gcds

Considering the greatest common divisor (gcd) of a pair or triple of numbers can help.

Problem Find all pairs of integers (x, y) such that

x4 + 2x2 y + y 3 = 0.

Solution Let gcd(x, y) = d. Thus we may write x = sd and y = td, where gcd(s, t) = 1.
The equation becomes
s4 d + 2s2 t + t3 = 0.
Now if p is any prime factor of s, we see that p | t3 and thus p | t. But since gcd(s, t) = 1,
this forces s = ±1 and thus d = −2t − t3 . Thus the complete set of solutions is given by

x = ±t(2 + t2 )
y = −t2 (2 + t2 ),

for any integer t. We can verify these solutions by direct substitution.

48 3 Diophantine equations

Problem Find all positive integral solutions to the equation

a2 + 2b2 = c2 .

Solution By dividing out by λ = gcd(a, b, c), we may assume that a, b and c are pairwise
coprime. Considering the equation modulo 4 shows us that a and c are both odd and that b
is even. Thus after writing b = 2d and rearranging we have
c−a c+a
2d2 = ·
2 2
and that gcd( c−a c+a c−a c+a
2 , 2 ) = gcd(a, c) = 1. Thus whichever one of 2 and 2 is odd must
be a perfect square, say u , and the other must be twice a square, say 2v 2 . Adding these
2

together yields
c = u2 + 2v 2 , a = ±(u2 − 2v 2 ) and d = uv.
Thus the most general solutions are given by

a = |u2 − 2v 2 |λ
b = 2uvλ
c = (u2 + 2v 2 )λ,

where λ is any positive integer and (u, v) are any pair of relatively prime positive integers.1

Note that the solutions are valid even if gcd(u, v) > 1.

3.8 Reduction of variables

Some problems do not ask for all solutions but only ask for the existence of an infinite number
of solutions. Reducing the number of variables by some well thought out substitutions can
often simplify the matter. The idea is to make as many things cancel out as possible, while
still leaving scope for infinitely many solutions.

Problem Show that the equation

a3 + b3 = c3 + d3 + e3

has infinitely many integer solutions where gcd(a, b, c, d, e) = 1 and a, b, c, d and e are all
distinct.

Solution We try the substitution a = x−s, b = x+s. The left-hand side becomes 2x3 +6s2 x.
A little more thought encourages us to try a similar thing on the right-hand side, namely
c = x − t, and d = x + t. Using these, the equation simplifies to

6x(s − t)(s + t) = e3 .

We have now reduced the equation to four variables. Trying s = 2 and t = 1 simplifies matters
even further to
18x = e3 .
1A very similar approach yields a formula for all Pythagorean triples, that is, triples of positive integers that
are the side lengths of a right-angled triangle.
3.9 Infinite descent 49

Choosing e = 18y finishes off the problem. We have discovered an infinite one-parameter
family of solutions given by

a = 182 y 3 − 2
b = 182 y 3 + 2
c = 182 y 3 − 1
d = 182 y 3 + 1
e = 18y.

It is clear that they are distinct with gcd(a, b, c, d, e) = 1.

3.9 Infinite descent

The following statement may seem pretty obvious: It is impossible to have an infinitely
descending sequence of positive integers. This is known as the principle of infinite descent.
This principle is equivalent to the fact that any set of positive integers has a smallest member.
The principle of infinite descent is particularly useful for proving that a Diophantine equation
has no solutions. For instance, suppose we could prove that given any solution to a Diophantine
equation there is another solution which is even smaller (but still positive). Then if we applied
this to a supposed minimal solution, we would have a contradiction.
By minimal we mean that we can measure the size of the solution in some sense: for example,
the sum of the absolute values of the variables, or perhaps, the absolute value of a particular
variable.

Problem Find all integral solutions to

x3 + 3y 3 + 9z 3 = 9xyz.

Solution Assume we have a solution (x, y, z) with |x| + |y| + |z| > 0. Then there is a solution
with |x| + |y| + |z| minimal. The equation implies 3 | x3 and so 3 | x. Thus x = 3a for some
integer a. Substitute this into the equation and divide the equation by 3 to find

9a3 + y 3 + 3z 3 = 9ayz.

In a similar way we obtain 3 | y so that y = 3b for some integer b and so we deduce

3a3 + 9b3 + z 3 = 9abz.

From this we obtain 3 | z so that z = 3c for some integer c and so we deduce

a3 + 3b3 + 9c3 = 9abc.

It only remains to note that (a, b, c) is also a solution to the original equation and satisfies
0 < |a| + |b| + |c| = |x|+|y|+|z|
3 . This contradicts the minimality of |x| + |y| + |z|. Thus there
is no solution with |x| + |y| + |z| > 0. Hence the only solution is x = y = z = 0.
50 3 Diophantine equations

3.10 Vieta jumping

Suppose that we have a Diophantine equation in two variables x and y which is a monic2
quadratic in x. If we have one solution (x, y) = (a, b) in integers, it turns out that we can
apply Vieta’s formula3 for the sum of the roots to generate another integer solution.

Problem Show that the Diophantine equation

x2 + y 2 − 4xy − 4 = 0

has infinitely many solutions.

Solution We try x = 0. This quickly yields y = 2 is a solution. So (0, 2) is a solution.

Now set y = 2 into the equation. We obtain the quadratic equation

x2 − 8x = 0.

Using Vieta’s formula, the sum of the roots is 8. Since x = 0 is one of the roots, the other
must be x = 8. So (8, 2) is a solution. By symmetry, (2, 8) is also a solution.
Next substitute y = 8 into the original equation. We obtain the quadratic equation

x2 − 32x + 60 = 0.

Again by Vieta’s formula, the sum of the roots is 32. Since x = 2 is one of the roots, the
other must be x = 30. So (30, 8) is a solution. By symmetry, so is (8, 30).
This procedure may be continued indefinitely. In general if (x, y) = (a, b) is a solution, we
may construct another solution by considering the quadratic equation

x2 − 4bx + b2 − 4 = 0.

From Vieta’s formula, the sum of the roots is 4b. Since x = a is one of the roots, the
other must be 4b − a. Thus (x, y) = (4b − a, b) is also a solution and by symmetry, so is
(x, y) = (b, 4b − a).
If we assume that 0 ≤ a < b, then b < 4b − a. Thus any solution (a, b) leads to a bigger
solution (b, 4b − a). Since (0, 2) is a solution, we can thus construct infinitely many solutions
using this procedure starting from (0, 2).

By using the technique of infinite descent, it is possible to prove that all integer solutions
(x, y), with 0 ≤ x ≤ y, to the given equation occur as a consecutive pair of the generated
sequence 0, 2, 8, 30, . . . , where the rule for the sequence is

x0 = 0, x1 = 2 and xn+2 = 4xn+1 − xn for n = 0, 1, 2 . . ..

See if you can prove this.

2A polynomial of degree n in x is monic if the coefficient of xn is equal to 1.
3 See section 9.4 if you don’t know about Vieta’s formulas.
3.11 Cyclotomic recognition 51

3.11 Cyclotomic recognition

The cyclotomic polynomials Φ1 (x), Φ2 (x), Φ3 (x), . . . are defined recursively by the relations
Y
xn − 1 = Φd (x)
d|n

for n ≥ 1.
From this definition it is easy to compute the first few cyclotomic polynomials to be
Φ1 (x) = x − 1
Φ2 (x) = x + 1
Φ3 (x) = x2 + x + 1
Φ4 (x) = x2 + 1
Φ5 (x) = x4 + x3 + x2 + x + 1
Φ6 (x) = x2 − x + 1
Φ7 (x) = x6 + x5 + x4 + x3 + x2 + x + 1
Φ8 (x) = x4 + 1.

Additionally, it is not too hard to show directly from the definition that
Φp (x) = xp−1 + xp−2 + · · · + x + 1
for any prime p.
Sometimes part of an algebraic expression can be seen to be part of a cyclotomic polynomial.
Since any cyclotomic polynomial is a factor of an expression of the form xn − 1, there is a
fair chance that the theorems of Euler or Fermat from section 2.12 might be helpful.

Problem Prove that the equation

x3 = y 16 + y 15 + · · · + y 2 + y + 9
has no solutions in positive integers.

Solution We astutely recognise that the RHS is equal to 8 plus the cyclotomic polynomial
Φ17 (y) = y 16 + y 15 + · · · + y 2 + y + 1. Rewrite the equation as
x3 − 8 = y 16 + y 15 + · · · + y 2 + y + 1.
Note that the RHS is a factor of y 17 − 1.
Let p be any prime factor of the RHS. Note that this implies that p - y. Then it follows that
y 17 ≡ 1 (mod p).

However, since p - y, we also know from Fermat’s little theorem that

y p−1 ≡ 1 (mod p).

Using the gcd trick from section 2.13 we deduce that

yd ≡ 1 (mod p),
where d = gcd(17, p − 1).
Thus there are two possibilities for d, namely, d = 1 or d = 17.
52 3 Diophantine equations

Case 1: d = 1
Then we have y ≡ 1 (mod p), and so,

y 16 + y 15 + · · · + y 2 + y + 1 ≡ 17 (mod p).

Thus p | 17 and therefore, p = 17.

Case 2: d = 17
Then 17 | p − 1 from which it follows that p ≡ 1 (mod 17).

We have shown that every prime factor p of the RHS satisfies either p = 17 or p ≡ 1 (mod 17).
Hence it follows that every factor of the RHS is congruent to 0 or 1 modulo 17.
However, since
x3 − 8 = (x − 2)(x2 + 2x + 4)
we see that x − 2 ≡ 0, 1 (mod 17).
But if x ≡ 2 (mod 17), then x2 + 2x + 4 ≡ 12 (mod 17).
If x ≡ 3 (mod 17), then x2 + 2x + 4 ≡ 2 (mod 17).
In both cases the other factor is not congruent to 0 or 1 modulo 17, a contradiction. Thus
the equation has no integer solutions.
 Plane geometry
4
Welcome dear reader to a domain where many fear to tread, and many stumble. The domain
is plane geometry, surely one of the most beautiful areas of mathematics.
In one sense, geometry is the easiest mathematical topic to learn for it is the epitome of logical
rigour. But in another sense, geometry is the most difficult subject to teach. It relies heavily
on ingenuity. Solving a geometry problem is more often dependent on noticing something
that is difficult to see, perhaps more so than elsewhere. But perseverance is an excellent
character trait for problem solving.
We shall assume you know all the basic theorems of plane geometry: congruent triangles,
similar triangles, basic circle theorems, cyclic quadrilaterals and special points of a triangle.
There are a number of pitfalls unique to geometry, into which many an inexperienced problem-
solver falls. They are incredibly easy to avoid but far too many fail to avoid them, because
they think these matters are not important.

Your chances of solving the problem are as good as your diagram. If you
don’t draw a good diagram in a geometry problem, unless it is very, very easy, you have
significantly lowered your chances of solving it.

Size matters. We admire the responsible sentiment of those who don’t want to use
up too much paper. But with a big accurate diagram, you’re likely to see things more
clearly and solve the problem. With a dozen small terrible diagrams you’re not going
to get anywhere. A good diagram should take up most of the page.

Many-coloured diagrams stimulate the mind. It is prettier to draw colourful

diagrams. Consider that almost all of your work in mathematics will be done in vast
swathes of monochromatic grey or black or blue. Wouldn’t you appreciate a change of
colour occasionally? A multicoloured diagram is often a better diagram, helping you
to solve the problem. Drawing different parts or aspects of the diagram in different
colours helps you to see these aspects independently, and see relationships between
them. Unfortunately for technical reasons this book is black and white with a little
blue and grey. But you do not have such excuses!

Accuracy is a virtue. If you draw freehand, you draw sloppily. If you draw accurately,
using ruler and compass, your diagram will be spot on the mark. And with a perfect
diagram, you can see very concretely the real situation the problem is asking about.
You can spot things, make guesses, get a feel for how all the points and lines relate, and
54 4 Plane geometry

so on. Does one of your angles look like a right angle? Does one of your quadrilaterals
look cyclic? Do those three points look collinear? Your accurate diagram will give you
all sorts of suggestions, if you look hard enough at it, and you can try and prove them.

4.0 Problems
1. Two parallel lines are tangent to a circle with centre O. A third line, also tangent to
the circle, meets the two parallel lines at A and B.
Prove that AO is perpendicular to OB.
2. Suppose that two circles are externally tangent at P . Let a common tangent touch the
circles at A and B.
Prove that triangle AP B is right-angled.
3. Let ABC be a triangle with incentre I. Suppose that X is the midpoint of the arc BC
not containing A on the circumcircle of triangle ABC.
Prove that X is the circumcentre of triangle BIC.
4. Let D, E, F be points on sides AB, BC, CA of triangle ABC such that DE = BE
and F E = CE.
Prove that the circumcentre of triangle ADF lies on the bisector of ∠DEF .
5. Given a triangle ABC, let the median from vertex A intersect the circumcircle of the
triangle again at K. A circle Γ passes through A and B such that BC is tangent to Γ.
Let L be the intersection of AK and Γ different from A.
Prove that BLCK is a parallelogram.
6. Let two circles intersect at A and B. Suppose that a common tangent to the two circles
meets them at P and Q.
If the line AB meets P Q at M , show that M is the midpoint of P Q.
7. ABC is a triangle, right-angled at C. The internal angle bisectors of angle BAC and
angle ABC meet BC and CA at P and Q, respectively. Let M and N be the feet of
the perpendiculars from P and Q to AB, respectively.
Find the size of ∠M CN .
8. Two circles Γ1 and Γ2 intersect at A and B. It is known that Γ2 passes through point
O, where O is the centre of Γ1 . Point X lies on the arc AOB, and AX intersects Γ1
again at Y .
Prove that XY = XB.
9. Let AD be an altitude and H be the orthocentre of 4ABC. Let line AD meet the
circumcircle of 4ABC at X.
(a) Prove that HD = DX.
(b) Prove that the circumradius of the triangles formed by H and any two vertices is
equal to the circumradius of 4ABC.
10. Two circles of equal radius intersect at A and B. The point C is on one of the circles
such that B is the midpoint of the arc AC.
Prove that AC is tangent to the other circle.
4.0 Problems 55

11. Two circles C1 and C2 intersect at A and B. Let P be a point on C1 and Q be a point
on C2 , so that P and Q lie on opposite sides of the line AB. Suppose further that
∠AP B + ∠AQB = 90◦ .
Prove that if O1 is the centre of C1 and O2 is the centre of C2 , then the triangles O1 AO2
and O1 BO2 are right-angled.
12. Let ABCD be a quadrilateral such that CD bisects ∠ACB. Suppose that

∠DAC + ∠DBC + ∠DCB = 90◦ .

Prove that D is the incentre of triangle ABC.

13. Circles C1 and C2 intersect at two distinct points P and Q. A line ` through P meets
the circles C1 and C2 in A and B, respectively. Let Y be the midpoint of AB and
suppose QY meets the circles C1 and C2 in X and Z, respectively.
Show that XY = Y Z.
14. Let AB and CD be distinct parallel chords of a circle and let P be a point on their
common perpendicular bisector. The lines AP and CP meet the circle again at Q and
R, respectively. Let S be the intersection of the lines BR and QD.
Prove that P S is parallel to AB.
15. Triangle ABC is a 40◦ -60◦ -80◦ triangle. The angle bisector at A meets BC at D. Let
E be the midpoint of CD and let F be the foot of the altitude from C to AB.
Prove that perpendicular bisector of EF intersects AC at its midpoint.
16. From a point O, two tangents are drawn to a given circle touching it at points A and B.
The chord AC is drawn parallel to the tangent OB, and OC intersects the circle at E.
Prove that the line AE bisects OB.
17. Let ABC be a triangle with incentre I, inradius r, circumcentre O, circumradius R and
semiperimeter s. Let Ia , Ib and Ic be the centres of the excircles opposite A, B and
C, respectively, and let ra , rb and rc be the radii of the excircles opposite A, B and C,
respectively.
Prove the following formulas.
1 1 1 1
(a) r = ra + rb + rc
(b) Area(4Ia Ib Ic ) = 2sR
(c) OI 2 = R2 − 2Rr
18. Let k be a semicircle with diameter AB. Let D be a point such that AB = AD and AD
intersects k at the point E. Let F be the point on the chord AE such that DE = EF .
Let BF extended meet k at the point C.
Show that ∠BAE = 2∠EAC.
19. The inscribed circle of triangle ABC is tangent to the sides AB and BC of 4ABC in
the points P and Q. The line P Q intersects the bisector of ∠BAC at the point S.
Prove that AS is perpendicular to SC.
20. Two chords AB and CD of a circle intersect at a point E inside the circle. Let M be
an interior point of the segment EB. The tangent line at E to the circle through D, E
and M intersects the lines BC and AC at F and G, respectively.
If AM/M B = t, find EG/EF in terms of t.
56 4 Plane geometry

21. The bisectors of the angles A and B of the triangle ABC meet the sides BC and CA
at the points D and E, respectively.
Assuming that AE + BD = AB, determine the size of ∠C.

22. Suppose that P is a point inside triangle ABC that satisfies ∠ABP = ∠ACP and
∠CBP = ∠CAP .
Prove that P is the orthocentre of the triangle.

23. Let P be an interior point of 4ABC, and let AP , BP and CP intersect BC, CA and
AB at D, E and F , respectively.
Prove that
AP BP CP AE AF AP
+ + =2 and + = .
AD BE CF EC FB PD
24. Convex quadrilateral P QRS satisfies P Q = QR and P Q is not parallel to SR. The
diagonals P R and QS intersect at T . The perpendicular bisectors of P R and QS
intersect at V .
Prove that P QT V is cyclic.

25. Triangle ABC satisfies ∠ABC = 2∠ACB. Point P , located inside the triangle, satisfies
P B = P C and AP = AB.
Prove that ∠BAC = 3∠P AC.

26. A triangle ABC satisfies ∠ACB > ∠ABC. The internal bisector of ∠BAC meets BC
at D. The point E on AB is such that ∠EDB = 90◦ . The point F on AC is such that
∠BED = ∠DEF .
Show that ∠BAD = ∠F DC.

27. Let A, B and C be three collinear points with B between A and C. Equilateral triangles
ABD, BCE and CAF are constructed with D and E on one side of the line AC and
F on the opposite side.

(a) Prove that the centroids of the triangles are the vertices of an equilateral triangle.
(b) Prove that the centroid of this triangle lies on the line AC.

28. Let ABC be an acute triangle. Let AD be the altitude on BC, and let H be any
interior point on AD. Lines BH and CH, when extended, intersect AC and AB at E
and F , respectively.
Prove that ∠EDH = ∠F DH.

29. Let C be a circle with centre O and let A and B be points on the circle such that
∠AOB = 90◦ . Let C1 and C2 be two circles internally tangent to C at points A and B,
respectively. Furthermore C1 and C2 are tangent to each other and have centres O1
and O2 , respectively. Circle C3 is located inside angle AOB. It has centre O3 and is
externally tangent to C1 and C2 and is internally tangent to C.
Prove that OO1 O3 O2 is a rectangle.

30. Let ABC be an acute triangle. Let M be the midpoint of BC and P be the point on
AM such that M B = M P . Let H be the foot of the perpendicular from P to BC.
The lines through H perpendicular to P B and P C meet AB and AC at Q and R,
respectively.
Show that BC is tangent to the circle through Q, H and R at H.
4.0 Problems 57

31. Triangle ABC is acute-angled with circumcircle Γ and orthocentre H so that AB 6= AC.
Let AH meet BC and Γ at D and E, respectively. Let F be the midpoint of BC. The
line tangent to circle DEF at D meets the lines AB and AC at M and L, respectively.
Prove that M D = DL.

32. Let ABC be a triangle with a right angle at A and area 4, and let S be its circumcircle.
Let S1 be the circle tangent to sides AB and AC and internally tangent to S. Let S2
be the circle tangent to rays AB and AC and externally tangent to S. Let r1 and r2
denote the respective radii of S1 and S2 .
Prove that r1 r2 = 44.

33. Angle A is the smallest in triangle ABC. The points B and C divide the circumference
of the triangle into two arcs. Let U be an interior point of the arc between B and C
which does not contain A. The perpendicular bisectors of AB and AC meet the line
AU at V and W , respectively. The lines BV and CW meet at T .
Show that AU = T B + T C.

34. Points X, Y and Z are located inside triangle ABC and satisfy

∠Y AC = ∠ZAB = 13 ∠A
∠ZBA = ∠XBC = 13 ∠B
∠XCB = ∠Y CA = 13 ∠C.

Prove that triangle XY Z is equilateral.1

35. Let ω be the incircle of triangle ABC. Let L, N and E be the points of tangency of
ω with the sides AB, BC and CA, respectively. Lines LE and BC intersect at the
point H and lines LN and AC intersect at the point J. (All the points H, J, N and E
lie on the same side of the line AB.) Let O and P be the midpoints of EJ and N H,
respectively.
Prove that p
Area(HJN E) = 4 Area(ABOP ) × Area(COP ).

36. Let ABCD be a convex quadrilateral in which the diagonals AC and BD are perpendic-
ular and the opposite sides AB and DC are not parallel. The perpendicular bisectors
of AB and CD meet at point P inside ABCD.
Prove that ABCD is cyclic if and only if triangles ABP and CDP have equal areas.

37. Consider five points A, B, C, D and E such that ABCD is a parallelogram and BCED
is a cyclic quadrilateral. Let ` be a line passing through A. Suppose that ` intersects
the interior of the segment DC at F and intersects line BC at G. Suppose also that
EF = EG = EC.
Prove that ` is the bisector of angle DAB.

38. Let ABCD be a convex quadrilateral which does not have any two sides of equal length.
Prove that ABCD is cyclic if and only if there exist points Q and R on line BD, one
strictly between B and D, the other outside of the segment BD, such that

∠DAQ = ∠DCQ = ∠BAR = ∠BCR.

1 This result is known as Morley’s theorem.
58 4 Plane geometry

39. A circle with centre O passes through the vertices A and C of triangle ABC and
intersects the segments AB and BC again at distinct points K and N , respectively.
The circumscribed circles of the triangles ABC and KBN intersect at exactly two
distinct points B and M .
Prove that angle OM B is a right angle.
4.1 Angle chasing 59

4.1 Angle chasing

One of the simplest techniques in geometry is to ‘chase’ angles. Given a diagram, insert all
the angles you can. Label one angle as α or some other symbol, and then work out what you
can in terms of α. Continuing this process disposes of many a geometry problem.
Warning! Doing an angle chase is much easier than writing down an angle chase in a proof,
step by step. The former just involves writing stuff on your diagram. The latter involves
explaining each step, in the correct order. As a result it’s sometimes hard to read an angle
chase when written down as a proof. (Just ask any Olympiad marker!) So don’t worry if
what’s written here is a little hard to follow. It’s best to draw your own diagram and work
through the proof on that diagram.
Let’s start by stating an underused but important theorem.

Alternate segment theorem Let A, B, C be points on a circle Γ, and let P A be a line

segment such that P lies on the opposite side of line AB as C. Then the line P A is tangent
to Γ at A if and only if ∠ACB = ∠P AB.

We will now see how angle chasing can make light work of a problem.

Problem In parallelogram ABCD, AC is longer than BD. Let P be a point on AC such

that BCDP is a cyclic quadrilateral.
Prove that BD is a common tangent to the circumcircles of triangle ADP and triangle ABP .

Solution

D C

A B

By the alternate segment theorem, it is sufficient to prove that ∠P DB = ∠DAP and

∠P BD = ∠BAP .
Since the quadrilateral BCDP is cyclic, we have ∠P DB = ∠P CB. Also ∠P CB = ∠DAP
because ABCD is a parallelogram. Putting these two pieces of information together gives
∠P DB = ∠DAP , which is one of the statements that we wanted to prove. The other follows
by an entirely analogous argument.

4.2 Cyclic quadrilaterals

It’s always great to find a cyclic quadrilateral. In fact it is positive progress, because they
have useful properties.
60 4 Plane geometry

Theorem

A quadrilateral ABCD is cyclic if and only if ∠ABC + ∠ADC = 180◦ .

A quadrilateral ABCD is cyclic if and only if ∠ACB = ∠ADB.

Even problems that don’t seem to involve any circles often feature cyclic quadrilaterals.
It’s worth remembering that discovering a cyclic quadrilateral is a valuable find because it
gives you new information that you could not otherwise obtain by standard angle-chasing
techniques.

Problem In triangle ABC, points D and E are located on the side BC such that AD is an
altitude and AE is an angle bisector. The point M on AE is such that BM is perpendicular
to AE and the point N on AC is such that EN is perpendicular to AC.
Prove that the points D, M , N are collinear.

Solution
A

B D E C

With three right angles floating around the diagram, you can be confident that there are also
cyclic quadrilaterals. In fact, we know that ABDM is cyclic because ∠ADB = ∠AM B = 90◦ .
We also know that ADEN is cyclic because ∠ADE + ∠AN E = 90◦ + 90◦ = 180◦ .
Now there are many ways to show that the points D, M , N are collinear, but our particular
approach will be to prove that ∠BDM + ∠N DC = 180◦ .
As with many plane geometry problems, we start by labelling a sensible angle. Here we will
use ∠BAC = 2α. Then we use this to label as many other angles in the diagram as possible.
For a start, we have ∠BAE = ∠CAE = α.
The cyclic quadrilateral ABDM tells us that
∠BDM = 180◦ − ∠BAM = 180◦ − ∠BAE = 180◦ − α.
The cyclic quadrilateral ADEN tells us that
∠N DC = ∠N DE = ∠N AE = α.
Therefore, ∠BDM + ∠N DC = (180◦ − α) + α = 180◦ , as required.

4.3 One step at a time

With any type of problem, not just maths problems, often a good tactic is to examine different
parts of the problem one by one and see what you can work out. Then putting it all together
4.3 One step at a time 61

you may obtain an answer.

This is pretty obvious advice but it’s often overlooked in geometry, since you usually draw
the whole diagram at once, and the whole diagram is often too complicated to see anything,
especially in harder problems.
Nobody can teach you how to separate out the parts of a difficult problem. Indeed getting
to know how to look at things the right way is part of the skill of problem solving. This
principle can be used to solve the following easier problem.

Problem Let ABCD be a square and P be a point on its side BC. The circle passing
through points A, B and P intersects BD once more at point Q. The circle passing through
points C, P and Q intersects BD once more at point R.
Prove that points A, R and P are collinear.

Solution First of all draw the entire diagram carefully. But after that it pays to draw just
a part of the diagram and see what we can glean. We will examine the situation with each
circle separately. So first let’s just examine the set-up with circle ABP Q.

A D

B P C

Since BD is the diagonal of a square, ∠ABD = ∠CBD = 45◦ .

Clearly ABP Q is cyclic, so ∠AP Q = ∠ABQ = 45◦ and ∠P AQ = ∠P BQ = 45◦ . Hence
AP Q is isosceles with AQ = P Q.
Similarly, letting ∠AP B = α we can chase all the angles around.

A D
45
α Q

45 45
45 α
B P C

Some of the other angles in the diagram are ∠BQP = ∠BAP = 90◦ −α and ∠QP C = 135◦ −α.
62 4 Plane geometry

Well, that’s all pretty interesting. Now let’s think about drawing in the circle around C, P, Q.
Actually, let’s just think about drawing in CQ—one step at a time!

A D
45
α Q

45 45
45 α
B P C

When you only look at this part of the diagram, it’s pretty clear that by some sort of symmetry,
AQ = CQ. You could prove this by saying that C is the reflection of A in the line BD, and
Q is on BD. Or you could prove ADQ and CDQ are congruent. Anyway, we now have
AQ = CQ = P Q. So CP Q is isosceles and ∠QCP = ∠QP C = 135◦ − α.
Now we’ll put the second circle in.

A D
Q

B P C

We now have ∠P CR = ∠P QR = 90◦ − α. Since ∠P CQ = 135◦ − α, we have ∠RCQ = 45◦ .

Thus ∠RP Q = ∠RCQ = 45◦ . But now ∠RP Q = ∠AP Q = 45◦ . Therefore, points A, R and
P are collinear as required.

4.4 Triangle centres

It is likely that you have come across the following special points of a triangle.2 See if you
can navigate through the following exercises.

1. Centroid Let ABC be a triangle with medians AX, BY and CZ meeting at centroid G.
Prove that the line segments XY , Y Z and ZX divide triangle ABC into four
smaller triangles which are congruent to each other.
2 If you don’t know about centroids, orthocentres, and so on, you can look them up on the internet.
4.4 Triangle centres 63

The triangle XY Z is called the medial triangle. Prove that it can be obtained by
performing a dilation3 on triangle ABC with centre G and factor − 12 .
Prove that G is the centroid of the medial triangle.
Prove that the three medians of a triangle divide it into six triangles of equal area.
Prove that
AG BG CG
= = = 2.
GX GY GZ
2. Orthocentre Let ABC be a triangle with altitudes AD, BE and CF meeting at
orthocentre H.

Draw triangle ABC, the three altitudes AD, BE, CF and triangle DEF .
Write down all six cyclic quadrilaterals which appear in the diagram.
Label every angle in the diagram in terms of ∠A, ∠B and ∠C.
What is the relationship between H and triangle DEF ?
What are the orthocentres of triangles AHB, BHC and CHA?

The four points A, B, C and H are called orthocentric.

3. Circumcentre Let ABC be a triangle with circumcentre O.

Draw triangle ABC, the perpendicular bisectors of the sides XO, Y O, ZO and
the medial triangle XY Z.
Write down all three cyclic quadrilaterals appearing in your diagram.
Label every angle in the diagram in terms of ∠A, ∠B and ∠C.
What is the relationship between O and triangle XY Z?

4. Incentre Let the incircle of triangle ABC have centre I and touch AB, BC and CA
at P , Q and R, respectively.

Draw triangle ABC, the angle bisectors AI, BI, CI, the points of tangency of the
incircle P , Q, R, the segments P I, QI, RI and triangle P QR.
Write down all three cyclic quadrilaterals appearing in your diagram.
Label every angle in the diagram in terms of ∠A, ∠B and ∠C.
Label the lengths AQ, AR, BR, BP , CP and CQ in terms of the side lengths a,
b and c of the triangle.
Prove that the area of triangle ABC is given by rs, where r is the radius of the
incircle and s = a+b+c
2 is the semiperimeter.

5. Excentres Let Ia , Ib and Ic be the excentres opposite A, B and C, respectively, of

triangle ABC. Suppose that the excircle centred at Ia touches AB, BC and CA at P ,
Q and R, respectively.

Draw triangle ABC, the internal angle bisector AIa , the external angle bisectors
BIa , CIa , the points of tangency of the excircle P , Q, R, the segments P Ia , QIa ,
RIa and triangle P QR.
Write down all three cyclic quadrilaterals appearing in your diagram.
Label every angle in the diagram in terms of ∠A, ∠B and ∠C.
3 See section 7.3.
64 4 Plane geometry

Label the lengths AQ, AR, BR, BP , CP and CQ in terms of the side lengths a,
b and c of the triangle.
Prove that the area of triangle ABC is given by ra (s − a), where ra is the radius
of the excircle opposite A and s is the semiperimeter.
What is the relationship between I and triangle Ia Ib Ic ?
What is the relationship between triangle Ia Ib Ic and the circumcircle of triangle
ABC?

4.5 Constructions
There’s a simple rule of thumb for deciding whether a geometry question is hard, or really
hard. If it can be solved without drawing any extra lines beyond the ones you’re given (and
maybe a couple more natural ones), it’s not that hard. If it requires you to make some
constructions of your own, it’s really hard.
Again, nobody can teach you to look at a diagram and say where to make a construction—this
is a truly creative business. But by doing enough problems it’s possible to get a feel for how a
certain situation works, enough that you can try inserting various extra bits, here and there.
Don’t be afraid to do so: if it doesn’t work, or your diagram becomes too cluttered, just draw
another one!

Problem Suppose that A, B, M are points on a circle such that M is the midpoint of the
arc AB. Let C be an arbitrary point on the arc AM B such that AC is longer than BC.
Let D be the foot of the perpendicular from M to AC.
Prove that AD = DC + CB.

Solution

P
M
x

A B

A much easier task than proving one length is equal to the sum of two lengths is proving that
one length is equal to another. With this in mind, we extend the line AC to the point P such
that CP = CB. Of course, what we now need to prove is that AD = DP .
But if AD = DP , then we would know that M lies on the perpendicular bisector of AP .
Since M also lies on the perpendicular bisector of AB, it must be the case that M is the
circumcentre of triangle ABP . Let’s aim to prove this using an angle chase.
4.5 Constructions 65

First, we let ∠AP B = x. Since we have constructed triangle BCP to be isosceles, we know
that ∠P BC = x and ∠P CB = 180◦ − 2x. From this, it follows that ∠ACB = 2x and since
ABCM is a cyclic quadrilateral, we also have ∠AM B = 2x.
Now look at what we have here. The chord AB subtends an angle 2x at M with AM = BM
and an angle x at P . Since P and M lie on the same side of AB, the point M is indeed the
circumcentre of triangle ABP . We now know that M D splits the isosceles triangle AM P
into two congruent triangles, so AD = DP .

By the way, as with most geometry problems, different solutions can be found. For the one
just discussed, it is possible to define a point Q on segment AD such that CD = DQ. Thus
it remains to show that AQ = BC. This can be done by proving that triangles AM Q and
BM C are congruent. See if you can angle chase out the details!

Problem A triangle ABC has squares P ABK and QACL constructed on its exterior. The
altitude AD of triangle ABC is extended to meet P Q at point M .
Prove that M is the midpoint of P Q.

Solution

R Q

P S
A L

B D C

First, let’s do a bit of angle chasing. If we let ∠CAD = α and ∠BAD = β, then we have
∠ACD = 90◦ − α and ∠ABD = 90◦ − β. Also, since angles on a straight line add to 180◦ ,
we have ∠QAM = 90◦ − α and ∠P AM = 90◦ − β.
There are several ways to proceed from here, but one way is as follows. Note that QA = AC
and ∠QAM = ∠ACD. These are quite similar situations, and by drawing a single line
segment, we can create a pair of congruent triangles—certainly a useful thing to do! So let R
be the foot of the perpendicular from Q to the line M D. This gives us the congruent triangles
QAR and ACD, as desired. But what we have done on the right side of the diagram, we can
similarly do on the left. So let S be the foot of the perpendicular from P to the line M D, so
that we have congruent triangles P AS and ABD. In particular, we have managed to prove
that QR = AD = P S.
Therefore, M is horizontally halfway between P and Q. This means that M must be the
midpoint of P Q, and we are done.
66 4 Plane geometry

(You could make this last statement a little more rigorous by showing that triangle M QR is
congruent to triangle M P S, so that M P = M Q.)

4.6 Exploit symmetry

When the diagram is symmetric, you can draw the same thing in different places.

Problem In rectangle ABCD, let M and N be the midpoints of BC and CD, respectively.
Let DM and BN intersect at P .
Prove that ∠M AN = ∠BP M .

Solution We use the symmetry of the rectangle to redraw the situation near A, over
near B. That is, let Q be the midpoint of AD and construct BQ. Then, by symmetry,
∠QBN = ∠M AN .

A Q D

B M C

All that we need to prove now is that BQ and M D are parallel. But once we observe that
BM and QD are equal and parallel, it follows that BQDM is a parallelogram and so BQ is
parallel to M D as desired.

4.7 Extend to the circumcircle

When you’ve got a triangle involved and some cevians (any lines through the three vertices of
the triangle such as medians, angle bisectors or altitudes), a useful construction is to extend
them to the circumcircle of the triangle. This might sound like a pretty random thing to
do, but it comes up sufficiently often that we’re devoting a section to it. Nobody has ever
explained to us a deep underlying reason as to why it should work so well, but it does.4
Here is a nice property associated with extending the angle bisectors to the circumcircle. See
if you can prove it.

Extensions of angle bisectors to circumcircle Let I be the incentre of triangle ABC.

If AI extended meets the circumcircle of triangle ABC at D, then BD = CD = ID. This
means that D is the circumcentre of triangle IBC.

A diagram of this configuration is found on page 85.

4 Forexample, this construction can be used with (i) the extended sine rule (section 4.9), (ii) Ceva’s theorem
(section 6.4) and (iii) a cyclic hexagon with concurrent diagonals as per the configuration found on page 95.
4.7 Extend to the circumcircle 67

There is one diagram that is particularly exciting! Take a triangle and its circumcircle. Draw
its altitudes and extend them to the circumcircle. Now, as an exercise, try to discover as
many interesting properties of this diagram as you can.
Here are a couple of such properties that you should try to prove for yourself.

Extensions of altitudes to circumcircle Let AD, BE and CF be the altitudes and

let H the orthocentre of triangle ABC. If AD, BE and CF are extended to meet the
circumcircle of triangle ABC at A0 , B 0 and C 0 , respectively, then HD = DA0 , HE = EB 0
and HF = F C 0 . This implies the following properties.
The reflections of H in the three sides of the triangle lie on the circumcircle.
The diagram has three kites in it, namely, HBA0 C, HCB 0 A and HAC 0 B.
The circumradius of triangle HBC is equal to the circumradius of triangle ABC.

Problem With notation as above, let P be a point on the circumcircle of triangle ABC.
Suppose that A00 is the intersection of lines P A0 and BC, B 00 is the intersection of lines P B 0
and CA, and C 00 is the intersection of lines P C 0 and AB.
Prove that the points A00 , B 00 and C 00 are collinear.

Solution Try drawing the diagram yourself. It looks quite complicated, unless it is large and
multicoloured. A carefully drawn diagram suggests that H also lies on the line through A00 ,
B 00 and C 00 . If this were true, then it suffices to prove that any two of A00 , B 00 , C 00 are collinear
with H. With this tactic in mind let’s try and prove that B 00 , C 00 and H are collinear. For
this, it is enough to prove that ∠BHC 00 = ∠B 0 HB 00 .

B0
A

B 00
C0

H
C 00

A00
B C

A0
P

We know that AB is the perpendicular bisector of C 0 H. Furthermore, C 00 lies on AB. Thus

∠BHC 00 = ∠BC 0 C 00 = ∠BC 0 P .
Similarly, AC is the perpendicular bisector of B 0 H. And B 00 lies on AC. Therefore, we have
∠B 0 HB 00 = ∠HB 0 B 00 = ∠BB 0 P .
68 4 Plane geometry

Finally, we note that ∠BC 0 P = ∠BB 0 P . Thus ∠BHC 00 = ∠B 0 HB 00 which implies B 00 , C 00

and H are collinear.
A similar argument shows that A00 , B 00 and H are collinear. Thus A00 , B 00 , C 00 and H are
collinear and we are done.

Actually, we are not done because we have run into diagram dependence issues. Although the
argument that A00 , B 00 and H are collinear is similar to the argument that B 00 , C 00 and H
are collinear, it is only similarish. There are some differences in the angle chase. Finally, the
whole argument depended on the particular diagram drawn and the position of P relative to
the other points on the circumcircle. All these issues need to be ironed out. See if you can do
it!5
As a final kicker, there is a really short solution to this problem involving Pascal’s theorem.6
See if you can find it.

4.8 Reverse reconstruction

In a geometry problem, the situation you must deal with is fairly concrete. However, just by
varying your logic a little, you can often find different approaches which work well. Sometimes
you assume something slightly different to the problem, and show that actually you have the
situation of the problem. Sometimes you can work backwards from the answer. Here is an
example.

Problem Point O lies inside square ABCD such that ∠OAB = ∠OBA = 15◦ .
Prove that triangle ODC is equilateral.

Solution Although there is a trigonometric approach to this problem, without trigonometry

the problem is difficult to approach directly. So instead, we’ll vary our assumptions by
approaching the problem from the reverse end.

B A B A
75 75
O E

60 60
C D C D

Let E be the point inside ABCD such that EDC is equilateral. Then we have a diagram
which looks the same, but by our assumption we know a different set of angles, and lots of
lengths are equal. We now aim to show that ∠EAB = ∠EBA = 15◦ , so that E and O are
the same point. (This follows since there is only one possible point O inside ABCD satisfying
the conditions ∠OAB = ∠OBA = 15◦ .)
As triangle CDE is equilateral we have CE = CD = CB. So triangle CBE is isosceles.
But since ∠BCE = 30◦ we have ∠CEB = ∠CBE = 75◦ and so ∠EBA = 15◦ . Similarly,
∠EAB = 15◦ , as desired. Therefore O = E and triangle ODC = EDC is equilateral.
5 For help on how to do this you might like to consult section 17.3.
6 See section 6.9.
4.9 Trigonometry 69

4.9 Trigonometry

Sine rule Let ABC be a triangle with side lengths a = BC, b = CA, c = AB and
circumradius R. Then we have
a b c
= = = 2R.
sin A sin B sin C

Problem Let ABC be an acute-angled triangle with circumcentre O and orthocentre H.

If AO = AH, find all possible values of ∠A.

Solution The sine rule automatically gives us the value of AO in terms of the side lengths
and angles of triangle ABC. The natural approach is to use trigonometry to express the
length AH in terms of the side lengths and angles of triangle ABC.
As we discussed earlier in this chapter, any angles involving the orthocentre, the feet of the
altitudes, and the vertices of the triangle can easily be found in terms of ∠A, ∠B and ∠C.
So our task shouldn’t be too difficult!

F
O
H

B D C

Let D, E and F be the feet of the altitudes as per the diagram. Considering triangle AEH,
we obtain
AE AE
= sin ∠C ⇒ AH = .
AH sin ∠C
Considering triangle AEB, we obtain

AE
= cos ∠A ⇒ AE = AB cos ∠A.
AB
Piecing these two pieces of information together and invoking the sine rule yet again, we end
up with
AB cos ∠A
AH = = 2R cos ∠A = 2AO cos ∠A.
sin ∠C
Given that AH = AO, we can cancel this equation to give cos ∠A = 12 , which in turn implies
that ∠A = 60◦ .

It turns out that if the triangle is not restricted to being acute, then there is another value of
∠A for which AH = AO. See if you can find (and prove) what it is.
70 4 Plane geometry

4.10 Areas
One simple approach to geometry problems, often neglected, is to consider areas. This means
more than remembering the formulas for the area of a triangle
1 1
∆= base × height and ∆= ab sin C.
2 2
The following ideas are useful.

Area properties

Two triangles with the same base length and the same height have equal area.
If two similar figures have corresponding lengths in the ratio a : b, then their areas are
in the ratio a2 : b2 .
The following problem is a variation on these ideas.

Problem Let ABCD be a quadrilateral. Let the midpoints of AB, BC, CD and DA be E,
F , G and H, respectively.
Prove that EF GH has half the area of ABCD.

Solution The first thing to notice is that EF GH is a parallelogram. Why? By the midpoint
theorem, both EF and GH are parallel to AC and half its length. Therefore EF k GH and
EF = GH. Similarly F G = HE and F G k HE. This is a commonly used fact. Now the
diagram looks nicer!

D
G
C

H
F

A E B

Since clearly 4AHE ∼ 4ADB with ratio 1 : 2 we have

1
|AHE| = |ADB|.
4
(Here we have used the notation |AHE| to mean the area of polygon AHE.)
Similarly,
1 1 1
|CF G| = |CBD|, |BEF | = |BAC| and |DCH| = |DAC|.
4 4 4

Note that
|ADB| + |CBD| = |BAC| + |DAC| = |ABCD|.
Adding together the equations above gives
2|ABCD| |ABCD|
|AHE| + |CF G| + |BEF | + |DCH| = = .
4 2
4.10 Areas 71

So these four triangles give half the area of ABCD. Now EF GH is what you get if you
remove these four triangles from ABCD. So |EF GH| = 12 |ABCD| as required.

Problem Suppose that each of the three main diagonals AD, BE and CF divide the convex
hexagon ABCDEF into two regions of equal area.
Prove that the three diagonals meet at a common point.

Solution Label the intersection points and lengths along the diagonals as shown. (The
diagram has been drawn assuming that the diagonals don’t meet at a common point.)

D E

d e
I h
C F
c G f
g i
H
b a

B A

We want to show that the three points G, H and I coincide. Our strategy is to prove that
g = h = i = 0.
Each of the main diagonals divides the hexagon into equal pieces of equal area. It follows
that |ABCD| = |DEF A| = |BCDE| = |EF AB| = |CDEF | = |F ABC|.
Now ABCD and BCDE overlap along BCDH, so we have

|ABH| = |DEH|.

Therefore,
1 1
ab sin ∠AHB = (d + g)(e + i) sin ∠DHE,
2 2
and since the angles are equal, this reduces to

ab = (d + g)(e + i).

Considering other pairs of quadrilaterals similarly gives

ef = (b + i)(c + h) and cd = (a + g)(f + h).

Multiplying all three equations together yields

abcdef = (a + g)(b + i)(c + h)(d + g)(e + i)(f + h).

All the quantities involved are non-negative, so this is ridiculous unless we have g = h = i = 0.
Hence AD, BE and CF must meet at a point.

Note that this solution is diagram dependent. There is one other way to draw the diagram.
Specifically H could lie inside quadrilateral CDEF (instead of F ABC as in the diagram).
But this case can be handled in much the same way.
72 4 Plane geometry

4.11 Relate to known diagrams

Most often, a construction is made because it relates one part of the diagram to another. If
nothing extra is drawn, then often a section that is hard to deal with cannot be related to
a well-understood portion. A construction is often the key. Knowing what construction to
make is a matter of insight and experience. The underlying philosophy is always: make it
nicer! If you know you can fit the diagram inside another diagram you have studied before,
draw it in by all means!

Problem Suppose that N is the midpoint of the side BC of triangle ABC. Construct
right isosceles triangles AM B and AP C on sides AB and AC outside the triangle where,
∠AM B = ∠AP C = 90◦ .
Prove that M N P is also a right isosceles triangle.

Solution There is a construction which simplifies this problem enormously. Motivation is

provided by the following considerations.

It’s hard to relate points from one of the isosceles triangles to the other.

We are used to drawing squares, not isosceles triangles, on the sides of triangles, in
some proofs of Pythagoras’ theorem. This is a situation with lots of nice things going
on.

Right isosceles triangles lie nicely in squares.

So, construct squares ABF G and ACDE as shown.

G
A P D

B N C

We see that M and P are the centres of these two squares. In fact, since BM = 12 BG
and BN = 12 BC, triangles BM N and BGC are similar. So, M N = 12 GC, and M N k GC.
Similarly, N P = 12 BE and N P k BE. Thus it suffices to show that GC = BE and GC ⊥ BE.
These facts are some of the nice properties of this diagram that you may already know from
elsewhere. Either way, note that ∠GAC = ∠GAB + ∠BAC = ∠EAC + ∠BAC = ∠BAE
and GA = BA and AC = AE. Therefore, triangles GAC and BAE are congruent. (In fact
one is a rotation of the other about A by 90◦ .) It follows that GC = BE and GC ⊥ BE as
required.
4.12 Create beautiful pictures 73

4.12 Create beautiful pictures

In the end, geometry is one of the greatest art forms known to humankind. And in solving
a geometry problem you are the artist, and you create beautiful pictures. Understand the
picture; move the pieces around and see what happens; help the picture explain itself; it is,
after all, confused, just like us, and seeks illumination.

Problem Let ABD be a triangle and let C be a point on the side BD, lying strictly between
B and D. Suppose that BC = 2CD, ∠ACB = 60◦ and ∠ADC = 45◦ .
Determine ∠BAD.

Solution Simple techniques will not suffice here. Trigonometry can provide a solution but
we’ll avoid it, and take the artistic approach.
We get the feeling that there might be a construction which illuminates the diagram. We
hope to discover something we didn’t know before about the diagram. For instance, we might
discover that some point is a centroid, or incentre, or another well-known point. We might
discover a cyclic quadrilateral, a tangent line, or something else useful.
Since DC : CB = 1 : 2, it might be worthwhile to make a construction so that C will be a
centroid of some triangle. But that does not seem to work, and so we will instead try to
make C the incentre of a triangle. Incentres, being the intersection of angle bisectors, always
give us lots of information about angles, and information about ratios, which is what we want.
So, paint a picture where C is the incentre of some triangle. A little angle chasing gives
∠DAC = 15◦ . So construct a point E such that ∠DAE = 30◦ and ∠ADE = 90◦ . Then C is
the incentre of triangle AED. Furthermore, AED is a 30◦ -60◦ -90◦ triangle. Label points as
shown with H being the intersection of EC and AD. Note that EHD is also a 30◦ -60◦ -90◦
triangle.

15

60 45
B C D

By the angle bisector theorem we have

AH AE
= = 2.
HD DE
74 4 Plane geometry

We were also given

BC
= 2.
CD
Hence triangles HCD and ABD are similar and therefore, ∠BAD = 60◦ .
 Important configurations in geometry
5
Solving problems in geometry very often involves recognising that a diagram contains a known
configuration within it. What follows is a partial list of such configurations, a number of
which you have already seen in this book and others that you will yet see. You should, in
time, be able to prove them all. However, the art is in learning to recognise the configuration
as part of a more complex diagram.
Some of these are well-known theorems and are quotable. Others are not so well known.
In most cases it is permissible to quote the result. However, just making use of the result
without explicitly saying that it is a known result has caused many students to lose up to
three marks at the IMO.
The configurations are organised into three categories.

A-List Extremely useful

B-List Very useful
C-List Useful

Each configuration is accompanied by one, two or three stars. This is an estimated rating
of how difficult the result is to prove using geometric methods.1 More stars mean that it is
more difficult to prove.
The configurations are purposely kept minimal since the idea is to be able to recognise the
minimal elements needed to infer something. The resulting incidence is generally indicated by
fattening and colouring the points or curves involved. Thus four fat points usually indicate
that the four points are concyclic and so on.
This chapter is clearly very different to all the other chapters. Try to become very familiar
with it!

1 That is, without resorting to grubby computational methods such as trigonometry, complex numbers or
coordinate geometry!
76 5 Important configurations in geometry

A1 Angle bisector and perpendicular bisector F

6 AC
AB = ⇒ ABCD is cyclic.
AB = AC ⇒ ABCD is a kite.

A
•
α α

B• •C
a a
•
D
 77

A2 Pivot theorem F

•
•
•

• • •

Variations: lines extended

•
•
•

•
• • •

Variations: tangents2

2 Inboth these variations, two points are allowed to coalesce into a single point resulting in a tangent incidence
rather than two distinct intersection points.
78 5 Important configurations in geometry

A3 Radical axis theorem F

•
•
•
• •
• •

Variations

Zero radius circle(s)

Some circles tangent
 79

A4 Similar switch F

4AP B ∼ 4CP D ⇔ 4AP C ∼ 4BP D.

B B
C C

D D

A A
• •
P P

Equivalent formulation via spiral symmetries

There is a spiral symmetry about P which sends segment AB to segment CD if and only if
there is a spiral symmetry about P which sends segment AC to segment BD.

Additional information: presence of two circles

The above configuration is often seen as a byproduct of two intersecting circles. Point P is
the centre of each spiral symmetry.

B
C

A
•
P

Additional information: presence of two more circles

• •

• •
•
• •
•

(Compare this with the four lines and four circles diagram in the C-List.)
80 5 Important configurations in geometry

B1 Radical axis bisects common tangent F

x
• x

Variation: tangent circles

•
 81

B2 Perpendicularity F

AC ⊥ BD ⇔ AB 2 + CD2 = AD2 + BC 2 .

Convex quadrilateral

B • D

Triangle

B
• D
C

Non-convex quadrilateral

A A

D C
B

C B D
82 5 Important configurations in geometry

B3 Alternate segment switch F

 83

B4 Ratios for collinearity F

(Points A, R, K are not permitted to be collinear for this diagram.)
AB KL
R, S, T collinear ⇔ = .
BC LM

•R •S •
T

K L M
84 5 Important configurations in geometry

B5 Points of contact of incircle and excircle F

• x
• • x
•
 85

B6 Circumcircle, incentre and excentre F

I = incentre
Ia = excentre
DB = DC = DI = DIa .

I
•

• •
B C

•
D

• Ia
86 5 Important configurations in geometry

B7 Simson line F

•
• •

• • •

Additional property FFF

H = orthocentre
Simson line bisects P H.

x
•
x
.
H
 87

B8 Pascal’s theorem FFF

• • •

Two points coincident

•
•
•

Pappus’ theorem

• •
•
88 5 Important configurations in geometry

B9 Desargues’ theorem FFF

Triangles are in perspective from a point if and only if they are in perspective from a line.

•
•

•
 89

B10 Quadrilateral and incircle F

Convex quadrilateral ABCD has an incircle ⇔ AB + CD = BC + AD.

D
B

Variations

AB + CD = BC + AD
B D

D B AB − CD = BC − AD

A
D

AB − CD = BC − AD

B
C
90 5 Important configurations in geometry

C1 Nine-point circle F
Midpoints of sides, feet of altitudes and midpoints of segments connecting vertices to ortho-
centre are all concyclic.

•
•
• •

•
• •
• •
 91

C2 Euler line F

H = orthocentre
N = nine-point centre
G = centroid
O = circumcentre

O
N G •
H • • 2
• 3
1
92 5 Important configurations in geometry

C3 Four lines and four circles F

•
 93

C4 Newton–Gauss line FF
Midpoints of the diagonals of a complete quadrilateral3 are collinear.

3A complete quadrilateral is the figure determined by four lines, no three of which are concurrent, and their
six points of intersection. The three pairs of points which are not already connected by the original four
lines determine the diagonals of the complete quadrilateral.
94 5 Important configurations in geometry

C5 Alternative characterisation of symmedian4 FF

Another characterisation of symmedian via tangent circles FF

4 The symmedian is the reflection of the median about the angle bisector.
 95

C6 Convex cyclic hexagon and diagonals F

Diagonals concurrent ⇔ AB · CD · EF = BC · DE · F A.

B D

•
E
A

F
96 5 Important configurations in geometry

C7 Quadrilateral, triangles and incircles F

•
if and only if
 97

C8 Median, inradius and chord of incircle FF

•
•

| {z }| {z }
m m

Variation: excentre

m m
z }| { z }| {

•
98 5 Important configurations in geometry

C9 Incentre and midpoints FF

M = midpoint BC
X = midpoint AD

X
•
•I

•
B D M C

Variation: excentre

M = midpoint BC
Y = midpoint ADa

Y•

M
B • C
Da

• Ia
 99

C10 Incentre, excentre, midpoint and contact points FF

M = midpoint BC
D = incircle contact point
Da = excircle contact point

A A

B M Da C B D M C

Ia
100 5 Important configurations in geometry

C11 Incentre and chord of incircle FF

• I
•

Variation: incentre and excentre

• •
I
•

• •

• •

•
Ia
 101

C12 Harmonic quadrilateral5 FFF

P , B, D collinear ⇔ AB · CD = AD · BC.

P
B
•
•
D•

5A cyclic quadrilateral is said to be harmonic if the products of its opposite sides are equal.
102 5 Important configurations in geometry

C13 Incentre and mixtilinear incircle6 FFF

I •
•
•

Variation: excentre and mixtilinear excircle

•
I
•a
•

6A mixtilinear incircle is a circle which is internally tangent to a triangle’s circumcircle and two of its sides.
There are three mixtilinear incircles associated with any triangle. A mixtilinear excircle is similar but is
instead externally tangent to the triangle’s circumcircle.
 103

C14 Butterfly theorem FFF

x x
|
• {z
•
}| {z
• }
m m
 Incidence geometry
6
An incidence basically describes a situation when something more specific occurs than might
otherwise be thought. For example, three lines are usually not concurrent, but sometimes
you may be asked to prove that three lines are concurrent. So ‘three lines concurrent’ is an
example of an incidence. Other examples include ‘three points collinear’ and ‘four points
concyclic’.
Closely related to incidences is the concept of locus. For our purposes, a locus is a set of
points satisfying some condition. We often think of a locus as the curve traced out by a
point following some rule. For example, a circle may be thought of as the locus of points
equidistant from a given point. A line may be thought of as the locus of points equidistant
from a given line and lying on a given side of the given line.
When asked to find the locus of a set of points it is important to know beforehand what the
answer is likely to be. This can be done by drawing a careful diagram and building up a
picture of the locus. Most loci tend to be lines or circles, but sometimes ellipses or even whole
regions can occur.
An example might be to find the locus of points P such that ∠AP B = 45◦ , where A and B
are given fixed points. We know that the angle subtended in a circle on one side of a chord
is constant, so we can see that the locus is the union of two circular arcs each of which is
three-quarters of a full circle.

6.0 Problems
1. Show that the orthocentre and the circumcentre of a triangle are isogonal conjugates of
each other.1

2. (a) Prove that the medians of a triangle are concurrent.

(b) Prove that the altitudes of a triangle are concurrent.
(c) Prove that the angle bisectors of a triangle are concurrent.
(d) Prove that the lines joining the vertices of a triangle with the points of tangency
of the incircle with the opposite sides are concurrent.
1 See the isogonal conjugates theorem in section 6.3 if you don’t know what isogonal conjugates are.
106 6 Incidence geometry

3. Find the locus of points P such that for fixed points A and B,
AP
= k,
BP
where k is a positive constant.
4. Let A, B, C and D be four distinct points on a line, in that order. The circles with
diameters AC and BD intersect at the points X and Y . The line XY meets BC at
the point Z. Let P be a point on the line XY different from Z. The line CP intersects
the circle with diameter AC at the points C and M , and the line BP intersects the
circle with diameter BD at the points B and N .
Prove that the lines AM , DN and XY are concurrent.
5. The segment AB is fixed and point M is a variable point on that segment. Squares SA
and SB are constructed on AM and M B and on the same side of AB.
Find the locus of the midpoint of the segment joining the centres of the two squares.
6. Let circles C1 and C2 intersect at A and B. The tangent to C1 at A meets C2 again
at P . The tangent to C2 at A meets C1 again at Q. Let M be the point on line AB
such that AB = BM .
Prove that A, P , M and Q are concyclic.
7. Let A be a fixed point on a fixed circle Γ. Let B and C be variable points on Γ and
AD 2
let D be a point on BC such that BD·DC is a fixed constant.
Find the locus of D as B and C vary over Γ.
8. Of all triangles with given base length and height, which one has the largest inradius?
9. Points A, B and C range over the circumference of a fixed circle.
Find the locus of the incentre of triangle ABC.
10. Let K, L, M and N be four collinear points on the possibly extended sides AB, BC,
CD and DA, respectively, of quadrilateral ABCD.
Prove that
AK BL CM DN
· · · = +1,
KB LC M D N A
where this is an equation in directed lengths.2
11. Let ABCDEF be a cyclic convex hexagon with vertices labelled clockwise.
Prove that
AB · CD · EF = BC · DE · F A,
if and only if the diagonals AD, BE and CF are concurrent.
12. Let ABCD be a cyclic quadrilateral. Let A1 and C1 be the respective feet of the
perpendiculars from A and C to BD, and let B1 and D1 be the respective feet of the
perpendiculars from B and D to AC.
Prove that A1 B1 C1 D1 is cyclic.
13. Let ABCD be a quadrilateral whose diagonals AC and BD are perpendicular. Let P ,
Q, R and S be the midpoints of the sides, and let T , U , V and W be the feet of the
altitudes from these midpoints to the opposite sides.
Prove that P , Q, R, S, T , U , V and W all lie on a circle.
2 By AK
this we mean that KB
is positive if K is between A and B, and negative otherwise.
6.0 Problems 107

14. Triangle ABC has sides which satisfy

2AB = AC + BC.

Prove that the midpoints of AC and BC and the incentre and circumcentre of triangle
ABC are concyclic.
15. Let M be the midpoint of the side AC of a triangle ABC and let H be the foot of
the altitude from B. Let P and Q be the orthogonal projections of A and C onto the
bisector of the angle at B.
Prove that the four points H, P , M and Q lie on the same circle.

16. Show that the tangents to the circumcircle of a triangle at two of its vertices meet on
the symmedian from the third vertex.
17. In an acute-angled triangle ABC let AD and BE be altitudes and let AP and BQ be
internal angle bisectors. Denote by I and O the incentre and the circumcentre of ABC,
respectively.
Prove that D, E and I are collinear if and only if P , Q and O are collinear.
18. Let ABC be an acute-angled triangle and let D, E and F be the feet of the perpendiculars
from A, B and C onto the sides BC, CA and AB, respectively. Let P , Q and R be
the feet of the perpendiculars from A, B and C onto the lines EF , F D and DE,
respectively.
Prove that the lines AP , BQ and CR are concurrent.
19. Let ABCD be a quadrilateral such that its opposite sides AB and CD meet at P and
its other opposite sides AD and BC meet at Q. Let X be the intersection of P Q and
CA. Let Y be the intersection of P Q and DB.
Show that
P X · QY = P Y · XQ.

20. Let ABCD be a given convex quadrilateral with sides BC and AD equal in length and
not parallel. Let E and F be variable points on the sides BC and AD, respectively,
and which satisfy BE = DF . The lines AC and BD meet at P , the lines BD and EF
meet at Q and the lines EF and AC meet at R. Consider all triangles P QR as E and
F vary as mentioned earlier.
Show that the circumcircles of all these triangles have a common point other than P .

21. Let I be the incentre of triangle ABC. A circle which is tangent to the circumcircle of
triangle ABC on the inside also touches CA and BC at D and E, respectively.
Show that I is the midpoint of DE.
22. Let ABCD be a convex quadrilateral that just happens to have both an incentre I and
a circumcentre O.3
Prove that the intersection of the diagonals of the quadrilateral lies on the line OI.

3 Such a quadrilateral is called bicentric.

108 6 Incidence geometry

6.1 Collinear points

One way to prove that three points A, B, C are collinear is to prove that ∠ABC = 0◦ or 180◦ .

Problem Let ABC and ADE be similar triangles whose vertices are labelled clockwise. Let
P be the second common point of the circumcircles of the triangles besides A.
Show that P must lie on the line connecting B and D.

Solution

B A

Look at the diagram. Equal angles abound and we have the chain of equalities

∠BP A = ∠BCA = ∠DEA = ∠DP A.

The first follows from the cyclic quadrilateral ABCP , the second follows from the similar
triangles ABC and ADE, while the third follows from the cyclic quadrilateral ADEP . But
seeing that B and D lie on the same side of the line AP , the equality ∠BP A = ∠DP A tells
us that P must lie on the line passing through B and D.

Are we done? No, we certainly are not! This is an opportune time to mention a common
pitfall in geometry known as diagram dependence. We only solved the problem for the diagram
shown. It is possible to have other diagrams where the relative positions of the points are
different, and our angle chase is a bit different. For instance, if triangle ADE were rotated
clockwise until D lay on ray AP beyond P , then it is no longer true that ∠DEA = ∠DP A,
but instead we would have ∠DEA = 180◦ − ∠DP A. See if you can identify all the different
configurations possible and solve in each case.4
Three other points in the diagram turn out to be collinear. What are they? Prove it!

Problem Let P be a point on the circumcircle of triangle ABC. Let D, E and F be the
feet of the perpendiculars from P to the lines BC, AC and AB, respectively.
Show that D, E and F are collinear.
4 You might like to consult section 17.3 for a possible way to deal with this without resorting to a large number
of case distinctions.
6.2 Menelaus’ theorem 109

Solution

A P

B D C

Since BAF is a straight line, it suffices to show that ∠BF D = ∠AF E. From cyclic
quadrilaterals P F BD and P F AE we know that ∠BF D = ∠BP D and ∠AF E = ∠AP E. If
we subtract ∠BP E from both these angles, then it suffices to prove that ∠EP D = ∠AP B.
However, both these angles are equal to ∠BCA—the first is because CDEP is cyclic, and
the second is because BAP C is cyclic.

This solution is not complete since we have not addressed issues of diagram dependence. We
leave this for the reader to complete.
The result from the preceding problem can be extended. It is known as Simson’s theorem.

Simson’s theorem Let P be any point in the plane of triangle ABC. Then the feet of
the perpendiculars from P to the (possibly extended) sides of the triangle are collinear if
and only if P lies on the circumcircle of the triangle. In such a case the line containing the
three feet of the perpendiculars is called the Simson line.

6.2 Menelaus’ theorem

Menelaus’ theorem can be used to prove collinearity of points.

Menelaus’ theorem If X, Y and Z lie on the three (possibly extended) sides BC, AC
and AB of a triangle ABC, then the three points X, Y and Z are collinear if and only if
AZ BX CY
· · = −1,
ZB XC Y A
where the segments are considered to have directed length.

The part about directed lengths in the statement of Menelaus’ theorem simply means that
−→ −→
AZ
the ratios take into account the directions of the vectors AZ, ZB, and so forth. Thus ZB is
a positive ratio if Z lies on segment AB, and is a negative ratio otherwise.
110 6 Incidence geometry

B C X

Problem Suppose that ABC is a triangle with circumcircle Γ in which the three tangents
to Γ at A, B and C meet the three opposite sides at X, Y and Z, respectively.
Prove that X, Y and Z are collinear.

Solution

Y
A

X B C

First, triangles XAB and XCA are similar. This follows from the alternate segment theorem,
which asserts that ∠XAB = ∠BCA. Thus we may write
XA XB AB
= = .
XC XA AC
Combining some of these equalities yields

BX AB 2
=− .
XC AC 2
We compute similar expressions for the other two ratios. These all multiply together and
cancel out to give −1. Thus X, Y and Z are collinear by Menelaus’ theorem.

6.3 Concurrent lines

One way of showing that three lines k, `, m are concurrent is to show that the point of
intersection of two of the lines lies on the third.
6.4 Ceva’s theorem 111

Problem Let P be a point in the plane of a triangle ABC. Reflect the lines P A, P B and
P C through the angle bisectors at A, B and C, respectively.
Prove that these three reflected lines are concurrent.

Solution Let pa , pb and pc denote the distances from P to the lines BC, CA and AB,
respectively. Similarly, for any point Q in the plane of triangle ABC, we let qa , qb and qc
denote the distances from Q to the lines BC, CA and AB, respectively.
Observe that Q lies on the line obtained by reflecting the line P A through the angle bisector
at A if and only if
qb pc
= .
qc pb

P Q
B C

(Can you locate the appropriate similar triangles to establish why this is true?)
Next, define Q to be the point which lies on the reflected line through A and the reflected
line through B. Therefore, the previous observation gives us the two equations
qb pc qc pa
= and = .
qc pb qa pc

Multiplying these two equations together, we obtain

qb pa
= ,
qa pb

which implies that Q lies on the reflected line through C as well. Therefore, the three reflected
lines are concurrent at Q.

The result of this problem is a known theorem.

Isogonal conjugates theorem If `A is a cevian through A of triangle ABC, the isogonal

conjugate of `A is found by reflecting `A in the angle bisector at A to get the line `0A .
The theorem states that three cevians `A , `B and `C are concurrent if and only if their
isogonal conjugates `0A , `0B and `0C are concurrent. The respective points of concurrency are
called isogonal conjugate points.

6.4 Ceva’s theorem

Ceva’s theorem is a highly useful criterion often used to prove that three lines concur.
112 6 Incidence geometry

Ceva’s theorem If X, Y and Z lie on the three (possibly extended) sides BC, AC and
AB of a triangle ABC, then the three lines (called cevians) AX, BY and CZ are concurrent
if and only if
AZ BX CY
· · = +1,
ZB XC Y A
where the segments are considered to have directed length.

Note that this is exactly the same expression as for Menelaus’ theorem except that we have +1
on the right-hand side instead of −1.

Problem Outside triangle ABC, points K, L and M are constructed in such a way that

∠M AB = ∠LAC, ∠KBC = ∠M BA and ∠LCA = ∠KCB.

Prove that the three lines AK, BL and CM are concurrent.

Solution Let x = ∠M AB, y = ∠KBC and z = ∠LCA and let AK, BL and CM intersect
BC, CA and AB at points X, Y and Z, respectively, as in the diagram.

A
x
x α
L

Y
M
Z

y z
β γ
B y X z C

By Ceva’s theorem it would suffice to prove that

AZ BX CY
· · = +1.
ZB XC Y A

Let P denote this product. Note that

BX |4ABX| |4KBX|
= = .
XC |4ACX| |4KCX|

Using addendo5 we have

1
BX |4ABK| 2 AB · BK sin(β + y)
= = 1 .
XC |4ACK| 2 AC · CK sin(γ + z)
5 Addendo a c a+c
is a simple yet highly useful little result. If r = b
= d
, then also r = b+d
.
6.5 Concyclic points 113

We obtain similar expressions for the other two ratios and thus compute after much cancelling
out that
KB LC M A
P = · · .
KC LA M B
Finally, we use the sine rule in triangle KBC to find
KB sin z
= .
KC sin y
We obtain similar expressions for the other two ratios so that we finally compute that P = 1.
Therefore, AX, BY and CZ are concurrent by Ceva’s theorem.

There is also a trigonometric version of Ceva’s theorem.

Trigonometric form of Ceva’s theorem If angles are marked as in the figure, then the
cevians are concurrent if and only if
sin α1 sin β1 sin γ1
· · = +1.
sin α2 sin β2 sin γ2

α2
α1

γ2 β1
γ1 θ β2
C X B

The proof of this is quite straightforward and may be carried out by using the sine rule six
times. For example,
CX AC XB AB
= and =
sin α1 sin θ sin α2 sin(180◦ − θ)
and so we obtain equations such as
sin α1 AB CX
= · .
sin α2 AC XB

6.5 Concyclic points

How can you show that four points A, B, C and D are concyclic?

Find a point O such that AO = BO = CO = DO.

Show that ∠ABC + ∠ADC = 180◦ and that C and D lie on opposite sides of the
line AB.
Show that ∠ACB = ∠ADB and that C and D lie on the same side of the line AB.
114 6 Incidence geometry

Problem Let ABC be a triangle with altitudes AD, BE, CF , medians AX, BY , CZ, and
orthocentre H. Let A0 , B 0 and C 0 be the midpoints of AH, BH and CH, respectively.
Prove that the nine points A0 , B 0 , C 0 , D, E, F , X, Y and Z all lie on a circle.6

Solution

A0
E

Z Y

B0 C0

B D X C

There are midpoints galore in this problem. In fact, six of the nine points which we will prove
are concyclic are defined as midpoints. Therefore, it seems like a prime opportunity to use
the midpoint theorem, which states that if Z is the midpoint of AB and Y is the midpoint of
AC, then Y Z is parallel to BC and half its length. Applied in triangle ABH, we obtain that
B 0 Z is parallel to AH, while applied in triangle ACH, we obtain that C 0 Y is also parallel to
AH. Applied in triangle ABC, we obtain that Y Z is parallel to BC, while applied in triangle
HBC, we obtain that B 0 C 0 is parallel to BC.
In summary, B 0 Z and C 0 Y are parallel to each other and to AH. Furthermore, Y Z and B 0 C 0
are parallel to each other and to BC. However, since AH is perpendicular to BC, B 0 C 0 Y Z
must be a rectangle. Similar arguments lead to the fact that C 0 A0 ZX and A0 B 0 XY are also
rectangles.
Therefore, if we let N be the midpoint of B 0 Y , then N is the centre of the circle circumscribing
the rectangle B 0 C 0 Y Z, as well as the centre of the circle circumscribing A0 B 0 XY , both of
which have B 0 Y as diameters. It follows that A0 , B 0 , C 0 , X, Y and Z all lie on a circle.
Now note that the line Y Z bisects AD and is perpendicular to it. In other words, the reflection
of A in the line Y Z is the point D. To paraphrase again, triangle AY Z is congruent to
triangle DY Z. However, triangle AY Z is also congruent to triangle XZY . In particular, we
have the equal angles ∠Y DZ = ∠Y XZ, so that the quadrilateral XY ZD is cyclic. Therefore,
the point D—and by a similar argument, the points E and F —lie on the circumcircle of
triangle XY Z. It follows that A0 , B 0 , C 0 , D, E, F , X, Y and Z all lie on a circle.

Here is an alternative solution which illustrates that there are many different ways to complete
an angle chase. It is also highlights the ‘one step at a time’ method described in section 4.3.
6 For obvious reasons, this circle is called the nine-point circle of triangle ABC.
6.5 Concyclic points 115

Solution Draw the triangle with only the extra points X, Y , Z and D marked.

Z Y

B D X C

Thus
∠Y DC = ∠Y CD = ∠XZY.
The first equality comes from the fact that triangle ADC is right-angled at D. Hence its
circumcentre is the midpoint of AC, namely Y , and so Y A = Y D = Y C. The second equality
comes from the fact that XZY C is a parallelogram, and thus has opposite angles equal.
Hence ∠Y DX = ∠Y DC = ∠XZY , which establishes that D lies on circle XY Z. Similarly,
points E and F lie on circle XY Z. Thus the six points X, Y , Z, D, E and F all lie on the
same circle.
Now draw the triangle with only the extra points D, E, F , A0 and H marked.

A0 E

F
H

B D C

The circle with diameter AH passes through points E and F due to the right angles at E
and F . Thus A0 is the centre of circle EAF H, and so ∠EA0 F = 2∠A.
Furthermore, ∠HDF = ∠HBF = 90◦ − ∠A from cyclic quadrilateral HDBF . Similarly,
∠HDE = ∠HCE = 90◦ − ∠A. Therefore,

∠EDF = ∠HDF + ∠HDE = 180◦ − 2∠A = 180◦ − ∠EAF.

This means that A0 lies on circle DEF . Similarly, points B 0 and C 0 also lie on circle DEF .
Since the circle through D, E and F is unique and contains the points X, Y , Z, A0 , B 0 and
C 0 , we conclude that all nine points lie on this same circle.
116 6 Incidence geometry

Problem Let D, E and F be points on the sides BC, CA and AB of triangle ABC.
Prove that the circumcircles of triangles AEF , DBF and DEC are concurrent.

Solution
A

P
F

B D C

Let the circumcircles of triangles AEF and DEC meet at P . Then since the quadrilaterals
AF P E and DP EC are cyclic, we have

∠BF P = 180◦ − ∠AF P

= ∠AEP
= 180◦ − ∠CEP
= ∠CDP
= 180◦ − ∠BDP.

Since ∠BF P + ∠BDP = 180◦ , the quadrilateral BF P D is also cyclic. Therefore, the
circumcircle of triangle DBF also passes through P .7

The result of this problem can be generalised as follows.

Generalised pivot theorem Let A, B, C, D, E, F and P be any seven distinct points

in the plane. Consider the following six statements.

D lies on line BC.

E lies on line AC.
F lies on line AB.
P lies on circle AEF .

P lies on circle BDF .

P lies on circle CDE.
If any five of these statements are true, then so is the sixth.

This is an extremely useful result and is well worth remembering. Many geometry problems
happen to include this set-up lurking as a subdiagram.
7 Thisargument is diagram dependent because it did not address the possibility of P lying outside of the
triangle. See section 17.3 to see how to deal with this and more.
6.6 Power of a point 117

6.6 Power of a point

Another useful condition for concyclicity is via the power of a point theorem. This is also
known as the intersecting chords theorem.

Power of a point theorem If lines AB and CD intersect at P , then A, B, C and D are

concyclic if and only if
P A · P B = P C · P D,
where the lengths are considered to be directed.

In addition, we define the power of a point P with respect to a circle Γ to be the real number
P A · P B, where P , A and B are collinear and A and B are points on Γ. We treat the lengths
as being directed so that the power is positive if P is outside Γ, and negative if P is inside Γ.
Note that the power of a point theorem ensures that this is a well-defined real number. That
is, we get the same result no matter how A and B are chosen on Γ, provided that P , A and
B are collinear.

Problem Two circles Γ1 and Γ2 intersect in two points A and B. Point P lies on the line
AB. A line passing through P intersects Γ1 at U and V . Another line passing through P
intersects Γ2 at X and Y .
Prove that the four points U , V , X and Y are concyclic.

Solution

U X
A

Γ1 Γ2
P

Y V

Using power of a point on Γ1 we have

P U · P V = P A · P B.

Using power of a point on Γ2 we have

P X · P Y = P A · P B.

Thus
P U · P V = P X · P Y,

and so using the power of a point theorem again we deduce that U , V , X and Y are
concyclic.
118 6 Incidence geometry

6.7 Radical axes

Sometimes locus considerations help us to establish incidences as in the following.
We define the radical axis of two circles to be the set (locus) of points which have equal power
(see section 6.6) with respect to both circles.

Theorem The radical axis of a pair of circles is a straight line.

In the case that the two circles intersect, then you can probably see that their radical axis
is simply the line joining their two points of intersection. However, the theorem is still true
even if the circles don’t intersect.
The radical axis concept is very important, especially in light of the following.

Radical axis theorem Given three circles, the three radical axes associated with the
three pairs of circles are either concurrent or parallel.

Problem Prove the radical axis theorem.

Solution Let the three circles be Γ1 , Γ2 and Γ3 . Let λ be the radical axis of Γ1 and Γ2 and
let µ be the radical axis of Γ2 and Γ3 . Suppose that λ and µ intersect at P .8
Since P lies on λ, it has equal power with respect to Γ1 and Γ2 . Similarly, since P lies on µ,
it has equal power with respect to Γ2 and Γ3 . It follows that P has equal power with respect
to Γ1 and Γ3 and consequently lies on the radical axis of Γ1 and Γ3 .
Thus the three radical axes all pass through P .

The concurrence of three radical axes is highly useful. Look out for them!

Problem A line parallel to the side BC of triangle ABC meets AB at F and AC at E.

Prove that A lies on the common chord of the circles whose diameters are BE and CF .

Solution
A

X
P
•
W
• γ
β
F E

• •
B C

8 If λ and µ are parallel, then P is undefined and so this argument is faulty. Can you deal with this case?
6.8 Ellipses 119

Let β be the circle with diameter BE and let γ be the circle with diameter CF . Let P and Q
be the two points where β and γ intersect. We want to show that line P Q passes through A.
We have two circles, and their common chord which is supposed to pass through A. Note
that there are two other obvious lines, namely AB and AC, also passing through A. This
looks like a candidate for the radical axis theorem!
We seek a third circle whose radical axes with β and γ are the lines AB and AC, respectively.
There is only one such candidate. It should pass through the intersection, W say, of line AB
and β and it should also pass through the intersection, X say, of line AC and γ. If we can
show that BW XC is cyclic, then applying the radical axis theorem to circles BW XC, β and
γ will tell us that BW , CX and P Q are concurrent. Since BW and CX intersect at A, this
means that P Q would also pass through A.
Thus it remains to show that BW XC is cyclic. Since BE is a diameter of β, we know that
∠F W E = ∠BW E = 90◦ . Thus W lies on the circle with diameter F E. Similarly X lies on
the circle with diameter F E. So F W XE is a cyclic quadrilateral.
Therefore,

∠AW X = ∠XEF (F W EX cyclic)

= ∠ECB (F E k BC).

Thus ∠AW X = ∠XCB implying that BW XC is cyclic as desired.

If you look at an accurately drawn diagram, it seems that W E, F X and P Q are also
concurrent. This is the same as saying the orthocentre of triangle AEF also lies on line P Q.
See if you can prove it! You only need the radical axis theorem applied once more. But which
circles should you apply it to?

6.8 Ellipses
Ellipses have certain highly useful properties. For example, an ellipse can be thought of as
the locus of points P such that the sum P A + P B is a given constant for fixed points A and
B. The points A, B are called the foci. These two points have a nice optical property: if
the inside boundary of an ellipse is made of reflective material, then any light emitted from
one focus, A say, will always pass through the other focus B. Thus if P is any point on the
ellipse and ` is a tangent at P , then the angle that AP makes with ` equals the angle that
BP makes with `, that is, ‘angle of incidence equals angle of reflection’.

. .
A B
120 6 Incidence geometry

Problem Let ABC be an acute triangle.

(a) For which point(s) P in the plane is the sum P A + P B + P C minimal?9

(b) Show how to construct such a point using a straightedge and compass.

Solution

(a) For any point P in the plane write f (P ) = P A + P B + P C. We wish to find the
minimum value for f (P ).
First we investigate the possibility that P might lie outside the triangle. The outside of
the triangle can be divided into six regions defined by the lines AB, AC and BC and
P could lie in any of these.

I
C
II
VI

V B IV
A III

Case 1: The point P lies in one of the regions I, III or V.

WLOG P lies in region I.

P .
C

B A

Consider the ellipse with foci at A and B such that P is on the ellipse. Since
triangle P AB lies inside the ellipse and C lies inside triangle P AB, we conclude
that C lies strictly inside the ellipse. Consequently, we have AP + BP > AC + BC.
Therefore,
f (P ) = AP + BP + CP > AP + BP > AC + BC = f (C).
Case 2: The point P lies in one of the regions II, IV or VI.
WLOG P lies in region II.

C .P
Q

B A
9 This problem is considered again in section 12.6.
6.8 Ellipses 121

Consider the point Q which is the intersection of segments AC and BP . Then

with the help of the triangle inequality AP + CP > AC, we have

f (P ) = AP + CP + BP > AC + BP > AC + BQ = AQ + CQ + BQ = f (Q).

So in both cases we have shown that for any point P lying outside the triangle, there
exists another point Q on the boundary of the triangle such that f (P ) > f (Q). So from
here on we may restrict our attention to points not lying outside the triangle.
Consider any position of a point P inside or on the boundary of triangle ABC such that
f (P ) = P A + P B + P C is minimal.10 If P were on the boundary of the triangle, it is
easy to see that P would have to be a foot of an altitude of the triangle. Consequently,
P cannot be at a vertex.
Consider the ellipse ε with foci at A and B and such that P is on ε. Consider also the
circle γ centred at C passing through P .

γ
C

P
Q
ε

B A

If γ is not tangent to ε, then it intersects ε in at least two points. In this case, any point
Q lying strictly inside the intersection of γ and ε would necessarily have CQ < CP and
AQ + BQ < AP + BP and so f (Q) < f (P ), a contradiction. (Even if Q lay outside the
triangle, this would still be a contradiction, because f (P ) was taken to be the global
minimum.)
Therefore, γ is tangent to ε at P . Let ` be the common tangent of γ and ε at P . Since
we know the angle of incidence is equal to the angle of reflection at P in ε and we
also know that ` ⊥ P C, this allows us to deduce that ∠AP C = ∠BP C. Similarly, we
deduce that ∠BP A = ∠CP A.
Hence all three angles around P are equal. Therefore, P is a point where all three
angles are equal to 120◦ . 
10 Aminimal value exists because f is a continuous function whose domain is now restricted to a closed and
bounded subset of the plane.
122 6 Incidence geometry

(b) We construct the point as follows. Erect equilateral triangles and their circumcircles
outwards on AB and BC. These circles both subtend 120◦ angles on AB and BC inside
triangle ABC and hence their intersection point is the desired point P . 

6.9 Pascal’s theorem

Here is another useful tool for proving collinearity or concurrence.

Pascal’s theorem Let A, B, C, D, E and F be any six points on any conic section. Let
X = AB ∩ DE, Y = BC ∩ EF and Z = CD ∩ F A. Then X, Y and Z are collinear.

Note that the theorem is true no matter what order the six points appear on the conic. The
theorem is also true in its limiting cases. For example, if B is permitted to approach and
then coincide with A, then the line AB becomes the tangent at A. In the context of using
Pascal’s theorem, the tangent at A is often written as AA. Some illustrative diagrams can be
found on page 87.
In most applications the conic section11 will be a circle. Sometimes it might be a pair of
lines.12
A useful way to remember what intersects what is that AB, DE are pairs of opposite sides of
the hexagon ABCDEF , as are BC, EF and CD, F A.

Problem Let ABC be an acute triangle with circumcentre O. Let AA1 be a diameter of
the circumcircle of triangle ABC. The tangent line to the circumcircle at A1 intersects the
line BC at point D. The line OD intersects the sides AB and AC of the triangle at points P
and Q, respectively.
Prove that OP = OQ.

Solution

.O

B
D A1

Since OA = OA1 , the problem is equivalent to proving that AQA1 P is a parallelogram. To

this end it suffices to prove that P A1 k AC. A similar argument will show that QA1 k AP .
11 Conic sections include circles, ellipses, hyperbolas, parabolas and pairs of straight lines.
12 This special case of Pascal’s theorem is known as Pappus’ theorem.
6.9 Pascal’s theorem 123

Thus we may now forget about point Q.

Let X be the point where A1 P meets the circumcircle for a second time. Note that proving
P A1 k AC is now equivalent to proving that ACA1 X is a rectangle or equivalently that CX
passes through point O. Let K be the intersection of AA1 and CX. We shall prove that
K = O.

•K
P •

B
•
D A1

It would suffice to prove that K also lies on DP for then K would lie on both DP and AA1 ,
ensuring that K = O. This is where Pascal’s theorem makes an entry. Indeed K = CX ∩ AA1 ,
D = A1 A1 ∩ BC and P = AB ∩ A1 X. It only remains to see if there is a Pascal configuration
corresponding to these lines and points.
The Pascal configuration virtually reveals itself! A1 A1 is a side (corresponding to the tangent
of the circle at A1 ), A1 X and AA1 are sides, as are XC and CB. So if the hexagon is
AA1 A1 XCB with points in that order, then we apply Pascal’s theorem to find that the points
K = AA1 ∩ XC, D = A1 A1 ∩ CB and P = A1 X ∩ BA are collinear as desired.
 Transformation geometry
7
Geometric transformations are often used in an effort to understand a particular geometry
problem better. The transformations considered in this chapter are spiral symmetries and
affine transformations. Reflections are discussed in section 12.2.
Every spiral symmetry may be described as a linear complex function.1 Understanding what
functions correspond to what spiral symmetries helps to understand their compositions.
An anticlockwise rotation of angle θ composed with a dilation of factor r = 6 0 about the same
point is called a spiral symmetry. If it is performed about the origin, it may be described by

f (z) = az,

where a = reiθ = r cos θ + ir sin θ. If it is performed about a general point P represented by

the complex number p, then we have f (z) − p = a(z − p). Thus

f (z) = az + (1 − a)p.

Hence a general spiral symmetry is described by

f (z) = az + b

for any complex number b provided that a 6= 0 or 1. The centre of the spiral symmetry is
b
given by p = 1−a . If a = 1, we also have the identity spiral symmetry f (z) = z. Furthermore,
if we throw in the translation functions f (z) = z + b for each complex number b, we have
the complete set of linear functions of the complex plane. In loose terms a translation can
be thought of as a spiral symmetry whose centre is infinitely far away and whose angle of
rotation is 0◦ . From here on, whenever we speak of a spiral symmetry, we implicitly include
the translations amongst them.
To summarise, the full family of spiral symmetries, now also including the translations, is
given by the set of functions
f (z) = az + b,
where a 6= 0. The rotation angle is θ = arg(a) and the dilation factor is r = |a|.
Consider what happens if we compose two such spiral symmetries. Suppose we do

f1 (z) = a1 z + b1
1 You might like to check out chapter 8 and section 17.1 if you would like a refresher on complex numbers.
126 7 Transformation geometry

followed by
f2 (z) = a2 z + b2 .
We obtain
f3 (z) = f2 (f1 (z)) = a3 z + b3 ,
where a3 = a1 a2 and b3 = a2 b1 + b2 .
If a1 = r1 eiθ1 , a2 = r2 eiθ2 and a3 = r3 eiθ3 , then

r3 = r1 r2 and θ3 = θ1 + θ2 .

Thus we have the following.

Theorem When composing spiral symmetries, we have the following.

Dilation factors multiply.

Rotation factors add.

Note also that in general the order of composition does matter. So, it would not necessarily
be that case that f1 (f2 (z)) = f2 (f1 (z)). For instance, the respective centres of the spiral
symmetries are usually different.
There are some important subgroups of the family of spiral symmetries.

Translations
A general translation is represented by f (z) = z + α for any complex number α. It is
easy to verify that the composition of two translations is a translation.
Rotations
A rotation is represented by f (z) = az + b for |a| = 1. Note that for our purposes, the
family of rotations includes all the translations. It is easy to verify that the composition
of two rotations is a rotation.
Dilations
A dilation is represented by f (z) = az + b for a ∈ R and a 6= 0. Note that for our
purposes, the family of dilations includes all the translations. It is easy to verify that
the composition of two dilations is a dilation.

There are also geometric transformations, known as affine transformations, which we will
discuss later in the chapter.

7.0 Problems
1. Given three parallel lines, show how to construct with straightedge and compass a point
on each line so that the points form an equilateral triangle.
2. If the opposite sides of a hexagon are equal and parallel, prove that the diagonals joining
opposite vertices are concurrent.
3. Show that the three medians of a triangle are concurrent.
4. Given three circles in the plane, show how to construct a point on each so that the
points form an equilateral triangle. Discuss under what circumstances this is even
possible.
7.0 Problems 127

5. Given two triangles in the plane which have corresponding sides parallel, show that the
lines joining the corresponding vertices are concurrent.

6. Let ABC be a triangle and let D, E and F be the midpoints of BC, AC and AB,
respectively.
−→ −→ −→
Prove that the vectors AD, BE and CF form a triangle.

7. Show how to construct, using straightedge and compass, a square whose vertices all lie
on the sides of a given triangle.

8. Two circles are internally tangent at T . A chord AB of the outer circle is tangent to
the inner circle at P .
Prove that T P bisects ∠AT B.

9. Two common tangents of two intersecting circles meet at a point A. Let B be a point of
intersection of the two circles, and C and D be the points in which one of the tangents
touches the circles.
Prove that the line AB is tangent to the circumcircle of triangle BCD.

10. Let ABC be a triangle. Triangles A0 BC, B 0 CA and C 0 AB are erected externally on
the sides of triangle ABC such that

∠AC 0 B + ∠BA0 C + ∠CB 0 A = 180◦ .

Let OA , OB and OC be the circumcentres of triangles A0 BC, B 0 CA and C 0 AB, respec-

tively.
Prove that ∠OA OC OB = ∠AC 0 B.

11. Let X and Y be the centres of the squares erected externally on sides AB and AC of
triangle ABC. Let M be the midpoint of BC.
Prove that M X and M Y are equal and perpendicular.

12. Triangle ABC has squares P ABK and QACL constructed on the exterior of the sides
AB and AC, respectively. Let AH be an altitude of triangle ABC with H on BC.
Prove that A, H and the midpoint of P Q are collinear.

13. Given triangle ABC, we erect similar isosceles triangles BCP , ACR and ABQ externally
on sides BC and AC and internally on side AB.
Prove that P QRC is a parallelogram.

14. Given a quadrilateral ABCD, we erect the four equilateral triangles ABP and CDR
externally and BCR and ADS internally on its sides.
Show that P QRS is a parallelogram.2

15. Chords AB and CD of circle Γ intersect at a point E inside Γ. The circle ω is internally
tangent to the figure bounded by segments AE and EC and arc AC (not containing B
or D) of Γ, touching arc AC at point F . A line ` containing the centre O of Γ, intersects
segments AE and DE at points P and Q, respectively, and satisfies EP = EQ. Line
EF intersects ` at point M .
Prove that the line through M parallel to the line AB is tangent to Γ.
2 This is a generalisation of the previous problem.
128 7 Transformation geometry

16. Let M be the midpoint of the altitude of triangle ABC from vertex A. Let I be the
incentre of the triangle. Let Y be the point of tangency of the excircle opposite vertex
A with side BC.
Prove that M , Y and I are collinear.
17. The incircle of triangle ABC has centre I and touches BC at D. Let E be the midpoint
of BC.
Prove that the line through E and I passes through the midpoint of the segment AD.
18. Two circles Γ1 and Γ2 lie inside and are internally tangent to a third circle Γ at points
A1 and A2 , respectively. A common external tangent touches Γ1 at T1 and Γ2 at T2 .
Let P be the intersection of lines A1 T1 and A2 T2 .
(a) Prove that P lies on Γ.
(b) Hence prove that A1 T1 T2 A2 is cyclic.
(c) Hence also prove that if Γ1 and Γ2 intersect at two points, then the extension of
their common chord also passes through P .
19. Let Q = A0 A1 A2 A3 be a quadrilateral in the plane. Given a point M0 in the plane we
define the sequence M0 , M1 , M2 , . . . of points in the plane as follows. If n ≡ i (mod 4),
then Mn+1 is the point obtained from rotating the point Mn anticlockwise about Ai by
90◦ .
(a) Prove that if M2008 = M0 for one point M0 in the plane, then we also have
M2008 = M0 for all points M0 in the plane.
(b) Suppose that Q is a parallelogram. Prove that if M2008 = M0 , then Q is a square.
Under what circumstances is the converse true?
(c) Suppose that Q is a general quadrilateral. Find simple necessary and sufficient
conditions on Q such that M2008 = M0 .
20. Let O, A, B, C, D and E be six points in the plane such that the 10 triangles which
have O as a vertex all have area at least 1.
√
(a) Prove that one of those 10 triangles has area at least 2.
(b) Is the result still true if we only consider O plus four more points?
21. A point P is permitted to vary over the interior of a fixed triangle ABC. The line AP
meets BC in A1 . Points B1 and C1 are defined similarly.
As P varies over the interior of the triangle, what is the maximum possible area of
triangle A1 B1 C1 ?
22. Circles Γ1 and Γ2 intersect at points A and B. A common tangent to the two circles
touches Γ1 at P1 and Γ2 at P2 . Points M1 and M2 are such that line M1 M2 is the
perpendicular bisector of AB, and

∠P1 M1 M2 = ∠P2 M2 M1 = 90◦ .

Line AM1 intersects Γ1 again at N1 . Line AM2 intersects Γ2 again at N2 . Lines N1 P1

and N2 P2 intersect at point P .
Prove that BP1 P P2 is a parallelogram.
23. Let ε be an ellipse inscribed in 4ABC.
Prove that the foci of ε are isogonal conjugates.
7.0 Problems 129

24. Let H be a strictly convex hexagon inscribed in an equilateral triangle. Suppose that
H has all of its sides of equal length.
Prove that the three main diagonals of H are concurrent.
25. Let ABCD be a convex quadrilateral with BA = BC. Denote the incircles of triangles
ABC and ADC by ω1 and ω2 , respectively. Suppose that there exists a circle ω tangent
to the ray BA beyond A and to the ray BC beyond C, which is also tangent to the
lines AD and CD.
Prove that the common external tangents of ω1 and ω2 intersect on ω.
130 7 Transformation geometry

7.1 Translations
Problem Given a segment AB and circles S1 and S2 , using a straightedge and compass,
show how to construct points P and Q, one on each circle, satisfying P Q k AB and P Q = AB.

Solution

P Q S2

S1 τ (S1 )

A B

Suppose that P and Q are located as required on the two circles with P on S1 , say. Then the
translation τ , which takes A to B, will also take P ∈ S1 to Q ∈ S2 . Thus if we apply the
translation τ to S1 ,3 any place where τ (S1 ) intersects S2 will be a candidate for Q. Finally,
we can invert the translation to recover P from Q.

In the above problem there may be 0, 1, 2, 3, 4, or infinitely many possible positions for the
segment P Q. Can you analyse under what conditions these occur?

7.2 Rotations
Problem Suppose ABC is a triangle. We erect three equilateral triangles P BC, QCA and
RAB externally on the sides of triangle ABC. Let K, L and M be their respective centroids.
Show that triangle KLM is equilateral.

Solution Note that triangles AM B, BKC and CLA are all isosceles and have 120◦ angles
at M , K and L, respectively. We capitalise on the fact that these three angles sum to 360◦ .

B C
M0

3 It is routine to translate a circle using straightedge and compass.

7.3 Dilations 131

Consider the three 120◦ clockwise rotations TM , TK and TL about M , K and L, respectively.
Note that
TM (A) = B, TK (B) = C and TL (C) = A.
We know that the composition TL ◦ TK ◦ TM of these rotations has angle sum 360◦ and thus
must be a translation. However,

TL (TK (TM (A))) = A.

Thus the composition is a translation which fixes the point A. Hence the composition must
be the identity transformation.
Consider now, what happens to the point M .

Under TM , point M remains fixed.

Under TK , point M is rotated to some point M 0 , say.

Finally under TL , point M 0 is rotated back to M .

From these observations we see that

M K = KM 0 , M L = LM 0 and ∠M KM 0 = ∠M LM 0 = 120◦ .

It follows that triangles M KL and M 0 KL are congruent (SSS). Consequently,

∠M KL = ∠M 0 KL = 60◦ and ∠M LK = ∠M 0 LK = 60◦ .

This is sufficient to show that triangle KLM is equilateral.

7.3 Dilations
One of the important things to remember about dilations is that if X is the centre of a
dilation f , then X, Y and f (Y ) are always collinear for any point Y .

Problem Prove that the centroid, orthocentre and circumcentre of a triangle are collinear.4

Solution
A

C0 B0

..
G O

B A0 C

4 The line common to all three points is called the Euler line.
132 7 Transformation geometry

Consider the dilation f with scale factor − 12 about the centroid G of triangle ABC. Note
that f maps triangle ABC to its medial triangle A0 B 0 C 0 which has the same centroid G0 = G
as triangle ABC.
Let H be the orthocentre of triangle ABC and let H 0 = f (H) be the orthocentre of triangle
A0 B 0 C 0 . The perpendicular bisectors of the sides of ABC are the altitudes of the medial
triangle and so O = H 0 . But H, G and H 0 = f (H) are collinear thanks to the dilation. Thus
H, G and O are collinear.

In fact we have proven more! Namely that H, G and O occur in that order on the line and
that HG : GO = 2 : 1.
Continuing these ideas, it is possible to deduce that if N is the circumcentre of the medial
triangle, then N also lies on the same line in between H and G. Furthermore,

HN : N G : GO = 3 : 1 : 2.

Point N is in fact the centre of the nine-point circle of the triangle.

With a little more thought can you see that there are two dilations which transform the
nine-point circle into the circumcircle?! One has centre G and dilation factor −2. The other
has centre H and dilation factor +2.

Problem For a pair of non-congruent circles neither of which lies inside the other, we define
their focal point to be the point of intersection of their pair of external common tangents. For
three circles in the plane, none of which lies inside the other, we thus have three focal points,
one for each pair of circles.
Prove that these three focal points are collinear.5

Solution

S3

S2
S1

P3 P2 P1

Let the circles be S1 , S2 and S3 and let r1 , r2 and r3 be their respective radii. Let P1 , P2
and P3 be the focal points of the pairs (S2 , S3 ), (S3 , S1 ) and (S1 , S2 ) of circles, respectively.
The intersection of the two common external tangents to a pair of circles is a centre of dilation
of the circles, and the dilation factor is equal to the ratio of the two radii.
5 Thisis known as Monge’s theorem. If you understand the proof of it, you should be able to state and prove
a similar result involving two internal focal points (the intersection point of the pair of internal tangents)
and an external focal point. In fact you should be able to state and prove a result even when one circle is
inside another.
7.4 Spiral symmetries 133

With this in mind, let D1 be the dilation with factor + rr32 , centred at P1 . Note that D1 sends
S2 to S3 . Similarly, let D2 be the dilation with factor + rr31 , centred at P2 and let D3 be the
dilation with factor + rr21 , centred at P3 . Thus

D1 (S2 ) = S3 , D2 (S3 ) = S1 and D3 (S1 ) = S2 .

Note that the composition D3 ◦ D2 ◦ D1 has dilation factor equal to

r3 r1 r2
· · = +1
r2 r3 r1
and so D3 ◦ D2 ◦ D1 must be a translation. However,

D3 (D2 (D1 (S2 ))) = S2 .

Thus the composition is a translation which fixes the circle S2 . Hence the composition must
be the identity transformation.
Consider now the line ` = P1 P2 . Since ` passes through P1 , we see that the dilation D1 leaves
` fixed (although points within ` do move along `). After next applying D2 to ` we see that `
still remains fixed because ` passes through P2 . Finally applying D3 to ` must leave ` fixed
because D3 ◦ D2 ◦ D1 is the identity transformation.
However, if P3 were not on `, then D3 would move `. Thus P3 is also on ` and hence collinear
with P1 and P2 .

7.4 Spiral symmetries

Problem Let P be a point inside triangle ABC such that ∠P BA = ∠P CA. Let K and L
be the feet of the perpendiculars from P to AB and AC, respectively. Let M be the midpoint
of BC.
Prove that KM = LM .6
PB
Solution Let θ = ∠P BK = ∠LCP , and let r = KB = PLCC
. We exploit the fact that
triangles P KB and P CL are similar, but oppositely oriented.

L
K

θ θ

B θ M θ C

M0
6 This problem is solved using complex numbers in section 8.4.
134 7 Transformation geometry

Consider the spiral symmetry SC centred at C which takes L to P . Note that SC has dilation
factor r and rotation factor θ. Consider also the spiral symmetry SB centred at B which
takes P to K. Note that SB has dilation factor 1r and rotation factor θ.
The composition SB ◦ SC has dilation factor 1 and rotation factor 2θ. It is thus a rotation
about some point. If we can show that this point is M , then we are done because the
composition takes L to K.
Since the only point fixed by a rotation is the centre of rotation, it suffices to show that M is
fixed under the composition. This is rather easy. Indeed suppose that SC (M ) = M 0 , then
triangles CM M 0 and CLP are similar. Thus M M 0 ⊥ BC. Since M is the midpoint of BC
we also have that triangles BM M 0 and CM M 0 congruent. Thus SB (M 0 ) = M . Hence M is
indeed fixed under the composition.

In fact we have proven not only that KM = LM but also that ∠KM L = 2θ, which is even
more than we were asked to prove!

7.5 Affine transformations

A simple affine transformation can be thought of as a stretch of the plane in one direction.
For example, the transformation (x, y) 7→ (x, ry), where r > 0 is a stretch in the direction of
the y-axis.
Composing such transformations with spiral symmetries yields the full set of affine trans-
formations. These may be represented as the set of functions F : R2 → R2 , where
(x, y) 7→ (ax + by + c, dx + ey + f ) and ad − bc 6= 0.
Unlike the family of spiral symmetries, affine transformations do not generally preserve angles.
Nevertheless, affine transformations have a number of useful properties.

Properties of affine transformations

1. Collinear points remain collinear.

2. Concurrent lines remain concurrent.
3. Parallel lines remain parallel.

4. Ratios of lengths of parallel segments are unchanged.

5. Ratios of areas are unchanged.
6. It is possible to transform any ellipse into any circle.

7. It is possible to transform any triangle into any triangle.

8. It is possible to transform any convex quadrilateral into a cyclic quadrilateral whose
diagonals intersect at any given angle.

Note that properties 6, 7 and 8 are not independent of each other. For example, if we choose
to transform an ellipse into a circle, then we lose the freedom to transform any triangle into
any triangle.

Problem Let G be the centroid of triangle ABC and let M be the midpoint of BC. Let X
and Y be on AB and AC, respectively, such that the points X, G and Y are collinear and so
7.5 Affine transformations 135

that XY is parallel to BC. Suppose that XC and BG intersect at Q and that Y B and GC
intersect at P .
Show that triangle M P Q is similar to triangle ABC.

Solution It is sufficient to prove that QM k AC, P M k AB and QP k BC. Conveniently

these statements are all invariant under an affine transformation.
Furthermore, the whole problem set-up may be defined purely in terms of information that is
invariant under affine transformations. For example, the centroid is just the intersection of
the three medians. But a median requires that a midpoint of a side be chosen and a midpoint
of a segment remains a midpoint after an affine transformation. The segment XY remains
parallel to BC, and so on.
Consider the affine transformation which turns ABC into an equilateral triangle. By the
foregoing discussion it suffices to prove the result only for the special case in which ABC is
equilateral! This is not particularly difficult and we leave it as an exercise.

Note that we could just have easily assumed that ABC was a right isosceles triangle and
used coordinate geometry.
 Complex numbers
8
Many geometry problems can be solved by simply tossing them onto the complex number
plane and then doing a few routine calculations. Arithmetical operations involving complex
numbers1 can be seen to have geometric interpretations. Indeed if α ∈ C is a constant, we
have the following table of interpretations for any complex number z.

Algebra Geometry
z±α Translation
z
z · α, (for α 6= 0) Spiral symmetry
α
Dilation if α ∈ R
Rotation if |α| = 1
z Reflection
|z| Length
arg(z) Angle
Roots of z n = 1 Vertices of regular n-gon

8.0 Problems
1. Suppose that points A and B in the plane are represented by the complex numbers α
and β.
Find geometric interpretations for the arithmetic, geometric and harmonic means of α
and β.

2. Show how to code the following properties using complex numbers.

(a) Two lengths are equal.

(b) Two lengths are parallel.
(c) Two lengths are perpendicular.
(d) Two angles are equal, or one is twice another.
1 Please refer to section 17.1 if you would like to brush up on complex number basics.
138 8 Complex numbers

(e) A length is constant.

(f) An angle is constant.
(g) Two triangles are similar. (Be careful! There are two cases.)
(h) A quadrilateral is cyclic.
3. Given a triangle ABC, find complex number expressions for the triangle’s centroid,
circumcentre and orthocentre in terms of the vertices. The expressions are particularly
simple if the triangle is inscribed in the unit circle of the complex plane.
4. Given triangle ABC, we erect similar isosceles triangles BCP , ACR and ABQ externally
on sides BC and AC and internally on side AB.
Prove that P QRC is a parallelogram.2
5. Given a quadrilateral ABCD, we erect the four equilateral triangles ABP and CDR
externally and BCR and ADS internally on its sides.
Show that P QRS is a parallelogram.
6. Devise a geometry problem based on the relation

(2 + i)(3 + i) = 5 + 5i.

7. Draw any quadrilateral. On each side draw a square lying outside the given quadrilateral.
Draw line segments joining the centres of opposite squares.
Show that the two line segments are equal in length and perpendicular.
8. Let Q = A0 A1 A2 A3 be a quadrilateral in the plane. Given a point M0 in the plane we
define the sequence M0 , M1 , M2 , . . . of points in the plane as follows. If n ≡ i (mod 4),
then Mn+1 is the point obtained from rotating the point Mn anticlockwise about Ai by
90◦ .
(a) Prove that if M2008 = M0 for one point M0 in the plane, then we also have
M2008 = M0 for all points M0 in the plane.
(b) Suppose that Q is a parallelogram. Prove that if M2008 = M0 , then Q is a square.
Under what circumstances is the converse true?
(c) Suppose that Q is a general quadrilateral. Find simple necessary and sufficient
conditions on Q such that M2008 = M0 .
9. Let ABCDE be a convex pentagon such that

∠BAC = ∠CAD = ∠DAE and ∠ABC = ∠ACD = ∠ADE.

The diagonals BD and CE meet at P .

Prove that the line AP bisects the side CD.
10. Consider a regular n-gon A0 A1 . . . An−1 and a point P on its circumcircle.
(a) Compute the value of
n−1
X
P A4i .
i=0
2 Some of these problems were also seen in Transformation geometry. It should not be a surprise that some
geometry problems are susceptible to both tactics. After all, the basic properties of spiral symmetries can be
explained using complex numbers.
8.0 Problems 139

(b) For which points P is the product

n−1
Y
P Ai
i=0

maximised or minimised?
(c) What results can you derive if P is allowed to vary over a circle concentric with
the circumcircle?
11. (a) Show there is an equiangular (i.e. having all internal angles equal) hexagon with
side lengths 1, 2, 3, 4, 5 and 6 in some order.
(b) Do the same for a 15-gon.
(c) Generalise to any n-gon, where n is not a power of a prime.
(d) If n has at least three different prime factors, show that we can construct an
equiangular n-gon with side lengths 12 , 22 , . . . , n2 in some order.
12. For which integers n ≥ 3 is it possible to find a convex n-gon with all its interior angles
equal but all its side lengths different positive integers?
13. If P (x), Q(x), R(x) and S(x) are all polynomials such that

P (x5 ) + xQ(x5 ) + x2 R(x5 ) = (x4 + x3 + x2 + x + 1)S(x),

prove that x − 1 is a factor of P (x).

14. Suppose
(1 + x + x2 + · · · + x10 )200 = a0 + a1 x + · · · + a2000 x2000 .

Determine the value of

a1 + a12 + a23 + · · · + a1992 .
140 8 Complex numbers

8.1 Addition ideas

In some contexts, complex numbers may be thought of as vectors. Expressions for the
midpoint and centroid are exactly the same as for vectors. Specifically, the midpoint M of
AB can be described as
α+β
µ= ,
2
where α, β and µ are the complex numbers representing the points A, B and M , respectively.
From here on, in a minor abuse of notation, we will simply identify the points with their
corresponding complex number. So, for example, not only will M represent the point M of
the geometrical configuration, but it will also denote the complex number represented by M
after tossing the figure onto the complex plane. Thus, in the above example,
1
M= (A + B).
2
The centroid of a triangle ABC is given by
1
G= (A + B + C).
3
More generally, the centroid of an n-gon A1 A2 . . . An is given by
1
(A1 + A2 + · · · + An ).
n

Problem Show that the midpoints of a quadrilateral form a parallelogram whose diagonals
intersect at the centroid of the original quadrilateral.

Solution Let the quadrilateral be ABCD. Let the midpoints of AB, BC, CD and DA be
E, F , G and H, respectively. If we toss the quadrilateral onto the complex plane, then we
have
1 1 1 1
E= (A + B), F = (B + C), G= (C + D) and H = (D + A).
2 2 2 2

D
G
C

H
F

A E B

We know that EF GH is a parallelogram if and only if EF is equal and parallel to HG. That
is, if
F − E = G − H.
A routine calculation shows in fact that they both equal 12 (C − A). Thus EF GH is indeed a
parallelogram.
8.2 Angles 141

Finally, we know that the centroid of a general quadrilateral ABCD is just

1
(A + B + C + D).
4
The centroid of a parallelogram is also the intersection of its diagonals.3 For EF GH this
point is
1
(E + F + G + H),
4
which easily simplifies to the same expression as we obtained for ABCD.

8.2 Angles
Given an angle ∠AOB, where O is at the origin of the complex plane, we can express
 
B
∠AOB = arg B − arg A = arg .
A

(This is only correct if ∠AOB is measured anticlockwise. If ∠AOB is measured clockwise,

A
then the correct expression is ∠AOB = arg A − arg B = arg B .)
Given four points A, B, X and Y in the plane, how can we express the angle between segments
AB and XY ? If we translate both these segments to the origin of the complex plane, we find
that the required angle is just  
B−A
arg .
Y −X
Given a complex number z, how can we find arg z? If z = reiθ , then z = re−iθ . Thus
z 2iθ
z =e . This form can be useful when you want to prove that two angles are equal.

Problem Let A and B be two fixed points on a circle Γ and let XY be a variable diameter
of Γ. Find the locus of points P defined by the intersection of AX and BY as XY varies
over Γ.

Solution A careful diagram suggests that P lies on a circle passing through A and B. We
can verify this by showing that ∠AP B is constant. This is equivalent to showing that the
complex number z = A−XB−Y has constant argument. This in turn is equivalent to showing that
z
z is constant.

A
X

O. P

3 Try and prove this fact for yourself. It’s not hard.
142 8 Complex numbers

We may assume that the centre O of Γ is the origin of the complex plane. Thus XY being
a diameter means that Y = −X. Let r be the radius of the circle. Thus for any complex
number α lying on the circle, we have αα = r2 .
We compute

z A−X B−Y
= ·
z B−Y A−X
r2 r2
A−X B − Y
= · r2 r2
B−Y A − X
AX
=
BY
A
=− .
B
1 z
. Thus ∠AP B = 21 ∠AOB, which is a constant.


But arg z = 2 arg z

Do not be content to finish here with the problem! The only instance of where we actually
used Y = −X was to simplify A·X A X z
B·Y to − B . In fact, if Y is constant, we still derive that z is
constant. But XY is constant is the same as saying that XY is a chord of constant length.
Thus our solution by complex numbers shows that the problem can in fact be generalised.
Can you discuss the significance of the negative sign?
Can you discuss how we account for both parts, that is, the major and minor arcs of the
circle which form the locus of P ?

8.3 Multiplication ideas

A rotation about the origin of the complex plane corresponds to multiplication by a complex
number of unit magnitude. If we wish to rotate about a point other than the origin, we must
first translate to the origin, do the rotation then translate back. Specifically, if A, X and Y
are three points such that Y is obtained by rotating X about A through an angle of θ, then
we may write
Y = A + (X − A)eiθ .

Problem Devise a geometry problem4 based on the equality

(2 + i)3 = 2 + 11i.

Solution We observe that the relation immediately tells us that

arg(2 + 11i) = 3 arg(2 + i).

Thus we may pose the following problem.

Problem Let ABC be a triangle such that AB ⊥ BC, AB = 2 and BC = 11.

Let D be a point on side BC such that BD = 1.
Prove that ∠DAB = 13 ∠CAB.
4 Thereare lots of complex number relations that give rise to potential geometry problems. You might like to
try and find some more.
8.3 Multiplication ideas 143

10

D
1
A 2 B

The solution would then run as follows.

Solution Toss triangle ABC onto the complex plane so that

A = 0, B=2 and C = 2 + 11i.
Then D = 2 + i. We can compute that
D3 = (2 + i)3 = 2 + 11i = C.
Therefore,
arg(C) = arg(D3 ) = 3 arg(D).
This means that ∠CAB = 3∠DAB.

Problem Let P AB and P XY be equilateral triangles oriented the same way. Let K, L and
M be the midpoints of P A, BX and P Y , respectively.
Prove that triangle KLM is equilateral.
πi
Solution Toss the figure onto the complex plane so that P = 0. Let ω = e 3 . Multiplication
by ω corresponds to a rotation of 60◦ anticlockwise. Thus we may write
B = ωA and Y = ωX.

Y
L

P K A
144 8 Complex numbers

We may also write

1 1 1
K= A, L = (ωA + X) and M = ωX.
2 2 2
Triangle KLM is equilateral if and only if
L − M = ω(K − M )
1 1 1 1 1
⇔ ωA + X − ωX = ωA − ω 2 X
2 2 2 2 2
⇔ X(ω 2 − ω + 1) = 0.
Since
πi 1 √
ω=e 3 = cos 60◦ + i sin 60◦ = (1 + i 3),
2
it is easy to check that ω 2 − ω + 1 = 0, as desired.

8.4 Similarity ideas

Suppose we wish to specify a triangle ABC up to similarity. Then it is enough to specify one
ratio of side lengths and the measure of the angle between these sides.
Let’s toss ABC onto the complex plane so that A is situated at the origin of the complex
plane. The angle at A is the same as arg B − arg C = arg( B AB
C ). The ratio AC is the same as
|B|
 
B B
|C| = C . These two pieces of information are simply the polar form of C . Thus it suffices
to know only the value of BC . If A is not at the origin, then translating the triangle so that A
is at the origin, shows that the similarity type of ABC depends only on B−A C−A .
In particular, suppose that we wish to locate B such that ∠BAC = θ and AB = rAC. If A
were at the origin, we would have B = zC, where |z| = r and arg z = θ. If A is not at the
origin, we find after translating to the origin that B − A = z(C − A). Thus
B = zC + (1 − z)A
and the complex number z determines the similarity type of triangle ABC.

Problem Let similar triangles AKB, BLC and CM A be constructed on the exterior of
triangle ABC.
Prove that the centroids of triangles ABC and KLM coincide.

Solution
A

M
K

C B

L
8.4 Similarity ideas 145

AK
Let z be the complex number such that arg z = ∠BAK and |z| = AB . Due to the similarity
of the three triangles we may write

K = zB + (1 − z)A, L = zC + (1 − z)B and M = zA + (1 − z)C.

The centroid of triangle KLM is just 13 (K + L + M ), which upon substitution in terms of A,

B, C and z, simplifies to 13 (A + B + C) and this is the centroid of triangle ABC.

Problem Let P be a point inside triangle ABC such that ∠P BA = ∠P CA. Let K and L
be the feet of the perpendiculars from P to AB and AC, respectively. Let M be the midpoint
of BC.
Prove that KM = LM .5

Solution Toss the figure onto the complex plane.

L
K

θ θ

B M C

Triangles P KB and P LC are similar but are oppositely oriented. Assume that BK CL
P K = P L = r.
◦
Point B is obtained by rotating P about K by 90 clockwise and dilating by a factor r. This
corresponds to multiplying by the complex number −ir. Thus

B = (−ir)P + (1 − (−ir))K = −irP + (1 + ir)K.

Similarly, point C is obtained by rotating P about L by 90◦ anticlockwise and dilating by a

factor r. This corresponds to multiplying by the complex number ir. Thus

C = irP + (1 − ir)L.

Finally, M = 12 (B + C), because it is the midpoint of BC. Therefore,

1
M= ((1 + ir)K + (1 − ir)L).
2

Now KM = LM if and only if

|K − M | = |L − M |
1 1
⇔ |(1 − ir)K − (1 − ir)L| = |(1 + ir)L − (1 + ir)K|
2 2
⇔ |K − L||1 − ir| = |K − L||1 + ir|.

This is true because r√is a real number and so 1 − ir and 1 + ir are complex conjugates with
common magnitude 1 + r2 .
5 This problem was solved using spiral symmetries in section 7.4.
146 8 Complex numbers

Note that in the above solution point A was not used at all in the calculation. This is
because the important part consisted of the two similar triangles P KB and P LC. In such a
calculational solution it is probably better only to draw in what you need to complete the
calculation. In this case point A and the segments AK and AL could be omitted.

8.5 Roots of unity

The nth roots of unity, that is, all n complex roots of the equation xn = 1, are the vertices of
a regular n-gon centred at 0. Thus questions about regular polygons can often be posed in
terms of roots of unity.
If ω 6= 1 is a root of unity, then ω is a root of

xn−1 + xn−2 + · · · + 1 = 0.

Since |ω| = 1, it is also true that

1
ω= = ω n−1 .
ω
2πik 2πi
The set of nth roots of unity may be given by {e n | k = 0, 1, . . . , n − 1}. Thus if ω = e n ,
then all the other nth roots of unity are just the powers of ω, that is, 1, ω, ω 2 , . . . , ω n−1 .

Problem Let A0 A1 . . . An−1 be a regular n-gon of circumradius r.

Prove that for any point P on the circumcircle of the n-gon, the value of
n
X
P A2k
k=0

is independent of P and find this value.

Solution Dilate the n-gon by a factor of 1r and toss it onto the complex plane so that each
2πi
Ak is an nth root of unity. Specifically, Ak = ω k , where ω = e n . Furthermore, since P is a
point on the circumcircle we may identify P with a complex number ρ satisfying

ρρ = |ρ|2 = 1.

A2 = ω 2 P
A1 = ω

A3

A0 = 1

A4

A6
A5
8.5 Roots of unity 147

Recall that
1 + ω + ω 2 + · · · + ω n−1 = 0.

We may now compute as follows.

P A2k = |ρ − ω k |2
= (ρ − ω k )(ρ − ω k )
= (ρ − ω k )(ρ − ω k )
= ρρ + (ωω)k − ρω k − ρω k
= |ρ|2 + |ω|2k − ρω k − ρω k
= 2 − ρω k − ρω k .

Thus
n−1
X n−1
X 
P A2k = 2 − ρω k − ρω k
k=0 k=0
n−1
X n−1
X
= 2n − ρ ω n−k − ρ ωk
k=0 k=0
n−1
X n−1
X
= 2n − ρ ωk − ρ ωk
k=0 k=0
= 2n − 0 − 0
= 2n.

Thus for a regular n-gon of unit circumradius the sum required is 2n. If the circumradius
is r, then the answer is 2nr2 , which is independent of P .

Problem Let p ≥ 3 be a prime number and let P be a convex p-gon with all its interior
angles equal and all its side lengths positive integers.
Prove that P is regular.

Solution Suppose that the side lengths of P in order were a0 , a1 , . . . , ap−1 . It would then
follow that
p−1
X
ak ω k = 0,
k=0
2πi
where ω = e p . (Can you see why?)
So ω is a root of the polynomial
p−1
X
f (x) = ak xk = 0.
k=0

Yet ω, being a root of unity, is also a root of

p−1
X
g(x) = xk = 0.
k=0

Thus ω is a root of
d(x) = gcd(f (x), g(x)).
148 8 Complex numbers

However, g(x) is irreducible.6 Thus d(x) = g(x) and so g(x) is a factor of f (x). Since f (x) and
g(x) have the same degree, we must have f (x) = a0 g(x). Consequently, a0 = a1 = · · · = ap−1
and therefore, P is regular.

Problem Suppose that

(1 + x + x2 + · · · + x10 )200 = a0 + a1 x + · · · + a2000 x2000 .

Determine the sum

a0 + a11 + a22 + · · · + a1991 .

Solution Let
2000
X
p(x) = aj xj = (1 + x + · · · + x10 )200 .
j=0
2πi
Let ω = e 11 . Note that p(wk ) = 0 provided that 11 - k.
We now substitute x = ω k for k = 0, 1, 2, . . . , 10 into p(x). This yields the following 11
equations.
2000
X
aj = 11200
j=0
2000
X
aj ω j = 0
j=0
2000
X
aj ω 2j = 0
j=0
..
.
2000
X
aj ω 10j = 0
j=0

On the LHS we see that the coefficient of aj is

10
X 10
X
ω jk = (ω j )k .
k=0 k=0

If 11 - j, then this sum is just zero. If 11 | j, then this sum is 11.

Thus
LHS = 11(a0 + a11 + a22 + · · · + a1991 ).

On the RHS we simply get 11200 . Thus

a0 + a11 + a22 + · · · + a1991 = 11199 .

6 To prove that g(x) = 1 + x + x2 + · · · + xp−1 is irreducible over Z whenever p is a prime, take the change
of variables x = y + 1, expand the brackets, then use the upstairs–downstairs technique modulo p from
section 9.10.
 Polynomials
9
Polynomials are just algebraic expressions of a particular type. However, they actually consti-
tute much more than a boring branch of algebra. In fact, polynomials have many interesting
properties and many amazing connections to other flavoursome areas of mathematics. This
need for lateral thinking provides a fertile breeding ground for mathematical problems, and
in this chapter we’ll sample a variety of those goodies. But in order to proceed, you’ll need
some basic knowledge about polynomials, but not too much. Let’s start by introducing a bit
of jargon, while various results concerning polynomials will appear, mostly without proof,
scattered throughout the chapter.

A polynomial is an expression of the form

P (x) = an xn + an−1 xn−1 + · · · + a0 ,

where a0 , a1 , . . . , an are numbers called coefficients and x is a variable. If the coefficients

are integers, then we call P (x) an integer polynomial or a polynomial over Z, and
similarly for Q, R and C.

We refer to a0 as the constant term and we refer to an as the leading coefficient, provided
that it is non-zero. If an is equal to 1, then we say that the polynomial is monic.

If an is non-zero, then we refer to n as the degree of the polynomial, and denote it by

deg P (x). By convention, the zero polynomial has degree equal to −∞. Note that the
degrees of two polynomials behave nicely under multiplication and addition.

deg P (x)Q(x) = deg P (x) + deg Q(x)

deg(P (x) + Q(x)) ≤ max(deg P (x), deg Q(x))

Equality almost always holds in the last line. The only time it does not is when the
leading terms of P (x) and Q(x) cancel each other out.

If P (r) = 0, then we say that r is a root or a zero of P (x).

We say that P (x) is a divisor or a factor of Q(x) if there exists a polynomial M (x) such
that P (x)M (x) = Q(x).
150 9 Polynomials

9.0 Problems
1. Let P (x) be a real polynomial satisfying

P (1) = 3, P (2) = 5 and P (3) = 2.

Determine the remainder when P (x) is divided by (x − 1)(x − 2)(x − 3).

2. Let P (x) be an integer polynomial such that P (2) is divisible by 5 and P (5) is divisible
by 2.
Prove that P (7) is divisible by 10.
3. If p, q, r are the roots of the polynomial x3 − x − 1, compute
1+p 1+q 1+r
p2 + q 2 + r2 and + + .
1−p 1−q 1−r

4. Find the cubic monic polynomial whose roots are the cubes of the roots of

x3 − x2 + x − 2 = 0.

5. For which positive integers n is the polynomial

(x + 1)n + xn + 1

divisible by the polynomial x2 + x + 1?

6. Let P (x) be an integer polynomial whose leading coefficient is odd. Suppose that P (0)
and P (1) are also odd.
Prove that P (x) has no rational roots.
7. Let P (x) be a monic polynomial of degree four with distinct integer roots a, b, c and d.
If P (r) = 4 for some integer r, prove that
1
r= (a + b + c + d).
4

8. Does there exist an integer polynomial P (x) such that

P (10) = 400, P (14) = 440 and P (18) = 520?

9. Find a polynomial P (x) such that P (x) is divisible by x2 + 1 and P (x) + 1 is divisible
by x3 + x2 + 1.
10. Find all real polynomials P (x) such that

xP (x − 1) = (x − 2)P (x).

11. Let P (x) be an integer polynomial such that the equation P (x) = 5 has five distinct
integer solutions.
Prove that the equation P (x) = 8 has no integer solutions.
12. Find all real polynomials P (x) such that

P (x)P (x + 1) = P (x2 ).
9.0 Problems 151

13. (a) Does there exist a non-constant integer polynomial P (x) such that the sequence
P (1), P (2), P (3), . . . consists only of primes?
(b) Does there exist a non-constant real polynomial P (x) such that the sequence
P (1), P (2), P (3), . . . consists only of primes?
14. Find all real polynomials P (x) such that, if a is a real number and P (a) is an integer,
then a must also be an integer.
15. (a) Prove that the polynomial

x5 − 5x3 + 10x2 − 15x + 20

is irreducible over Q.
(b) Let P (x) = an xn + an−1 xn−1 + · · · + a0 be an integer polynomial and suppose
that there exists a prime p such that

p2 - a0 , p | a0 , p | a1 , ..., p | an−1 and p - an .

Prove that P (x) is irreducible over Q.1

(c) Prove that the polynomial

2xn + 4xn−1 + 12x + 3

is irreducible over Q.
16. (a) Show that there is a unique polynomial p(x) with integer coefficients in the set
{0, 1, 2, 3, 4, 5, 6, 7, 8, 9} such that p(−2) = p(−5) = 10.
(b) What happens if p(−2) = p(−5) = 2009 instead?
17. Let p(x) be a polynomial with integer coefficients. Suppose that there are integers
x1 , x2 , . . . , xn such that

p(x1 ) = x2 , p(x2 ) = x3 , ..., p(xn−1 ) = xn and p(xn ) = x1 .

Prove that x1 = x3 .
18. (a) Which polynomials can be written as a finite sum of cubes of polynomials with
real coefficients?
(b) Which polynomials can be written as a finite sum of cubes of polynomials with
integer coefficients?
19. Let a1 , a2 , . . . , a100 , b1 , b2 , . . . , b100 be distinct real numbers. For each i and j the number
ai − bj is written in the ith row and jth column of a 100 × 100 table. Suppose that the
product of the numbers in each column is equal to 1.
Prove that the product of the numbers in each row is equal to −1.
20. Show that the set of real numbers x which satisfy the inequality
70
X k 5
≥
x−k 4
k=1

is a union of disjoint intervals, the sum of whose lengths is 1988.

1 This result is commonly known as Eisenstein’s criterion.
152 9 Polynomials

21. Let p(x) be a polynomial of degree n with the property that

k
p(k) = , for k = 0, 1, 2, . . . , n.
k+1

Determine p(n + 1).

22. Let p(x) be a polynomial of degree n, with the property that

p(k) = 2k , for k = 0, 1, 2, . . . , n.

Determine p(n + 1).

23. Let p(x) be a polynomial of degree 10. Suppose that

p(0) = p(2) = p(4) = p(6) = p(8) = p(10) = 0

and
p(1) = p(3) = p(5) = p(7) = p(9) = 1.

What is the value of p(11)?

24. Let p(x) be a polynomial of degree 3n such that

p(0) = p(3) = · · · = p(3n) =2

p(1) = p(4) = · · · = p(3n − 2) = 1
p(2) = p(5) = · · · = p(3n − 1) = 0,

and p(3n + 1) = 730.

Find n.

25. For any polynomial p(x) of degree at most n, prove that

n  
X n+1
p(n + 1) = (−1)n−k p(k).
k
k=0

26. Let k be a given positive integer and let F (x) be an integer polynomial satisfying

0 ≤ F (c) ≤ k, for c = 0, 1, . . . , k + 1.

(a) If k ≥ 4, prove that F (0) = F (1) = · · · = F (k + 1).

(b) Find counterexamples for when k < 4.

27. Let k be an integer and let

p(x) = x6 + x5 + x4 + x3 + x2 + k.

(a) Prove that p(x) does not contain a cubic factor.

(b) If k = 9 or 43, prove that p(x) is irreducible.

28. Let P (x) and Q(x) be polynomials whose coefficients are all equal to 1 or 7.
If P (x) divides Q(x), prove that 1 + deg P (x) divides 1 + deg Q(x).
9.0 Problems 153

29. Find all polynomials p(x, y) with real coefficients such that

p(u, p(v, w)) = p(u + v, w)

for all real u, v, w.

30. The graph of a monic cubic polynomial contains the vertices of exactly one square.
What is the area of the square?
31. Prove that every real polynomial can be multiplied by a non-zero real polynomial to
obtain a polynomial whose exponents are all divisible by 1000.

32. Find all integer polynomials f (x) such that f (a) = f (b) for infinitely many pairs of
integers a, b with a 6= b.
33. Let F (x) be a real polynomial of degree n.
Show that F (x) has n real roots if and only if it is not possible to write

F (x)2 = G(x)2 + H(x)2

for non-zero real polynomials G(x) and H(x) of different degrees.

34. Let P (x) be a polynomial of degree n > 1 with integer coefficients and let k be a positive
integer. Consider the polynomial

Q(x) = P (P (· · · P (P (x)) · · · )),

where P occurs k times.

Prove that there are at most n integers t such that Q(t) = t.
154 9 Polynomials

9.1 Identity theorem

We start with a very basic proposition: a polynomial which is zero infinitely often must
actually be the zero polynomial. This is stated as the identity theorem below, along with a
couple of simple variants.

Identity theorem
If a polynomial has infinitely many roots, then it is the zero polynomial.
If two polynomials satisfy P (x) = Q(x) for infinitely many values of x, then the
polynomials P (x) and Q(x) are equal.

If two polynomials of degree at most n satisfy P (x) = Q(x) for n + 1 values of x, then
the polynomials P (x) and Q(x) are equal.

You should definitely try your hand at proving the identity theorem. It should be well within
your grasp, particularly if you are acquainted with the following important results.

Factor theorem and remainder theorem The factor theorem states that the number
r is a root of P (x) if and only if P (x) is divisible by x − r. More generally, the remainder
theorem states that P (r) = c if and only if the remainder after dividing P (x) by x − r is c.

Problem Find all real polynomials P (x) which satisfy

P (0) = 0 and P (x2 + 1) = P (x)2 + 1.

Solution If we substitute x = 0 into the given equation, we obtain P (1) = 1. If we substitute

x = 1, we obtain P (2) = 2. If we substitute x = 2, we obtain P (5) = 5.
Continuing in this fashion, we will find infinitely many values of x for which P (x) = x. Thus,
P (x) − x is a polynomial with infinitely many roots so, by the identity theorem, it’s the zero
polynomial. Therefore, P (x) = x.

This proof is a little sloppy. The following is a brief sketch of how you should write it down,
given that you want to produce a completely rigorous proof.
You might start by explicitly defining the sequence 0, 1, 2, 5, . . . that we are interested in by
defining a sequence a0 , a1 , . . . which satisfies

a0 = 0 and an+1 = a2n + 1 for n = 0, 1, 2 . . ..

Next, you would show that this is an increasing sequence. Then you would prove that
P (an ) = an for each n, by induction.
If you have done all of this correctly, then you now have infinitely many values of x for which
P (x) = x and can invoke the identity theorem just as we did above.

9.2 Division algorithm

If you are given two polynomials, then you can add them, subtract them or multiply them to
obtain another polynomial. However, you cannot in general divide them to obtain another
polynomial. This situation should remind you of the one we have when we deal with integers.
In that case, we know that for any two integers a and b 6= 0, there is a unique way to write
9.3 Fundamental theorem of algebra 155

a = qb + r, where 0 ≤ r < |b|. One of the lessons you will hopefully learn from this chapter
is the fact that many concepts from number theory have natural analogues in the world of
polynomials. The division algorithm provides just one example.

Division algorithm for polynomials For any two polynomials A(x) and B(x) 6= 0, there
is a unique way to write
A(x) = Q(x)B(x) + R(x),
where deg R(x) < deg B(x).

You can convince yourself that this is true by performing polynomial long division on a few
well chosen pairs of polynomials.

Problem Let P (x) be a real polynomial satisfying P (a) = A and P (b) = B, where a 6= b.
Determine the remainder when P (x) is divided by (x − a)(x − b).

Solution The division algorithm asserts that there is a unique way to write
P (x) = Q(x)(x − a)(x − b) + R(x),
where R(x) is linear or a constant. Our goal, of course, is to determine R(x). Plugging the
values a and b into this equation, we find that
R(a) = P (a) = A and R(b) = P (b) = B.
Therefore, the two points (a, A) and (b, B) lie on the graph of y = R(x), which we know to
be a line. So we can easily deduce that the desired remainder is
A−B aB − Ab
R(x) = x+ .
a−b a−b
Of course, the answer is nicely symmetric, as we should have expected.

9.3 Fundamental theorem of algebra

One of the most important results concerning polynomials, as evidenced by its grand name,
is the fundamental theorem of algebra. This hugely powerful theorem provides us with a first
route of attack against many polynomial problems.

Fundamental theorem of algebra A complex polynomial P (x) of degree n can be

completely factorised as

P (x) = A(x − r1 )(x − r2 ) · · · (x − rn ),

where r1 , r2 , . . . , rn are complex numbers. The roots r1 , r2 , . . . , rn may include repetition,

and the number of times a root occurs is called its multiplicity.

For the remainder of this chapter, we will never again mention polynomials with complex
coefficients. So, you might be wondering, why is the fundamental theorem of algebra so useful
to us? It’s simply because we’ll be dealing with polynomials whose coefficients are integers,
rationals and real numbers, and these are all just particular examples of complex numbers.

Problem Find all real polynomials P (x) which satisfy the equation
(x − 16)P (2x) = 16(x − 1)P (x).
156 9 Polynomials

Solution By the fundamental theorem of algebra, we may factorise the polynomial as

P (x) = A(x − r1 )(x − r2 ) · · · (x − rn ).

Now substitute this into both sides of the given equation—a good way to start many a
polynomial problem. The left-hand side is
 r1   rn 
A(x − 16)(2x − r1 ) · · · (2x − rn ) = 2n A(x − 16) x − ··· x − ,
2 2
while the right-hand side is

16A(x − 1)(x − r1 ) · · · (x − rn ).

Since these should be equal, we can compare leading coefficients to obtain

2n A = 16A,

from which it follows that n = 4. We can now compare roots to obtain the fact that the
following two sets contain exactly the same elements.
n r1 r2 r3 r4 o
16, , , , = {1, r1 , r2 , r3 , r4 }
2 2 2 2
r1
Without loss of generality, let r1 = 16 which implies that 2 = 8. Then, without loss of
generality, let r2 = 8 which implies that r22 = 4.
Proceeding in this fashion yields

P (x) = A(x − 2)(x − 4)(x − 8)(x − 16),

where A can be any real number. As usual, you should substitute this back into the original
equation to check that it actually works, and you will find that it certainly does.

9.4 Vieta’s formulas

If you want to describe a polynomial to somebody, then you can either give them all of its
coefficients or give them just the leading coefficient and all of its complex roots.

an xn + an−1 xn−1 + · · · + a0 or an (x − r1 )(x − r2 ) · · · (x − rn )

There are nice relations between these two descriptions, which you obtain by expanding out
the second expression and comparing coefficients with the first.

Vieta’s formulas If the polynomial an xn + an−1 xn−1 + · · · + a0 has roots r1 , r2 , . . . , rn ,

then the following formulas hold.
an−1
r1 + r2 + · · · + rn = −
an
an−2
r1 r2 + r1 r3 + · · · + rn−1 rn =
an
..
.
a0
r1 r2 · · · rn = (−1)n
an

Observe that the signs on the right-hand sides of these equations alternate between negative
and positive. The first equation involves the sum of the roots, the second equation involves
9.4 Vieta’s formulas 157

the sum of the products of the roots taken two at a time, the third equation involves the sum
of the products of the roots taken three at a time, and so on. Such expressions, known as
elementary symmetric functions, appear in the following important result.

Fundamental theorem of symmetric polynomials A polynomial in the variables

r1 , r2 , . . . , rn is called symmetric if it remains the same when you swap ri with rj for any i
and j.
The theorem tells us that every symmetric polynomial in the variables r1 , r2 , . . . , rn can be
expressed in terms of the elementary symmetric functions.

Problem Suppose that P (x) is a degree three polynomial with roots p, q, r such that
   
1 1
P +P − = 82P (0).
5 5

Determine the value of

1 1 1
+ + .
pq qr rp

Solution If we let P (x) = ax3 + bx2 + cx + d, then the given equation becomes
   
a b c a b c
+ + +d + − + − + d = 82d.
125 25 5 125 25 5
This simplifies to
2
b + 2d = 82d,
25
from which we quickly obtain
b
= 1000.
d
But from Vieta’s formulas, we know that
1 1 1 −a(p + q + r) b
+ + = = = 1000.
pq qr rp −apqr d

Now let’s turn our attention to something just a little more difficult.

Problem Given a positive integer n, let 1, r1 , r2 , . . . , rn be the roots of the polynomial

P (x) = xn+1 − 1.
Show that
1 1 1 n
+ + ··· + = .
1 − r1 1 − r2 1 − rn 2

Although the left-hand side of the equation appears to involve a symmetric expression of the
roots of P (x), the root 1 is missing. So our approach might be to find a polynomial whose
roots are precisely r1 , r2 , . . . , rn . That polynomial would be
P (x)
= xn + xn−1 + xn−2 + · · · + 1.
x−1
Subsequently, your strategy might be to expand out the given expression, rely on Vieta’s
formulas, and hope for the best. But such an approach would create a huge mess, requiring
you to put the given expression over a common denominator, followed by excessive amounts
of algebra. However, a much slicker approach is available.
158 9 Polynomials

Solution Rather than considering the polynomial with r1 , r2 , . . . , rn as roots, we consider

the polynomial with 1 − r1 , 1 − r2 , . . . , 1 − rn as roots.
For ease of notation, let’s use the substitution sk = 1 − rk for k = 1, 2, . . . , n. Since each rk
satisfies the equation rkn+1 = 1, each sk satisfies
(1 − sk )n+1 = 1.

Therefore, s1 , s2 , . . . , sn are roots of the polynomial

(1 − x)n+1 − 1.

Since this polynomial has degree n + 1, it must have n + 1 complex roots counting multiplicity.
So just as we previously excluded the root 1, we now must exclude the root 0. In summary,
s1 , s2 , . . . , sn are precisely the roots of the polynomial
(1 − x)n+1 − 1
Q(x) = .
x
Now if we write Q(x) = an xn + an−1 xn−1 + · · · + a0 , then Vieta’s formulas tell us that
1 1 1 s2 s3 · · · sn + · · · + s1 s2 · · · sn−1 a1
+ + ··· + = =− .
s1 s2 sn s1 s2 · · · sn a0
n+1
−1
In order to obtain the coefficients of Q(x), we simply need to expand (1−x)x using the
good old binomial formula.
     
n+1 n+1 2 n+1
Q(x) = − +x −x + · · · + (−1)n+1 xn
1 2 3
Therefore, we conclude that
n+1


1 1 1 2 n
+ + ··· + = n+1

= .
s1 s2 sn 1
2

9.5 Integer polynomials

Within the class of real polynomials lies the class of integer polynomials, which is a fascinating
world of its own. There are many new techniques and results which only apply to integer
polynomials. One example is the following, which is known as the rational root theorem.

Rational root theorem Let P (x) = an xn + an−1 xn−1 + · · · + a0 be an integer polynomial.

If p(x) has a rational root which can be expressed as rs in lowest terms, then r | a0 and s | an .

Problem Prove the rational root theorem.

r
Solution If sis a root of the polynomial, then we have the equation
 r n  r n−1 r
an + an−1 + · · · + a1 + a0 = 0.
s s s
It’s only natural now to clear the equation of fractions. Multiplying both sides by sn gives
an rn + an−1 rn−1 s + · · · + a1 rsn−1 + a0 sn = 0.
Since r divides every term of the equation except a0 sn , it follows that r also divides a0 sn
itself. However, we’ve assumed that r and s are relatively prime, so it must be the case that
r | a0 . An analogous argument shows that s | an .
9.5 Integer polynomials 159

The following is another useful little result, very easy to prove, which applies only to integer
polynomials.

Lemma If P (x) is a polynomial with integer coefficients, then

a − b | P (a) − P (b)

for all integers a and b.

Problem Suppose that P (x) is an integer polynomial and a is an integer satisfying

P (P (P (P (a)))) = a.

Prove that P (P (a)) = a.

Solution Consider the sequence defined by a0 = a and an+1 = P (an ) for all non-negative
integers n. The sequence consists entirely of integers and we have a4 = P (P (P (P (a)))) = a0 .
It follows that a5 = a1 , a6 = a2 , a7 = a3 , and so on. In other words, the sequence is periodic
with period four.
By the lemma above, we have

ak − ak−1 | P (ak ) − P (ak−1 ) = ak+1 − ak

for every positive integer k. In particular, we have the following chain of divisibilities.

a1 − a0 | a2 − a1 | a3 − a2 | a0 − a3 | a1 − a0

If a term appearing in this chain is equal to 0, then they must all be, since 0 doesn’t divide
any non-zero integer. In this case, we would have a1 = a0 or equivalently, P (a) = a, which
immediately implies that P (P (a)) = a.
Otherwise, all terms appearing in the chain are non-zero. Since a | b implies that b = 0 or
|a| ≤ |b|, we have the following chain of inequalities.

|a1 − a0 | ≤ |a2 − a1 | ≤ |a3 − a2 | ≤ |a0 − a3 | ≤ |a1 − a0 |

Naturally, this implies the following chain of equalities.

|a1 − a0 | = |a2 − a1 | = |a3 − a2 | = |a0 − a3 | = |a1 − a0 |

If a1 − a0 = a2 − a1 = a3 − a2 = a0 − a3 = m, then it’s simple to deduce that m = 0,

contradicting the fact that each term in the chain is non-zero.
So there must be a value of k for which ak − ak−1 = −(ak+1 − ak ), which implies that
ak+1 = ak−1 . However, this implies that ak+2 = ak , ak+3 = ak+1 , ak+4 = ak+2 , and so on.
Sooner or later, since the sequence is periodic, we obtain an equation which is equivalent to
a2 = a0 , which implies that P (P (a)) = a.

If you fully understand the previous solution, then you should have no problem in proving
the following, more general, statement.

If P (x) is an integer polynomial and a is an integer satisfying

P (P (· · · P (P (a)) · · · )) = a,

then P (P (a)) = a.
160 9 Polynomials

9.6 Complex numbers

It’s an amazing fact that, even if you only care about real polynomials, it pays to know
something about complex numbers. For example, the fundamental theorem of algebra asserts
that every complex polynomial of degree n has precisely n complex roots. In the case when
the polynomial has real coefficients, these roots must obey the following theorem.

Conjugate root theorem If P (x) is a real polynomial, then z is a root of multiplicity

m if and only if its conjugate z is a root of multiplicity m. Therefore, the roots of a real
polynomial are real or occur in complex conjugate pairs.

If you are familiar with complex conjugation, you should be able to prove the conjugate root
theorem without too much trouble.

Problem Prove that any real polynomial can be written as a product of real linear and real
quadratic polynomials.

Solution The conjugate root theorem allows us to list the real roots of P (x) as r1 , r2 , . . . , rk
and the non-real roots as z1 , z1 , z2 , z2 , . . . , zm , zm . In other words, for some real number A,
we can write

P (x) = A(x − r1 ) · · · (x − rk )(x − z1 )(x − z1 ) · · · (x − zm )(x − zm ).

However, if z = a + bi for real numbers a and b, then

(x − z)(x − z) = (x − a − bi)(x − a + bi) = (x − a)2 + b2 .

This means that each complex conjugate pair of roots contributes a real quadratic factor to
P (x). Clearly, each real root contributes a real linear factor to P (x), so we’re done.

9.7 Algebraic trickery

Sometimes, solving a polynomial problem just comes down to thinking of the right algebraic
identity or manipulation. Therefore, you should try to become an algebraic mathemagician
and know all the tricks! The first one we’ll look at involves a simple substitution which you
can think of as shifting the graph of the polynomial to the left or right.

Problem Show that the polynomial

P (x) = x7 − 2x5 + 10x2 − 1

has no real root greater than 1.

Solution Normally it’s quite difficult to tell, simply by looking at a polynomial, whether
or not it has a real root greater than 1. However, it’s often easier to tell whether or not a
polynomial has a real root greater than 0, as we’ll soon see.
Motivated by these reasons, let’s set x = y + 1 and substitute into the expression for P (x) to
obtain

P (y + 1) = (y + 1)7 − 2(y + 1)5 + 10(y + 1)2 − 1

= y 7 + 7y 6 + 19y 5 + 25y 4 + 15y 3 + 11y 2 + 17y + 8.
9.7 Algebraic trickery 161

Note that P (x) has a real root greater than 1 if and only if P (y + 1) has a real root greater
than 0. However, it’s clear by inspection that P (y + 1) is positive whenever y is positive. So
we can conclude that P (y + 1) has no real root greater than 0. Therefore, P (x) has no real
root greater than 1.

For the next problem, we use the fact that the product of two sums of two perfect squares is
another sum of two perfect squares. This follows from the algebraic identity

(A2 + B 2 )(C 2 + D2 ) = (AC − BD)2 + (AD + BC)2 .

You could prove this simply by expanding both sides and verifying that they are equal. For
the complex numbers expert, a more insightful approach is to recognise that the result is
equivalent to
|A + Bi| × |C + Di| = |(A + Bi) × (C + Di)|.
Either way, this is one of those remarkable formulas that every good mathemagician should
know!
Furthermore, this identity is not just true when A, B, C and D are numbers, but any
quantities that you can add, subtract and multiply. In particular it is true if A, B, C and D
are functions, and of course these include polynomials.

Problem Let P (x) be a real polynomial such that P (x) ≥ 0 for all real x. Prove that it’s
possible to write
P (x) = F (x)2 + G(x)2
for real polynomials F (x) and G(x).

Solution If P (x) has a real root r with odd multiplicity, then the sign of the graph y = P (x)
will change from positive to negative or vice versa at x = r. Therefore, every real root of
P (x) must have even multiplicity.
Now we use the result from section 9.6 to write

P (x) = A(x − r1 )2 · · · (x − rk )2 (x − z1 )(x − z1 ) · · · (x − zm )(x − zm ).

Here, the real number A must be non-negative to ensure that P (x) ≥ 0 for all real x. Therefore,
we can write
P (x) = R(x)2 Z1 (x)Z2 (x) · · · Zm (x),
where √
R(x) = A(x − r1 )(x − r2 ) · · · (x − rk )
and
Zi (x) = (x − zi )(x − zi ).

However, if z = a + bi for real numbers a and b, then

(x − z)(x − z) = (x − a − bi)(x − a + bi) = (x − a)2 + b2 .

Therefore, we can write each of Z1 (x), Z2 (x), . . . , Zm (x) as a sum of two perfect squares.
Since R(x)2 = R(x)2 + 02 is also a sum of two perfect squares, we can repeatedly use our
mathemagical identity

(A2 + B 2 )(C 2 + D2 ) = (AC − BD)2 + (AD + BC)2 ,

to write the product P (x) = R(x)2 Z1 (x)Z2 (x) · · · Zm (x) as a sum of two perfect squares.
162 9 Polynomials

9.8 Irreducibility
By the fundamental theorem of algebra, every polynomial can be factorised into linear factors,
if we are allowed to use complex numbers. But given a polynomial with integer coefficients,
one might ask whether or not it can be factorised into two polynomials, each with integer
coefficients themselves. Of course, you can take out a factor of 1 or perhaps some larger
integer, but this is totally boring. So we say that a polynomial F (x) is reducible over Z if it
can be written as a product F (x) = G(x)H(x), where G(x) and H(x) are integer polynomials
of positive degree. If this is not possible, then we say that F (x) is irreducible over Z. Similarly
if F (x) = G(x)H(x), where G(x) and H(x) are rational polynomials of positive degree, we
say that F (x) is reducible over Q and otherwise, that it is irreducible over Q.
Clearly, if an integer polynomial is reducible over Z, then it’s reducible over Q, because every
integer is certainly rational. More surprisingly, the converse is true as well.

Gauss’ lemma If an integer polynomial is reducible over Q, then it’s reducible over Z. So,
for an integer polynomial, reducibility over Z and reducibility over Q are the same thing.

Showing that a given polynomial is irreducible can be very difficult. We’ll see a few different
techniques as we go on, but for the following problem, we’ll use a rather direct approach.

Problem If a1 , a2 , . . . , an are distinct integers, prove that the polynomial

F (x) = (x − a1 )(x − a2 ) · · · (x − an ) − 1

is irreducible over Z.

Solution Suppose that F (x) = G(x)H(x), where G(x) and H(x) are integer polynomials of
positive degree. Clearly, F (x) has degree n, so we know that

deg G + deg H = n.

Now substituting x = ak for some value of k yields

G(ak )H(ak ) = −1.

Therefore, G(ak ) and H(ak ) must be integers which multiply to give −1, which leads to
G(ak ) = 1 and H(ak ) = −1 or vice versa. In either case, we have

G(ak ) + H(ak ) = 0.

In summary, we have deduced that the polynomial

P (x) = G(x) + H(x)

has the distinct roots a1 , a2 , . . . , an , and possibly more. So, if P (x) is a non-zero polynomial,
then it would have to have degree at least n. But the degree of P (x) is at most the maximum
of deg G and deg H, which is strictly less than n by assumption. So this case cannot occur.
The other possibility is that P (x) is the zero polynomial. But then G(x) = −H(x) and so,

F (x) = −G(x)2 .

There is a problem here: the leading coefficient of −G(x)2 is clearly negative, while the
leading coefficient of F (x) is clearly positive.
From these contradictions, we can conclude that F (x) is irreducible over Z.
9.9 Factorisation 163

We finish this section with a useful fact concerning irreducible integer polynomials which you
might like to prove.

Proposition Suppose that r is a root of an irreducible rational polynomial P (x). If r is

also a root of a rational polynomial Q(x), then it must be the case that P (x) | Q(x).

9.9 Factorisation
Yet another piece of algebraic trickery that arises in polynomial problems is the difference of
perfect powers factorisation.

xn − y n = (x − y)(xn−1 + xn−2 y + · · · + y n−1 )

This often appears when one is dealing with geometric series.

Problem Prove that there are no primes in the integer sequence

10001, 100010001, 1000100010001, . . ..

Solution The trick is to write x = 10, so that we can express the terms of the sequence as

1 + x4 + x8 + · · · + x4k .

Now we apply two tactics. The first is to use the factorisation above to write this geometric
series in closed form, and the second is to further factorise the result, which features differences
of perfect squares.
x4k+4 − 1 (x2k+2 + 1) (x2k+2 − 1)
4
= ×
x −1 (x2 + 1) (x2 − 1)
Now the roots of x2 − 1 are ±1 and it’s easy to check that these are both roots of x2k+2 − 1.
Similarly, the roots of x2 + 1 are ±i, and it’s easy to check that these are both roots of
x2k+2 + 1 if k is even, or x2k+2 − 1 if k is odd.
In either case, after dividing the numerator by the denominator, we are left with the product
of two integer polynomials. For k ≥ 2, both of these polynomials are guaranteed to have
positive degree. Furthermore, it’s easy to check (and you should do so right now) that they
cannot be equal to 1 when x = 10. Thus, every term of the sequence after the first is a
product of two numbers greater than 1 and hence, not prime.
For the first term of the sequence, our clever polynomial argument doesn’t work. However,
we can compute directly that 10001 = 73 × 137.

9.10 Polynomials modulo p (upstairs–downstairs)

Many of the tricks and techniques from number theory carry over to the study of polynomials.
We’ve already witnessed the division algorithm for polynomials in action and now it’s time to
consider modular arithmetic for polynomials. We will be reducing the coefficients—but never
the exponents—of an integer polynomial modulo p. For example, it makes sense to write a
statement like

5x6 − x4 + 2x3 − 10x2 + x + 3 ≡ 4x4 + 2x3 + x + 3 (mod 5).

In this way, polynomials can be added, subtracted, and multiplied modulo p.

164 9 Polynomials

It’s almost always useful to take p to be prime, in which case you can rely on the division
algorithm along with many of the results you already know and love which hold for polynomials
in general. For example, we require p to be a prime for the following simple, though useful,
statement to be true.
Theorem Let p be a prime. Suppose that

F (x)G(x) ≡ 0 (mod p),

then
F (x) ≡ 0 (mod p) or G(x) ≡ 0 (mod p).

Another particularly useful true statement that requires p to be prime is the following analogue
of the fundamental theorem of arithmetic.
Unique factorisation for polynomials modulo p If p is a prime, then any factorisation
of a polynomial into irreducible factors modulo p is unique up to the order of its factors.

(We did not mention it earlier, but unique factorisation also holds for ordinary integer
polynomials.)
One of the most useful reasons to consider polynomials modulo p is to prove irreducibility. If
F (x) can be factorised as F (x) = G(x)H(x), then we know that F (x) ≡ G(x)H(x) (mod p).
In other words, if a polynomial is reducible over the integers, then it’s reducible over the
integers modulo p. The contrapositive of this statement is particularly useful: if a polynomial
is irreducible over the integers modulo p, then it’s irreducible over the integers.
The beauty of this technique is that modulo p there are only finitely many polynomials of
degree lower than a given polynomial. This means that you can check them all! Furthermore,
you are free to choose whichever value of p happens to work for you. However, one must
beware! If a polynomial is reducible over the integers modulo p, then it definitely does not
follow that it’s reducible over the integers.

Problem Prove that the polynomial

F (x) = x5 − x2 + 1

is irreducible over Z.

Solution We simply consider F (x) modulo 2. For the remainder of this proof, all congruences
are assumed to be considered modulo 2.
Suppose that F (x) is reducible over the integers modulo 2. Then we can write

F (x) ≡ G(x)H(x),

where G(x) and H(x) have positive degree.

Since deg G + deg H = 5, one of G(x) and H(x) must have degree at most 2. Without loss
of generality, assume that it’s G(x). So G(x) has degree 1 or 2 and, furthermore, we may
assume that G(x) is irreducible.
Working modulo 2, there are only two polynomials of degree 1, namely, x and x + 1. It’s easy
to check that neither is a factor of F (x) modulo 2. For example, you can use polynomial
division to obtain the following.

x5 − x2 + 1 ≡ (x4 + x) · x + 1
x5 − x2 + 1 ≡ (x4 + x3 + x2 ) · (x + 1) + 1
9.10 Polynomials modulo p (upstairs–downstairs) 165

Similarly, there are only four polynomials modulo 2 and of degree 2, namely, x2 , x2 + 1, x2 + x
and x2 + x + 1. To make things even simpler, we have

x2 ≡ x · x, x2 + 1 ≡ (x + 1) · (x + 1) and x2 + x ≡ x · (x + 1),

so that x2 + x + 1 is the only irreducible quadratic modulo 2. Since

x5 − x2 + 1 ≡ (x3 + x2 ) · (x2 + x + 1) + 1,

this is also not a factor of F (x). So F (x) is irreducible over the integers modulo 2 and hence,
irreducible over the integers.

Often, we think of working with integers as upstairs and working with integers modulo p
as downstairs. For more difficult problems, we may need to move between upstairs and
downstairs quite often, transferring information between the two levels.

Problem Prove that the polynomial

F (x) = xn + 5xn−1 + 3

is irreducible over Z for every integer n > 1.

Solution In this solution, downstairs will refer to working modulo 3. Suppose that F (x) is
reducible, so that we can write F (x) = G(x)H(x) for polynomials G(x) and H(x) of positive
degree.
Downstairs, we have
F (x) ≡ xn−1 (x + 5),
which, by unique factorisation modulo p, can only be non-trivially factorised into two
polynomials as
xk (x + 5) × xn−1−k ,
for some k = 0, 1, 2, . . . , n − 2. So, without loss of generality, we have

G(x) ≡ xk (x + 5) and H(x) ≡ xn−1−k .

If k is positive, then moving upstairs tells us that the constant terms of both G(x) and H(x)
are divisible by 3. This implies that the constant term of F (x) is divisible by 9, a rather
blatant contradiction.
So, k = 0 and our equations downstairs become

G(x) ≡ x + 5 and H(x) ≡ xn−1 .

Going back upstairs we must have deg G(x) ≥ 1 and deg H(x) ≥ n−1. But G(x)H(x) = F (x),
which has degree n. Thus, deg G(x) = 1 and deg H(x) = n − 1.
Since G(x) is linear it has a rational root, which we can express as rs in lowest terms. But
then rs is a root of F (x) and the rational root theorem implies that r | 3 and s | 1. In short,
F (x) must have ±1 or ±3 as a root.
We can substitute these four values into the expression xn + 5xn−1 + 3 to verify that they are
certainly not roots, which grants us our desired contradiction.
166 9 Polynomials

9.11 Polynomials modulo P (x)

Perhaps surprisingly, we can also reduce a polynomial modulo another polynomial. All the
standard rules of modular arithmetic still apply; but, as usual, it’s best to see an example.

Problem Let a, b, c and d be positive integers.

Show that the polynomial
x4a+3 + x4b+2 + x4c+1 + x4d
is divisible by x3 + x2 + x + 1.

Solution There are a number of ways to solve this problem. One is to find the roots of
x3 + x2 + x + 1 and show that they’re also roots of the given polynomial. But we’re here to
learn about polynomial modular arithmetic, so our approach will be to show that the given
polynomial is congruent to zero modulo x3 + x2 + x + 1.
Since x4 − 1 = (x − 1)(x3 + x2 + x + 1), we have
x4 ≡ 1 (mod x3 + x2 + x + 1).
So by working modulo x3 + x2 + x + 1, we obtain
x4a+3 + x4b+2 + x4c+1 + x4d = x3 (x4 )a + x2 (x4 )b + x(x4 )c + (x4 )d
≡ x3 + x2 + x + 1
≡ 0.

9.12 Lagrange interpolation

The identity theorem tells us that, in general, two points determine a line, three points
determine a quadratic, four points determine a cubic, and so on. So somebody could say
to you that P (x1 ) = y1 , P (x2 ) = y2 , . . . , P (xn+1 ) = yn+1 and challenge you to work out the
degree n of the polynomial P (x). Of course, you could just get scared and run away, but
it would be much better if you knew that there is a big fat formula called the Lagrange
interpolation formula to answer this question precisely. Even better than knowing the formula
itself, which is quite unwieldy, is to understand where it comes from.

Problem If x1 , x2 , . . . , xn+1 are distinct numbers, find a polynomial of degree at most n

which satisfies
P (x1 ) = P (x2 ) = · · · = P (xn+1 ) = 0,
except that P (xk ) = yk , for one particular value of k.

Solution If yk = 0, then the problem is easy, since the identity theorem forces us to take
P (x) = 0. Otherwise, the problem is almost as easy, since we know all of the roots of the
polynomial.
P (x) = A(x − x1 ) · · · (x\
− xk ) · · · (x − xn+1 ).
That strange looking hat notation is telling the term (x − xk ) to politely leave the building—it
is excluded from the product. We can then substitute x = xk to derive an expression for A.
The final answer is
(x − x1 ) · · · (x\
− xk ) · · · (x − xn+1 )
P (x) = yk .
\
(xk − x1 ) · · · (xk − xk ) · · · (xk − xn+1 )
9.13 Root focus 167

Problem If x1 , x2 , . . . , xn+1 are distinct numbers, find a polynomial of degree at most n

which satisfies
P (x1 ) = y1 , P (x2 ) = y2 , . . . , P (xn+1 ) = yn+1 .

Solution Simple! We just add up the polynomials that we obtained in the previous problem.
So the final answer is
n+1
X (x − x1 ) · · · (x\
− xk ) · · · (x − xn+1 )
P (x) = yk .
k=1
\
(xk − x1 ) · · · (xk − xk ) · · · (xk − xn+1 )

It’s this result that is known as the Lagrange interpolation formula. Once again, remember
that it’s not the actual formula itself that you should commit to memory, but the method
used to arrive at the formula.

9.13 Root focus

It just might be possible to get some information about a root of a polynomial. Then with
some insight you just might be able to use this to your advantage as in the following problem.

Problem If p is an odd prime and n is a positive integer, prove that the polynomial
F (x) = xn + x + p
is irreducible over Z.

Solution Suppose that F (x) is reducible, so that we can write

F (x) = G(x)H(x)
for integer polynomials G(x) and H(x) of positive degree.
Since F (0) = G(0)H(0) = p, we may assume without loss of generality that
G(0) = ±1 and H(0) = ±p.

So, if the complex roots of G(x) are r1 , r2 , . . . , rk , then with the help of Vieta’s formulas, we
have the equation
|r1 r2 · · · rk | = 1.

If all these complex roots had magnitude greater than 1, then their product would be greater
than 1, a contradiction. Hence there must be some complex root r of G(x) which satisfies
|r| ≤ 1.
But any root of G(x) is also a root of F (x). So F (x) has a complex root r which satisfies
|r| ≤ 1. This is a very interesting piece of information indeed! This was certainly not
something that was obvious from the problem’s statement.
Continuing, this implies that
rn + r + p = 0
for some r ∈ C with |r| ≤ 1.
Now think about what this equation means geometrically. It means that the triangle in the
complex plane whose vertices are rn , rn + r and rn + r + p = 0 has side lengths |rn | ≤ 1,
|r| ≤ 1 and p ≥ 3. But this directly contradicts the triangle inequality, so we conclude that
xn + x + p is irreducible over Z.
 Functional equations
10
Functional equations are simply equations involving functions! To read this chapter, you’ll
need to have some familiarity with the concept of a function and other basic notions, some of
which may be found in section 17.2. To solve functional equations, you’ll need an assortment
of standard and not-so-standard techniques, many of which may be found in this chapter. As
is common for mathematical Olympiad problems, the concepts themselves are generally quite
simple, but they will be applied in particularly tricky ways.
Hopefully, you are already familiar with the following standard notation, which we’ll use
consistently throughout this chapter.

Z = {integers} Q = {rationals} R = {reals}

+ +
N = {positive integers} Q = {positive rationals} R+ = {positive reals}

10.0 Problems
1. Find all functions f : R → R such that

f (x − f (y)) = 1 − x − y

for all real numbers x and y.

2. Find all functions f : R → R such that

xf (x) + f (1 − x) = x3 − x

for all real numbers x.

3. Find all functions f : R → R such that

f (x + y)f (x − y) = 2x + f (x2 − y 2 )

for all real numbers x and y.

4. Find all functions f : R → R which satisfy the equations

f (xy) = xf (y) + yf (x) and f (x + y) = f (x1001 ) + f (y 1001 )

for all real numbers x and y.

170 10 Functional equations

5. Can you find a continuous function g : R → R such that

g(g(x)) = x

for all x ∈ R, and such that g is not a linear function?

6. Find all functions f : Z → Z such that

f (m + f (n)) = f (m) + n

for all integers m and n.

7. Find all functions f : R \ {0} → R such that

 
1
f (x) = xf
x

for all non-zero real numbers x and

f (x) + f (y) = 1 + f (x + y)

for all non-zero real numbers x and y with non-zero sum.

8. Prove that there exists no function f : Z → Z which satisfies the equation

f (m + f (n)) = f (m) − n

for all integers m and n.

9. (a) Find all strictly increasing or strictly decreasing functions f : R → R such that

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

for all real numbers x and y.

(b) Prove that there is no strictly increasing or strictly decreasing function f : R → R
such that
f (x + f (y)) = f (x) + y 2
for all real numbers x and y.

10. Find all continuous functions f : R → R satisfying

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

for all x, y ∈ R.

11. Prove that there exists no function f : Z → Z which satisfies the equation

f (f (n)) = n + 1

for all integers n.

12. Find all functions f : R → R such that

f (x + y) + f (x + z) + f (y + z) ≥ 3f (x + 2y + 3z)

for all x, y, z ∈ R.
10.0 Problems 171

13. Find all functions f : Z → Z such that

f (m + n) + f (mn) = f (m)f (n) + 1

for all integers m and n.

14. Find all functions f : R → R such that

f (x2 − y 2 ) = (x − y)(f (x) + f (y))

for all real numbers x and y.

15. Find all functions f : Q → R such that

f (xy) = f (x)f (y) − f (x + y) + 1

for all rational numbers x and y.

16. Find all functions f : N+ → N+ such that

f (f (m) + f (n)) = m + n

for all positive integers m and n.

17. Find all functions f : N+ → N+ such that

f (f (m)f (n)) = mn

for all positive integers m and n.

18. Prove that there exists no function f : N+ → N+ which satisfies the equation

f (n) = f (f (n − 1)) + f (f (n + 1))

for all integers n > 1.

19. Find all functions f : N+ → N+ such that

f (n) + f (f (n)) = 2n

for all positive integers n.

20. Find all functions f : Q → Q such that

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

for all rational numbers x and y.

21. Find all monotonic functions f : R → R satisfying

f (4x) − f (3x) = 2x

for all x ∈ R.
22. Let T denote the set of all ordered pairs (a, b) of non-negative integers.
Find all functions f : T → R such that

0
 if ab = 0,
f (a, b) =
 1 + f (a + 1, b − 1) + f (a − 1, b + 1)
 otherwise.
2
172 10 Functional equations

23. Let R+
0 be the set of non-negative real numbers.
Find all functions f : R+ +
0 → R0 such that
√ √
f (x + y − z) + f (2 xz) + f (2 yz) = f (x + y + z)
for every x, y, z ∈ R+
0 such that x + y ≥ z.

24. Find all functions f : R+ → R+ such that

f (w)2 + f (x)2 w 2 + x2
=
f (y 2 ) + f (z 2 ) y2 + z2
for all positive real numbers w, x, y, z satisfying wx = yz.
25. Let n be a given positive integer.
Find all functions f : R → R such that
y
xn f (y) − y n f (x) = f
x
for all real x, y with x 6= 0.
26. Find all functions f : R → R such that
f (f (x) + y) = 2x + f (f (y) − x)
for all real numbers x and y.
27. Determine all bijections f : R → R such that
(i) f (x) is strictly increasing, and
(ii) f (x) + f −1 (x) = 2x for all real x.
28. Find all functions f : R → R such that
f (x4 + y) = x3 f (x) + f (y)
for all real numbers x and y.
29. Let S be the set of real numbers greater than −1. Find all functions f : S → S such
that
(i) f (x + f (y) + xf (y)) = y + f (x) + yf (x) for all x and y in S, and
f (x)
(ii) x is strictly increasing on the intervals −1 < x < 0 and 0 < x.
30. Find all non-decreasing functions f : R → R such that
(i) f (0) = 0, f (1) = 1, and
(ii) f (a) + f (b) = f (a)f (b) + f (a + b − ab) for all real numbers a, b such that a < 1 < b.
31. Let R+ + +
0 be the set of non-negative real numbers. Suppose f is a function f : R0 → R0
satisfying the conditions
(i) f (a) = 0 if and only if a = 0
(ii) f (ab) = f (a)f (b)
(iii) f (a + b) ≤ f (a) + f (b) + |f (a) − f (b)|
for all a, b ∈ R+
0.
Prove that
f (a + b) ≤ f (a) + f (b)
for all a, b ∈ R+
0.
10.1 Cauchy’s functional equation 173

10.1 Cauchy’s functional equation

Cauchy’s functional equation asks us to find functions which satisfy

f (x + y) = f (x) + f (y).

However, the choice of domain and codomain that we are interested in will have a massive
effect on what the solutions look like. This is one of the simplest and most fundamental
functional equations, and its solution should be familiar to any competent problem solver!

Problem Let N0 denote the set of non-negative integers.

Find all functions f : N0 → R such that

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

for all non-negative integers x and y.

The idea is to work out the value of f (x) for more and more values of x until we have the
entire function.
Before we proceed, here is a useful strategy for solving functional equations. It’s often a good
idea to guess what the solutions are and, in this case, you should be able to find at least
one without too much trouble! You should keep the suspected answer in the back of your
mind while solving the functional equation, yet remain open to other possibilities. For this
particular problem, one of the easiest solutions to find is f (x) = x. But then you might also
notice that f (x) = 2x is a solution and, in fact, so is f (x) = cx for any real number c.

Solution So how do you start on a problem like this? For almost any functional equation, a
good starting point is the substitution of values for x and y. It’s a free world, so you can use
whatever numbers you like, provided they are in the function’s domain. In practice, however,
simple substitutions such as setting the variables to be 0, 1, −1 or equal to each other are the
best to start with. For this problem, the substitution x = y = 0 yields f (0) + f (0) = f (0)
which then implies that f (0) = 0.
Next, you might try to figure out the value of f (1). But, try as you might, you won’t be
able to do it. If you are too narrow-minded and believe that f (x) = x is the only solution,
then you could get stuck at this point. That’s why you need to remain open-minded to other
possibilities, because if f (x) = cx is a solution for any real number c, then f (1) can take on
any value whatsoever!
With this in mind, let f (1) = c so that we can try to prove that f (x) = cx for all non-negative
integers x. In order to obtain the value of f (2), it’s natural to substitute x = y = 1, which
leads to
f (2) = f (1) + f (1) = 2c.
In fact, putting y = 1 in the functional equation gives

f (x + 1) = f (x) + f (1) = f (x) + c.

Therefore, if f (x) = cx, then f (x + 1) = cx + c = c(x + 1).

From here you should be able to prove by induction that f (x) = cx for all non-negative
integers x and some real number c.
It’s easy, and absolutely necessary, to check that every function of this type satisfies the given
functional equation. As a courtesy to the reader, we will not usually present such routine
checks, but leave them as exercises!
174 10 Functional equations

Problem Find all functions f : Q → R such that

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

for all rational numbers x and y.

Solution Note that all the hard work required to solve the previous problem carries over to
this problem as well. So we already know that f (x) = cx for all non-negative integers x and
some real number c.
Since we have already deduced that f (0) = 0, it makes sense to try the substitution y = −x
in the functional equation. This leads to

f (0) = f (x) + f (−x) ⇒ f (−x) = −f (x).

This piece of information tells us that f (x) = cx holds for every integer, whether positive,
negative or zero.
Now that we have the function on all of Z, it’s time to extend our net to all of Q. The trick
is to notice that Cauchy’s functional equation implies that

f (x + y + z) = f (x + y) + f (z) = f (x) + f (y) + f (z).

Hence, we can expand out sums of not only two numbers, but three numbers, or four numbers,
or even n numbers.1 In particular, for any integer m and positive integer n, we have
 
m m m m m m
f + + ··· + =f +f + ··· + f .
|n n {z n} | n n {z n }
n times n times

Therefore, f (m) = nf ( m
n ) which implies that
m f (m) cm
f = = .
n n n
We’ve now deduced that f (x) = cx for all rational numbers x and some real number c. Once
again, you should check that functions of this type do indeed satisfy the given functional
equation.

We have solved Cauchy’s functional equation where the domain is N0 , Z and Q. So what are
the solutions to Cauchy’s functional equation when the domain is R? You could be forgiven
for believing that all the solutions are linear, as they were in the previous cases, but that just
isn’t true! Although f (x) = cx is a solution, there are in fact many, many more, all of which
are crazy and none of which can be described by any compact formula. To eliminate these
crazy solutions, it’s necessary to impose an extra condition. For example, it’s known that
the solutions to Cauchy’s functional equation on the real numbers are given by f (x) = cx, if
we are also given that f is continuous, that f is monotonic, or that f is bounded on some
interval.

10.2 Guess and hope

We earlier advocated the strategy of trying to guess the solutions to a functional equation.
Although guessing itself is rarely a substantial step toward a proof, a correct guess can
sometimes make the problem a great deal simpler.
1 To be rigorous you could use induction to get the corresponding result for n numbers.
10.3 Substitutions 175

Here’s the idea. Say you have a functional equation and you think that f (x) = x3 is the only
solution. Then the function g(x) = f (x) − x3 would be pretty simple because it would be the
zero function. Another simple function would be h(x) = fx(x)
p
3 or perhaps even k(x) = 3 f (x).
This motivates us to use one of these three substitutions to simplify the original functional
equation. Sometimes, but certainly not always, you may end up with a much easier problem.

Problem Find all functions f : Q → R such that

f (x + y) = f (x) + f (y) + 2xy

for all rational numbers x and y.

Solution The first thing you might notice is the fact that this looks like Cauchy’s functional
equation, except for that pesky 2xy term which ruins everything. You could try similar
methods to those used to solve Cauchy’s functional equation, but the 2xy term makes things
a little difficult.
The second thing you might notice is the fact that 2xy arises when you expand (x + y)2 . In
fact, the given functional equation bears an uncanny resemblance to the formula

(x + y)2 = x2 + y 2 + 2xy.

In fact, this verifies that f (x) = x2 is a solution; maybe not the only solution, but a solution
nonetheless.
Now it’s time to guess and hope! Let’s substitute g(x) = f (x) − x2 , that is, f (x) = g(x) + x2 ,
in the hope that g(x) satisfies a simpler functional equation. This leads to

g(x + y) + (x + y)2 = g(x) + x2 + g(y) + y 2 + 2xy,

and we can cancel terms from both sides to leave

g(x + y) = g(x) + g(y).

You guessed it—Cauchy’s functional equation! We already know that the solutions to this are
given by g(x) = cx for some real number c. Retracing our steps, we find that f (x) = x2 + cx
for some real number c. You can easily substitute this back into the original functional
equation to verify that it is indeed a solution.

10.3 Substitutions
Suppose that you stumble upon a magic black box in which you can place an object and
another object will emerge from the other side. If you want to know how the box works,
you might try passing different objects through it and observing the output, until you can
deduce something about the behaviour of the magic black box. Similarly, if you stumble upon
a functional equation and want to solve it, a natural approach is to substitute particular
numbers or, better still, algebraic expressions for the variables.

Problem Find all functions f : R → R such that f (x) 6= 0 for x 6= 0 and

f (f (x) + y) = f (x2 − y) + 4f (x)y

for all real numbers x and y.

176 10 Functional equations

Solution An obvious first substitution is y = 0, because it knocks out one of the terms on
the right-hand side. In fact, it leads to the equation

f (f (x)) = f (x2 ).

Trying the substitution x = y = 0, we discover that f (f (0)) = f (0). You could continue
trying various combinations of x and y equal to simple numbers like 0, 1 and −1, but you
probably wouldn’t get too far with this approach.
The idea here is to look for algebraic substitutions, rather than numerical ones. In particular,
we would like to find algebraic substitutions which provide nice cancellation. For example,
the appearance of f (f (x) + y) on the left-hand side motivates us to try y = −f (x), which
yields a rather interesting looking equation.

f (0) = f (x2 + f (x)) − 4f (x)2

Furthermore, the term f (x2 − y) on the right-hand side motivates us to try y = x2 , which
turns up yet another interesting looking equation.

f (f (x) + x2 ) = f (0) + 4f (x)x2

You probably can’t help but notice that these two equations bear quite a resemblance to each
other. In fact, we can use them to eliminate the unwieldy term f (f (x) + x2 ) and obtain

f (0) + 4f (x)2 = f (0) + 4f (x)x2 ⇒ f (x)2 = f (x)x2 .

For x = 0, this equation simply tells us that f (0) = 0. But if x 6= 0, then f (x) 6= 0, and so we
can divide through by f (x) ending up with

f (x) = x2 .

Therefore, the only possible solution is f (x) = x2 and you can check that it does indeed
satisfy the original functional equation.

This problem involved some algebraic trickery! Such deviousness is often the result of a
long period of time spent playing around and substituting different combinations of the
variables involved. In the end, this is a matter of intuition and experience. However, note
what motivated us to substitute y = x2 . We turned f (x2 − y) into f (0). You should always
be on the lookout for substitutions which give similarly nice results.

10.4 Injective, surjective and bijective

A function is said to be injective or one-to-one if it doesn’t take the same value twice. That
is, f (x) = f (y) implies that x = y. But how do you prove that a function is injective? The
most direct way is just to assume that f (x) = f (y) and use the functional equation somehow
to deduce that x = y. But there are other ways, such as invoking the fact that a function
which is strictly increasing or strictly decreasing must be injective. The big advantage of
having an injective function f is that you can cancel f from both sides of an equation.
A function is said to be surjective or onto if it takes on all possible values in the codomain.
That is, for every b in the codomain, there exists a in the domain such that f (a) = b. But
how do you prove that a function is surjective? The most direct way is to take an arbitrary
b in the codomain and use the functional equation somehow to find a value of a for which
10.4 Injective, surjective and bijective 177

f (a) = b. The big advantage of having a surjective function f is that you can then substitute
f (a) = b for any value of b in the codomain.
A function is said to be bijective if it is both injective and surjective. The big advantage of
having a bijective function f is that there exists an inverse function f −1 which satisfies

f −1 (f (x)) = x and f (f −1 (x)) = x.

These three properties are supremely important when solving functional equations. It’s a
standard tactic, when confronted with a difficult problem, to try to prove that the function
involved has some of these properties. All of the previous discussion might sound completely
cryptic to you and, if that’s the case, you’ll probably feel more enlightened after considering
the following example.

Problem The function f : R → R satisfies

f (xf (x) + f (y)) = f (x)2 + y

for all real numbers x and y.

Prove that f is bijective.

Solution Think about holding x constant. For example, let x = 0 and vary y. Then the
right-hand side can attain any real value. Since the left-hand side is f applied to some
expression, this proves that f is surjective.
You might be a little suspicious of this argument, but we can also express it in more algebraic
terms. To prove that f is surjective, for any real number b we must be able to find a real
number a such that f (a) = b. This value of a is simply f (b − f (0)2 ), since the functional
equation with x = 0 and y = b − f (0)2 implies that

f (f (b − f (0)2 )) = b.

If the extra algebra didn’t help your understanding, that’s fine—it was probably just compli-
cating what is actually quite a simple argument.
Now we turn our attention to showing that f is injective. To do this, we assume that
f (y1 ) = f (y2 ) and hope to deduce that y1 = y2 . But if f (y1 ) = f (y2 ), then we have

f (xf (x) + f (y1 )) = f (xf (x) + f (y2 ))

⇒ f (x)2 + y1 = f (x)2 + y2 .

Therefore, we can conclude that y1 = y2 and so f must be injective.

Since we have shown that f is both injective and surjective, we now know that f is bijective.

If you examine the previous proof closely, you’ll see that we relied on two particular features
of the functional equation. The first is that the left-hand side involves only f (y), but not y.
The second is that the right-hand side involves only y, but not f (y). Variables which appear
in this way often provide the key to proving the fact that a function is injective, surjective or
bijective. Unfortunately, this particular example doesn’t really demonstrate just how helpful
it is to know that f is bijective. But we’ll see various illustrative examples of this as we
progress.
178 10 Functional equations

10.5 The associative2 trick

One of the most common strategies when solving a functional equation is to look for one
expression which can be evaluated in two different ways. Doing so will often provide some
extra information that is crucial to solving the problem.
One such possibility involves nested functions. For example, f (g(h(x))) can be evaluated in
two different ways
f (g (h(x))) or f ( g(h (x)))
|{z} |{z}
depending on which composition of functions is evaluated first.

Problem Do there exist functions f : R → R and g : R → R such that

f (g(x)) = x2 and g(f (x)) = x3

for all real numbers x?

Solution The following chain of implications tells us that the function f must be injective.

f (a) = f (b) ⇒ g(f (a)) = g(f (b)) ⇒ a3 = b3 ⇒ a=b

Now we notice that the given information about f and g allows us to calculate the expression
f (g(f (x))) in two different ways.

f (g (f (x))) = f (x)2 and f ( g(f (x)) = f (x3 )

|{z} |{z}

Therefore,
f (x)2 = f (x3 )

for all values of x. In particular, we know that

f (−1) = f (−1)2 , f (0) = f (0)2 and f (1) = f (1)2 .

It follows that f (−1), f (0) and f (1) are all equal to 0 or 1 and so two of them are equal to
each other. This directly contradicts the fact that f is injective, so we conclude that there
cannot exist functions satisfying the conditions of the problem.

Note that the associative trick is particularly useful where the expression f (f (x)) is known.
For example, if
f (f (x)) = x + 1,

then applying the associative trick to f (f (f (x))) tells us that

f (x + 1) = f (x) + 1.

2 Anoperation ∗ is said to be associative if f ∗ (g ∗ h) = (f ∗ g) ∗ h always holds true. In our setting here, the
operation is function composition. Other examples of associative operations include the ordinary arithmetic
operations of addition and multiplication.
10.6 Exploit symmetry 179

10.6 Exploit symmetry

Another context for evaluating one expression in two different ways is when symmetry arises
as illustrated in the following example.

Problem Find all functions f : Z → Z such that

f (m + f (n) + mf (n)) = m + mn + f (n)

for all integers m and n.

Solution The crucial observation is the following. If we substitute m = f (p) for any integer p,
the left-hand side is a symmetric expression in p and n. In other words, it remains the same
when we swap p and n. However, the right-hand side is not symmetric in p and n.

f (f (p) + f (n) + f (p)f (n)) = f (p) + f (p)n + f (n)

Since swapping p and n in the above equation leaves the left-hand side unchanged, it must
also leave the right-hand side unchanged. Thus

f (p) + f (p)n + f (n) = f (n) + f (n)p + f (p).

Hence
f (p)n = f (n)p
for all integers p and n.
Substituting p = 1, we deduce that f (n) = f (1)n. Thus any solution to this functional
equation must be of the form f (n) = cn.
However, as with all functional equations, we must check our solutions. If we plug f (n) = cn
into the original functional equation, we deduce that c = 1. So, the only possible solution is
f (n) = n.

10.7 Involutions
An involution is a function which is its own inverse. That is, a function f which satisfies

f (f (x)) = x

for all x. Involutions can be surprisingly useful and crop up in the most mysterious ways.

Problem Find all functions g : R \ { 32 } → R such that

 
1 2x
x − g(x) = g
2 3x − 2

for all real numbers x 6= 32 .

Solution Though it might look scary and complicated, this functional equation is actually
simpler than most. For example, it only contains one variable x. All the equation does is
2x
relate two different values of g, that is, the value at x and the value at 3x−2 . This begs the
180 10 Functional equations

2x 2x
following question: if x is related to 3x−2 , then what is 3x−2 related to? With this in mind,
2x
let’s replace x with 3x−2 in the functional equation and hope for the best.

2x
!
2( 3x−2 )
   
2x 2x 1
−g = g 2x
3x − 2 3x − 2 2 3( 3x−2 )−2

Yes, it looks messy, but a little algebra should convince you that
2x
2( 3x−2 ) 4x
2x = = x.
3( 3x−2 )−2 6x − 2(3x − 2)

So we now have the equations

     
1 2x 2x 2x 1
x − g(x) = g and −g = g(x).
2 3x − 2 3x − 2 3x − 2 2

You can think of these as two simultaneous equations, which you can solve for both g(x) and
2x
g( 3x−2 ). Do this and you’ll find that

4x(x − 1)
g(x) = ,
3x − 2
which is indeed the solution to the functional equation.

2x
What’s going on here? Obviously there’s something special about the expression 3x−2 . In
2x
fact if you let f (x) = 3x−2 , then f (f (x)) = x, which means that it’s an involution!

10.8 Fixed points

A fixed point of a function f is a value of x for which f (x) = x. Despite being such a simple
concept, it’s often an extremely useful strategy to consider the fixed points of a function. In
fact, interesting deductions often arise from the existence of fixed points.

Problem Find all functions f : R+ → R+ such that

(i) f (xf (y)) = yf (x) for all positive real numbers x and y, and

(ii) f (x) < 1000 for all x > 1000.

Solution As a first step, let’s prove that the function must be bijective. Think about
holding x constant. For example, set x = 1 and vary y. Then the right-hand side can attain
any positive real value. Since the left-hand side is f applied to some expression, this proves
that f is surjective.
If we assume that f (y1 ) = f (y2 ), then it is certainly true that

f (xf (y1 )) = f (xf (y2 )) ⇒ y1 f (x) = y2 f (x).

Therefore, we can deduce that y1 = y2 and so f must be injective.

Since we’ve shown that f is both injective and surjective, we now know that f is bijective.
10.8 Fixed points 181

Let’s use these facts to determine the value of f (1). The big advantage of having a surjective
function lies in the fact that there exists a positive real number k for which f (k) = 1.
Substituting y = k in the functional equation gives

f (xf (k)) = kf (x) ⇒ f (x) = kf (x) ⇒ k = 1.

Therefore, we have f (1) = f (k) = 1. This is a fairly indirect way to find f (1), but the
technique is surprisingly common and useful. Finding just one value of the function, like
f (1), can often lead to a windfall of other benefits. For example, if we now set x = 1 in the
functional equation, we obtain the fact that

f (f (y)) = y.

In other words, f is an involution.

A useful substitution in this case is x = y, which implies that

f (xf (x)) = xf (x).

This is interesting because it means that the number xf (x) is a fixed point of f for any value
of x. Such a fact seems to imply that f must have a whole lot of fixed points. On the other
hand, you can’t have too many fixed points, for the second condition guarantees that every
fixed point is less than or equal to 1000. All this discussion really motivates us to concentrate
on the fixed points of f . We have the following observations.

Since f (1) = 1, the number 1 is a fixed point of f .

Suppose that a is a fixed point of f , so that f (a) = a. Now substitute x = a1 and y = a
into the functional equation to obtain
       
f (a) 1 1 1 1
f = af ⇒ f (1) = af ⇒ f = .
a a a a a
1
Hence, a is also a fixed point of f .
Suppose that a and b are fixed points of f , so that f (a) = a and f (b) = b. Now
substitute x = a and y = b into the functional equation to obtain

f (af (b)) = bf (a) ⇒ f (ab) = ab.

Hence, ab is also a fixed point of f . It follows that, if a is a fixed point of f , then so are
a2 , a3 , a4 , . . ..

Now let’s piece these three statements together in a clever way to show that 1 is the only
fixed point of f .
Suppose that a > 1 is a fixed point, then the fixed points

a2 , a3 , a4 , . . .

would eventually become greater than 1000, a contradiction.

1
Similarly, if a < 1 is a fixed point, then so is a > 1 and the fixed points

1 1 1
, , ,...
a2 a3 a4
would eventually become greater than 1000, another contradiction.
182 10 Functional equations

Thus, the only possible fixed point of f is 1. And yet we also know that xf (x) is a fixed point
for any value of x. The only way that this can happen is if xf (x) = 1 or, in other words, if
1
f (x) =
x
for any value of x.
As always, you should substitute this solution back into the functional equation to make sure
that it actually works.

Of course, there was a great deal of algebraic trickery involved in this solution. Rather than
remarking bewilderedly ‘I could never think of that!’, you should examine the solution again
and instead ask ‘What factors are present in this problem that would lead me to try that
technique?’ For all, well almost all, mathematicians are ordinary mortals. And most likely
any solution you read to a difficult problem, however short, has involved a great deal of effort.

10.9 Somewhere versus everywhere

We come now to a common fallacy that is the bane of many a beginner functional equation
solver! It concerns the relationship between individual points and the entire domain, between
local information and global information, between somewhere and everywhere.
A prime example to illustrate this point is the functional equation we introduced in section 10.4.
In that section, we demonstrated that the function is bijective. Now we capitalise on our
previous work and solve the functional equation once and for all.

Problem Find all functions f : R → R which satisfy

f (xf (x) + f (y)) = f (x)2 + y

for all real numbers x and y.

Solution Since we’ve already shown that f is bijective, we know that there exists a real
number k for which f (k) = 0. Substituting x = k into the functional equation immediately
implies that
f (f (y)) = y,
which means that f is an involution.
The trick now is to employ the ‘one expression, two ways’ philosophy. Note that the expression
xf (x) is invariant—that is, does not change—when you replace x with f (x). This follows
from our newly obtained result which states that f (f (x)) = x. So when we replace x with
f (x) in the given functional equation, we obtain

f (f (x)f (f (x)) + f (y)) = f (f (x))2 + y ⇒ f (xf (x) + f (y)) = x2 + y.

Comparing this with the original functional equation leads directly to

f (x)2 = x2 ⇒ f (x) = ±x.

And this is where we should tread carefully, for herein lies a trap for the unsuspecting
functional equation enthusiast. A beginner might now conclude that there are two solutions,
namely, f (x) = x and f (x) = −x. But this reasoning is entirely incorrect, because all we
have deduced is that for each real number x, the value of f (x) is either x or −x. There are
10.10 Completely multiplicative functions 183

still infinitely many possibilities, since the function could conceivably take the value x at
some places and −x at others!
Such crazy solutions to the functional equation seem improbable though, so let’s try to
eliminate them. To do this, suppose that there exist two distinct real numbers x and y such
that f (x) = x and f (y) = −y. In fact, we might as well assume that x and y are non-zero,
since f (0) = 0 in either case. Now substitute these values into the functional equation to
obtain
f (x2 − y) = x2 + y.
However, we know that f (x2 − y) = ±(x2 − y). If we take the positive sign, we obtain y = 0
and if we take the negative sign, we obtain x = 0. Both cases contradict the fact that x and
y are non-zero.
Therefore, we can finally conclude that the only solutions to the functional equation are
f (x) = x and f (x) = −x.

If you really want to make sure that you never fall into the ‘somewhere versus everywhere’
trap, then you should return to the problem from section 10.3 and solve it without the
condition that f (x) 6= 0 for x 6= 0.

10.10 Completely multiplicative functions

Thus far, our focus has mainly been on functional equations where R is the domain. Many of
the same techniques work when the domain is N+ or Q, but sometimes you need to pull out
a brand new trick. The one we’ll look at here is based on the prime factorisation of positive
integers. Recall the fundamental theorem of arithmetic, which asserts that every positive
integer n can be uniquely expressed as a product of primes

n = pa1 1 pa2 2 · · · pakk ,

where p1 , . . . , pk are primes and a1 , a2 , . . . , ak are positive integers. In fact, if we allow negative
powers of primes, then we may uniquely express any positive rational number as well.
These concepts will be particularly useful for dealing with completely multiplicative functions.
These are functions f which satisfy

f (mn) = f (m)f (n)

for every m and n in the domain.

Problem Find all completely multiplicative functions f : N+ → N+ .

The unsuspecting reader might assume that, much like other functional equations we’ve seen,
the only solution is given by f (n) = n. But he or she could not be further from the truth.

Solution A first step would be to substitute simple numbers such as m = n = 1 into the
functional equation f (mn) = f (m)f (n). This would give us the fact that

f (1) = 1.

Now observe that the functional equation allows us to write

f (kmn) = f (km)f (n) = f (k)f (m)f (n).

184 10 Functional equations

More generally, you can prove by induction that

f (m1 m2 · · · mk ) = f (m1 )f (m2 ) · · · f (mk ).

Thus, if we take the prime factorisation n = pa1 1 pa2 2 · · · pakk , then we have

f (n) = f (pa1 1 pa2 2 · · · pakk ) = f (p1 )a1 f (p2 )a2 · · · f (pk )ak .

So the function can be completely described by its values at the primes, namely,

f (2), f (3), f (5), f (7), . . ..

If you still believe that f (n) = n is the only solution, then you could try to go further and
show that f (p) = p for every prime p. Unfortunately, you won’t get anywhere! This is because
we’ve deduced as much as we can possibly deduce.
In fact, as long as we take f (1) = 1, we can set f (2), f (3), f (5), f (7), . . . to be any arbitrary
positive integer sequence and any such choice will give us a completely multiplicative function
f : N+ → N + .
You shouldn’t just believe this, so check it for yourself. In conclusion, this functional equation
has infinitely many solutions given by the previous description.

10.11 Well-ordering of N+
One property of N+ not shared by Q or R is the fact that every subset of N+ has a smallest
element. It seems so simple, yet this property can be incredibly useful, so much so that it has
the grand sounding name of well-ordering.

Problem The function f : N+ → N+ satisfies

f (n + 1) > f (f (n))

for each positive integer n.

Prove that f (n) = n for each positive integer n.

Solution The first thing you should do is try to construct such a function. You’ll probably
find that f (n) = n is an easy one to come up with, but other possibilities tend to fail because
you obtain various inequalities leading to other inequalities leading to contradictions. It’s
hard to see why these fail in general. Using the well-ordering of N+ allows us to consider
minimality, which will aid us considerably here.
The idea is to suppose that f (k) is the smallest member of the set S = {f (1), f (2), f (3), . . .}.
Of course, there might be more than one such k, but we only need one of them. The functional
equation tells us that
f (k) > f (f (k − 1)).
This seems to contradict the fact that f (k) is the smallest member of S. The only way to
avoid this contradiction is to have k = 1. In this way the inequality doesn’t arise because the
domain of f is the set of positive integers. This sleight of hand proves that f (1) is the unique
smallest member of S.3
3 Theterm ‘unique smallest’ is a little imprecise. After all, S only has one smallest member anyway. What we
mean is that if m is the smallest member of S, then the only number that satisfies f (n) = m, is n = 1. The
term has an analogous meaning for ‘unique second smallest’, and so on, later in the proof.
10.11 Well-ordering of N+ 185

Next, what’s the second smallest member of S? It must be f (k) for some k ≥ 2. Again, there
might be more than one such k, but we only need one of them. The functional equation tells
us that
f (k) > f (f (k − 1)).
If f (k − 1) > 1, we would have f (f (k − 1)) > f (1). This would contradict the fact that f (k)
is the second smallest member of S. So we must have f (k − 1) = 1. This implies that the
smallest member of S is 1 and that k = 2. Thus another sleight of hand proves that f (2) is
the unique second smallest member of S.
What’s the third smallest member of S? It is f (k) for some k ≥ 3. The functional equation
tells us that
f (k) > f (f (k − 1)).
If f (k − 1) > 2, we would have f (f (k − 1)) > f (2) > f (1). This would contradict the fact
that f (k) is the third smallest member of S. We certainly can’t have f (k − 1) = 1 because
this occurs only for k = 2. So we must have f (k − 1) = 2. This implies that the second
smallest member of S is 2 and that k = 3.
This argument can be continued inductively4 to show that the members of S written in
increasing order are
f (1) < f (2) < f (3) < · · · .
It also shows that the members of S written in increasing order are

1 < 2 < 3 < ···.

From this we deduce that f (n) = n is the only solution.

4 Thisproof cannot be considered complete as it currently stands. Can you complete the proof by writing out
the details of the induction yourself?
 11
Inequalities

An inequality is a mathematical problem which asks you to prove that some expression is
always greater than (or equal to) some other expression. For example, you might have to
show that for any real numbers a, b, c, the following holds.

a2 + b2 + c2 ≥ ab + bc + ca

Although daunting at first, you should be able to solve a whole variety of inequalities after
mastering a handful of techniques and tricks. That is exactly what this chapter will help you
to achieve.

11.0 Problems
1. Farmer Brown wants a rectangular paddock of area A next to a long straight river, with
a fence on the other three sides.
What is the minimum length of fencing that is required?

2. (a) Use the AM–GM inequality to find the maximum value of xyz, where x, y, z are
positive real numbers satisfying x + 2y + 3z = 3.
(b) Use the AM–GM inequality to find the minimum value of x + y + z, where x, y, z
are positive real numbers satisfying xy 2 z 3 = 108.

3. Prove that if a, b, c are positive real numbers satisfying abc = 1, then

(a + b)(b + c)(c + a) ≥ 8.

4. Let x and y be positive numbers such that x + y = 1. Show that

  
1 1 9
1+ 1+ ≥ .
x y 4

5. If a, b, c are positive, prove in at least three different ways that

a b c 3
+ + ≥ .
b+c c+a a+b 2
188 11 Inequalities

6. Use the rearrangement inequality to prove the following, where a, b, c are positive real
numbers.
(a) a2 b + b2 c + c2 a ≤ a3 + b3 + c3
1 1 1 a b c
(b) + + ≤ 2+ 2+ 2
a b c b c a
(c) abc bca cab ≤ aab bbc cca
7. Use the Cauchy–Schwarz inequality to find the minimum value of

x2 + y 2 + z 2 ,

where x, y, z are real numbers satisfying 3x + 4y + 5z = 10.

8. For positive real numbers a, b, c, prove that
r
a+b+c ab + bc + ca
≥ .
3 3

9. Let a, b, c be positive numbers satisfying

1 1 1
+ + = 1.
a b c
Prove that
(a − 1)(b − 1)(c − 1) ≥ 8.

10. For positive real numbers a, b, c, d, prove that

1 1 4 16 64
+ + + ≥ .
a b c d a+b+c+d

11. Let A, B, C be the angles of a triangle. If the triangle is acute, use Jensen’s inequality
to prove the following inequalities. Also determine which of these inequalities remain
true if the triangle is obtuse.
√
3 3
(a) sin A + sin B + sin C ≤
2
3
(b) cos A + cos B + cos C ≤
2
1
(c) cos A cos B cos C ≤
8
12. If a1 , a2 , . . . , an are distinct positive integers, prove that
a1 a2 an 1 1 1
+ 2 + ··· + 2 ≥ + + ··· + .
12 2 n 1 2 n

13. If a1 , a2 , . . . , an are positive numbers whose product is 1, prove that

(2 + a1 )(2 + a2 ) · · · (2 + an ) ≥ 3n .

14. Let a, b, c be positive real numbers such that abc = 1. Prove that
bc ca ab
+ + ≤1
b5 + c5 + bc c5 + a5 + ca a5 + b5 + ab
and determine when equality holds.
11.0 Problems 189

15. If a, b, c are positive real numbers, prove that

9 2 2 2 1 1 1
≤ + + ≤ + + .
a+b+c a+b b+c c+a a b c

16. Prove that if a, b, c are positive real numbers with sum less than 1, then

abc(1 − a − b − c) 1
≤ .
(a + b + c)(1 − a)(1 − b)(1 − c) 81

17. Suppose that the numbers x, y, z are greater than or equal to 1 and satisfy
1 1 1
+ + = 2.
x y z
Prove that √ √ p √
x+y+z ≥ x−1+ y−1+ z − 1.

18. For positive real numbers x, y, z, show that

x y z 9
+ + ≤ .
(x + y)(x + z) (y + z)(y + x) (z + x)(z + y) 4(x + y + z)

19. Let x, y, z be positive real numbers which satisfy xyz = 1. Show that

x3 y3 z3 3
+ + ≥ .
(1 + y)(1 + z) (1 + z)(1 + x) (1 + x)(1 + y) 4

20. Let a1 , a2 , . . . , an be positive real numbers satisfying

1 1 1 1
+ + ··· + = .
a1 + 100 a2 + 100 an + 100 100

Prove that √
n
a1 a2 · · · an ≥ 100(n − 1).

21. Let n ≥ 3 be an integer, and let a2 , a3 , . . . , an be positive real numbers such that
a2 a3 · · · an = 1. Prove that

(1 + a2 )2 (1 + a3 )3 · · · (1 + an )n > nn .
190 11 Inequalities

11.1 Squares are non-negative

A simple though useful inequality is the fact that, when you multiply a real number by itself,
the answer is always zero or positive. In short, squares are non-negative.

Problem Prove that

x4 > 8x − 9

for every real number x.

Solution The given inequality is equivalent to

(x2 − 1)2 + 2(x − 2)2 > 0.

You should check this right now with pen and paper to make sure that we’re not lying to you!
Once you’ve done that, the inequality should be almost self-evident since the left-hand side is
a sum of squares, which we know to be non-negative. Furthermore, the first term is zero only
when x = ±1, while the second term is zero only when x = 2. So it’s impossible for them to
be 0 simultaneously, and this guarantees that the left-hand side is not only non-negative, but
always positive.

Of course, the strategy we used here involved taking all terms to the left-hand side and then
magically discovering that it could be expressed as a sum of squares. Although this may not
have been apparent to you from the outset, spotting squares becomes much easier after one
has had some square-spotting practice.

Problem Farmer Black wants a rectangular paddock of area A with a fence on each side.
What is the minimum length of fencing that is required?

Solution Let the side lengths of Farmer Black’s rectangular paddock be x and y. We are
given that xy = A and would like to find the minimum possible value of 2x + 2y. Your
intuition probably tells you that this will occur when x and y are equal to each
√ other, in
which case the rectangular paddock would actually be a square with perimeter 4 A. So let’s
try to prove that
√
2x + 2y ≥ 4 A,

or equivalently,
x+y √
≥ xy.
2
After squaring both sides1 , collecting all terms on the left-hand side, and factorising, the
inequality becomes

x2 + 2xy + y 2 ≥ 4xy ⇔ x2 − 2xy + y 2 ≥ 0 ⇔ (x − y)2 ≥ 0.

Of course, this is true because squares are non-negative. Therefore, we have

√ succeeded in
proving that the minimum length of fencing that Farmer Black requires is 4 A.
1 Generally LHS ≥ RHS is not equivalent to LHS2 ≥ RHS2 . But if it is known that LHS ≥ 0 and RHS ≥ 0,
then it is equivalent.
11.2 AM–GM inequality 191

11.2 AM–GM inequality

√
In the previous section, we used the inequality x+y
2 ≥ xy, which can be expressed as follows:
the average of two non-negative numbers is at least as large as the square root of their product.
More generally, it’s true that the average of n non-negative numbers is at least as large as
the nth root of their product.

AM–GM inequality If a1 , a2 , . . . , an are non-negative real numbers, then

a1 + a2 + · · · + an √
≥ n a1 a2 · · · an .
n
Furthermore, equality occurs if and only if a1 = a2 = · · · = an .

The inequality is so named because the left-hand side is known as the arithmetic mean—a
fancy name for what we usually call the average—while the right-hand side is known as the
geometric2 mean.

Problem If a, b, c are positive and satisfy a + b + c = 1, prove that

1 1 1
+ + ≥ 27.
ab bc ca

Solution Since we have the condition a + b + c = 1, the expression we wish to minimise is

1 1 1 a+b+c 1
+ + = = .
ab bc ca abc abc
In other words, we would like to maximise the product abc.
The AM–GM inequality is often the key when dealing with inequalities which involve sums
and products. Applying it to a, b, c, we obtain

a+b+c √
3 1 √
3 1
≥ abc ⇒ ≥ abc ⇒ abc ≤ .
3 3 27
It follows that
1 1 1 1
+ + = ≥ 27,
ab bc ca abc
as desired.

11.3 Rearrangement inequality

Suppose that you are on a television game show and that there are three piles of money in
front of you: one consisting of $5 notes, one consisting of $20 notes and one consisting of $100
notes. For your prize, you are allowed to take 100 notes from one pile, 200 notes from another
pile and 300 notes from the remaining pile. What should your strategy be? Well, the greedy
capitalist inside you would probably take 300 $100 notes, 200 $20 notes and 100 $5 notes, in
order to maximise your winnings. And of course, in order to minimise your winnings, you
would do the exact opposite, that is, take 300 $5 notes, 200 $20 notes and 100 $100 notes.
This greedy principle is the basis for the following.
2 Thearithmetic mean can be thought of as the average with respect to the addition operation, while the
geometric mean can be thought of as the average with respect to the multiplication operation.
192 11 Inequalities

Rearrangement inequality Let (a1 , a2 , . . . , an ) and (x1 , x2 , . . . , xn ) be two sequences

of real numbers. Suppose we seek a rearrangement (i.e. permutation) (b1 , b2 , . . . , bn ) of
(x1 , x2 , . . . , xn ) which maximises the expression

E = a1 b1 + a2 b2 + · · · + an bn .

Then E is maximised when the sequences (a1 , a2 , . . . , an ) and (b1 , b2 , . . . , bn ) are sorted the
same way.
If, on the other hand, we are trying to minimise E, then this occurs when the sequences
(a1 , a2 , . . . , an ) and (b1 , b2 , . . . , bn ) are sorted the opposite way.

Two sequences of numbers (a1 , a2 , . . . , an ) and (b1 , b2 , . . . , bn ) are said to be sorted the same
way if the largest ai is in the same position as the largest bi , the second largest ai is in the
same position as the second largest bi , and so on. On the other hand, they are said to be
sorted the opposite way if the largest ai is in the same position as the smallest bi , the second
largest ai is in the same position as the second smallest bi , and so on.
For future convenience, we’ll refer to the expression

a1 b1 + a2 b2 + · · · + an bn

as the product of the sequences (a1 , a2 , . . . , an ) and (b1 , b2 , . . . , bn ).

Problem If x1 , x2 , . . . , xn are positive real numbers, show that

x21 x2 x2 x2
+ 2 + · · · + n−1 + n ≥ x1 + x2 + · · · + xn .
x2 x3 xn x1

Solution The left-hand side of the inequality looks suspiciously like the product of two
sequences,  
2 2 2 1 1 1
(x1 , x2 , . . . , xn ) and , ,..., .
x2 x3 x1

You should be able to see that the right-hand side can also be written as a product of the
same two sequences rearranged, that is,
 
1 1 1
(x21 , x22 , . . . , x2n ) and , ,..., .
x1 x2 xn

So, by the rearrangement inequality, the desired result will follow once we prove that these
latter two sequences are sorted the opposite way. But this is certainly true since
1 1
x2i ≤ x2j if and only if ≥ .
xi xj

In other words, if x2i is the largest term of the first sequence, then x1i is the smallest term
of the second sequence; if x2i is the second largest term of the first sequence, then x1i is the
second smallest term of the second sequence; and so on.

The rearrangement inequality is a surprisingly useful result, so you should take the time to
really understand the previous argument before moving on.

Problem For positive real numbers a, b, c, prove that

aa bb cc ≥ ab bc ca .
11.4 Cauchy–Schwarz inequality 193

Solution This inequality involves products of powers, whereas the rearrangement inequality
involves sums of products. This suggests that we should take the logarithm of both sides.
And we can do this without any change of the inequality sign, since log x is an increasing
function on the set of positive real numbers. Thus, we obtain the equivalent inequality

a log a + b log b + c log c ≥ b log a + c log b + a log c.

Note that the left-hand side is a product of the two sequences

(a, b, c) and (log a, log b, log c).

These are sorted the same way, since the log function is increasing, as we already mentioned.
So by the rearrangement inequality, the left-hand side is certainly greater than or equal to
the product of the two sequences

(a, b, c) and (log c, log a, log b),

which just so happens to be the right-hand side of the inequality.

As a final comment on this section we note that the rearrangement inequality generalises
to more than two sequences of numbers. Here we state the version for three sequences of
numbers.3
Rearrangement inequality for three sequences of numbers Let (a1 , a2 , . . . , an ),
(x1 , x2 , . . . , xn ) and (y1 , y2 , . . . , yn ) be three sequences of positive real numbers. Suppose we
seek rearrangements (b1 , b2 , . . . , bn ) of (x1 , x2 , . . . , xn ) and (c1 , c2 , . . . , cn ) of (y1 , y2 , . . . , yn )
which maximise the expression

E = a1 b1 c1 + a2 b2 c2 + · · · + an bn cn .

Then E is maximised if the three sequences are sorted the same way.

Note that in contrast to the two sequence version of the rearrangement inequality, there is no
simple criterion for minimising E.

11.4 Cauchy–Schwarz inequality

Cauchy–Schwarz inequality If (x1 , x2 , . . . , xn ) and (y1 , y2 , . . . , yn ) are two sequences of

real numbers, then

(x21 + x22 + · · · + x2n )(y12 + y22 + · · · + yn2 ) ≥ (x1 y1 + x2 y2 + · · · + xn yn )2 .

Furthermore, equality occurs if and only if we have the equal ratios

x1 : y1 = x2 : y2 = · · · = xn : yn .

The Cauchy–Schwarz inequality is often neglected by the budding inequality problem solver,
mainly because it is difficult to know when and how to use it. However, it can be a mighty
weapon in the hands of an expert. Here are some general tips.

If there are squares or square roots, try Cauchy–Schwarz.

3 Notethat the numbers in the three sequence version must all be positive. This is in contrast to the two
sequence version where they only need to be real.
194 11 Inequalities

If there are products or fractions, try Cauchy–Schwarz.

If all else fails, try Cauchy–Schwarz.

Problem Prove the AM–HM inequality4 , which states that if a1 , a2 , . . . , an are positive real
numbers, then
a1 + a2 + · · · + an n
≥ 1 1 .
n a1 + a2 + · · · + a1n
Furthermore, show that equality occurs if and only if a1 = a2 = · · · = an .

Solution Since a1 , a2 , . . . , an are positive real numbers, we can simply take the two sequences
 
√ √ √ 1 1 1
( a1 , a2 , . . . , an ) and √ , √ ,..., √
a1 a2 an

and substitute them into the Cauchy–Schwarz inequality.

We obtain  
1 1 1
(a1 + a2 + · · · + an ) + + ··· + ≥ n2 ,
a1 a2 an
which gives the desired result after some rearranging.
Equality occurs if and only if

√ 1 √ 1 √ 1
a1 : √ = a2 : √ = · · · = an : √ ,
a1 a2 an

which is equivalent to the fact that a1 = a2 = · · · = an .

Problem Let a, b, c be positive real numbers such that abc = 1. Prove that

1 1 1 3
+ + ≥ .
a3 (b + c) b3 (c + a) c3 (a + b) 2

Solution A common starting strategy is to massage the given inequality into a nicer looking
form by using substitutions or other algebraic trickery. In this case, it would be nice to get
rid of the unsightly cubes appearing in the denominator of eachterm. We  can improve the
situation greatly if we opt for the clever substitution (a, b, c) = x1 , y1 , z1 . This will help to
bring some of the terms into the numerator—almost always a good thing—and will also give
us the equally nice constraint xyz = 1. As an example, the first term on the left-hand side
would become
1 1 x3 x3 yz x2
3
= 1 1 1 = 1 1 = = .
a (b + c) x3 ( y + z ) y + z
y+z y+z

Similar expressions hold for the other two terms.

So we now have the task of proving the inequality

x2 y2 z2 3
+ + ≥ ,
y+z z+x x+y 2

for positive real numbers satisfying xyz = 1.

4 The inequality is so named because the right-hand side is known as the harmonic mean.
11.5 Power means inequality 195

This inequality not only looks much friendlier, but also seems to be a prime candidate for the
Cauchy–Schwarz inequality. To obtain the left-hand side, it makes sense to let one of our
sequences be  
x y z
√ ,√ ,√ .
y+z z+x x+y
When using Cauchy–Schwarz, you should always look for nice cancellation, and this is most
easily achieved if we let the other sequence be
√ √ √

y + z, z + x, x + y .

Using these two sequences, the Cauchy–Schwarz inequality tells us that

 2
y2 z2

x
+ + (2x + 2y + 2z) ≥ (x + y + z)2 ,
y+z z+x x+y

which in turn implies that

x2 y2 z2 x+y+z
+ + ≥ .
y+z z+x x+y 2

All that remains is to prove that x + y + z ≥ 3. But this is an immediate consequence of

using the AM–GM inequality with the numbers x, y, z, along with the fact that xyz = 1.

11.5 Power means inequality

Power means inequality For positive real numbers a1 , a2 , . . . , an , define

r1
√ ar1 + ar2 + · · · + arn

M0 = n
a1 a2 · · · an and Mr =
n

for every non-zero real number r.

If r > s, then Mr ≥ Ms , and equality occurs if and only if a1 = a2 = · · · = an .

We often call M2 the quadratic mean (QM), M1 the arithmetic mean (AM), M0 the geometric
mean5 (GM), and M−1 the harmonic mean (HM). You can see just how powerful this result
is from the fact that the AM–GM inequality and the AM–HM inequality are simply special
cases.

Problem Suppose that x, y, z are positive real numbers which satisfy xyz = 1.
If r > s > 0, prove that
xr + y r + z r ≥ xs + y s + z s .

Solution The inequality seems reminiscent of the power means inequality with n = 3, which
can be rearranged to give the following.
 rs −1
xs + y s + z s

r r r s s s
x + y + z ≥ (x + y + z )
3
5 Actually,M0 looks like the oddball here. Even though we can’t directly put r = 0 into the definition of Mr ,
it turns out that Mr converges to the geometric mean as r → 0.
196 11 Inequalities

So all that remains to be proved is that

 s  r −1
x + ys + zs s
≥ 1.
3
However, this follows from the fact that rs − 1 > 0, which is true by assumption, and the fact
that
xs + y s + z s √
≥ 3 xs y s z s = 1,
3
which is true by the AM–GM inequality.

In fact, this is a particular case of the following more general result, which you should now
be able to prove on your own. It’s definitely a handy little inequality to have under your belt
and will reappear before the end of the chapter.

A useful inequality Suppose that x1 , x2 , . . . , xn are positive real numbers which satisfy
x1 x2 · · · xn = 1. If r > s > 0, then

xr1 + xr2 + · · · + xrn ≥ xs1 + xs2 + · · · + xsn .

11.6 Jensen’s inequality

Any time you have an expression in your inequality which can be written as

f (x1 ) + f (x2 ) + · · · + f (xn ),

it’s worth checking whether f is convex (like a smile) or concave (like a frown).6 More often
than not, Jensen’s inequality can be used in such cases.

Jensen’s inequality If the real numbers x1 , x2 , . . . , xn lie on an interval where the func-
tion f is convex, then
 
f (x1 ) + f (x2 ) + · · · + f (xn ) x1 + x2 + · · · + xn
≥f .
n n

If they lie on an interval where the function f is concave, then the inequality is reversed.

Problem Let x1 , x2 , . . . , xn be positive real numbers whose sum is 1. What is the minimum
value of the following expression?
x1 x2 xn
+ + ··· +
1 + x2 + · · · + xn 1 + x1 + x3 + · · · + xn 1 + x1 + · · · + xn−1

Solution Note that this expression can be written more simply as

x1 x2 xn
+ + ··· + .
2 − x1 2 − x2 2 − xn
The form f (x1 ) + f (x2 ) + · · · + f (xn ) now appears, which suggests using Jensen’s inequality
with
x
f (x) = .
2−x
6 Itis important to know that functions like f (x) = x2 and f (x) = ex are convex while f (x) = log x is concave
on the set of positive real numbers. For those who have learnt calculus, you can use the fact that a function f
is convex when f 00 (x) ≥ 0 and concave when f 00 (x) ≤ 0.
11.7 Substitutions 197

Indeed, the function f is convex on the interval [0, 1] and since the numbers x1 , x2 , . . . , xn all
lie on this interval, we have
x1 x2 xn
+ + ··· + = f (x1 ) + f (x2 ) + · · · + f (xn )
2 − x1 2 − x2 2 − xn
 
x1 + x2 + · · · + xn
≥ nf
n
 
1
= nf
n
n
= .
2n − 1
n
This shows that the expression is always greater than or equal to 2n−1 . To finish we need to
demonstrate that this value can actually be attained. Simply taking x1 = x2 = · · · = xn = n1
does the job.

Problem If the numbers a, b, c are greater than 1, prove that

3 1 1 1
√
3
≤ + + .
1+ abc 1+a 1+b 1+c
1
Solution You might think of trying Jensen’s inequality with a function like f (x) = 1+x .
This helps with the RHS. By changing variables and modifying our function accordingly, we
can deal with the LHS too.
Consider the function
1
f (x) =,
1 + ex
which is convex for x positive. Then Jensen’s inequality allows us to deduce that
 
1 1 1 1 1
x+y+z ≤ + + .
1+e 3 3 1 + ex 1 + ey 1 + ez

If we now substitute a = ex , b = ey , c = ez , which is allowed since a, b, c are greater than 1,

we obtain the desired inequality.

11.7 Substitutions
Substitutions can often help to simplify an inequality. In particular, you should always be on
the lookout for useful substitutions, such as the following.

Given abc = 1, consider the substitution x = an , y = bn , z = cn for some positive

 or

x y z
negative real number n. Other substitutions worth considering are (a, b, c) = y , z , x ,
 
which may break symmetry, or (a, b, c) = xyz2 , xz xy
y2 , z2 which retains symmetry.
 
x y z
If a + b + c = 1, you could try (a, b, c) = x+y+z , x+y+z , x+y+z .

Consider a substitution which makes an ugly denominator look nicer, even if it makes
the numerator look uglier.
Consider a substitution which makes an ugly constraint look nicer, even if it makes the
inequality look uglier.
198 11 Inequalities

If a, b, c are the side lengths of a triangle, it’s almost always useful to try the substitution
(a, b, c) = (y + z, z + x, x + y). In fact, a, b, c are the side lengths of a triangle if and
only if x = b+c−a2 , y = c+a−b
2 , z = a+b−c
2 are all positive.
We call this the incircle substitution because the incircle of a triangle with side lengths
a, b, c divides the sides into segments of lengths x, y, z. (See section 12.9.)

Problem If a, b, c are positive, show that

a b c
+ + ≥ 1.
b + 2c c + 2a a + 2b

Solution Whenever fractions are concerned, it’s almost always better for ugliness to occur
in the numerator rather than the denominator. In this particular inequality, we can introduce
the following substitution to move the ugliness from the bottom of each fraction to the top.
x = b + 2c 9a = 4y + z − 2x
y = c + 2a ⇔ 9b = 4z + x − 2y
z = a + 2b 9c = 4x + y − 2z

After performing these substitutions and tidying up (something which you should try on your
own with pen and paper), the inequality looks a lot more palatable.
   
y z x x y z
4 + + + + + ≥ 15
x y z y z x
And it’s not too difficult to see that this inequality is true, since each bracket is greater than
or equal to 3 by the AM–GM inequality.

11.8 Addition and multiplication of inequalities

Here is a neat way to create an interesting and difficult inequality from a boring and simple
one. As an example, we’ll start with the well-known inequality
√
x + y ≥ 2 xy.
Now, pick a few variables such as a, b, c and write down all possible combinations of the
inequality using these.
√ √ √
a + b ≥ 2 ab b + c ≥ 2 bc c + a ≥ 2 ca
Adding or multiplying these together produces the following two bigger and better inequalities.
√ √ √
a + b + c ≥ ab + bc + ca
(a + b)(b + c)(c + a) ≥ 8abc

Many interesting inequalities are created using this technique. One way to solve them is to
reverse engineer them. In other words, your task is to discover the original chunks which were
added or multiplied together to create the inequality. As you will see, this does not involve
guesswork alone, but also intelligence, experience, and sometimes luck.

Problem Let a, b, c be the lengths of the sides of a triangle.

Prove that √ √ √ √ √ √
a+b−c+ b+c−a+ c+a−b≤ a+ b+ c.
11.8 Addition and multiplication of inequalities 199

Solution We begin by using the incircle substitution

a = y + z, b = z + x, c = x + y,

which was mentioned in section 11.7. This is equivalent to

b+c−a c+a−b a+b−c

x= , y= , z=
2 2 2

and transforms the inequality into the form

√ p √ √ √ √
2x + 2y + 2z ≤ y + z + z + x + x + y.

A very naive and incorrect approach is to guess

√ that√the first term on the left is less than or
equal to the first term on the right, so that 2x ≤ y + z. However, this simply cannot be
true for all positive x, y, z. So we have to dig a little deeper to find the chunks which are
hiding. To uncover them, let’s write the left-hand side as
√ √ √ √ √ √
2y + 2z 2z + 2x 2x + 2y
+ + .
2 2 2

And now it seems far more plausible that each term on the left is smaller than the corresponding
term on the right. All that remains is to show that
√ p √
2x + 2y ≤ 2 x + y,

which is easily proved by squaring, simplifying, and resorting to the AM–GM inequality. Then
adding this inequality with the two others obtained by permuting the variables yields the
desired result.

Problem Prove that, for all positive real numbers a, b, c,

abc abc abc

+ + ≤ 1.
a3 + b3 + abc b3 + c3 + abc c3 + a3 + abc

Solution This time, the chunks are much harder to uncover, but they look like the following.

abc c
≤ ⇔ a2 bc + b2 ca + c2 ab ≤ ca3 + cb3 + c2 ab.
a3 + b3 + abc a+b+c

After collecting all terms on the left-hand side and factorising, this latter inequality simply
becomes
−c(a + b)(a − b)2 ≤ 0,

which is obviously true. Now the solution to the original problem follows immediately, since
we can simply sum the three corresponding inequalities.

abc abc abc a+b+c

+ + ≤ =1
a3 + b3 + abc b3 + c3 + abc c3 + a3 + abc a+b+c

The trick in this solution was to break the number 1 into three chunks, each one depending
on a, b and c.
200 11 Inequalities

11.9 Expand and conquer

Most of the inequalities that we have seen are symmetric, which means that they look exactly
the same after permuting the variables. Note that if you are expanding a symmetric expression
in three variables and you see the term a2 b, then you would certainly expect to see also the
terms a2 c, b2 a, b2 c, c2 a, c2 b.
Inequalities can also be cyclic, which means that they look the same after cycling the variables.
Note that if you are expanding a cyclic expression in three variables and you see the term
a2 b, then you would certainly expect to see also the terms b2 c and c2 a, but not necessarily
b2 a, c2 b and a2 c.
We can take advantage of these facts when expanding algebraic expressions by using symmetric
sum notation and cyclic sum notation. The following examples should give you some idea of
how it works for three variables, although you can use this notation for more variables.
X
a2 b = a2 b1 c0 + a2 c1 b0 + b2 c1 a0 + b2 a1 c0 + c2 a1 b0 + c2 b1 a0
sym = a2 b + a2 c + b2 c + b2 a + c2 a + c2 b
X
a2 b = a2 b1 c0 + b2 c1 a0 + c2 a1 b0
cyc = a2 b + b2 c + c2 a
X
a2 = a2 b0 c0 + a2 c0 b0 + b2 c0 a0 + b2 a0 c0 + c2 a0 b0 + c2 b0 a0
sym = 2a2 + 2b2 + 2c2
X
a2 = a2 b0 c0 + b2 c0 a0 + c2 a0 b0
cyc = a2 + b2 + c2

Problem For positive real numbers a, b, c, prove that

a + b − 2c b + c − 2a c + a − 2b
+ + ≥ 0.
b+c c+a a+b

Solution It might seem like a maniacal thing to do, but we’re going to multiply both sides
of the inequality by (a + b)(b + c)(c + a). Since the inequality is cyclic in nature, it would be
foolish for us not to use cyclic sum notation. We want to prove
X
(a + b − 2c)(c + a)(a + b) ≥ 0,
cyc

which is equivalent to
X
a3 + 2a2 b + b2 c + b2 a − 2c2 a − 2c2 b − a2 c ≥ 0.
cyc

This might seem even messier than when we began, but remember that in cyclic sum notation,
we have equations like
X X X
a2 b = b2 c = c2 a = a2 b + b2 c + c2 a.
cyc cyc cyc

So, this observation allows us to cancel a lot of the terms until we are left with something
which almost looks nice. X
a3 + a2 b − 2a2 c ≥ 0
cyc
11.10 Homogeneous inequalities 201

It would be fantastic if we could prove that the sum a3 + a2 b − 2a2 c was non-negative, but
this fact just isn’t true for arbitrary positive real numbers a, b, c.
However, remember that in cyclic sum notation, the term a2 b is the same as the term c2 a.
Plugging this into the inequality and factorising gives
X X
a3 + c2 a − 2a2 c = a(a − c)2 ≥ 0,
cyc cyc
which is now obviously true, using the fact that squares are non-negative.

11.10 Homogeneous inequalities

Many inequalities that we’ve seen are homogeneous, meaning that all terms have the same
degree, in some sense. For example, the expression x3 +x2 y−y 3 is homogeneous of degree three,
while the expression x3 + y is not. More precisely, we say that an inequality is homogeneous
of degree n if multiplying all of the variables by any positive constant λ has the effect of
multiplying the entire inequality by λn . Note this effectively leaves the inequality to be
proven unchanged because dividing by λn returns the original inequality. With homogeneous
inequalities, we are free to set the sum or product of the variables to be a particular number
and just prove the inequality with that particular constraint. Hopefully, you will soon see
what we mean.

Problem If a, b, c are positive, prove that

   
 a b  c a+b+c
1+ 1+ 1+ ≥2 1+ √
3
.
b c a abc

Solution First, we note that the inequality is homogeneous, since plugging in the values ra,
rb and rc gives
   
 ra  rb  rc  ra + rb + rc
1+ 1+ 1+ ≥2 1+ √ 3
,
rb rc ra rarbrc
which is identical to the original inequality after some cancellation. This means that if we
can prove the inequality for (a, b, c), then it must hold true for (ra, rb, rc) for any positive
real number r. In particular, if we can prove the inequality in the case a + b + c = 1, then
it follows that the inequality is true for any positive value of a + b + c. Similarly, if we can
prove the inequality for abc = 1, then it follows that the inequality is true for any positive
value of abc.
In this particular case, fixing abc = 1 gets rid of the unsightly cube root for us and the
inequality takes the following form
 
 a b  c
1+ 1+ 1+ ≥ 2(1 + a + b + c).
b c a
Expanding and simplifying leaves us with
a2 c + b2 c + b2 a + c2 a + c2 b + a2 b ≥ 2(a + b + c).

The rest of the inequality can be handled by cleverly pairing the six terms on the left-hand
side as shown and applying the AM–GM inequality to each pair.
√ √ √
(a2 b + a2 c) + (b2 c + b2 a) + (c2 a + c2 b) ≥ 2 a4 bc + 2 b4 ca + 2 c4 ab
3 3 3
= 2(a 2 + b 2 + c 2 )
≥ 2(a + b + c)
202 11 Inequalities

The very last step here is a simple application of the useful inequality stated in section 11.5.

11.11 Muirhead’s inequality

Muirhead’s inequality, also known as majorisation, is a hefty weapon. Many symmetric
inequalities can be smashed with this technique along with brute force calculations.

Muirhead’s inequality Let A = (a1 , a2 , . . . , an ) and B = (b1 , b2 , . . . , bn ) be two sequences

of real numbers which satisfy the following inequalities.

a1 ≥ a2 ≥ · · · ≥ an
b1 ≥ b2 ≥ · · · ≥ bn

a 1 ≥ b1
a 1 + a 2 ≥ b1 + b2
a1 + a2 + a3 ≥ b1 + b2 + b3
..
.
a1 + a2 + · · · + an−1 ≥ b1 + b2 + · · · + bn−1
a1 + a2 + · · · + an−1 + an = b1 + b2 + · · · + bn−1 + bn

(Note that the last line really is an equality.)

The sequence A is said to majorise the sequence B and this is expressed by writing A  B.
If A  B and x1 , x2 , . . . , xn are non-negative real numbers, then
X X
xa1 1 xa2 2 · · · xann ≥ xb11 xb22 · · · xbnn .
sym sym

Problem Prove that if a, b, c are positive real numbers with product 1, then
a4 b + ab4 + a4 c + ac4 + b4 c + bc4 ≥ 2ab + 2bc + 2ac.

Solution Note first that since abc = 1 we may replace ab with a2 b2 c, bc with ab2 c2 and ac
with a2 bc2 thus making the inequality homogeneous. Then if we write it using symmetric
sum notation, the inequality takes the form
X X
a4 b1 c0 ≥ a2 b2 c.
sym sym

Since the triple (4, 1, 0) majorises the triple (2, 2, 1), we may invoke Muirhead’s inequality to
finish off the problem.

Problem Let x, y and z be positive real numbers such that xyz = 1.

Prove that
x5 − x2 y5 − y2 z5 − z2
+ + ≥ 0.
x5 + y 2 + z 2 y 5 + z 2 + x2 z 5 + x2 + y 2

Solution We suppress the brute force part here! Suffice to say that after homogenising,
clearing denominators and rearranging, the inequality takes the form
X X
x10 yz + 4x7 y 5 + x6 y 3 z 3 ≥ x8 y 2 z 2 + 2x6 y 5 z + 2x6 y 4 z 2 + x5 y 5 z 2 .
sym sym
11.12 Weighted inequalities 203

It would be nice if we could match up sequences on the LHS to majorise sequences on the
RHS, but unfortunately this cannot be done. The sequence (10, 1, 1) on the LHS majorises
all of the sequences associated with the RHS. The sequence (7, 5, 0) on the LHS majorises all
the sequences on the RHS except for (8, 2, 2). But the sequence (6, 3, 3) on the LHS does not
majorise anything on the RHS. The solution to our conundrum is to use the AM–GM on the
LHS so as to smooth out the strong (10, 1, 1) term with the weak (6, 3, 3) term. This yields

x10 yz + x6 y 3 z 3 ≥ 2x8 y 2 z 2 .

It is now sufficient to prove

X X
x8 y 2 z 2 + 4x7 y 5 ≥ 2x6 y 5 z + 2x6 y 4 z 2 + x5 y 5 z 2 .
sym sym

We are now done because (8, 2, 2) majorises (5, 5, 2) while (7, 5, 0) majorises (6, 5, 1) and
(7, 5, 2).

The expansion and collection of like terms which was suppressed in the above solution would
require an enormous amount of accuracy, time and perseverance if performed without using
symmetric sum notation. Even with this notationX it is somewhat unwieldy. A useful notation
that can be used is to write [i, j, k] to mean xi y j z k . Of course one must understand how
sym
to manipulate the notation. But it’s not that difficult. Perhaps the hardest part is to deduce
that

[i, j, k][p, q, r] = [i + p, j + q, k + r] + [i + p, j + r, k + q]
+ [i + q, j + p, k + r] + [i + q, j + r, k + p]
+ [i + r, j + p, k + q] + [i + r, j + q, k + p],
! !
X X
which represents the expansion of xi y j z k xp y q z r .
sym sym
Note that most expansions are simpler, such as

[i, j, k][p, 0, 0] = 2[i + p, j, k] + 2[i, j + p, k] + 2[i, j, k + p].

11.12 Weighted inequalities

Some inequalities have more general versions that involve weights. These include not just the
AM–GM–HM, but the entire power means inequality. Furthermore, the already very general
Jensen’s inequality also generalises with weights.

Weighted power means inequality For positive real numbers a1 , a2 , . . . , an , and positive
real numbers w1 , w2 , . . . , wn (called weights), define
r1
w1 ar1 + w2 ar2 + · · · + wn arn

1
M0 = (aw 1 w2 wn w1 +w2 +···+wn
1 a2 · · · an ) and Mr =
w1 + w2 + · · · + wn

for every non-zero real number r.

If r > s, then Mr ≥ Ms , and equality occurs if and only if a1 = a2 = · · · = an .
204 11 Inequalities

Weighted Jensen’s inequality If the real numbers x1 , x2 , . . . , xn lie on an interval where

the function f is convex, and the weights w1 , w2 , . . . , wn are any positive real numbers, then
 
w1 f (x1 ) + w2 f (x2 ) + · · · + wn f (xn ) w1 x1 + w2 x2 + · · · + wn xn
≥f .
w1 + w2 + · · · + wn w1 + w2 + · · · + wn

If they lie on an interval where the function f is concave, then the inequality is reversed.

Note that in both weighted inequalities, choosing w1 = w2 = · · · = wn = 1 recovers the

ordinary versions of the inequalities.
Choosing the weights carefully can lead to very powerful results.

Problem Prove the Cauchy–Schwarz inequality from the weighted AM–HM inequality,
where the variables in the Cauchy–Schwarz inequality are positive real numbers.

Solution We prove only the version for two sets of three variables but the proof generalises
quite easily for two sets of n variables. The Cauchy–Schwarz inequality for two sets of three
variables is
(a21 + a22 + a23 )(b21 + b22 + b23 ) ≥ (a1 b1 + a2 b2 + a3 b3 )2 .

The weighted AM–HM inequality for three variables x1 , x2 , x3 , and corresponding weights
w1 , w2 , w3 , is
w1 x1 + w2 x2 + w3 x3 w1 + w2 + w3
≥ w1 w2 w3 ,
w1 + w2 + w3 x1 + x2 + x3

which rearranges as
 
w1 w2 w3
(w1 x1 + w2 x2 + w3 x3 ) + + ≥ (w1 + w2 + w3 )2 .
x1 x2 x3

This looks very similar to the Cauchy–Schwarz inequality. In fact it is identical if we put
ai
wi = ai bi and xi = .
bi

Note that the change of variables at the end of the solution is reversible via
√
√ wi
ai = w i xi and bi = √ .
xi

This demonstrates that the weighted AM–HM inequality and the Cauchy–Schwarz inequality
are in fact the same inequality but stated in different variables!
Using weighted means it is possible to prove the more general Hölder’s inequality.

Hölder’s inequality Let a1 , a2 , . . . , an and b1 , b2 , . . . , bn be two sets of non-negative real

numbers and let p and q be positive real numbers satisfying p1 + 1q = 1, then
1 1
(ap1 + ap2 + · · · + apn ) p (bq1 + bq2 + · · · + bqn ) q ≥ a1 b1 + a2 b2 + · · · + an bn .

In fact this generalises again to more than two sets of variables. The full version is a bit
unwieldy and would rarely be used, so we just state the version for three sets of variables
from which the generalisation should be apparent.
11.12 Weighted inequalities 205

Hölder’s inequality for three sets of variables Let a1 , a2 , . . . , an , b1 , b2 , . . . , bn and

c1 , c2 , . . . , cn be three sequences of n non-negative real numbers and let p, q and r be positive
real numbers satisfying p1 + 1q + 1r = 1, then

n
! p1 n
! q1 n
! r1 n
X X X X
api bqi cri ≥ ai bi ci .
i=1 i=1 i=1 i=1

Here is one more example of how a difficult inequality can be proved using weighted means.

Problem Prove that if a, b, c are positive real numbers, then

a b c
√ +√ +√ ≥ 1.
a2 + 8bc 2
b + 8ac 2
c + 8ab

Solution Apply Jensen’s inequality to the convex function

1
f (x) = √
x

and weights a, b, c in the following way.

af (a2 + 8bc) + bf (b2 + 8ac) + cf (c2 + 8ab) a(a2 + 8bc) + b(b2 + 8ac) + c(c2 + 8ab)
 
≥f
a+b+c a+b+c

This may be rewritten as

3
a b c (a + b + c) 2
√ +√ +√ ≥√ .
2
a + 8bc 2
b + 8ac 2
c + 8ab a + b3 + c3 + 24abc
3

Thus it suffices to prove

(a + b + c)3 ≥ a3 + b3 + c3 + 24abc.

However, this is very easy to prove after the LHS is expanded. We leave the proof of this last
part to the reader.
 Geometric inequalities
12
Geometric inequalities, even simple looking ones, can be fiendishly difficult. Of course, it
helps to have some mastery of geometry as well as some mastery of inequalities. We will look
at a whole new bag of tricks for solving geometric inequalities in this chapter.

12.0 Problems
1. If ABCD is a convex quadrilateral, which point P minimises the sum

P A + P B + P C + P D?

2. If the point P lies inside triangle ABC, prove that

AP + BP < AC + BC.

3. Let ABC be a triangle and let X, Y and Z be the midpoints of BC, CA and AB,
respectively.
Prove that
3
(AB + BC + CA) < AX + BY + CZ < AB + BC + CA.
4

4. Two points A and B lie on different sides of a line `.

(a) Determine the point X on ` which minimises the difference between the lengths of
AX and BX.
(b) Determine the point X on ` which maximises the difference between the lengths
of AX and BX.
5. Of all the triangles with a given base and a given area, which one has the smallest
product of its side lengths?
6. An ant starts on the boundary of a circular disc with radius one metre and walks in a
straight line. Every now and then, it turns left by 60◦ or right by 60◦ , alternating each
time. When the ant reaches the boundary of the disc again, it decides to stop for a rest.
What is the maximum distance that the ant could have travelled?
208 12 Geometric inequalities

7. Two planets A and B lie in space, near to a long, straight asteroid belt which can be
considered as a straight line.
What is the shortest path from A to B via the asteroid belt?

8. Let A and B be two points on a circle.

(a) Determine the point M on the circle which maximises the value of AM 2 + BM 2 .
(b) Determine the point M on the circle which minimises the value of AM 2 + BM 2 .

9. Consider a point M inside a given angle.

Which line through M cuts off a triangle of minimal area from the angle?

10. Determine the point M inside acute triangle ABC which minimises the value of

AM × BC + BM × AC + CM × AB.

11. Let triangle ABC have orthocentre H and circumradius R.

Prove that
AH + BH + CH ≤ 3R.

12. Let P , Q, R and S be points on the sides AB, BC, CD and DA of a parallelogram
ABCD, respectively.
Prove that
P Q + QR + RS + SP ≥ 2AC,
where AC is the shorter diagonal of the parallelogram.

13. Let ABC be a triangle. Let K, L and M be points on BC, AC and AB, respectively.
Let X, Y and Z be points on LM , M K and KL, respectively. Let E1 , E2 , E3 , E4 , E5 ,
E6 and E denote the areas of triangles AM Y , CKY , BKZ, ALZ, BM X, CLX and
ABC, respectively.
Show that p
E ≥ 8 6 E1 E2 E3 E4 E5 E6 .

14. A bee flies for four metres, ending where it began.

Prove that its path can be enclosed in a sphere of radius one metre.

15. Let M and N be the midpoints of sides AD and BC of the convex quadrilateral ABCD.
Prove that
2M N ≤ AB + CD,
with equality if and only if AB is parallel to CD.

16. Consider a quadrilateral of area A and with consecutive side lengths a, b, c and d.
Prove that
ac + bd ≥ 2A,
with equality if and only if the quadrilateral is cyclic and its diagonals are perpendicular.

17. Consider the hexagon formed in a triangle by drawing the three tangents to the incircle
which are parallel to the sides of the triangle.
2
Prove that the perimeter of the hexagon is less than or equal to 3 times the perimeter
of the triangle.
12.0 Problems 209

18. In the plane we are given 5 distinct points A, B, C, P and Q, no three of which are
collinear.
Prove that

AB + BC + CA + P Q < AP + AQ + BP + BQ + CP + CQ.

19. Let ABCDEF be a convex hexagon satisfying

AB = BC, CD = DE and EF = F A.

Prove that
BC DE FA 3
+ + ≥ ,
BE DA F C 2
and determine when equality occurs.
20. Let P be a point inside triangle ABC.
Prove that one of the three angles ∠P AB, ∠P BC, ∠P CA is at most 30◦ .
21. In an acute triangle ABC, let O be the circumcentre and P be the foot of the altitude
from A. Suppose that
∠BCA ≥ ∠ABC + 30◦ .

Prove that
∠CAB + ∠COP < 90◦ .

22. Let ABCDEF be a convex hexagon whose opposite sides are parallel. Let RA , RC and
RE be the radii of circles F AB, BCD and DEF , respectively.
If P is the perimeter of the hexagon, prove that
P
RA + RC + RE ≥ .
2
210 12 Geometric inequalities

12.1 Triangle inequality

Many geometric inequalities boil down to the well-known fact that the shortest path between
two points is given by a straight line segment. For example, it will usually be shorter to travel
in a straight line from A to C than to travel from A to B and then from B to C. This is
precisely the content of the triangle inequality.

Triangle inequality If A, B and C are points, then

AC ≤ AB + BC,

with equality if and only if B lies on the line segment AC.

Problem If M is the midpoint of the side AC in triangle1 ABC, prove that

2BM < AB + BC.

Solution The form of this geometric inequality suggests that we should try to construct
a triangle whose side lengths are AB, BC and 2BM . The natural way to construct a line
segment of length 2BM in our diagram is to extend BM to a point D, where BM = DM .
Alternatively, we can describe D as the unique point in the plane which makes ABCD a
parallelogram.

A D

B C

Applying the triangle inequality in triangle ABD implies that

BD ≤ AB + AD
⇒ 2BM ≤ AB + BC.
Equality occurs if and only if A lies on the line segment BD, which is impossible since ABC
is a triangle.

12.2 Reflection principle

The following problem will give us our first taste of the reflection principle, a beautiful
technique which can be used to solve many geometric inequalities. It’s particularly useful for
problems in which one has to minimise a sum of distinct lengths.

Problem Two towns A and B lie on the same side of a long, straight river.
What is the shortest path from A to B via the river?
1 Inmost of our problems, including this one, triangle refers to a non-degenerate triangle, that is, one in which
the three vertices are not collinear.
12.2 Reflection principle 211

Solution It’s clear that such a shortest path must consist of two straight line segments, AP
and P B, where P is some point on the river. But where should we put the point P ? The
trick is to imagine a ghost town B 0 , located at the reflection of town B through the river, as
shown in the diagram.

A.
. B

.
B0

Then for any point P on the river, the distance P B is exactly the same as the distance P B 0 .
By the triangle inequality, we have

AP + P B = AP + P B 0 ≥ AB 0 ,

with equality if and only if P lies on the line segment AB 0 .

So we should put the point P at the intersection of the line segment AB 0 with the river.

One thing you might notice from this solution is that, in order to obtain the shortest path,
the line segments AP and P B should meet the river at equal angles. This is analogous to the
physical principle which states that when a beam of light is reflected by a mirror, the angle
of incidence equals the angle of reflection.
Let’s now try to use the reflection principle to solve a more substantial problem.

Problem Consider the quadrilateral ABCD with AB = 16, BC = 8, CD = 8 and ∠ABC =

∠BCD = 90◦ . Let P be the point on AB such that AP = 2 and let Q be the point on CD
such that DQ = 3.
Find the length of the shortest path which begins at P , meets the side DA at a point W ,
then meets the side BC at a point X, then meets the side DA at a point Y , then meets the
side AB at a point Z, and then finally ends at Q.

P
W

D
Z Q

B X C
212 12 Geometric inequalities

Solution It might look like a crazy problem, but you’ll see just how easy it is with the help
of the reflection principle. Perhaps surprisingly, we will reflect not once, not twice, not thrice,
but four times!

First take the quadrilateral and reflect it through its side AD.
Then take the new quadrilateral and reflect it through its side BC.
Then take the new quadrilateral and reflect it through its side AD.
Finally, take the new quadrilateral and reflect it through its side AB.

The end result should be a diagram consisting of the original quadrilateral on the far left,
and four reflected copies.

P0
X0
0
W
Y0

Q0
Z0

As you can see, any path

P →W →X→Y →Z→Q
can be copied onto our new diagram to obtain a path

P 0 → W 0 → X 0 → Y 0 → Z 0 → Q0 .

Furthermore, since reflections keep lengths the same, we can be sure that P W = P 0 W 0 ,
W X = W 0 X 0 , XY = X 0 Y 0 , Y Z = Y 0 Z 0 and ZQ = Z 0 Q0 . Of course, in order to minimise
the length of the path from P 0 to Q0 on our new diagram, we should simply take the straight
line segment P 0 Q0 . We can then copy this back on to our original quadrilateral to obtain the
desired shortest path.
As for the path’s length, it can be calculated using Pythagoras’ theorem, since we know
that Q0 is 40 units across√and 9 units down from P 0 . So the length of the shortest path with
the desired properties is 402 + 92 = 41.

We’ll use the reflection principle to full effect in the solution of the following interesting
problem, first posed by Fagnano way back in the 18th century.

Problem Let ABC be an acute triangle and let X, Y and Z be points on the sides BC,
CA and AB, respectively.
Where should X, Y and Z be located so that the perimeter of triangle XY Z is minimal?

Solution First, we’ll minimise the perimeter of triangle XY Z, where X is some fixed point
on BC. The trick is to reflect X in the sides AC and AB to obtain the points P and Q,
respectively.
12.3 Transformations 213

Z
Y

B X C

Then by the triangle inequality,

XY + Y Z + ZX = P Y + Y Z + ZQ ≥ P Q,

with equality if and only if Y lies on the intersection of AC and P Q, while Z lies on the
intersection of AB and P Q. So for a fixed point X on BC, we can find Y and Z which
minimises the perimeter of triangle XY Z. In fact, we know that the minimal perimeter in
this case is precisely the length of P Q.
Now observe that, since P and Q were obtained by reflection we have

AP = AQ = AX.

Furthermore,

∠P AQ = ∠P AX + ∠QAX = 2∠CAX + 2∠BAX = 2∠BAC.

So, triangle AP Q is always an isosceles triangle whose angle at A is equal to 2∠BAC.

Since ∠BAC is fixed, the length P Q is minimised precisely when AP = AQ = AX is
minimised. This means that we should take X to be the foot of the altitude from A.
By a similar argument, Y should be the foot of the altitude from B, and Z should be the
foot of the altitude from C.

When stating Fagnano’s problem, we were careful to mention that ABC is an acute triangle.
What happens when it’s right-angled or obtuse?

12.3 Transformations
We’ve already seen that reflections are particularly useful for solving geometric inequalities,
but they’re certainly not the only transformations which come in handy. In general, transfor-
mations which preserve lengths, known as isometries, may also be useful. In fact, a common
approach is to use an isometry to create a situation in which the triangle inequality can be
employed.

Problem Two towns A and B lie on different sides of a long, straight river with parallel
banks.
Where is the best place to build a bridge, perpendicular to the banks of the river, such that
the path from A to B via the bridge is shortest?
214 12 Geometric inequalities

Solution Suppose that the bridge touches the north bank at P and the south bank at Q, so
that the width of the river is the distance P Q. Now consider the translation which takes B
towards the river by the distance P Q. Then the path from A to B via the bridge has length

AP + P Q + QB = (AP + P B 0 ) + P Q.

A .

Q .
B0
. B

By the triangle inequality, we know that

AP + P B 0 ≥ AB 0 ,

with equality if and only if P lies on the line segment AB 0 .

Therefore, to minimise the path from A to B via the bridge, we should place the point P at
the intersection of AB 0 with the north bank of the river.

12.4 Trigonometry
Trigonometry is often useful for solving geometric inequalities, because we can use various
trigonometric identities as well as the fact that sin θ ≤ 1 and cos θ ≤ 1 for all angles θ. Here
is a nice illustrative example.

Problem Let P be a point inside triangle ABC and let the feet of the perpendiculars from P
to the sides BC, CA, AB be D, E, F , respectively.
Find the point P which maximises the value of
PD · PE · PF
.
PA · PB · PC
Solution Consider the angles a1 and a2 labelled in the following diagram.

a1 a2

E
F
P

B D C
12.5 Parametrisation 215

Using some trigonometric trickery, we have the following.

PF PE
· = sin a1 sin a2
PA PA
1 1
= cos(a1 − a2 ) − cos(a1 + a2 ) (product to sum)
2 2
1 1
≤ − cos A
2 2
A
= sin2 (double angle formula)
2

Equality holds here if and only if a1 = a2 or, in other words, when P lies on the angle bisector
from A. Writing down the other two analogous inequalities, multiplying them together, and
taking the square root gives

PD · PE · PF A B C
≤ sin · sin · sin ,
PA · PB · PC 2 2 2
with equality if and only if P lies on all three angle bisectors.
So the maximum value is achieved when P is the incentre of the triangle.

12.5 Parametrisation
When solving a geometric inequality, it’s sometimes possible to write the expression that
you’re trying to minimise or maximise in terms of certain parameters, such as angles or
lengths. This can often help to turn a geometric inequality into an algebraic inequality.

Problem Let I be the incentre of triangle ABC and let the angle bisectors from A, B and
C meet the opposite sides at A0 , B 0 and C 0 , respectively.
Prove that
AI · BI · CI 8
0 0 0
≤ .
AA · BB · CC 27

Solution The diagram is brimming with angle bisectors while the inequality is brimming
with ratios, which suggests that we might be able to use the angle bisector theorem to our
advantage.

B0
0
C
I

B A0 C
216 12 Geometric inequalities

AI
A natural approach is to try to rewrite the expression AA 0 in terms of some nicer lengths. So

let’s aim to write it in terms of a = BC, b = CA and c = AB, the side lengths of the triangle.
AI
We can’t get the expression AA 0 directly from the angle bisector theorem. However, we can

get the related expression IA0 from using the angle bisector theorem in triangle ABA0 or
AI

ACA0 . The end result is

AI AB AC
= = .
IA0 BA0 CA0
x z x+z
Now we use the simple fact2 that if y = w, then y+w is also equal to them.

AI AB + AC b+c IA0 a
0
= 0 0
= ⇒ =
IA BA + CA a AI b+c

AI
Now we have all the information we need to determine AA0 .

AA0 IA0 + AI IA0 a+b+c AI b+c

= = +1= ⇒ =
AI AI AI b+c AA0 a+b+c

This concludes the parametrisation of the geometric inequality in terms of the side lengths of
the triangle. We have now left geometry behind and all that remains is to prove the algebraic
inequality    
b+c c+a a+b 8
≤ ,
a+b+c a+b+c a+b+c 27
where the only restriction on a, b, c is that they are positive real numbers which obey the
triangle inequality. Applying the AM–GM in the obvious way, we obtain

b+c c+a a+b

!3
+ +
   
b+c c+a a+b a+b+c a+b+c a+b+c
≤
a+b+c a+b+c a+b+c 3
8
= .
27

The inequality is still true if the point I is allowed to be any point inside the triangle. However,
the proof of this requires a different, but not difficult, approach. See if you can prove it!

12.6 Ptolemy’s inequality

All of the geometric inequalities that we’ve solved so far involve fairly well known facts from
geometry or inequalities, along with some tricky techniques. The following theorem is less
well known but we’ll need it in our toolkit.
Ptolemy’s inequality For any four points A, B, C and D, we have

AB · CD + BC · DA ≥ AC · BD.

If the four points do not lie on a line, then equality occurs if and only if ABCD is cyclic
with the points A, B, C and D lying in that order around the circle.

Unfortunately, Ptolemy’s inequality is of no use to us if we don’t know when and where to

apply it. So let’s take a look at an example.
2 This simple fact is called addendo.
12.6 Ptolemy’s inequality 217

Problem In triangle ABC, the angle bisector from A meets the circumcircle again at P .
Similarly, the angle bisector from B meets the circumcircle again at Q and the angle bisector
from C meets the circumcircle again at R.
Prove that
AB + BC + CA < AP + BQ + CR.

Solution

A
Q

B C

Observe that P bisects the arc between B and C on the circumcircle since AP is the angle
bisector from A. In particular, this means that we have
∠A
∠P BC = ∠P CB = ⇒ P B = P C.
2

Now points A, B, P and C are cyclic in that order, so we may use the equality form of
Ptolemy’s inequality to find

AB · P C + P B · AC = AP · BC.

Since P B = P C we may divide both sides by P B · AP to find

AB + AC BC ∠A
= = 2 cos ,
AP PB 2
where the last equality is seen by dropping a perpendicular from P to BC.
Since cos ∠A
2 < 1 we have
AB + AC < 2AP.
Now add this to the two analogous inequalities

BC + BA < 2BQ and CA + CB < 2CR,

and divide by 2 to obtain the desired result.

It seems that Ptolemy’s inequality is a good idea when there are isosceles triangles around.
In the following problem, which was posed by Fermat, we’ll see that it’s an even better idea
when there are equilateral triangles.
218 12 Geometric inequalities

Problem Let ABC be a triangle with all angles less than 120◦ .
Find a point P such that
PA + PB + PC
3
is minimal.

Solution The first thing we’re going to do is find a bound for P B + P C. And the way we’re
going to do that is to use Ptolemy’s inequality with an equilateral triangle. But where is
the equilateral triangle, you might be wondering? Well, let’s just construct the point X so
that triangle BCX is equilateral and external to our original triangle. Now using Ptolemy’s
inequality in BP CX gives
BP · CX + P C · XB ≥ BC · P X.

Cancelling out the terms CX = XB = BC gives us the inequality

P B + P C ≥ P X.

B C

So we’ve learnt that using an equilateral triangle gives us nice cancellation in Ptolemy’s
inequality. We now have
PA + PB + PC ≥ PA + PX
and you just can’t help but use the triangle inequality on this last expression. That gives
P A + P B + P C ≥ AX.

But can we actually achieve equality? Yes we can! For equality to occur in Ptolemy’s
inequality, we need the point P to lie on the circumcircle of triangle BCX on the minor arc
BC. For equality to occur in the triangle inequality, we need P to lie on the line segment AX.
So P should be the intersection of the circumcircle of triangle BCX and the line segment
AX. In fact, the minimal value of P A + P B + P C will simply be the length of AX.
3A slightly weaker version of this problem was solved in section 6.8.
12.7 Locus and tangency 219

Note that the conditions on P imply that ∠BP C = 120◦ . But there was nothing special
about the side BC. We could just as well have constructed our equilateral triangle on CA or
AB in which case we would have arrived at the result CP A = 120◦ or AP B = 120◦ .
So it turns out that P is the unique point in triangle ABC such that

∠BP C = ∠CP A = ∠AP B = 120◦ .

This is known as the Fermat point.

One other thing worth wondering about is why we required triangle ABC to have all angles
less than 120◦ . Please analyse this for yourself.

12.7 Locus and tangency

Recall from chapter 6 that a locus is often thought of as a curve traced out by a point
following some rule.
Tangency occurs when two curves just touch each other. What do locus and tangency have
to do with geometric inequalities? Read on to find out!

Useful facts Let A and B be two points in the plane.

The locus of points M such that ∠AM B is constant is the union of two circular arcs
symmetric about AB.
The locus of points M such that AM 2 + BM 2 is constant is a circle whose centre is
the midpoint of AB.

The locus of points M such that AM + BM is constant is an ellipse whose foci are A
and B.

Problem Of all the triangles with a given base and a given perimeter, which one has the
greatest area?

Solution

M?

A B

Suppose that the given base is AB and that the given perimeter is p. Consider the ellipse
whose foci are A and B such that, for every point M on the ellipse,

AB + AM + BM = p.
220 12 Geometric inequalities

This is all valid because p and AB are constants.

We would like to find the point M on the ellipse which maximises the area of triangle ABM .
In other words, we would like to find the point M on the ellipse that is furthest away from
the line AB. This occurs precisely when M lies on the perpendicular bisector of AB and
triangle ABM is isosceles with base AB.

In the previous problem, the notion of locus was crucial to the solution. In the next, we’ll see
how the notion of tangency also comes into play.

Problem If A and B are two points on the same side of a line `, determine all points P on
` such that ∠AP B is maximised.

Solution First, we consider the case when AB is parallel to `. Then there is a unique circle
which passes through A and B and is tangent to ` at a point T .
Every point P on ` apart from T lies on the same side of AB and outside the circle. Therefore,
for every point P on `, we have the inequality

∠AP B ≤ ∠AT B,

with equality if and only if P = T .

.
A

A . . B .
B

T T1 X T2

Now, consider the case when the line through AB meets ` at some point X. Then there are
two circles which pass through A and B and are tangent to ` at points, which we’ll call T1
and T2 , one on each side of X.
Every point P on ` apart from T1 , but on the same side of X as T1 , lies on the same side of
AB and outside the circle passing through T1 . Therefore, for every point P on ` on the same
side of X as T1 , we have the inequality

∠AP B ≤ AT1 B,

with equality if and only if P = T1 . Similarly, for every point P on ` on the same side of X
as T2 , we have the inequality
∠AP B ≤ AT2 B,
with equality if and only if P = T2 .
It is possible that ∠AT1 B = ∠AT2 B, in which case both points T1 and T2 are solutions to
the problem. If r1 and r2 are the radii of circles ABT1 and ABT2 , respectively, then by the
extended sine rule we have
AB AB
2r1 = = = 2r2 .
sin ∠AT1 B sin ∠AT2 B
12.8 Isoperimetric inequalities 221

This implies that the two circles have equal radius and so AB is perpendicular to `. Conversely,
if AB is perpendicular to `, then by symmetry both T1 and T2 are solutions to the problem.
On the other hand, if we assume without loss of generality that ∠AT2 B < AT1 B, then for
every point P on `, we have the inequality ∠AP B < AT1 B, with equality if and only if
P = T1 .

12.8 Isoperimetric inequalities

Given a rope, what is the maximum area that you can enclose? Perhaps your intuition tells
you that the largest possible area is enclosed by a circular boundary, but how can you be sure?
Your intuition is indeed correct, although trying to construct a rigorous proof is an extremely
difficult task. We simply state this isoperimetric, literally ‘same perimeter’, inequality below
without proof, along with a variant of it pertaining to polygons.

Isoperimetric inequalities
Among all plane figures of a given perimeter, the circle has the maximum area.
Equivalently, among all plane figures of a given area, the circle has the minimum
perimeter.
Among all n-gons of a given perimeter, the regular n-gon has the maximum area.
Equivalently, among all n-gons of a given area, the regular n-gon has the minimum
perimeter.

Problem Prove that among all polygons with given side lengths in a given order, the cyclic
one has the maximal area.

Solution Implicit in the statement of this problem is the fact that if someone gives you a
polygon, then you can always construct a cyclic polygon with the same side lengths in the
same order, and this polygon is unique. We won’t discuss a proof of this fact here, but you
should certainly think about why it is true.
First, if the polygon were not convex, then turning a concave part of the polygon inside out
would yield a polygon of greater area.4 Thus we may assume that the polygon is convex.
Now suppose that P is a convex polygon with given side lengths in a given order which cannot
be circumscribed by a circle. Let Q be the unique cyclic polygon with the same side lengths
in the same order. The circumcircle of Q is naturally divided into the polygon itself as well
as a number of segments. Now consider the shape obtained by gluing those segments onto
the corresponding sides of P .

P −→ Q −→ P

4 This is very sloppy, because the new polygon might cross itself. See if you can iron out the details of this.
222 12 Geometric inequalities

This shape5 has the same perimeter as the circumcircle of Q. So, by the isoperimetric
inequality, its area is smaller than the area of the circumcircle of Q. However, after removing
the segments from each shape, we obtain the fact that the area of P is less than the area
of Q. So we can conclude that, among all convex polygons with given side lengths in a given
order, the cyclic one has the maximal area.

12.9 Incircle substitution

Incircle substitution Suppose that triangle ABC has sides of length BC = a, AC = b

and AB = c. Let the incircle of triangle ABC be tangent to BC at X, AC at Y and AB
at Z.
The incircle substitution is given by putting

a = y + z, b = x + z, and c = x + y,

where x = AY = AZ, y = BX = BZ and z = CX = CY .

x
x

z
y

B y z C

Problem If a, b, c are the side lengths of a triangle, prove that

abc ≥ (b + c − a)(c + a − b)(a + b − c).

Solution Using the incircle substitution, the inequality transforms to something which looks
much nicer, that is,
(y + z)(z + x)(x + y) ≥ 8xyz.
This follows from multiplying the three inequalities
√ √ √
y + z ≥ 2 yz, z + x ≥ 2 zx and x + y ≥ 2 xy

together. Each of these can be proved by using the AM–GM inequality or the fact that
squares are non-negative.

12.10 Triangle formulas

Sometimes knowing lots of triangle formulas can help turn a geometric inequality into an
algebraic inequality.
5 You should think about why the circular arcs around P do not intersect each other.
12.10 Triangle formulas 223

Problem Prove that R ≥ 2r, where R and r are respectively the circumradius and inradius
of a triangle.

Solution Perhaps surprisingly, one solution to this problem arises from considering the
area A of the triangle. In particular, we have the three formulas
abc p
A= , A = rs, and A = s(s − a)(s − b)(s − c).
4R
Here, a, b and c are the side lengths of the triangle and s = a+b+c
2 denotes the semiperimeter.
If you’ve never seen these before, then now is definitely the time to try to prove them. The
third one is particularly interesting and is often referred to as Heron’s formula. From these,
we can deduce the equation
abcrs
= s(s − a)(s − b)(s − c),
4R
from which
r (s − a)(s − b)(s − c)
= .
4R abc
However, the result of the previous problem implies that
8abc ≥ (s − a)(s − b)(s − c).
Thus, it follows that
r 1
≤ ,
4R 8
which simplifies to give R ≥ 2r.

This result is often known as Euler’s inequality because it follows immediately from the
following interesting result.

Euler’s theorem in geometry The distance d between the circumcentre and the incentre
of a triangle is given by p
d = R2 − 2Rr.

You might like to prove this result by yourself, thereby providing an alternative proof of
Euler’s inequality. As a treat, we’ll finish with the most elegant proof of this chapter—another
proof of Euler’s inequality.

Solution We start with a simple observation: the smallest circle which touches or crosses
all three sides of a triangle is the incircle.
Recall that the midpoints of the sides of a triangle form what is known as the medial triangle.
It is always similar to the original triangle and exactly half the size. Therefore, the circumcircle
of the medial triangle6 has radius R2 .
Observe that the circumcircle of the medial triangle passes through the midpoints of all three
sides of the original triangle. So, from our earlier simple observation, the circumcircle of the
medial triangle is at least as large as the incircle.7
Hence, we may conclude that
R
≥ r,
2
which can be equivalently written as R ≥ 2r.
6 This is also the nine-point circle.
7 Infact, as a carefully drawn diagram suggests, the incircle is always internally tangent to the nine-point
circle. But this is more difficult to prove.
224 12 Geometric inequalities

B C
 Combinatorics
13
In a very loose sense, combinatorics is the area of mathematics concerned with counting. By
this, we don’t mean calling out the numbers 1, 2, 3, . . . or anything like that. Combinatorics
is rather about counting in a clever way, or counting without counting. Actually, this isn’t
a particularly good description of combinatorics either. Pretty much anything that doesn’t
quite fit into another mathematical category tends to be described as combinatorial. This
makes combinatorics a treasure trove of mathematical delights!

13.0 Problems
1. How many whole numbers from 1 to 2003 are not divisible by 2, 7 or 11?
2. A number is said to be ordered if each digit is greater than or equal to the digit on its
left.
How many three-digit numbers are ordered?
3. How many eight-digit phone numbers begin with the number 9, are even, and have the
third, fourth and fifth digits in decreasing order?
4. How many odd numbers are there in the 2003rd row of Pascal’s triangle?
5. The number 3 can be expressed as a sum of one or more positive integers (taking order
into account) in four ways:

3 = 1 + 2 = 2 + 1 = 1 + 1 + 1.

How many ways can a positive integer n be so expressed?

6. In how many ways can a person obtain a score of 6 in a test containing 5 questions,
each one marked out of 7?
7. How many ways are there to place two kings on a chessboard so that no two attack
each other?
8. A spider has one sock and one shoe for each of its eight legs.
In how many different orders can the spider put on its socks and shoes, assuming that
on each leg the sock must be put on before the shoe?
226 13 Combinatorics

9. Prove that
n     
X n k n−m n
=2 .
k m m
k=m

10. Prove in at least two different ways that

 2  2  2  
n n n 2n
+ + ··· + = .
0 1 n n

11. Prove in at least two different ways that if m, n are positive integers and 1 ≤ k ≤ n,
then           
n m n m n m m+n
+ + ··· + = .
0 k 1 k−1 k 0 k

(This is a generalisation of the previous problem.)

12. Find the number of k-tuples (S1 , S2 , . . . , Sk ) satisfying

S1 ⊆ S2 ⊆ · · · ⊆ Sk ⊆ {1, 2, . . . , n}.

13. Let there be given nine lattice points in 3-dimensional space.

Show that there is a lattice point on the interior of one of the line segments joining two
of these points.

14. In a circus, there are n clowns who dress and paint themselves up using a selection of
12 distinct colours. Each clown is required to use at least five different colours. One
day the ringmaster of the circus demands that no two clowns have exactly the same set
of colours, and no more than 20 clowns may use any one particular colour.
Find the largest number n so that the ringmaster’s demand can be met.

15. In a maths competition, 2n students take part. Each student submits a different problem
and all 2n problems are collected, shuffled, and then handed back to each participant.
The distribution is considered fair if there are n participants receiving the problems
proposed by the other n participants.
Prove that the number of ways in which the problems can be distributed in a fair way
is a perfect square.

16. An n × n matrix which has entries coming from the set S = {1, 2, . . . , 2n − 1} is called a
silver matrix if, for each i = 1, 2, . . . , n, the ith row and the ith column together contain
all the elements of S.
Prove the following.

(a) There is no silver matrix for n = 1997.

(b) Silver matrices exist for infinitely many values of n.

17. An 8 × 8 square is divided into 64 unit squares, and must be covered by√64 black and 64
white isosceles right-angled triangles whose side lengths are 1, 1 and 2, respectively.
Each unit square must be covered by 2 triangles. A covering is said to be awesome if
any two triangles sharing a common side are of distinct colours.
How many different awesome coverings are there?
13.0 Problems 227

18. A chess master who has 11 weeks to prepare for a tournament decides to play at least
one game every day, but in order to conserve his energy he decides not to play more
than 12 games in any seven-day period.
Show there are consecutive days during which he plays exactly 21 games.
19. Let S = {0000000, 0000001, . . . , 1111111} be the set of all binary sequences of length 7.
The distance between two elements s1 , s2 ∈ S is the number of places in which s1 and
s2 differ.
Show that if T ⊂ S and |T | > 16, then T contains two elements whose distance is at
most 2.
20. A rectangular chessboard has 5 rows and 2008 columns. Each square is painted either
red or blue.
Determine the largest integer N which guarantees that, no matter how the board is
painted, there are two rows which have matching colours in at least N columns.
21. Let n > 0 be an integer. We are given a balance scale and n weights of masses
20 , 21 , . . . , 2n−1 , respectively. We are to place each of the n weights on the balance scale,
one after another, in such a way that the right pan is never heavier than the left pan.
At each step we choose one of the weights that has not yet been placed on the balance
scale, and place it on either the left pan or the right pan, until all of the weights have
been placed.
Determine the number of ways in which this can be done.
22. Up until now the national library of the small city state of Sepharia has had n shelves,
each shelf holding at least one book. The library recently bought k new shelves. The
books will be rearranged, and the librarian has announced that each of the now n + k
shelves will hold at least one book. Call a book privileged if the shelf on which it
will stand in the new arrangement is to hold fewer books than the shelf where it was
previously located.
Prove there are at least k + 1 privileged books in the national library of Sepharia.
23. Let n and k be positive integers with k ≥ n and k − n an even number. Let 2n lamps
labelled 1, 2, . . . , 2n be given, each of which can be either on or off. Initially all the
lamps are off.
We consider sequences of steps: at each step one of the lamps is switched from on to off
or from off to on.
Let N be the number of such sequences consisting of k steps and resulting in the state
where lamps 1 through n are all on, and lamps n + 1 through 2n are all off.
Let M be the number of such sequences consisting of k steps resulting in the state
where lamps 1 through n are all on, and lamps n + 1 through 2n are all off, but where
none of the lamps n + 1 through 2n is ever switched on.
N
Determine the ratio M.

24. A group of 21 girls and 21 boys took part in a mathematical contest. Suppose that
(i) each contestant solved at most six problems, and
(ii) for each girl and each boy, at least one problem was solved by both of them.
Prove that there was a problem that was solved by at least three girls and at least three
boys.
228 13 Combinatorics

13.1 Addition and multiplication

Suppose that the Irish restaurant chain O’Donald’s has five types of burger and three types
of drink. If you would like a burger or a drink, then you have 5 + 3 = 8 choices. On the other
hand, if you would like a burger and a drink, then you have 5 × 3 = 15 choices.
Many combinatorics problems boil down to applying these addition and multiplication
principles in various rather ingenious ways. To make sure that you have some familiarity with
these ideas, consider the following basic examples of combinatorial questions. If necessary,
you should read over these carefully until you fully understand what is going on—especially
because what goes on here, though basic, is extremely useful.

Problem In how many ways can n people stand in a line?

Solution There are n choices of position for the first person to stand in the line. After the
first position has been chosen, there are n − 1 positions remaining for the second person. And
after the second position has been chosen, there are n − 2 positions remaining for the third
person, and so on.
Therefore, the number of ways to choose the order in which n people stand in a line is

n × (n − 1) × (n − 2) × · · · × 2 × 1.

You probably already know that this number is written as n! and is called ‘n factorial’.

Problem How many three-digit numbers contain no zeros or nines as digits?

Solution There are 8 choices for the first digit, 8 choices for the second digit and 8 choices
for the third digit, that is, any number from 1 to 8.
Hence, there are 83 = 512 such numbers.

Problem Each letter in Morse code is a sequence of at most four dots and dashes. How
many letters are possible?

Solution There are 21 = 2 letters with 1 signal, 22 = 4 letters with 2 signals, 23 = 8 letters
with 3 signals, and 24 = 16 letters with 4 signals.
Hence, the answer is 2 + 4 + 8 + 16 = 30.

So Morse code only just copes with the 26 letters of the English alphabet!

13.2 Subtraction and division

Sometimes the best way to solve a counting problem is to count too much—a technique often
known as overcounting. If you have counted k too many objects, then the answer is obtained
by subtracting k. On the other hand, if you have counted each object k times, then the
answer is obtained by dividing by k.
Many combinatorics problems boil down to applying these subtraction and division principles
in conjunction with the addition and multiplication principles we discussed in the previous
section.

Problem How many five-digit positive integers have at least two digits the same?
13.3 Binomial identities 229

Solution The number of five-digit positive integers is 99999 − 10000 + 1 = 90000.

The number of five-digit positive integers with no two digits the same is 9×9×8×7×6 = 27216.
This is because there are 9 ways to choose the first digit (because it cannot be zero), 9 ways
of choosing the second digit because it cannot be the same as the first, 8 ways of choosing
the third digit because it cannot be equal to either of the first two digits, and so on.
Hence, the answer is 90000 − 27216 = 62784.

Problem In how many ways can you choose a four-flavour combination from 10 ice-cream
flavours?

Solution Well, if you actually cared about the order in which you chose them, then the
answer would simply be 10×9×8×7. But since the order doesn’t matter, we have overcounted
each combination by a factor of 4 × 3 × 2 × 1—the number of ways of reordering the four
flavours that we have chosen.
Therefore, the answer is
10 × 9 × 8 × 7
= 210.
4×3×2×1
This can be conveniently expressed using factorial notation as
 
10 10!
= .
4 4! 6!

Problem In how many ways is it possible to arrange the letters of the word recurrence?

Solution There are 10! ways to rearrange the letters, but again we have overcounted. The
three occurrences of the letter r can be rearranged in 3! ways among themselves. In general,
if a letter occurs k! times, then there are k! ways of rearranging the occurrences of that letter
among themselves. Since the letters r and e occur three times, the letter c twice and the
letters u and n once, the answer is
 
10 10!
= = 50400.
3, 3, 2, 1, 1 3! 3! 2! 1! 1!

13.3 Binomial identities

A lot of fun can be had playing around with binomial coefficients. These lie in the intersection
of algebra, combinatorics, polynomials and even calculus, giving rise to many problems with
multiple solutions from different areas of mathematics.
Recall the following basic identities.

Symmetry    
n n
=
k n−k
Binomial theorem
n  
X n k n−k
(x + y)n = x y
k
k=0

Addition formula      
n n−1 n−1
= +
k k−1 k
230 13 Combinatorics

In-and-out formula    
n−1 n
n =k
k−1 k

Problem For every positive integer n, prove that

       
n n n n
− + − · · · + (−1)n = 0.
0 1 2 n

Solution We provide two very distinct proofs of this fact. In the next section we provide
a third proof! This is a rather common feature of binomial identities and it pays to be
comfortable with the various types of proof.

Method 1: Pascal’s triangle

Recall that we can obtain the nth row of Pascal’s triangle from the (n − 1)th row by
adding the two terms above each element of the
nth row as per the addition formula
above. In particular, with the convention that nk = 0 if k < 0 or k > n, we have
     
n n−1 n−1
= +
0 −1 0
     
n n−1 n−1
= +
2 1 2
     
n n−1 n−1
= +
4 3 4
..
.
and so n0 + n2 + n4 + · · · is equal to the sum of all the binomial coefficients on the

(n − 1)th row of Pascal’s triangle. Similarly,

     
n n−1 n−1
= +
1 0 1
     
n n−1 n−1
= +
3 2 3
     
n n−1 n−1
= +
5 4 5
..
.
and so n1 + n3 + n5 + · · · is also equal to the sum of all the binomial coefficients on

the (n − 1)th row of Pascal’s triangle. Therefore,

           
n n n n n n
+ + + ··· = + + + ··· ,
0 2 4 1 3 5
which is equivalent to what we wanted to prove. 
Method 2: Algebra
The binomial theorem states that
       
n n 0 n n 1 n−1 n 2 n−2 n n 0
(x + y) = x y + x y + x y + ··· + x y .
0 1 2 n
All we need to do now is substitute x = −1 and y = 1 to obtain the desired result.
13.4 Bijections 231

Problem For every positive integer n, prove that

       
n n n n
+2 +3 + ··· + n = n2n−1 .
1 2 3 n

Solution

Method 1: Algebra
We use the in-and-out formula to write k nk = n n−1



k−1 . Then the left-hand side becomes
     
n−1 n−1 n−1
n +n + ··· + n = n2n−1 ,
0 1 n−1
n−1
X 
n−1
where = (1 + 1)n−1 is a consequence of the binomial theorem.
k
k=0

Method 2: Calculus
Substituting y = 1 into the binomial formula, we obtain
         
n n n 2 n 3 n n
(x + 1)n = + x+ x + x + ··· + x .
0 1 2 3 n

Now look at what happens when we differentiate both sides.

       
n n n 2 n n−1
n(x + 1)n−1 = +2 x+3 x + ··· + n x
1 2 3 n

All we need to do now is substitute x = 1 to obtain the desired result.

13.4 Bijections
Another way to prove combinatorial identities is to look for combinatorial interpretations for
both sides of the identity. This is really a good technique and a useful skill to have. Here we
discuss two problems we considered earlier but present bijective solutions.

Problem For every positive integer n, prove that

       
n n n n
− + − · · · + (−1)n = 0.
0 1 2 n

Solution The identity is equivalent to the fact that, for every positive integer n,
X n X n
= .
k k
k odd k even

If S is a set with n elements, the left-hand side counts the number of subsets of S with an
odd number of elements while the right-hand side counts the number of subsets of S with an
even number of elements.
Now that we have a combinatorial interpretation for both sides of the equation, it makes sense
to look for a one-to-one correspondence—a bijection to use more precise terminology—between
the objects described by the left-hand side and the objects described by the right-hand side.
232 13 Combinatorics

In other words, we want to find a rule for turning a subset of S with an odd number of
elements into a subset of S with an even number of elements, and vice versa.
We start by picking some fixed s ∈ S. The rule is to add the element s to your set if it doesn’t
already contain it, and to remove the element s from your set if it does already contain it.
Since we are either adding or subtracting a single element, this rule certainly turns a subset
of S with an odd number of elements into a subset of S with an even number of elements,
and vice versa. It is a bijection because the rule can be reversed. In fact, the rule is its own
inverse!1 This completes our combinatorial proof.

Problem For every positive integer n, prove that

       
n n n n
+2 +3 + ··· + n = n2n−1 .
1 2 3 n

Solution The idea is to find a set of objects which can be counted in two ways—one of which
produces the left-hand side while the other produces the right-hand side. In this particular
case, the form of the left-hand side gives a strong hint as to what that set might be.
Each term of the left-hand side is of the form k nk which suggests that, from



a set of n people,
we would like to choose a committee of k people. This can be done in nk ways. Furthermore,
we would like to choose a president from this committee. This can be done in k ways. As k
ranges from 1 up to n, we are choosing committees of every possible size. So what we have
shown is that the left-hand side counts the number of ways to choose a committee, with one
person designated the president, from a set of n people.
Of course, all that remains is to show that the number of ways to choose a committee, with one
person designated the president, from a set of n people also happens to be n2n−1 . Whereas
we earlier counted by choosing the committee first and then the president, we will now count
by choosing the president first and then the remainder of the committee. But this is easy,
because there are n choices for the president and for each of the remaining n − 1 people, there
are two choices. Each person is either in the committee or is not. Thus the number of ways of
choosing a committee with president is n2n−1 . This completes our combinatorial proof.

13.5 The supermarket principle

This is a common trick in counting problems. Although it is a well-defined technique, it does
not seem to have a well-defined name but is another example of bijective combinatorics. We
shall call it the supermarket principle because of the following problem.

Problem The supermarket has an unlimited supply of apple tarts, chocolate muffins and
cheesecakes.
How many ways are there to buy seven cakes from the supermarket?

Solution You buy your seven cakes and take them to the checkout, where you put them
on the conveyor belt. Just to make things easy, you place all the apple tarts together first,
then the chocolate muffins, and then the cheesecakes. To make things really easy for the
stressed-out checkout attendant, you place some of those conveyor belt dividers between each
type of cake.
Now there will be nine objects on the conveyor belt: seven cakes and two dividers. (If you
buy no apple tarts, for example, the first object will be a divider.) Choosing which of these
1 As we saw in section 10.7, a rule or a function which is its own inverse is called an involution.
13.6 Pigeonhole principle 233

nine
objects are the dividers uniquely determines what cakes you have bought. So the answer
is 92 = 36.

In this question, we established a bijective correspondence.

   
ways of buying 7 ways of placing 2
↔
cakes of 3 types dividers among 7 cakes

You should convince yourself that this is actually correct. That is, for each way of buying
the prescribed cakes, there is a way of placing the dividers; and conversely, for each way of
placing the dividers, there is a way of buying cakes so that the dividers are placed there. Both
of these are quite obvious once you understand what is going on, but often this is difficult to
express!
Now for a less supermarket-oriented and slightly more subtle example. There are certainly
other solutions to it, but perhaps this is the most elegant.

Problem Find the number of three-digit numbers whose digit sum is 10.

Solution Represent a number by a collection of ‘units’ (1s) and ‘dividers’ (∆s). For example,
325 is represented by
1 1 1 ∆ 1 1 ∆ 1 1 1 1 1.
Now every such three-digit number corresponds to a way of placing two dividers among ten
1s. However the correspondence is not bijective! This is because the ‘conveyor belt’

∆1∆111111111

would correspond to the number 019, which is not a three-digit number. You can’t have a
divider in the first place. But provided that you start with a 1, the correspondence is bijective.
(You should make sure of this.)
Ignoring the initial 1, the answer is the number of ways of placing two dividers amongst the
remaining nine 1s. There are 11 objects
in total and so the number of ways of nominating
two of them to be dividers is simply 11 2 = 55.

13.6 Pigeonhole principle

Although the pigeonhole principle was introduced in sections 1.9 and 1.10, it is a powerful
combinatorial tactic which we showcase here.

Problem Given 27 distinct odd positive integers less than 100, prove there is a pair of them
whose sum is 102.

Solution Consider the following sets of odd positive integers.

{1, 101}, {3, 99}, . . . , {49, 53}, {51}

Note that each of the first 25 sets consists of two numbers whose sum is 102, while the 26th
set consists of a single number. Furthermore, the 26 sets form a partition of all the odd
positive integers less than 100. So given 27 numbers (pigeons), some set (pigeonhole) has two
different numbers (pigeons) chosen (living) from (in) it! Those two numbers add to 102.
234 13 Combinatorics

Problem The prime factorisations of r + 1 positive integers a1 , . . . , ar+1 together involve

only r primes p1 , . . . , pr .
Prove that there is a non-empty subset of these integers whose product is a perfect square.

Solution Recall that a positive integer is a perfect square if and only if the exponent of
every prime in its prime factorisation is even. So given a product of powers of these primes
n = pα αr
1 · · · pr , define the parity pattern of n to be (α1 , . . . , αr ), where the αi are considered
1

modulo 2.
Clearly when you multiply two numbers, you add their parity patterns term by term
modulo 2, and if you can divide, then you subtract their parity patterns modulo 2. We just
need to show there is a non-empty subset of our r + 1 integers whose product has parity
pattern (0, . . . , 0).
So, consider all 2r+1 subsets of our r + 1 integers, and consider their parity patterns. But
there are only 2r possible parity patterns. By the pigeonhole principle there are two distinct
subsets X and Y of {a1 , . . . , ar+1 } whose products have the same parity pattern v.
If X and Y were disjoint, then the product of the elements of X ∪ Y would have parity
pattern 2v = (0, . . . , 0), as required.
If X and Y overlap, let D = X ∩ Y . We claim that the product of the elements of X ∪ Y \ D
is a perfect square. Indeed consider the equality,
      
Y Y Y Y Y
 x  y =  x  y  d2  .
x∈X y∈Y x∈X\D y∈Y \D d∈D

Y Y
The LHS has parity pattern (0, . . . , 0) because x and y have the same parity pattern.
x∈X y∈Y
Y
Also d2 has parity pattern (0, . . . , 0). It follows that
d∈D

  
Y Y
 x  y
x∈X\D y∈Y \D

also has parity pattern (0, . . . , 0) and so the product of the elements of X ∪ Y \ D is a perfect
square, as required.

13.7 Principle of inclusion–exclusion

Here is a basic idea of set theory.2

|X ∪ Y | = |X| + |Y | − |X ∩ Y |
|X ∪ Y ∪ Z| = |X| + |Y | + |Z| − |X ∩ Y | − |Y ∩ Z| − |Z ∩ X| + |X ∩ Y ∩ Z|

This can be generalised as per the following theorem.

2 Why not draw the corresponding Venn diagrams to see what is going on?
13.8 Double counting 235

Principle of inclusion–exclusion
X
|X1 ∪ · · · ∪ Xn | = (−1)k−1 |Xi1 ∩ · · · ∩ Xik | .
{i1 ,...,ik }⊆{1,...,n}

It can also be inverted, since

(S \ X1 ) ∪ · · · ∪ (S \ Xn ) = S \ (X1 ∩ · · · ∩ Xn )

and
(S \ X1 ) ∩ · · · ∩ (S \ Xn ) = S \ (X1 ∪ · · · ∪ Xn ),
which yield
X
|X1 ∩ · · · ∩ Xn | = (−1)k−1 |Xi1 ∪ · · · ∪ Xik | .
{i1 ,...,ik }⊆{1,...,n}

Problem An absent-minded postman has n letters to deliver to n different addresses.

In how many ways can he deliver the mail, one letter to each address, so that no letter is
delivered to its correct address?

Solution This problem really involves permutations, which we consider to be bijections from
the set {1, 2, . . . , n} to itself or equivalently, arrangements of the numbers 1, 2, . . . , n. What
we are seeking is the number Dn of derangements of the set {1, 2, . . . , n}. A derangement is a
permutation of a set which leaves no element fixed.
For i = 1, 2, . . . , n, let Xi denote the set of permutations which fix the number i. The number
of elements in Xi is easy to count and it is simply (n − 1)!. Furthermore, we know that
Xi1 ∩ Xi2 ∩ · · · ∩ Xik is simply the set which fixes the number i1 , fixes the number i2 , and so
on, up to the number ik . This is a set which is also easy to count and its number of elements
is simply (n − k)!. Note that the total number of ways of choosing k of the Xi is equal to nk .

What does the set X1 ∪ X2 ∪ · · · ∪ Xn represent? It simply represents the number of

permutations that fix 1 or 2 or 3, and so on, up to n. In other words, it is the set of
permutations that fix at least one of 1, 2, . . . , n. So the number of elements it contains is
simply n! − Dn . Therefore,

X
|X1 ∪ · · · ∪ Xn | = (−1)k−1 |Xi1 ∩ · · · ∩ Xik |
{i1 ,...,ik }⊆{1,...,n}
n  
X
k−1 n
⇒ n! − Dn = (−1) (n − k)!
k
k=1
n
X n!
⇒ Dn = n! − (−1)k−1
k!
k=1
(−1)n
 
1 1 1 1
= n! − + − + ··· + .
0! 1! 2! 3! n!

13.8 Double counting

Double counting is short for counting the same thing in two different ways.
236 13 Combinatorics

Problem At a party each person knew exactly 22 others. For any pair of people X and Y
who knew one another, there was no other person at the party whom they both knew. For
any pair of people X and Y , who did not know one another, there were exactly 6 other people
whom they both knew.
How many people were at the party?

Solution This problem has an obvious graph theory interpretation, where vertices represent
people and edges represent a mutual acquaintanceship between two persons. Define a vee to
be a triple of persons such that exactly two of the three pairs of acquaintances know each
other. We count the number of vees in two different ways.3
Suppose there
are n people at the party. Concentrating on vertices we see that each vertex
contributes 22
2 = 231 vees since each vertex has 22 edges emanating from it. Thus the total
number of vees is 231n.
On the other hand adding the degrees of each vertex gives 22n, but this overcounts the
number of edges by a factor of two. Therefore, the total number of edges is 11n. This means
that the total number of pairs of vertices not connected by an edge is
 
n
− 11n.
2

Each such non-edge makes a vee with 6 other vertices. Thus the total number of vees is
  
n
6 − 11n .
2

Equating the two expressions for the number of vees yields

  
n
231n = 6 − 11n .
2

This is easily solved for n and yields n = 100.

Problem Let pn (k) be the number of permutations of the set {1, 2, . . . , n} which have
exactly k fixed points.
Prove that
n
X
kpn (k) = n!.
k=0

Solution We find a combinatorial interpretation for the sum. We are counting each per-
mutation, but counting each with multiplicity equal to the number of its fixed points. So
permutations without any fixed points, that is, derangements, are not counted at all; permu-
tations with one fixed point are counted once; those with two fixed points are counted twice;
and so on. Consider a permutation as the numbers 1 to n, written in some order.
For each permutation with one or more fixed points, colour in one of the fixed points as
our
Pn ‘favourite’. Then the set of ‘permutations with favourite fixed points’ has precisely
k=0 kpn (k) elements. This is a combinatorial interpretation for the sum.
For example, here are the ‘permutations with favourite fixed points’ for the case n = 3.

123, 123, 123, 132, 213 and 321.

3 In the graph theory interpretation a vee actually looks like the letter V.
13.9 Injections 237

How many of these objects are there? Well, with 1 as the favourite fixed point, there are
(n − 1)! permutations—one for each permutation of the other objects. Similarly, with 2 as the
favourite fixed point, there are also (n − 1)! permutations. There are (n − 1)! permutations
for each individual favourite fixed point.
This gives n × (n − 1)! = n! of them overall.4

13.9 Injections
Problem A permutation (x1 , x2 , . . . , x2n ) of the set {1, 2, . . . , 2n}, where n is a positive
integer, is said to be good if
|xi − xi+1 | = n
for at least one i in {1, 2, . . . , 2n − 1}, and is said to be bad otherwise.
Show that, for each n, there are more good permutations than bad permutations.

Solution Since we want to show there are more of one thing than another, we don’t construct
a bijection, but an injection! We find a map from bad permutations to good permutations
which is injective but not surjective: this will show there are more good permutations.
First think about what a good permutation means. For given x ∈ {1, . . . , 2n}, there is only
one y in the set for which |x − y| = n. So a permutation is good if and only if we see at least
one of the pairs
(1, n + 1), (2, n + 2), . . . , (n, 2n)
occurring as adjacent numbers in the permutation (in any order). A bad permutation is one
where we see none of these pairs of numbers adjacent. In what follows, it will be convenient
to define the notation i ∗ n as follows.

i + n if 1 ≤ i ≤ n
i∗n=
i − n if n + 1 ≤ i ≤ 2n

Now take any bad permutation. Here is how to make a good one out of it. Suppose x1 = i.
Then i ∗ n can’t occur as x2 , but must occur later on. Thus the permutation looks like
(i, A, i ∗ n, B),
where A is a non-empty sequence, and B is another (possibly empty) sequence, neither of
which contains a pair of the form (j, j ∗ n). We make our permutation good by putting i and
i ∗ n together to make
(A, i, i ∗ n, B).
Thus we have a map φ from the set of bad permutations to the set of good permutations
given by
φ(i, A, i ∗ n, B) = (A, i, i ∗ n, B).

We first show our map φ is not surjective. Any good permutation in the image of φ is of the
form (A, i, i ∗ n, B), where A is non-empty and the only adjacent pair of the form (j, j ∗ n)
occurs if j = i. Clearly there are more good permutations than these. For example, any
permutation of the form (1, n + 1, . . .) is a good permutation that is not in the image of φ.
Now we show φ is injective. Suppose that φ(i1 , A1 , i1 ∗ n, B1 ) = φ(i2 , A2 , i2 ∗ n, B2 ). Then
from the definition of φ we have
(A1 , i1 , i1 ∗ n, B1 ) = (A2 , i2 , i2 ∗ n, B2 .)
4 There are many other solutions to this problem. See if you can find any others.
238 13 Combinatorics

However from our earlier observation the only adjacent pair of the form (j, j ∗ n) on the LHS
occurs at j = i1 . Similarly, the only adjacent pair of the form (j, j ∗ n) on the RHS occurs at
j = i2 . This allows us to deduce that i1 = i2 . Then it follows that A1 = A2 and B1 = B2 .
Therefore, φ is injective.

13.10 Recursion
As a first example of the technique of recursion, we solve the problem about the absent-minded
postman that we saw in section 13.7.

Problem An absent-minded postman has n letters to deliver to n different addresses.

In how many ways can he deliver the mail, one letter to each address, so that no letter is
delivered to its correct address?

Solution Suppose that Dn is the number of derangements of n objects. The trick here is to
relate Dn to some of the previous values D1 , D2 , . . . , Dn−1 .

We will take a derangement of n − 1 letters and construct n − 1 distinct derangements

of n letters. Suppose that in our derangement of n − 1 letters, letter 1 is delivered
to address j. Then we adjust this by delivering letter 1 to address n and letter n
to address j. This gives a derangement of n letters. In general, we could take any
i ∈ {1, 2, . . . , n − 1} and suppose that in our derangement of n − 1 letters, letter i is
delivered to address j. Then we adjust this by delivering letter i to address n and letter
n to address j. This gives a derangement of n letters.
So given a derangement of n − 1 letters and a number i ∈ {1, 2, . . . , n − 1}, we have
constructed a derangement of n letters. You can check that these are indeed all different.
Thus we have constructed
(n − 1)Dn−1
derangements of n letters.

Unfortunately, we have not accounted for every possible derangement of n letters. The
ones we are missing are precisely those where letter n is delivered to address j and
letter j is delivered to address n for some j ∈ {1, 2, . . . , n − 1}. But in this case, notice
that the remaining letters and addresses form a derangement on n − 2 letters. Therefore,
for each j there are Dn−2 ways in which this can happen. But since we can take any
j ∈ {1, 2, . . . , n − 1} there are
(n − 1)Dn−2
cases in total, where the letter n has swapped addresses with another letter.

Putting our two pieces of information together we have

Dn = (n − 1)(Dn−1 + Dn−2 ).

This is a nice simple recursion for Dn which we can manipulate as follows.

Dn − nDn−1 = −(Dn−1 − (n − 1)Dn−2 )

Letting
En = Dn − nDn−1 ,
13.10 Recursion 239

for n = 1, 2, . . ., we see that

En = −En−1 = En−2 = · · · = (−1)n−2 E2 = (−1)n−2 (D2 − 2D1 ) = (−1)n .

Here we have used D2 = 1 and D1 = 0.

Therefore,
Dn − nDn−1 = (−1)n ,
and so,
Dn Dn−1 (−1)n
− = .
n! (n − 1)! n!

Since this is true for all integers n ≥ 1 we also have the same equation for n replaced
successively by n − 1, n − 2, . . . , 1. Upon summing these equations, all the middle terms on
the left-hand side cancel out leaving us with

Dn D0 (−1)n (−1)n−1 (−1)1

− = + + ··· + ,
n! 0! n! (n − 1)! 1!

which is equivalent to the result we obtained earlier in section 13.7.

Problem In town A there are n girls and n boys such that each girl knows each boy. In
town B there are n girls and 2n − 1 boys such that girl k knows boys 1, 2, 3, . . . , 2k − 1, and
only these boys. Let A(n, r) denote the number of different ways in which r girls from town A
can dance with r boys from town A, forming r pairs where the girl knows the boy. Similarly,
let B(n, r) denote the number of different ways in which r girls from town B can dance with
r boys from town B, forming r pairs where the girl knows the boy.
Prove that A(n, r) = B(n, r), for r = 1, 2, . . . , n.

Solution
We can calculate the number A(n, r) very easily.
The number of ways of choosing
r girls is nr and the number of ways of choosing r boys is nr . The number of ways of pairing
them up is r!. Therefore, we have
 2
n n!2
A(n, r) = r! = .
r (n − r)!2 r!

To show that
n!2
B(n, r) =
(n − r)!2 r!
directly is a very difficult task indeed. However, what we can do is relate B(n, r) to the values
B(n − 1, r − 1) and B(n − 1, r) in the following way.
We establish a recurrence relation for B(n, r). Let n ≥ 2 and 2 ≤ r ≤ n. There are two cases
for a desired selection of r pairs of girls and boys.

Case 1: Girl n is dancing.

The remaining r − 1 girls can choose their partners in B(n − 1, r − 1) ways while girl n
can choose her partner from any of the unchosen 2n − r boys.
This contributes (2n − r)B(n − 1, r − 1) to the value of B(n, r).

Case 2: Girl n is not dancing.

This case is easy because there are simply B(n − 1, r) possible choices.
240 13 Combinatorics

So for every n ≥ 2 and 2 ≤ r ≤ n, we have the recursion

B(n, r) = (2n − r)B(n − 1, r − 1) + B(n − 1, r),

with initial conditions

B(n, r) = 0, if r > n
B(n, 1) = 1 + 3 + 5 + · · · + (2n − 1) = n2 .

It is directly verified that the numbers A(n, r) satisfy the same initial conditions and recurrence
relations, from which it follows that A(n, r) = B(n, r) for all n and r ≤ n. We leave it to the
reader to do the algebra that verifies this.

13.11 Double counting via tables

Often a problem can be interpreted in terms of a table. The information is somehow recorded
in the cells of the table. This gives a really clear way of understanding what is happening.
Then by examining the situation in the table horizontally on the one hand and vertically on
the other hand, we have a natural way to do some double counting.

Problem In a competition, there are a contestants and b judges, where b ≥ 3 is an odd

integer. Each judge rates each contestant as either pass or fail. Suppose k is a number such
that, for any two judges, their ratings coincide for at most k contestants.
Prove that
k b−1
≥ .
a 2b

Solution It is advantageous to organise the information of this problem in a table. We do

so by having one row for each contestant and one column for each judge. We place a 1 or a 0
in a cell if the judge corresponding to that column has given the contestant corresponding to
that row a pass or a fail, respectively.
The tactic now is to find something in the table which we can double count. What we
will count is the number of matches between two judges’ ratings. More precisely, we are
counting the number of instances where two different judges Jx and Jy agree on the result of
contestant C. Let the number of these instances be N .

J1 J2 J3 J4 J5 J6 J7
C1 0 1 0 0 0 0 1
C2 1 0 1 0 1 0 1
C3 1 1 1 0 0 0 0
C4 1 1 0 0 1 1 0
C5 0 1 0 1 0 1 1

Think vertically.
b


For each pair of judges, we know that they agree in at most k places. Since there are 2
pairs of judges, we conclude that
 
b
N ≤k .
2
13.12 Combinatorial reciprocal principle 241

Think horizontally.
Suppose a contestant has received m passes and n
fails, where m + n = b. The m judges
m
who awarded the contestant a pass contribute 2 agreements, while the n judges who
n


awarded the contestant a fail contribute 2 agreements.
Therefore, looking along a row, the number of agreements is
   
m n m(m − 1) n(n − 1)
+ = +
2 2 2 2
m 2 + n2 b
= − .
2 2

Since m + n = b is constant, m2 + n2 is minimised when m and n are as close to each

other as possible. Since b is odd this occurs when {m, n} = { b−1 b+1
2 , 2 }. So we have
" 2  2 #
m 2 + n2 b 1 b−1 b+1 b
− ≥ + −
2 2 2 2 2 2
 2
b−1
= .
2

Summing over the number of rows, we obtain

 2
b−1
N ≥a .
2

All that remains is to put these two pieces of information together. What we have shown is
that
 2  
b−1 b
a ≤N ≤k .
2 2

It follows that
 2 
k b−1 b
≥
a 4 2
b−1
= .
2b

13.12 Combinatorial reciprocal principle

The what principle, I hear you say? Yes, this is another principle to which we gave a name,
because it seemed really interesting at the time, and because nobody else seems to have
named it. Here’s the principle: if you have a set S of objects falling into k different categories,
then
X 1
= k.
number of objects in the same category as x, including x
x∈S

To help you understand what this is saying, write down some examples and then try to prove
the principle. Once it is well understood, this principle can solve some otherwise difficult, or
even apparently unapproachable, problems.
242 13 Combinatorics

Problem Students from 13 different countries participated in the 491st International Math-
ematics Bonanza. Each student belonged to one of five different age groups.
Prove that there were at least nine participants in the Bonanza who had more fellow
participants in his or her age group than fellow participants from his or her own country.

Solution Given a student x, let Ax denote the number of students (including x) in the same
age group as x, and let Cx denote the number of students from the same country. So by the
combinatorial reciprocal principle,
X 1 X 1
=5 and = 13.
x
Ax x
Cx

Thus X 1 1
− = 8.
x
Cx Ax

Since Ax and Cx are positive integers, we have

1 1
− <1
Cx Ax
for each x.
So the number of x for which C1x − A1x > 0 is more than 8. That is, there are at least 9
students for which C1x > A1x , that is, Ax > Cx . Thus we are done.
 Graph theory
14
Graph theory is a truly remarkable area of mathematics. In one sense, the concept of a
graph is so easy that every human being has some innate understanding of it. On the other
hand, graph theory offers some notoriously difficult problems and is the site of much active
mathematical research today. In between the very simple and the very complicated lies a
vast region of interesting and amazing results, some of which we’ll soon see.
When we use the word graph in this chapter, we don’t mean a complicated representation
of a function on a set of coordinate axes or anything like that. We simply mean a diagram
obtained by joining dots.
In the usual graph theory terminology, the dots are called vertices while the lines connecting
them are called edges. The edges of a graph merely represent relationships between the
vertices. In particular, we usually don’t care how a graph is drawn in the plane. In fact, we
consider two graphs G and H to be the same—the technical word is isomorphic—if there is a
one-to-one correspondence between their vertices, such that two vertices are connected by an
edge in G if and only if the two corresponding vertices are connected by an edge in H. Note
that a graph is allowed to consist of more than one connected component.

1 4

2 5
3

2 5
3 4
1
graph with graph with
two components multiple edges a pair of isomorphic graphs

For most of our purposes, we will assume that a graph is finite and has no loops (a loop is an
edge connecting a vertex with itself) or multiple edges (two or more edges connecting the
same pair of vertices). However, there are situations where considering loops and multiple
edges is useful.
244 14 Graph theory

14.0 Problems
1. Remembering that we view two graphs as being the same if they are isomorphic, how
many graphs are there with at most five vertices? How many of these are planar?1
2. At a party with 100 people, in any set of four there is at least one person who is mutually
acquainted with the other three.
Given that there are three people who are mutually unacquainted with each other,
prove that the remaining 97 people must know everyone at the party.
3. In a country with n towns, there are some one-way roads connecting pairs of towns.
It’s known that for any two towns you can drive from one of them to the other one.
Prove that there is a town from which it’s possible to drive to any other town.
4. Show that if every vertex in a graph has degree k, then the graph contains a path of
length k. (See section 14.7 for the definition of a path.)
5. Consider a 3 × 3 × 3 cube made from 27 unit cubes. These smaller cubes are rooms
which have doors on each of their faces.
Is it possible to start in the central room and visit every other room exactly once
without ever leaving the large cube?
6. Consider a tennis tournament in which each of n people played against every other
person exactly once.
(a) Show that it is possible to label the players P1 , P2 , . . . , Pn in such a way that P1
beat P2 , P2 beat P3 , and so on to Pn−1 beat Pn .
(b) Suppose it is known that there is no loop of players Q1 , Q2 , . . . , Qk such that Q1
beat Q2 , Q2 beat Q3 , and so on up Qk−1 beat Qk and Qk beat Q1 , for some k ≥ 3.
Prove that there is a unique way to rank the players so that each player beat
everyone below them.
7. A snozzberry has the shape of a convex polyhedron such that three faces meet at every
vertex, and each face is either a pentagon or a hexagon. Each pentagon is surrounded by
five hexagons while each hexagon is surrounded by three hexagons and three pentagons.
How many faces does a snozzberry have?
8. My wife and I were invited to a dinner party attended by four other couples, making
a total of 10 people. A certain amount of handshaking took place subject to two
conditions: no one shook his or her own hand and no couple shook hands with each
other. Afterwards, I became curious and asked everybody else at the party how many
people they shook hands with.
Given that I received nine different answers, how many hands did I shake?
9. An n-domino consists of two squares which share an edge, where each square is labelled
by a number from 1 up to n.
(a) Prove that there are n+1


2 dominoes.
(b) For which values of n is it possible to arrange the dominoes in a line so that
adjacent halves of neighbouring dominoes have the same label?
10. Prove that a graph is bipartite if and only if it has no cycles of odd length.
1 Some of the problems in this section use terminology that is defined later in the chapter.
14.0 Problems 245

11. If G is a graph with V vertices and E edges, prove that the following statements are
equivalent to each other.

(i) The graph G is a tree.

(ii) The graph G is connected, and the removal of any edge leaves G disconnected.
(iii) Any two vertices of G are connected by exactly one path.
(iv) There are no cycles in G, but the addition of any edge creates a cycle.

12. A convex polyhedron with an even number of edges is given.

Prove that an arrow can be placed on each edge so that each vertex is pointed to by an
even number of arrows.

13. Prove that the complete bipartite graph K3,3 is not planar.

14. Seventeen people correspond by mail with one another, each one with all the rest. In
their letters only three different topics are discussed and each pair of people deals with
only one of these topics.
Prove that there are at least three people who write to each other about the same topic.

15. (a) During a meeting of 2n people, more than n2 handshakes took place, with no pair
of people shaking hands more than once.
(i) Prove that there must have been three people who all shook hands with each
other.
(ii) Is the problem still true if exactly n2 handshakes took place?
(b) What is the maximal number of edges that a graph on n vertices can have such
that the graph contains no triangle?

16. (a) Prove that, at any party with nine people, there must exist four mutual friends or
three mutual strangers.
(b) Show that this is not true for a party with eight people.

17. A Platonic solid is a convex polyhedron in which each vertex is surrounded by the same
number of congruent regular polygons.
Prove that there are exactly five Platonic solids and determine the number of vertices,
edges and faces of each.

18. Consider a planar graph whose faces, including the infinitely large one, are all triangles.
If each vertex is coloured red, green or blue, prove that the number of faces whose
vertices are all different colours is even.

19. There are 1985 participants at an international meeting. In each set of three participants,
there are at least two who speak the same language.
Given that no one speaks more than five languages, prove that there are at least 200
participants who speak the same language.

20. Each edge of the complete graph K9 is coloured either blue, or red, or left uncoloured.
Find the smallest value of n such that whenever n edges are coloured, there necessarily
exists a monochromatic triangle.
246 14 Graph theory

21. A prism with pentagons A1 A2 A3 A4 A5 and B1 B2 B3 B4 B5 as top and bottom faces is

given. Each of the sides of the pentagons and each of the line segments Ai Bj , for all
i, j = 1, 2, 3, 4, 5, is coloured either red or green. It is also known that each triangle
whose vertices are vertices of the prism and whose sides have all been coloured, has two
sides of a different colour.
Show that all 10 sides of the top and bottom faces are the same colour.
22. In a certain country there are n towns, where n ≥ 4. Initially there are no roads
connecting the towns. A road may be built between towns A and B if there exist two
other towns X and Y such that there is no road between towns A and X, there is no
road between towns X and Y and there is no road between towns Y and B.
What is the maximum number of roads that can be built?
23. At a school some students are friends. A friendship is a set of two students who are
friends with each other. A trio is a set of three students who are all friends with each
other.
Prove that
2 3
t2 ≤ f ,
9
where f is the number of friendships and t is the number of trios.
24. Among a group of 120 people, some pairs are friends. A weak quartet is a set of four
people containing exactly one pair of friends.
What is the maximum possible number of weak quartets?
14.1 Degree 247

14.1 Degree
Let’s start with a few fundamental definitions in graph theory. If one of the endpoints of the
edge e is the vertex v, then we say that e and v are incident. If two vertices u and v are
incident to the same edge, then we say that u and v are adjacent. Now define the degree of a
vertex v to be the number of edges incident to v and denote it by deg(v).

Problem Show that at any party, there are always at least two people with exactly the
same number of friends at the party.

This is the first of many party problems which we will examine. In fact, you can think of
every graph theory problem as a party problem in disguise. Vertices represent people, while
edges represent mutual acquaintance. For the time being, we won’t deal with the case of
celebrities or forgetful people, that is, where A knows B but B doesn’t know A. Furthermore,
we don’t count people as knowing themselves, so that the graph has no loops, and we don’t
allow people to know each other twice over, so that the graph has no multiple edges. The
degree of a vertex represents the number of people at the party that a person knows so, in
some sense, degree is a popularity index!
Throughout the chapter, we will switch between party language and graph language, depending
on the circumstances. This particular problem can be translated into graph theory terminology
in the following way.

Show that in any graph, there exist two vertices with the same degree.

Solution With the goal of obtaining a contradiction, suppose that each person knows a
different number of people at the party.
If there are n partygoers, then they can know 0, 1, 2, . . . , n − 1 people, and these are the only
possibilities. Since there are n people and n possibilities for the number of people they know,
there must be one person who knows 0 people, one person who knows 1 person, and so on,
up to one person who knows all n − 1 other people at the party. However, it’s impossible for
there to be two people at the party, one who knows no one else and one who knows everybody
else. This is the desired contradiction, so we can conclude that there exist two people who
know the same number of people at the party.

At any party, if you ask everyone in the room, including yourself, how many hands they shook,
and add up all of the answers, then you will always end up with an even number. In fact, the
sum will be twice the number of handshakes that have occurred during the party. We can
state this in the language of graph theory in the following way.

Handshaking lemma In any graph, the sum of the degrees of all the vertices is equal to
twice the number of edges.

This is true because each edge contributes two to the sum of the degrees, one for each vertex
incident to it. A simple corollary of the handshaking lemma is the fact that the number of
vertices of odd degree in any graph must be even.

Problem Is it possible to build a house with exactly eight rooms, each with three doors,
and such that exactly three of the house’s doors lead outside?

Solution If you could build such a house, then you could construct a corresponding graph
in the following way. Let there be nine vertices, one for each room and one to represent the
248 14 Graph theory

outside of the house. Place an edge between two vertices if there is a door between the two
corresponding areas. The conditions of the problem assert that every vertex of our graph has
degree three.
Since there are nine vertices, the sum of the degrees is 9 × 3 = 27. But the handshaking
lemma tells us that there should be 13 12 edges in our graph, which is clearly impossible!

14.2 Directed graphs

There are many graph theory problems which involve tournaments, one-way roads and the
like. These are easily represented by a directed graph. This is a graph in which every edge has
a direction, usually indicated by an arrow. For instance, in the case of a tournament, players
can be represented by vertices and matches can be represented by directed edges, where the
direction of an edge points from the winner to the loser.
Each vertex of a directed graph has both an indegree, which counts the number of incoming
edges, as well as an outdegree, which counts the number of outgoing edges. We will denote
the indegree and outdegree of a vertex v by indeg(v) and outdeg(v), respectively. Sometimes,
we may want to refer to the total degree of a vertex, which is simply indeg(v) + outdeg(v).

Problem Consider a squash tournament in which each of n people plays against every other
person exactly once. Let Lk and Wk be the number of losses and wins of the kth player,
respectively.
Prove that
L21 + L22 + · · · + L2n = W12 + W22 + · · · + Wn2 .

Solution We rearrange the given equation and seek to prove that

(L1 + W1 )(L1 − W1 ) + (L2 + W2 )(L2 − W2 ) + · · · + (Ln + Wn )(Ln − Wn ) = 0.

In the directed graph which represents this tournament, the total degree for the kth player is

indeg(v) + outdeg(v) = Lk + Wk = n − 1.

Since this is a constant, we can divide the previous equation through by n − 1 to obtain

(L1 − W1 ) + (L2 − W2 ) + · · · + (Ln − Wn ) = 0.

So our aim is to prove that

L1 + L2 + · · · + Ln = W1 + W2 + · · · + Wn .

But this is now quite simple, since the left-hand side represents the sum of the indegrees,
which is equal to the number of edges in the graph. Similarly, the right-hand side represents
the sum of the outdegrees, which is also equal to the number of edges in the graph.

14.3 Connected graphs, cycles and trees

There’s quite a bit of terminology in graph theory, but thankfully, the vast majority of it
seems to make perfect sense.
14.3 Connected graphs, cycles and trees 249

A connected graph is one in which it’s possible to walk between any two vertices along
the edges. In other words, the entire graph consists of only one piece.
A cycle is a sequence of distinct vertices v1 , v2 , . . . , vn with v1 adjacent to v2 , v2 adjacent
to v3 , and so on, with vn adjacent to v1 .
A tree is a connected graph which has no cycles.

Problem If G is a graph with V vertices and E edges, prove that G is a tree if and only if
it is connected and E = V − 1.

Solution First, we will prove that if G is a tree, then G is connected and E = V − 1. Of

course, the fact that G is connected follows immediately from the definition of a tree, so it’s
only necessary to prove that E = V − 1.
We will proceed by induction on the number of vertices.
The base case is rather trivial, since you can see for yourself that the statement is true if G
has 1 or 2 vertices.
Suppose now that the statement holds for all trees with n vertices and consider an arbitrary
tree with n + 1 vertices. The main idea is to prove that there exists a vertex with degree 1.
Start at a random vertex and start walking along a path which never visits the same vertex
twice. Sooner or later, you must get stuck in one of two ways: either you return to a vertex
you already visited, or you reach a dead end. The former case implies that G contains a cycle,
which is a contradiction. The latter case gives us the vertex of degree 1 that we are looking
for. Now we simply remove this vertex and the single edge incident to it. This leaves us with
a tree with n vertices and, by the inductive hypothesis, n − 1 edges. Hence, our original tree
must have n + 1 vertices and n edges and so satisfies E = V − 1.
Now we aim to prove the converse, that if G is connected and E = V − 1, then G is a tree.
Of course, the fact that G is connected is immediate, so it’s only necessary to prove that G
has no cycles.
Again, we will proceed by induction on the number of vertices.
The base case is rather trivial, since you can see for yourself that the statement is true if G
has 1 or 2 vertices.
Suppose now that the statement holds for all G with n vertices and consider an arbitrary
graph with n + 1 vertices which is connected and satisfies E = V − 1. Once again, the main
idea is to prove that there exists a vertex with degree 1.
From the handshaking lemma, we can deduce that if every vertex in the graph has degree
at least 2, then V ≤ E, which is a contradiction. Furthermore, since G is connected, there
are no vertices with degree 0. This gives us the vertex of degree 1 that we are looking for.
Now we simply remove this vertex and the single edge incident to it. This leaves us with a
graph which is connected with n vertices and n − 1 edges. By the inductive hypothesis, such
a graph has no cycles. Furthermore, adding in a new edge incident to a vertex with degree 1
cannot create any cycles. In other words, G is a tree.

The solution to the previous problem should give you some ideas which you can use to prove
the following important results.

Theorem Consider a graph with V vertices and E edges.

If E ≤ V − 2, then the graph is not connected.

If E ≥ V , then the graph contains a cycle.
250 14 Graph theory

14.4 Complete graphs and bipartite graphs

There are certain types of graph which seem to pop up in all sorts of problems.

The complete graph Kn is the graph consisting of n vertices with an edge between every
pair of vertices.
A bipartite graph is one whose vertices can be coloured black and white such that every
edge is incident to one black vertex and one white vertex.
The complete bipartite graph Km,n is the graph consisting of m black vertices and n
white vertices with an edge between each black and each white vertex.

The best way to remember jargon like this is to put it to good use!

Problem In the parliament of Frenemia each member is friends with exactly one other
member and enemies with exactly one other member.
Prove that the members of parliament can be divided into two chambers so that no chamber
contains a pair of mutual friends or a pair of mutual enemies.

Solution As is often the case, our first step will be to phrase this in graph theory language.

In a graph, the edges are coloured red and blue in such a way that each vertex is
incident to one red edge and one blue edge. Prove that the graph is bipartite.

We will give a procedure for colouring the vertices black and white such that every edge is
incident to one black vertex and one white vertex.
If you try drawing some graphs which satisfy the conditions of the problem, then you’ll find
that they all consist of a bunch of cycles, each one with an even number of vertices.
With this in the back of our minds, let’s start by taking any vertex and colouring it black.
Now take another vertex adjacent to it and colour it white. And take another vertex adjacent
to this one and colour it black. Continue walking around the graph, alternately colouring
vertices black and white until you get stuck. Clearly, this can only happen in one of two
ways: either you reach a dead end, or you meet a vertex which has already been coloured.
The former case actually never arises, since every vertex has degree two while a dead end is a
vertex of degree one. In the second case—and you should carefully think about why this is
so—we must return to the vertex at which we started.
In summary, we have traversed a cycle, alternately colouring vertices black and white. Now a
problem arises if the cycle has an odd number of vertices. But we’ve been told that the edges
alternate between red and blue, so there must be an even number of vertices along the cycle.
At this stage, we have successfully managed to colour some of the vertices black and white.
If we have coloured every vertex, then we are done, but if not, then we simply repeat the
process, starting at an uncoloured vertex.
This recipe will eventually colour every vertex in such a way that each edge is incident to one
black vertex and one white vertex, so the graph is certainly bipartite.

14.5 Pigeonhole principle

Let’s return to the graph theory party scene! We will assume that at a graph theory party,
every pair of people either know each other, in which case we call them friends, or they do
14.6 Euler trails 251

not know each other, in which case we call them strangers. We will use vertices to represent
people, red edges to represent friends and blue edges to represent strangers. Therefore, every
party is simply a complete graph with each edge coloured red or blue.

Problem Prove that, at any party with six people, there must exist three mutual friends or
three mutual strangers.

Solution In graph theory terminology, this problem translates to the following.

Prove that given a complete graph on six vertices with each edge coloured red or
blue, there exists a monochromatic2 triangle.

Consider a random partygoer A. Of the five edges incident to A, the pigeonhole principle
guarantees that at least three of them are the same colour. Without loss of generality, let
these edges be red and let them join A to the party people B, C and D. If BC is red, then
triangle ABC is red. If CD is red, then triangle ACD is red. If DB is red, then triangle ADB
is red. So to avoid a red triangle, the edges BC, CD and DB must all be blue, which forces
triangle BCD to be blue. So we simply cannot avoid having a monochromatic triangle.

14.6 Euler trails

In the eighteenth century, the inhabitants of Königsberg, now known as Kaliningrad, liked to
walk along the Pregel River. They would use the city’s seven bridges to cross over to one
of the two islands in the river or to the other bank of the river. And they often wondered
whether or not it was possible to design a walking tour which crossed each bridge exactly
once. Here is a map of the situation.

The people tried in vain for many years until the great mathematician Leonhard Euler proved
that it was impossible. His first move was to reduce the map to a graph, with each vertex
representing a land mass and each edge a bridge as in the following diagram.
2 The word monochromatic simply means one-coloured.
252 14 Graph theory

• •

In this section only, we will allow a graph to have loops and multiple edges. So what we
would like to know is whether it’s possible to walk around the graph, traversing every edge
exactly once. Such a walk is known as an Euler trail.

Problem Prove that the Königsberg graph has no Euler trail.

Solution Let’s suppose that the Königsberg graph has an Euler trail and hope for a
contradiction.
We’ll start with the obvious fact that it must start somewhere and end somewhere. Between
the start and end, each time we visit a vertex, we must have walked along two edges incident
to it, one going in and one coming out. Thus every vertex other than the start and end must
have even degree. So if there exists an Euler trail, then there can be at most two vertices of
odd degree. But you can see for yourself that the graph has four vertices of odd degree, a
contradiction.

This proof shows that, if a graph has an Euler trail, then it must have at most two vertices
of odd degree. However, by the handshaking lemma, we know that any graph has an even
number of vertices of odd degree. So for a graph to have an Euler trail, the number of vertices
of odd degree must be either 0 or 2. It turns out that the converse of this statement is true
as long as the graph is connected.

Problem Prove that a graph has an Euler trail if and only if it is connected and has 0 or 2
vertices of odd degree.

Nailing down the details of this proof requires quite a bit of work. We will only present the
main ideas of the proof and leave it up to you, dear reader, to sort out all of the details!

Solution Suppose that a connected graph has 0 or 2 vertices of odd degree. If there are 2
vertices of odd degree, call one of them S and one of them F . If there are 0 vertices of odd
degree, pick any vertex and call it both S and F . We will prove that there exists an Euler
trail which starts at S and finishes at F .
We will proceed by strong induction on the number of edges.
For a connected graph with one edge, it’s easy to find an Euler trail!
Now assume that we have a graph with E edges and that the result is true for any graph
with fewer than E edges. The idea is to start at S and commence walking randomly, never
traversing the same edge twice. Since there are only finitely many edges, sooner or later you
will get stuck somewhere. It’s impossible to get stuck at a vertex with even degree, because
each time we can walk in, there will always be an edge along which we can walk out. In fact,
it turns out (and you should think about why) that the only place we can get stuck is at F .
If our walk so far has traversed every edge of the graph, then we are done! Otherwise, we’ve
missed some of the edges and these remaining edges must form a number of smaller connected
14.7 Paths 253

graphs G1 , G2 , . . . , Gn . Each of these components Gk must have all its vertices of even degree
(think carefully about why this must be so). And each such Gk intersects our previously
constructed walk at some vertex vk . Furthermore, each Gk has fewer than E edges, so the
inductive hypothesis guarantees that there is an Euler trail for each Gk which starts and
finishes at vk .
Now we modify our original walk as follows. Start the same way but for each k whenever you
reach one of the vertices vk , take a detour along the Euler trail for Gk . Our new walk now
traverses all the edges from the original walk, along with all the edges from the Gk . So it’s
an Euler trail for the original graph.

That was quite a difficult argument, so take some time to think about it carefully. The idea of
the random walk which cannot be stopped turns out to be a very useful one in graph theory.

14.7 Paths
A walk is any sequence of vertices v1 , v2 , . . . , vn with v1 adjacent to v2 , v2 adjacent to v3 , and
so on, with vn−1 adjacent to vn .
A path is any walk with distinct vertices.

Problem In Eulerland, there are 100 cities and two airlines, Air Gauss and Air Jordan.
For any two cities in Eulerland, exactly one of the companies provides direct flights in both
directions between them. It’s known that there are two cities a and b such that it is impossible
to travel from a to b using only Air Jordan flights.
Prove that it’s possible to travel between any two cities in Eulerland using only Air Gauss
flights.

Solution We can use vertices to represent cities and edges to represent flights, but how can
we represent the two airlines? Simple! We use red edges to represent Air Gauss and blue
edges to represent Air Jordan. It turns out that a whole variety of graph theory problems
involve colouring things in. In graph theory language, the problem translates to the following.

Suppose that each edge of the complete graph on 100 vertices is coloured red or
blue. It’s known that there are two vertices a and b such that there is no red path
from a to b.
Prove that there is a blue path between any two vertices.

Let A denote the set of all vertices which are connected to a by a red path, including a
itself. Similarly, let B denote the set of all vertices which are connected to b by a red path,
including b itself. And finally, let C denote the set of remaining vertices. (It is possible that
C is empty.) We have the following observations.

Every edge between a vertex in A and a vertex in B must be blue—for otherwise, there
would be a red path from a to b.

Every edge between a vertex in A and a vertex in C must be blue—for otherwise, that
vertex in C could be reached by a red path from a.

Every edge between a vertex in B and a vertex in C must be blue—for otherwise, that
vertex in C could be reached by a red path from b.
254 14 Graph theory

We will now prove that for two arbitrary vertices u and v, there exists a blue path between
them. We’ve already established that there exists a blue edge, and hence a blue path, between
any two vertices which are in different groups. So all that remains is to consider when u and
v lie in the same group.

If u and v lie in A, consider the blue path u → b → v.

If u and v lie in B, consider the blue path u → a → v.

If u and v lie in C, consider the blue path u → a → v.

Therefore, in all possible cases, there is a blue path between u and v.

14.8 Extremal principle

In chapter 1, we discussed the many advantages of using the extremal principle, that is,
considering the minimum or maximum of some value. The extremal principle is very useful
in graph theory as you can see for yourself in the following problems!

Problem In the country of König, it’s possible to travel by plane between any two of the
cities, although you might have to take several flights, stopping at other intermediate cities
along the way. By a journey, we mean a sequence of flights which never visits the same city
twice. Let m be the maximum possible number of flights on a journey between two cities in
König.
Prove that any two journeys of length m must have at least one city in common.

Solution Clearly, we can interpret cities as vertices, flights as edges, and journeys as paths.
If we define the length of a path to be the number of edges that it traverses, then the problem
can be restated as follows.

If the longest path in a connected graph has length m, prove that any two paths
of length m in the graph must share a vertex.

To obtain a contradiction, assume there exist two paths P1 and P2 of length m in the graph
that don’t share a vertex. Let P1 join vertex A to vertex B and let P2 join vertex C to vertex
D. The aim is to find a path with length greater than m, which will then contradict the fact
that m is the maximum length of a path.
How might we construct such a path? Well, since the graph is connected, there must be a
path from a vertex on P1 to a vertex on P2 . In fact, let X be a vertex on P1 and Y a vertex
on P2 such that there exists a path from X to Y of minimal length. Using such a path of
minimal length ensures that it is a path which cannot pass through any other vertices of P1
or P2 , because otherwise it would not be of minimal length.

A X B
• • • • • • • • •

• • • • • • • • •
D Y C
14.9 Count and count again 255

Note that either the path from A to X or the path from B to X has length at least m 2.
Without loss of generality, let it be from A to X. Similarly, either the path from C to Y or
the path from D to Y has length at least m 2 . Without loss of generality, let it be from C to
Y . Now the path from A to X to Y to C passes through no vertex twice, and has length
strictly greater than m. However, this contradicts the fact that m is the maximum length of
a path. Therefore, any two paths of length m in the graph must share a vertex.

Problem There are several boys and girls at a party, where each girl dances with at least
one boy but no boy dances with every girl.
Prove that there exist two boys B1 and B2 and two girls G1 and G2 such that B1 dances
with G1 , B2 dances with G2 , B1 does not dance with G2 and B2 does not dance with G1 .

It’s quite difficult to attack this problem directly. As we’ll soon see, the best approach is to
consider the boy who dances with the most girls. Like some other proofs in this book, this
may appear as surprising as a magician pulling a rabbit out of a hat! But the point, dear
reader, is for you to carefully read over the proof and ask how you could have thought of
it yourself. What features of the problem lead you to think about the extremal principle?
And why would you apply it in this particular way? If you understand the answers to these
questions, then you will soon be ready to start pulling rabbits out of your own hat!

Solution Of course, vertices will represent people while edges represent couples who dance
together. It’s useful to use black vertices for the boys and schematically place them at the
top of the diagram and to use white vertices for the girls and schematically place them at the
bottom of the diagram.

We’ll start by considering one of the boys, Max say, who dances with the maximum number
of girls. Now there is a girl, Anne say, who doesn’t dance with Max. And Anne must dance
with some boy, Bob say, who isn’t Max. So if one of Max’s dance partners, of whom there
are maximally many, does not dance with Bob, then we’re done!
But if Bob danced with all of Max’s dance partners, as well as with Anne, then Bob would
have danced with more girls than Max. This contradiction means that there must be some
girl, Carly say, who dances with Max but not Bob.
So we can take B1 , B2 , G1 and G2 to be Max, Bob, Carly and Anne, respectively.

14.9 Count and count again

The handshaking lemma asserts that, in any graph, the sum of the degrees of all the vertices
is equal to twice the number of edges. One way to prove this is to consider the number of
endpoints of edges. Counting these in one way, we obtain the sum of the degrees of all the
vertices and counting these in another way, we obtain twice the number of edges.
256 14 Graph theory

We have already witnessed the virtues of double counting in section 13.8 and here, we’ll see
that it can be particularly useful when dealing with graphs.

Problem In a senate, there are 30 senators and each pair of them are either mutual friends
or mutual enemies. Each senator has exactly 6 enemies and every group of 3 senators forms a
commission.
Find the total number of commissions whose members are either all mutual friends or all
mutual enemies.

Solution This problem can be paraphrased purely in graph theory terminology.

Each edge of the complete graph K30 is coloured either red or blue. Each vertex
is incident to exactly 6 blue edges.
Find the total number of monochromatic triangles.

Let X be the number of monochromatic triangles. We define a vee to be a pair of edges

which are incident to the same vertex. The trick here is to count colourful vees—that is, those
consisting of one red edge and one blue edge—in two different ways.

If we concentrate on one particular vertex, we see that it’s incident to 6 blue edges and
23 red edges. This means that there are 6 × 23 = 138 colourful vees formed from the
edges incident to one particular vertex.
Since there are 30 vertices altogether, the total number of colourful vees is
30 × 138 = 4140.
30


Note that there are 3 = 4060 triangles in our graph. Since X of these are monochro-
matic, we know that 4060 − X of these are not. But in a monochromatic triangle, there
are no colourful vees, while in a triangle which is not monochromatic, there are two
colourful vees.
Therefore, the total number of colourful vees is
2(4060 − X) = 8120 − 2X.

Now we simply need to equate these two results and we end up with
4140 = 8120 − 2X
⇒ X = 1990.

14.10 Planar graphs

When trying to solve graph theory problems, obviously you draw graphs. Unfortunately, the
edges sometimes cross and that just looks plain ugly. It would be nice to draw our graphs
without any edges crossing and, when this is possible, we call the graph planar.
A planar graph drawn without edges crossing will always divide the plane into regions, one of
which is infinitely large, which we call faces. One of the most useful theorems concerning
planar graphs is the following.

Euler’s formula For a connected planar graph with V vertices, E edges and F faces,
V − E + F = 2. When using Euler’s formula, we always include the infinitely large face on
the outside.
14.11 Polyhedra 257

Problem Suppose a planar graph with E edges divides the plane into F faces.
Prove that 3F ≤ 2E.

Solution Let’s count the number E of edges. Imagine cutting the planar graph along its
edges. We get a collection of polygons and one figure which might be called an ‘anti-polygon’
corresponding to the outside face. Since each original edge splits into two edges, the total
number of newly formed edges is 2E. However, each of our F newly formed polygons, including
the outside anti-polygon, has at least three edges. Thus the number of newly formed edges is
at least 3F . It follows that 3F ≤ 2E.

Problem Prove that K5 , the complete graph on five vertices, is not planar.

Solution To obtain a contradiction, let’s suppose that K5 is planar.

Given that V = 5 and E = 10, Euler’s formula tells us that if we could draw K5 in the plane
without edges crossing, then we would have F = 7. But from the previous problem, 2E ≥ 3F
and so 20 ≥ 21. Contradictions don’t come more blatantly than that!

Problem If a graph is planar, prove that it has at least one vertex of degree less than or
equal to five.

Solution In the hope of finding a contradiction, let’s start with the assumption that there
exists a planar graph, all of whose vertices have degree at least six.
Suppose that this graph has V vertices, E edges and divides the plane into F faces. Then
the sum of the degrees is at least 6V , so the handshaking lemma asserts that 6V ≤ 2E or
equivalently,
E
V ≤ .
3
2E
From the previous problem, we know that F ≤ 3 .
Substituting these two inequalities into Euler’s formula, we obtain

E 2E
2=V −E+F ≤ −E+ = 0,
3 3
an obvious contradiction.

14.11 Polyhedra
If you take a bunch of polygons and glue them together so that no side is left unglued, then
the resulting object is usually called a polyhedron. Typical examples include the tetrahedron,
the square pyramid, the cube and the soccer ball with its pentagonal and hexagonal patches.
The corners of the polygons are called vertices, the sides of the polygons are called edges and
the polygons themselves are called faces. We say that a polyhedron is convex if, for each
plane which lies along a face, the polyhedron lies on one side of that plane. So, for example,
the cube is convex while the polyhedron formed by gluing three congruent cubes together to
form an L shape is not.
The fact that every polyhedron is a graph is a rather simple statement. This is because if
you have a polyhedron and simply ignore the faces, then what you have left over is just a
bunch of vertices connected in pairs by edges, or in other words, a graph.
258 14 Graph theory

A more interesting statement is the following fact. Every convex polyhedron corresponds to a
planar graph. So why is this true? Well, suppose that your polyhedron is made from some
sort of rubbery material, like a balloon. If you pop the balloon by removing one of the faces,
then what remains is a rubbery sheet with the vertices and edges still drawn on it. Now
just stretch this out flat onto a table and there you have your planar graph. Note that this
planar graph has the same number of vertices, edges and faces as the original polyhedron.
The face that we removed from the polyhedron now corresponds to the infinitely large face of
the planar graph. In particular, Euler’s formula applies to all convex polyhedra.

Problem Suppose that you have a convex polyhedron and you are told that each face is a
quadrilateral or a hexagon and that three faces meet at every vertex. Furthermore, every
quadrilateral face shares an edge with four hexagonal faces, while every hexagonal face shares
an edge with three quadrilateral faces and three hexagonal faces.
Deduce the number of quadrilateral faces and the number of hexagonal faces of the polyhedron.

Solution Let V be the number of vertices, E the number of edges, Q the number of
quadrilateral faces, and H the number of hexagonal faces of the polyhedron. We will deduce
these four values by writing down four equations that they satisfy and solving them.

The first equation comes from applying the handshaking lemma to the polyhedron.
Since three faces meet at every vertex, every vertex of the polyhedron has degree three.
Therefore, the sum of the degrees is simply 3V and we obtain the equation

3V = 2E. (1)

The second equation comes from double counting the edges via face contributions. Each
quadrilateral contributes four edges and each hexagon contributes six edges. But each
edge has been double counted due to the fact that it is adjacent to two faces so we find
that 4Q + 6H = 2E, or equivalently,

E = 2Q + 3H. (2)

The third equation comes from a clever double counting argument. The trick here is to
count the number of times a quadrilateral face shares an edge with a hexagonal face.
This happens four times for each quadrilateral face. In other words, the number is 4Q.
Arguing in a different way, we can say that this happens three times for each hexagonal
face. In other words, the number is 3H. Of course, these numbers must be the same, so
we must have
4Q = 3H. (3)

The fourth equation is simply Euler’s formula. Since F = Q + H we may write

V − E + Q + H = 2. (4)

It is not hard to solve equations (1)–(4). For example, using (3) we have H = 43 Q. Put this
into (2) to find E = 6Q. Put this into (1) to find V = 4Q. Finally put everything into (4) to
obtain Q = 6. We may then go on to find H = 8, E = 36 and V = 24.

14.12 Graph theory and inequalities

This nice intersection of mathematical topics doesn’t appear too often, but is very interesting
indeed.
14.12 Graph theory and inequalities 259

Problem Given a graph with n vertices and E edges, prove that the number of triangles is
at least
4E 2 nE
− .
3n 3
Solution The basic idea of the solution is rather simple and elegant. Start by labelling the
vertices v1 , v2 , . . . , vn , and suppose that they have degrees d1 , d2 , . . . , dn , respectively.
Next take two vertices vi and vj which are connected by an edge. Then there are di − 1 edges
which emanate from vi to the rest of the graph and dj − 1 edges which emanate from vj to
the rest of the graph. Since there are n − 2 other vertices, then by the pigeonhole principle a
triangle will be created if
(di − 1) + (dj − 1) > n − 2.
In fact, we can guarantee that the edge between vi and vj will be involved in at least
(di − 1) + (dj − 1) − (n − 2) = di + dj − n
triangles.

• • • • •

• •
vi vj

If we sum over all possible edges, then we will count each triangle at most three times. So,
letting T denote the total number of triangles, we have the inequality
X X
3T ≥ (di + dj − n) = (di + dj ) − nE.
edges edges

2
Since we are aiming for the result 3T ≥ 4E n − nE, all that remains to be proved is the
inequality
X 4E 2
(di + dj ) ≥ .
n
edges

Note that in this sum over edges, the number di appears once for each edge that is incident
to the vertex vi . That is, di occurs in the sum di times. So we can write this inequality
equivalently as
n n
!2
2 2
X 4E (2E) 1 X
d2i ≥ = = di .
i=1
n n n i=1
Here, we’ve used the handshaking lemma to express 2E as the sum of the degrees.
We’ve finally arrived at a point where the problem is reduced to algebra. One way to finish
off the problem is to invoke the power means inequality3 , the QM–AM in particular.
From the power means inequality, we know
r
d21 + d22 + · · · + d2n d1 + d2 + · · · + dn
≥
n n
and this is exactly what we need to complete the proof.

3 If you’re unsure of what this means, then you might like to consult section 11.5.
 Games and invariants
15
Games? Who said games? We always seem ready to play a game. Why? Because games are
fun! Games often turn up in mathematics problems and if there is one important point you
should remember, it is this:
Play the game!
The reason is that you gain insight into what the problem is all about. In fact you might
need to play the game for quite a while before you notice something that just might be the
key element in solving the problem.

15.0 Problems
1. Amy and Ben play the game of misère noughts and crosses on a 3 × 3 square array. On
Amy’s turn, she can place an X in any vacant square, while on Ben’s turn, he can place
an O in any vacant square. The players take turns to place their symbol, with Amy
going first. Any player who gets three in a row (horizontally, vertically or diagonally)
immediately loses the game. The game is considered drawn if there is no winner after
all squares have been filled.
(a) Which player, if any, has a winning strategy?
(b) Answer the same question if the game is played on a three-dimensional 3 × 3 × 3
cubic array.
2. Initially, there are n coins on a table and two players take turns to remove a number of
them. The number of coins removed must belong to the set S and a player wins by
removing the last coin.
Find the winning and losing positions in the following cases.
(a) S = {1, 2, 3, 4, 5, 6}
(b) S = {1, 2, 3, . . . , k}
(c) S = {1, 3, 4}
(d) S = {1, 3, 8}
(e) S = {1, 2, 4, 8, 16, 32, 64, 128, . . .}
(f) S = {p | p is a prime or equal to 1}
262 15 Games and invariants

3. A 16 × 16 square grid is constructed from sixteen 4 × 4 smaller square grids called boxes.
The game of Ukodus is played on the 16 × 16 grid as follows. Two players write, in
turn, numbers from the set {1, 2, . . . , 16} in different squares. The numbers in each row,
column and box of the 16 × 16 grid must be different. The loser is the one who is not
able to write a number.
Which player has a winning strategy?

4. At the start of a game, the numbers 1 and 2 are each written 10 times on a blackboard.
Two players take turns to erase two of the numbers, replacing them with a 1 if they are
different and with a 2 if they are the same. The first player wins if the last number on
the board is 1, while the second player wins if it is 2.
Which player has a winning strategy?

5. Sixty-three squares on an 8 × 8 chessboard are tiled with 3 × 1 rectangles.

What are the possible locations for the square which remains untiled?

6. (a) Show that it’s possible to tile an m × n chessboard with 4 × 1 rectangles if and
only if 4 divides m or 4 divides n.
(b) Show that it’s possible to tile an m × n chessboard with k × 1 rectangles if and
only if k divides m or k divides n.

7. On the island of Trichroma, there are 10 blue, 15 red and 20 yellow chameleons. If
two chameleons of different colours meet, they both simultaneously change to the third
colour.
Is it possible for all of the chameleons to eventually be the same colour?

8. Lex plays a game on an 8 × 8 chessboard which contains an X in the bottom-left corner.

All other squares contain an O. Any three consecutive squares in a row or column is
called a playable set, if either

(i) it contains exactly two squares with an X, one of which must be the middle square
of the playable set, or
(ii) it contains exactly two squares with an O, one of which must be the middle square
of the playable set.

At any stage Lex is permitted to choose any playable set and simultaneously change
every X into an O, and every O into an X.
After a finite number of such changes, Lex has reached a configuration with exactly one
square containing an X.
Which squares could it be?

9. Two players start with the number 1 and take turns to multiply it by an integer from 2
to 9. The winner is the first player to obtain a number greater than or equal to 1000.
Which player has a winning strategy?

10. A coin is placed in each square of a 4 × 4 grid so that all are showing heads apart
from the coin in the first row and second column. You are allowed to flip all of the
coins along a row, along a column, or along a line parallel to one of the diagonals. In
particular, you are allowed to flip the coin in any corner square.
Prove that it’s impossible for all of the coins to show heads.
15.0 Problems 263

11. A positive integer is written in each square of an m × n chessboard. You are allowed to
add or subtract the same integer from any two squares which share a common side, as
long as the resulting numbers are both non-negative.
When is it possible to reduce all of the numbers to zero?

12. Is it possible to fill a 7 × 9 × 11 box with 3 × 3 × 1 boxes?

13. The numbers 2, 3 and 6 are written on a blackboard. You are allowed to replace two of
them, say a and b, with the numbers
3a 4b 4a 3b
+ and − .
5 5 5 5

Prove that it is impossible for a number greater than 7 to appear on the blackboard.

14. A class of students is lined up in order of height with the tallest person at the front. At
any stage a student who is not in the first or second position is permitted to move two
places towards the front.
For what size class is it possible that the students could end up in the reverse order of
height?

15. In the game of Fifteen we have 15 unit squares, numbered from 1 to 15. They are
arranged in order in a 4 × 4 square array leaving a space of one unit square in the
bottom-right corner. Any square adjacent to the space is free to slide into the space.

(a) By using these sliding moves is it possible to swap the positions of the squares
labelled 14 and 15, as shown in the diagram below?
(b) How many different positions are achievable in this game?

1 2 3 4 1 2 3 4

5 6 7 8 ? 5 6 7 8
−→
9 10 11 12 9 10 11 12

13 14 15 13 15 14

16. A real number is written in each square of an m × n grid. If the sum of the numbers in
a row or column is negative, then you are allowed to switch the signs of the numbers in
that row or column.
Prove that the sum of the numbers in each row and column will eventually be non-
negative no matter in what order the rows and columns are chosen.

17. Fifty coins of various denominations lie in a row. Ollie picks up a coin from one end of
the row, then Ellie picks up a coin from one end of the row of remaining coins. They
alternate in this way until they each have 25 coins.
Prove that Ollie can guarantee to win at least as much money as Ellie.

18. A deck of n cards labelled 1, 2, . . . , n is shuffled. If the top card is labelled k, then the
order of the top k cards is reversed.
Prove that after finitely many of these shuffles, the card labelled 1 will eventually be on
top of the deck.
264 15 Games and invariants

19. Two players play a game involving a knight on an 8 × 8 chessboard. The first player
places the knight on the board and the second player makes a knight’s move.1 The
two players then take turns to make a knight’s move, but may not place the piece on a
square it has already visited.
If the player who is unable to move is considered the loser, which player has a winning
strategy?
20. A game consists of circular discs numbered from 1 to 20 that are allowed to slide freely
along a track, which is a closed loop. Furthermore, there is a section of the track on
which exactly four discs can fit, which can be swivelled around its centre so that the
order of the four discs is reversed.
(a) From any given initial configuration, is it possible to move the discs into a configu-
ration where all 20 numbers are in order clockwise around the track?
(b) How many different positions are achievable in this game?
(c) Answer the same questions if there are 21 discs instead.

21. There are three amoebas sitting at the points (0, 0), (0, 1) and (1, 0) of the coordinate
plane. Every now and then an amoeba splits into two separate amoebas, one of which
will move one unit upwards, while the other moves one unit to the right. They do this
in such a way that no two amoebas ever sit at the same point.
Is it possible for the amoebas to split in such a way that the points (0, 0), (0, 1) and
(1, 0) will eventually be unoccupied?
22. The game of Domination is played on an m × n chessboard. Two players take turns to
place a domino so that it covers two adjacent squares of the chessboard. Dominoes are
not allowed to overlap and the player who cannot move loses.
(a) Which player has a winning strategy if Domination is played on a 3 × 3 board?
(b) Which player has a winning strategy if Domination is played on an m × n board,
where m and n are even?
(c) Which player has a winning strategy if Domination is played on an m × n board,
where m is odd and n is even?
23. Two people play a game involving n coins on a table. The first player takes at least
one, but not all, of the coins. The players then take turns to take at least one coin, but
no more than was taken on the previous move. The player who takes the last coin is
considered the winner.
For which values of n does the second player have a winning strategy?
1Achess knight moves either two squares horizontally and one square vertically or two squares vertically and
one square horizontally.
15.0 Problems 265

24. Chomp is played with an m × n grid of chocolate. Two players take turns to eat a
square of chocolate, along with every square which is above and to the right of it.
Unfortunately the bottom-left square is poisonous, so that the player who is forced to
eat it is considered the loser.

(a) Determine a winning strategy for the first player on a 2 × n grid.

(b) Determine a winning strategy for the first player on an n × n grid.
(c) Show that there exists a winning strategy for the first player on any size grid apart
from 1 × 1.

25. Initially, the number 2 is written on a blackboard. Two players take turns to erase the
number N from the blackboard and replace it with the number N + d, where d is one
of the divisors of N satisfying 0 < d < N .

(a) If the player who first writes a number greater than 12345 loses the game, which
player has a winning strategy?
(b) If the player who first writes a number greater than 123456 loses the game, which
player has a winning strategy?

26. Write the numbers 1, 4, 5, 8, 6, 2, 3, 9 and 7 in that order, around a circle. A move
consists of changing three consecutive numbers (a, b, c) to (a + 1, b − 2, c + 1).
Can you change all of the numbers to 5 using such moves?

27. Ten girls sitting around a circular table are playing a game with N cards. Initially, one
girl is holding all of the cards. Each minute, if there is at least one girl holding at least
two cards, one of them must pass a card to each of her two neighbours. The game ends
if no girl is holding more than one card.

(a) Prove that if N ≥ 10, then it is impossible for the game to end.
(b) Prove that if N < 10, then the game must eventually end.

28. Consider a chocolate bar in the shape of an equilateral triangle, with sides of length n,
divided by grid lines into equilateral triangles of side length 1. Two players take turns
to break off a triangular piece along one of the grid lines and pass the remaining block
of chocolate to the other player. A player who is unable to move or who leaves an
equilateral triangle of side length 1 is declared the loser.
For which values of n does the second player have a winning strategy?

29. On an infinite grid, n2 pieces are arranged in an n × n block of squares, one piece per
square. A move consists of jumping a piece horizontally or vertically over a piece in
an adjacent occupied square to an unoccupied square immediately beyond. The piece
which has been jumped over is then removed.
Find those values of n for which this game can end with only one piece remaining.

30. Consider the unit squares formed by the integer lattice points in the x-y plane. At n
squares north of the x-axis is a princess whom her loyal subjects wish to rescue. In
each square below the x-axis is one pawn. (These are the princess’s loyal subjects.)
But the pawns can only move by jumping either horizontally or vertically over another
immediately adjacent pawn to a vacant square, after which the pawn that has been
jumped over is removed.
For which n is it possible that one of the princess’s loyal subjects can reach her by
landing on her square?
266 15 Games and invariants

31. Amy and Ben play a game of 11-in-a-row on an infinite two-dimensional square array.
They take turns to choose a vacant square and mark it. Amy goes first and marks
squares with an X, while Ben marks squares with an O. A player wins by being the
first to mark 11 consecutive squares vertically, horizontally or diagonally.
(a) Show that Amy can prevent Ben from winning.
(b) Show that Ben can prevent Amy from winning.
(c) Show that (a) and (b) are still true if they are playing 9-in-a-row instead.2

2 Forthe general case of n-in-a-row, it is known that the first player can force a win for n ≤ 5 and that the
second player can force a draw for n ≥ 8. As of September 2014, the status for n = 6, 7 is unknown.
15.1 Number invariants 267

15.1 Number invariants

There are lots of fun problems which look something like this.

You are given the configuration A and a set of legal moves which change the
configuration. Can you use these moves to end up with the configuration B?

If the answer is yes, then you could prove this by simply demonstrating a sequence of legal
moves which takes configuration A to configuration B. But if the answer is no, then you have
to be much trickier. More often than not, you will need to use the idea of an invariant. An
invariant is something—for example, a number—which we can associate to every configuration
such that performing a legal move doesn’t change it. So if the value of the invariant for
configuration A differs from its value for configuration B, then the task is impossible to
achieve, no matter how hard you try. All of this probably sounds quite cryptic and will
remain so until we see some examples in action.

Problem Given some numbers, we may choose two of them, say a and b, and replace them
with the single number a + b.
Prove that if we start with the numbers

1, 2, 3, . . . , 100

and apply the operation 99 times, we always end up with the same final number.

Solution It should be obvious that, after applying the operation 99 times, we will be left
with only one number. And it should seem intuitively clear that, no matter what order we
choose to perform the additions, the final number will be the sum of the original numbers.
This is certainly true, but we can state it in the language of invariants by associating to the
numbers a1 , a2 , . . . , an the sum

I = a1 + a2 + · · · + an .

The number I is an invariant because it doesn’t change when we apply the operation.
Therefore, it must be the same for both the initial and final configurations. Indeed, we have

1 + 2 + · · · + 100 = 5050,

so the final number will always be 5050.

Problem Given some numbers, we may choose two of them, say a and b, and replace them
with the single number ab + a + b.
Prove that if we start with the numbers
1 1 1
1, , , . . . ,
2 3 100
and apply the operation 99 times, we always end up with the same final number.

Solution Again, it should be obvious that, after applying the operation 99 times, we will be
left with only one number. Although this problem is more difficult than the previous one, it’s
still easily solved once you stumble upon the correct invariant. We simply associate to the
numbers a1 , a2 , . . . , an the value

I = (a1 + 1)(a2 + 1) · · · (an + 1).

268 15 Games and invariants

Removing the numbers a and b from our list divides the value of I by

(a + 1)(b + 1).

However, adding the number ab + a + b to our list multiplies the value of I by

ab + a + b + 1 = (a + 1)(b + 1).

So, the number I is an invariant because it doesn’t change when we apply the operation.
The value of I for the initial configuration can be determined using the following veritable
feast of cancellation.
     
1 1 1 1 2 3 4 101
I= +1 +1 + 1 ··· + 1 =
·
·
··· = 101
1 2 3 100 1 2
3
10
0

Since the invariant must be the same for both the initial and final configurations, the final
number will always be 100.

Of course, the difficulty in this solution lies in finding the actual invariant to use. A good
problem will often leave clues which can guide you in the right direction. For example, an
observation that might lead you to discover the correct invariant for this problem is the fact
that ab + a + b almost factorises as (a + 1)(b + 1) − 1.

15.2 Parity
Another extremely useful invariant is parity, which essentially means oddness or evenness.
You should always keep your eyes open for a parity argument.

Problem Given some numbers, we may choose two of them, say a and b, and replace them
with the difference |a − b|. Suppose that we start with the numbers

1, 2, 3, . . . , 2n,

for some odd positive integer n.

If we apply the operation 2n − 1 times, show that we always end up with an odd number.

Solution Since we always end up with an odd number, a likely candidate for an invariant is
the parity of the sum of the numbers. Initially, we have

2n(2n + 1)
1 + 2 + 3 + · · · + 2n = = n(2n + 1),
2
which is odd.
To show that the parity of the sum of the numbers doesn’t change after each operation,
suppose that we remove the numbers a and b, where we assume without loss of generality
that a ≥ b. Then the sum of the numbers would change by

−a − b + (a − b) = −2b,

which is an even number.

So the parity of the sum of the numbers is indeed an invariant. This shows that the final
number must be odd.
15.3 Modular arithmetic invariants 269

Problem You are given an 8 × 8 chessboard where the squares are coloured black and white
in the usual way. You are allowed to switch the colours of all the squares in a row or column.
Can you end up with exactly one black square on the board?

Solution The first thing you should do is take out some pen and paper, draw a chessboard,
and try to obtain exactly one black square. Playing around with the problem in this way
should convince you, sooner or later, that you probably cannot do it. In fact, you might even
be able to conjecture that it’s impossible to obtain an odd number of black squares. So let’s
try to show that the parity of the number of black squares is an invariant.
We note that if a row or column has x black squares, then after switching the colours, it will
have 8 − x black squares. So the change in the number of black squares is

−x + (8 − x) = 8 − 2x,

obviously an even number.

Therefore, the parity of the number of black squares is indeed an invariant and, since there
are initially 32 black squares, it’s impossible to end up with exactly one black square on the
board.

15.3 Modular arithmetic invariants

Parity is the same thing as considering an integer modulo 2. But in the world of invariants,
sometimes modulo 2 just isn’t good enough and you might need to consider integers modulo
other numbers.

Problem A magic lolly machine has the property that if two of Stephen’s Stupendous
Smarties are inserted, then three of Justin’s Jumbo Jaffas come out. Also, if three of
Stephen’s Stupendous Smarties are inserted, then two of Justin’s Jumbo Jaffas come out.
The reverse also occurs, that is, if two of Justin’s Jumbo Jaffas are inserted, then three of
Stephen’s Stupendous Smarties come out. And if three of Justin’s Jumbo Jaffas are inserted,
then two of Stephen’s Stupendous Smarties come out.

(a) If I want to turn two Jaffas into exactly 61 Jaffas, what is the minimum number of
Smarties that I also end up with?

(b) Can I turn one Jaffa and one Smartie into 10 Jaffas and no Smarties?

Solution

(a) The first thing you should do is take out some pen and paper, pretend you have a magic
lolly machine, and work out how many Jaffas and Smarties you can obtain. For example,
you might come up with the following possibilities, created from just two Jaffas.

Jaffas 2 0 3 1 7 1 13 1 28 0 63 61
Smarties 0 3 1 4 0 9 1 19 1 43 1 4
Difference 2 −3 2 −3 7 −8 12 −18 27 −43 62 57

It seems that if I want to turn two Jaffas into 61 Jaffas, then I might have to create at
least four Smarties.
270 15 Games and invariants

If we’re looking for an invariant, we could think about the sum of the number of Jaffas
and Smarties, but this doesn’t seem to be very useful.
The difference, on the other hand, is very interesting indeed. In fact, it seems that
J −S (mod 5)
is an invariant, where J is the number of Jaffas and S is the number of Smarties.
Let’s prove this by considering the effect of one operation of the magic lolly machine.
The pair (J, S) can become any of
(J − 2, S + 3), (J − 3, S + 2), (J + 3, S − 2) or (J + 2, S − 3).
In all of these cases, the difference changes from J − S to J − S ± 5, so we can now
conclude that J − S (mod 5) is indeed an invariant.
Initially, we only have two Jaffas and J − S ≡ 2 (mod 5). So to end up with 61 Jaffas,
we need at least four Smarties in order to maintain J − S ≡ 2 (mod 5).
But be careful! While the invariant allows us to decide that certain (J, S) combinations
are not possible, it doesn’t necessarily allow us to decide which ones actually are possible.
So we still need to show that one can obtain 61 Jaffas and four Smarties, but this we
have already accomplished in the table above. 
(b) Consider the problem of turning one Jaffa and one Smartie into 10 Jaffas and no
Smarties. Our invariant does not exclude this as a possibility, since J − S ≡ 0 (mod 5)
for both cases. However, observing that we need at least two Jaffas or at least two
Smarties to use the magic lolly machine tells us that this task is impossible. 

15.4 Colouring invariants

You may have come across the following classic puzzle before. The idea behind its beautifully
simple solution can be generalised to solve far more difficult problems.

Problem Consider an 8 × 8 chessboard, where the top-right and bottom-left squares have
been removed.
Is it possible to tile this mutilated chessboard with 2 × 1 rectangles?

Solution The first thing you should do is take out some pen and paper, draw a mutilated
chessboard, and try to tile it with 2 × 1 rectangles. However, I can tell you right now that
you will fail, not because your tiling skills are poor, but because the task is impossible!
Perhaps surprisingly, the key to this problem is the standard black-and-white colouring of the
chessboard. This is because a 2 × 1 rectangle will always occupy two adjacent squares on the
chessboard and hence, cover one black square and one white square. Therefore, any part of
the chessboard that can be tiled with 2 × 1 rectangles must have the same number of black
and white squares.
Now we note that the standard chessboard has 32 squares of each colour, while the mutilated
chessboard is obtained by removing two squares of the same colour. Since there are now 30
squares of one colour remaining and 32 squares of the other colour, it is impossible to tile the
mutilated chessboard with 2 × 1 rectangles.

We were lucky in this problem, because the standard 8 × 8 chessboard came with a colouring
which helped our cause, free of charge. But sometimes, as in the next problem, you have to
invent your own colouring.
15.4 Colouring invariants 271

Problem Show that a 10 × 10 chessboard cannot be tiled with 4 × 1 rectangles.

Solution The tactic is to find a colouring of the chessboard such that any 4 × 1 rectangle
on the board occupies one square of each colour. Of course, this means that we require four
colours, which we will call 0, 1, 2 and 3. Working along the bottom row of the chessboard,
we may as well label the first four squares 0, 1, 2 and 3, in that order. After that, every
square in the row must be coloured according to the repeating pattern 0, 1, 2, 3, 0, 1, 2, 3, and
so on. If we apply the same argument along the columns, we might end up with the following
colouring.

1 2 3 0 1 2 3 0 1 2
0 1 2 3 0 1 2 3 0 1
3 0 1 2 3 0 1 2 3 0
2 3 0 1 2 3 0 1 2 3
1 2 3 0 1 2 3 0 1 2
0 1 2 3 0 1 2 3 0 1
3 0 1 2 3 0 1 2 3 0
2 3 0 1 2 3 0 1 2 3
1 2 3 0 1 2 3 0 1 2
0 1 2 3 0 1 2 3 0 1

We call this a modulo 4 colouring, because if we label the rows and columns 0, 1, 2, . . ., then
the square in row i and column j is coloured i + j modulo 4.
This colouring certainly obeys the rule that a 4 × 1 rectangle on the board always occupies
one square of each colour. Of course, we’re hoping that there are not the same number of
squares of each colour. One way to verify this is to simply count them and you would indeed
find that this is true. However, that is rather pedestrian, so let’s use a slicker, more stylish,
approach.
We simply note that it is quite easy to demonstrate a tiling of the entire board except for
the 2 × 2 square in the top-right corner. The tiled part of the board must certainly contain
the same number of squares of each colour, otherwise we wouldn’t have been able to tile it.
However, the remaining part of the board does not because there is one square coloured 0,
two squares coloured 1, one square coloured 2 and no squares coloured 3. Hence there cannot
be the same number of squares of each colour on the entire chessboard. We conclude that a
10 × 10 chessboard cannot be tiled with 4 × 1 rectangles.

For other problems, you might need to use a modulo n colouring for some other positive
integer n. Or something completely different might be needed! Using the notation (i, j) to
represent the square in row i and column j, other useful colourings include the following.

Colour (i, j) according to i (mod 2). This yields a striped pattern.

Colour (i, j) according to (i (mod 2), j (mod 2)). This uses four colours but is different
from the modulo 4 colouring used in the above solution.
272 15 Games and invariants

15.5 Monovariants
An invariant is something which doesn’t change when you perform a particular move. On
the other hand, a monovariant is a value which always gets larger or always gets smaller
when you perform a particular move. For example, if you keep spending money without ever
earning any, then you will never again have as much money as when you started spending.
Here the monovariant is obviously the amount of money that you have. This idea is crucial
to solving many problems, including the following.

Problem Given some numbers, we may choose two of them, say a and b, and replace them
with the numbers
b a
a+ and b − .
2 2

If we start with a set of non-zero numbers S and keep applying the operation, show that we
can never again obtain the set S.

Solution Let the numbers be a1 , a2 , . . . , an and consider the sum of squares

M = a21 + a22 + · · · + a2n .

We will determine the change in M after we replace two of the numbers, say a and b.
2 
a2 b2

b a 2
change in M = a+ + b− − a2 − b2 = + ≥0
2 2 4 4

So M is not an invariant but a monovariant, because it never decreases. In fact, since S

consists of non-zero numbers, M is guaranteed to increase after the first operation. So the
same value of M can never again be obtained by applying the operation. Therefore, we can
never again obtain the set S.

For some reason, squares often feature in monovariant problems, as they did in the previous
one. Next, we will see how the idea of a monovariant can help when it doesn’t seem like it
should.

1
Problem A unit fraction is a number of the form n, where n is a positive integer.
Prove that every rational number between 0 and 1 can be expressed as a sum of finitely many
distinct unit fractions.

Solution We will show that this can be achieved using the greedy algorithm.3 You should
convince yourself that the solution to the problem follows immediately once we have proven
the following statement.

Given a rational number 0 < r < 1, subtract the largest unit fraction less than or
equal to r. Prove that, if we continue to do this, we must eventually reach the
number 0 after finitely many subtractions. Furthermore, we never subtract the
same unit fraction more than once.
3 That is, take as much as you can at each stage.
15.6 Invariants as cost 273

The idea is to show that the numerator of the rational number is a monovariant which
decreases until we reach 0.
Write the number r as
a
r= ,
b
where a and b are relatively prime positive integers.
1
Let the largest unit fraction less than or equal to r be m, so that

1 a 1
≤ < .
m b m−1

Therefore, after one step, we have the fraction

a 1 am − b
− = .
b m bm
a 1
However, the inequality b < m−1 implies that am − b < a.
This means that the numerator of the fraction strictly decreases after each step4 , until we
eventually reach the number 0.
It should be clear that we never subtract the same unit fraction more than once because
1 1
if m is the largest unit fraction less than or equal to r, then r − m is too small to be able to
1
subtract m again.

15.6 Invariants as cost

In our increasingly capitalistic world, we regularly think about money. So it’s sometimes
useful to think of an invariant as the cost of something, like we do in the following example.

Problem Initially, there is a pawn placed in each square in the bottom four rows of an 8 × 8
chessboard. If two pawns are in adjacent squares of the same row, you are allowed to remove
them and add a pawn in the row above.
Is it possible to place a pawn in the top row of the chessboard?

Solution Let’s number the rows in order so that 1 is the lowest row while 8 is the highest
row. The idea behind this problem is as follows: since two pawns in row R can become one
pawn in row R + 1, we should consider the cost of a pawn in row R + 1 to be twice the cost
of a pawn in row R.
So suppose that pawns in row 1 cost $1, pawns in row 2 cost $2, pawns in row 3 cost $4,
and so on, so that pawns in row 8 cost $128. The total value of the pawns initially on the
chessboard is
8 × ($1 + $2 + $4 + $8) = $120.

Since this cost is invariant, it’s impossible to place a pawn in the top row of the chessboard,
which would cost $128.

is, of course, the possibility that am−b

4 There
bm
is not in lowest terms. However, in this case, cancelling the
common factor decreases the numerator even further.
274 15 Games and invariants

15.7 Permutation parity

A permutation is just a rearrangement of objects. More formally a permutation of a set S is
a bijection f : S → S.
For finite sets there is a convenient two-line notation that represents a permutation. For
example, if S = {1, 2, 3, 4}, then the notation
 
1234
2314

represents the permutation f (1) = 2, f (2) = 3, f (3) = 1 and f (4) = 4.

A shorter one-line notation is just to write

(2 3 1 4)

to mean the same thing.

We can combine permutations by applying them successively. Thus if we applied (2 3 1 4) and
then followed this by (1 2 4 3), the net result is (2 4 1 3). This is written as
    
1234 1234 1234
(1 2 4 3)(2 3 1 4) = (2 4 1 3) or = .
1243 2314 2413

Note that because permutations are functions, we write the order of composition of permuta-
tions from right-to-left, as shown in the above example.
It turns out that permutations have a parity invariant which is highly useful.

Parity of a permutation The parity of a permutation is defined as the parity of the

number of inversions in the permutation.

A pair i < j is said to be an inversion if j occurs before i in the permutation. For example,
the permutation (2 3 1 4) has inversions (2, 1) and (3, 1). Thus (2 3 1 4) is an even permutation.
The permutation (1 2 4 3) has (4, 3) as its only inversion and so it is an odd permutation. The
composition (2 4 1 3) has three inversions so it is an odd permutation.
The important thing you need to know about permutation parity is the following.

Parity of compositions of permutations When composing two permutations, their

parities behave in the same way as ordinary parity does for addition.

In particular we have the following.

The composition of two even permutations is an even permutation.

The composition of two odd permutations is an even permutation.

The composition of an even permutation and an odd permutation is an odd permutation.

These can be proved via the following exercises which we leave for you to do.

A transposition is a permutation which swaps two elements. Prove that applying a

transposition to a permutation changes its parity.

Prove that any permutation can be written as a composition of transpositions.

15.8 Combinatorial games 275

Thus conclude that the parity of a permutation is simply the parity of the number of
transpositions which can be used to express it.5

The following illustrates how permutations and parity may be used to solve a problem quickly.

Problem Andrew, Brenda and Chris are competing in a three-person race. At some point
in the race Andrew is winning, with Brenda coming second and Chris third. From here on
until the end of the race it is noted that there are 36 times when their relative order changes.
The race ends with Brenda finishing third.
If at no point in time did all three draw level, who won the race?

Solution The original order may be notated as [A,B,C] (Andrew first, Brenda second, Chris
third). Each change in the order is a transposition which is an odd permutation. Since there
are 36 such permutations, the net result is an even permutation of [A,B,C].
We are given that B comes last, so the order is either [A,C,B] or [C,A,B]. But [A,C,B] is an
odd permutation, whereas [C,A,B] is an even permutation. Thus the final order must be
Chris first, Andrew second and Brenda third.

15.8 Combinatorial games

For the remainder of this chapter, we consider combinatorial games, which are games that
satisfy the following conditions.

There are two players who take turns to move.

There is no luck involved.

The game always ends after a finite number of moves.

Each player’s moves are known to the other player.

Either one player wins and the other loses or the game results in a draw.

The most important result concerning combinatorial games is the following.

Fundamental theorem of combinatorial games In a combinatorial game, either there

exists a winning strategy for one of the players or both of them can force a draw.

Problem There are two piles on a table, one containing 1000 coins and the other containing
1001 coins. Two players take turns to remove at least one coin from one of the piles. The
winner is the player who removes the last coin from the table. Since a draw is impossible, the
fundamental theorem of combinatorial games guarantees that there exists a winning strategy
for one of the players.
Which player has a winning strategy?

Solution The easiest way to determine which player has a winning strategy is simply to
come up with a winning strategy which works!
5 This also proves that no matter how we write a permutation as a composition of transpositions (and there are
many different ways of doing this) the parity of the number of transpositions is always the same. Specifically,
it is equal to the parity of the permutation.
276 15 Games and invariants

In this game, the first player has a winning strategy which we can describe as follows: take
coins from the larger pile to leave two piles of the same size.
For this strategy to work, the first player must always see two piles of different sizes on their
move. This is certainly true for their first move since the piles contain 1000 and 1001 coins.
Subsequently, the second player is always faced with two piles of the same size which they
must turn into two piles of different sizes. So the first player continues to see two piles of
different sizes when it is their turn to move.
But we still have to show why the second player has no chance of winning against this strategy.
The reason for this is that, as we have already mentioned, the second player always leaves
two piles of different sizes. However, to win the game, you must leave two piles of equal size,
both with zero coins. Since the second player can never win the game as long as the first
player follows this plan, what we have described is a winning strategy for the first player.

15.9 Position analysis

Often, a winning strategy doesn’t present itself very easily but can still be found by using a
technique called position analysis. If you are about to move and you have a strategy which
allows you to win, then we call the current position of the game a winning position. On the
other hand, if you are about to move and you do not have a strategy which allows you to
win, then we call the current position of the game a losing position. One corollary of the
fundamental theorem of combinatorial games is the following extremely useful result.

Theorem In a combinatorial game with no draws, every position of the game can be
categorised as winning or losing. Furthermore, it must be the case that
from a winning position, it is possible to move to a losing position, and

from a losing position, it is impossible to move to a winning position.

The idea of position analysis is to use this result to analyse enough small cases until we see
can some general patterns. If all goes well, then we should be able to describe all of the
winning positions, all of the losing positions, and perhaps even a winning strategy.

Problem Initially, there are n coins on a table and two players take turns to remove 1, 2, 3
or 4 coins. A player wins if he removes the last coin.
Find the winning and losing positions.

Solution In this problem, we can describe the position of the game by the number of coins
on the table. It is clear that 1, 2, 3 and 4 are all winning positions because we can simply
remove all of the coins on the table if it is our move.
But what happens when there are 5 coins on the table? Well, the only possibilities are for us
to leave 1, 2, 3 or 4 coins on the table, thereby leaving our opponent in a winning position.
So 5 must be a losing position.
But if 5 is a losing position, then 6, 7, 8 and 9 must all be winning positions. That is because
from these positions, we can remove 1, 2, 3 or 4 coins to leave 5 coins on the table, thereby
leaving our opponent in a losing position.
Continuing this argument gives us the following table.
15.10 The copycat strategy 277

Winning 1, 2, 3, 4, 6, 7, 8, 9, 11, 12, 13, 14, 16, 17, 18, 19, . . .

Losing 5, 10, 15, 20, . . .

At this stage, it seems like a safe bet that the losing positions are simply the multiples of 5.
To prove that this is true, all we need to do is demonstrate that

from a non-multiple of 5, it is possible to move to a multiple of 5, and

from a multiple of 5, it is impossible to move to a multiple of 5.

Both of these statements are obvious! For the first, we use the fact that a non-multiple of 5
must be of the form 5k + 1, 5k + 2, 5k + 3 or 5k + 4 and subtracting 1, 2, 3 or 4 from each of
these, respectively, leaves a multiple of 5.
For the second, we note that the difference between two multiples of 5 is still a multiple of 5
and certainly cannot be 1, 2, 3 or 4.

Note that not only have we determined the winning and losing positions, we have also
uncovered a simple winning strategy: always reduce the number of coins to a multiple of 5.

15.10 The copycat strategy

The notion of symmetry is a most powerful one in mathematics. Here, we will see how to use
it to beat your friends at certain games, simply by being a copycat.

Problem Next to a square table is a pile of circular coins, all the same size. Two players
take turns putting a coin on the table so that it doesn’t touch any other coin. The player
who cannot do so loses the game.
Show that the first player can always win.

Solution The first player can win by placing the first coin at the centre of the table. Now,
wherever the second player places their coin, the first player simply copies them by placing
their coin symmetrically opposite.
After each of the first player’s moves, the configuration of coins is symmetric under a 180◦
rotation of the table. Thus, whenever the second player can place a coin on the table, the
first player can also do so by placing their coin symmetrically opposite.6 So by using this
copycat strategy, the first player can always win.

15.11 Pairing strategies

Many games involve the movement of pieces into various positions. Sometimes, a strategy
can be given by pairing up the positions so that wherever the first player can go, the second
player should go in the position that it is paired with.

Problem Alice and Bob play a game on a large7 grid where they take turns to choose a
square and mark it. Alice moves first and marks squares with an X while Bob marks squares
6 Itis possible that a coin may overlap a symmetrically opposite coin. But this occurs only if the coin lies over
the centre of the table. This is ruled out by the first player’s very first move.
7 A 100 × 200 grid is a nice large grid for the purposes of this problem.
278 15 Games and invariants

with an O. They play until one of the players marks a row or a column of five consecutive
squares, and this player wins the game. If no player marks a row or column of five consecutive
squares, then the game is declared a draw.
Show that Bob can prevent Alice from winning.

Solution Label the board as shown in the diagram, repeating the pattern in all directions.

1 2 3 3 1 2 3 3
1 2 4 4 1 2 4 4
3 3 1 2 3 3 1 2
4 4 1 2 4 4 1 2
1 2 3 3 1 2 3 3
1 2 4 4 1 2 4 4
3 3 1 2 3 3 1 2
4 4 1 2 4 4 1 2

Each square is paired with the neighbouring square that contains the same label.
Suppose that whenever Alice plays, Bob plays in the neighbouring square with the same label.
In this way, Alice can never occupy both squares of such a pair.
But the pairing was chosen very carefully so that any block of five consecutive squares in a
row or column contains such a pair. Needless to say, you should check that this is true for
yourself. Hence, Alice can never mark a row or a column of five consecutive squares.

Since Bob can prevent Alice from winning, the fundamental theorem of combinatorial games
states that he must have a winning strategy or both players can force a draw. In the next
section, we will use a sneaky technique to show that the latter is the case.

15.12 Strategy stealing

In the previous problem we showed that Bob could prevent Alice from winning. But as
unlikely as it seems, could Bob force a win for himself? This is where strategy stealing shows
its usefulness.

Problem Alice and Bob play a game on a large grid where they take turns to choose a
square and mark it. Alice moves first and marks squares with an X while Bob marks squares
with an O. They play until one of the players marks a row or a column of five consecutive
squares, and this player wins the game. If no player marks a row or column of five consecutive
squares, then the game is declared a draw.
Show that Alice can prevent Bob from winning.

Solution The main idea is that Alice’s extra move at the start of the game can never hurt
her chances of winning. However, this intuition doesn’t constitute a proof in itself, but can
be turned into one by using the concept of strategy stealing.
15.12 Strategy stealing 279

In order to obtain a contradiction, suppose that Bob has a winning strategy. In other words,
it is possible to write a book which describes how Bob can win, no matter how Alice plays.
Suppose now that Alice manages to steal this second player’s strategy book and plays as
follows. She simply places her first move randomly on the grid, ignores the fact that she has
moved, and then pretends that she is the second player. She continues to do this until the
book tells her to move in the square where she moved first. Since she has already done this,
she can use this move to mark a different random square on the grid. Continuing in this way,
we see that any book which provides a winning strategy for the second player can be used to
provide a winning strategy for the first player! This blatantly contradicts the fundamental
theorem of combinatorial games. So we must conclude that there is no winning strategy for
the second player at all. In other words, Alice can prevent Bob from winning.

Strategy stealing is brimming with trickery, so have a read through the above solution again
until it’s well understood. After that, you may have a look at the following example of
strategy stealing at its best.

Problem In the game Double Chess, the rules of chess are changed so that White and Black
alternately make two legal moves at a time.
Show that Black doesn’t have a winning strategy.

Solution In order to obtain a contradiction, let us suppose to the contrary that Black does
have a winning strategy. Then it must be the case that whatever White does on the first
move, Black can win from the resulting position.
So what would happen if White started by moving a knight out and then back to the square
that it was originally on? At this stage, the board looks exactly the same as it did initially,
but it’s now Black’s turn to move. Remember that, by assumption, Black can force a win
from this position. But if that is the case, couldn’t White have just mirrored Black’s strategy
from the very beginning of the game? Of course this is possible and so White also seems to
have a winning strategy. This contradicts the fundamental theorem of combinatorial games,
because White and Black cannot both have winning strategies. So our original assumption
must have been wrong and we conclude that Black doesn’t have a winning strategy.
 Combinatorial geometry
16
Combinatorial geometry is an exotic hybrid. Usually a combinatorial problem is posed in a
geometric setting. The usual ideas in combinatorics are often insufficient to take into account
extra constraints that come from the geometric setting. But in this chapter we will see ideas
that can be used for such problems.

16.0 Problems
1. Several points lie in the plane such that the area of the triangle formed by any three of
them is no more than 1.
Show that all the points lie in a triangle of area no more than 4.

2. Let K be a set of points in R3 such that every triangle with vertices in K has a side of
length at most 1.
Prove that there exist two spheres S1 and S2 , both of radius 1, such that K ⊆ S1 ∪ S2 .

3. Let b, r be positive integers. Suppose we are given 2r red points and 2b blue points in
the plane such that no three points are collinear.

(a) Show there exists a line in the plane with exactly b blue points and r red points
on each side.
(b) Can you generalise this question to three dimensions?

4. Each point on the perimeter of an equilateral triangle is coloured either black or white.
Is it always possible to find three points of the same colour which are also the vertices
of a right-angled triangle?

5. Prove that any convex polygon of area 1 lies inside a rectangle of area 2.

6. Show that for all n ≥ 4 there exists a convex hexagon which can be dissected into n
congruent triangles.

7. (a) Show that for all n ≥ 4, any cyclic quadrilateral can be dissected into n cyclic
quadrilaterals.
(b) What if the original quadrilateral is not cyclic?
282 16 Combinatorial geometry

8. A Platonic solid is a convex polyhedron in which each vertex is surrounded by the same
number of congruent regular polygons.
Prove that there are exactly five Platonic solids and determine the number of vertices,
edges and faces of each.

9. (a) Let S be a set of n ≥ 3 convex sets in the plane with empty intersection. Show
that there exist three members of S with empty intersection.
(b) Let S be a set of n ≥ 4 convex sets in space with empty intersection. Show that
there exist four members of S with empty intersection.
(These results are known as Helly’s theorem.)

10. There exist rectangles which can be dissected into squares of different sizes.
Can you find an example of this using just nine squares?1

11. Show that it is not possible to dissect a cube into finitely many cubes of different sizes.

12. (a) Show that any polygon of n sides can be cut into triangles.2
(b) Show that this can be done by cutting along just n − 3 diagonals of the polygon
resulting in n − 2 triangles.
(c) There are examples where we can get away with fewer than n − 2 triangles. What
is the least possible number of triangles?
(d) Show that any polyhedron can be dissected into tetrahedra.

13. For any set of points S, we say that S admits distance d, if there are two points in S
such that the distance between them is d.

(a) We wish to colour every point of the plane with finitely many colours in such a way
that no colour admits distance 1. Let χ be the minimal such number of colours for
which this is possible.
Show that 3 ≤ χ ≤ 9.
(b) Can you show that 4 ≤ χ ≤ 7?3
(c) If we only colour the rational points of the plane4 , show that χ = 2.

14. All points in space are coloured in one of three colours.

Prove that one of these colours admits all distances.

15. We are given

n points on a circle. We join each pair of such points with a line segment
forming n2 chords. These chords divide the circle into regions.
Find a sharp upper bound for the number of regions formed.

16. A cube of side length 30 contains 999 blue points in its interior, no four of which are
coplanar. Consider also the eight vertices of the cube which are coloured red.
Among the 1007 coloured points considered, prove that four of them, including at least
one blue point, form the vertices of a tetrahedron whose volume is less than 9.
1 This is called squaring a rectangle. It is also possible to square a square, but doing this requires at least 21
squares.
2 It is the non-convex case that is particularly tricky!
3 The number χ is known as the chromatic number of the plane. As of September 2014, it is still an open

problem to determine the exact value of χ. The bound given here is the best known so far. As for the
corresponding problem in three-dimensional space, it is known that 6 ≤ χ ≤ 15.
4 One can also prove that χ = 2 for the rational points of three-dimensional space.
16.0 Problems 283

17. Determine all integers n ≥ 4 for which there exist n points A1 , A2 , . . . , An in the plane
and real numbers r1 , r2 , . . . , rn satisfying the following two conditions.

(i) No three of the points A1 , A2 , . . . , An lie on a line.

(ii) For each triple 1 ≤ i < j < k ≤ n we have

Area(4Ai Aj Ak ) = ri + rj + rk .

18. A square ABCD is given. A triangulation of the square is a partition of the square into
triangles such that any two triangles are either disjoint, share only a common vertex,
or share only a common side. A good triangulation of the square is a triangulation in
which all the triangles are acute.

(a) Give an example of a good triangulation of the square.

(b) What is the minimal number of triangles required for a good triangulation?

19. Find all possible sets S of n ≥ 3 points in the plane such that every perpendicular
bisector of every pair of distinct points of S is an axis of symmetry of S.

20. Each point of the coordinate plane is coloured using finitely many colours. Let O denote
the origin. For each point X different from O, let C(X) be the circle with centre O and
radius
α(X)
OX + ,
OX
where α(X) is the angle measured clockwise in radians that OX makes with the positive
x-axis.
Prove there exists a point Y such that α(Y ) > 0, and the colour of Y appears on the
circle C(Y ).

21. For any set S of five points in the plane, no three of which are collinear, let M (S) and
m(S) denote the largest and smallest areas, respectively, of triangles determined by
three points from S.
What is the minimum possible value of M (S)/m(S)?

22. One is given a finite set of points in the plane, each point having integer coordinates.
Is it always possible to colour some of the points in the set red and the remaining points
white, in such a way that for any straight line L parallel to either one of the coordinate
axes the difference (in absolute value) between the numbers of white points and red
points on L is not greater than 1?

23. Each point in a square is coloured in one of 2005 colours.

Prove that there exists a rectangle with vertices of the same colour.

24. Determine whether or not there exist two disjoint infinite sets A and B of points in the
plane satisfying the following two conditions.

(i) No three points of A ∪ B are collinear and the distance between any two is at
least 1.
(ii) There is a point of A in any triangle whose vertices are in B, and there is a point
of B in any triangle whose vertices are in A.
284 16 Combinatorial geometry

25. We are given n ≥ 2 distinct lines in the plane such that no two lines are parallel and
such that the lines are not all concurrent through a single point.
Prove that there exists a point through which exactly two of the lines pass.5
26. Let n, k be positive integers and let S be a set of n points in the plane for which no
three points of S are collinear, and for every point P of S there are at least k points
of S equidistant from P .
Prove that
1 √
k< + 2n.
2
27. Consider a square of side length a positive integer n. Suppose that there are (n + 1)2
points in the interior of the square.
Show that three of these points define a (possibly degenerate) triangle of area at most 12 .

5 This is the dual of Sylvester’s theorem.

16.1 Proof by contradiction 285

16.1 Proof by contradiction

This is one of the most basic techniques in mathematics in general. Recall from section 1.5
that the method is to show that if the proposition is false, then it follows that some nonsense
is true. Of course this is a situation we cannot tolerate and so the proposition must be true.

Problem Is it possible to colour every point on a circle using the two colours white and
blue so that there is no isosceles triangle whose vertices all have the same colour?

Solution After experimenting for a while you may be convinced that it is not possible. So
assume for the sake of contradiction that it is possible.
Then certainly we can pick two points A and B of the same colour, say white. If X is the
point on the circle such that the arc lengths XA and AB are equal, then triangle AXB is
isosceles and so X must be blue. Similarly the point Y is blue where Y satisfies AB = BY .

P
A • B
◦
• •
◦

X Y
• •

Consider now the point P , say, on the minor arc of the circle halfway between A and B. Since
AP = P B, P cannot be white. Furthermore, since XP = Y P , point P cannot be blue.
Thus P cannot be any colour, which is a contradiction.

This proof is not quite complete because it may occur that the points A, B, X, Y and P are
not all distinct. For instance, if ABX forms an equilateral triangle, then X = Y .
We leave it for you to think about how to overcome this problem.
Actually, there is a really short solution to this problem as follows.

Solution Let ABCDE be any regular pentagon inscribed in the circle. Then by the
pigeonhole principle, three of its vertices must be the same colour.
However any three points of a regular pentagon form an isosceles triangle!

16.2 Extremal principle

Sometimes picking out something that is extremal in some sense can be just what needs to
be focused on to solve a problem completely.

Problem We are given a set of discs in the plane with pairwise disjoint interiors. Each disc
is tangent to at least six other discs of the family.
Prove that there are infinitely many discs in the set.
286 16 Combinatorial geometry

Solution Assume that the family is finite. Then there is a disc, D say, of minimal radius r.
Thus there are at least six discs around D and of radius at least r. However there is only
room for there to be exactly six discs, all of radius equal to r around D.

We may apply the same argument to each of these discs, thus generating infinitely many discs
of radius r at ever increasing distances from D. This is a contradiction.

Problem A closed6 and bounded7 shape S in the plane has the property that any two
points of S can be connected by a semicircular arc which lies completely in S.
Find all possibilities for the figure S.

Solution Since S is closed and bounded, there exist points A, B ∈ S such that the distance
d = AB is maximal.8 Let α be the semicircular arc lying in S which joins A to B and let γ
be the full circle defined by α.
Let P be any point on α. Then we know that there is a semicircular arc β lying in S which
joins B to P . If β lies outside of γ, then there exists a point on β, whose distance from A is
greater than d.9 This is a contradiction.

α
β
P

A B

Thus β lies inside γ. This is true for all points P ∈ α. Since the set of such β covers the
entire interior of γ, we see that the interior of γ is a subset of S.
Since S is closed, it follows that the boundary of γ is also a subset of S. Finally, since any
point lying outside of γ would give rise to a distance in S greater than d, we conclude that S
is in fact the closed disc described by γ.
6 We say that a set is closed if it contains its boundary. For example, the solid disc {(x, y) | x2 + y 2 ≤ 1} is
closed. So is the unit circle {(x, y) | x2 + y 2 = 1}. But the disc {(x, y) | x2 + y 2 < 1} is not closed because it
is missing some (in fact all) of its boundary, namely the unit circle.
7 We say that a set is bounded if there is a real number R > 0 such that the distance between the origin and

any point of the shape is at most R. The three examples in the previous footnote are all bounded because
all points of those shapes lie within distance 1 of the origin.
8 The existence of two such points is a consequence of the fact that a continuous function on a compact (i.e.

closed and bounded) set achieves a maximal value. In this case the compact set is S ×S = {(X, Y ) | X, Y ∈ S}
and the function is f : S × S → R, f (X, Y ) = XY .
9 For example, a point which lies very close to B on β would be such an offending point.
16.3 Perturbation 287

Problem Given n points in the plane, no three of which are collinear, show it is possible to
join them up in sequence so that we have a broken line consisting of n − 1 segments, no two
of which cross each other.

Solution There does not seem to be anything obviously extremal to look at here. However,
if you drew a really long path through the points, it seems likely that the path would contain
lots of intersections. A shorter path probably would contain fewer intersections. The shortest
path hopefully would contain none. Let us prove that this is indeed the case.
Let A1 A2 . . . An be a path of minimal length. Suppose that Aj Aj+1 crosses Ai Ai+1 (i < j).
Then consider the quadrilateral Ai Aj Ai+1 Aj+1 . The diagonals intersect at a point P inside
the quadrilateral.

Aj−1 Ai+2
Aj−1 Ai+2
Ai+1
Aj Ai+1
Aj
P −→
P
Ai Aj+1 Ai Aj+1

From the triangle inequality we have

Ai P + Aj P > A i Aj and Ai+1 P + Aj+1 P > Ai+1 Aj+1 .

Adding these two equations yields

Ai Ai+1 + Aj Aj+1 > Ai Aj + Ai+1 Aj+1 .

Thus the path

A1 A2 . . . Ai Aj Aj−1 . . . Ai+2 Ai+1 Aj+1 Aj+2 . . . An
is of shorter length than A1 A2 . . . An , which is a contradiction.

16.3 Perturbation
Sometimes objects such as lines or points may be not quite in the position that you want
them. A miniscule jiggle of the configuration can sometimes rectify this. We illustrate this
with another way to approach the previous problem.

Problem Given n points in the plane, no three of which are collinear, show it is possible to
join them up in sequence so that we have a broken line consisting of n − 1 segments, no two
of which cross each other.

Solution The points lie in the x-y plane. So if we label the points P1 , P2 , . . . , Pn according
to increasing x-coordinate, then we could simply join P1 to P2 , P2 to P3 , and so on.
However, it may not be the case that the x-coordinates are all distinct. Surely we can rotate
the configuration in the plane so that all x-coordinates are distinct! Indeed all we have to
do is rotate the

configuration so that the y-axis is not parallel to any of the lines formed by
joining all n2 pairs of points.
288 16 Combinatorial geometry

16.4 Induction
More than likely, a problem for which we can build larger examples out of smaller examples
can be approached by mathematical induction.

Problem Given 1002 distinct points in the plane, we join every pair of points with a line
segment and colour its midpoint red.
Show that there are at least 2001 red points.

Solution We prove by the induction the more general statement that for n ≥ 2 points there
are at least 2n − 3 red points.
The result is clearly true for n = 2.
Suppose now that the result is true for n = 2, 3, . . . , m where m ≥ 2. Consider an arrangement
of m + 1 points. By using a perturbation argument we may assume that all points have
distinct x-coordinates. Label them as A1 , A2 , . . . , Am , Am+1 by increasing x-coordinate.
By the inductive assumption we have at least 2m − 3 red points from the midpoints of
A1 , A2 , . . . , Am . Can we find two more red points by using Am+1 ? Yes! The midpoints of
Am+1 Am−1 and Am+1 Am are distinct and both are to the right of all red points considered
so far. Thus we have at least 2m − 1 = 2(m + 1) − 3 red points in all. This completes the
induction.

As an extension, can you determine where equality occurs in the above problem?

16.5 Discrete intermediate value theorem

The ordinary intermediate value theorem states that if f (x) is a continuous function defined
on some interval which achieves the value f (a) somewhere and the value f (b) somewhere
else, then f (x) achieves all values between f (a) and f (b). In particular, if f (x) is positive
somewhere and negative somewhere else, then it is necessarily zero somewhere in between.
There is a discrete analogue of this that involves sequences of integers.

Discrete intermediate value theorem If a1 , a2 , . . . , an is a sequence of integers with

the property that |ai − ai+1 | ≤ 1 for i = 1, 2, . . . , n − 1 and is such that ai < 0 and aj > 0
for some 1 ≤ i, j ≤ n, then ak = 0 for some 1 ≤ k ≤ n.

The following problem illustrates how useful this can be.

Problem We are given 2n + 1 blue points in the plane such that no three are collinear and
no four are concyclic.
For every pair of blue points A, B, show that there exists a circle passing through A, B with
n − 1 blue points inside it, three points on its boundary and n − 1 blue points outside it.

Solution Consider any two blue points A and B and orient the plane so that AB is vertical.
For each point P on the perpendicular bisector of AB there is an associated circle ΓP having
centre P and passing through A and B. The idea is to consider what happens as P varies
from far to the left of AB to far to the right of AB.
Suppose there are a blue points to the left of the line AB and b blue points to the right of
the line AB. WLOG a ≥ b. Since a + b = 2n − 1, a and b are of opposite parity and so we
must have a > b.
16.6 Convex hull 289

There is a point Pleft , such that the corresponding circle contains in its interior all of the a
blue points located to the left of AB but none of the b blue points to the right of AB. There
is also a point Pright , such that the corresponding circle contains in its interior all of the b
blue points located to the right of AB but none of the a blue points to the left of AB.10
As P varies continuously between Pleft and Pright the circle ΓP will meet the remaining 2n − 1
blue points one by one. When ΓP meets the ith such blue point, let ai be given by
ai = I(ΓP ) − O(ΓP ),
where I(ΓP ) is the number of blue points lying inside ΓP and O(ΓP ) is the number of blue
points lying outside ΓP . Note that ai is even because I(ΓP ) and O(ΓP ) have the same parity
due to I(ΓP ) + O(ΓP ) = 2n − 2.

•
A
•
•

• •
•
B

If the first point that ΓP meets is to the left of AB, then a1 = a − 1 − b. If it is to the right of
AB, then a1 = a − (b − 1). Either way we have a1 ≥ a − b − 1 ≥ 0 because a > b. Similarly,
if the last point that ΓP meets is to the left of AB, then a2n−1 = b − (a − 1). If it is to the
right of AB, then a2n−1 = b − 1 − a. Either way we have a2n−1 ≤ b − a + 1 ≤ 0 because
b < a. To summarise we have a1 ≥ 0 and a2n−1 ≤ 0.
What happens when ΓP goes from meeting the ith blue point to meeting the (i + 1)th blue
point?

If both such points are to the left of AB, then I(ΓP ) decreases by 1 while O(ΓP )
increases by 1. Thus ai+1 = ai − 2.
If both points are to the right of AB, then I(ΓP ) increases by 1 while O(ΓP ) decreases
by 1. Thus ai+1 = ai + 2.
If the points are on opposite sides of AB, then I(ΓP ) and O(ΓP ) remain unchanged.
Thus ai+1 = ai .

In all cases we have |ai+1 − ai | ≤ 2. Finally, applying the discrete intermediate value theorem
to the sequence 12 a1 , 12 a2 , . . . , 12 a2n−1 guarantees that ai = 0 for some i. Then the circle
corresponding to this i satisfies the conclusion of the problem.

16.6 Convex hull

Consider a wooden board with some nails hammered into it. What happens when you take a
rubber band and stretch it tightly around the area occupied by the nails? You end up with a
10 You might like to think of a reason for why Pleft and Pright must exist.
290 16 Combinatorial geometry

polygon whose angles are all less than or equal to 180◦ . The area occupied by this polygon is
called the convex hull of the nails.
The notion of convexity is defined as follows. A set S is convex if for any two points A and B
in S, the whole line segment AB lies entirely in S. It is easily shown that the intersection of
convex sets is also a convex set. The convex hull of a set T is defined to be the intersection of
all convex sets containing T . It is in fact the smallest convex set containing T .

Problem Let S be a set of five distinct points in the plane.

Show there exist three points A, B and C of S such that

108◦ ≤ ∠ABC ≤ 180◦ .

Solution It is tempting to oversimplify by asserting that the five points form a pentagon
and so one of its interior angles is at least 108◦ . Although this is true, that interior angle
might also be more than 180◦ . This is where the convex hull becomes useful.
Consider the perimeter of the convex hull. If it contains all five points, then we may use the
above argument since all interior angles are at most 180◦ .

If the convex hull is a quadrilateral, divide the quadrilateral into two triangles as shown. The
fifth point must be inside one of those triangles. Using line segments, join that fifth point to
the vertices of the triangle in which it lies. Since these segments define three angles around
the fifth point, none of which exceeds 180◦ , and whose sum is 360◦ , it follows that one of
those angles is between 120◦ and 180◦ .
If the convex hull is a triangle, again we find a triangle with a point (in fact two points) in its
interior leading to an angle between 120◦ and 180◦ .
If the convex hull is a segment, then all points are collinear and so three of them form a 180◦
angle.

16.7 Euler’s formula

Euler’s formula was discussed in sections 14.10 and 14.11. Recall that for convex polyhedra
Euler’s formula tells us
V − E + F = 2,
where V , E and F are the respective numbers of vertices, edges and faces. Recall further
that this is also valid for planar graphs remembering that we count the infinitely large face
on the outside as one of the faces.

Problem The square ABCD contains n points P1 , P2 , . . . , Pn in its interior such that no
three of the n + 4 points A, B, C, D, P1 , . . . , Pn are collinear.
16.8 Pigeonhole principle 291

(a) Show that it is possible to subdivide the square into triangles in such a way that the
vertices of each triangle are among the n + 4 given points.
(b) Show that the number of resulting triangles in (a) is always the same no matter how
the subdivision is carried out.

Solution

(a) It is easy to establish using induction that a subdivision is always possible. Indeed each
extra point would land inside some triangle. Subdividing this triangle into three further
triangles by joining the interior point to the three vertices loses the original triangle
but creates three smaller triangles. This construction yields 2n + 2 triangles. 

Note that such a proof by induction that all subdivisions as described in the problem lead
to the same number of triangles is faulty. This is because not all such subdivisions can be
constructed inductively in this way from subdivisions with fewer triangles.11 So instead for
part (b) we resort to Euler’s formula along with a counting argument.

(b) Let T be the number of triangles. Then F = T + 1 because of the infinitely large outside
face that has four sides. Each edge belongs to two faces. Since each triangle has three
edges and the outside face has four edges we have
3T + 4
E= .
2

Substituting this into Euler’s formula where V = n + 4 yields

3T + 4
n+4− + T + 1 = 2.
2
Thus T = 2n + 2. 

16.8 Pigeonhole principle

Problem Six points are given inside an equilateral triangle of area 4.
Prove that among the nine points which include the three vertices of the triangle and the six
given points, three of these form a triangle of area at most 1.

Solution Divide the triangle up into four equilateral triangles of area 1. Since we have nine
points in total, by the pigeonhole principle at least three of these points lie inside or on the
boundary of one of these four triangles and thus define a triangle of area at most 1.
4
In fact more is true! We can sharpen the result from 1 to 13 as follows.
Suppose that we place one of the six points inside the triangle, and use this point to subdivide
the triangle into three smaller triangles. Next place a second point of the six points. This
will fall inside one of the three smaller triangles and we can use our second point to subdivide
this smaller triangle into three even smaller triangles, making five triangles in all. Continuing
in this fashion placing one point at a time until all six points are placed results in the original
triangle being subdivided into 13 triangles.
4
Thus one of these triangles has area at most 13 .
11 There are many examples of this. Can you find one?
292 16 Combinatorial geometry

16.9 Colouring
Questions about colouring points in the plane often use the pigeonhole principle.

Problem Let S be a disc. The points of S are painted in finitely many colours.
Show that for every n ≥ 3 there exist infinitely many congruent polygons with n sides
contained in S such that all of them have their vertices painted in the same single colour.

Solution This problem initially seems quite daunting. The number of colours and the value
of n are allowed to be quite large. The best way to start is to examine simple cases. We
begin with the simplest case where we have only two colours and our congruent polygons are
triangles.
Consider any regular pentagon P , lying entirely in the interior of S. By the pigeonhole
principle three of the vertices of P have the same colour. However, there are infinitely many
disjoint translates of the set of five vertices of P which also lie in S. By the preceding
argument each of these contains a monochromatic triangle. Thus we have infinitely many
such monochromatic triangles.
But the triangles we are considering only come in one of two congruence types. Thus applying
the infinite pigeonhole principle, one of these congruence types contains infinitely many
monochromatic triangles.
We now have infinitely many congruent monochromatic triangles. Finally, a second application
of the infinite pigeonhole principle allows us to conclude that one of the two colours contains
infinitely many such congruent monochromatic triangles.
We leave it to the reader to work out how to solve the problem in its full generality as stated.
For m colours, the issue is basically to find a number k so that whenever P has k vertices
there are always n vertices of P of the same colour.

A couple of comments are in order here. First, P did not have to be regular. In the case
P is not regular, we have perhaps up to 53 = 10 different congruence types of triangle for

each translate of P . But that is still finitely many. Second, why did we choose P to be a
pentagon? The answer is that a pentagon has enough vertices so that if we colour them using
two colours, a monochromatic triangle always appears. So P could have had any number of
vertices greater than or equal to 5 and the proof would still work.
 17
Appendices

17.1 How do complex numbers work?

√ √
‘But −1 is not a real number. It can’t exist!’ The number i = −1 is not real in the sense
that it is not on the real number line. However, we represent it geometrically by a point one
unit directly above the real number 0.

Basic arithmetic
Computations with numbers involving i can be done by treating i as an unknown quantity,
but replacing i2 with −1 every time it occurs.1
Complex numbers include ones such as 2 − 3i. This lies 3 units directly below the number 2.
In fact considering all expressions of the form a + bi (a, b ∈ R), we fill up the entire plane.
This is called the complex plane.
Addition and subtraction are very easy to do. For example,

(2 − 3i) − (1 − 7i) = 1 + 4i.

Multiplication is almost as easy. For example,

(2 − 3i)(1 − 7i) = 2 − 17i + 21i2 = −19 − 17i,

since i2 = −1.
Division is easy once you think of rationalising the denominator. For example,

2 − 3i (2 − 3i)(1 + 7i) 2 + 11i − 21i2 23 + 11i 23 11

= = = = + i.
1 − 7i (1 − 7i)(1 + 7i) 1 − 49i2 50 50 50

One of the amazing things about complex numbers is the fundamental theorem of algebra.
This states that any non-constant polynomial with complex coefficients has at least one
complex root.
1A
√
similar thing is done when √
you are working with surds such as 2. You treat it as an unknown quantity
except that you can replace ( 2)2 with 2.
294 17 Appendices

√
To deal with quantities like
√ −1√we have to go beyond the real numbers and invent i. However,
for the quantities like i and −i we don’t need to invent any more numbers. They are
roots of the polynomials x2 − i = 0 and x2 + i = 0, respectively. The fundamental theorem of
algebra guarantees that both these polynomials have roots. In fact you might like to check by
hand that
√ √
   
1+i i−1
i=± √ and −i = ± √ .
2 2

So far we have considered complex numbers to be of the form a + bi (a, b ∈ R). This is called
the Cartesian form. The real numbers a and b basically tell us the x-coordinate and the
y-coordinate of the complex number so we can plot it in the complex plane if we wish.

Polar form
There is another form that is very useful called the polar form. Take a non-zero complex
number z = a + bi in Cartesian form. Draw a line segment from the number 0 to the number
√ z
in the complex plane. Using Pythagoras’ theorem, the length r of this segment is r = a2 + b2
and the angle θ the segment makes when measured anticlockwise from the positive x-axis
satisfies a = r cos θ and b = r sin θ.

bi z = a + bi

θ
a

The elements r and θ make up the polar form and we can write

z = r(cos θ + i sin θ).

This is often abbreviated as

z = r cis θ.
The number r is called the magnitude and the angle θ is called the argument of the complex
number. We use the notation r = |z| and θ = arg z. Note that θ is defined modulo 360◦ .
A marvellous property of the polar form is that it behaves so neatly under multiplication.
When two complex numbers are multiplied together, their magnitudes multiply and their
arguments add. Specifically, if

z1 = r1 cis θ1 and z2 = r2 cis θ2 ,

then
z1 z2 = r1 r2 cis(θ1 + θ2 ).
This property is not at all obvious but it is most important. If you want to try and prove it
(it’s not that hard), you might find the compound angle formulas for sin(x + y) and cos(x + y)
helpful.
17.1 How do complex numbers work? 295

A spectacularly amazing result is the following formula2 due to Euler.

eiθ = cos θ + i sin θ

Here θ is measured, not in degrees, but in radians.3 Once you know this amazing result it is
very easy to prove the rule for multiplying complex numbers in polar form. In fact you can
also prove the trigonometric compound angle formulas from this!
Here is an example of the power of the polar form. Suppose we are searching for complex
numbers z such that
z 3 = 1,
that is, cube roots of 1. Write z = reiθ . Then we have

r3 e3iθ = 1.

The polar form for 1 has |1| = 1 and arg(1) = 0. Thus we have r3 = 1 and 3θ = 0 and so
r = 1 and θ = 0.
However, we have missed something, namely, that the arg function is defined modulo 2π (in
radians). Thus we also need to check 3θ = ±2π, ±4π, ±6π, . . .. These lead to just two other
values (modulo 2π) namely, θ = 2π 4π
3 and θ = 3 .
So we have not one but three cube roots of 1, two of which are non-real complex numbers. If
you plot all three cube roots in the complex plane, you will see that they form the vertices
of an equilateral triangle inscribed in the unit circle. As a curiosity try computing their
Cartesian forms.
In general if n is a positive integer, the n roots of z n = 1 form the vertices of a regular n-gon
inscribed in the unit circle.

2 You might be wondering, ‘How do I even compute e to the power of i?’ Unfortunately, a discussion of what
this even means would take us too far afield. An internet search could be most illuminating.
3 Note that 360 degrees is equal to 2π radians.
296 17 Appendices

17.2 Function notation

A few examples should suffice for you to get the hang of it. If we write

f : N+ → R,

this means that f is a function defined for all elements of N+ (the set of positive integers)
and taking values in the set R (the set of real numbers). An example of such a function is
f (x) = −x.
The set that appears immediately after the colon (N+ in our example) is called the domain
of the function.
The set that appears after the arrow (R in our example) is called the codomain.
The set of values that f actually achieves (in our case it is the set of negative integers) is
called the image.
Note that if we tried to define the function f : N+ → N+ satisfying f (x) = −x, then no such
function exists! Indeed changing the domain or codomain can drastically alter things.
17.3 Directed angles 297

17.3 Directed angles

One of the most effective ways to deal with diagram dependence is through the use of directed
angles. This often helps avoid a large number of case distinctions.
For two lines m and n, the directed angle between them is denoted by ∠(m, n). It is the
angle by which one may rotate m anticlockwise to obtain a line parallel to n. Directed angles
have the following properties. Each of these properties is quite straightforward to prove.

All directed angles are considered modulo 180◦ . In algebraic terms this means that
∠(m, n) = ∠(m, n) + 180◦ .
∠(n, m) = −∠(m, n).

For any line k we have ∠(m, n) = ∠(m, k) + ∠(k, n).

Points A, B, C are collinear if and only if ∠(AB, AC) = 0◦ .
Points A, B, C, D are concyclic if and only if ∠(AD, AC) = ∠(BD, BC).4

If A, B, C are three points on a circle Γ and m is a line through A, then m is tangent

to Γ if and only if ∠(m, AB) = ∠(AC, BC).5

As an example, here is how you can prove part of the generalised pivot theorem as shown in
section 6.5. We shall prove that if D, E and F lie on lines BC, AC and AB, respectively,
and if P is the second intersection point of circles AEF and CDE, then P also lies on circle
DBF .

Solution Since A, E, P and F are concyclic, as are C, E, P and D, we know

∠(F P, F A) = ∠(EP, EA) and ∠(EP, EC) = ∠(DP, DC).

But EC and EA both define the same line, thus we may deduce that

∠(F P, F A) = ∠(DP, DC).

It only remains to notice that F A and F B define the same line as do DC and DB. Thus

∠(F P, F B) = ∠(DP, DB),

which means that D, B, F and P all lie on the same circle.

Of course in reality you would have solved this question by doing an angle chase on the
diagram. But in the write-up, using directed angles helps you to avoid having to deal with
any case distinctions, such as what happens if P is outside of triangle ABC or if any of the
points D, E or F are on the extensions of the sides of the triangle.

4 This one is especially useful because depending on the order of A, B, C, D around the circle you either have
the classic ‘bow tie’ configuration (angles standing on the same arc) or you have that opposite angles in a
convex quadrilateral add to 180◦ .
5 This is a limiting case of the preceding property.
298 17 Appendices

17.4 Some useful triangle formulas

In triangle ABC, let a, b, c stand for the lengths of the sides opposite A, B, C, respectively.
Similarly, let α, β, γ be the measures of ∠CAB, ∠ABC, ∠BCA, respectively. Additionally,
let s = a+b+c
2 be the semiperimeter, r the inradius, R the circumradius and 4 the area of
the triangle.

Area Formulas

1
4= × base × height
2
= rs
abc
=
4R
p
= s(s − a)(s − b)(s − c) (Heron’s formula)
1
= bc sin α.
2

Formulas involving a cevian

b d c

C x D y B

If AD is the internal angle bisector, then we have the angle bisector theorem
b x
=
c y
as well as
ab
x=
b+c
ac
y=
b+c
d2 = bc − xy.
If AD is the median, then we have Apollonius’ theorem.
 a 2
2d2 + 2 = b2 + c2
2
If AD is a general cevian, then we have Stewart’s theorem.
b2 y + c2 x = a(xy + d2 )
 Index

addendo, 112, 216 Ceva’s theorem, 111, 112

adjacent, 247 trigonometric version, 113
affine transformation, 134 cevian, 66, 111, 112, 298
alternate segment switch, 82 Chinese remainder theorem, 34, 35
alternate segment theorem, 59 chromatic number of the plane, 282
altitude, 63, 90 circumcentre, 63, 91
extension to circumcircle, 67 circumcircle, 66, 85, 94, 102
AM–GM inequality, 191, 216, 222 closed set, 286
AM–HM inequality, 194 codomain, 296
angle bisector, 63, 76 coefficient, 149
extension to circumcircle, 66 collinear points, 77, 78, 83, 86–88, 91, 93,
angle bisector theorem, 73, 215, 216, 298 98, 101, 102, 108
angle chasing, 59 angle condition, 108, 297
Apollonius’ theorem, 298 dilation property, 131
arithmetic mean, 191, 195 Menelaus’ theorem, 109
associative trick, 178 Pascal’s theorem, 122
ratio condition, 83
base-n, 31
colouring invariants, 270
Beatty’s theorem, 26
combinatorial games, 275
bicentric quadrilateral, 107
copycat strategy, 277
bijection, 231, 232
bijective function, 176, 177, 180 pairing strategies, 277
binomial identities, 229 position analysis, 276
binomial theorem, 230 strategy stealing, 278
bipartite graph, 250 combinatorial geometry, 281
bounded set, 286 combinatorial interpretations, 231
bounding arguments, 44 combinatorial reciprocal principle, 241
butterfly theorem, 103 combinatorics, 225
bijections, 231
Carmichael number, 23, 38 binomial identities, 229
case bash, 12, 44 double counting, 235, 240
Cauchy’s functional equation, 173, 174 inclusion–exclusion principle, 234
Cauchy–Schwarz inequality, 193, 204 injections, 237
centroid, 62, 91, 140 overcounting, 228

299
300 Index

pigeonhole principle, 233 diagram dependence, 68, 108, 109, 297

supermarket principle, 232 difference of perfect squares, 29
complete bipartite graph, 250 dilation, 126, 131
complete graph, 250 Diophantine equation, 39
complete quadrilateral, 93 direct proof, 5
completely multiplicative function, 183 directed angle, 297
completing the square, 5 directed graph, 248
complex numbers, 137, 160, 293 directed length, 109
Cartesian form, 294 discrete intermediate value theorem, 288
comparing algebra and geometry, 137 division algorithm, 29
complex plane, 293 polynomial, 154
polar form, 294 divisor, 43
similar triangles, 144 polynomial, 149
composition of spiral symmetries, 126 domain, 296
concave function, 196 double counting, 235, 240, 255
concurrent circles, 92, 116
concurrent lines, 88, 95, 97, 110 edge, 243
Ceva’s theorem, 111 Eisenstein’s criterion, 151
isogonal conjugates, 111 elementary symmetric function, 157
radical axis theorem, 118 ellipse, 119, 219
concyclic points, 77–79, 85, 86, 90, 100, Euclid’s algorithm, 29
113 Euler line, 91, 131
angle condition, 113, 297 Euler phi function, 20, 36, 46
cyclic quadrilateral, 59, 60 Euler trail, 251
power of a point, 117 Euler’s complex exponential formula, 295
Ptolemy’s inequality, 216 Euler’s formula (graph theory), 256, 258,
conjugate 290
complex conjugate pairs, 160 Euler’s inequality, 223
surds, 27 Euler’s theorem (geometry), 223
conjugate root theorem, 160 Euler’s theorem (number theory), 36, 46
connected graph, 249 excentre, 63, 85, 97–100, 102
constant term, 149 excircle, 84, 98, 100
contradiction, 8, 285 exhaustion, 12
contrapositive, 7 extremal principle, 16, 254, 285
converse, 5
convex function, 196 face, 256
convex hull, 289 factor
copycat strategy, 277 polynomial, 149
cycle, 249 factor theorem, 154
cyclic hexagon, 95 factorisation, 43
cyclic quadrilaterals, 59 difference of perfect powers, 163
cyclic sum notation, 200 difference of perfect squares, 29
cyclotomic polynomials, 51 into primes, 24
polynomials (mod p), 163
dealing with digits, 25 Fagnano’s problem, 212
degree Fermat point, 120, 217, 219
indegree, 248 Fermat’s little theorem, 35
of a polynomial, 149 Fibonacci sequence, 2, 12, 30
of a vertex, 247 fixed point, 180
outdegree, 248 floor function, 26
derangement, 235, 236 function
Desargues’ theorem, 88 bijective, 176, 177
Index 301

codomain, 296 path, 253

completely multiplicative, 183 planar, 256
concave, 196 tree, 249
convex, 196 vertex, 243
domain, 296 walk, 253
fixed point, 180 graph theory, 243
image, 296 Euler’s formula, 256
injective (one-to-one), 176 handshaking lemma, 247
inverse, 177 greatest common divisor, 29, 47
involution, 179 greedy algorithm, 31, 272
monotonic, 43
somewhere versus everywhere, 182 Hölder’s inequality, 204
surjective (onto), 176 handshaking lemma, 247, 255
functional equation, 169 harmonic mean, 194, 195
Cauchy’s functional equation, 173 harmonic quadrilateral, 101
fundamental theorem Helley’s theorem, 282
of algebra, 155, 160, 293 Heron’s formula, 223, 298
of arithmetic, 24, 183 homogeneous inequality, 201
of combinatorial games, 275
of symmetric polynomials, 157 identity theorem, 154
if and only if, 6
games, 261 image, 296
Gauss’ lemma, 162 incentre, 63, 85, 97–100, 102
gcd, 29, 31, 47 incidence geometry, 105
gcd trick, 37, 51 incident, 247
generator, 37 incircle, 63, 84, 89, 96, 98, 100, 222
geometric inequalities, 207 incircle substitution, 198, 222
geometric mean, 191, 195 inclusion–exclusion principle, 234
geometric transformations, 125 indegree, 248
geometry induction, 9, 288
area, 70 inductive hypothesis, 9
combinatorial, 281 strong, 11
constructions, 64 inequalities, 187
diagram dependence, 108 addition and multiplication, 198
important configurations, 75 cyclic sum notation, 200
incidence, 105 expand and conquer, 200
need for good diagrams, 53 geometric, 207
plane geometry, 53 homogeneous, 201
reverse reconstruction, 68 majorisation, 202
transformations, 125 reverse engineering, 198
trigonometry, 69 substitutions, 194, 197
graph, 243 symmetric sum notation, 200
bipartite, 250 useful [i, j, k] notation, 203
complete, 250 weighted, 203
complete bipartite, 250 inequality
connected, 249 AM–GM, 191
cycle, 249 AM–HM, 194
directed, 248 Cauchy–Schwarz, 193, 204
edge, 243 Hölder’s, 204
Euler trail, 251 Jensen’s, 196
face, 256 Muirhead’s, 202
isomorphic, 243 power means, 195
302 Index

quadratic mean, 195 Muirhead’s inequality, 202

rearrangement, 191, 193 multiplicity, 155
squares are non-negative, 190
triangle, 210 Newton–Gauss line, 93
weighted Jensen’s, 204 nine-point centre, 91, 132
weighted power means, 203 nine-point circle, 90, 114, 132, 223
infinite descent, 49 number invariants, 267
infinite pigeonhole principle, 16, 292 number theory, 19, 39
infinitude of primes, 8, 37
one-to-one function, 176
injective function, 176, 237
onto function, 176
integer polynomial, 149
orthocentre, 63, 86, 90, 91
integer polynomials, 158
outdegree, 248
intermediate value theorem, 288
overcounting, 228
intersecting chords theorem, 117
invariants, 261 pairing strategies, 277
as cost, 273 Pappus’ theorem, 87, 122
colouring, 270 parallel lines, 99
modular arithmetic, 269 parametrisation, 215
monovariants, 272 parity
number invariants, 267 modulo 2 invariants, 268
parity, 268 permutation, 274
permutation parity, 274 parity pattern, 234
inverse function, 177 Pascal’s theorem, 87, 122
involution, 179,√232 Pascal’s triangle, 230
irrationality of 2, 8 path, 253
irreducible polynomial, 162 perfect squares, 46
isogonal congugates, 111 permutation
isomorphic, 243 inversion, 274
isoperimetric inequalities, 221 parity, 274
transposition, 274
Jensen’s inequality, 196
perpendicular bisector, 63, 76
known diagram, 72, 75 perpendicularity
condition for, 81
Lagrange interpolation formula, 166 perspective triangles, 88
leading coefficient, 149 perturbation, 287
locus, 105, 118, 119, 219 pigeonhole principle, 13, 15, 16, 25, 233,
logic and deduction, 5 250, 291, 292
pivot theorem, 77, 116
majorisation, 202 planar graph, 256
median, 62, 94, 97 plane geometry, 53
Menelaus’ theorem, 109 Platonic solid, 245, 282
midpoint theorem, 70, 114 polyhedron, 257
mixtilinear circle, 102 polynomial, 149
modular arithmetic, 34, 46 coefficient, 149
polynomial, 45, 163, 166 constant term, 149
modular arithmetic invariants, 269 cyclotomic, 51
modulo n colouring, 271 degree, 149
Monge’s theorem, 132 division algorithm, 155
monic, 50, 149 divisor, 149
monotonicity, 43 factor, 149
monovariants, 272 factor theorem, 154
Morley’s theorem, 57 identity theorem, 154
Index 303

integer, 149, 158 rearrangement inequality, 191, 193

irreducible, 162 recursion, 238
leading coefficient, 149 reducible, 162
modular arithmetic, 45, 163, 166 reduction of variables, 48
monic, 50, 149 reflection principle, 210
rational root theorem, 158 remainder theorem, 154
reducible, 162 reverse reconstruction, 68
remainder theorem, 154 root, 149, 167
root, 149, 167 roots of unity, 146
symmetric, 157 rotation, 126, 130, 142
unique factorisation (mod p), 164
upstairs–downstairs, 163, 165 similar triangles
Vieta’s formulas, 156 complex numbers, 144
zero, 149 similar switch, 79
polynomial modulus, 45, 166 Simson line, 86, 109
position analysis, 276 Simson’s theorem, 86, 109
power means inequality, 195 spiral symmetry, 79, 125, 126, 133
power of a point, 117 complex number description, 125, 126
powers of two, 28 composition, 126
primes subgroups of, 126
factorisation into, 24, 183 square root, 27
infinitude of, 8, 37 squares are non-negative, 190
primitive root, 37 Stewart’s theorem, 298
proof, 1 strategy stealing, 278
by contradiction, 8, 285 strong induction, 11
by exhaustion, 12 inductive hypothesis, 12
by induction, 9 substitutions
direct proof, 5 functional equations, 175
Ptolemy’s inequality, 216 inequalities, 194, 197
Pythagoras’ theorem, 6 number theory, 48
polynomial, 160
quadratic discriminant, 46, 47 supermarket principle, 232
quadratic equations, 46, 50 surjective function, 176
quadratic formula, 5 symmedian, 94
quadratic mean, 195 characterisations of, 94
quadrilateral symmetric polynomial, 157
bicentric, 107 symmetric sum notation, 200, 202, 203
complete quadrilateral, 93 symmetry
conditions for incircle to exist, 89, 96 functional equation, 179
cyclic, 59 geometry, 66
harmonic, 101
midpoints of diagonals, 93 tangency, 119, 219
midpoints of sides, 70, 140 tangent circles, 77, 80, 94, 96, 102
Newton–Gauss line, 93 telescoping, 18
Tower of Hanoi, 4
radical axis, 78, 118 transformation geometry, 125
bisects common tangent, 80 transformations, 125
radical axis theorem, 78, 118 affine, 134
rational root theorem, 158 dilation, 126, 131
ratios geometric inequalities, 213
addendo, 112, 216 reflection, 210
condition for collinearity, 83 rotation, 126, 130, 142
304 Index

spiral symmetry, 125, 126, 133 upstairs–downstairs, 163, 165

translation, 126, 130
translation, 126, 130 vertex, 243
Vieta jumping, 50
tree, 249
Vieta’s formulas, 50, 156
triangle centres, 62, 85, 91, 99, 100, 102
triangle formulas, 222, 298 walk, 253
triangle inequality, 17, 210, 218 weighted Jensen’s inequality, 204
triangles weighted power means inequality, 203
perspective, 88 well-ordering, 184
WLOG, 44, 45
similar, 79, 144
trigonometry, 69 Zeckendorf’s theorem, 12
geometric inequalities, 214 zero, 149
version of Ceva’s theorem, 113 zero polynomial, 154