_OXFORD SCIENCE PUBLICATIONS

THE MATHEMATICAL
OLYMPIAD
HANDBOOK
AN INTRODUCTION TO
The Mathematical Olympiad Handbook
«■
 The Mathematical
Olympiad Handbook
An introduction to problem solving
based on
the first 32 British Mathematical Olympiads
1965-1996

A. GARDINER
School of Mathematics
University of Birmingham, UK

Oxford New York Tokyo

OXFORD UNIVERSITY PRESS
1997

1 REN i uNi VtiOi *

PETERBOROUGH, ONTAR
 3 ci'5 /<?97

Oxford University Press, Great Clarendon Street, Oxford 0X2 6DP

Oxford New York
Athens Auckland Bangkok Bogota Bombay Buenos Aires
Calcutta Cape Town Dar es Salaam Delhi Florence Hong Kong
Istanbul Karachi Kuala Lumpur Madras Madrid Melbourne
Mexico City Nairobi Paris Singapore Taipei Tokyo Toronto Warsaw
and associated companies in
Berlin Ibadan

Oxford is a trade mark of Oxford University Press

Published in the United States

by Oxford University Press Inc., New York

© A. Gardiner, 1997

All rights reserved. No part of this publication may be

reproduced, stored in a retrieval system, or transmitted, in any
form or by any means, without the prior permission in writing of Oxford
University Press. Within the UK, exceptions are allowed in respect of any
fair dealing for the purpose of research or private study, or criticism or
review, as permitted under the Copyright, Designs and Patents Act, 1988, or
in the case of reprographic reproduction in accordance with the terms of
licences issued by the Copyright Licensing Agency. Enquiries concerning
reproduction outside those terms and in other countries should be sent to
the Rights Department, Oxford University Press, at the address above.

This book is sold subject to the condition that it shall not,

by way of trade or otherwise, be lent, re-sold, hired out, or otherwise
circulated without the publisher’s prior consent in any form of binding
or cover other than that in which it is published and without a similar
condition including this condition being imposed
on the subsequent purchaser.

A catalogue record for this book is available from the British Library

Library of Congress Cataloging in Publication Data

Gardiner, A. (Anthony), 1947-
The Mathematical Olympiad Handbook: an introduction to problem
solving based on the first 32 British mathematical olympiads
1965-1996 / A. Gardiner.
1. Mathematics-Problems, exercises, etc. 2. British Mathematical
Olympiad. I. Title.
QA43.G345 1997 510'.76~dc21 97-17647
ISBN 0 19 850105 6

Typeset by Technical Typesetting Ireland

Printed in Great Britain by Biddles Ltd., Guildford & King’s Lynn
... a mathematical problem should be difficult in order to
entice us, yet not completely inaccessible lest it mock our
efforts. It should be a guidepost on the tortuous path to
hidden truths, ultimately rewarding us by the satisfaction
of success in its solution.
David Hilbert, 1900
 Digitized by the Internet Archive
in 2019 with funding from
Kahle/Austin Foundation

https://archive.org/details/mathematicalolym1997gard
 Preface
The supply of problems in mathematics is
inexhaustible; no sooner is one problem solved
than numerous others spring forth in its place.
David Hilbert, 1990

This is unashamedly a book for beginners. Unlike most Olympiad problem

books, my aim has been to convince as many people as possible that
Mathematical Olympiad problems are for them and not just for some bunch
of freaks. I have tried never to use one slick, unmotivated step where three
down-to-earth steps would do just as well. Once you are convinced that these
problems are do-able and are worth the effort required to solve them, you
will be ready to move on to other books of Olympiad problems—some of
which are listed in the section ‘Some books for your bookshelf.

Who is the book for?

In mathematics, as in music and sport, there are many youngsters who are
capable of performing at a much higher level than is required by the ordinary
school curriculum. While many youngsters have the necessary potential, the
final ability is not God-given: it has to be developed—and that requires effort
and commitment. Such students may remain in their peer groups, but need
higher goals to aim at. This book is for such students aged 15 or 16 and
above.
The book presumes an initial commitment on the part of you—the reader.
At first, this commitment need go no further than a basic technical compe¬
tence, combined with a willingness to struggle on when faced with a challeng¬
ing mathematical problem. But sooner or later you will have to make the
effort to follow up the mathematical ideas which lie behind these problems,
and start reading other books to master some unfamiliar mathematics on
your own. Reader-friendly mathematics books can be hard to find: to guide
you, the section ‘Some books for your bookshelf contains a short list of books
which you may find helpful, together with a very brief summary of the
mathematics covered by each one.

What is in this book?

The main part of this book contains problems from the first 32 British
Mathematical Olympiad (BMO) papers 1965-96, starting with 1996 and
viii Preface

working backwards. This is followed by a section in which the problems are

considered in turn, with each problem being followed by a page or so of text.
The first part of this text should be read only after you have had a good go at
the problem for yourself. You should then use these fresh ideas for a second
attack on the problem. Only later—whether or not you have succeeded in
solving the problem—should you work through the whole ‘Outline solution’.
If you failed to solve the problem, the text should convince you that there is
at least one ‘obvious’ line of attack which you could have discovered for
yourself. If on the other hand you managed to solve the problem on your
own, the text may still open your eyes to some things you never thought of.
However, the text itself does not present a complete solution. To obtain a
complete solution you must fill in for yourself the details in the ‘obvious’
approach outlined in the text.
Most of the problems in this book can be solved in many different ways.
The first correct solution that one finds to a problem is almost never the
simplest. It is only after one has succeeded in solving a problem that one can
sometimes look back and see that there is in fact a much easier solution—if
only one looks at things in the right way!
The papers from the years 1965-74 are different in style, in that each
contains ten or eleven (not necessarily original) problems—including the
occasional mechanics problem. I have included the statements of the prob¬
lems both for completeness and because they include many nice examples;
however, I have not provided any discussion, hints, or outline solutions for
these problems.
Most Olympiad problems do not require advanced mathematics for their
solution. However, they do require familiarity with some basic ideas which
receive little attention in most secondary school curricula. The section ‘A
little useful mathematics’ provides a brief introduction to the simplest and
most basic topics. You may find it helpful to work through this section for a
first time before tackling the problems in the rest of the book. Later, as you
meet problems that use some of these ideas, you may need to work through
some parts of the section in more detail.

Why was the book written?

Textbook exercises and examination questions are not meant to be particu¬

larly demanding: textbook exercises are intended to help all students practise
some standard technique, and exam questions are supposed to give ordinary
students an opportunity to show what they have learned. Thus both these
types of problems should be relatively straightforward for the large number of
students who are capable of performing at a higher level. This does not mean
they should be skipped! Indeed, it is perhaps more important for able
 Preface IX

students than for others that they receive a solid grounding in basic tech¬
nique. But this is only a beginning: such students need a regular diet which
goes beyond mere textbook exercises and examination questions.
This book should be especially useful to those who are preparing to take
part in Olympiad-type competitions. Taking part in competitions, and working
through old competition questions, provides one very valuable source of
extension and enrichment. Young musicians and sportsmen understand per¬
fectly well that the harder one practises or trains, the more likely one is to
benefit from performing or competing at a high level. In contrast, mathemat¬
ics competitions are frequently tackled without any preparation at all! This
can result in an experience that turns out to be memorable for all the
wrong reasons; instead of catching an exciting glimpse of new mathematical
horizons, the poorly prepared student can sometimes be rendered helpless by
a kind of mathematical stage-fright.
Take-home competitions, which allow students one to three weeks to solve
five or six problems, can sometimes get round this lack of preparation. Faced
with take-home problems that are sufficiently accessible, talented but poorly
prepared students can perform creditably and can learn a lot by working hard
and training ‘on the job’. (Examples of this approach working successfully on
a large scale in the UK include the Scottish Mathematical Challenge, the
Merseyside Challenge, and the Birmingham-based BUMMPS and Junior
PEST.) However, it is unreasonable to expect talented but poorly prepared
students to pull themselves up by their own bootstraps in a timed event which
allows them just one to three hours to solve five or six challenging, non¬
standard problems.
The fault here lies not with the timed, written format! Timed, written
papers have an invaluable role to play: they stress the importance of thinking
both creatively and quickly, and force students to present simple, clear
solutions. For administrative reasons, any large national competition is al¬
most bound to involve a timed paper. Moreover, the kind of questions
traditionally set in such competitions have tremendous potential for introduc¬
ing talented students to an accessible, yet completely new mathematical
world.
In the past it has been very difficult for ordinary schools to help their best
students prepare for, and hence benefit from, timed written papers of this
kind. As a result many very able students have performed in a way which may
sometimes have been discouraging rather than stimulating. This book should
help all schools and colleges to introduce able students in their last two or
three years at school to a new world of challenging mathematical problems.
The book is unlikely to succeed unless it also persuades teachers that these
Olympiad problems, though by no means easy, are more accessible than they
may have previously imagined. The experience of tackling and solving (or of
failing to solve and being annoyingly frustrated by) hard mathematical prob¬
lems is an eternal source of delight and disappointment. I hope that this little
X Preface

book will open up this world of delights and disappointments to many who
would otherwise have looked on bemusedly from the sidelines.
While I accept full responsibility for the structure of the book and the style
of the text, the contents owe much to many people. When the project was
first begun back in 1991, much encouragement and support was given by
Richard Higson and Chris Nash. The problems themselves are due to a
number of other people, whose individual contributions have been blurred by
the passing of time (in the early years, stalwart work was done by Margaret
and Walter Hayman; in the middle years, the mainstays were Robert Lyness
and David Monk; more recent papers stem from the work of the Problems
Group of the British Mathematical Olympiad Committee). The section ‘A
little useful mathematics’ had its roots in the original tentative draft by John
Deft of a booklet designed to help students prepare for the BMO. Dozens of
others have contributed in many different ways: in particular, Christopher
Bradley, Zad Khan, Adam McBride, David Monk, and Robert Pargetter
commented in detail on various drafts, and helped to reduce the number of
errors. I can only hope that each one of them will feel that this presentation
remains faithful, at least in part, to what they were trying to achieve. To all of
them, I express my thanks.
Finally, to those readers who are entering this mathematical world for the
first time, may I say ‘Bon voyage’.
 Contents

Problems and problem solving 1

How to use this book 3
Part I Background
A little useful mathematics 7
Introduction 7
1. Numbers 7
2. Algebra 11
3. Proof 18
4. Elementary number theory 24
5. Geometry 34
6. Trigonometric formulae 39

Some books for your bookshelf 41

Part II The problems
Background to the problem papers 53
32nd BMO, 1996 54
31st BMO, 1995 55
30th BMO, 1994 55
29th BMO, 1993 56
28th BMO, 1992 57
27th BMO, 1991 57
26th BMO, 1990 58
25th BMO, 1988 59
24th BMO, 1987B 60
23rd BMO, 1987A 61
22nd BMO, 1986 61
21st BMO, 1985 62
20th BMO, 1984 63
19th BMO, 1983 64
18th BMO, 1982 65
17th BMO, 1981 66
16th BMO, 1980 67
15th BMO, 1979 67
 xii Contents

14th BMO, 1978 68

13th BMO, 1977 69
12th BMO, 1976 69
11th BMO, 1975 71
10th BMO, 1974 72
9th BMO, 1973 74
8th BMO, 1972 76
7th BMO, 1971 78
6th BMO, 1970 79
5th BMO, 1969 81
4th BMO, 1968 82
3rd BMO, 1967 83
2nd BMO, 1966 84
1st BMO, 1965 86

Part III Hints and outline solutions

32nd BMO, 1996 89
31st BMO, 1995 95
30th BMO, 1994 99
29th BMO, 1993 103
28th BMO, 1992 109
27th BMO, 1991
115
26th BMO, 1990
122
25th BMO, 1988
128
24th BMO, 1987B
136
23rd BMO, 1987A
142
22nd BMO, 1986
152
21st BMO, 1985
162
20th BMO, 1984
168
19th BMO, 1983
177
18th BMO, 1982
181
17th BMO, 1981
188
16th BMO, 1980
193
15th BMO, 1979
196
14th BMO, 1978
202
13th BMO, 1977
207
12th BMO, 1976
213
11th BMO, 1975
219
Appendix The International Mathematical Olympiad:
UK teams and results, 1967-96
225
 Problems and problem solving
However unapproachable these problems may seem to us and
however helpless we stand before them, we have, nevertheless,
the firm conviction that their solution must follow by a finite
number of purely logical processes. [...]
This conviction of the solvability of every mathematical
problem is a powerful incentive to the worker. We hear within
us the perpetual call: There is the problem. Seek its solution.
You can find it by pure reason; for in Mathematics there is no
ignorabimus.
David Hilbert, 1990

David Hilbert (1862-1943) was one of the most outstanding mathematicians

of the modern era. At the International Congress of Mathematicians in Paris
in 1900, he presented 23 major research problems which he felt would be
important for the development of mathematics in the twentieth century.
These problems all seemed very hard, but in bringing them to the attention of
other mathematicians Hilbert felt that the need to stress that this should not
be used as an excuse to put off trying to solve them.
Hilbert’s judgement that these problems would play a central role in the
mathematics of the present century was remarkably accurate. But the most
interesting thing for us here is his rallying call: However unapproachable
these problems may seem at first sight, and however helpless we stand before
them, we have the firm conviction that their solution must be possible by
purely logical processes. There is the problem. Seek its solution. You can find
it by pure reason.
When faced with an unfamiliar and apparently very difficult mathematical
problem, there is always the temptation to imagine that it is too hard, that
progress towards a solution requires some trick or technique that we have not
yet learned, and that the solution is therefore beyond our powers. This
defeatist view is all the more plausible because it must sometimes be true! It
is nevertheless misguided. Let me try to explain why.
During the nineteenth century it became clear that the more scientists
discovered about nature, the more they realized how little they knew, and that
one could never hope to discover the whole truth. This realization was
summed up by one philosopher, Emil du Bois Reymond, in the phrase
‘ignoramus et ignorabimus’—ignorant we are and ignorant we shall remain.
This catch phrase certainly caught on! And as the new century dawned, David
Hilbert felt it important to state as clearly as he could that mathematics is
2 Problems and problem-solving

different. In mathematics, said Hilbert, we can tackle problems ‘with the firm
conviction that their solution must follow by a finite number of purely logical
processes’. As if to underline his assertion, one of his 23 outstanding prob¬
lems was solved almost immediately.
Hilbert was talking about mathematical research. However, his principle
applies just as well to solving Olympiad problems. When mathematicians
begin to explore a problem, they take it for granted that it must be possible to
make progress towards a solution using only the tools currently available. Of
course, they are sometimes wrong; but that is not the point. Mathematicians
know perfectly well that, strictly speaking, the assumption is irrational (in that
it cannot be justified logically, and is in general clearly false). But it is an
invaluable working hypothesis. Though strictly illogical, the assumption that
eveiy mathematical problem can be solved has justified itself so often in
practice that it has become a powerful conviction—one which is psychologi¬
cally invaluable each time we experience that feeling of helplessness when
trying to get to grips with a hard mathematical problem.
Faced with an unfamiliar mathematical problem, the mathematician—
whether young or old—is like someone with a hopelessly small bunch of
‘keys’ who is trying to open up some fiendishly difficult Chinese puzzle box.
At first glance the surface seems totally smooth, without a single visible crack.
If you were not convinced that it is indeed a Chinese puzzle box, and that it
can in fact be opened, you would soon give up. Knowing (or rather believing)
that it can be opened, you may be willing to keep searching until you
eventually begin to discern the slightest hint of a crack here and there. You
still have no idea how the pieces are meant to move, or which of your ‘keys’
may help you open up the first layer of the puzzle. But by trying the most
appropriate-looking keys in the most promising cracks, you eventually stum¬
ble on one that fits exactly, and the pieces begin to move.
In a good puzzle, success is never the result of pure chance. Indeed,
once one has discovered the way in, one often realizes that it should have
been obvious where to begin. But things usually become ‘obvious’ only in
retrospect. If the puzzle is unfamiliar and well-made, success may require
persistence, faith, and much time. Exactly the same is true of any good
mathematics problem. So never give up too easily, and always be prepared to
look back after solving a problem to see what you should perhaps have done
differently. That is how you learn.
 How to use this book
We must know, we will know.
David Hilbert, 1930

You have no choice but to tackle the problems in this book using the bunch
of ‘keys’, or mathematical techniques, that you already know—no matter how
limited they may be. It is also important to learn new tricks, and to revise old
ones, as you go along. Thus you may choose to begin by working carefully
through the section ‘A little useful mathematics’. But you must not let this
interfere with your basic hypothesis that every problem has to be solved using
only the techniques that you already know.
In this book, the text at the beginning of the outline solution for each
problem is meant to convince you that the problem could indeed have been
solved using only the techniques that you know. Do not read this text until you
have tried really hard to solve the problem for yourself .
Sometimes the text may not succeed in convincing you that you could have
solved the problem on your own—although even then it should show how,
using only the techniques that you know, you could have got further than you
did. On other occasions the text will only convince you that you need to learn,
or brush up, some bit of mathematics which is needed for that particular type
of problem. When this happens you should first look back at ‘A little useful
mathematics’. If that does not help, I suggest you consult one of the books
listed in ‘Some books for your bookshelf’.
As a guide I suggest that you allow yourself at least one hour to work on
each problem before consulting the relevant text in ‘Hints and outline
solutions’. Whenever possible you should give yourself time by ‘sleeping on
the problem’ before looking at the text the next day. If you make a habit of
this, you will discover the curious fact that, whenever you have been thinking
hard about a problem, all sorts of useful ideas come into your head just as
you begin to relax.
 Part I
Background
 A little useful mathematics
... what is clear and easy to understand
attracts us, what is complicated repels us.
David Hilbert, 1900

Introduction

The problems which are set in the British Mathematical Olympiad (BMO)
are not all that hard. However, they can seem rather hard if you are not
familiar with certain important ideas which are often skated over, or ignored
altogether, in school mathematics. This section introduces some of the most
basic of these ideas. If you work on it before tackling the rest of the book,
you will learn a lot of useful mathematics and will get far more out of tackling
the problems than if you simply try to solve the problems without any
preparation. In particular, make sure that you tackle all the exercises marked
with a single asterisk (*). (Harder exercises, which you may well find too
difficult on a first reading, are marked with a double asterisk (**).) Whenever
blanks are left in the text, you should make sure you know how to fill them in
before reading further.

1. Numbers

Our real number system KdQdZdN has several layers, like an onion. R
denotes the set of all real numbers; including, for example,

-6\, -3, l-y/2, 0, i/2, j, 3/l0, e, 1 +1/3, 666.

If we remove the irrational numbers (those numbers which cannot be ex¬

pressed as the ratio of two integers), we are left with the set <0 of all rational
numbers, or fractions; that is, numbers such as

1 61 -21
0,
16
-4, 5 > 2 > 1, 1.414, 29 > D3 •

Each rational number can be expressed in lowest terms as a quotient m/n,

where m and n are whole numbers having no common factors and n + 0; for
8 A little useful mathematics

example, 1.414= 1414/1000 = 707/500. If we take only those rational num¬

bers m/n with denominator n = 1, we obtain the set Z of integers'.

Z= {..., -4, -3, -2, -1,0,1,2,3,4,5,...}.

The positive integers

N = {1,2,3,4,5,6,7,8,...}
are often called the natural numbers.
Whenever we are working with numbers, the kind of operations which are
permissible, and the kind of statements and deductions which are valid, will
depend on which kind of numbers we are using.

(1.1) Every real number x has a negative —x, which is also a real number. If
x is a rational number, so is —x. If x is an integer, so is —x. But if x is a
natural number, — x is not a natural number.

(1.2) A real number x has an inverse \/x provided that x is non-zero. If x is

a (non-zero) real number, then 1 /x is also a real number.
(*) If x is a non-zero rational number, is 1 /x always a rational number?
(*) Suppose that x is a non-zero integer. When exactly is 1/x also an
integer?
If a problem is restricted to integers, one has to be very careful about writing
fractions (since the quotient of one integer by another will usually not be an
integer), or about taking square roots (since, for example, when x is an
integer, Vx" will usually not be an integer).
One also has to be careful about ‘cancelling’ the common factor V on each
side of an equation of the form ‘ax = ay’ to conclude V. x =y\ Cancelling is a
short way of ‘multiplying both sides of the equation by 1 /a’—and this is
possible only if a is non-zero. Thus, whenever one is tempted to cancel two
expressions that look equal, one must always think twice. For example, given
the equation ‘6(x - l)/x =x2 - T, it is an excellent move to factorize the
right hand side (RHS) to obtain ‘6(x - l)/x = (x - lXx + 1)’. It is then
tempting to cancel the common factor ‘x — T and conclude

‘.\ 6/x=x + 1
.’. x2 +x — 6 = 0, so x = 2 or x = —3’.
However, the cancellation is valid only ifx- 1 ¥= 0, so the above argument is
invalid and we have missed one solution. The cancellation must be rewritten
as:

‘6(x - l)/x =x2 - 1 = (x - l)(x + 1)

.-. either (a) x - 1 = 0, or (b) 6/x =x + 1;
.-. either (a) x = 1, or (b) x2 +x - 6 = 0, whence x = 2 or x = -3
 Numbers 9

(1.3) Suppose that x and y are any real numbers satisfying x-y = 0. If x ¥= 0,
then x has an inverse 1/x, so we can multiply both sides of x-y = 0 to obtain
y = (l/x)-0; that is, y = 0. This proves the important result that

x -y = 0 => either x = 0 or y = 0 .

(*) Find all real numbers t such that t(t + 2) = 3.

(Hint: Do not try to solve this directly. Instead, take everything to one
side; then factorize to obtain a product equal to zero; finally, use (1.3).)

If t can be any real number, one should never try to solve t2 = at - a directly
using the factorization t2 = a(t — 1). This factorization tells you almost noth¬
ing about t because a real number can be factorized on infinitely many
different ways: for example,

2 = 2-1 = (2/3)-3 = V2-t/2 =(v/3- lKv/3 + 1) = ... .

Instead, one must take everything to one side, factorize, and use

‘x-y = 0 => eit-er x = 0 or y = O’.

The warning in the previous paragraph depended on the fact that t could be
any real number. In contrast, if t and a are integers, then a factorization such
as t2 = a(t — 1) tells you a lot because t — 1 must then be a factor of t2\ But
two consecutive integers, such as t — 1 and t, always have a highest common
factor equal to 1, so there are only two possible ways in which t — 1 can divide
t2, namely, if t =_ or if t =_. [Whenever blanks are left in the text,
you should make sure you know how to fill them in.]
Another way of finding all integer solutions of the equation t2 = at - a is
to forget for the moment that t and a are integers, and to follow the advice
‘take everything to one side’. Thus we must find all (integer) solutions
of the equation t2 — at + a = 0. If we use the formula for solving quadratic
(see the Algebra section below), we see that the two real solutions are
t = [a ± \/(a2 - 4a)]/2. For 21 to be an integer, \l(a2 - 4a) must be an integer.
Hence a2 - 4a must be a perfect square. But a2 - 4a = (a - 2)2 - 4 is also 4
less than a perfect square. Hence the only possible solutions are a - 2 = 2
and a - 2 = -2; fortunately, in each of these cases the expression [a ±
]/(a2 - 4a)] is not only an integer, but is also even, so the expression for t has
an integer value and can be calculated in each case.
The above solution uses real number operations carefully and intelligently
(avoiding division unless the quotient is an integer, and avoiding square roots
unless the number happens to be a perfect square) to solve a problem about
integers. Sometimes this is impossible, or unnatural, so one must develop
special ways of working with integers alone—ways which depend only on
addition, subtraction, and multiplication. This leads to elementary number
theory—the study of divisibility (of one integer by another), of highest
10 A little useful mathematics

common factors (of pairs of integers), of properties of prime numbers, and of

prime factorization. We will come back to these topics in Section 4.

(1.4) Two natural numbers (in N) can be added or multiplied and the result is
always another natural number. In contrast, subtraction and division may give
answers outside IU Two integers (in Z) can be added, multiplied, or sub¬
tracted and the result is always in Z, but division will usually give an answer
outside Z. The set 0 of rational numbers and the set R of real numbers
admit all four operations (although ‘dividing by 0’ is forbidden).

(1.5) Once a unit of length has been chosen, the real numbers R can be made
to correspond precisely to the set of all points on the x-axis. These form a
continuum with no gaps. A real variable x (or t, or ...) can vary continu¬
ously, so we can take limits, draw graphs, use calculus, and so on. The rational
numbers 0 correspond to a subset of points on the x-axis. This subset of
rational points is very crowded, or dense, on the x-axis.

(*) Prove that between any two points a and b (a =£ b) on the x-axis there
are infinitely many rational points.

Nevertheless, the rational numbers leave masses of gaps—even if your eye

cannot see them!

(*) In Section 3 (on ‘Proof’) we prove that yj2 is not a rational number. Use
this fact to show that between any two rational numbers a and b (a ¥= b)
there are infinitely many irrational numbers.

(1.6) The rational number 1/3 is perfectly well-defined. However the decimal
for 1/3, namely ‘0.333333...(forever)’ (usually written as 0.3) is more compli¬
cated in that it only provides us with a sequence of better and better
approximations to 1 /3:

3 33 333 3333
To’Too’ iooo’ loooo”" ‘
No matter how far along this sequence one goes, the terms are never quite
equal to 1/3—though the error at the first stage is less than 1/10, at the
second stage is less than 1 /102, and so on. We say that this sequence
converges to 1/3, and that the sequence has limit equal to 1/3.
Some convergent sequences of rational numbers have a rational limit: for
example, the sequence

2 3 4 5
T’2’3’4”"
of rational numbers, the nth term of which is equal to (n 4- l)/n, tends to the
rational number 1: indeed, (n + l)/n = 1 + (1/n), so that the nth term is just
 Algebra. 11

‘1 /n’ more than 1, and the excess l/n ‘tends to 0’ as n gets larger. But most
convergent sequences of rational numbers have no rational limit. For
example, the sequence

14 141 1414
1? 10 ’100’ 1000’"'

of rational numbers, the nth term of which is given by the first n significant
figures in the decimal expansion of yfl, clearly tends to the irrational number
n.
(*) Prove that 0.9 = 1 and that 0.49 = 1/2. What simple fractions are 0.09,
0.109, and 0.4455 equal to? (Prove your assertions.)
(**) Prove that a real number written as a decimal is equal to a rational
number if and only if the decimal eventually recurs.

2. Algebra

One of the turning points in the development of modern science occurred

around 1600 when Galileo observed that if one wants to understand the
‘Book of Nature’, then one must learn the language in which that ‘Book’ is
written—namely mathematics. At the time, that language consisted chiefly of
number, geometry, and numerical relationships expressed in geometrical
form. Since then it has become increasingly clear that the true ‘language’ of
mathematics is algebra (a subject which only developed during the seven¬
teenth century).

2.1. Quadratic equations

If a, b, and c are any given real numbers with a # 0, then the quadratic
equation

ax2 + bx + c = 0
has solutions

—b ± ]/(b2 - 4ac)
x =---. (2.1)
2a

This formula can be used whether or not the quadratic can be easily
factorized. The expression (b2 - 4 ac) is called the discriminant: if the dis¬
criminant is negative, the equation has no real roots; if the discriminant is
zero, the equation has one (repeated) root; if the discriminant is positive, the
equation has two real roots.
12 A little useful mathematics

Real roots will in general be irrational. But if the coefficients are integers
or rationals, and the discriminant is a perfect square, the quadratic ax2 +
bx + c factorizes nicely and the roots are rational numbers. Conversely, if the
coefficients are integers or rationals and you know that one root x is rational,
then the discriminant must be a perfect square. (Why?)

(*) Show that it is impossible to find non-zero integers x and y satisfying

x2 +xy -y2 = 0.

The quadratic formula (2.1) works for arbitrary coefficients a, b, and c

(provided that a # 0). Its derivation is based on the following very useful
technique called completing the square.

(*) Fill in the missing coefficients in the following algebraic calculation:

ax2 + bx + c = a(x2 +_x +_) = a((x +_)2 +_)

:.ax2+bx + c = 0 <=> (x +_)2 +_=0 <=> {x+_)=+/_

<=> x = -_+ 'J_

2.2. Difference of two squares and generalizations

You will certainly need the familiar identity

a2 - b2 = (a - b){a +b). (2.2)

It is also important to know that this is the very simplest instance of an

infinite family of easy identities.

(*) Complete the factorization xn — 1 = (x - lXxn~1 + ... +1).

(*) Complete the factorization xn —yn = (x— yXxn~1 + ... +y”_1).

While the difference of two nth powers can always be factorized, the
situation for the sum of two nth powers is rather different. But do not be
misled by the fact that a2 + b2 does not factorize.

(*) Complete the factorization x3 + 1 = (* + IX...).

(*) Complete the factorization x3 + y3 = (x + y)(...).

(**) Complete the factorization ;c3 +y3 +z3 - 3xyz = (x +y + z\...).

(*) For which values of n can we factorize xn + 1 = (x + 1)(...) and

xn +yn = (x + y)(...)?

(*) Prove that 255 + 1 is divisible by 33.

(*) Prove that 19001"0 - 1 is divisible by 1991.

 Algebra 13

2.3. Function notation

One of the most important ideas in mathematics is the idea of a function. A

function is a rule which tells us how to operate on an ‘input’ number to
obtain an ‘output’ number. We can write a function in various ways: for
example,

f:x^>x2 + 3 or fix)=x2 + 3.

For this function /(5) = 28, because when we substitute 5 in place of x we get
the output 28.
You should be prepared to work:

(1) with polynomial functions

pix) =anxn + an_1xn~1 + ... +axx + aQ,

where the coefficients a0,a1,...,an_1,an will usually be given integers

(but may sometimes be rational numbers, real numbers, complex num¬
bers, or even unknowns);

(2) with rational functions

rix) = pix)/qix), where pix) and qix) are polynomials;

(3) and also with the trig (trigonometric) functions sin x, cos x, tan x, and so
on, with the exponential function e*, and with the natural logarithm
function In x.

2.4. Remainder Theorem for polynomials

Any polynomial fix) of degree n>\ can be divided by x — a to give

fix) = (x-a)-q(x) +r, (2.4)

where the quotient qix) is a polynomial of degree n — 1 and the remainder r

is a constant.

(2.4.1) The Remainder Theorem. The Remainder Theorem states that the
remainder r in the identity (2.4) is always equal to fia).

(*) Prove this.

(2.4.2) The Factor Theorem. In particular, a linear expression x- a divides a

polynomial fix) (with no remainder!) precisely when fia) = 0.

(*) Derive this from (2.4.1).

14 A little useful mathematics

Example. Suppose that we want to factorize fix) = 6x4 + 5x3 - 39x2 +

4x + 12. If fix) = (x- a)-gix) with a an integer (and g(x) a polynomial with
integer coefficients), then a must divide the constant term 12 (Why?). Thus it
is natural to try values of a which divide 12.
(*) (a) Calculate /(l), /(-1), /(2), /(- 2). Since /(2) = 0, we know that x-2
must be a factor.
(b) Complete the factorization fix) = ix- 2)i6x3... -6).
(c) Let g(x) = 6x3 + 17x2 -5x — 6. Why do you not need to calculate
g(l)? Or g(-l)? Or g( —2)? Calculate g(2), g(3), and g(-3).
(d) Factorize fix) completely as a product of four linear factors.

2.5. Binomial Theorem

Given a set of four objects—say, {1,2,3,4}—there are exactly six different
ways in which one can choose two of these four objects (provided that the
order in which the two are chosen does not matter); namely, 12, 13, 14, 23, 24,
and 34. In general, given a set of n objects—say, {1,2,3,..., n}—let
denote the number of ways of choosing exactly r of them (provided that the

order in which the r are chosen does not matter). Thus = 6.

To obtain a formula for | ” j we need one preliminary idea. Given any r

objects, we first need to know how many ways there are of putting these
objects in order (1st, 2nd,..., rth). The object to put in the 1st position can be
chosen in r different ways; and once the 1st position has been filled, the 2nd
position can be filled in r — 1 ways. Since there are r ways of filling the 1st
position, and for each way of filling the 1st position there are r — 1 ways of
filling the 2nd position, there are r X (r - 1) ways of filling the first two
positions. Similarly, there are r X ir — 1) X (r — 2) ways of filling the first
three positions, and r X (r — 1) X (r — 2) X ... X 3 X 2 X 1 = r! ways of filling
all r positions.
We now count in two different ways the number of ways of choosing r
objects in order from {1,2,3,..., n). On the one hand, we could choose the 1st
object in n different ways; and for each choice of the 1st object there are
n - 1 ways of choosing the 2nd object; and for each choice of the first two
objects there are n-2 ways of choosing the 3rd object; and so on until,
having chosen the first r— 1 objects, there remain n - ir - 1) ways of
choosing the rth object: there are thus

n X (n - 1) X in - 2) X ... X in - ir - 1)) = n\/in - r)! (2.5.1)

ways of choosing r objects in order from {1,2,3,..., n). However, another way
of counting the same set could go like this: there are exactly |" j ways of
choosing r objects from {1,2,3,...,n] (provided that the order of choosing
 Algebra 15

does not matter); and for each such choice of r objects, there are r! different
ways of putting them in order: there are thus

(")xr! (2.5.2)

ways of choosing r objects in order from {l,2,3,...,n}. Equating (2.5.1) and

(2.5.2) we obtain

(")“ 0T=7jiri- (2 5'3>

(Note: There is exactly one way of choosing n objects from {1,2,3,..., n}— you
have to choose them all. We thus want (” J = 1. This suggests that in the
formula

n! 1
0! X n! 0!
we must take 0! = 1.)

(*) (a) Check that = and

l <- 0 th row

l 1 st row

(b) Use (2.5.3) to prove that 1 2 1

13 3 1
(?)+Ui)-(^)- 14 6 4 1
(c) Use (a) and (b) to show that |”j is
the rth entry in the nth row of I • • • « 1
Pascal’s triangle (shown on the
right).

Binomial Theorem. For any positive integer n,

<i+*>MoM?M2)*2+(3)*3+-+("h (2-5-4)
Proof. When multiplying out two identical brackets of the form (1 + jc), we
obtain four terms 1*1, 1 x, x-1, and x-x. When multiplying out n such
brackets every term will be of the form xr-Y~r = xr for some r (0 <r<n).
To obtain a term xr we have to choose a factor ‘x’ from r of the n brackets,
and a factor T from the remaining n — r brackets: the number of such terms
is therefore equal to the number of ways of choosing r brackets (for the
factor ‘x’)—that is, |^ j. Q.E.D.
16 A little useful mathematics

(,) (a) Prove 2“ = (^ J + (" ) + ( ") + (3 ) +

+
")•
(b) Prove + + ... = + + +

Because of (2.5.4), the numbers I ” I are called binomial coefficients

2.6. Inequalities

(2.6.1) Suppose that a> b and that k is any real number. Then the following
are always true:

(1) a + k > b + k;

(2) ka > kb if k is positive, but ka < kb if k is negative;

(3) a2 > b2 provided that both a and b are positive.

(*) 2> -3, but 22 is not greater than (— 3)2. Explain! (Thus x>y does not
always imply that x2 > y2.)

(2.6.2) Suppose that a > b and c > d. Then we also have:

(4) a + c > b + d;

(5) ac > bd provided that the four numbers a, b, c, and d are all positive.

(*) Find real numbers a, b, c, and d such that a > b and c > d, but such that
ac < bd.

(*) Prove that the square of any real number x is always ^0. Prove that
x2 = 0 if and only if x = 0.

Many complicated-looking inequalities depend only on this simple fact that

squares cannot be negative.

(*) Use this fact to prove that, for any real numbers a and b, we have

a2 + b2 > 2ab,

and that equality (a2 + b2 = lab) occurs if and only if a = b.

(*) Prove that (a + b)/2 > ffab) whenever a and b are positive real
numbers.

The LHS in the above inequality ‘(a + b)/T is the average, or arithmetic
mean (AM), of the two numbers a and b; the RHS V(afc)’ is the geometric
 Algebra 17

mean (GM) of a and b. The inequality (a + b)/2 > \/(ab) is often called the
‘AM-GM inequality’ for two variables a and b.

(*) Look at your proof of the inequality (a + b)/2> \/(ab) to discover

exactly when equality can hold.

(*) Prove that a2 + b2 + c2 > ab + be + ca for any three real numbers a, b,

and c.

(* *) Prove the AM-GM inequality for three variables:

a +b +c 3_
---> vabc whenever a, b, and c > 0.

3_
If a, b, c > 0, prove that (a + b + c)/3 = fabc if and only if a = b = c.

2.7. Recurrence relations

The -nth term <~»f a sequence is often denoted by un. A recurrence relation for
the sequence (un) is a way of defining later terms of the sequence by a
formula involving one or more earlier terms. Thus:

(1) = 1, un = un _ j + 3 defines the sequence 1,4,7,10,13,..., 3n - 2,...;

(2) = 0, un = 2un_ j + 1 defines the sequence 0,1,3,7,15,...,2"~1 - 1,...;

(3) = 1, un = un_x + n defines the sequence

1,3,6,10,15,..., n(n -f- l)/2,

(4) ul = \, u2 = 1, un = un_l + un_2 defines the sequence 1,1,2,3,5,8,... of

Fibonacci numbers;

(5) uQ = 1, un =nurl_1 defines the sequence 1,1,2,6,24,120,... of factorials.

(***) Let ux = 1, un = (n — 1 )un_1 + 1. For which values of n is un divisible

by nl When you think you know, try to give a proper proof that your
answer is correct.

A sequence of numbers (un) is often generated by some implicit property,

rather than directly from a recurrence relation. In such a case, one step in
understanding the sequence (so that we can eventually find a formula for the
nth term un as a function of n) is to find a recurrence relation which the
sequence satisfies.

Example. How many regions are created in the plane by 20 straight lines,
no two of which are parallel and no three of which pass through the same
point?
18 A little useful mathematics

Solution. Let rn be the number of regions created by n such lines.

(*) Check that r0 = 1, r1 = 2, and r2 = 4.

(*) Prove that rn = rn_ x + n.

(Hint: Given n lines, remove one line L to leave n — 1 lines, and rn_ x
regions. The line L crosses all the other n- 1 lines; so replacing L cuts
exactly n regions in two.)

(*) You could now use the recurrence relation to find r20. However, it is
much better to find a formula for rn in terms of n (and then put n = 20
to solve the original problem).

(**) (a) Let sn denote the number of ways of expressing n as a sum of

positive integers. Thus sx = 1, s2 = 2, and s3 = 4 (the four ways are
3, 2 + 1, 1+2, and 1 + 1 + 1). Prove that sn = s„ _ 1 + sn _ 2 + •.. +
s1 + 1. Hence calculate s]0. Find a formula for sn in terms of n.
(b) Suppose that we decide not to distinguish between T + T and
‘2 + 1’; Let an denote the number of ways of expressing n as a sum
of positive integers when the order of the terms does not matter.
Thus cr3 = 3. Calculate an (l<n<6). Find cr10. Express the nth
term an in terms of earlier terms. Then try to find a formula for crn
in terms of n.

These examples illustrate a general principle in all counting problems.

When trying to find ‘How many?’ objects of a specified kind there are, the
key idea is to count some collection in two ways and then to equate the
answers. This is exactly what a recurrence relation does: for example, sn is
defined to count the number of ways of writing n as a sum of positive
integers; counting the same set in another way we obtain s„_1+s„_2 +
... +Sj + 1. Equating the two answers allows us to obtain a very nice formula
for sn in terms of n. The apparently similar problem for an is much harder.

3. Proof

A substantial part of Olympiads (and of all serious mathematics) is concerned

not with obtaining answers, but with the methods you use to obtain those
answers—and in particular with proofs. Learning to present reasons and
proofs in a clear and correct manner is a fundamental part of mathematics.
You will not find it easy, but you cannot do mathematics without it. Without
clear proofs mathematics sinks to the level of mere opinion.
Whenever you want to solve a problem, one way of getting started is to
experiment with special cases and to try to come up with a conjecture which
you believe to be true. But guesses are more often wrong than right. Even
 Proof 19

when your guess happens to be correct, if you cannot prove it propertly, or

cannot be bothered to give a proper proof, there is no reason why other
people should take you seriously.
The sequence of rational numbers

(2/1)1, (3/2)2, (4/3)3, (5/4)\...,((« + 1)/«)"

is in some ways ‘natural’. Each term seems to be larger than the one before
(although this is not easy to prove), and all the terms seem to be less than 3
(which is not easy to prove either). This suggests strongly that as the terms
increase, they get closer and closer to some number < 3. But it is not at all
easy to decide for sure which real number the sequence tends to. And even if
you manage to guess, it is far from easy to prove that your guess is eorrect.

(*) (a) What real number do you think the sequence (2/l)1,(3/2)2,
(4/3)3,... converges to? (You may be able to guess on the basis of
numerical experiment, even though you are unlikely to be able to
prove at this stage that your guess is correct.)
(b) Suppose you were to use a calculator to work out the nth term
un = ((n + 1 )/n)n when n = 1010, or n = 1020. Why should you not
believe the ‘answer’? (Try to explain before you try it.)

(**) Prove that the sequence (un) is increasing— that is, prove that, for each
n > 1,

(n/(n- 1))" 1 < ((n + l)/n)”.

(**) Prove that every term of the sequence (un) is less than 3—that is, prove
that, for each n > 1,

((n + 1 )/n)n < 3.

We look briefly at four standard methods of proof that you will certainly
need, and one variation which we shall use in the next section.

3.1. Proof by deduction

At bottom, all proofs depend on deduction. That is, you start out from what
you know to be true, and work logically through a sequence of steps until you
arrive at the conclusion that you originally wanted to prove.

Example. If x satisfies |x2 - 2 + 1| = 1, prove that x = 0 or x = 2.

Proof. Suppose that |x2-2x + l| = l. If we factorize x2 - 2x + 1 the

equation becomes |(x — l)21 = 1. Since (x — l)2 > 0 for all x, |(x—1)2| =
(x - l)2 for all x, so we can drop the modulus signs and the equation to be
20 A little useful mathematics

solved is just (x - l)2 = 1. Subtract 1 from both sides to obtain x2 -2x = 0.

Now factorize and use (1.3): x(x — 2) = 0; therefore x = 0 or x = 2. Q.E.D.

This example has been written out in detail so that it is easy to read. The
same proof could be written out using ‘=>’ or in place of words. Although
harder for a beginner to read, these are in some ways clearer:

Claim \x2 — 2x+l\ = l ==> x = 0 or x = 2.

Proof |x2 - 2x + 1| = 1 => ](x-l)2| = l Proof |x2 — 2x + 1| = 1

=> (x — l)2 = 1 .-. (x - l)2 = |(x - 1)2| = 1

=> x2 — 2x = 0 .•. xz — 2x = 0

=> x(x — 2) = 0 x(x — 2) = 0

=> x = 0 or x = 2 x = 0 or x = 2. Q.E.D.

To solve a mathematical problem, you may need to calculate some number,

to simplify some expression, or to prove some non-obvious fact. Whatever you
are trying to do, it is natural (and sensible) while looking for ideas and while
working in rough to try to make progress both forwards from what you know,
and backwards from what you want to find or prove. This kind of rough work
helps to narrow the gap between what is given and what is sought, and often
suggests an approach which might help you to get from one to the other.
However, when you come to use these ideas in your final solution, you must
only use what you know and what you are given: a deduction must never start
from what you are trying to prove.

3.2. Distinguishing between a statement and its converse

One reason why students get confused is that they often fail to distinguish
between a given statement (such as ‘if |x2 - 2x + 1| = 1, then either x = 0 or
x = 2’), and the converse statement (in this case ‘if x = 0 or x = 2, then
|x2-2x + i| = r).

(3.2.1) Usually, when a statement is true, its converse is false.

(3.2.2) More importantly, a statement and its converse are logically com¬
pletely different. So even when a statement and its converse both happen to
be true, the proofs of the two statements are likely to be quite different.
To prove |x2 — 2x + 1| = 1 => x = 0 or x = 2’ it is no good writing some¬
thing like

‘Suppose x = 0 or x = 2.
Then substituting in |x2 - 2x + 1| shows that |x2 - 2x + 1| = 1.’
 Proof 21

These two lines prove ‘x = 0 or x = 2 => |x2 - 2x + 1| = 1’, which is the

converse of what was wanted. This confusion must be confronted and
resolved.
The statement ‘|x2 — 2x + 1| = 1 => x = 0 or x = T can be expressed
either as a necessary condition, or as a sufficient condition, depending on
whether one is more interested in the conclusion ‘x = 0 or x = 2’, or the
hypothesis ‘|x2 — 2x + 1| = T. If one is mainly interested in knowing when the
conclusion ‘x = 0 or x = 2’ holds, one might say that

The condition ‘| x2 — 2x+l|=l’isa sufficient condition for

‘x = 0 or x = 2’ to hold.
If one is more interested in the hypothesis ‘|x2 - 2x + 1| = F, one might say
that

The condition ‘x = 0 or x = 2’ is a necessary condition for

‘|x2 — 2x + 1| = 1’ to hold.
(3.2.3) Sometimes you are asked to prove an ‘if-and-only-if’ statement (such
as ‘|x2 — 2x + 1| = 1 if and only ifx = 0 or x = 2’). You then have to prove two
things:

(1) the ‘if’ part (in our example: ‘if x = 0 or x = 2 then |x2 — 2x + 1| = 1’);

and

(2) the ‘only if ’ part (‘if |x2 — 2x + 1| = 1, then either x = 0 or x = 2’).

That is, in the language of the previous paragraph, you have to prove that the
condition ‘|x2 — 2x+l\=V is both necessary and sufficient for ‘x = 0 or
x = 2’ to hold.

3.3. Proof by contradiction

One way of showing an assertion is true is to show instead that it cannot
possibly be false.

(3.3.1) f2 is not a rational number.

Proof. Suppose that this claim were false. Then f2 would have to be a
rational number, so that we would have f2 = m/n for some integers m and
n. By cancelling we can clearly ensure that m and n are not both even. But
then

f2 = m/n => 2 = m2/n2 => 2n2 = m2 => m2 is an even integer.

Now for m2 to be even, m itself must be even—say m = 2M (because if m

were odd, its square m2 would have to be odd!). Thus f2 = 2M/n.

\j2 = 2M/n => 2 = 4M2/n2 => n2 = 2M2 => n2 is an even integer.

22 A little useful mathematics

But then n must also be even. Hence m and n are both even, contrary to our
assumption, so we have a contradiction. Thus the claim cannot possibly be
false, and so must be true. Q.E.D.

3.4. Proof by induction

Often in mathematics we want to prove not just one statement, but infinitely
many statements all at once. For example, instead of proving, as we did
above, V2 = 21/2 is irrational’, we might want to prove the following.

(3.4.1) ‘ fl = 21/r is irrational for every integer n > 1’. That is, the (2")th
root of 2 is irrational for every integer n > 1.

It is important to realize that the words and symbols in (3.4.1) conceal

infinitely many different claims—one for each value of n > 1. Proof by
induction is a miraculous way of escaping from the apparently hopeless task
of having to prove each claim in turn—a task that could never be completed.
To illustrate h-;w this miracle works, we give first a proof of (3.4.1).

Proof of claim (3.4.1): ‘21/2" is irrational’ holds for every integer n > 1. For
each individual integer n, let C(n) denote the single claim

‘21/r is irrational’.

We prove that C(n) is true for every n> 1 in just two steps:

(a) C(l) is certainly true (this is precisely (3.3.1)).

(b) Let n be any integer > 1.

Claim: Suppose that we have already proved that C(n) is true for some
particular n > 1. Then C(n + 1) must also be true.
Proof: Suppose C(n + 1) were false. Then 21/2<" + " is not irrational, so
must be rational: 21/2‘" + 1> = a/b for some integers a and b. But then
21/2” = (21/2<"+ ’)2 = a2/b2 would be a rational number, contradicting
C(n). Hence C(n + 1) cannot possibly be false, and so must be true.

We now know two things:

(a) C(l) is true, and (b) whenever C(n) is true (for some n > 1),

so is C(n + 1).

Combining (a) and (b) shows that C(2) is true. Feeding this into (b) shows that
C(3) must be true—and so on, thus proving that C(n) is true for all n> 1.
Q.E.D.

Step (b) of any induction proof is often wrongly understood as ‘going from
 Proof 23

n to n + V. In reality, the only way to prove that C(n + 1) is true in step (b) is
to somehow reduce things in a way which allows us to use the key assumption
that C(n) is true. This idea (of reducing from ‘n + T to ‘n’) comes out clearly
in our second example.

(3.4.2) ‘72n + 1 + 152n + 1 is exactly divisible by 22’ for every integer n > 0.

Proof. For each individual integer n> 0, let C(n) denote the single claim

<~j2n + i _|_ i52« + ! ^ a multiple of 22’.

(1) Then C(0) says that ‘71 + 151 is a multiple of 22’, which is correct.

(2) Let n be any integer >0:

Claim: Suppose that we have already proved that C(n) is true for some
particular value of n > 0. Then C(n + 1) must also be true.
Proof: We have to prove C(n + 1); that is, we must show that

<72(/i + i)+i + 152(» + 1)+1 is a multiple of 22’.

The algebra of powers allows us to rewrite this expression:

'jUn + 1)+ 1 _|_ -j^2(rt + l)+l _ y2n + 3 j^2n + 3

= 72(72n + 1 + 152n + 1) + (152 - 72)152” + 1

= 72(72n + 1 + 152" + 1) + (15 - 7)(15 + 7)152n + 1.

If C(n) is true, then the first bracket (72n +1 + 152n + 1) is certainly a

multiple of 22, so that the first term on the RHS is definitely a multiple of
22. Also, the second term on the RHS has a factor of (15 + 7) = 22. Thus
the whole RHS is a multiple of 22. Hence the LHS must be a multiple of
22, so that C(n + 1) is true.

We now know two things:

(a) C(0) is true, and (b) whenever C(n) is true (for some n ^ 0),

so is C(n + 1).

Combining (a) and (b) shows that C(l) must be true. Feeding this into (b)
shows that C(2) is true—and so on, thus proving that C(n) is true for all
n > 0. Q.E.D.

3.5. Strong form of proof by induction

This is a minor, but frequently useful, variation on Section 3.4. Suppose we

want to prove that some assertion C(n) is true for every n > 1. It is
24 A little useful mathematics

sometimes much easier to use a slightly different (although logically equiva¬

lent) form of proof by induction. Suppose we manage to prove both

(a) that C(l) is true, and

(b) that whenever C(l), C(2), ...,C(n) are all true, so is C(n + 1).

Then combining (a) and (b) shows that C(2) is true. Feeding this (and (a)) into
(b) shows that C(3) must be true. Feeding all this back into (b) shows that
C(4) must be true—and so on, once again proving that C(n) is true for all
n ^ 1. We shall see an example of this in (4.2.1).

4. Elementary number theory

Before we launch into a little ‘theory’, have a go at this number puzzle.

(*) (a) Take any two-digit number, subtract the sum of the digits, then divide
your answer by nine. What do you get? Explain!
(b) What happens if you do the same, starting with a three-digit number?

It is tempting to do exactly as instructed: choose a particular two-digit

number, do the stated calculation, and write down an answer. But mathemat¬
ics is about insight, not just answers. We often try to convey this by setting
problems that have a slightly surprising answer which demands an ‘explana¬
tion’. However, an answer can only be surprising if you remain alert.
It is easy to miss the first surprising thing in part (a)—namely that the
answer (after subtracting the sum of the digits and dividing by nine) always
seems to be an integer! The second surprise is that the answer is in fact a very
special integer.
If in part (a) you take a particular two-digit number (say 73), you might just
guess what is going on. You should then realize that ‘Explain’ means ‘prove
that the answer is always equal to the tens digit of the number you started
with’. A particular example can sometimes lead to a promising guess; and you
can then try to find some way of proving that your guess is correct.
However, this strategy has its limitations. Suppose that you were to try a
similar approach in part (b). If you started with a particular number—say,
124—you would land up with the answer 13! This suggests no obvious ‘guess’.
Mathematical problems cannot usually be solved by guessing on the basis of
particular examples. Instead, one needs an approach that will reveal the inner
structure of the problem: in this case (as in many others) this means using
algebra. If we start with any two-digit starting number ‘ab’ = 10a + b, and
then subtract the sum of the digits a + b and divide by 9, we obtain precisely
a. Similarly, if we start with any three-digit number ‘abc’, subtract the sum of
the digits a + b + c, and then divide by 9, we obtain precisely 11 a +b.
 Elementary number theory 25

4.1. Divisibility

These simple problems illustrate the general principle underlying elementary

number theory—namely that questions of divisibility can often be resolved by
carefully restricting school algebra to the set Z of integers. Because division is
usually impossible in Z, the arithmetic and algebra of integers is a little
different from that for rational numbers or real numbers. The central idea is
that of divisibility. This, despite its name, depends only on multiplication—not
on division.
If m and n are integers, we say that m divides n (exactly!) if n = m-k for
some integer k. The number m is then called a factor of n, n is called a
multiple of m, and n is said to be divisible by m. The equation n = m-k is a
factorization of n as a product of the two integers m and k. Thus factors
come in pairs: m and k = n/m.

(*) Prove that every integer n divides 0. Which two integers divide 1? Which
integers does 1 divide?

(*) Suppose that x and y are integers and that x divides (x +y)2. Prove that
x must -hen divide y2. Is it true that x always has to divide y ' (If so,
prove it; if not, give an example to show that it may be false.)

When told that m and n are integers and that m divides n, many students
find it hard to resist the temptation to interpret this fact by writing fractions
such as n/m. This is most unhelpful, for it is hard to remember that this
expression which looks like a fraction is in fact an integer. The information
that one number divides another should always be interpreted as an equation
involving multiplication with integer factors, and avoiding fractions:

‘m divides n’ means ‘n=m-k for some integer k\

When you work with the resulting equations, you then know that all the
symbols stand for integers. As we shall see, it is then possible to reason in
ways that simply don’t work if you mix fractions and integers.

(*) Suppose that x is a positive integer such that x divides x2 + 1. Write this
as an equation. Manipulate this equation to obtain a product of two
integers which is equal to 1. Hence find all possible values for x.

(*) Suppose that x is an integer (not necessarily positive) such that x + 3

divides x2 + 5x + 5. What does this tell you about x?

4.2. Primes and prime factorization

A prime number is a natural number n ^ 2 which cannot be factorized as a
product of smaller natural numbers. The first few prime numbers are 2, 3, 5,
7, 11, 13, 17, and 19. There are just 25 prime numbers less than 100. A
natural number n > 2 which is not prime is said to be composite.
26 A little useful mathematics

(*) Can the difference between a two-digit number and its ‘reverse ever be a
prime number?
The importance of prime numbers stems from the fact that every positive
integer (other than 1) can be ‘broken down’ into a product of prime factors.

Example. 360 = 2 X 180 = 2x2x90 = 2x2x2x45

= 2X2X2X3X15 = 2X2X2X3X3X5.

The reason why 1 is not included as a prime number is that unlike 2, 3, 5,

and so on—1 does not help us to break down, or factorize, other integers.

(4.2.1) Every natural number n > 2 can be completely factorized as a product

of prime numbers.

Proof. For each n > 2, let C(n) denote the single claim

‘n can be completely factorized as a product of prime numbers’.

(1) C(2) is certainly true, since n = 2 is a prime number and so is already

completely factorized as a ‘product’ of just one prime number.

(2) Let n be any integer >2.

Claim: Suppose we have already proved that C(2), C(3),..., C(n) are all
true for this particular value of n. Then C(n + 1) must also be true.
Proof: If n + 1 is a prime number, then it is already ‘completely factorized
as a product of just one prime number’, so that C(n + 1) is true. If n + 1
is not a prime number, then it can be factorized as a product of two
smaller numbers n + 1 = m-k. Since m < n + 1, we know that C(m) is
true; hence m can be completely factorized as a product of prime
numbers; say, m = pxp2 pr- Similarly, k<n + 1, so that C{k) is true
and k can be completely factorized as a product of prime numbers; say,
k = q1q2 qs- But then n + 1 =plp2 ••• pr-qxq2 ••• qs can also be factor¬
ized as a product of prime numbers, so C(n + 1) is true.
Hence, by the strong form of proof by induction, C(n) is true for every n > 2.
Q.E.D.

(4.2.1) shows that every natural number n > 2 can be factorized in at least
one way as a product of primes. In Section 4.4 we shall prove that each
natural number n ^ 2 can be factorized in exactly one way as a product of
primes.

(**) Prove that the difference between any integer n and any number n'
obtained from n by permuting the decimal digits of n is always
composite.
 Elementary number theory 27

4,3. The division algorithm and highest common factors

We shall describe the algorithm first, then look at what it does.

Example. Given any two natural numbers, say p = 481 and q = 91, we first
divide the smaller into the larger number to obtain a remainder r:

481 = 91 X 5 + 26.

We then do the same all over again, this time with the pair q = 91, r = 26,
and so on:

91 = 26X3 + 13,
26 = 13 X 2 + 0.

Since the remainders get smaller each time, we will always reach a stage at
which the last remainder is equal to 0.

(*) Why must the remainders get smaller each time?

In the above example, the last equation occurs at the third stage. This
equation with remainder = 0 shows that 13 divides 26 exactly. If we use this
fact in the previous equation, we see that 13 divides the RHS = 26 X 3 + 13
exactly, and so must divide the LHS = 91 exactly. But then in the first
equation 13 divides the RHS = 91 X 5 + 26 exactly, and so must divide the
LHS = 481. Hence

13 is a common factor of p = 481 and q = 91.

(*) Let x be any other factor of 481 and 91. Then the first equation shows
that x must divide 481 — 91 X 5 = 26 exactly. The second equation then
shows that x must divide 13. Why?

Hence 13 is the highest common factor (HCF) of 481 and 91. We write this as

13 = HCF(481,91).

The miraculous thing about the division algorithm is that it allows you to find
the HCF of two natural numbers without factorizing them. Moreover, although
it is called the ‘division’ algorithm, you don’t actually have to do any division
at all. All that is needed when implementing the algorithm is to do repeated
subtraction and to stop each time when the remainder r is smaller than the
‘divisor’ q. The division algorithm has many important consequences.
In the above example, the second equation shows that 13 = 91 — 3 X 26. If
we then substitute for 26 from the first equation, we obtain

HCF(481,91) = 13 = 91 - 3 X 26 = 91 - 3 X (481 - 5 X 91)

= 16X91-3X481.
28 A little useful mathematics

(* *) Use the division algorithm to find HCF(23 579,7651). Then substitute

backwards to find integers x and y such that

HCF(23 579,7651) = x • 23 579 + y • 7651

(4.3.1) If we apply the division algorithm to any two natural numbers p and q
we obtain a sequence of equations. The last non-zero remainder d is the
HCF of p and q. Substituting backwards in the sequence of equations then
shows how to express d = HC¥(p,q) as a combination of p and q with
integer coefficients x and y:

d = HCF (p,q) = x-p +y-q.

(*) Suppose that a, m, and n are integers satisfying the equation m2 = an,
and that m and n have highest common factor equal to 1. Prove that
n = 1 or n = —1.

(*) Use the previous exercise to prove that, if t and a are integers satisfying
t2 - a(t — 1), then t = 2 or t = 9.

Many problems involving integers x and y (say) are easier to solve once we
know that x and y ‘have no common factors’; that is, HCF(x, y) = 1. When
HCF(x, y) = d # 1, it may be possible by writing x = d-x' and y = d-y' to
reduce to a similar problem involving x' and y' (with HCF(x',y')= 1). This
trick of first removing common factors often makes hard problems easier.

(*) Suppose that you know that x and y are integers and that xy divides
x2 +y2. Prove that x = ±y.

(4.3.2) Let p be a prime number, and suppose that m and n are integers such
that p divides the product m-n exactly. One way in which this could happen
is if p divides m exactly. Suppose that p does not divide m exactly. Then
HCF(/?, m) = 1 (Why?) so that by (4.3.1) we have

HCF(p, m) = 1 — x'p +y-m for some integers x and y.

Therefore n =xpn +ymn.

Now p divides the RHS of this last equation exactly (Why?); so p must divide
the LHS exactly—-that is, p divides n. Hence

if p is a prime number and p divides the product mn,

then p divides m, or p divides n.

(*) The number p = 4 divides the product m-n = 2 X 6 = 12, but 4 does not
divide either of the two factors 2 and 6. Explain!
 Elementary number theory 29

(4.3.3) If k is an integer and k2 = a-b for some natural numbers a and b

with HCF(a, b) = 1, then a and b must both be perfect squares.

(**) Prove this.

(Hint: Denote this assertion by C(k). Check that C(l) is true. Then
suppose that C(l), C(2),..., C(n) are all true, and prove that C(n + 1)
must then be true: let (n + l)2 = a -b with HCF(a, b) = 1; if a = 1, then
a = l2 and b = (n + l)2 are both perfect squares; if a + \, choose a
prime p which divides a, show that p2 divides a, reduce to
((n + \)/p)2 = (a/p2)-b, and then use the fact that C((n + \)/p) is
true.)

(*) k = 6 is an integer, and 62 = 2 X 18 is a perfect square, but neither 2 nor

18 are perfect squares. Explain!

(**) Let d and z be integers such that d2 divides z2. Prove that d divides z.

Three integers x, y, and z that satisfy the equation x2 +y2 =z2 are called
a Pythagorean triple. A Pythagorean triple x, y, z such that HCF(x, y) = 1 is
said to be primitive.

(*) Let x, z be an unknown Pythagorean triple. Show that HCF(x, y) =

HCF(y, z) = HCF(z, x): let this number be d (say). Thus x = d-x', y =
d-y', and z = d-z' for some integers x', y', and z'. Show that x', y', z',
is a primitive Pythagorean triple; that is, show that HCF(x',y') =
HCF(y',z') = HCF(z',x')= L

(*) Let p and q be any two integers.

(a) Show that x=p2 — q2, y = 2pq, z =p2 + q2 is always a Pythagorean
triple.
(b) Find values of p and q in (a) to produce the Pythagorean triple 3, 4,
5. Can you find values of p and q to produce the Pythagorean triples

6,8,10; 5,12,13; 9,12,15; 8,15,17; 15,20,25; 15,36,39?

(c) What conditions on p and q guarantee that the Pythagorean triple in

part (a) is primitive?

Part (b) of the previous exercise should have convinced you that the
formulae in part (a) do not produce all Pythagorean triples. To find general
formulae which will generate all Pythagorean triples, one has to go through
two stages: first remove common factors, then find a formula for all primitive
Pythagorean triples. It is not hard to show that, with minor additional
restrictions (namely, (i) HCKp, q) = 1 and (ii) one of p and q is odd and the
other is even), the formulae in (a) produce all primitive Pythagorean triples;
hence a general Pythagorean triple always has the form x = d(p2 - q2),
y - 2dpq, z = d(p2 + q2).
30 A little useful mathematics

(**) Suppose that x and y are positive integers and that 2xy divides x +y2.
Show that x must be a perfect square. What else .can you conclude
about xl
(*) Suppose that x and y are integers and that xy divides (x— y)2.
(a) Show that any prime number which divides x must divide y as well.
(b) Suppose that p is a prime number such that p3 divides x. Show that
p3 must also divide y.

(*) Find all pairs of positive integers x, y:

(a) such that x-y divides x+y;
(b) such that x+y divides x-y.

4.4. Uniqueness of prime factorization and counting factors

We do not distinguish between two ‘different’ prime factorizations (such as

2x2x3 and 2 X 3 X 2) if they involve exactly the same prime factors in a
different order.

(4.4.1) Every natural number n > 2 can be factorized in just one way as a
product of prime numbers.

Proof. We have already proved that every natural number can be factorized
as a product of prime numbers in at least one way. We now have to show that
a natural number n > 2 can never be factorized in more than one way. Let
C(n) denote this assertion for the particular natural number n.

(1) C(2) is true (2 is prime, so it can be factorized in just one way—namely

as ‘2’).

(2) Suppose that for some particular value of n>2we have already proved
that C(2), C(3),..., C(n) are all true. We show that C(n + 1) must then
be true.

If n + 1 is a prime number, then C(n + 1) is clearly true.

What if n + 1 is not a prime number? In that case we suppose that n + 1
can be factorized in two ways and show that they are in fact identical. Suppose
that n + 1 =pxp2 •"Pr = cli(h are two ways of factorizing n + 1 as a
product of prime numbers. Then px is a prime number which divides a
product q1-(q2 "• qs)• Thus px must divide one of the two factors: either px
divides qx or px divides (q2 ••• qs).
If px divides qx, then px=qt (Why?). If Pl divides (q2"-qs) =
q2'(<h ■" qs\ then px must divide one of the two factors. Continuing in this
way we see that, whatever happens, px is equal to qv or p, is equal to q2, or
Px is equal to q3,..., or px is equal to qs. We may assume that px = q2 and
cancel these two primes to give p2p3 pr = q2q3 ■■■ qs = m (say). Then, since
C(m) is true, the remaining p’s (p2,...,pr) must be the same as the
 Elementary number theory 31

remaining q’s (q2,...,qs), so the two original factorizations were in fact

identical. Q.E.D.

The above result is what makes prime numbers important: every natural
number can be broken down uniquely as a product of prime numbers. This
has many important consequences; we shall content ourselves here with one
relatively modest consequence—a simple way of counting how many different
factors a number has.

(*) How many factors does 360 have?

{Hint: 360 = 23 • 32 • 5. Each factor of 360 involves a power of 2 (for which
there are four choices), a power of 3 (three choices), and a power of 5
(two choices).)

The key to the previous problem is the uniqueness of prime factorization, in

that any factor of 360 must be the product of at most three 2’s, at most two
3’s, and at most one 5. This simple idea is enough to prove the general result.

(4.4.2) If n =Pi'P22 •"p°rj where pl,p2,...,pr are different prime numbers,

then n has exactly (a1 + 1 )(a2 + 2)---{ar + 1) different factors.

(*) Which number less than 50 has the most factors? Which number less
than 100 has the most factors? Which number less than 1000 has the
most factors?

4.5. Last digits, working with remainders, and congruences

(*) My calculator thinks that (92)5 = 6590815 230. Why is this answer obvi¬
ously wrong? What do you think the right answer should be? (Do not use
a calculator!)

(*) My calculator thinks that (2)32 = 4 294 967 300. What do you think the
right answer should be this time?

Some problems about integers can be solved by considering only the last
figure, or units digit. When we add or multiply two numbers, the units digit of
the answer can be obtained by simply adding, or multiplying, the units digits
of the two starting numbers. The remaining digits have no effect on the units
digit of the answer.

(*) Find the units digit of the answer to each of the following:
19 X 93, 19 X 94, 1993 X 1994, 1993, 1994, 19931"4, 19941"3.

(*) Let 7V = 19931"4 + 19941"3.

(a) Show that N is not a perfect square.
(b) If N were a perfect cube, what would the units digit of its cube root
have to be?
32 A little useful mathematics

Although some problems about integers can be solved by thinking about

units digits, most cannot be solved in this way. One reason is that the units
digit of a number written in base 10 depends on the choice of ‘base’, and so is
not a property of the number itself. The number 21 is equal to 2 X 10 + 1,
and so has units digit T in base 10; but in base 9 it would have units digit
equal to ‘3’.

(*) 1 is a perfect square, but 11, 101, and 1001 are not perfect squares. Can
any other number of the form 10000...00001 ever be a perfect square?

The previous problem can be solved very simply—but not by looking at the
units digit in base 10. We must first generalize the underlying idea.
Everyone knows that

‘odd X odd = odd’ and that ‘odd + odd = even’.

These familiar facts and those of the previous few paragraphs are special
cases of a simple, but much more general theory—the theory of congruences.

(4.5.1) Let us start by choosing a ‘base’ n. Then each integer N can be written
as a multiple of n plus a remainder r:

N = q-n + r, where 0^r<n.

We write this as

N = r (mod n)

and say that ‘N is congruent to r modulo n\ More generally, we write

a = b (mod n) whenever n divides a — b.

Examples. 11 = 1 (mod 10); 78 = 66 (mod 12); 28 = 13 (mod 5).

(4.5.2) Let N' be another integer with remainder r' modulo n; that is,
N' =<?'-n + r'. Show that to find the remainders for N + N' and for N-N'
modulo n, we only have to look at r + r' and r-rthat is,

N + N' = r + r' (mod n) and N-N' = r-r' (mod n).

(*) (a) 5.5 = 25 = 5 (mod 10), so the remainder ‘5’(mod 10) is congruent to
its own square. Which other remainders (mod 10) are congruent
to their own squares? Which remainders (mod 10) are congruent to
other squares?
(b) What does part (a) tell you about the units digits of perfect squares
(written in base 10)?
(c) Which one of 5328 and 5329 could be a perfect square? What must
it be the square of? (Do not use a calculator!)
 Elementary number theory 33

(*) (a) Which remainders (mod4) are congruent to perfect squares?

(b) Suppose that x, y, and z are integers satisfying x2 +y2 =z2. Could
x and y both be odd? (If so, find an example; if not, prove it is
impossible.)

(*) (a) Suppose that 3 -x = 2(mod 10). What can you say about x?
(b) Suppose that 3 — 2x = 2 (mod 10). What can you say about x?
(c) Suppose that 3 — 2x = 1 (mod 10). What can you say about x?

(**) Suppose that (3 — x)-(x — 7) = 0(mod 10). What can you say about x?

Congruence ‘=(modn)’ behaves just like the more familiar equality *=’ for
addition and for subtraction. In general, as the previous problem and the next
two problems show, you have to be a little careful when using multiplication
—and even more careful when using division or cancellation!

(*) 2x5 = 0(mod 10), although neither 2 nor 5 is = 0(mod 10). Explain!
(*) 3 X 8 = 3 X 4(mod 12), but we cannot cancel the 3’s on each side. Why
not?
However, when the ‘base’ n happens to be a prime number, the congruence
relation = (mod n)’ behaves almost exactly like the more familiar equality
‘= ’. Here are two examples.
(*) Let p be a prime number and let x and y be natural numbers satisfying

x-y = 0 (mod p).

Prove that either x = 0(mod p) or y = 0(mod p).

(*) Let p be a prime number and suppose that a-x = a-y (mod p). Prove that

either a = 0(mod p), or x =y (mod p).

Hence we can cancel a on both sides of such a congruence—provided
that a is not congruent to 0(mod p).
(*) Let p be a prime number. Then (p — 1)! =£ 0(mod p).
(Hint: Use proof by contradiction. Suppose that p divides (p - 1)! =
(p - \)-(p - 2)!. Since p cannot divide the first factor p - 1, p must
divide the second factor (p — 2)!. Now repeat.)

The next fact is often very useful.

(4.5.3) Fermat's Little Theorem. Let p be any prime number. Then

ap = a (mod p) for every integer a.

Proof, (a) Suppose that a = 0(modp). Then ap = 0(modp), so ap =

a (mod p).
(b) Suppose that a # 0(mod p). Then none of
a-l,a-2,a-3, ...,a-(p — 1)
34 A little useful mathematics

is divisible by p (since if p divides a-i, p must divide either a or i, both of

which are impossible). Moreover, no two of a-l,a-2,...,a-(p — 1) are
congruent (mod p) (since if a-i=a-j (mod p), then i =7 (mod /?))• But there
are only p — 1 different non-zero remainders (mod p). Thus

a-\,a-2,...,a-(p - 1) must be congruent to 1,2,...,p - 1

in some order. Hence the product of the first set of numbers must be
congruent to the product of the second set of numbers.

.-. (a-lMa-2)-...•(<!•(/>- 1)) = 1 *2-...-(/? — 1) (modp)

ap~l-(p - 1)! = (p - 1)! (mod/?)

ap~l - 1 (mod p)

(where we can cancel (p- 1)!, because it is not = 0(mod /?)). Q.E.D.

Example. 255 + 1 is exactly divisible by 11.

Proof. 255 = (25)11 = (32)11, and 3211 = 32(mod 11).

.-. 255 + 1 = 32n + 1 = 32 + 1 = 33 = 0(mod 11). Q.E.D.

(*) Give another proof (cf. Section 2.2) that 255 + 1 is exactly divisible by 33.
(Do not use a calculator.)

(*) Give another proof that 19901"0 - 1 is exactly divisible by 1991.

5. Geometry

In Olympiad problems there is a tendency to assume that students are

familiar with the basic geometry of lines, triangles, and circles. You will have
met most of the relevant facts, but may not have spent enough time on them
to become truly familiar with them. Thus you may find it helpful, before
working on this section, to revise the basic properties of:

(1) angles in triangles;

(2) angles and parallel lines (alternate angles, supplementary angles, and
so on);

(3) circle theorems (angles in the same segment, angles at the centre of the
circle, angles in a semi-circle, cyclic quadrilaterals, and so on);

(4) congruence of triangles (side-side-side and side-angle-side criteria);

(5) similarity of triangles.

 Geometry 35

In particular, if you wish to claim that two triangles are ‘congruent’ (or
‘similar’), the order in which the vertices are listed matters: ‘ AA'B'C' is
congruent to A ABC’ means ‘A'B' =AB, B'C' = BC, C'A1 = CA, A A' =
A A, and so on’. If you need to work on elementary geometry, you may find
one of the general books listed in ‘Some books for your bookshelf helpful.

5.1. Pythagoras3 Theorem and the cosine rule

The ability to spot similar triangles and to handle the corresponding ratios of
sides is absolutely fundamental, and may need a lot of practice. To illustrate,
we begin with a proof of Pythagoras’ Theorem, using similar triangles.

(5.1.1) Claim. If A ABC has a right angle at A, then BC2 =AC2 +AB2.

Proof. Let the perpendicular from A to

BC hit BC at D. Then A ADC = ABAC =
90°, and AACD = AACB, so that ACAD =
A ABC. Thus AD AC and A ABC are simi¬
lar (angles equal in pairs). Hence AC/BC =
DC/AC. Similarly, A ADB = 90° and
AABD= ACBA, so that A DBA and A ABC are similar (angles equal in
pairs). Hence AB/BC = DB/AB.

.-. BC = DC + DB =AC2/BC +AB2/BC, so BC2 = AC2 +AB2. Q.E.D.

In a triangle ABC it is often convenient to call the angle ABAC at the

point A just ‘A’, and to denote the length of the side BC opposite angle A
by ‘a’. Thus A ABC has angles A, B, and C and sides a, b, and c.

A
(5.1.2) The cosine rule (Pythagoras’ Theorem
for non-right-angled triangles). In any trian¬
gle ABC, we have

a2 = b2 + c2 — 2be cos A
(*) Prove this.

5.2. Areas of triangles

We start with the most basic fact of all.

(5.2.1) The area of any triangle is equal to ‘half the base X the height’. It
follows that:
(1) triangles with equal bases and equal heights have equal areas;

(2) triangles with equal bases have areas proportional to their heights;
36 A little useful mathematics

(3) triangles with equal heights have areas proportional to their bases;

(4) if A ABC and A DEF are similar, then the ratio of their areas is equal to
(AB/DE)2 (and to (BC/EF)2, and to (CA/FD)2).

(5.2.2) Prove that the area of any triangle ABC is equal to |i»csin A.

(5.2.3) Angle bisector theorem. If the line AX

is drawn bisecting the angle A of AABC,
then BX:XC=AB:AC.

(5.2.4) Example. In the triangle on the right

the numbers represent the areas of the sepa¬
rate parts. We have to find the unknown
area x. Realizing the importance of trian¬
gles, we draw an extra line dividing the
quadrilateral into two triangles. If we now
view the left sloping edge as ‘base’, then the
triangle marked ‘a’ has the same height as
the triangle marked ‘5’, so their bases must
be in the ratio a: 5. Similarly, the triangle
a + b + 8 has the same height as the triangle
5 -I-10, so their bases are in the ratio (a + b + 8): 15. But these two pairs of
triangles have the same bases, so that a/5 = (a + b + 8)/15, from which we
see that 2a — b = 8.

(*) Now us the right sloping edge as base, look at the area ratios of two pairs
of triangles, and hence show that b/8 = (a + b + 5)/18. Hence find x.

5.3. Intersecting chords theorem

(5.3.1) If chords AB and CD in a circle meet at X, then AX XB = CXXD.

Proof.
A BAD = A BCD (angles in the same segment)
A ADC = A ABC (angles in the same segment)
A AXD = A CXB (vertically opposite angles)
.-. Triangles AXD and CXB are similar.
.*. AX: CX = XD: XB
.'. AX-XB = CX XD as required. Q.E.D.
 Geometry 37

(*) State and prove the corresponding theorem when

the chords AB and CD meet outside the circle.

(*) Suppose that chords CD and AB meet at X

outside the circle. As D approaches C, the line
CD becomes more and more like the tangent to
the circle at C, and CX XD tends to (CX)2. State
and prove the corresponding theorem in this case.

5.4. Some special points associated with a triangle

(5.4.1) The circumcentre of a triangle (usually labelled
O) is the centre of the circle passing through all three
vertices. It must therefore be equidistant from all
three vertices, and so lies at the intersection of the
three perpendicular bisectors of the sides of the
triangle.

(*) Prove that if we start with an arbitrary triangle

AABC, the three perpendicular bisectors of the
sides intersect at the point O.

When three or more lines pass through the same

point, we say they are concurrent. We can therefore
rephrase (5.4.1) as follows: ‘In any triangle, the per¬
pendicular bisectors of the three sides are concurrent’.

(5.4.2) The sine rule. Given any triangle AABC, let its
circumcircle have radius R. Draw the diameter
BA' through B. Then ABCA'= 90° (Why?) and
ACA'B = A CAB (Why?). Thus sin A = sin CA'B =
a/BA' = a/2R. Hence a/sin A = 2 R. Similarly,
b/sin B = 2R and c/sin C = 2R. We thus obtain the
sine rule:

In any triangle A ABC, a/sin A = 6/sin B = c/sin C = 2R.

(5.4.3)
(a) Let ABCD be a cyclic quadrilateral (that is, the four vertices A, B, C,
and D lie in order on the circumference of some circle). Show that
sin A = sinC, whence either A = C or A + C= 180°; similarly, either
B = D or B + D = 180°. Show that A = C only if A = C = 90°. Hence
conclude that we always have A + C = 180° = B + D.

(b) Find a more straightforward proof of the fact that opposite angles of a
cyclic quadrilateral have sum equal to 180°.
38 A little useful mathematics

(c) Let ABCD be any quadrilateral satisfying A + C — 180° — B + D. Show

that the circumcircle of A ABC passes through D.

(Parts (a) and (c) together prove that a quadrilateral ABCD is cyclic if and
only if opposite angles add up to 180°.)

(5.4.4) The incentre I of a triangle is the centre

of the circle which just touches all three sides.
Prove that the incentre lies on the bisector of
each angle. Conclude that the bisectors of the
three angles of any triangle are concurrent at
the point I.

(*) Prove that the radius r of the incircle of AABC is equal to A/s, where
A is the area of the triangle and s = (a + b + c)/2 is half the perimeter.
{Hint: Express the area A in terms of r.)

(5.4.5) The orthocentre H of a triangle is the

point at which the three altitudes meet. (An
altitude is a perpendicular from one vertex to
the opposite side.)

(*) Given any triangle ABC, construct triangle A'B'C', where A'B' passes
through C and is parallel to AB, B'C' passes through A and is parallel
to BC, and C'A' passes through B and is parallel to CA. Prove that the
orthocentre H of A ABC is the circumcentre of triangle A'B'C'.

(*) Prove that, if D, E, and F are the feet of the three altitudes of triangle
ABC, then H is the incentre of triangle DEF.

(5.4.6) The centroid G of a triangle is the point

at which the three medians meet. (A median is a
line joining a vertex to the mid-point of the
opposite side.)

(*) Prove that in any triangle, the three medians are concurrent.

(*) Prove that the centroid G is two-thirds of the way down each median.

(*) Prove that any triangle in which O = I = G = H must be equilateral.

Does the same conclusion hold if only some of the points O, I, G, and
H coincide?

(**) Prove that in any triangle the points O, G, and H are always collinear
(that is, they lie on one straight line, called the Euler line) and
OG: GH = 1:2.
 Trigonometric formulae 39

If the triangle were cut from a sheet of card, the centroid would be the
physical centre of gravity.

6. Trigonometric formulae

Our experience of elementary geometry is rooted in the idea that angles are
measured in degrees. However, the beginnings of higher mathematics (ele¬
mentary calculus, limits, and so on) reveal that one should really measure
angles in radians-, that is, an angle at C should be measured by the arc-length
it determines on the unit circle centred at C. Thus 360° becomes ‘27t
(radians)’, 180° becomes V (radians)’, and so on. In some contexts it does not
matter whether one uses degrees or radians; in other contexts important
results may only be true if angles are measured in radians. In particular, the
trig functions sin and cos have to be thought of as functions of angles
measured in radians if one wants to use calculus, or to work with limits. Thus,
at the very least, you must learn to be flexible, to think in both degrees and
radians, and to sort out when it is essential to work in radians (and why).
You should know by heart the values of

sinO, sin 7t/2, sin7r, sin7r/4, sin7r/3, sin7r/6;

COSO, COS7T/2, COS 77, COS 77/4, COS 77/3, cos 77/6.

You should also know that ‘sin’ is an odd function, and that ‘cos’ is an even
function; that is,

sin(-x) = -sin(x), cos(-*) = cos(*).

You should also memorize (and know how to prove) all the basic trig
formulae.

(6.1) sin2/! -1- cos2^4 = 1.

(6.2) sin2A = 2sin A -cos A.

(6.3) cos2 A = 2cos2A -1 = 1 - 2sin2/! = cos2A - sin2/!.

(6.4) Use these to derive the following:

(a) 1 + tan2/! = sec2/!;

(b) sin 2 A = 2 tan A/( 1 + tan2A);

(c) cos 2 A = (1 - tan2/l)/(l + tan2A);

(d) tan 2A = 2 tan A/(l - tan2/!).

40 A little useful mathematics

(6.5) sin(^l + B) = sin A cos B + cos A sin B.

(6.6) Use (6.5) to prove that sin(7r/2 — A) = cos A. Hence derive the
following:

(a) cos(/4 ±B) = cos A cos B + sin A sin B\

(b) tan(v4 ±B) = (tan A ± tan B)/{ 1 + tan A tan B)\

(A±B\ (A+B
(c) sin A ± sin B = 2 sin cos
2 2 y
A+B
(d) cos A + cos B = 2 cos cos
2 Y)
A+B (A-B
(e) cos A - cos B = -2 sin sin

It is important to know quick ways of deriving, or checking, these formulae

so that you can always be sure that you are using the correct formula. You
must also be ready to use these formulae to prove less familiar results.

You are now ready to have a go at some of the problems in the rest of the
book. We all find these problems hard—that is what makes them worth
tackling. Good mathematicians do not find maths easy; they simply worry
away at every problem until they begin to see what is going on. Half the
battle is to learn that things get clearer the longer you keep struggling on. So
do resist the temptation to look at the solutions too soon.
 Some books for your bookshelf
If we do not succeed in solving a mathematical
problem, the reason frequently consists in our
failure to recognise the more general standpoint
from which the problem before us appears only
as a single link in a chain of related problems.
David Hilbert, 1900

The headings used in this section are intended as a rough indication of the
contents, and should not be interpreted too strictly. Books which are more
demanding are marked with an asterisk (*).
The only useful reference is one that is available. I have therefore re¬
stricted myself in the main to books that are (a) in print, or (b) should be
available in decent libraries, or (c) may be available in school store cupboards
or second-hand bookshops. However, list F gives details of selected resources
available from other countries, together with the addresses from which they
are available: it would be wise to check prices before ordering, and to include
payment with your order.
If you need to learn, or to revise a topic, the first place to look is in the
textbooks currently used in your school. If that does not help, you may find
that your teachers can help you locate older textbooks which cover the
relevant topic in greater depth. School textbooks have the advantage of
including lots of exercises to help you come to grips with the ideas. All you
need is one book that helps you. One book that you really read is better than
a whole library of ‘useful’, but unread, references. The school textbooks listed
here are proven texts which may have something to offer, but are no longer
widely used—they are not necessarily ‘the best’.
More advanced books, not written for a school audience, tend to include
fewer worked examples and fewer exercises for you to practise on. You
should always read them with a pencil in your hand, and with plenty of rough
paper available—actively exploring details as you go along. Learning from
books can be difficult, but it is a skill worth mastering.

A. School textbooks

S. L. Parsonson (1970/71). Pure mathematics (2 vols). Cambridge University

Press. [An excellent pure mathematics A-level text, teaching ideas, not
42 Some books for your bookshelf

rules, and covering everything except calculus; with lots of good (but
tough) exercises and answers.]

B. Problem solving books, each covering a variety of topics

Ed J. Barbeau, William O. Moser, and Murray S. Klamkin (1995). Five

hundred mathematical challenges. Mathematical Association of America.
[Three hundred pages of problems and solutions to help lay the technical
foundations for higher olympiads.]
Judita Cofman (1990). What to solve? Problems and suggestions for young
mathematicians. Oxford University Press. [An adventure in serious problem
solving, built around 120 tough problems—classified by topic and by
solution technique.]
Judita Cofman (1995). Numbers and shapes revisited. Oxford University Press.
[Sequel to ‘What to solve?’.]
Heinrich Dorrie (1965). 100 great problems of elementary mathematics. Dover.
[A tour through elementary mathematics guided by classical problems.]
A. Gardiner (1987). Discovering mathematics. Oxford University Press. [A
hands-on introduction to mathematical ‘research’ using elementary
problems.]
George T. Gilbert, Mark I. Krusemeyer, and Loren C. Larson (1993). The
Wohascum County problem book. Mathematical Association of America.
[One hundred and thirty challenging problems with full solutions.]
Derek Holton, Problem solving series (15 booklets). Available from the
Mathematical Association (259 London Road, Leicester LE2 3BE). [A
chatty introduction to some serious problem solving. Booklets 1, 2, 4, 5,
and 6 available as a single book from Canadian Mathematics Competitions
(see list F).]
Ross Honsberger (1970). Ingenuity in mathematics. Mathematical Association
of America.
Ross Honsberger (1973). Mathematical gems. Mathematical Association of
America.
Ross Honsberger (1976). Mathematical gems II. Mathematical Association of
America.
Ross Honsberger (1985). Mathematical gems III. Mathematical Association of
America.
Ross Honsberger (1978). Mathematical morsels. Mathematical Association of
America.
Ross Honsberger (1979). Mathematical plums. Mathematical Association of
America.
Ross Honsberger (1985). More mathematical morsels. Mathematical Associa¬
tion of America.
 Some books for your bookshelf 43

Ross Honsberger (1995). From Erdos to Kiev: problems of Olympiad caliber.

Mathematical Association of America.
Ross Honsberger (1995). Episodes in nineteenth and twentieth century
Euclidean geometry. Mathematical Association of America. [Honsberger is
a beachcomber, who picks up interesting topics as he wanders along the
mathematical seashore, and then cuts and polishes them until they sparkle
with life—wonderful stuff (includes some exercises and solutions).]
* Loren C. Larson (1983). Problem-solving through problems. Springer. [A
remarkable array of problems and ‘take it or leave it’ solutions.]
Hans Rademacher and Otto Toeplitz (1966). The enjoyment of mathematics.
Princeton. [Twenty-eight readable chapters covering elementary topics
from the whole of mathematics, including many neglected areas everyone
should meet: now available as a paperback (Dover 1994).]
Hugo Steinhaus (1979). One hundred problems in elementary mathematics.
Dover. [A classic—elementary problems with genuine mathematical
content.]

C. Books covering essentially one topic

1. Numbers

Ivan Niven (1961). Numbers rational and irrational. Mathematical Association

of America. [A classic elementary treatment covering all the things one
should know (and a bit more); includes exercises and answers.]
David Wells (1986). The Penguin dictionary of curious and interesting numbers.
Penguin. [Fascinating facts presented in a way which makes you wonder
why, and want to know more.]

2. Proof
Books on ‘proof are nearly always a disappointment. Proof has to be
mastered in context; so the best books on proof are to be found in other parts
of this list!

3. Elementary number theory

* Alan Baker (1984). A concise introduction to the theory of numbers.

Cambridge University Press. [If the other books listed here are too easy,
this one may help you to see things in a new light!]
Albert H. Beiler (1964). Recreations in the theory of numbers. Dover. [Not a
textbook—and none the worse for that! Uses serious recreational topics as
a springboard for an interesting, though non-systematic, adventure. Lots of
problems and some solutions.]
44 Some books for your bookshelf

R. P. Burn (1982). A pathway into number theory. Cambridge University Press.

[Very much a DIY course—800 problems (and very little text) that should
let you teach yourself. The hands-on supplement to Davenport’s
exposition.]
H. Davenport (1992). The higher arithmetic, an introduction to the theory of
numbers (6th edn). Cambridge University Press (also Dover 1983). [One of
the best popular expositions of elementary number theory from A to Z.]
* I. Niven, H. Zuckerman, and H. Montgomery (1991). An introduction to the
theory of numbers (5th edn). John Wiley. [One of the best, and most
complete, number theory textbooks.]
Oystein Ore (1967). Invitation to number theory. Mathematical Association of
America. [Excellent gentle introduction—with exercises and answers.]

4. Geometry
* H. S. M. Coxeter (1961). Introduction to geometry. John Wiley. [An amazing
panoply of geometry—with exercises and solutions. Inevitably each topic is
touched on rather lightly, so be prepared to look elsewhere to fill in the
gaps.]
* H. S. M. Coxeter and S. L. Greitzer (1967). Geometry revisited. Mathematical
Association of America. [Very much a second course. If you know all the
geometry in ‘A little useful mathematics’ and want to use it in less familiar
settings, here’s your chance (with exercises and solutions).]
C. V. Durell (1921). A concise geometry. Bell.
C. V. Durell (1920). Modem geometry. Macmillan.
H. G. Forder (1930). A school geometry. Cambridge University Press.
H. S. Hall and F. H. Stevens (1903). A school geometry parts I-VI. Macmillan.
[If your elementary geometry is weak, these old-fashioned texts (reprinted
many times) provide a systematic approach with lots of excellent problems.]
H. G. Forder (1931). Higher course geometry (2 vols). Cambridge University
Press. [Sequel to ‘A school geometry’. Lots and lots of excellent problems.]
Harold R. Jacobs (1974). Geometry. W. H. Freeman. [A beautifully produced
compendium of basic school geometry (and a bit more)—with exercises
and selected answers.]
Serge Lang and Gene Morrow (1983). Geometry: a high school course. Springer.
[A coherent modern treatment of school geometry (with exercises).]
* Z. A. Melzak (1983). Invitation to geometry. John Wiley. [A fine exposition
of some unusual topics, focused more on problems than on theory.]
D. Pedoe (1995). Circles: a mathematical view. Mathematical Association of
America.
David Wells (1991). The Penguin dictionary of curious and interesting geometry.
Penguin. [From Acute-angled triangle dissections to Zonohedra. Every
page is bursting with things to do, questions to pursue, results to try to
prove.]
 Some books for your bookshelf 45

5. Trigonometry

C. V. Durell and A. Robson (1930). Advanced trigonometry. Bell.

H. S. Hall and S. R. Knight (1893). Elementary trigonometry. Macmillan.
[Good old school texts (reprinted many times) with lots of problems.]

6. Algebra

Ed Barbeau (1989). Polynomials. Springer. [A glimpse of higher mathematics

through the study of elementary polynomials. Lots and lots of problems.]
S. Barnard and J. M. Child (1925). A new algebra. Macmillan.
H. S. Hall and S. R. Knight (1897). Algebra for beginners. Macmillan.
H. S. Hall and S. R. Knight (1885). Elementary algebra. Macmillan. [Good old
school texts (reprinted many times) with lots of problems.]
H. S. Hall and S. R. Knight (1887). Higher algebra. Macmillan. [Uses algebraic
technique to introduce many related topics from higher mathematics in an
old-fashioned way. Lots of problems.]

7. Combinatorics (counting)

V. Boltyansky and I. Gohberg (1985). Results and problems in combinatorial

geometry. Cambridge University Press. [A nice introduction to combinato¬
rial geometry (e.g. given a sphere, into how many pieces must one cut it if
all the pieces are to have smaller diameter than the original sphere?). Out
of print, but worth finding.]
Ivan Niven (1965). Mathematics of choice: how to count without counting.
Mathematical Association of America. [Excellent introduction to
factorials, binomial coefficients, recurrence relations, inclusion-exclusion,
partitions, induction, etc.]
N. Ya. Vilenkin (1971). Combinatorics. Academic Press. [A wonderful collec¬
tion of counting problems used to motivate the theory (out of print, so try
a good library).]
A. M. Yaglom and I. M. Yaglom (1987). Challenging mathematical problems
with elementary solutions, Vol. 1 (Combinatorial analysis and probability
theory). Dover. [One hundred excellent (but tough) problems—with the
associated theory discussed in the text; includes full solutions.]

D. Collections of Olympiad problems, or of mathematical techniques

for Olympiad problems

* Samuel L. Greitzer (1978). International mathematical olympiads 1959-1977.

Mathematical Association of America.
46 Some books for your bookshelf

* Murray S. Klamkin (1986). International mathematical olympiads 1978-1985,

and forty supplementary problems. Mathematical Association of America.
[These two books cover all the International Olympiads up to 1985. Most
of each book is given over to presenting solutions to each problem.
Invaluable for the enthusiast, but not much ‘motivation’ to help beginners.
Problems from earlier IMOs tend to be easier than more recent ones.]
Murray S. Klamkin (1988). USA mathematical olympiads 1972-1986. Mathe¬
matical Association of America. [Similar to the preceding two
books—though the problems are simpler than the recent IMOs. The book
includes a useful booklist.]
Hungarian problem book I based on the Eotuos competitions 1884-1905.
Mathematical Association of America (New Mathematical Library, Vol.
11).
Hungarian problem book II based on the Eotvos competitions 1906-1928.
Mathematical Association of America (New Mathematical Library, Vol.
12). [English translations (by Elvira Rapaport) of Joszef Kiirschak’s
‘Problems of the mathematics contest’. Classic problems (with full solu¬
tions) from the motherland of mathematical problem solving.]
A. M. Yaglom and I. M. Yaglom (1987). Challenging mathematical problems
with elementary solutions, Vol. 2. Dover. [Another eighty problems on a
variety of topics, together with a brief discussion of the relevant theory;
with full solutions.]
D. O. Schlarsky, N. N. Chentzov, and I. M. Yaglom (1962). The USSR
olympiad problem book. W. H. Freeman (recently reprinted by Dover 1993).
[An excellent collection of tough problems. Worth hunting down.]

E. Puzzle books

Ed Barbeau (1995). After Math. Ward and Emerson Inc. (available from
Ward and Emerson, Six O’Connor Drive, Toronto, ON M4K 2K1, Canada).
[Lots of lovely puzzles and brain teasers presented with a light touch for
the general reader (with solutions and comments on the problems).]
A. Gardiner (1987). Mathematical puzzling. Oxford University Press. [Elemen¬
tary, but genuinely mathematical problems for beginners to cut their teeth
on. Reprint available from UK Mathematics Foundation, School of Mathe¬
matics, University of Birmingham, B15 2TT, UK.]
Tony Gardiner (1996). Mathematical challenge. Cambridge University Press.
[Six hundred quality multiple-choice problems to stretch younger sec¬
ondary pupils.]
Tony Gardiner (1997). More mathematical challenges. Cambridge University
Press. [More demanding problems for interested students aged 11-15.]
Martin Gardner (1965). Mathematical puzzles and diversions. Pelican.
Martin Gardner (1966). More mathematical puzzles and diversions. Pelican.
 Some books for your bookshelf 47

Martin Gardner (1977). Further mathematical puzzles and diversions. Pelican.

Martin Gardner (1978). Mathematical carnival. Pelican.
Martin Gardner (1981). Mathematical circus. Pelican.
Martin Gardner (1987). Riddles of the sphinx. Mathematical Association of
America.
Martin Gardner (1995). New mathematical diversions. Mathematical Associa¬
tion of America. [Real mathematics presented in a chatty way for the
serious amateur. Numerous other books by the same author.]
Boris A. Kordemsky (1975). The Moscow Puzzles. Pelican. [Hundreds of
excellent puzzles (with solutions) involving elementary mathematics;
justifiably popular.]
Raymond Smullyan (1981). What is the name of this bookl Pelican.
Raymond Smullyan (1984). Alice in puzzleland. Penguin. [Off-beat, but serious
logic problems to sharpen your wits. Numerous other books by the same
author.]
Ian Stewart (1989). Game, set and math. Penguin. [Real mathematics prob¬
lems presented in a lively and stimulating way, with lots for readers to get
their teeth into. Numerous other books by the same author.]
David Wells (1992). The Penguin book of curious and interesting puzzles.
Penguin. [Five hundred and sixty-eight classic puzzles (and solutions).
Invaluable.]

F. Books available from non-standard sources

I. Available from the Australian Mathematics Trust, University of

Canberra, P.O. Box 1, Belconnen ACT 2616, Australia

J. Edwards, D. King, and P. O’Halloran (eds) (1986). All the best from the
Australian mathematics competition. Australian Mathematics Competition.
P. O’Halloran, G. Pollard, and P. Taylor (eds) (1992). More of all the best from
the Australian mathematics competition. Australian Mathematics Competi¬
tion. [Excellent multiple choice problems for all secondary school ages.]
W. P. Galvin, D. C. Hunt, and P. O’Halloran (eds) (1988). An olympiad down
under. Australian Mathematics Competition Ltd. [Details of the 1988 IMO
in Australia—including all the problems submitted but not used.]
A. Plank and N. Williams (eds) (c. 1992—no date). Mathematical toolchest.
Australian International Centre for Mathematical Enrichment, Canberra.
[A slim compendium—including more results than you will ever need, and
additional references, but no proofs or exercises.]
J. B. Tabov and P. J. Taylor (1996). Methods of problem solving (Book 1).
Australian Mathematics Trust. [Lots of problems (and full solutions) on
proof by contradiction, induction, rotations, barycentric co-ordinates, and
invariants.]
48 Some books for your bookshelf

P. J. Taylor (ed.) (1993). International mathematics Tournament of the Towns

1980-84. Australian Mathematics Trust.
P. J. Taylor (ed.) (1992). International mathematics Tournament of the Towns
1984-89. Australian Mathematics Trust.
P. J. Taylor (ed.) (1994). International mathematics Tournament of the Towns
1989-93. Australian Mathematics Trust. [Marvellous tough Russian
problems for ages 14-18.]

In addition, the Australian Mathematics Trust (same address) publishes

annual reports on its own multiple choice competitions and olympiads,
including problems and solutions.

2. Available from Canadian Mathematical Society, 577 Kinjj Edward,

Ottawa, Ontario KIN 6N5, Canada

W. Moser and E. Barbeau (1978). The first ten Canadian mathematics olympiads
1969-1978. Canadian Mathematical Society.
C. M. Reis and S. Z. Ditor (1988). The Canadian mathematics olympiads
1979-1985. Canadian Mathematical Society. [Problems, solutions, and
results from the Canadian olympiads.]
E. Barbeau, M. Klamkin, and W. Moser (1975-80). 1001 problems in high
school mathematics (Books 1-4). Canadian Mathematical Society. [Lots
and lots of lovely problems (plus solutions) at sub-olympiad level.]

The Canadian Mathematical Society also publish an excellent problem

journal called Crux Mathematicorum: see list G.

3. Available from Canadian Mathematics Competitions, Faculty of

Mathematics, University of Waterloo, Waterloo, Ontario N2L 3G1,
Canada

Peter Booth, Bruce Shawyer, and John Grant McLoughlin (1995). Shaking
hands in Comer Brook and other math problems. Waterloo Mathematics
Foundation. [Two hundred problems for senior secondary students (with
solutions).]
Canadian Mathematics Competition (1988-96). Problems, problems, prob¬
lems, Vols 1-6. Waterloo Mathematics Foundation. [Multiple choice
problems from the Canadian Mathematics Competitions—on all sec¬
ondary levels.]
D. Holton (1993). Let's solve some math problems. Waterloo Mathematics
Foundation. [Booklets 1, 2, 4, 5, and 6 from the ‘Problem solving series’
(see list B), improved and bound together as one.]
 Some books for your bookshelf 49

4. Available from MathPro Press, P.O. Box 713, Westford MA

01886-0021, USA

Stanley Rabinowitz (1992). Index to mathematical problems, Vol. 1, 1980-84.

MathPro Press. [Enough problems here for a very long lifetime!]
D. Fomin and A. Kirichenko (1994). Leningrad mathematical olympiads
1987-1991. MathPro Press. [Hundreds of excellent problems from one of
the most famous olympiads.]
L. Zimmerman and G. Kessler (1995). ARML-NYSML contests 1989-1994.
MathPro Press. [The UK has nothing to match the American ‘math
leagues’, where teams get together for a couple of days to compete on all
sorts of math problems. These are excellent problems that have proved
their worth.]

5. Available from Centre for Excellence in Mathematical Education,

885 Red Mesa Drive, Colorado Springs, CO 80906, USA

Alex Soifer (1987). Mathematics as problem solving. Centre for Excellence in

Mathematical Education. [The title says it all. Excellent problems used to
suggest something of the essence of mathematics and how to do it better.]
Alex Soifer (1990). How does one cut a triangle. Centre for Excellence in
Mathematical Education. [Lots of problems (and solutions) involving
dissections of triangles.]
V. Boltyanski and A. Soifer (1991). Geometric etudes in combinatorial mathe¬
matics. Centre for Excellence in Mathematical Education. [Lovely prob¬
lems from tiling and combinatorial geometry.]
Alex Soifer (1994). Colorado Mathematical Olympiad. Centre for Excellence in
Mathematical Education. [The first ten Colorado Olympiads together with
‘further exploration’ of some of the more challenging problems.]
Alex Soifer (1997). Mathematical colouring book. Centre for Excellence in
Mathematical Education. [Lots of tough and interesting problems involv¬
ing the combinatorics of colouring.]

6. Available from Science Culture Technology publishing, AMK Central

Post Office, P.O. Box 0581, Singapore 915603

Marcin Kuczma (1996). 40th Polish mathematics olympiad 1989/1990. SCT

Publishing. [Twenty-four problems—with complete multiple solutions
—from the six sets of questions for the 40th Polish olympiad.]
50 Some books for your bookshelf

7. Available from Brendan Kelly Publishing Inc., 2122 Highview Drive,

Burlington, Ontario L7R 3X4, Canada
Ravi Vakil (1996). A mathematical mosaic: patterns and problem solving.
Brendan Kelly Publishing. [A wonderfully lively and refreshing introduc¬
tion to selected topics in number theory, combinatorics, geometry, infinity,
complex numbers and calculus. Excellent off-beat material for serious
students, written by a double IMO gold medallist.]

8. Available from Deakin University Press, Geelong, Victoria 3217,

Australia
Terence C. S. Tao (1992). Solving mathematical problems: a personal perspec¬
tive. Deakin University Press. [A mature introduction to the gentle art of
solving olympiad problems—full of wisdom and good taste, written by
another recent IMO gold medallist.]

9. Available from the Academic Distribution Center, 1216 Walker

Road, Freeland, MD 21053, USA
Marcin E. Kuczma (1994). Problems: 144 problems of the Austrian-Polish
mathematics competition. The Academic Distribution Centre. [A compila¬
tion of the problems posed at the first 16 Austrian-Polish mathematics
olympiads 1978-1993, with complete solutions.]

G. Problem solving journals

One of the best ways to develop mathematical problem solving is to make a

regular habit of tackling challenging problems on an appropriate level.
Problem-solving journals, or journals with a regular student ‘Problem corner’,
are a key resource.
The Mathematical Gazette (published by The Mathematical Association, 259
London Road, Leicester, LE2 3BE). [Three substantial volumes each year
containing interesting articles and reviews. Each issue contains ‘Student
problems’ and a ‘Problem corner’ with solutions to previously posed
problems.]
Mathematical Spectrum (published by the Applied Probability Trust, Hicks
Building, The University, Sheffield S3 7RH). [Three issues each year
aimed at interested secondary students, containing articles and a regular
problem corner (with solutions).]
Crux Mathematicorum (published by the Canadian Mathematical Society, 577
King Edward, P.O. Box 450, Station A, Ottawa, ON KIN 6N5, Canada).
[The problem solver’s bible! Eight issues each year, packed with original
problems (and solutions) and olympiad problems from around the world.]
 Part II
The problems
 The problems
The mathematician’s strength is refined by fire
through solving problems; s/he discovers new
methods and fresh ways of looking at things [...]
Mathematicians of past centuries devoted
themselves with passionate zeal to the solution
of specific difficult problems; they appreciated
the significance of difficult problems.
David Hilbert, 1900

Background to the problem papers

In the early years up to 1973, the British Mathematical Olympiad consisted of

a single round, and selection for the UK International Mathematical Olympiad
team each year was based strongly on the results of this one paper. By 1974 it
had become clear that one could not select an IMO team on the basis of a
single paper. From that year onwards, the BMO was followed by a second
three and a half hour paper for 20-40 invited candidates (the Further
International Selection Test—or FIST). Thus the BMO became, in effect, a
two-stage event. However, both rounds remained strongly selective.
Recently, partly as a result of the expansion of other mathematical compe¬
titions, the number of candidates wanting to take part in the BMO has
increased (the 1995 record entry was around 850, from 300 schools and
colleges). Most of these candidates want an additional challenge beyond their
school mathematics, but have no experience of the kind of problem usually
associated with mathematical ‘olympiads’. Thus the BMO has had to adjust.
In 1992 the old BMO was re-christened ‘BMO, Round V, and it was
implicitly accepted that this would no longer be a mainly selective paper, but
an event designed to challenge up to a 1000 able, if often technically weak,
youngsters in their last years at school.
In the same year (1992), the old FIST became ‘BMO, Round 2’—with a
purely selective function. (Entry to Round 2 is by invitation only: in recent
years the top 50 or so candidates in Round 1 have been invited to enter
Round 2 as of right, together with a further 50 or so candidates chosen on the
basis of other evidence.)
This book contains the BMO Round 1 papers only. These have generally
54 The problems

been taken in January each year. The slight hiatus in the dating of the
1987-9 papers stems from a short-lived attempt to move the date from
January to December: the 23rd BMO was held as usual in January 1987
(1987A); the 24th BMO was held in December 1987 instead of January
1988 (and so is dated 1987B); and the 25th BMO was held in December 1988
(dated 1988). The 26th BMO reverted to the more convenient January date in
1990—hence the apparent gap in 1989. Up to 1975, three hours was allowed
for each paper; in 1976 the style of problems changed and the time allowed
was increased to three and a half hours.

32nd British Mathematical Olympiad, 1996

1. Find as efficiently as possible all pairs (m, n) of positive integers satisfying

the following two conditions:
(a) two of the digits of m are the same as the corresponding digits of n,
while the other digits of m are both 1 less than the corresponding
digits of n;
(b) both m and n are four-digit squares.

2. A function / is defined for all positive integers and satisfies

fi\) = 1996,

and

/(1) +/(2) + ... +f(n) = n2f(n) for all n > 1.

Calculate the exact value of /(1996).

3. Let ABC be an acute-angled triangle, and let O be its circumcentre. The

circle through C, O, and B is called S. The lines AB and AC meet the
circle S again at P and Q respectively. Prove that the lines AO and PQ
are perpendicular. [ Note: Given any triangle XYZ, its circumcentre is the
centre of the circle which passes through the three vertices X, Y, and Z.]

4. For any real number x, let [x] denote the greatest integer which is less
r n
than or equal to x. Define q(n) = for n = 1,2,3,.... Determine
[M
all positive integers n for which q(n) > q(n + 1).

5. Let a, b, and c be positive real numbers. Prove that:

(a) 4(a3 + £>3) > (a +b)3;
(b) 9(a3 +b3 +c3)^(a +b + c)3.
 30th BMO, 1994 55

31st British Mathematical Olympiad, 1995

Find the first positive integer whose square ends in three 4s. Find all
positive integers whose square ends in three 4s. Show that no perfect
square ends in four 4s.

2. ABCDEFGH is a cube of side 2.

(a) Find the area of the quadrilateral AMHN,
where M is the midpoint of BC, and N is
the midpoint of EF.
(b) Let P be the midpoint of AB, and Q the
H
midpoint of HE. Let AM meet CP at X,
and let HN meet FQ at Y. Find the
length of XY.

3. (a) Find the maximum value of the expression x2y -y2x when 0 <1
and 0 <y < 1.
(b) Find the maximum value of the expression

x2y +y2z + z2x - *2z-y2x - z2y

when 0 < x < 1, 0 < y < 1, and 0 < z < 1.

4. Triangle ABC has a right angle at C. The internal bisectors of angles

BAC and ABC meet BC and CA at P and Q respectively. The points M
and N are the feet of the perpendiculars from P and Q to AB. Find angle
MCN.

5. The seven dwarfs walk to work each morning in single file. As they go they
sing their famous song: ‘High-low-high-low, it’s off to work we go,... .
Each day they line up so that no three successive dwarfs are either
increasing or decreasing in height. Thus the line-up must go up-
down-up-down- ... or down-up-down-up-.... If they all have different
heights, for how many days can they go to work like this if they insist on
using a different order each day? What if Snow White always came along
too?

30th British Mathematical Olympiad, 1994

1. Starting with any three-digit number n (such as n = 625), we obtain a new

number f(n) which is equal to the sum of the three digits of n, their three
products in pairs, and the product of all three digits.
(a) Find the value of n/f(n) when n = 625. (The answer is an integer!)
(b) Find all three-digit numbers n such that the ratio n/f(n) = 1.
56 The problems

2. In triangle ABC the point X lies on BC.

(a) Suppose that ABAC = 90°, that X is the midpoint of BC, and that
A BAX is one third of ABAC. What can you say (and prove!) about
triangle ACX?
(b) Suppose that ABAC = 60°, that X lies one third of the way from B to
C, and that AX bisects ABAC. What can you say (and prove!) about
triangle ACX?
3. The sequence of integers u0,u1,u2,u3,... satisfies u0 = 1 and
un+lun_l=kun for each n^l,
where k is some fixed positive integer. If w20oo= 2000, determine all
possible values of k.
4. The points Q and R lie on the circle y, and P is a point such that PQ and
PR are tangents to y. The point A lies on the extension of PQ, and y' is
the circumcircle of triangle PAR. The circle y' cuts y again at B, and AR
cuts y at the point C. Prove that 4 PAR = A ABC.
5. An increasing sequence of integers is said to be alternating if it starts with
an odd term, the second term is even, the third term is odd, the fourth is
even, and so on. The empty sequence (with no terms at all!) is considered
to be alternating. Let A(n) denote the number of alternating sequences
which only involve integers from the set {1,2,...,n}. Show that ^4(1) = 2
and A(2) = 3. Find the value of 4(20), and prove that your value is
correct.

29th British Mathematical Olympiad, 1993

1. Find, showing your method, a six-digit integer n with the following

properties:
(a) the number formed by the last three digits of n is exactly one greater
than the number formed by the first three digits of n (so n might look
like 123124);
(b) n is a perfect square.
2. A square piece of toast ABCD of side length 1 and centre O is cut in half
to form two equal pieces, ABC and CDA. If the triangle ABC has to be
cut into two parts of equal area, one would usually cut along the line of
symmetry BO. However, there are other ways of doing this. Find (with
proof!) the length and location of the shortest straight cut which divides
the triangle ABC into two parts of equal area.
3. For each positive integer c, the sequence un of integers is defined by
Wj = l, u2 = c, un = (2n + l)un_l — (n2 — l)un_2 (« > 3).
For which values of c does this sequence have the property that
Uj divides Uj whenever i <;'?
 27th BMO, 1991 57

[Note: If x and y are integers, then x dividesy if and only if there exists an
integer z such that y =xz. For example, x = 4 divides y= -12, since we
can take z = -3.]

4. Two circles touch internally at M. A straight line touches the inner circle
at P and cuts the outer circle at Q and R. Prove that CQMP = /LRMP.
5. Let x, y, and z be positive real numbers satisfying

|<jry+yz+zx<3.
Determine the range of values for (a) xyz and for (b) x+y +z.

28th British Mathematical Olympiad, 1992

1. (a) Observe that the square of 20 has the same number of non-zero digits
as the original number. Does there exist a two-digit number other than
10, 20, and 30 the square of which has the same number of non-zero
digits as the original number? If you think there is one, then give an
example. If you claim that there is none, then you must prove your
claim.
(b) Does there exist a three-digit number other than 100, 200, and 300 the
square of which has the same number of non-zero digits as the original
number?
2. Let ABCDE be a pentagon inscribed in a circle. Suppose that AC, BD,
CE, DA, and EB are parallel to DE, EA, AB, BC, and CD respectively.
Does it follow that the pentagon has to be regular? Justify your claim.
3. Find four distinct positive integers the product of which is divisible by the
sum of every pair of them. Can you find a set of five or more numbers with
the same property?
4. Determine the smallest value of x2 + 5y2 + 8z2, where *, y, and z are
real numbers subject to the condition yz +zx +xy = —1. Does x2 + 5y2 +
8z2 have a greatest value subject to the same condition? Justify your
claim.
5. Let / be a function mapping positive integers into positive integers.
Suppose that
f(n + l)>f(n) and f(f(n)) = 3n for all positive integers n.
Determine /(1992).

27th British Mathematical Olympiad, 1991

1. Prove that the number 3" + 2 X 17", where n is a non-negative integer, is

never a perfect square.
58 The problems

2. Find all positive integers k such that the polynomial x2k + 1+x+l is
divisible by the polynomial xk+x+\. For each such k, specify the
integers n for which xn +x + 1 is divisible by xk +x + 1.
3. The quadrilateral ABCD is inscribed in a circle of radius r. The diagonals
AC and BD meet at E. Prove that if AC is perpendicular to BD, then

(*) EA2 3 4 + EB2 + EC2 + ED2 = 4r2.

Is it true that if (*) holds, then AC is perpendicular to BD? Justify your

answer.
4. Find, with proof, the minimum value of (x + yXy + z), where x, y, and z
are positive real numbers satisfying the condition xyz(x +y + z) = 1.
5. Find the number of permutations (arrangements)

j\i j2, J3’ J4’ Js’ it of I? 2, 3, 4, 5, 6

with the property

for no integer n, 1 < n < 5,

do fy, j2,.. •, jn form a permutation of 1,2,..., n.

6. Show that if x and y are positive integers such that x2 +y2 — x is divisible
by 2xy, then x is a perfect square.
7. A ladder of length / rests against a vertical wall. Suppose that there is a
rung on the ladder which has the same distance d from both the wall and
the (horizontal) ground. Find explicitly, in terms of / and d, the height h
from the ground that the ladder reaches up the wall.

26th British Mathematical Olympiad, 1990

1. Find a positive integer the first digit of which is 1 and which has the
property that, if this digit is transferred to the end of the number, the
number is tripled.

2. ABCD is a square and P is a point on the line AB. Find the maximum
and minimum values of the ratio PC/PD, showing that these occur for
the points P satisfying APXBP=AB2.

3. The angles A, B, C, and D of a convex quadrilateral satisfy the relation

cos A + cos B + cos C + cos D = 0.

Prove that ABCD is either a trapezium or is cyclic.

4. A coin is biased so that the probability of obtaining a head is p, 0 <p < 1.

 25th BMO, 1988 59

Two players, A and B, throw the coin in turn until one of the sequences
HHH or HTH occurs. If the sequence HHH occurs first, then A wins. If
HTH occurs first, then B wins. For what value of p is the game fair (that
is, such that A and B have an equal chance of winning)?
5. The diagonals of a convex quadrilateral ABCD intersect at O. The
centroids of triangles AOD and BOC are P and Q; the orthocentres of
triangles AOB and COD are R and S. Prove that PQ is perpendicular to
RS. [Note: The centroid of a triangle is the intersection of the lines joining
each vertex to the midpoint of the opposite side; the orthocentre is the
intersection of the altitudes.]

6. Prove that if x and y are rational numbers satisfying the equation

x5 +y5 = 2x2y2,
then 1 — xy is the square of a rational number.

25th British Mathematical Olympiad, 1988

1. Find all integers a, b, and c for which

(x — a)(x — 10) + 1 = (x + b)(x + c) for all x.

2. Points P and Q lie on the sides AB and AC respectively of triangle ABC
and are distinct from A. The lengths AP and AQ are denoted by x and y
repectively, with the convention that x > 0 if P is on the same side of A as
B, and x < 0 on the opposite side; and similarly for y. Show that PQ
passes through the centroid of the triangle if and only if
3xy = bx + cy,

where b = AC and c = AB.

3. The lines OA, OB, and OC are mutually perpendicular. Express the area
of triangle ABC in terms of the areas of triangles OBC, OCA, and OAB.
4. Consider the triangle of numbers on j
the right. Each number is the sum of
three numbers in the previous row: l l i
the number above it and the numbers
immediately to the left and right of 12321
that number. If there is no number in
. . ... n . 1 3 6 7 6 3 1
one or more of these positions, (J is
used. Prove that, from the third row
on, every row contains at least one even number.
5. None of the angles of a triangle ABC exceeds 90°. Prove that
sin A + sin B + sin C > 2.
60 The problems

6. The sequence {an} of integers is defined by a, =2, a2 = 7, and

for n> 2.
“2 <an + l
an - 1

Prove that an is odd for all n > 1.

24th British Mathematical Olympiad, 1987B

1. Find all real solutions x of the equation

Ax + 1972098- 1986t/(x + 986049))

+ Ax + 1974085 - 1988V(x + 986049)) = 1.

[Note: A always denotes the non-negative square root of y.]

2. Find all real-valued functions / which are defined on the set D of natural
numbers x ^ 10, and which satisfy the functional equation

fix+y) =/(x)-/(y),

for all x,y eD.

3. Find a pair of integers r, s such that 0 < s < 200 and

45 r 59
— > - > —.
61 5 80

Prove that there is exactly one such pair r, s.

4. The triangle ABC has orthocentre H. The feet of the perpendiculars from
H to the internal and external bisectors of angle BAC (which is not a right
angle) are P and Q. Prove that PQ passes through the midpoint of BC.

5. For any two integers m and n with 0 < m < n, numbers d(n, m) are
defined by

d(n,0) = d(n, n) = 1 for all n > 0,

and

md(n,m) = md(n - 1, m) + (2n - m)d(n - 1, m - 1) for 0 < m < n.

Prove that all of the d(n, m) are integers.

6. Show that, if x and y are real numbers such that lx2 + 3xy + 3y2 = 1, then
the least positive value of (x2 +y2)/y is
 22nd BMO, 1986 61

23rd British Mathematical Olympiad 1987A ,

1. (a) Find, with proof, all integer solutions of

a3 + = 9.

(b) Find, with proof, all integer solutions of

35x3 + 66 x2y + 42 xy2 + 9 y3 = 9.

2. In a triangle ABC, ABAC = 100° and AB = AC. A point D is chosen on

the side AC so that AABD = ACBD. Prove that AD + DB = BC.
3. Find, with proof, the value of the limit as n -»°° 0f the quotient

[Note: j denotes a binomial coefficient.]

4. Let P(x) be any polynomial with integer coefficients such that

P(21) = 17, P(32) = -247, P(37) = 33.

Prove that if P(N) = N + 51, for some integer N, then N = 26.

5. A line parallel to the side BC of an acute angled triangle ABC cuts the
side AB at F and the side AC at E. Prove that the circles on BE and CF
as diameters intersect on the altitude of the triangle drawn from A
perpendicular to BC.
6. If x, y, and z are positive real numbers, find, with proof, the maximum
value of the expression
xyz
(1 +x)(x +y)(y +z)(z + 16)

7. Let n and k be arbitrary positive integers. Prove that there exists a

positive integer x such that \x(x + 1) - k is divisible by 2”.

2.2nd British Mathematical Olympiad 1986 ,

N
1. Reduce the fraction — to its lowest terms when

N = 2 244 851485 148 514 627, D = 8118811881 188118000.

62 The problems

2. A circle S of radius R has two parallel tangents, tx and t2. A circle Sx of

radius rx touches 5 and t{, a circle S2 of radius r2 touches S and t2, Sx
also touches S2. All of the contacts are external. Calculate R in terms of
rx and r2.
3. Prove that if m, n, and r are positive integers and

1 + m + m/3 = (2 + /3) ,

then m is a perfect square.

4. Find, with proof, the largest real number K (independent of a, b, and c)

such that the inequality

a2 + b2 + c2 > K(a + b + c)2

holds for the lengths a, b, and c of the sides of any obtuse-angled triangle.

5. Find, with proof, the number of permutations

crx, a2, ^3> • • •»1,2,3,..., n

such that

ar<ar+2 forl<r<«-2 and ar<ar+3 forl<r<n-3.

[Note: A permutation is any ordered arrangement of the numbers

1,2,3,...,n.]

6. AB, AC, and AD are three edges of a cube. AC is produced to E so that

AE = 2 AC, and AD is produced to F so that AF = 3AD. Prove that the
area of the section of the cube by any plane parallel to BCD is equal to
the area of the section of tetrahedron ABEF by the same plane.

21st British Mathematical Olympiad 1985 ,

1. Circles Sj and S2 both touch a straight line p at the point P. All points of
S2, except P, are in the interior of Sx. The straight line q: (i) is perpendic¬
ular to p\ (ii) touches S2 at R; (iii) cuts p at L; and (iv) cuts at N and
M, where M is between L and R.
(a) Prove that RP bisects Z_ MPN.
(b) If MP bisects A RPL, find, with proof, the ratio of the areas of the
circles Sx and S2.

2. Given any three numbers a, b, and c between 0 and 1, prove that not all
of the expressions a( 1 - b), Ml - c), and c(l - a) can be greater than
 20th BMO, 1984 63

3. Given any two non-negative integers n and m with n^m, prove that

U)^("»1)+3("m2) + -+(» + 1-">(S)-U:l)-

[Note: denotes the binomial coefficient

r(r - l)(r - 2).. .(r - s + 1)/^!.]

4. The sequence fn is defined by fQ = 1 and fx = c, where c is a positive
integer, and for all n> 1 by

fn = 2fn-l~fn-2 + 2-

Prove that, for each k > 0, there exists h such that fkfk + x =fh.

5. A circular hoop of radius 4 cm is held fixed in a horizontal plane. A

cylinder with radius 4 cm and length 6 cm rests on the hoop with its axis
horizontal, and with each of its two circular ends touching the hoop at two
points. The cylinder is free to move subject to the condition that each of
its circular ends always touches the hoop at two points. Find, with proof,
the locus of the centre of one of the cylinder’s circular ends.

6. Show that the equation x2 + y2=z5 +z has infinitely many solutions in

positive integers x, y, and z having no common factor greater than 1.

20th British Mathematical Olympiad, 1984

1. P, Q, and R are arbitrary points on the sides BC, CA, and AB respec¬
tively of triangle ABC. Prove that the triangle the vertices of which are
the centres of the circles AQR, BRP, and CPQ is similar to triangle ABC.

2. Let an be the number of binomial coefficients | j

” (0 < r < n) which leave
remainder 1 on division by 3, and let bn be the number which leave
remainder 2. Prove that an > bn for all positive integers n.

3. (a) Prove that, for all positive integers m,

2m — 1
<mt
m

(b) Prove that, if a, b, c, d, and e are positive real numbers, then

64 The problems

4. Let TV be a positive integer. Determine, with proof, the number of

solutions x in the interval 1 < x < N of the equation x2 — [x2] = (x— [x])2.
[Note: For a real number x, [x] denotes the largest integer ^x.]
5. A plane cuts a right circular cone with vertex V in an ellipse E, and meets
the axis of the cone at C. The point A is an extremity of the major axis of
E. Prove that the area of the curved surface of the slant cone with V as
vertex and E as base is (VA/AC) X (area of E).

6. Let a and m be positive integers. Prove that if there exists an integer x

such that a2x — a is divisible by m, then there exists an integer y such that
both a2y — a and ay2 -y are divisible by m.

7. The quadrilateral ABCD has an inscribed circle. To the side AB we

associate the expression uAB= p/s\n A DAB) + p2{s\n A ABC), where px
and p2 are the lengths of the perpendiculars from A and B respectively to
the opposite side CD. Define uBC, uCD, and uDA similarly, using in each
case the perpendiculars to the opposite side. Show that uAB = uBC =
UCD UDA •

19th British Mathematical Olympiad, 1983

1. In triangle ABC with circumcentre O, AB=AC, D is the midpoint of

AB, and E is the centroid of triangle ACD. Prove that OE is perpendicu¬
lar to CD.

2. The Fibonacci sequence {Fn} is defined by

^ = 1, F2 = 1, Fn=Fn_1+Fn_2 (»> 2).

Prove that there are unique integers a, b, and m such that 0 < a < m,
0 < b < m, and Fn - anbn is divisible by m, for all positive integers n.

3. Given any real number a * -1, the sequence xv x2, x3,... is defined by

*i=«, and xn + x=x2n+xn for all n^l.

Let yn = 1/(1 +xn)■ Let Sn be the sum, and let Pn be the product, of the
first n terms of the sequence y1,y2,y3,.,. . Prove that aS +P = 1 for
all n. n

4. The two cylindrical surfaces

x2 + z2 = a2, z > 0, \y\<a,

and

y2 +z2 = a2, z > 0, x\

 18th BMO, 1982 65

intersect. Together with the plane z — 0 they enclose a dome-like shape

which we shall call a cupola. The cupola is placed on top of a vertical
tower of height h, the horizontal cross-section of which is a square of side
2a. Find the shortest distance over the surface of the cupola and tower
from the highest point of the cupola to a corner of the base of the tower.

5. Given ten points inside a circle of diameter 5, prove that the distance
between some pair of the given points must be less than 2.

6. Consider the equation

(*) A2p + 1 -x2) + \/(3x +p + 4) = \/(x2 + 9x + 3p + 9),

in which x and p are real numbers and the square roots are to be real
(and non-negative). Show that if x and p satisfy (*), then

(x2 + x-p)(x2 + 8x + 2p + 9) = 0.

Hence find all real numbers p for which (*) has just one solution x.

18th British Mathematical Olympiad 1982 ,

1. The convex quadrilateral PQRS has area A; O is a point inside PQRS.
Prove that if

2 A = OP2 + OQ2 + OR2 + OS2,

then PQRS is a square with O as its centre.

2. When written in base 2, a multiple of 17 contains exactly three digits 1.

Prove that it contains at least six digits 0, and that if it contains exactly
seven digits 0, then it is even.

3. If sn = l + ^ + f + ^ + ... + ^ and n > 2, prove that

n(n + 1— n < sn < n — (n — 1 )nb,

where a and b are given in terms of n by an = 1 and b(n — 1) = — 1.

4. For each choice of real number ux, a sequence ul,u2,u3,... is defined by

the recurrence relation uin—un_x + \f (n > 2). By considering the curve
x3 =_y + H, or otherwise, describe, with proof, the behaviour of un as n
tends to infinity.

5. A right circular cone stands on a horizontal base of radius r. Its vertex V

is at a distance l from each point on the perimeter of the base. A plane
section of the cone is an ellipse with lowest point L and highest point H.
66 The problems

On the curved surface of the cone, to one side of the plane VLH, two
routes, Rx and R2, from L to H are marked: R1 follows the semi¬
perimeter of the ellipse, while R2 is the route of shortest length. Find the
condition that Rx and R2 intersect between L and H.

6. Prove that the number of sequences ax, a2,..., an, with each a, = 0 or 1

and containing exactly m occurrences of ‘OF, is I + i r

17th British Mathematical Olympiad 1981 ,

1. The point H is the orthocentre of triangle ABC. The midpoints of BC,
CA, and AB are A', B', and C respectively. A circle with centre H cuts
the sides of triangle A'B'C' (produced if necessary) in six points: Dx and
D2 on B'C'; Ex and E2 on C'A'; and Fx and F2 on A’B’. Prove that
ADX =AD2 = BEX = BE2 = CF1 = CF2.

2. Given positive integers m and n, Sm is equal to the sum of m terms of the

series

(h + 1) — (h + 1 )(n + 3)

+ (n + 1 )(n + 2)(n + 4) — (n + 1 )(n + 2)(n + 3)(n + 5) + ... ,

the terms of which alternate in sign, with each term (after the first) equal
to the product of consecutive integers with the last but one integer
omitted. Prove that Sm is divisible by m!, but not necessarily by m!(n + 1).

3. Let a, b, and c be any positive numbers. Prove that:

(a) a3 +b3 + c3 > a2b + b2c + c2a;
(b) abc > (a + b — c\b + c — a\c + a — b).

4. Given n points in space such that no plane passes through any four of
them, let S be the set of all tetrahedra the vertices of which are four of
the n points. Prove that a plane which does not pass through any of the n
points cannot cut more than n2(n — 2)2/64 of the tetrahedra of S in
quadrilateral cross-sections.

5. Find, with proof, the smallest possible value of |12m - 5"|, where m and n
are positive integers.

6. Given distinct non-zero integers at (1 < i < n), let p, be the product of
all the factors (a, — a,),(a, — u2),...,(a, — an) except for the zero factor
(u, - a,). Prove that if k is a non-negative integer, E"= x al^/pi is an integer.
 15th BMO, 1979 67

16th British Mathematical Olympiad, 1980

1. Prove that the equation xn +yn = zn, where n is an integer >1, has no
solution in integers x, y, and z with 0 <x^n and 0 <y < n.

2. Find a set S of seven consecutive positive integers for which a polynomial

Pix) of degree 5 exists with the following properties:
(a) all of the coefficients in P(x) are integers;
(b) P(n) = n for five numbers n e S, including the least and the greatest;
(c) Pin) = 0 for some n e S.

3. Given a semi-circular region with diameter AB, P and Q are two points
on the diameter AB, and R and S are two points on the semi-circular arc
such that PQRS is a square. C is a point on the semi-circular arc such that
the areas of the triangle v4.BC and the square PQRS are equal. Prove that
a straight line passing through one of the points R and S and through one
of the points A and B cuts a side of the square at the incentre of the
triangle.

4. Find the set of real numbers a0 for which the infinite sequence {a„} of real
numbers defined by an + 1 = 2n — 3an in > 0) is strictly increasing. [Note:
The sequence {an} is strictly increasing if an <an + l for all n > 0.]

5. In a party of ten people, you are told that among any three people there
are at least two who do not know each other. Prove that the party contains
a set of four people, none of whom knows the other three.

15th British Mathematical Olympiad, 1979

1. Find all triangles v4BC for which AB + AC = 2 and AD + BD = >/5, where

AD is the altitude through A, meeting BC at the point D.

2. From a point O in three-dimensional space three given rays, OA, OB, and
OC, emerge, with Z. BOC = a, L.COA = /3, and ZAOB = y, 0 < a, (5, y <
tt. Prove that, given 2s > 0, there are unique points X, Y, and Z on OA,
OB, and OC respectively such that the triangles YOZ, ZOX, and XOY
have the same perimeter 2s. Express OX in terms of 5, sin(a/2), sin( (5/2),
and sin(y/2).

3. S = [ax, a2,..., an) is a set of distinct positive odd integers. The differences
|a, - aj\ (!<*'<;< n) are all distinct. Prove that

£ a, > \nin2 + 2).

/=!
 The problems
68

4. The function / is defined on the rational numbers and takes only rational
values. For all rationals x and y,

fix+fiy)) = fix)fiy).

Prove that / is a constant function.

5. If n is a positive integer, denote by pin) the number of ways of expressing

n as the sum of one or more positive integers. Thus pi4) = 5, as there are
five different ways of expressing 4 in terms of positive integers; namely,

1 + 1+ 1 + 1, 1 + 1+2, 1 + 3, 2 + 2, and 4.

Prove that pin + 1) — 2pin) + pin — 1) ^ 0 for each n > 1.

6. Prove that there are no prime numbers in the infinite sequence

10001,100010001,1000100010001,... .

14th British Mathematical Olympiad, 1978

1. Determine, with proof, the point P inside a given triangle ABC for which
the product PLPMPN is a maximum, where L, M, and N are the feet
of the perpendiculars from P to BC, CA, and AB respectively.

2. Prove that there is no proper fraction m/n with denominator n < 100, the
decimal expansion of which contains the block of consecutive digits T67’,
in that order.

3. Show that there is one and only one sequence {un} of integers such that

ul = l, ul<u2, and u3n + 1 = un_lun + x foralln>l.

4. An altitude of a tetrahedron is a straight line through a vertex which is

perpendicular to the opposite face. Prove that the four altitudes of a
tetrahedron are concurrent if and only if each edge of the tetrahedron is
perpendicular to its opposite edge.

5. Inside a cube of side 15 units there are 11000 given points. Prove that
there exists a sphere of unit radius containing at least six of the given
points.
 12th BMO, 1976 69

6. Show that if n is a non-zero integer, 2 cos n9 is a polynomial of degree n

in 2 cos 6. Hence or otherwise, prove that if k is rational, then cos kit is
either equal to one of the numbers 0, ± or ±1, or is irrational.

13th British Mathematical Olympiad, 1977

1. A non-negative integer fin) is assigned to each positive integer n in such a

way that the following conditions are satisfied:
(a) fimn) =f(m) + fin), for all positive integers m, n\
(b) f{n) = 0 whenever the units digit of n (in base 10) is a ‘3’; and
(c) /(10) = 0.
Prove that fin) = 0, for all positive integers n.
2. The sides BC, CA, and AB of a triangle touch a circle at X, Y, and Z
respectively. Prove that the centre of the circle lies on the straight line
through the midpoints of BC and of AX.
3. (a) Prove that if x, y, and z are non-negative real numbers, then
xix -y)ix -z) + y(y -z)iy -x) + ziz -x)iz -y) > 0.
(b) Hence or otherwise, show that, for all real numbers a, b, and c,

a6 + b6 + c6 + 3a2b2cz > 2ib3c3 + c3a3 + a3b3).

4. The equation x3 + qx + r = 0, where r ^ 0, has roots u, v, and w. Express
the roots of the equation r2x3 + q3x + q3 — 0 in terms of u, v, and w.
Show that if u, v, and w are real, then this latter equation has no root in
the interval —1 <x < 3.
5. The regular pentagon A1A2A3A4A5 has sides of length 2a. For each
i = 1,2,...,5, Kt is the sphere with centre At and radius a. The spheres
K1,K2,..., K5 are all touched externally by each of the spheres Px and P2
—also of radius a. Determine, with proof, whether or not Pj and P2 have
a common point.
6. The polynomial 26ix+x2 +x3 + ... +xn), where n > 1, is to be decom¬
posed into a sum of polynomials, not necessarily all different. Each of
these polynomials is to be of the form axx + a2x2 + a2x3 + ... +anxn,
where each a, is one of the numbers 1,2,3,...,n and no two a, are equal.
Find all the values of n for which this decomposition is possible.

12th British Mathematical Olympiad, 1976

1. Find, with proof, the length d of the shortest straight line which bisects the
area of an arbitrary given triangle, but which does not pass through any of
70 The problems

the vertices. Express d in terms of the area A of the triangle and one of its
angles. Show that there is always a shorter line (not straight) which bisects
the area of the given triangle.

2. Prove that if x, y, and z are positive real numbers, then

x y z 3

y+z z+x x+y 2

3. SVS2,---, S50 are subsets of a finite set E. Each subset contains more than
half the elements of E. Show that it is possible to find a subset F of E
having not more than five elements, such that each 5) (1 < i < 50) has an
element in common with F.

4. Prove that if n is a non-negative integer, then 19 X 8" + 17 is not a prime

number.

5. Prove that, if a and /3 are real numbers, and r and n are positive integers
with r odd and r < n, then

(r-1)/2 , \ / \

(r~ l)/2 / \/ \
- E (r-,)(r7,)(“/’)'(o + <,) •
[Note: j denotes the coefficient of xs in the expansion of (1 +x)m.]

6. A sphere with centre O and radius r is cut by a horizontal plane distance

r/2 above O in a circle K. The part of the sphere above the plane is
removed and replaced by a right circular cone having K as its base and
having its vertex V at a distance 2r vertically above O. Q is a point on
the sphere on the same horizontal level as O. The plane OVQ cuts the
circle K in two points, X and Y, of which Y is the further from Q. P is
a point of the cone lying on VY, the position of which can be determined
by the fact that the shortest path from P to Q over the surfaces of
cone and sphere cuts the circle K at an angle of 45°. Prove that VP =
V'3-r/v'(l + 1/V5). [Note: In a spherical triangle ABC, the sides are arcs of
great circles, centre O, and the sides are measured by the angles that they
subtend at O. You may find the following spherical triangle formulae
useful: sin a/sin A = sin b/sin B = sin c/sin C; cos a = cos b cos c +
sin b sin ccos A.\
 11th BMO, 1975 71

11th British Mathematical Olympiad, 1975

1. Given that x is a positive integer, find (with proof) all solutions of

vT + V2 + + ]/(x3 - 1) 400.

[Note: [z] denotes the largest integer <z.]

2. The first n prime numbers 2,3,5,...,pn are partitioned into two disjoint
sets A and B. The primes in A are ax,a2,...,ah, and the primes in B are
bv b2,..., bk, where h + k = n. The two products
h k
n and n bf>
i =1 ( = 1

are formed, where the ct, and /3( are any positive integers. If d divides the
difference of these two products, prove that either d = 1 or d >pn.

3. Use the pigeonhole principle to solve the following problem. Given a point
O in the plane, the disc S with centre O and radius 1 is defined as the set
of all points P in the plane such that \OP\ < 1, where OP is the distance
of P from O. Prove that if S contains seven points such that the distance
from any one of the seven points to any other is > 1, then one of the seven
points must be at O. [Note: The pigeonhole principle states that if more
than n objects are put into n pigeonholes, some pigeonhole must contain
more than one object.]

4. In a triangle ABC, three parallel lines AD, BE, and CF are drawn,
meeting the sides BC, CA, and AB in D, E, and F respectively. The
points P, Q, and R are collinear and divide AD, BE, and CF respectively
in the same ratio k: 1. Find the value of k.

5. Let m be a fixed positive integer. You are given that

(20m) + (2r)COS9+(22m)COS2#+-+(2m)COs2me

= (2 cos \ 6 )2m cos m 6,

where there are 2m + 1 terms on the LHS; the value of each side of this
identity is defined to be The function gm(6) is defined by

«"<9,= (2(T) + (2f )COS2e+ (2r)COs49+ ■■■+[2m)COS2me-

Given that there is no rational k for which a = kit, find the values of a
for which limm^ Jgm(at)/fm(a)] =
72 The problems

6. Prove that, if n is a positive integer greater than 1, and x >y > 1, then

xn+'-l ^ yn+1 - 1 •
x(xn~x - 1) > y(yn~l ~ 1)
7. Prove that there is only one set of real numbers x^, x2,..., xn such that
2 1
(1 -Xj)2 3 4 5 + Uj ~X2)2 + ...+(*„_!-*„) + *« = n + 2 •

8. The interior of a wine glass is a right circular cone. The glass is half filled
with water and is then slowly tilted so that the water reaches a point P on
the rim. If the glass is further tilted (so that water spills out), what fraction
of the conical interior is occupied by water when the horizontal plane of
the water level bisects the generator of the cone furthest from F?

10th British Mathematical Olympiad, 1974

1. The curves A, B, and C are related in such

a way that B ‘bisects’ the area between A
and C; that is, the area of the region U is
equal to the area of the region V at all
points P of the curve B. Find the equation
of the curve B given that the equation of
curve A is y = jx2 and that the equation of
curve C is y = \x2.

2. A domino can be represented by an unordered pair of integers. Thus

. 'can be represented as (1,5) or (5,1) and the double * .

as (2,2). The set of all 15 dominoes containing two integers from 1, 2, 3,

4, and 5 is partitioned into three subsets of five dominoes. The dominoes
in each subset form a closed chain; that is, (a,bXb,c)(c,dXd,eXe,a),
where a, b, c, d, and e need not all be different. How many distinct
partitions are there? (The order of the three subsets in the partition is
immaterial.)

3. Prove that it is impossible for all the faces of a convex polyhedron to be

hexagons.

4. M is a 16 X 16 matrix. Each entry in the leading diagonal and each entry

in the bottom row (that is, the 16th row) is 1. Every other entry of the
matrix is Find the inverse of M.

5. A bridge deal is defined as the distribution of 52 ordinary playing cards

among four players, so that each player has 13 cards. In a bridge deal,
 10th BMO, 1974 73

what is the probability that just one player has a complete suit? (Leave
your answer in factorials.)

6. The points X and Y are the feet of the perpendiculars from P to CA and
CB respectively, where P is in the plane of triangle ABC. If PX = PY,
and the straight line through P which is perpendicular to AB cuts XY at
Z, prove that CZ bisects AB.
1. The roots of the equation x3 = bx + c (be # 0, b and c real) are a, (3,
and y. If

(3 = pa2 + qa + r, y = p(32 + q (3 + r, a=py2 + qy + r,

determine p, q, and r in terms of b and c. State a condition which

ensures that p, q, and r are real.
8. Let n be an odd prime number. It is required to write the product
n- 1
nu+o
i=i

as a polynomial
n—l

L aixi-
7=0

By considering the product n”=1U + i) in two ways, establish the

relations

an-2 = n(n ~ D/2\,

2an_3 = n(n - Din - 2)/3! + an_2(n - Din - 2)/2!,

(n — 2)ay = n + an_2(n - 1) + an_3(n - 2) + ... +3a2,

(n- 1 )a0 = 1 +an_2 + ... +av

Prove that n \ a; (;' = 1,2,..., n - 2) and that n\(a0 + 1). Prove also that,
when x is an integer,

n\(x+ DU + 2)...(x + n - 1) -xn_1 + 1.

Hence deduce Wilson’s Theorem and Fermat’s Theorem; namely, that
when n is prime and x is not a multiple of n:
(a) n | (n — 1)! + 1;
(b) n \xn~1 — 1.
[ Note: p | q means p divides q leaving no remainder.]
74 The problems

9. A vertical uniform rod of length 2 a is hinged at its lower end to a

frictionless joint secured to a horizontal table. It falls from rest in this
unstable position on to the table. Find the time occupied in falling.
Comment on your answer. [Note: You may quote the result
/(cosec x)dx = log |(tan \x)\ if you wish.]

10. A right circular cone the vertex of which is V and the semi-vertical angle
of which is a has height h and uniform density. All points of the cone
the distances of which from V are less than a or greater than b, where
0 < a < b < h, are removed. A solid of mass M is left. Given that the
gravitational attraction that a point mass m at P exerts on a unit mass at
O is (Gm/OPz)OP, prove that the magnitude of the gravitational attrac¬
tion of this solid on a unit mass at V is

§GM(1 + cos a)/(a2 + ab + b2).

9th British Mathematical Olympiad, 1973

1. (a) Two fixed circles are touched by a variable circle at P and Q. Prove
that PQ passes through one of two fixed points.
(b) State a true theorem about ellipses (or if you like about conics in
general) of which the result in part (a) is a particular case.

2. Nine points are given in the interior of the unit square. Prove that there
exists a triangle of area < | the vertices of which are three of the points.

3. A curve consisting of the quarter circle x2+y2=r2, x, y > 0, together

with the line segment x = r, — h < y < 0, is rotated about x = 0 to form a
surface of revolution which is a hemisphere on a cylinder. A string is
stretched tightly over the surface from the point on the curve
(rsin 0,r cos 6) to the point (— r, -h) in the plane of the curve. Show
that the string does not lie in a plane if tan 6>r/h. [Note: You may
assume spherical triangle formulae such as cos a = cos b cos c +
sin b sin c cos A, or sin A cot B = sin c cot b — cos c cos A. In a spherical
triangle the sides a, b, and c are arcs of great circles and are measured
by the angles that they subtend at the centre of the sphere.]

4. You have a large number of congruent equilat¬

eral triangles on a table and you want to fit n of
these together to make a convex equiangular
hexagon (that is, one the interior angles of which
are each 120°). Obviously, n cannot be any posi¬
tive integer. The smallest n is 6, the next smallest
is 10, and the next 13. Determine conditions for
possible n.
 9th BMO, 1973 75

5. There is an infinite set of positive integers of the form 2" - 3 with the
following property Q:

no two members of the set have a common prime factor.

Here is an outline of a proof of this fact.

Suppose that there is a finite set S = {2m‘ - 3,2m2 - 3,..., 2m* - 3}
with property Q having k members. Let the prime factors of
these k numbers be pl,p2,...,pr Consider the number N =
2(Pi-1Xp2-i) - (p,-i)+i gy Fermat’s Theorem, ap~1 = 1 (modp) for

every prime p that does not divide a. Hence N -3 = -l(mod pr),

for each r (1 < r < t), and N — 3 may be added to S to give a larger
set with property Q.
Give a properly expanded and reasoned proof that there is an infinite set
of positive integers of the form 2" - 7 with property Q.

6. In answering general knowledge questions (framed so that each question

is answered ‘yes’ or ‘no’) the teacher’s probability of being correct is a
and a pupil’s probability of being correct is (3 or y according as the pupil
is a boy or a girl. The probability of a randomly chosen pupil agreeing
with the teacher’s answer is Find the ratio of the number of boys to
girls in the class.

7. The life-table issued by the Registrar-General of Draconia shows, out of

each 10000 live births, the number y expected to be alive x years later.
When x = 60, y = 4820. When x = 80, y = 3205. At age 100, all Draconi-
ans are put to death. For 60<x<100 the curve y=Ax( 100—x) +
B/(x — 40)2 fits the figures in the table very closely, A and B being
constants. Determine the life-expectancy (in years correct to one decimal
place) of a Draconian aged 70.

a
8. Call the matrix M = the companion matrix for the mapping
az + b
cz + d
(a) Prove that, if A/, is the companion matrix for T■ (i = 1,2), then M^M2
is the companion matrix for the mapping T{T2.
(b) Find conditions on a, b, c, and d so that T4 = / but T2 #/.

9. Let Lr equal the determinant

x y 1
a + c cos 6r b + c sin 6r 1
l + n cos dr m + n sin 6r 1

Show that the lines Lr = 0, r= 1,2,3, are concurrent and find the
co-ordinates of the point at which they meet.
76 The problems

10. Construct a detailed flowchart for a computer program to print out all
positive integers up to 100 of the form a2 — b2 — c2, where a, b, and c are
positive integers and a ^ b + c. (There is no need to print in ascending
order or to avoid repetitions.)

11. (a) Two uniform rough right circular cylinders, A and B, with the same
length, have radii a and b and masses M and m respectively.
Cylinder A rests with a generator in contact with a rough horizontal
table. Cylinder B rests on A, initially in unstable equilibrium, with its
axis vertically above that of A. Equilibrium is disturbed, B rolls on
A, and A rolls on the table. In the subsequent motion, the plane
containing the axes makes an angle 6 with the vertical. Draw dia¬
grams showing angles, forces, etc., for the period during which there
is no slipping. Write down equations which will give on elimination a
differential equation for 6, stating the principles used. Indicate how
the elimination could be done (you are not asked to do it).
(b) Such a differential equation is, with k = M/m,

6(4 + 2cos 6 — 2cos20 + 9k/2) + 62 sin j(2cos 6—1)

3g(l + &)sin 6
a +b

Obtain 6 in terms of 6. [Note: The moment of inertia of a uniform

cylinder about its axis is j(mass) X (radius)2.]

8th British Mathematical Olympiad, 1972

1. 5 is a set of elements {a,b,...} each pair of which are distinct. R is a

relation holding or not holding between every ordered pair of distinct
elements of S according to the following rules:
(a) either aRh or bRa, but not both;
(b) if aRb and bRc, then cRa.
Find the largest number of elements that S can have, proving your result.

2. Let a, b, c, and d be integers, with a # b. Show that there are at most four
different points on the hyperbola

(x + ay + c)(x + by + d) = 2

each of the co-ordinates of which are integers. Find a set of conditions

which are both necessary and sufficient for there to be four such points.

3. In a plane, two circles of unequal radii intersect at A and B, and through

an arbitrary point P a straight line L is to be constructed so that the two
 8th BMO, 1972 77

circles intersect equal chords on L. By considering distances from the

point of intersection of L with AB (produced if necessary), or otherwise:
(a) find a ‘ruler and compass’ construction for L; and
(b) state the region of the plane in which P must lie for the construction
to be possible.

4. A point P is on a smooth curve in a plane containing two fixed points A

and B; PA = a, PB = b, and AB = c. The angle APB = 0, where c2 = a2 +
b2 — lab cos 6. Show why

sin20(d.s)2 = (da)2 + (db)2 - 2(da)(dfr)cos 6,

where s measures the distance along the curve. The point P moves along
the curve so that at time t during the interval (\T <t <T) its distance
from A and B are given, respectively, by hcos(t/T) and ksinU/T),
where h, k, and T are positive constants. Prove that the speed of P during
this interval varies as cosec 6.

5. In a right circular cone the semi-vertical angle of which is d, a cube is

placed so that four of its vertices are on the base and four on the curved
surface. Prove that as 6 varies the maximum value of the ratio of the
volume of the cube to the volume of the cone occurs when sin 6 =

6. Tk = k — 1, for k = 1,2,3,4. For all k > 3, T2k_l = T2k_2 + 2k~2 and T2k =
T2k_5 + 2k. Show that, for all k > 0,

l + Ta-.-lTZ*-'] and 1 + T2k = [f 2*-'],

where [x] denotes the greatest integer not exceeding x.

7. Starting from px = 2, qx = 1, rx = 5, and sx = 3, sequences of integers are

formed by using the following relations, which hold for all positive integral
n:

Pn+ 1 f 3^„, tfn + 1 ~ ^PnQn’ rn~Pn^',<ln’ Sn ~ Pn + Q.n%

Prove that:
(a) pn/qn > ^3 > rn/sn\
(b) p„/qn differs from i/3 by less than sn/(2rnq2).

8. Three children, A, B, and C, build cairns of stones on a beach and try to

knock down each other’s cairns. Each time a child aims at a cairn, he has a
fixed chance of hitting it: A has a chance of f of hitting the cairn at which
he aims, B a chance of f, and C a chance of Shots aimed at one cairn
do not hit another. A player whose cairn is hit is out of the game. A game
is won by being the only remaining player. The rule of play is for the
players in the game to throw simultaneously, choosing with equal probabil¬
ity between his two opponents’ cairns (when there are two). Find the
probability that C wins, with A being eliminated in the first round.
78 The problems

9. A rocket, free of gravitational forces, is accelerating in a straight line. Its

mass is M and that of its fuel is m=mQe~kt. The relative velocity of
escape of the fuel is v = v0e~kt. If m0 is small compared with M, show
that the terminal velocity of the rocket is approximately \m0v0/M greater
than its initial velocity.

7th British Mathematical Olympiad, 1971

1. (a) Factorize (x +y)7 - (x7 +y7).

(b) Prove that there is no integer n for which 2n3 + 2n2 + 2n + 1 is a
multiple of 3.

2. x1 — 9 and xr+1 = 9Xr for all positive integral r. Prove that the last two
digits of x3 written to base 10 are the same as the last two digits of x4
written to base 10. What are these two digits?

3. Of a regular polygon of 2n sides there are n diagonals which pass through

the centre of the inscribed circle. The angles which these diagonals
subtend at two given points A and B on the circumference are
ava2,a3,...,an and bvb2,b3,...,bn. Prove that
n n

X) tan2ai= £ tan2h,.
i= 1 i=l

4. Given a set of n + 1 positive integers, none of which exceeds 2n, prove by

induction or otherwise that at least one member of the set must divide
another member of the set.

5. The triangle ABC has angles A, B, and C, in descending order of

magnitude. Circles are drawn such that each circle cuts each side of the
triangle internally in two real distinct points. The lower limit to the radii of
such circles is the radius of the incircle of the triangle ABC. Show that the
upper limit is not R, the radius of the circumcircle, and find this upper
limit in terms of R, A, and B.

6. (a) I(x) = fcx fix, u)du, where c is a constant. Show why

(b) Find lim0 0 cot 6 sin(t sin 6).

(c) Git) = /o cot d sinO sin d)dd. Prove that G'iTr/2) = 2/77.

7. Two real numbers h and k are given, such that h > k > 0. Find the
probability that two points chosen at random on a straight line of length h
should be at a distance of less than k apart.
 6th BMO, 1970 79

8. A is a 3 X 2 matrix and B is a 2 X 3 matrix. The elements of each matrix

are numbers. AB = M and BA = N. det M = 0 and det N # 0. Also, M2 =
kM, where k is a number. Determine det N in terms of k, proving your
result. You may assume the usual rules for combining matrices. [ Note: k is
not a matrix; kC is defined as the matrix each element of which is k times
the corresponding element of C.]

9. Two uniform solid spheres of equal radii are so placed that one is directly
above the other. The bottom sphere is fixed and the top sphere, initially at
rest, rolls off. If the coefficient of friction between the two spherical
surfaces is n, show that slipping occurs when 2 sin 6 = /x,(17cos 6 - 10),
where 0 is the angle that the line of centres makes with the vertical. [ Note:
The moment of inertia of a solid sphere of radius r and mass M about a
diameter is |Mr2.]

6th British Mathematical Olympiad, 1970

1. (a) Express the finite series

1 1 1
~- + -- + ... H-
log2fl log 3# log„fl

as a quotient of logarithms to base 2. Does the series converge as

n * co?
(b) Evaluate the sum of the coefficients of the polynomial

p(x) = (1 +x -x2)3(1 - 3x+x2)2,

and the sum of the coefficients of its derivative.

2. Sketch the curve the equation of which is given by x2 + 3xy + 2y2 + 6x +

12y + 4 = 0. About which point does it possess some symmetry property?

3. You are probably aware of the remarkable fact that the three points of
intersection of three appropriate pairs of trisectors of the internal angles
of any triangle T form the vertices of an equilateral triangle t. Given a
real constant A > 0, consider the class of all triangles T the area of
which equals A. Find, with this class, the maximum value of the area of
the triangle t. Discuss whether there is a minimum value.

4. S is any set of n positive integers the sum of which is not divisible by n.

Prove that it is always possible to choose a subset of S, the sum of the
members of which is divisible by n.

5. A cube of wood is to be sawn up into at least 300 separate pieces. What is

the minimum number of sawcuts which need to be made? (Each sawcut is
80 The problems

planar, and it is assumed that pieces are not moved relative to each other
—so that the block remains intact as a cube until after the last cut has
been made.)

6. In the region |x| < a, y(x) is defined by the differential equation

dy/dx = /(x), where f(x) is a given even, continuous function of x.
Prove that:
(a) y(—a) — 2y(0) + y(a) = 0;
(b) fa_a y(x)dx = 2ay(0).
A numerical step-by-step solution is calculated starting from the point
(-o,0) using 2N equal x-wise steps, according to the usual scheme
g(xn+1) = g(x„) + Sx-g'(xn). Prove that in general this solution will not
satisfy the result (a) above. Suggest a modified step-by-step procedure
which would ensure that (a) is satisfied.

7. The two base angles B and C of an isosceles triangle ABC are equal to
50°. The point D lies on BC, so that Z BAD = 50°, and the point E lies
on AC, so that Z ABE = 30°. Find Z BED.

8. Eight electric light bulbs are arranged in a row, each controlled by its
own on/off switch. State how many different patterns (or ‘states’) of lit
bulbs there are (including the pattern of all bulbs off). Denote the N
distinct states by St, i= 1,2,3Let ni be the number of switches
which have to be altered to change from 5, to Si+l (with the convention
that SN+l = iS^). Find the smallest possible value of n = Efii n,; that is,
find the smallest value for the total number of switch-alterations which
have to be made so as to run through all the possible states of the lights
in some order, before returning to the initial state.

9. Find rational numbers x and y such that

^(2^3 — 3) =x1/4 —y1/4.

10. Unlimited supplies of straight rods (of negligible thickness) are available,
the length of each rod being an integral multiple of 1 m. The rods are laid
out on a large horizontal surface to form as many separate right-angled
triangles as possible. Let MO denote the number of different right-
angled triangles the shortest side of which is a rod of length i metres.
(For values of i for which no such right-angled triangle exists, we define
M0 = 0.)
(a) Find some kind of ‘formula’ for MO.
(b) Prove that, for any given positive integer Af, there exists a value of i
such that MO > M.
(c) Discuss whether M0 ->00 as i -> oo.
 5th BMO, 1969 81

5th British Mathematical Olympiad, 1969

1. Find the condition on the distinct real numbers a, b, and c such that
(x — a)(x — b)/(x — c) takes all real values for real values of x. Sketch
two graphs of (x - a\x - b)/(x - c) to illustrate:
(a) a case in which the condition is satisfied; and
(b) a case in which the condition is not satisfied.

2. Find all the real solutions of cos x + cos5x + cos7x = 3.

3. A square piece of plywood is cut up into n2 equal square pieces. These

are then rearranged to make four rectangles with one square piece left
over, in such a way that the nine dimensions of the five shapes (including
the edge length of the square piece) are all different. Find the minimum
value of n for this situation to be possible, and specify the shapes of the
four rectangles for this value of n. State all other values of n for which
the situation is possible, justifying your answer.

4. Find all possible pairs (a, b) of integers satisfying a2 - 3ab - a + b = 0.

5. A long corridor of unit width has a right-angled corner in it. A rigid

length of pipe (the thickness of which may be neglected) lies on, and is
everywhere in contact with, the plane floor of the corridor. The length of
the pipe (which may be curved) is defined as the straight-line distance
between its two ends. Find the maximum length of pipe subject to the
condition that it can be moved along both arms of the corridor and round
the corner without leaving contact with the floor.

6. Prove that, if a, b, c, d, and e are positive integers, any common factor of

(ae + b) and (ce + d) is also a factor of (ad — be).

7. Let f(x) denote a real-valued function of the real variable x, not

identically zero, and differentiable at x = 0.
(a) If f(x)-f(y) =f(x+y) for all real values of x and y, prove that /(x)
is differentiable any number of times for all x, and that X/=o /DO =
1/(1 -/(l)) when /(1) < 1.
(b) If f(x)-fiy) +f(x—y) for all real values of x and y, find /(x).

8. A square has sides of length x. All four vertices of the square lie on the
sides of a circumscribing triangle. The incircle of the triangle has radius
r. Prove that 2r > x > /2 • r.

9. Let Pin) be the proposition that ‘a polygon of n2 sides can be drawn so

that its vertices lie at the n2 points of a square n X n array’. State, with
proof, the truth or otherwise of P(4) and P(5). Formulate, with some
justification, a conjecture about the truth of Pin) for n> 6. [Note: A
polygon satisfies the following common-sense properties: (a) no two sides
82 The problems

coincide; (b) no two sides intersect unless they are adjacent, non-parallel
sides, in which case their intersection is one of the polygon’s vertices.]

10. Describe a ruler-and-compass method of constructing an equilateral

triangle the area of which equals that of a given triangle.

4th British Mathematical Olympiad, 1968

1. A circle C of unit radius rolls without slipping along the outside of the
circle with centre the origin and radius 2 in the (x, y)-plane. A fixed point
P of C is originally at the position (2,0). Find equations, giving the
co-ordinates (x, y) of P in terms of a suitable parameter 6, as C rotates.
Sketch the locus S described by P, showing all tangents parallel to the x-
and y-axes.

2. Cows are put out to pasture when the grass has reached a certain height.
Thereafter, as the cows eat the grass, the grass continues to grow as the
pasture is consumed. If 15 cows can consume the grass in three acres of
pasture in four days, while 32 cows can consume the grass in four acres in
two days, how many cows will be required to consume the grass in six
acres in three days?

3. The ‘distance’d between two points (x1; yx) and (x2, y2) in the (x, y)-plane
is defined by

d = |x2-Xj \ + \y2-y1 |.

Using this notion of distance, find the locus of all points (x,y) satisfying
x > 0, y>0, and which are equidistant from the origin and from a fixed
point (a, b) in the plane, where a > b. Distinguish the cases according to
the signs of a and b.

4. Two spheres of radii a and b are tangent to each other and a plane is
tangent to these spheres at different points. Find the radius of the largest
sphere which can pass between the first two spheres and the plane.

5. Given x, y, and z such that

sin x + sin y + sin z = 0

and
cos x + cos y + cos z = 0,
prove that

sin 2x + sin 2y + sin 2z = 0

and
cos2x + cos 2 y + cos 2 z = 0.
 3rd BMO, 1967 83

6. If ax,a2,...,a7 are integers, and bx,b2,...,b-1 are the same integers

rearranged, show that the value of

(ax - bx)(a2 - b2)...(a7 - b7)

is necessarily even.

7. A knock-out ping-pong tournament is arranged among n people and a

new ball is used for each game. How many balls are needed?

8. A chord of length / divides the interior of a circle of radius r into two

regions, Dx and D2. A circle S of maximal radius is inscribed in Dx; the
area of the part of Dx outside S is A. Show that A is greatest when the
area of Dx exceeds that of D2, and when

1677T
/=-7.
16 + 7T 2

9. Find the lengths of the sides of a triangle if its altitudes have lengths 3, 4,
and 6

10. The faces of a tetrahedron are formed by four congruent triangles. If a

is the angle between a pair of opposite edges of the tetrahedron, show
that

sin(5 - C)
cos a =
sin(5 + C) ’

where B and C are the angles adjacent to one of these edges in a face of
the tetrahedron.

11. The sum of the reciprocals of a set of n different positive integers is

equal to one. If n = 3 show that there is only one such set and find it.
Find such a set for n = 4, 5, and more generally for any value of n > 3.

12. Find, with proof, the maximum number of points which can be placed on
the surface of a sphere of unit radius such that the distance between any
two of the points is:
(a) at least ^2;
(b) greater than 'J2.

3rd British Mathematical Olympiad, 1967

1. If a and [3 are the roots of x2 +px + 1 = 0, and y and 8 are the roots of
x2 + qx+ 1 = 0, show that

(a — y)(/3 — y)(a + 8)( (3 + 8) = q2 -p2.

84 The problems

2. By putting y = tx, or otherwise, draw the graph of x8 + xy + ys - 0. Show

clearly all turning values of x and y, and the behaviour of the graph
when x and y become large.
3. (a) The lengths a and b of two sides of a triangle satisfy a > b. The
lengths of the corresponding altitudes are ha and hb. Prove that
a + ha>b + hb. Discuss all cases in which equality occurs.
(b) If the circumcentre and the incentre of a triangle A coincide, prove
that A is equilateral.
4. When is it possible to find points equidistant both from two given points
P and Q and from a given straight line ABl When it is possible, show
how to construct such points with ruler and compasses only.

5. Show that fit) = (t - sin tXir-1 - sin t) is increasing in the interval

0 < t < 7t/2.
6. Find all real numbers x in the interval 0 < 27r such that

2cos x < |7(1 + sin2x) — 7(1 — sin2x)| < 72.

[Note: If t is any positive number, 7f is the positive square root of t. If x
is any real number, |x| = 7(x2).]
7. Four real numbers xx, x2, x3, and x4 are such that the sum of any one of
them with the product of the other three is equal to 2. Find all possible
solutions.
8. Find all non-negative integers n such that 5” — 4" is divisible by 61.
9. If a, (3, and y are the angles of a triangle with af3y#0, show that
cos2a + cos2/3 + cos2y ^ f. Find all such triangles for which equality holds.
Show also that cos2a + cos2j3 + cos2y does not have a greatest value.
10. Alan, Brian, Colin, and David collect stamps. Alan collects British stamps
issued before 1900 and foreign stamps. Brian collects British stamps
issued after 1900 and foreign special issues. Colin collects foreign
stamps issued before 1900 and British special issues. David collects
foreign stamps issued after 1900 and British special issues. Are there any
stamps which no-one collects? Are there any stamps which everyone
collects? What stamps could David give away that would be of interest to
Alan but not to Brian?

11. In a certain town, the streets are arranged in a rectangular grid. A man
wishes to go from one place to another place m streets east and n streets
north. How many shortest paths are there?

2nd British Mathematical Olympiad, 1966

1. Find the least and the greatest values of (xA +x2 + 5)/(x2 + l)2 for real
values of x.
 2nd BMO, 1966 85

2. Give the conditions under which all the roots of the equations

±/U-a) ± ij(x-b) ± A* - c) = 0

are real, where a, b, and c are distinct real numbers.

3. Sketch the graph of y2 =x2(x + l)/(x — 1). Find all turning values of y,
and describe the behaviour of the graph when x and y become large.

4. The points A, B, C, and D are four consecutive vertices of a regular

polygon. If

1 11
AB ~ AC + AD’

how many sides must the polygon have?

5. A square nut of side a is to be turned by means of a spanner, the hole in

which consists of a regular hexagon of side b. Find the conditions on a
and b for this to be possible.

6. Find the largesl interval of values x for which the expression

y = A* — 1) + v/[x + 24 — 10 A* — 1)]
has a constant value. [Note: is defined for t > 0 to be that non-negative
number whose square is f.]

7. Prove that A, ^3, and A cannot be terms of the same arithmetic

progression.

8. The faces of a cube are coloured by six colours, so that each face has a
different colour. Find how many cubes of distinct appearance can be
produced in this way. Show also that 1680 regular octahedra of distinct
appearance can be produced by colouring the eight faces of a regular
octahedron with eight given colours so that each face has a different
colour.

9. If a, f3, and y are the angles of a triangle, find:

(a) the smallest possible value of tan(a/2) + tan(/3/2) + tan(y/2);
(b) the largest possible value of tan(a/2)tan(/3/2)tan(y/2).

10. One hundred students, all of different heights, are arranged in a square
of 10 rows by 10 columns. In each row the tallest student is selected, and
the shortest of these tall students is labelled A. In each column the
shortest student is selected, and the tallest of these short students is
labelled B. If A and B are different persons, find which of them is the
taller and why.

11. (a) Prove that, among any 52 integers, there must always exist two
integers such that either the sum or the difference of these two
integers is divisible by 100.
86 The problems

(b) Prove that, given any 100 integers, none of which is divisible by 100, it
is possible to find two or more of these integers the sum of which is
divisible by 100.

1st British Mathematical Olympiad, 1965

1. Sketch the graph of y = (x2 + \)/{x + 1). Find all turning values, and
describe the behaviour of the graph when x or y become large.
2. A pupil is swimming at the centre of a circular pond. At the edge of the
pond there is a teacher, who wishes to catch the pupil, but who cannot
swim. The teacher can run four times as fast as the pupil can swim, but
not as fast as the pupil can run. Can the pupil escape from the teacher?
Justify your answer.
3. Prove that, for every integer n:
(a) n3 — n is divisible by 3;
(b) n1 — n is divisible by 7;
(c) n13 — n is divisible by 13.
4. How many zeros are there at the end of the number 100!? [Note: n\
denotes the product of all the integers from 1 to n inclusive.]
5. Prove that the product of four consecutive integers is always one less
than a perfect square.

6. Let (2 +1/2)n = P + F, where P is an integer and 0 < F < 1. Show that the
positive integer n can be chosen so that F > 0.999.
7. Find the remainders upon dividing the polynomial x + x3 + jc8 9 10 + x21
x81+;c243:

(a) by * — 1;
(b) by x2 - 1.

8. Determine all values of the coefficient a for which the equations x2 -+■
*** + 1 = 0 and x2 + x + a = 0 have at least one common root.
9. If a, b, and c are any three positive real numbers, prove that

(a + b)(b + c)(c + a) > 8abc.

Show that equality holds only when a = b = c.

10. A chord of length i/3 divides a circle of unit radius into two regions. Find
the rectangle of maximum area which can be inscribed in the smaller of
these two regions.
 Part III
Hints and outline solutions
' . . .

.
 Hints and outline solutions
It has come to me in a flash! One’s
intelligence may march about and about a
problem, but the solution does not come
gradually into view. One moment it is not.
The next it is there.
William Golding, Rites of Passage

32nd British Mathematical Olympiad 1996 ,

1. Find as efficiently as possible all pairs (m,n) of positive integers
satisfying the following two conditions:
(a) two of the digits of m are the same as the corresponding digits of
n, while the other two digits of m are both 1 less than the
corresponding digits of n\
(b) both m and n are four-digit squares.

(1) It may help to start by tackling the corresponding ‘three-digit’ problem.

Determine all pairs (m, n) of positive integers satisfying:

(a) two of the digits of m are the same as the corresponding digits of n,
while the other digit of m is 1 less than the corresponding digit of n
(as in, say, 263 and 273);
(b) both m and n are three-digit squares.

If m has three digits then m must look like ‘abc\ If n has two digits the
same as m (and in the same positions as m), with the other digit one larger
than for m, then n must look like

‘ab(c + 1)’, or ‘_or ‘-’.

(2) If m =‘abc’ and n = ‘ab(c + 1)’, then the two squares ‘abc’ and ‘ab(c + 1)’
differ by _. Hence the only possibilities for m and n are ‘_’ and
‘_so m is not ‘positive’; (anyway, ‘000’ and ‘001’ are not really
‘three-digit’ numbers). In the second case the two squares ‘abc’ and ‘a(b + l)c’
differ by exactly _; a short check should convince you that there are no
90 Hints and outline solutions

solutions. In the third case the two squares differ by exactly-, and there
is just one solution: namely m =_, n =_.
(3) Before moving on to the ‘four-digit’ problem, you should try to improve
on the above trial-and-error approach. How can you be sure that there is no
solution in the second case (that is, with n — m = 10)? How do you know that
there is only one solution in the third case? The key to a more mathematical
approach lies in writing the squares m and n as squares: m =x2 and n =y2.
Then

n - m = y2 -x2 = (_)(-).

There is more to this factorization than meets the eye, for the two factors on
the RHS differ by 2x, which is e*e*. Hence either:

(3.1) both factors must be o**, so n — m is odd; or

(3.2) both factors must be e*e*, so n — m is a multiple of *ou*.

Now check the three cases in (2) properly. In the first case,

n — m = 1; :.y—x=\=y+x => y =_, x =_

In the second case,

n - m = 10, which is e*e*, but not a multiple of *ou*;

hence there are no solutions.

In the third case,

n — m = 100;

.-. either y -x=y +x =_, so x = 0 and m is not ‘positive’;

or y — x — _____, y + x =_(since y — x and y + x must both be e*e*),
y =_, x =_, so m =_, n =_ is the only solution.

(4) If m and n both have four digits, and satisfy condition (a) in the question,
then there are exactly *i* possible positions for the pair of digits common to
both m and n. Hence, if m has the form ‘abed’, n has the form

lab{c + \){d + 1)’, or ‘_’, or ‘_’, or ‘_

or *_or ‘_’.

You must now analyse each of these six cases separately using the factoriza¬
tion in (3) above, and the conditions (3.1) and (3.2). (Note: In each case there
may be no solution, or one solution, or more than one solution.)
 32nd BMO, 1996 91

2. A function / is defined for all positive integers and satisfies

/(1) = 1996,
and
/(1) +/(2) + ... +f(n) = n2f(n) for all n > 1.

Calculate the exact value of /(1996).

(1) If you have never seen anything like this before, it seems reasonable to
start calculating: /(1) = 1996 (given); /(2) =_ (from /(1) +/(2) = 22/(2));
/(3) =_, and so on. Use this approach to find /(2), /(3), and /(4).
(2) If you persevere, and keep your wits about you, you might just notice
something interesting (though you must be careful not to let the numbers
1996, 1, 2, 3, 4, etc. obscure what is going on). If you are lucky, you may even
be able to guess a value for /(1996). But this would not answer the question!
In mathematics, the word ‘determine’ means more than just ‘guess’ or ‘find’;
it means that you have to show exactly why your value is correct. In other
words, you have to ‘find the correct value, andproue it is correct’’. For this, it is
not the values of /(2), /(3), and so on, that matter, but their form. Thus it is
important to express /(2) and /(3) in a form that reveals what is really going
on:

/( 2) = •/(l),
(22 - 1)

1
/(3) = 2 i [/(l)+/(2)] = ^ /(1) + /(l)
3—1 3—1 (22 - 1)

1
•/( 1).
(22 — 1) (32 - 1)

Write out the calculation which shows that

1
/(4) = •/(l).
(22 — 1) (32 - 1) (42 — 1)

(3) Now guess what you expect to be the corresponding expression for f(n) in
terms of /(1), and prove that your guess is correct (by induction on n).
(4) Even at this stage it is important to resist the temptation simply to
substitute n = 1996. Factorize each of the factors (r2 - 1) in the denominator
of your (proven!) expression for /(n), and cancel to obtain a greatly simplified
formula for /(n) in terms of n and /(1). Finally, substitute n = 1996.
92 Hints and outline solutions

3. Let ABC be an acute-angled triangle, and let O be its circumcentre.

The circle through C, O, and B is called S. The lines AB and AC meet
the circle S again at P and Q respectively. Prove that the lines AO and
PQ are perpendicular.

(1) The first thing to do in any geometry problem is to get out your ruler and
compasses (and a sharp pencil), and draw a good diagram. As you construct
the diagram, try to understand what it is that you have to prove. In your
diagram, it is highly likely that the line segments AO and PQ do not even
meet! So perhaps you should extend the line segment AO to meet PQ at the
point X (say). You have to show that A AXP is a right angle. You would
scarcely be asked to ‘prove’ this if it was obvious, so you must expect to have
to do something for yourself.
(2) The simplest thing you could hope for is that the required result
‘ z. AXP = 90°’ may be equivalent to something which is a little more obvious.
In the question there is no mention of the point X; all of the information
given is about the points A, B, C, O, P, and Q. Use the triangle AXP to
reformulate the required result ‘ A AXP = 90°’ into an equivalent statement
involving only A BPQ and A BAO.
(3) Somewhere you must expect to use the fact that the four points B, P, Q,
and C lie on a *i***e, and so form a ****i* quadrilateral. Use this to
reformulate the required result into a statement involving ABCA and A BAO
only.
(4) One reason why (3) is a good move is that you now know that the
required result ‘ A AXP = 90°’ is equivalent to a statement about angles in the
original triangle ABC. Read the question again carefully. What does the fact
that ‘O is the circumcentre of triangle ABC’ tell you about triangle OAB1
What does it tell you about triangle OBC1 What does it tell you about
triangle OCA? (If you did not mark in the line segments BO and CO when
you drew the original diagram, it should now be clear that this was a mistake.
You should always expect to have to mark important lines or points which are
not explicitly mentioned in the question—although you must be selective,
since if you mark too many points and lines the diagram becomes a mess.)
How does this prove what you want to prove about A BAO + A BCA1
[Alternatively, A BAO = AXAB = A ABO (since AO = BO, so A AOB is
isosceles); AAOB = 2 X AACB (since O is the centre of the circle through
A, B, and C). Hence 180° = A XAB + A ABO + A AOB = 2(Z_ BAO +
ABCA).]

4. For any real number x, let [x] denote the greatest integer which is less

than or equal to x. Define q(n) = for n = 1,2,3,... . Determine

Un)
all positive integers n for which q(n) > q(n + 1).
 32nd BMO, 1996 93

(1) One of the amazing things about mathematics is the way one can start
out in a complete fog, yet still manage to fumble one’s way to the required
goal! At first sight it is not at all clear what this question is about, nor how
one is meant to begin. The unfamiliar ‘[x]’ makes the whole thing look much
worse. But don t just sit around feeling glum: start by calculating a few values
q(2), q(3), g(4), g(5),..., just to see how this strange looking function
q(n) behaves. When n = 1, ^1 =_; so [^1] =_; so l/[\/l] =_;
so q( 1) = [1/[a/1]] =_. Now do the same to find q(2), q(3), <7(4), etc.
(2) Before you can answer the question you need an idea about how to begin.
The first thing that you are almost bound to notice is that the nasty looking
denominator [Vai] jumps whenever n moves from a number just below a
perfect **ua*e to the next **iia*e. So suppose that we choose a square m2
and concentrate on values of n between two successive perfect squares m2
and (m + l)2; that is, with n ^ m2 but n < (m + l)2. Then y/n > __ and
vfo <_• Hence [v^] =_. For such values of n, the exact value of
q(n) = [n/m] will depend on where n is in the interval m2 < n < (m + l)2:

• if m2 < n < m2 + m, then q{n) = [n/m] =_;

• if m2 + m < n < m2 + 2m, then q(n) = [n/m] =_;
• if n = m2 + 2m, then q(n) = [n/m] =_.

Thus for all of these values, q(n) increases as n increases.

(3) Hence the only value of n for which we may find q(n) > q(n + 1) is
n =_. Calculate q(m2 + 2m) and q((m + l)2).
(4) This more or less finishes off the question. But it may be worth reflecting
on the way in which a question that seemed so inaccessible a little while ago
now seems so simple.

5. Let a, b, and c be positive real numbers. Prove that:

(a) 4(a3 + b3) ^ (a + b)3;
(b) 9(a3 + b3 + c3)>(a+b + c)\

If you have a clever idea about how to solve both parts at once, you will
probably not need this outline solution. So I shall begin as though you have
not yet solved the problem.
(1) The first inequality (a) almost invites you to multiply out the bracket on
the RHS and collect up terms. Do this: you should be able to write out the
expansion of (a + b)3 without doing any work, using the coefficients 1, 3, 3,
and 1 from Pascal’s triangle. (‘Multiplying out’ is often the wrong thing to do;
but unless you have a better idea, the cubed terms on both sides of (a), and
the fact that all terms land up with the same coefficient ‘3’, suggest that it
might lead here to a simpler version of the problem.)
94 Hints and outline solutions

(2) This shows that the required inequality is equivalent to proving that

a3 - a2b - ab2 + b3 ^ 0 whenever a, b > 0.

The LHS of this inequality is just asking to be factorized (the first and third
terms have a common factor a, while the second and fourth terms have a
common factor b\ both the resulting terms then have a common factor
a2 — b2). If you complete this factorization, the LHS has three factors—one
of which is positive, while the other two are equal; so we have ‘(something
positive) times (a perfect square)’, which must therefore be ^ 0. Hence the
whole expression is always $5 0, as required.
(3) The steps in (1) and (2) above start out with the inequality that is to be
proved, and transform it into an equivalent inequality that you can prove.
While it is possible to write this out correctly (provided that you are very
careful to stress that ‘a and b satisfy 4(a3 + b3) > (a + b)3 if and only if
a3 — a2b - ab2 + b3 > O’), it is better to make a habit of never starting out
from what you have to prove. Instead, you should start from something that is
indisputably true and move step by step to deduce what you wish to prove.
For example:

‘Suppose that a, b > 0. Then a + b > 0, and (a - b)2 ^ 0.

.-. (a + b)(a - b)2 = a3 - a2b - ab2 + b3 > 0

.•. 3(a3 — a2b — ab2 + b3) > 0
4(a3 + b3) ^ a3 + 3a2b + 3ab2 + b3 = (a + b)3.’

[An alternative approach to (a) is to notice that a3 +b3 can itself be

factorized (see the ‘Algebra’ section in ‘A little useful mathematics’). One
(positive) factor cancels with one of the factors of (a + b)3. It follows that the
required inequality is equivalent to 4(a2 - ab + b2) >(a+ b)2\ which simpli¬
fies to a2 - lab + b2 ^ 0, or (a - b)2 > 0.]
(4) Unfortunately—as you have probably discovered if you are reading this
hint in search of inspiration—it is not easy to make these simple-minded
methods work when the number of variables increases from two (a, b) to
three (a,b,c)l But that only means that you need an idea, rather than blind
calculation. One bright idea you might think of is to try to use part (a). You
know that a, b, and c are all positive. Thus you can safely use the result of (a)
for each of the three paris ‘a, b\ ‘b, c\ and ‘c, a' separately:

.-.4(a3 + b3) > (a + b?, 4(b3 + c3) > (b + c)\ 4(c3 + a3) > (c + af.

Now add these three inequalities, simplify the LHS, and multiply out the
RHS. The resulting inequality is certainly not quite what you want, but it may
come in handy later on.
 31st BMO, 1995 95

(5) You want to prove the inequality 9(a3 + b3 + c3) ^ (a + b + c)3. To suc¬
ceed, you must certainly not start by assuming what you want to prove; but it
is a good idea to start with the LHS ‘9(a3 + b3 + c3)’, and to try to find a
succession of true inequalities that will lead you to the RHS. So, start with
9(a3 + b3 + c3) and use the inequality that you proved in (4) to show that

9(a3 + b3 + c3) > 3(a3 + b3 + c3 + a1 2b + b2a + b2c + c2b + c2a + a2c).

(6) Now work in rough to compare the RHS of the inequality proved in (5)
with the RHS (a + b + c)3 of the inequality that you are trying to prove. Show
that

3(a3 + b3 + c3 + a2b + b2a + b2c + c2b + c2a + a2c) > (a + b + c)3

is true, provided that 2a3 + 2b3 + 2c3 > babe holds. Then finish off the
question.

31st British Mathematical Olympiad 1995 ,

1. Find all squares ending in three 4s. Show that no square ends in four 4s.

(1) The first number that ends in three 4s is 444, which is not a square (since
it lies between 212 = 441 and 222 = 484). The very next candidate is _,
and this just happens to be equal to (_)2. Thus you should certainly have
managed to find at least one such square.
(2) The challenge is to find all squares N2 which end in three 4s. The first
thing to realize is that you only need to worry about the last three digits of N
(since if N = 1000/: +x, then N2 = (1000/: +x)2 = 1000(_) +x2,
so the only bit which affects the last three digits of N2 is the final term x2).
You can now use one of two approaches.
(3) For the last digit of x2 to be a 4, the last digit of x must be _ or_.
The last two digits of x2 are determined by the square of the last two digits of
x (Why?). If you square in turn 02, 12, 22, 32, 42, 52, 62, 72, 82, 92 and 08, 18,
28, 38, 48, 58, 68, 78, 88, 98 (no calculators!) you find that only_, _, _,
and _ give two 4s. You can now repeat this exercise to discover which
endings x have squares ending in three 4s. (This approach is rather messy.
See if you can streamline it to make it more mathematical.)
(3') Alternatively, you know from (1) that 382 ends in three 4s. Hence, if
N2 also ends in three 4s, then N2 = 382 (mod 1000). Therefore N2 - 382 =
(N - 38XW + 38) = 0 (mod 1000), or 1000 | (N - 38XN + 38). Now HCF{N -
38, N+ 38) = HCF(N - 38, (A + 38) - (N - 38)) = HCF'(N - 38,76). Hence
96 Hints and outline solutions

HCF(N - 38, N + 38) must divide 76 = 21 2 X 19. Since we know that 2x5 =
1000 \{N + 38XN - 38), it follows that 531 TV — 38 or 53) AT -h 38; moreover,
both N + 38 and N- 38 must be divisible by 4 (Why?). Hence

N2 ends in three 4s =* either N + 38 or N — 38 is divisible by 4 X 125,

=> N = 500A: ± 38 for some k.

Conversely, if N = 500A: ± 38, then N2 ends in three 4s (Why?).

(4) Finally, you should check that, when N has this form, then N never
ends in four 4s.
[A better method here is to observe the following. If A2 ends in four 4s,
then N = 2M is even, and M2 ends in at least two Is (Why only two?). But
then M2 = 11 (mod 100), so M2 would be congruent to 3 (mod 4), which is
impossible!]

2. ABCDEFGH is a cube of side 2.

(a) Find the area of the quadrilateral AMHN,
where M is the midpoint of BC, and N is
the midpoint of EF.
(b) Let P be the midpoint of AB, and Q the
midpoint of HE. Let AM meet CP at X,
and let HN meet FQ at Y. Find the
length of XY.

(1) Start with (a). We can only calculate ‘area’ for two-dimensional figures, so
you should first show that the plane through A, M, and H passes through N
too. (It is enough to show that AMHN is a parallelogram.) A ABM has
a *i*** a***e at _, so we can use Pythagoras to calculate AM =_.
Similarly, we can find MH, HN, and NA. All four sides of AMHN are e*ua*,
so AMHN is a_. (Do not fall into the trap of assuming that AMHN
is a **ua*e!). The diagonal AH has length _ (Why?), and MN has
length _(Why?); hence area( AMHN) —_.
[Alternatively, represent AM and AN as vectors and take the cross
product.]
(2) Now for (b). Take A as the origin, and find the co-ordinates of X (using
2-D geometry in ABCD). Then find the co-ordinates of Y. Then use
Pythagoras (in 3-D) to^calculate the length of XY.
[Or use vectors: AB = 2i, AF = 2j, AD = 2k, and AX = A(2i + k) = i +
M2k+ i), so ix = __, and AX =_i +_k. Similarly, AY =_j +
(—k +—i); .•. XY = —k + — j +_i, so the required length
XY =_.]
 31st BMO, 1995 97

3. Maximize (a) x2y-y2x and (b) x2y +y2z+z2x-x2z-y2x-z2y,

when 0 <x, y, z < 1.

(1) Start with (a). It is not hard to discover what seems to be the maximum
value; namely, -, when x =_ and y =_. The challenge is to prove
that no other x and y can do better!
(2) Here is one way. Since 0<x<l, and xy ^ 0, x2y =x-xy Thus
x2y -y2x=x-xy - y2x ^ xy - y2x = x(y -y2) ^y - y2 (since x<l and
y-y2 =>'(1 -y) > 0)- Finally, observe that y-y2 = \ - (± -y)2 < ± (since
the squared term is always >0). Moreover, x2y-y2x=± when x =_,
y=—•
[Alternatively, for each fixedx (0 <x «: 1), differentiating with respect to y
shows that x2y-y2x has maximum value _ at y =_. Hence the
maximum over all x is at x =_, y =_.]
(3) Now for part (b). The expression can take positive values (for example,
when x =_, y =_, z =_); so the maximum is certainly positive.
When x=y (or y=z, or z=x), the expression equals _; hence the
remainder theorem gives

(*) x2y +y2z + z2x —x2z-y2x-z2y = (jc -y)(y - z)(x - z)

= (x -y)(z -y)(z -x) = (y -z)(z -x)(y -x).

For a maximum, either:

(i) all brackets in (x — yXy — z\x — z) are positive, so x >y > z; or

(ii) exactly two brackets in (x-yXy - z\x - z) are negative, so z >x >y or
y >z^x and all of the brackets in one of the other factorizations in (*)
are positive!

Relabelling the variables if necessary, you may assume that x^y^z. Then
(x — yXy —zXx — z) < (1 -yXy ~z)( 1 - z) = a ■ (b - a) ■ b = b2a - a2b, where
a = l-ye[0,l], and 6 = l-ze[0,l], Now apply part (a) (with x = b, y = a)
to obtain a maximum when a =_, b -_.
[There are many other correct ways. But beware: most arguments are
flawed!]

4. Triangle ABC has a right angle at C. The internal bisectors of angles

BAC and ABC meet BC and CA at P and Q, respectively. The points
M and N are the feet of the perpendiculars from P and Q to AB. Find
angle MCN.

(1) Draw an accurate, decent-sized diagram.

(2) You should always expect to have to add important lines and points that
98 Hints and outline solutions

are not mentioned explicitly in the question. The segments PM and QN are
both perpendicular to AB; this should suggest that it might be a good idea to
draw the perpendicular CL from C to AB. (Then AACB, AANQ, AALC,
ACLB, and APMB are *i*i*a* (Why?). Hence AN: AL =AQ: AC.)
(3) BQ bisects the angle ABC. What does this tell you about the triangles
BQC and BQN1 (Explain.) What can you conclude about triangle QNCl Use
this (and the fact that QN and CL are parallel) to prove that CN bisects
ALCQ.
(4) Prove that CM bisects A LCP.
(5) Deduce that AMCN = AMCL + ALC7V = \ALCB + \ALCA
= \(ALCB + ALCA) =_.

5. The seven dwarfs walk to work each day in single file, with heights
alternating up-down-up-down-... or down-up-down-up-.... If they
all have different heights, for how long can they continue like this with
a new order every day? For how long could they continue if Snow White
always came along too?

(1) We need a recurrence for S(n)—the number of arrangements with n

dwarfs. The answer to the first and second parts will then be S(7) and S(8)
respectively. Let the dwarfs be numbered 1,2,3,...,n, where dwarf i may be
assumed to have height i. Let U(n) be the number of possible up-down-...
arrangements, and D(n) the number of down-up-... arrangements for n
dwarfs. Suppose that you are given some up-down-up-... arrangement. If,
for each i (1 < i < n), you replace dwarf i by dwarf (n + 1) -1, the
up-down-up-... arrangement becomes a down-up-down-... arrangement
—and conversely. Hence U(n) = £>(«). (This simple proof has to be given; the
result isn’t obvious.) When n > 2, U(n) = D(n) = S(n)/2; when n = 0 or 1
one has to be slightly careful, since then U(n) = D(n) = S(n) = 1.
(2) Count the number 5,(«) of possible (up-down-... or down-up-...)
arrangements in which the tallest dwarf V is in the (i + l)th position. To
obtain such an arrangement, you must first choose i dwarfs from {1,2,...,
n - 1} to go in the first i positions; then order these i dwarfs (up-down-...
or down-up-...); finally, you must order the remaining n-(i + 1) dwarfs
(up-down-...). Since U(i) = D(i), the formula is the same irrespective of
whether the first i dwarfs go ‘up-down-... ’ or ‘down-up-... ’; namely,

S,(n) = (nT X) xUiOxUCn-i-l).

(3) When i > 1, U(i) = S(i)/2; but when / = 0 or 1, t/(i) = S(i) = 1. Hence
we may use

S(n)= £ 5,(n),
i= 0
 30th BMO, 1994 99

with starting values U(0) = 5(0) = 1 and U(l) = 5(1) = 1, to calculate 5(2),
5(3), etc.:

5(2)= (Jjc/(0)C/(1) + | jJc/(l)t/(0) = l-M + 1-1-1 =_;

50) = ( q)t/(0)£/(2) + ( ^ )t/(l)C/(l) + (^ )t/(2)f/(0)

= 1-1-1 +2-1-1 + 1-1-1 =_;

S(4) = ( q ]l/(0)£/(3) + (3 Jt/(l)l/(2) + (3)U(2)U(l)

+ ^Jc/(3)C/(0) =_;

5(5) = ;
5(6) = ;
5(7) = ;
5(8) =

30th British Mathematical Olympiad, 1994

1. Starting with any three-digit number n (such as n = 625), we obtain a

new number f(n) which is equal to the sum of the three digits of n,
their three products in pairs, and the product of all three digits.
(a) Find the value of n/f(n) when n — 625. (The answer is an integer!)
(b) Find all three-digit numbers n such that the ratio «//(«)= 1.

(1) Part (a) is only included to help you to check that you have read the
question. (If you calculate n/f(n) and obtain an integer, then you have
probably got the right idea. Otherwise, go back and read the question again
more carefully.)
(2) Now for part (b). In the BMO ‘find’ means ‘find them all, and prove that
you have found them all’. By all means work in rough to find what you think
is the complete list. But do not then think that this solves the problem: it is
only a beginning. The first mathematical step is straightforward algebra: if
n = ‘abc’ = 100a + 10b + c, then f(n) = abc + ab + be + ca + a + b + c.
Hence n=f(n) precisely when

99a + 9b — abc 3-ab + be + ca.

That is,

(*) n=f(n) => (9 — c)b = a(bc + b + c — 99).

100 Hints and outline solutions

Now b, c < 9, so be + b + c - 99 <_. Since a > 0; it follows that the RHS

of (*) is <_. But b > 0, and c < 9 (so 9 — c 5* 0); hence the LHS of (*) is
_. Thus f(n) — n is true if and only if both sides of (*) are equal to
_. Since a # 0 (as n is a three-dig\i number), we must have be + b + c = 99;
hence b = c = 9. Thus n = 199, 2_, _, _, _, -> ->
__, or_.
[Alternatively, use (*) to solve for c. Use this expression for c to show
c > 9, with equality only if b — 9. Hence conclude that b = 9 = c.\

2. In triangle ABC the point X lies on BC.

(a) Suppose that /l BAC = 90°, that X is the midpoint of BC, and that
ABAX is one third of ABAC. What can you say (and prove!) about
triangle ACX1
(b) Suppose that ABAC = 60°, that X lies one third of the way from B
to C, and that AX bisects ABAC. What can you say (and prove!)
about triangle A CXI

(1) In your attempt at part (a) you may not have taken the injunction ‘(and
prove!)’ seriously enough. ‘Prove’ in mathematics means ‘lay out logically,
and explain each step clearly in terms of basic facts or theorems’. Do not be
satisfied with baby language: for example, rather than ‘X is halfway along BC
so X must be halfway up (from AB to C) and halfway across (from B to
AC)’, you should use *i*i*a* **ia***e*. (If you join the midpoint X of BC
to the midpoint M of AC, you must explain logically why XM is perpendicu¬
lar to AC. And if you choose M on AC so that XM is perpendicular to AC,
you must explain why M has to be the midpoint of AC.)
(2) One possible approach to part (a) is to use the sine rule to find AB from
triangle AXB in terms of sin(AAXB), then find AC from triangle AXC in
terms of A AXC. [Alternatively, use the area formula ‘(\)ab sin C’ in each
triangle.] Hence conclude that AB: AC = ^3, so tan(Z. ACB) =_, whence
AACX=_. Hence AACX is isosceles with base angles equal to 60°, so it
must in fact be e*ui*a*e*a*. (Note: It is not enough to derive a trig equation
relating certain angles in the diagram, and then to simply ‘assert’ that your
favourite solution is the only possible solution. If you use such an approach,
you must prove that there is only one possible solution.)
[Alternatively, you could find AX from both triangles and conclude that
tan (AABX)= 1/a/3. Or you could observe that the circumcircle of A ABC
has diameter BC (since ABAC is a right angle), so X is the centre. Hence
AX = BX = CX, and so on.]
(3) Part (b) is related to part (a). But this does not mean that you can assume
what needs to be proved (namely that AB = 90°) and work backwards! You
could use the area formula ‘(\) AB ■ AX ■ sin_\(\)AC-AX-sin_= 1:2’
to deduce that AC: AB =_. [The sine rule in triangles AXB and AXC
 30th BMO, 1994 101

gives the same result.] The cosine rule on A ABC (for A A) then gives
BC: AB =-, and the converse of Pythagoras (or the cosine rule on
A ABC (for AB)) gives AABC =_. It is then not hard to show that
AX = XC, or that AAXC is _. (Your final solution must be laid out
carefully, giving reasons for each step.)

3. The sequence of integers u0,uvu2,u3,... satisfies u0 = 1 and

= for each n > 1,

where k is some fixed positive integer. If w2000 = 2000, determine all

possible values of k.

(1) Think carefully before writing down lots of equations (or you may just go
round in circles). Make sure you understand, and avoid, the following
common mistakes.

(a) Once one term repeats, you cannot conclude that the whole sequence
must repeat. Instead you have to show not only that u6 = u0, but also that
u7 = ux (since it is u6 and u7 together that determine u8, and so on).

(b) There are several pieces of information in the question for you to juggle.
You must not only take them all into account, but you must show the
reader how you are doing this. It is not enough simply to declare ‘The only
possible values of k are _’.

(2) Let ux = x. Then u2 =_, u3 =_, u4 =_, u5 =_, u6 =_,

and un =_.
(3) Since each term is determined by the two preceding terms, and u6 = 1 = u0
and u1 = ul=x, the sequence must recur. Hence u0 = u6 = ul2 = ..., so
w2000 = ul-
(4) But u2 = kx, where k and x are integers, and k is positive. Hence
w2ooo = 2000 implies that k and x must both divide 2000. Now 2000 = 24 X 53,
so k and x are both of the form 2a-5b. Moreover, u5=k/x must be an
integer, so 2000 = 24-53 = (k/x)-x2. Therefore x =_, _, _, _,
_, or _, and k =_, _, _, _, _, or _. Each of these
pairs x, k gives rise to a sequence of integers satisfying the conditions in the
question.

4. The points Q and R lie on the circle y, and P is a point such that PQ
and PR are tangents to y. The point A lies on the extension of PQ,
and y' is the circumcircle of triangle PAR. The circle y' cuts y again at
B, and AR cuts y at the point C. Prove that A PAR = A ABC.
102 Hints and outline solutions

(1) This is in fact the easiest question on the paper; but that does not
automatically make it popular! Very little geometry is taught nowadays, so if
you are interested in mathematics, here is a truly wbnderful branch of
mathematics for you to explore on your own. Everything you need here may
be found in ‘A little useful mathematics’. Start by drawing an accurate
diagram. (A good diagram always helps you to think more clearly.)
(2) The chord AP in the circle y' subtends angles at B and at R. Hence
A ABP — L ARP.
(3) The line PR is tangent to the circle y at R. The angle between the
tangent PR and the chord RC is equal to the angle RBC subtended in the
opposite segment. Hence A PRA = A RBC.
(4) Z. ABC = A ABP - LCBP = L RBC - A PBC = A RBP
= A_ (angles subtended by the chord RP on y' are equal).

5. An increasing sequence of integers is said to be alternating if it starts

with an odd term, the second term is even, the third term is odd, the
fourth is even, and so on. The empty sequence (with no terms at all!) is
considered to be alternating. Let A(n) denote the number of alternat¬
ing sequences which only involve integers from the set (1,2,..., n). Show
that A( 1) = 2 and A(2) = 3. Find the value of A(20), and prove that
your value is correct.

(1) It is easy to think that you have ‘solved’ a problem like this one, and yet
score very few marks. One reason here is that it is tempting just to guess the
Fibonacci connection, to assume that one’s guess is correct, and then calcu¬
late as though no further proof is needed. The whole essence of mathematics
is that guessing is useful only in that it focuses your attention on what has to
be proved. You then have to find some way of proving that your guess is
correct. That is the challenge here. The solution is easy once you find it. But
it is not obvious.
(2) When n = 1, the only alternating sequences are the empty sequence and
the sequence T’; so A( 1) = 2. When n = 2, the only alternating sequences are
the empty sequence, the sequence T, and the sequence ‘12’; so A(2) = 3.
(3) Alternating sequences from {1,2,...,n) are of three types. First, there is
the empty sequence; next there are those sequences that begin with a T
(type A); all other alternating sequences begin with an odd number ^3
(type B).
(4) Each type A sequence arises uniquely by first adding 1 to each term of
some alternating sequence from {1,21}, and then sticking a ‘1’ at the
front. Each type B sequence arises uniquely by adding 2 to each term of some
 29th BMO, 1993 103

alternating sequence from {1,2,...,n — 2}. Since this latter operation also
includes the empty sequence, this proves that

A(n) =A(n — 1) +A(n — 2).

(5) Now use A(l) = 2, A(2) = 3, and the recurrence relation in (4) to calcu¬
late ,4(20).

29th British Mathematical Olympiad 1993 ,

1. Find, showing your method, a six-digit integer n with the following

properties:
(a) the number formed by the last three digits of n is exactly one
greater man the number formed by the first three digits of n (sc n
might look like 123124);
(b) n is a perfect square.

(1) Look at the condition (a): ‘The number formed by the last three digits is
one more than the number formed by the first three digits.’ What exactly is
this saying? Let the number formed by the first three digits be x; then the
number formed by the last three digits has to be _. So if the six-digit
number n =‘abcdef\ then x = ‘abc\ ‘def’ = x + 1, and n=x+ 1 +_.
You are also told that n is a square, say n=y1 2\ your equation for n can
therefore be written as y2 —x + 1 + 103 4x.
(2) You should now feel an urge to take the T to the LHS and to factorize
both sides. (When working with real numbers, this would usually be a bad
move. But when working with integers, factorizations can be highly instruc¬
tive!) Factorize the resulting LHS y2 — 1 = (_)•(_), and the RHS
x+103x =_•(_). Finally, factorize 1001=_•_•_. Hence
(y — l)(y + 1) =x-l • 11 • 13.
(3) Now try all possibilities in turn (such as y — 1 = la, y + 1 = 143b, or vice
versa). Provided that you are systematic and show your method clearly, you
will have anwered the question.
(4) However, the mathematician in you should want to find a deductive
approach which will generate all possible solutions—such as the following,
which is based on the useful fact that 1001 = 7-11-13. If (y-lXy + l) =
7 • 11 • 13 x, then two of the primes 7, 11, and 13 must divide one of factors on
104 Hints and outline solutions

the LHS, while the other prime divides the other factor. (Since y < 10 , at
most two of the three primes divide each factor.) There are 3x2 cases:

Case 1 y + 1 = 143k, y — 1 — Ik' Case 2 y +1 = 91k, y- 1 = Ilk'

:.k = 3, k' = 61, x= 183 .’. k = ,k' = , x=
(N
II ••• y=- ,y1 2 3 =
II

Case V y — 1 = 143k, y + 1 = Ik' Case 2' y-1 =■ 91k, y + 1 = Ilk'

.•. k ,k' , *=
*
= =

ll
ll
II

to

y= ,y2 (not six-digit)

V:

•••
II
II

Case 3 y + 1 = Ilk, y — 1 = 13k'

k =_, k' =_, * =-

Case 3' y - 1 = Ilk, y + 1 = 13k'

k =_, k' =_, x =-
y =_, y2 =_ (not six-digit).

2. A square piece of toast ABCD of side length 1 and centre O is cut in

half to form two equal pieces, ABC and CDA. If the triangle ABC has
to be cut into two parts of equal area, one would usually cut along the
line of symmetry BO. However, there are other ways of doing this. Find
(with proof!) the length and location of the shortest straight cut which
divides the triangle ABC into two parts of equal area.

(1) Beware of assuming that a line which bisects the area must go through
some special point—such as the centroid.
(2) There are two essentially different ways of cutting the toast. You may cut
off either one of the 45° angles (say A), or the right angle at B. Both ways
have to be considered—although any method which works in one case ought
to work just as well in the other case; thus the second case should be
relatively easy.
(3) It is not unreasonable to feel that, because
the angle at A is smaller than the angle at B,
the shortest cut which cuts off half the triangle
by cutting off the angle A ought to be shorter
than the shortest cut which cuts across the angle
B, so we consider this first. Let XY cut the
triangle in half; then area( ABC) =_, so
area(AXY) =_. Let AX = x and AY = y.
Then |area (ABC) = area (AXY) = \ -xy • 1/yJl. C
Thus xy =_. Hence as x and y vary over all possible cuts, the product
xy remains constant.
 29th BMO, 1993 105

(4) How can this help you find c, the length of the cut? It should be clear
that you have little choice but to use the *o*i*e rule to find c in terms of x
and y, giving c2 =_. You can then substitute for y (using the
fixed value of xy) to find an expression for c2 in terms of x.
(5) To find the shortest cut it is natural (if you know calculus) to differentiate
the formula for c2 with respect to c and solve dc/dx = 0. You must then
explain clearly why this gives a minimum.
(6) If you do not know calculus, or do not wish to use calculus, you should
use the constant value of xy to express c2 in terms of (x +y), and conclude
that ‘c is a minimum when x + y is a minimum’. To find the minimum value
of c for this kind of cut you can then use the AM-GM inequality:

x +y
11
= vT
1 with equality if and only if x = y = -—
1 \
2
v/2 / l /

[An even quicker way is to write c2 = (x -y)2 + constant, and observe that
this is a minimum when x = y.]
(7) You must now repeat all this for a cut which cuts off the right-angled
corner at B, to see if the shortest cut in this case is longer or shorter than the
shortest cut in the previous case. The same method works and is left as an
exercise.

3. For each positive integer c, the sequence un of integers is defined by

u1 = 1, u2 = c, un = (2n + 1)«„_i — (n2 — l)w„_2 (n > 3).

For which values of c does this sequence have the property that

ut divides Uj whenever i <;'?

(1) Faced with a problem like this, there is really only one way of getting
started; namely, work out the next few terms of the sequence in terms of c.
But be careful! A little algebraic slip at this stage can lead to a lot of
misdirected work. So work out the next term u3 in terms of c: u3 =_.
(2) If you now think about the last sentence of the question, you will see that
this tells you something about the two integers u2 and u3, and that this in
turn tells you that the values of c that you want must satisfy the condition
that c divides _.
(3) You should now be able to find a shortlist which includes all possible
values of c. However, the last sentence of the question is a much stronger
condition than just saying ‘u2 divides u3, so some of these ‘possible’ values
may not work. If c divides a, and c divides b, then c divides the difference
a - b. You know that c divides 7c, and you have just shown that c must divide
u3 =_; hence c divides their difference, namely_.
 Hints and outline solutions
106

(4) This tells you that there are only four possibilities for c (since c > 0),

c =_, or-, or-, or-

You can now test each of these possible values in turn (smallest first).
(5) If c =_, the sequence goes _, —, —, —, —> — = — >so us
doesn’t divide u6.
(6) If c =_, the sequence goes , ,—,—,—which looks
promising! How can you prove that the sequence has the required property
in this case? To have any hope of showing that ui divides Uj whenever i ^ j,
you really need a *o**u*a for the nth term of the sequence so try to guess
a formula, and then prove it (by i**u**io* on n).
Conjecture. un =_.
Proof. The formula certainly works when n = 1 (or you would not have
chosen it). Thus it will be enough to suppose the formula is valid for the first
n- 1 terms, and show that it must then be valid for the nth term as well. The
definition of un says that

un = (2n + l)n„_j - (n + l)(n - l)u„_2-

Since you are assuming that your formula is correct for un_1 and for un_2,
you can substitute for these terms on the RHS and simplify. Do this and
complete the proof.
(7) If c =_, the sequence goes _,_,_,_,_,..., which again
looks promising! I leave you to guess and prove a general formula in this
case.
(8) Finally, if c =_, the sequence goes_,_,_,_,..., so u3 does
not divide «4.

4. Two circles touch internally at M. A straight line touches the inner

circle at P and cuts the outer circle at Q and R. Prove that
Z. QMP = Z. RMP.

If you have learned a little geometry you should enjoy this problem. The
problem is clearly about angles, circles, and tangents. So you expect to have to
use the three basic facts about such things; namely,

• the angles in the same segment are equal,

• the opposite angles of a cyclic quadrilateral add to 180°,
• the angle between a chord and a tangent is equal to the angle in

(In one sense, the first of these three facts should be enough, since the other
two follow from it. However, you should learn to use them all. In this problem
the shortest solution avoids using the first fact explicitly.)
 29th BMO, 1993 107

(1) Draw your own diagram! So far only four points (namely M, P, Q, and R)
and four line segments (namely MQ, MP, MR, and QR) have names. Two
other as-yet-unlabelled points are screaming out to be given names. Which
points are they?
(2) Let QM and MR meet the inner circle at X and Y respectively. Three
line segments (namely _, _, and _) then beg to be drawn in to
create lots and lots of ways you can use the three basic facts listed above.
(3) It is tempting now to mark in as many pairs of equal angles as you can.
This is a good move; but it can mess up your diagram. So do this on a rough
diagram, while keeping your best diagram clear until you see which pairs of
angles will help you to solve the problem.
(4) From your rough diagram it should soon become clear that, although
QMP is equal to several other angles, and Z PMR is equal to several other
angles, there is no direct way of proving that these two angles are equal by
simply using our first basic fact. You are clearly going to need the other two
facts. The second fact shows

ZQXP = 180° — ZMXP = Z- (cyclic quadrilateral).

This means that the two triangles QXP and PYM have two angles equal.
What about the angles Z.QPX and LPMY9 Z.QPX is the angle between the
tangent QP and the chord PX, and so is equal to one of the angles you are
interested in: /LQPX= Z_(angles between tangent and chord).
(5) Thus if only you could show that Z QPX = Z PMY, then you would be
finished. One slightly backhanded way of doing this is to show that Z PQX =
zLMPY. To finish things off you have to realize that we have still not used
one of the most important hypotheses in the question! Which one? If two
circles touch internally at M, then they must have a common *a**e** at M.
So you can use our third basic fact to show that the angle between the
tangent at M and the chord MR is equal to the very interesting angle
subtended by the chord MR at Q on the outer circle—namely Z_—and
the very interesting angle subtended by the chord MY at the point P on the
inner circle—namely Z_.
[Alternatively, you may have noticed two lines in your original diagram
which look remarkably parallel! It is always worth drawing a careful diagram.
And though one should never assume that lines that look parallel must be
parallel, a diagram can suggest useful things to prove. Prove that these two
lines are in fact parallel, and use this to solve the problem.]

5. Let x, y, and z be positive real numbers satisfying

Determine the range of values for (a) xyz and (b) x+y + z.
108 Hints and outline solutions

(1) The upper bound in (a) is pleasantly easy if you know the AM GM
inequality for three variables, which says that

a +b+c 31~—-—-
‘whenever a, b,c > 0, we have - > V \a-o-c) ,

with equality if and only if a = b = c\

To obtain an upper bound for xyz, all you need to do is choose a, b, and c in
a way which will allow you to use one of the two inequalities involving
xy + yz + zx which are given in the question. Once you notice this, there is
only one choice, namely a =_, b =-, and c =-. Then

ci + b + c 3,-—
1>---^yj(a-b-c) => xyz <-,

with equality when x = y = z =-, xyz =-

Finally, note that if x=y = z =_, then xy +yz + zx =_, so the inequal¬
ities in the question are satisfied. (To show that you know that this upper
bound is exact, you must show explicitly that xyz can actually take this
boundary value by specifying explicit values of x, y, and z which satisfy the
inequalities given in the question!)
(2) A lower bound for xyz may seem much more difficult. Since the upper
bound used the right-hand inequality xy + yz + zx < 3, you might expect the
lower bound to use the other given inequality, xy +yz + zx^\. It doesn’t. In
fact, xyz can take any positive value < 1. The proof is not hard, but requires a
change of focus. Since x, y, and z are all positive, you can be sure that
xyz >_. To show this is the best possible lower bound, you have to show
that it is possible to choose values for x, y, and z which make their product
xyz as small as ever you please. To keep things simple, try x =y = 1, say. Then
z only has to satisfy | <x2 + 2xz = 1 + 2z < 3; so you can choose z as small
as you like and make xyz =z as small as you like. Thus xyz > 0 is best
possible.
(3) Now for part (b). Once you have understood how to show that xyz can
take arbitrarily small positive values, you should be able to show that
‘x+y +z can be arbitrarily large’. (For example, take x = l/n =y for n 1,
and choose a suitable value of z such that | <x2 + 2xz = (1 /n2) + 2z/n < 3:
for example, z = n.)
(4) The proof of a lower bound for x+y+z is a little more awkward. It
helps to notice that x=y=z=| gives the value x+y+z=_. This
seems hard to beat! However, you don’t have to spot this before getting
started (although you will need it later to show that the lower bound is exact.)
Suppose that you want to prove that ‘x+y+z^_’. You must certainly
use the inequalities given in the question. These involve the expression
xy + yz + zx, so you need to link the expression xy +yz+zx with x+y+z in
 28th BMO, 1992 109

some way. One way in which you should certainly be tempted to try is to
look at

(x +y +zf = x1 2 +y2 +z2 + 2(xy + yz+ zx) ^x2 +y2+z2 + 2-(_).

: (5) Thus you would like to prove ‘x2 +y2 + z2 > j. You may either:

(a) use the AM-GM inequality a2 + b2 ^ 2ab three times to finish the
solution; or

(b) rewrite x2 +y2 +z2 - xy — yz — zx as the sum of three squares.

[Alternatively, look up the Cauchy-Schwarz inequality and see if you can

use it to produce a shorter solution.]

28th British Mathematical Olympiad 1992 ,

1. (a) Observe that the square of 20 has the same number of non-zero
digits as the original number. Does there exist a two-digit number
other than 10, 20, and 30 the square of which has the same number
of non-zero digits as the original number? If you think there is one,
then give an example. If you claim that there is none, then you must
prove your claim.
(b) Does there exist a three-digit number other than 100, 200, and 300
the square of which has the same number of non-zero digits as the
original number?

(1) Start with part (a). Let the number be N = ‘ab’; that is, N= 10a +b. If
b — 0, then the only solutions are _,_, and ___ (since 40, 50, 60, 70,
80, and 90 have squares with two non-zero digits).
(2) Thus you may assume that b + 0. Hence N2 has a non-zero units digit,
and can only have one other non-zero digit. Since the first digit (whether this
be the hundreds or the thousands digit) has to be non-zero, the tens digit must
be zero. Since

N2 = (10fl + bf = 10 V + 2 ab -10 + b2,

it therefore follows that ‘2ab + any carry from b2’ must be a multiple of
_; in particular, ‘any carry from b2’ must be even. There are very few
possibilities:

(i) b < 3, so 10 I 2ab implies a = 5; hence N =_, _, or_: none of

these work.

(ii) b = 5, which contradicts the fact that 2ab + 2 = 10a + 2 should be a

multiple of 10.
 Hints and outline solutions
no

(N = .), or a = OV = _
(iii) b~l, and 10 114a + 4, so a =
neither of which work.

(iv) b~ 8, and 10 1 16a + 6, so a = (N = .), or a = (N = X

neither of which work.
(N = .), or a = (N = ),
(v) b = 9, and 10 118a + 8, so a =
neither of which work.
Hence N = 10, 20, and 30 are the only such two-digit numbers.
(3) Now for part (b). Let the number N = ‘abc’; that is, N = 10 a + 10b + c.
If c = o, then N = lOrc, where n is a two-digit number the square of which
has the same number of non-zero digits as n, so the only solutions are
TV =_5_5 or_(by part (a)). Thus you may assume that c # 0. If
b = 0, then N2 = 10V + 102-2ac + c2. The first and last digits then have to
be non-zero, so all other digits must be zero. Hence c ^ 3 (as c ^ 4 gives rise
to a carry from c2, and hence a non-zero tens digit). Moreover, a < 3 (since if
a ^ 4, then a2 > 10, so 10 V would give rise to non-zero digits in both the
ten thousands and hundred thousands). But then the hundreds digit will be
non-zero. Hence there are no solutions with b = 0. Thus we may assume that
b =£ 0, so

N2 = 10V + 103-2ab + 102(2 ac + b2) + 10-2bc + c2

has just three non-zero digits—the first digit, the last digit, and one other.
(4) To find an example, it is tempting to try c = 1 and to choose b =- to
kill off the tens digit. Then

N2 = 104(a2 + a) + 103-2 + 102(_) + 1,

and there are then just two values of a which make the hundreds digit zero;
namely, a =_or a =_. You must then check whether N2 has just
three non-zero digits.
(5) When you think about each of these examples, you should realize that
each is closely related to another example. Finding all solutions, with a proof
that you have not missed any, requires a little more care; but it is worth
doing. (There are fewer than ten examples altogether.)

2. Let ABCDE be a pentagon inscribed in a circle. Suppose that AC, BD,

CE, DA, and EB are parallel to DE, EA, AB, BC, and CD
respectively. Does it follow that the pentagon has to be regular? Justify
your claim.

(1) You should certainly begin by drawing a pentagon ABCDE inscribed in a

circle, with the pentagon looking non-regular (to avoid misleading yourself).
 28th BMO, 1992 111

(2) To show that the pentagon has to be regular you must show that all five
angles are equal and that all five sides are equal. Look first at angles:

• 2. EAD = Z EBD = Z ECD (angles in the same segment)

• Z EAD = Z BDA (since AE is parallel to BD)
= EDBC (since AD is parallel to BC)
® ABCA= ABAC (since AD is parallel to BC)
= AADE (since AC is parallel to ED)
= A ABE (angles in the same segment).
(3) Repeating the reasoning in the final bullet point of (2) once more will
show that Z ABE = Z EAD. It follows that Z ABC = 3 X Z ABE. Similarly,
Z BCD = 3 X Z ECD, and so on, so all five angles of the pentagon ABCDE
are equal and each angle is trisected by the diagonals at that corner.
(4) Hence A.BAC= Z-BCA, so AABC is isosceles, whence AB = BC. Simi¬
larly, BC = CD, and so on, so the five sides of the pentagon ABCDE are all
equal. Hence the pentagon is regular.
[Alternatively, there is a lovely short proof: If Z ABC = 6, then Z BCE =
180° — 6 (since CE is parallel to AB). But ABCE is a cyclic quadrilateral, so
Z BAE = 6. Continuing in this way, we see that all five angles are equal. It
remains to prove that the sides are equal.]
You might like to consider the following variations: (a) What if only four of
the diagonals are parallel to the opposite sides? Does it still follow that the
pentagon has to be regular? (b) What if you are told not that the pentagon is
inscribed in a circle, but that all the sides have the same length? Does it still
have to be regular? What if only four of the diagonals are parallel to their
opposite sides?

3. Find four distinct positive integers the product of which is divisible by

the sum of every pair of them. Can you find a set of five or more
numbers with the same property?

(1) It is natural to begin by trying to construct such a set of four numbers

a <b < c < d in the simplest imaginable way: ‘If a = 1, and b = 2, then
a + h = 3, so you could choose c = 3. Then a + c = 4 and b + c = 5, and
abc = 6, so d = 10 would make abed = 60, which is divisible by the sum of
each pair except a + d and c + d.’
(2) Various attempts to improve on this should lead you to suspect that a = 1
will not work. It is then natural to try a = . ‘If a = 2 and b = 3, then
a+b= , so it is natural to try c =_. Then a+c = ...’. If this is
successful, fine; otherwise, one might try a = 2, b =_. Provided that you
are flexible, imaginative (and persistent) this naive approach should sooner or
later produce a set of four numbers which work.
112 Hints and outline solutions

(3) When looking for a set of five numbers, it is natural to try to build on the
set of four numbers you have already found. (This strategy is not always the
right one in mathematics, but it is always worth trying.) In this case it should
work.
(4) At the same time, a little reflection on the set of four numbers which you
first stumbled upon should suggest that they are part of a natural sequence
ax<a2<a3< .... If you produce such a sequence and prove that it has the
property that

‘for every m ^ 4, axa2a3 ...am is divisible by each sum a, + (1 <i < m)\

you will have answered the question as fully as required. (Let a, = 4i — 2.

Suppose that i < j. Then a, + = 4(i + j - 1). If i + j is odd, then we
may write 4(i +j — 1) = 2xk with k odd; now k <(/+; — l)/2 <j, so
k 11 -3 -5-.. .(2; — 1), whence 2xk\axa1...aj (since x<;). If i+j is even,
then i+j — 1 = 2m — 1 is odd with m < j, so a, + = 2 am I axa2 ... a;.)
(5) There remains the (unstated) question as to whether the sequence {a,} in
(4) is the only sequence of this kind.

4. Determine the smallest value of x2 + 5y2 + 8z2, where x, y, and z are

real numbers subject to the condition yz + zx + xy = — 1. Does x2 +
5y2 + 8z2 have a greatest value subject to the same condition? Justify
your claim.

Problems of this kind can seem impossibly difficult at first. However, when
you eventually see how to do them you are likely to kick yourself, for at this
level many inequalities are simply disguised versions of the basic fact that
squares cannot be negative.
(1) The given expression consists of squares with positive coefficients and so
can never be negative. But it is not enough just to say that it is always ^ 0, for
the expression can never actually equal 0(x=y=z = 0 does not satisfy the
constraint ‘xy+yz + zx= — 1’).
(2) Is there some way in which you can write the expression x2 + 5y2 + 8z2
in terms of perfect squares which will bring in terms involving xy and yz, and
zx? Of course there is! You expect terms such as ^ to appear whenever the
bracket you are squaring contains an x term and a y term. So the vague
question above can now be sharpened up in the following way. Can you
rewrite the given expression x2 + 5_y2 + 8z2 as a sum of squares of two or
more separate brackets in such a way that the brackets between them
contribute 1-x2, 5-y2, and 8-z2, plus mixed terms which can be simplified
using xy +yz + zx= —1? (Once you have done this you should obtain an
obvious lower bound just by observing that the squares have to be >0.
However, you should not stop there. How do you know that the lower bound
 28th BMO, 1992 113

that you have found is the best possible one—corresponding to the ‘smallest
possible value’ of the given expression? The answer is that you don’t, unless
you manage to find specific values of x, y, and z (a) which make each of the
squared brackets equal to zero, and (b) which also satisfy the constraint
xy +yz + zx — -1.)
(3) The first thing to notice is that (ax + ... )2 contributes a2x2. Since we may
assume that a ^ 0, to land up with 1 -x2 you have to have a = 1 with 1 -x in
just one bracket. And to land up with 5-_y~ you have to have (+l)-y in one
bracket and (+2)-y in another. And to land up with 8-z2 you have to have
(±2)-z in one bracket and (+2)-z in another. This suggests just two brackets
with relatively few possibilities:

(4) You have to choose the signs (and which y term goes in which bracket) to
try to ensure that the mixed terms involving xy, yz, and zx all have the same
coefficient. This coefficient turns out to be -4, so the constraint shows that

(*) x2 + 6y2 + isz2 = (x + 2y + 2z)2 + (y - 2zf - 4(xy+yz+2x)

> 0 + 0 + 4
(5) If you want to choose x, y, and z actually to achieve the value ‘4’, you
must first make the two squares on the RHS of (*) equal to 0 by insisting that
y = 2z and that x = —2y — 2z= —3y. You must then fix up these values of x,
y, and z to satisfy xy +yz + zx= -1. For example, if you try z = 1, y =_,
and x =_ to satisfy the condition, then xy+yz + zx =_; hence to
satisfy the correct constraint while keeping the brackets equal to 0 it is
enough to divide each of the trial values of x, y, and z by _. Hence the
expression x2 + 5y2 + 8z2, subject to the constraint xy +yz + zx = -1, really
does attain its smallest value 4.
(6) How about a greatest value? You may be inclined to suspect that the
answer in this case may be ‘No’, and to try to show that the expression can
take arbitrarily large values. If you decide to try this, it is important to
simplify as much as possible. Could you perhaps get rid of one of the three
variables? For example, if z = 0, then the constraint becomes x = —(1 /y). If
you then let y become larger and larger, x tends to zero; thus the given
expression x2 + 5_y2 + 8z2 just goes on getting larger and larger without
bound. Hence it has no greatest value.

5. Let / be a function mapping positive integers into positive integers.

Suppose that

f(n + \)>f(n) and /(/(«)) = 3n for all positive integers n.

Determine /(1992).
114 Hints and outline solutions

(1) Successful problem-solving often depends on clinging to the optimistic

assumption that, although unfamiliar problems often seepi impossible at first,
they can usually be solved once you understand what is going on. This
question looks unfamiliar, and is at first sight very abstract in that it involves
an unknown function / which is not actually given, being described only in
terms of some of its properties. Despite all this, you have no option but to
take courage and set to! Where should one begin? The obvious place (since /
maps positive integers to positive integers) is to start thinking about the value
of fin) when n =_. Since /(1) has to be a positive integer, you certainly
know that /(1)>1. You also know that /(/(D) =_• It follows that /
cannot be the identity function. This may not seem to help, until you realize
that this, combined with the condition fin + 1) > fin), implies that fin) > n
(Why?). Use this fact to prove that /(D = 2.
(2) Use the same ideas to find /(2), /(3), and /(6).
(3) Once you know /(3) and /(6), the condition fin + l)>fin) should tell
you exactly what /(4) and /(5) have to be.
(4) You are now in business, since the condition fifin)) = 3n allows you to
fill in quite a few other values (such as f(7), /(8), /(9), and so on). It may still
not be clear what is going on, but at least the horrible feeling that you cannot
even begin should have begun to recede.
(5) If you construct a table of values for the function, you should realize that
there are long stretches where fin + 1) —fin) + 1, followed by stretches
where the value of fin) goes up in jumps. How large are these jumps? Later
on you will have to come back and prove the important bits, but the main
thing at this stage is to convince yourself that what seemed like a totally
inaccessible problem is in fact much more manageable than you thought. The
first step in making sense of what is going on is often to risk yourself by
making some kind of a guess as to what you expect to find in the next bit of
the table—that is, to formulate simple conjectures, and then to test them
against the facts to see if they stand up. If your guess turns out to be wrong,
stand back and try to improve it. You should decide fairly quickly that you
think you know where the jumps occur. It is not too hard then to decide what
you expect the value of /(1992) to be.
(6) All that then remains is to decide what to prove, and how to prove it. This
is not nearly as hard as it may appear provided that you sort out what to
prove first:

(a) Prove first (by induction on n) that /(3") =_ and that /(2 • 3") =_.

(b) Then prove that /(3" + k) =_for each k, 0 < k < 3”.

(c) Finally, deduce that /(2-3n + k) = 3n+1 + 3k for each k, 0<A:<3/1.

You can then use the proven facts (a), (b), and (c) to calculate /(2-36) =
_, and /(2-36 + 534) =_.
 27th BMO, 1991 115

27th British Mathematical Olympiad, 1991

1. Prove that the number 3" + 2 X 17", where n is a non-negative integer,

is never a perfect square.

(1) Let un = 3" + 2 X 17”. Work out the first four or five terms, keeping a
look-out for anything which might be helpful. (Of course, you could simply
look at the numbers that you obtain for each of the first few terms and see
whether they happen to be perfect squares. But that would miss the point!
I The challenge here is to prove that un can never be a perfect square for any
value of n. So what you should be looking for is something about the values
you are getting which has a chance of being true for every un, and which
might give you a clue as to why un can never be a square.)
(2) As n increases it takes longer to calculate un (and the risk of making an
error increases!). This makes it important not to waste information. What was
the first value of n you shouid have used in (1)? {Hint: the answer is not
n = 1.)
(3) Look carefully at the first few terms un. Do you notice anything which is
incompatible with being a perfect square and which might be true for later
terms as well? (In many mathematical problems it can be a disadvantage to
know too much. Fortunately, there are only two basic facts about perfect
squares, and most people know only one of them! So you are unlikely to be
handicapped here by knowing too much.)
(4) Write out the first 12 perfect squares and compare your list with the
numbers u0, uv u2,... . What differences stand out? This should give you an
idea. However, the crucial word in Question 1 is the first word: ‘Prove’. It is
not enough just to assert that the units digit of un follows some pattern (that
would get you at most 1 mark out of 4!). Patterns can deceive. It is your job
not just to assert, but to prove that what you think really happens does
indeed go on for ever.
(5) Any problem in which you have to prove that something is true for every
integer n > 0 should immediately suggest the method of proof by induction.
This leaves you to decide exactly what to prove and how to prove it. (Almost
any approach should convince you that it is much easier to work with the two
parts 3” and 2 X 17" separately.) You should formulate, and prove by induc¬
tion, a statement about the units digit of any number of the form 3" (n > 0).
Do the same for 2 X 17". Hence solve Question 1.
[You might like to try the following variation. For which values of n could
4" + 3 X 18" conceivably by a perfect square? Examine these values of n
more carefully and hence decide (and prove your assertion) whether it can in
fact ever be a perfect square.]
 116 Hints and outline solutions

2. Find all positive integers k such that the polynomial x2k+1 +x+ 1 is
divisible by the polynomial xk+x+\. For each such k, specify the
integers n such that xn +x + 1 is divisible by xk +x + 1.

(1) The first part of the question simply asks when x2k + +x + 1 is divisible
'i

by xk + x+ 1. The natural approach is therefore to try to carry out the long

division and see what happens. Do this, keeping track of the remainder as you
go along. (The difference between this and ordinary long division of polyno¬
mials is that here you are not trying to find the answer! Instead, you want to
discover which values of k could conceivably leave no remainder.)
(2) At some point in the division you should get a non-zero remainder r{x)
which does not involve k. Find r(x). If x*+1 +x + 1 is to divide x2k + l +x + 1
exactly, with no remainder, then xk + x+ 1 must divide this remainder r(x)
exactly. Hence the degree of xk +x+ 1 must be less than or equal to the
degree of the polynomial r(x). Now check each of the four possible values of
k to see which, if any, of them actually works.
(3) In the second part of the question, the appearance of three letters x, n,
and k can easily mislead you. Things are not nearly as complicated as they
seem, since the letter k merely stands for the value you found in the first
part! Rewrite the second part of the question, replacing the letter k by its
actual value. Then carry out the long division oi xn + x + \ by xk + x + \ and
find all values of n for which the division can end with remainder equal to
zero.
(4) Whether or not your first approach actually works, it is always worth
looking back to see if there is some way of improving, or simplifying, what you
did. How can you be sure at a glance that 7 divides 6993 exactly without
actually doing the division! (The answer has something to do with 6993 + 7.) A
polynomial fix) divides exactly into another polynomial g(x) precisely when
fix) also divides the sum gix) +fix); similarly, fix) divides g(x) precisely
w2^en fix) divides gix) — fix). Use this idea with fix) =xk +x + 1, g(x) =
*2*+1 +*+ L How does this simplify the problem? (You may still feel that
you prefer the original approach. However, this idea of manipulating polyno¬
mials like numbers is very powerful. At some stage in any mathematical
problem you have to know something, or work it out on the spot, or be
prepared to go away and learn it!

• Factorize x2k+1 -xk in the obvious way.

• Show that xk and xk +x + 1 can have no common factors. Hence con-
clude that xk+x+l divides x2k+1+x+l precisely when x*+x l +
divides x 1 - 1.
• Hence solve the first part of Question 2.

Use the same idea (or put * = exp(27ri/3) and use the Remainder
Theorem) to find all values of n for which xn+x + \ is divisible by
x +x + 1 (without remainder).)
 27th BMO, 1991 117

3. ABCD is a quadrilateral inscribed in a circle of radius r. The diagonals

AC and BD meet at E. Prove that if AC is perpendicular to BD, then

(*) EA2 + EB2 + EC2 + ED2 = 4r2.

Is it true that if (*) holds then AC is perpendicular to BD1 Give a

reason for your answer.

(1) In any geometry question, you must begin by drawing a suitable diagram.
(2) All those right angles (and the squares in the equation (*) that you are
trying to prove) should suggest that you need ****a*o*a*’ theorem. Use this
theorem to write down as many equations as you can. Choose two of these
equations and add to obtain an equation of the form

EA2 + EB2 + EC2 + ED2 = ...(**).

(3) The RHS of your equation (**) involves terms such as AB2 which do not
appear in (*), and does not yet involve the crucial quantity r2. The need to
involve r2 should give you itchy fingers, so that you can scarcely resist the
temptation to mark in the centre O of the circle in your diagram and then to
draw in the four obvious radii.
(4) This does not finish off the problem, but it should begin to get you
excited. For there are now four new triangles staring you in the face, each
with two sides of length r and with one other side which appeared in one of
the (many!) equations that you wrote down in (2). Use the *o*i*e rule (on
triangles AOB and COD or on triangles BOC and DOA) to substitute for
AB2 + CD2 (or for BC2 + DA2) in terms of r.
(5) Show that the equation (*) is true if cos A AOB + cosACOD = 0 (or
cos A BOC + cos A DOA = 0); that is, if A AOB + A COD = 180°. Then use
the fact that AC and BD are perpendicular to complete your solution to the
first part of the question.
[Alternatively, you may prefer a solution to the first part which avoids the
cosine rule:

EA2+EB2=AB2 and EC2 + ED2 = CD2.

So we only have to show that AB2 + CD2 = (2r)2. Let BE be the diameter
through B. Then

A AFB = A ADB (angles in the same segment)

A FAB = 90° (angle in a semi-circle)
A DEA = 90° (given)
.-. ZL ABF = 90° - A AFB = 90° - A ADE = A DAE.
AF = CD (chords subtending equal angles)

• (2r)2=AB2+AF2 (since AABF is right-angled)

= AB2 + CD2.]
118 Hints and outline solutions

(6) For the second part of the question you must think carefully about the
method that you have just used to answer the first part. The two sides of
equation (*) turned out to be equal because, when AC is perpendicular to BD,
both sides are equal to AB1 2 + CD2 (or BC2 + DA2). (For the LHS we used
Pythagoras (a special case of the cosine rule) on the right-angled triangles
ABE and CDE meeting at E, while for the RHS we used the cosine rule on
the triangles AOB and COD meeting at O.) You could do exactly the same
for any quadrilateral even when AC is not perpendicular to BD: all you need
for (*) to hold is that cosine terms which arise at E should cancel with the
cosine terms which arise at O. When the diagonals AC and BD are
perpendicular, the cosine terms on each side sum to zero and so balance
automatically. Another easy way of making the cosine terms on each side
balance is if the diagonals happen to cross at the point _ (that is, when
ABCD is a *e**a***e). Hence the converse is *a**e.
[You might like to find out what one can say about an arbitrary cyclic
quadrilateral ABCD if all you know about it is that it satisfies the condition
(*).]

4. Find, with proof, the minimum value of (x +yXy +z), where x, y, and
z are positive real numbers satisfying the condition xyz(x +y + z) = 1.

(1) This question may look quite unlike any sort of problem you are familiar
with. But that does not mean that it is beyond you! Good mathematical
problems are often like this. When you first read them you are left totally
bemused. But as you struggle to find a way to begin, and as you try out one or
two ideas, you gradually realize that they may not be all that hard. If you then
stick at it, a complete solution should emerge sooner rather than later. (After
experiencing this initial-frustration-leading-to-eventual-success a few times,
you will begin to realize that you should never be put off by first impressions.)
One off-putting feature of Question 4 is that there are three unknowns, x, y,
and z. If there were only two unknowns (x and y) you might interpret a
‘condition’ involving x and y as the equation of a curve in two dimensions.
You could then sketch the curve (geometry), or solve the equation to find y
in terms of x (algebra). The appearance of three unknowns makes the
question appear more forbidding. Admittedly the geometry is a little more
difficult: the equation xyz(x+y + z)= 1 is the equation of a surface in the
positive ‘octant’ (x > 0, y > 0, z > 0)—and we all find 3-D geometry harder to
visualize than 2-D geometry. But, as Descartes showed when he invented
co-ordinate geometry, algebraic fools can rush in and succeed where geomet¬
ric angels fear to tread! Algebra often generalizes easily, and can be applied
routinely; whereas geometry requires real thought and insight.
(2) As the equation xyz(x + y + z) = 1 stands, it is difficult to relate it to the
expression (x +y)(y +z) which we are trying to minimize. It would be much
 27th BMO, 1991 119

easier if the two could somehow be combined, so that the condition xyz(x +
y+z)= 1 becomes incorporated in the expression (x+yXy+z): you could
then concentrate on one expression and forget about the extraneous condi¬
tion. The most obvious way of incorporating the condition xyz(x +y +z)= 1
is to use it to *u***i*u*e for one of the variables in the expression
(x +yXy Tz).
(3) But which variable should you eliminate? The expression (x + yXy +z) is
only partly symmetrical in x, y, and z. Which one of the three variables in
the expression (x+yXy +z) sticks out like a sore thumb? Use the equation
xyz(x+y+z) = 1 to express this variable in terms of the other two. Then
substitute this back in the expression (x +yXy +z) and simplify.
(4) This does not solve the problem. But it certainly looks like progress in
that the expression you have to minimize now has only two variables and is
beautifully symmetrical. So you no longer need to be so frightened by it! You
might even try substituting a few easy values for x and z to see what values
the expression takes. (The fact that you have substituted for y using the
condition xyz(x +y +z) = 1 means that this condition is now automatically
satisfied.) For example, find the value of y when x = z = 1. What is the value
of the expression (x +yXy + z) in this case?
(5) To finish things off properly you must now prove (using the AM-GM
inequality for two variables, or otherwise) that for any positive real number u,
u + (1 /u) > 2. Hence complete your solution of Question 4.
(6) Having struggled to produce a solution, it is always worth looking back
over it to see if there is anything you can learn from it. You should now
realize the importance of incorporating the condition xyz(x +y +z) = 1 into
the expression (x+yXy+z) in some way. Can you see another way to
rewrite (x +yXy + z) which will allow you to substitute very simply using the
fact that xyz(x + y +z) = 1, and which leads very easily to the same simplified
expression as in (3)?

5. Find the number of permutations (arrangements)

jv ?2> ?3’ ?4> it of 2, 3, 4, 5, 6

with the property

for no integer n, 1 < n < 5, do jx, j2,..., j„

form a permutation of 1,2

(1) At a first reading it may not even be clear what the question is saying. Do
the symbols ]\, j2, ;3, j4, j5, and ;6 stand for six different permutations of 1,
2, 3, 4, 5, and 6? Or is % j2, j3, ;4, j5, j6’ a single permutation of T, 2, 3, 4, 5,
6’? One of these two interpretations makes a nonsense of the property at the
end of the question and so is obviously wrong! So which is correct?
120 Hints and outline solutions

(2) Once you have sorted out what the words mean, you may still not see how
to begin. In that case you could begin to work your way into the question by
trying to calculate the corresponding number of permutations for some easier
cases. Let pn be the number of permutations jiJi’-'-’Jn °f 1>2, ...,n with
the property:

(*) for no k < n is j\ ,j2,. .., fy a permutation of 1,2,..., k.

Clearly, px = 1. Work out p2. Then work out p3.

(3) You should not expect to find some simple-minded pattern: mathematics
is much more interesting than that! It is easy to get the wrong idea about why
it is often worth looking at what happens for small values of n. The answers
themselves are not very interesting. Instead, you should be looking for a
manageable method which might help you to calculate p6—a number which
in this case may well be around 500, since the number of all permutations of
1, 2, 3, 4, 5, and 6 is 6! =_. (If you try to count without a suitable
method, you are bound to make mistakes—probably already when calculating
p4, and almost certainly when calculating p5 and p6.) Try to think of a
general method for calculating p4. Use your method to find p4. Finally, write
down (in some systematic way) all permutations jv j2, j3, j4 of 1, 2, 3, 4
satisfying the condition (*) in (2) above, count them, and so check your value
for p4.
(4) There are many possible general methods. Perhaps the most simple-
minded one is to observe that p4 counts

all permutations jx, j2, ;3, j4 of 1, 2, 3, 4,

except those which do not satisfy (*).

(a) How many permutations jv j2, j3, j4 of 1, 2, 3, 4 have j\ = 1? (b) How

many have {jx,j2} = {1,2}? (c) How many have 0'i,72,73} = {1,2,3}? What
value does this suggest for p4?
(5) Unless you are alert, the values you obtained for p4 in (3) and in (4) will
be different! What can have gone wrong? The method in (4Xa) counts all
permutations of the form 1, whereas the method in (4Xb)
counts all permutations of the form 1, 2, _, _ or 2, 1, _, _. Clearly,
some permutations get counted in both (4Xa) and (4Xb)’(and again in (4Xc)X
To avoid this, (4)(b) for example should be adjusted so that it only counts
permutations of the form 2, 1, (that is, permutations j2, j3, j4 of
1, 2, 3, 4 such that {jx,j2} = {1,2} but jx ¥= 1). How many of these are there?
(6) Show that p4 = 4! ~(Pl X 3! +p2 X 2! +p3 x 1!). Find p5. Then find p6.
[You might like to try to find a formula for pn.]

6. Show that if x and y are positive integers such that x2+y2-x is

divisible by 2xy, then x is a perfect square.
 27th BMO, 1991 121

(1) Like many good mathematics problems, this one looks a little strange at
first sight. Working your way into a problem can take a long time. You have
to show that, under the given conditions, the positive integer x has to be a
square. There are clever ways of doing this, but, as so often with Olympiad
problems, there is also a very simple-minded approach—provided that you
have done your homework on ‘A little useful mathematics’. How can one
recognize when r is a perfect square? One way is to imagine the number x
already factorized as a product of prime powers * =p“ -pu2 ■... -p”. Then x
will be a perfect square precisely when all the exponents u,v,...,w are e*e*.
(2) Suppose that p is a prime number which divides x. Show that p must
also divide y. Hence show that p1 2 3 must also divide x.
(3) This certainly looks promising, but we need a more general observation
for a complete solution. Suppose that p is a prime number such that some
odd power p2l~l divides x. Show that then p2‘ must divide x. Hence write
out a complete solution to Question 6.
[Alternatively, you may interpret the given condition ‘x2 +y2 -x is divisi¬
ble by 2xy’ as an equation; solve it (as a quadratic in _); hence find a
different solution to Question 6.]

7. A ladder of length / rests against a vertical wall. Suppose that there is a

rung on the ladder which has the same distance d from both the wall
and the (horizontal) ground. Find explicitly, in terms of l and d, the
height h from the ground that the ladder reaches up the wall.

(1) Draw a diagram and mark the lengths /, d, and h.

(2) It is natural to try to solve the problem directly. Let the foot of the ladder
be distance / from the wall. Use *i*i*a* triangles to show that f/h =
(f—d)/d. Hence find / in terms of h and d. Then use Pythagoras’ Theorem
to obtain an equation involving only h, d, and /. Hence try to find an
expression for h in terms of / and d.
(3) The difficulty with this straightforward approach is that the equation that
one obtains linking h, l, and d is a quartic in h (that is, it involves h4), and
you are unlikely to know a formula for the roots of such an equation. You
would obviously prefer a *ua**a*i* equation—that is, one involving only h
and h2 (as well as l and d). One possible way of finding such an equation for
h is to notice that, if h does indeed satisfy a quadratic, then the quartic you
obtained in (2) must factorize as the product of two quadratics. So it is
natural to try to factorize the quartic that you found in (2).
(4) While this must be possible, there is no reason why the coefficient of h2,
the coefficient of h, and the constant term in each quadratic factor should be
simple monomials (such as Id, or l2): they might, for example, involve square
roots. This makes the task of factorizing the quartic rather hard. It may be
easier to look for a direct way of producing a quadratic. For that you need an
122 Hints and outline solutions

idea which explains why the length h should satisfy a quadratic. Look again at
the diagram that you drew in (1). Any equation which has h as one root will
automatically have another root corresponding to one of the other lengths in
your diagram. Which length is it? Why must the roots come in pairs like this?
(At first sight you may think there is an asymmetry between the vertical
height h up the wall and the horizontal distance / from the wall to the foot
of the ladder. But if you turn the diagram through a right angle it should be
clear that geometrically the two lengths h and / are indistinguishable. Thus
one should expect that any equation which has h as one root will have / as
another root and that alebraically there will be no way of distinguishing
between them.)
(5) If the roots of a quadratic x2 -px + q are a and /3, what are p and ql
Thus all you need to do is to find expressions for h+f and for h •/ in terms
of / and d only. {Hint: If h2 — ph + q = 0, show that p2 = l2 + 2q and dp = q.
Hence find p and q.)

26th Britirh Mathematical Olympiad, 1990

1. Find a positive integer the first digit of which is 1 and which has the
property that, if this digit is transferred to the end of the number, the
number is tripled.

(1) Which is the ‘first digit’? And what is meant by ‘transferred to the end of
the number’? A moment’s thought should make it clear that the first digit
must be the one on the left-hand end and that ‘transferred to the end’ must
mean ‘moved to the right-hand end'. (Whenever you are faced with a slightly
unusual problem, it is tempting to think that the question is unclear in some
way. But a more careful reading will nearly always show that there is only one
possible interpretation. In this case, if the 1 were moved to the left-hand end,
the final number could not be even twice as large as the original number!
Thus the original number must start out with a 1 on the left-hand end, and
the 1 must get moved to the right-hand end.)
(2) Since we do not know how long the original number is, we may as well
write it as 1_. We must then solve either

2X
1 3]znn
So it is really a question about multiplication (or division). Which operation is
easier to think about—multiplication, or division?
(3) Concentrate on the first version—the easier one. The only unusual thing
 26th BMO, 1990 123

about this is that we are given the multiplier ‘3’, and the units digit ‘1’ of the
answer, and we have to work out the units digit of the original number:

• What must the units digit of the original number be?

• Which other previously unknown digit can you now fill in?
• What does this now tell you about the tens digit of the original number?
» Keep going.

(4) Of course it is important to remember that the two extended blanks

‘_’ in (2) represent exactly the same string of digits. So once you know the
units digit of the original number, you immediately know the *e** digit of
the answer. Knowing this and the multiplier ‘3’ then allows you to work out
what the tens digit of the original number must be. But this is the same as the
*u***e** digit of the answer—and so on.
(5) How do you know when to stop? (How many different answers are
there?)
(6) Solving a problem is only the beginning. When you come to write out
your final solution you should always look for a more mathematical approach.
The method used above is easy to carry out, but not so easy to explain in
words. One reason for this difficulty is that you have found a procedure that
works, and this tends to block further thought about why it works. A
non-mathematician may be content to have a method that works; but the
mathematician always wants to know what makes a procedure work. One
approach here is to denote the extended blank ‘_’ in the original number
by V:

• The blank in the answer may be the same string of digits; but they have
been shifted one place to the left. Thus the blank in the answer is not x,
but _.
• If jc has n digits, then the 1 at the front denotes not 1, but _, so the
original number was x +_.

(7) The question now gives rise to an equation involving x and n:

• What is this equation?

• How can you solve it to find (n and) *? How many possible answers are
there?
• What is the connection between these answers and the (recurring)
decimal for f ?

[You might like to try the following variation. Find a positive integer the
first digit of which is a and which has the property that, if this digit is
transferred to the end of the number, the resulting number is exactly b times
the original:
 124 Hints and outline solutions

(a) For which integer values of a and b does this problem have a solution?

(b) Give a complete description of all possible solutions for each pair a, b.]

2. ABCD is a square and P is a point on the line AB. Find the maximum
and minimum values of the ratio PC/PD, showing that these occur for
the points P given by AP X BP — AB2.

(1) If P is a point on the line AB, you may think that PC/PD is a maximum
when P = A, and that PC/PD is a minimum when P — B. Why is this wrong?
(2) As always, you have to sort out what the question really means: that is
part of the challenge. In this case it clearly depends on what is meant by ‘the
line AB’. Even if you jump to the conclusion and (mis)interpret this as
meaning ‘the line segment between A and B\ you should notice two things.
First, if P=A, then the required condition AP X BP=AB2 does not hold, so
something is clearly wrong. Second, the ratio PC /PD clearly increases as P
moves from B towards A, and so presumably goes on increasing for a while if
P continues to move past A on the line BA. (Gould it increase for ever?) All
this suggests that ‘the line AB’ must mean the infinite line through A and B.
(3) Suppose that A is to the left of B. Imagine P moving along the infinite
line AB:

• What happens to the ratio PC/PD as P whizzes off past B to +°o.

• What happens to the ratio PC/PD as P whizzes off past A to -oo.

If A is the top left-hand corner of the square ABCD, then the point P = Pmax
at which PC/PD is a maximum must occur when P is somewhere to the left
°f A. The point P' =Pmin, at which P'C/P'D is a minimum, occurs when
P'D/P'C is a maximum-, hence Pmin must be exactly as far to the right of B
as Pmax was to the left of A.
(4) Now that you know what you are looking for, it is natural to take a
general point P and call something ‘x’; then use Pythagoras to express PC
and PD in terms of x, and find an expression for PC/PD in terms of x;
finally, find the value of x at which this expression is a maximum or a
minimum, and then solve the original problem.
(-*) There are many other approaches. For example, the answer ‘AP X BP =
AB ’ is given! And, while products occur relatively rarely in geometry, one
often comes across *a*io*— especially in triangles. Thus, the given relation
AP X BP =AB should suggest looking at PA/AB (which is equal to PA /AD
-that is, tan A PDA). Let A PDA = 6. Find PD in terms of 6 and the side 5
of the square ABCD (using A PAD). Find (PC/PD)2 in terms of 6 (using
the cosine rule in APDC). Show that (PC/PD)2 = 1 + cos20 + sin20. Con
elude that, for PC/PD to be a maximum or a minimum, t = tan 6 must
satisfy t + t - 1 = 0. Hence complete the solution.
 26th BMO, 1990 125

[You might like to try the following variation. Let C and D be any two
points in the plane and let k > 0 be any real number:

(a) Show that the locus of points P such that PC/PD = k is (i) a circle with
C inside if k < 1; (ii) the perpendicular bisector of the segment CD if
k = 1; (iii) a circle with D inside if k> 1.

(b) Show that the point P on the line AB at which PC/PD is a minimum
corresponds to the value of k for which the circle in part (a) just touches
the line AB.}

3. The angles A, B,C, and D of a convex quadrilateral satisfy the relation

cos A + cos B + cos C + cos D = 0.

Prove that ABCD is either a trapezium or is cyclic.

(1) A convex quadrilateral is a quadrilateral with no ‘dents’—that is, with all

angles < 180°. It is often a good idea to begin by writing down the simplest
facts which seem to be relevant. They may not be the most important, but
one should certainly not overlook them. (Moreover, putting something down
on paper can help to get things moving and give one a feeling of progress.)
What do you know about the four angles A, B,C, and D in any quadrilateral
ABCD? Use this fact to express cos D in terms of A, B, and C.
(2) So, if you wish, you can eliminate D from the given equation cos A +
cos B + cos C + cos D = 0. But that would spoil the symmetry, so perhaps you
should stay with the original equation for a while. What do you know about
the angles A and C (or B and D) of a cyclic quadrilateral ABCD1 In what
sense do the angles A, B, C, and D in a trapezium ABCD satisfy something
similar?
(3) Your answers to the two questions in (2) suggest that, whenever a convex
quadrilateral ABCD satisfies cos A + cos B + cos C + cos D = 0, you should
try to prove that either two opposite angles, or two adjacent angles, have sum
equal to _. It is time to go back to the original question. What could you
possibly do with a sum of four cosines? (You really have very little choice, for
there is only one obvious thing to do with a sum of cosines!)
(4) But even if you had to make a choice, you should realize that, while one
can never conclude very much from an equation of the form ‘sum of
things = O’, one can conclude much more from an equation of the form
‘product of things = O’. So you should instinctively want to change the sum
on the LHS of the original equation into a product:

• What is the trig identity that expresses cos X + cos 7 as a product of

cosines?
126 Hints and outline solutions

• Apply this to cos A + cos B and to cos C + cos D.

• Factorize the expression for cos A + cos B + cos C + cos D. (Why is
cos j(A + B) = —cos j(C + D)?)
• Use the identity for cos X — cos Y to express the sum cos A + cos B +
cos C + cos D as a product of three terms.

How does this solve the problem?

(5) When you think you have managed to solve a problem completely, it is
often a good idea to stand back and take a long hard look at what you have
done. Read the question again carefully, check your working, think about
what you actually used in achieving your solution, and so on. And don’t be
surprised if you discover a mistake: progress in mathematics often consists of
taking two steps forward and one step backwards. Even when you don’t find a
mistake, you will often find that you have overlooked something! The original
question refers to a convex quadrilateral; that is, a quadrilateral that does not
have any ‘dents’. Where does your solution make use of the fact that the
quadrilateral ABCD is convex?
[You might like to use the same ideas to find all quadrilaterals ABCD with
the property that sin A + sin B + sin C + sin D = 0—or to prove that none
exist.]

4. A coin is biased so that the probability of obtaining a head is p,

0 <p < 1. Two players, A and B, throw the coin in turn until one of the
sequences HHH or HTH occurs. If the sequence HHH occurs first,
then A wins. If HTH occurs first, then B wins. For what value of p is
the game fair (that is, such that A and B have an equal chance of
winning)?

(1) Probability questions are rarely ‘easy’.

HHHH
But when dealing with a sequence of HHH
repeated tosses like this, one can often
obtain a better picture of what is going
on by drawing a simple tree-diagram like
the one here. Circle those nodes where
A wins and put a cross where B wins. Try Start
to make sense of what you find.
(2) Denote the probability that A wins
the whole game by a and the probability
that B wins by /3. If the first toss is a tail,
this helps neither A nor B. So the proba¬
TTT TTTT
bility that A wins a game that has al¬
ready started out T... is still _; similarly, the probability that B wins
such a game is _. Thus the probability that A wins a game and that the
 26th BMO, 1990 127

first toss is T is (1 -p) X_. Similarly, HHH

the probability that B wins a game and
that the first toss is T is (1 —p) X_.
(3) A can win in four ways:

at 1, with probability _
via 3, with probability X

via 5, with probability X

via 6, with probability X

Hence a = + X +
solve for a in terms of p.
(4) Do the same for /3. Finally, choose p such that a = (3.
[Give an example of two sequences of tosses for which neither A nor B
wins. It should therefore come as a slight surprise that a + (3 =_. Can you
prove this directly?]

5. The diagonals of a convex quadrilateral ABCD intersect at O. The

centroids of triangles AOD and BOC are P and Q\ the orthocentres of
triangles AOB and COD are R and S. Prove that PQ is perpendicular
to RS.

(1) Every mathematics problem has to be understood before you can begin to
solve it. A convex quadrilateral is one without dents (that is, with all angles
< 180°). You are told that the quadrilateral ABCD is convex. What does this
tell you about the point O?
(2) Drawing a diagram proves absolutely nothing. But it is a very good way of
making sure that you understand the problem. The very act of drawing can
sometimes suggest ways of solving the problem. And an accurate drawing may
reveal coincidences which give you other ideas. Draw an accurate diagram
(for an arbitrary convex quadrilateral ABCD). Mark the triangles AOD,
BOC, AOB, and COD. Check the meaning of centroid and orthocentre (in ‘A
little useful mathematics’). Mark P, Q, R, and S on your diagram. Make sure
that PQ and RS look perpendicular: if they don’t, your diagram must be
inaccurate (or else the question is wrong).
(3) Perhaps the most obvious way of proving two lines are perpendicular is to
use vectors and to show that some dot product is equal to zero. What would
be a good point to choose as the origin? Suppose that the points A, B,C, D,
R, and S have position vectors a, b, c, d, r, and s. Find the position vectors of
P and Q in terms of these. What do you know about the lines OR and AB?
What does this tell you about the vectors r and a - b? What do you know
about s and c — d?
(4) Show that PQ is perpendicular to RS provided that s (a - b) = r (d - c).
128 Hints and outline solutions

Use the many pairs of perpendicular lines in the diagram (such as SC, BD
and SD, AC) to show that s (a - b) = a d - b e. Then do the same for
r • (d — c), and hence solve the problem.
(5) At this point one tends to sit back and feel vaguely smug, when one
should in fact look back and see what one can learn. Strictly speaking, a
convex quadrilateral could have one angle equal to 180°. In such cases it is
sometimes worth looking back to check what happens when one of the angles,
say A, is equal to 180°? Where is the point 01 Where is the point R1 And
where is the centroid P of triangle AOD1

6. Prove that if x and y are rational numbers satisfying the equation

x5 +y5 = 2 x2y2

then 1 — xy is the square of a rational number.

(1) Equations in rational numbers are a bit like equations in integers. But it
is not always a good idea to go from an equation in two unknown rationals x
and y to an equation in four unknown integers p, q, r, and s by writing
x=p/q and y = r/s. Instead, you should try to simplify.
(2) If x and y are rational numbers, then so is x/y. Although the given
equation is not quite homogeneous (that is, having all terms of the same
degree), it is tempting to divide both sides by y5 to obtain

(x/y)5 + 1 = 2 (x/y)2/y.

Now put x/y = t. (If x and y are rational (and y ¥= 0), then so is t.) Obtain an
expression for y in terms of t. Obtain an expression for x in terms of t.
(3) Show that if t is any rational ¥= — 1, substituting these two expressions for
y and x in the original equation gives a rational solution. How does this solve
the original problem?

25th British Mathematical Olympiad, 1988

1. Find all integers a, b, and c for which

(x - aXx - 10) + 1 = (x + bXx + c) for all x.

(1) This is not an equation to be solved! It is meant to be an identity, with

the two sides being equal for all x. Hence the two sides must be equal as
polynomials, with the same coefficients on each side. The coefficient of x2 on
each side is obviously 1: so far, so good! How about the coefficient of x? To
make those equal we need

—a — 10 = b + c.
 25th BMO, 1988 129

And how about the constant terms on each side? To make those equal we
must have

10a + 1 = he.

That gives us two equations in three unknowns. If you eliminate a and take
everything to one side you obtain one equation involving two unknowns, b
and c, of the form be +_b +_c +_= 0. And then you seem to be
stuck!
(2) Go back and read the question again: ‘Find all integers a, b, and c...\
Aha! So a, b, and c are not any old unknowns; they have to be integers. The
most important thing about integers is the way in which they factorize.
Rewrite the equation obtained in (1) in the form

_= 1.

Then factorize the LHS and find all solutions. (This method only works
because we know that b and c are integers. If b and c were arbitrary real
unknowns, you should automatically make the RHS = zero before trying to
factorize.)
(3) Now that you have managed to find a solution, it is time to reflect. Look
back at the question. Can you see how your final equation ‘( X )= V
could have been obtained directly from the identity given in the question?
What value should you choose to substitute for x in the original identity?

2. Points P and Q lie on the sides AB and AC respectively of triangle

ABC and are distinct from A. The lengths AP and AQ are denoted by
x and y respectively, with the convention that jc > 0 if P is on the same
side of A as B, and x < 0 on the opposite side; and similarly for y.
Show that PQ passes through the centroid of the triangle if and only if

3xy = bx + cy,

where b = AC and c = AB.

(1) The whole question reeks of vectors. Which is the obvious point to choose
for the origin 0? Suppose that AB = c has length c and AC = b has length
—> —>

b. Express AP and AQ in terms of these.

(2) One basic property of vectors is that a point X lies on the line joining P
and Q precisely when

OX = \OP + (1 - X)OQ

for some real number A: if A = 0, then X = Q; if A — 1, then X — P; if A —

then X is the midpoint of PQ. Let G be the centroid of triangle ABC. There
130 Hints anti outline solutions

is a simple expression for AG in terms of b and c. What is it? Use this to

show that
_ x y
AG = A — c + (1 - A) — b
c b

(that is, that G lies on PQ) precisely when

3 xy = bx + cy.

3. The lines OA, OB, and OC are mutually perpendicular. Express the
area of triangle ABC in terms of the areas of triangles OBC, OCA, and
OAB.

(1) The first move in any geometric question must be to draw a diagram. It is
not easy to draw good 3-D diagrams. But in this case if we think of OA, OB,
and OC as the (positive) x-, y-, and z-axes, it should be possible to produce
something reasonable.
(2) What is ‘special’ about all three triangles OBC, OCA, and OAB? Your
answer should suggest that the areas of these three triangles are best
expressed in terms of the three lengths OA = x, OB = y, and OC = z. Use
these lengths to find the areas of A OBC, A OCA, and A OAB.
(3) How about the area of A ABC? The simplest ways of finding the area
of A ABC (|base X height, and \absmC) require other information about
A ABC. We don’t yet know the ‘height’, or any of the angles, but we should
be able to find them now that we know the lengths of the three sides. Use
OA =x and OB=y to find AB, and hence the length of the altitude OD
from O to AB. Hence calculate the height CD of triangle ABC, and the area
of triangle zl.BC. Find a formula giving the area of A ABC in terms of the
areas of A OAB, A OBC, and A OCA.
(4) This direct approach should succeed. An alternative is to use the formula
(Heron’s formula) which gives the area of triangle ABC purely in terms of
the lengths of the three sides AB = c, BC = a, and CA = b. Let 2s = a + b + c
be the perimeter of the triangle. Then

(area AABC)2 = s(s - a)(s - b)(s - c).

Use the known expressions for a, b, and c (in terms of x, y, and z) to answer
the original question. (The algebra may look horrible at first, but if you
calculate intelligently the expression for (area A ABC)2 in terms of x, y, and
z simplifies dramatically.)
(5) The formula area = jab sin C’ may suggest a connection with vectors.
The dot product (or scalar product) of two vectors a and b is a real number,
and is very useful when trying to ‘resolve’ a (along b and perpendicular to b).
This is not what we want here, since a • b = |a| |b|cos d depends on the cosine
 25th BMO, 1988 131

of the angle 6 between the two vectors. However, there is another product of
a and b the cross product (or vector product), a X b: this is a vector of
length |a| |b| sin 6 (= the area of the parallelogram spanned by a and b) and
with direction perpendicular to a and b. Hence

area of triangle ABC = \\ AB XAC|.

At first sight this may just look like a complicated way of saying ‘area =
jbc sin A\ But it has one tremendous advantage: once you are used to it,
calculating vector products is straightforward!
Let OA=xi, OB=yj, and OC=zk, where i, j, and k are mutually
perpendicular unit vectors (forming a right-handed system). Write down AB
and AC in terms of i, j, and k. Use the fact that j Xj = k=-jXi,j X k = i =
—kXj, kxi=j = — ixk, to work out AB XAC purely by algebra, without
worrying about ‘angles’. Hence answer the original question.

4. Consider the triange of numbers on

the right. Each number is the sum
l
of three numbers in the previous
row: the number above it and the l l l
numbers immediately to the left
and right of that number. If there is 1 2 3 2 1
no number in one or more of these
1 3 6 7 6 3 1
positions, 0 is used. Prove that,
from the third row on, every row
contains at least one even number.

(1) There is only one obvious way to begin to get some feeling for this kind of
simple-but-unfamiliar array: write out the first ten or so rows for yourself. Do
you notice any patterns at all (perhaps familiar, perhaps not) in the array of
numbers you produce?
(2) Each number in the next row only depends on the three numbers
immediately above it. Hence the left-right symmetry in the first few rows is
bound to continue. Thus, in each row, the first half is just the reverse of the
last half. There are several other things that you should have noticed. The
way in which the left-hand sloping edge is produced shows that it must consist
entirely of l’s. The way in which the next sloping diagonal is generated shows
that it must consist of the natural numbers 1,2,3,... (Why?). You should
recognize the numbers in the next sloping diagonal. Can you prove that the
nth number in this sloping diagonal really is what you think it is?
(3) It follows from these initial observations that most rows have an even
number in either the second or the third position. So, if we want, we can
forget about those rows and concentrate on the rest. Suppose that we label
132 Hints and outline solutions

the rows by the sequence of natural numbers in the second sloping diagonal
(the top row ‘1’ is then the Oth row; the next row ‘1 1 1’ is the 1st row; and so
on.) For which values of n does the nth row have odd numbers 1, odd, odd,...
in the first three positions?
(4) Look at the 5th and 9th rows. What do you notice about the numbers in
the fourth position in each row?
(5) Of course, what you noticed for the 5th row and the 9th row may not be
true for the 105th row and the 109th row. The original question is deliber¬
ately vague about where one should look in a row to find the even numbers.
And the first ten rows tend to suggest that there is no simple general pattern.
However, (4) may suggest that the original vague question could perhaps be
resolved by proving something much more specific; namely, that the fourth
number in the (4rc + l)th row is always even, for every n ^ 1. Let fm denote
the fourth number in the rath row:

• Show that + \m{m + 1) + fm.

• Use fm = £7L2(/, -/,_!> to find a formula for fm. (Use the formula for
£r=u2-)
• Explain why this solves the original problem.

(6) Mathematicians, like politicians, succeed by applying the principle ‘divide

and rule’. The above approach starts by picking off the easy cases. Half the
rows have an even number in the second position; and half the remaining
rows have an even number in the third position. This allows one to concen¬
trate on every fourth row—the (4n + l)th rows, n ^ 1. This strategy can be
very effective. But afterwards it is often worth looking back to see if there is a
more uniform approach. If you really want to gain insight into the way odd
and even numbers occur in this curious number triangle, you will have to look
at much more than the first few rows. Unfortunately, the whole process of
adding three numbers to find the next number gets in the way, and soon leads
to errors. This is particularly annoying, since you don’t really care whether
the middle number in the eighth row is 1107 or 1109 as long as you know that
it is odd. This suggests simply writing ‘O’ for odd and ‘E’ for even, and
adding in the obvious way (O + O = E,
O + E = O, and E + E — E). Take a large 0
sheet of paper and write out the first 20 o o o
rows of the number triangle, entering ‘O’
for odd and ‘£’ for even entries. You O E o E o
already know that the two outside diago¬
nals consist entirely of O’s (namely l’s). o o e o e o o
What other striking patterns of O’s and.
E’s do you notice? Which of these patterns can you prove goes on for ever, as
you suspect?
 25th BMO, 1988 133

(7) You may have noticed one or two places where odd numbers seem to
occur. It seems much harder to pin down anything very useful about what you
really want to know—namely about where the even numbers occur. Still, you
should at least have noticed, and been able to prove, that there is one vertical
column which consists entirely of odd numbers. It is not clear how that might
help, but perhaps you should bear it in mind.
(8) Whenever you want to prove something (such as ‘every row after the first
two contains at least one even number’) and have no idea how to begin, there
is one truly marvellous way of getting started, which has the double advantage
of making the whole problem much clearer and of giving you additional
information apparently for nothing. (Does this seem too good to be true?
Well, there is a hitch; but the advantages are so impressive that you probably
won’t even notice.) You want to show that, except for the Oth and 1st rows,
every row contains at least one even number. All that you know for sure is
that the first, the last, and the middle number in each row is odd. Apart from
that, you are pretty well stuck! So what do you do? Just relax! Stop trying to
prove directly that every row must contain at least one even number. Instead,
star*; thinking about what it would mean if the thing you want to prove true
were actually false: that is, if some row were to contain no even numbers at
all. If you can show that this is only possible for the first two rows, then you
will be home and dry.
The advantage of this switch is that instead of trying to think about some
arbitrary nth row, you can concentrate on one particular row about which
you know something very specific; namely, that it consists entirely of odd
numbers. That is, not only does the number triangle begin as you know it
begins; but somewhere further down the triangle there would be a row
consisting entirely of odd numbers. At this stage you do not know how far
down the triangle this ‘odd’ row is, so you don’t know how long the row is.
But just suppose that there is a row somewhere. What can you say about the
previous row? And what can you say about the row before that? How does
this solve the original problem?

5. None of the angles of a triangle ABC exceeds 90°. Prove that

sin A + sin B + sin C > 2.

(1) What an amazing inequality! Where on earth could it have come from?
And how good is it? What is the value of the LHS for an equilateral triangle?
What about an isosceles right-angled triangle? What about an isosceles
triangle with a very small apex angle? What about your favourite acute-
angled triangle? Can you find an obtuse-angled triangle for which the
inequality is false?
(2) Well, how are you going to prove it? The LHS mentions all three angles
A, B, and C; but angle C clearly depends on A and B. Thus it may seem
 Hints and outline solutions
134

natural to substitute for C in terms of A and 5, and to use the basic trig
formulae to rewrite the LHS. Rewrite sin A + sin B as a, product, take out a
common factor; then use the formula for cos X + cos 7 to express the LHS as
a product; finally, use sin +B) = cos(C/2) to obtain the original LHS in
the form 4cos(yl/2)cos(5/2)cos(C/2). Since the angles A/2, 5/2, and C/2
are all <45°, this product is clearly > ^2 . This shows that the inequality to be
proved depends on the fact that, within the range (0,45°], we must somehow
play off the exact value of C/2 against the values of A/2 and 5/2.
(3) Go back to the less symmetrical expression for the LHS, namely
2cos(C/2Xcos j(A + 5) + cos j(A - 5)). For each given value of C, the
value of the first factor ‘cos(C/2)’ and the value of A+ B are both fixed;
hence we need to find a lower bound for cos j(A — 5), that is, an upper
bound for \{A - 5). Therefore, if C is fixed, A + 5 (= 180° - C) is fixed, so
to find a lower bound for cos^(/l —5) we must take A as large as possible
and 5 as small as possible; that is, A =_, 5 =-. To obtain a lower
bound for the original LHS, we must therefore choose C (< 90°) so that
2cos(C/2Xsin(C/2) + cos(C/2)) = sinC + cosC+1 is as small as possible.
(For this, all you need is that, since 0 < C < 90°, both sin C and cos C lie
strictly between 0 and 1, so sin C + cos C > sin1 2 3C + cos2C = 1.)
[Alternatively, working in terms of radians rather than degrees, you might
like to prove that when 0 < 0 < tt/2 we have sin 0 > 20/7r (with equality only
for 0 = 77-/2). Hence sin A + sin 5 + sin C > (2/nXA + 5 + C)!]

6. The sequence {an} of integers is defined by ax = 2, a2 = 7, and

1 a2n 1
(*) -T<0n+i-for n > 2.
2 an_x 2

Prove that an is odd for all n > 1.

(1) How curious! Given the inequalities for an + 1, it is not at all clear how an
behaves. The way in which an+1 is defined seems completely inscrutable.
Still, the only way of making it less inscrutable is to use it to work out the
next few terms. The whole question takes it for granted that, once we know
an_x and an, the inequalities should determine an + x uniquely. Use (*) with
n = 2 to find a3. Is it uniquely determined? Use (*) with n = 3 to find a4. Is it
uniquely determined?
(2) Find a5, a6, a7, a8, and a9. Are they all odd?
(3) By now it should at least be clear that the inequalities defining an + x are
not as complicated as they seemed. Once we know that an_x and an are
integers, it follows that a2n/an_x is a rational number; an + x is then just the
nearest integer to this rational number. But why on earth should this ‘nearest
 25th BMO, 1988 135

integer’ always be odd? And why is the fraction a\/an_ j always so close to
the nearest integer? Go back and work out the actual differences an+x —
a2n/an-\ when n = 2, 3, 4, 5, 6, 7.

(4) At this stage you should ‘smell a rat’, even if you don’t know how to catch
it. Examine successive ratios an + x/an, l<n<8. Do you notice anything
interesting about this sequence of ratios? Can you rearrange the inequality
(*) so that it involves successive terms of the sequence {an+x/an}7
(5) This more symmetrical version of the defining inequality may turn out to
have certain advantages. But it still conceals rather than reveals why it might
be true that ‘an is odd for all n > V. There may be mathematical techniques
that could help here, but if one doesn’t know them, then one has to try
something more straightforward. Could there be some simpler rule which
defines our sequence {an}, perhaps like the rule for generating Fibonacci
numbers? Look carefully at the ratios of successive terms

0-2 O^ O^ O^

a1 0-2 ^3 #4

The fractional part of each ratio may be a little hard to understand; but there
is something remarkably constant. This suggests that each an+x consists of
some fixed multiple k-an of the previous term an, plus an extra bit. What
looks like the most promising value of k7 Can you make sense of the ‘extra
bit’ that is needed each time? If you write out the sequence {an + x — kan) of
‘extra bits’ required, it should not be long before you think you know what is
going on.
(6) You should now have one official defining rule (*) for the sequence {an},
and another much simpler rule which seems to define the same sequence.
However, all of this is wild (one hopes, inspired) guesswork; you don’t know
that the two sequences are the same. They certainly start in the same way;
but the defining rules look so different that this could easily be an accident.
To keep the two sequences apart in your mind, give the new, simpler
sequence a different name {bn}: that is, {bn} is the sequence in which bx = 2,
b2 = 7, and bn + x is given by the simple recurrence bn + x = 3bn + 2bn_x. You
have to prove that this is just an alternative description of the original
sequence. But there is no point doing this unless you have some idea how this
will help you to solve the original problem.
(7) Explain why bn has to be odd for all n > 1. Now take a deep breath and
try to show that bn = an for all n. Clearly, {bn} is a sequence of integers.
Suppose that you know that bx = ax for all i < n. Show that

1
<
2b,n

Hence solve the original problem.

136 Hints and outline solutions

24th British Mathematical Olympiad, 1987B

_____ _1- —

1. Find all real solutions x of the equation

Ax + 1972 098 - 1986 Ax + 986 049))

+ Ax + 1974 085 - 1988 Ax + 986 049)) = 1.

(1) This is Question 1 on the paper, so it ought to be reasonably easy. You

may be tempted to try to get rid of the root signs by squaring both sides. If so,
just stand back and think a little before you begin. How many times are you
going to have to square things to get rid of all the square root signs?
(2) Suppose that A(a - cvfo) + Ad - evfc) = 1. Square both sides. Then isolate
the ‘large’ root on one side and square both sides again. Finally, equate
coefficients on both sides; that is, equate the constant terms on each side,
and equate the coefficients of A on each side. It looks as though this method
may work. (There are technical problems in ‘equating coefficients’. There is
also the extra complication that each time we square both sides we introduce
‘false’ solutions.)
(3) Is there not some simpler way? Is it perhaps possible that the two long
expressions under the square roots in the original question can actually be
written as perfect squares? If so the outside square roots will just melt away.
(a) Try to see whether we might in fact be able to write

x + 1972 098 - 1986 A x + 986 049) = (_)2.

Can you work out what has to go in the bracket on the RHS?
(b) Can you do the same for the other expression

x + 1974085 - 1988Ax + 986 049) = (___)2?

(4) This means that the LHS of the original equation simplifies considerably.
However, one must be careful about simply taking square roots, since the
expressions would go in the brackets on the RHS of the identities in (3Xa)
and (3)(b) could easily be negative, and V’ indicates the non-negative root.
The only way to avoid a mistake here is to use modulus signs:

A...)2 = l(...)|.

The equation to be solved now takes the form

\y - a\ + \y - (}\ = 1.

What is y? What are a and fB?

 24th BMO, 1987B 137

(5) Solving a modulus equation—even a simple looking one-requires consid¬

erable care. The key is to realize that

‘Iy — a|’ means ‘the distance from y to a\

Show that \y — a| + |y — (a + 1)| = 1 precisely when y lies between a and

a + 1. Use this to solve the original equation.
[Alternatively, you might prefer to change variable by putting x +
986049 = u right at the start.]

2. Find all real-valued functions / which are defined on the set D of

natural numbers x > 10, and which satisfy the functional equation

fix+y)=fix)-fiy),

for all x, y e D.

(1) Which familiar functions / have the property that fix+y) =fix)-fiy)l
Your first guess has to be that the unknown functions / in the question may
be just these familiar functions in disguise: after all, the functional equation
is just a fancy way of writing one of the familiar index laws for powers. But
why does the question restrict to integers x 2* 10? Is this merely a smoke¬
screen? Or could you have missed something? Perhaps things will become
clearer once you get going.
(2) Let / be an arbitrary unknown function with the property that
fix +y) =f(x)-f(y) for all integers *2* 10. What can you say about /(2*)?
About /(3;c)? About /(nx)?
(3) Suppose that /(10) = n. What can you say about /(20)? About /(30)?
About /(110)? About /(x-10)?
(4) What does /(3x) = /(§x + §x) tell you about /(§*)? About /(§x)? About
fi%x)l About /(§*)? Suppose that /(10) = a. What can you say about /(15)?
About /(12)? About /(ll)? About /(25)? Does the value /(10) = a determine
the value of fix) for all natural numbers x 2* 10? Now complete the solution.
(Let fix) = b. Then /(2x) =_ and filOx) =_. But fix-10) =_
from (3). Hence b =_.)

3. Find a pair of integers r, s such that 0 < s < 200 and

45 > / > 59
61 ^ 3 ^ 80 •

Prove that there is exactly one such pair r, s.

(1) At first glance, one feels that it should be easy to find a rational number
r/s between |f and §. Check that §=!=!>!!>§• It follows that the
138 Hints and outline solutions

required number r/s should lie between H and 80 4. Since you waiU
5 < 200, it is natural to try 59|/80 = 119/160; this certainly lies between w
and and has a denominator s < 200. Does it lie between ^ and gj-, or
between ff and fg?
(2) Notwithstanding this initial disappointment, it may be worth trying to find
a rational number r/s of the required type by intelligent trial and error. In a
problem like this it is important to have a go, and to reflect on what you find.
You may be lucky and succeed relatively quickly. Even if the problem turns
out to be harder than you expected, you should still have gained some
valuable insight into why the problem is perhaps harder than it looks. In this
case the difficulty lies in the fact that the upper and lower bounds gj and gg
are surprisingly close. Work out — H as a fraction.
(3) The chances that a fraction with denominator <200 should lie in this
very small gap are pretty slim. However, there is a standard fact which can
often be quite useful. Suppose that a/b <c/d, where b,d> 0. Where does
(a + c)/(b + d) lie relative to a/b and c/dl (Does this depend on the values
of a, b, c, and dl Or can you prove that (a + c)/(b + d) is ‘always < a/b’!
Or is it ‘always > c/d’? Or is it ‘always strictly between a/b and c/d'T)
(4) Step (3) should certainly allow you to find a pair of integers r, s with the
required property. The sting in the tail is that you now have to show that this
pair is the only possible pair with 0 < 5 < 200. This depends on something you
already noticed in step (2); namely, that 45 X 80 — 59 X 61 = 1. Suppose that
a/b <c/d with cb-ad= 1. Then step (3) guarantees that (a + c)/{b + d)
lies between a/b and c/d. Suppose that x/y also lies between a/b and c/d.
Show that the extra condition cb - ad = 1 implies that y>b+d. (x/y < c/d
implies that cy - xd > 1, and x/y > a/b implies -xd < -(ady + d)/b; hence
l<cy-xd<(y- d)/b.)
(5) Now use step (4) three times (for §§ < |f, for §§ < and for < |f) to
complete your solution of the original problem.

4. The triangle ABC has orthocentre H. The feet of the perpendiculars

from H to the internal and external bisectors of angle BAC (which is
not a right angle) are P and Q. Prove that PQ passes through the
midpoint of BC.

(1) While you may know a little about the circumcentre O (where the
perpendicular bisectors of the three sides meet), the incentre I (where the
angle bisectors meet), and the centroid G (where the medians meet),
the orthocentre H—where the perpendiculars AA\ BB', and CC' from
each vertex to the opposite side, the ‘altitudes’, meet—is somehow less
memorable. Still, you must not let that frighten you. Draw a diagram!
(2) The diagram tends to be a little messy and gives little clue as to why PQ
should pass through the midpoint of BC. The most striking thing is all those
 24th BMO, 1987B 139

right angles produced by the altitudes AA', BB', and CC' and the rectangle
HPAQ. The right angles in the figure described in the question seem to be
scattered all over the place, with no obvious connections. You need an idea
to link them. What simple geometrical shape produces lots and lots of related
right angles?
(3) The diagram is full of hidden *i***e*. The circle with diameter PQ
establishes a connection between six different right angles in your diagram.
What are they? Since ABB'C is a right angle, the point B' lies on the
semicircle with diameter BC (and the important midpoint of BC is the
*e***e of this circle). Which other right angle in the diagram lies on this
same semicircle? The supplementary right angle ABB'A at B' lies on the
two other hidden semicircles. What are their diameters?
(4) You have to prove something about the line PQ, so it seems sensible to
begin, as in (3), by looking for a circle which involves the right angles at both
P and Q. (There is only one real choice.) Which six labelled points in your
diagram lie on the circumference of this circle? (Four of these points are
obvious; to find the last two, observe that the circle has diameters PQ and
_.) Where is the centre of this circle? What can you say about AB'AP
and AB'C'P (subtended by the same **o** B'P)1 What can you say about
AC'AP and AC' B'Pl What can you conclude about triangle PB'C' (given
that AP bisects ABAC)1
(5) You have to prove that PQ passes through the centre of the circle with
diameter BC. In which two points does this circle meet the circle with
diameter PQ1 Use the fact that PB' = PC' to prove that PQ is
*e**e**i*u*a* to B'C'. Hence complete the solution.

5. For any two integers m and n with 0 < m < n, numbers d(n, m) are
defined by

d(n,0) = d(n, n) = 1 for all n > 0,

and

md(n, m) = md(n - 1, m) + (2n - m)d(n - 1, m - 1) for 0 < m < n.

Prove that all of the d(n, m) are integers.

(1) There are times in mathematics when one has to take a deep breath and
get started, in the hope that things will become clearer once you get stuck in.
They certainly won’t get any clearer if you avoid getting stuck in! Your only
hope is to read the question carefully in the hope of recognizing something
vaguely familiar in this welter of symbols. (John von Neumann, one of the
greatest mathematicians of the twentieth century, expressed this very well
when a student complained that he didn’t understand something. ‘Young
 Hints and outline solutions
140

man’, said von Neumann, ‘in mathematics you don’t have to understand
things. You just get used to them.’) The first condition, din, 0) — din,n) — 1
for all n > O’, should immediately remind you of binomial coefficients
(| n j = | n j = i for an n > o). The second condition then looks like a compli¬

cated variation on the rule for generating Pascal’s triangle. Like binomial
coefficients, the numbers din, m), n ^ m > 0, can be arranged in a triangular
array. For convenience, imagine this array in the first quadrant with the
number din, m) in m ^ 0) at the point with co-ordinates in, m). Enter the
numbers din, 0) = 1 and din, n) = 1 (all n > 0) in your array.
(2) Triangular arrays of numbers bordered by l’s are not uncommon in
mathematics. As with Pascal’s triangle, there is usually some rule which
allows one to calculate an unknown entry in terms of ‘earlier’ entries in the
triangle. The rule in the question is just like the rule for Pascal’s triangle,
except that we have to add suitable multiples of two entries in the previous
column:

• Calculate d(2,l), di3,1), d(4,l), and di5,l>, enter ihe values in your
array.
• Calculate di3,2), di4,2), and d(5,2); enter the values in your array.
• Calculate di4,3) and d(5,3); enter the values in your array.

(3) The numbers din, m) with m = 1 in > 1) should be thoroughly familiar.

You are told that di 1,1) = 1. Suppose that m = 1 and n > 2; then the first
term on the RHS of the recurrence din, 1) = ... + ... is T •din - 1,1)’, while
the second is just ‘(2n - 1)-1’. It follows that din, 1) is obtained by adding the
first n odd numbers (for which there is a very simple formula). The numbers
din, 2) in ^ 2) are harder to work out, but the answers should still make you
suspicious. Calculating d(4,3) and di5,3) is a little messy; but the answers
should make you even more suspicious.
(4) So far, the numbers din,m) all seem to be **ua*e*. But what are they
squares of? It should not take you too long to guess the answer to this
question: after all, you only know one other number triangle of this kind to
compare it with. This should leave you in the position of being almost certain
what din, m) is in fact equal to. If you could prove that din, m) is what you
think it is then this would certainly show ‘that all the din, m) are integers’.

Use induction (on the sum n + m) to prove that ‘din, m) = |^j whenever
n > m > O’.

(0 If n + m = 0, then n = m = 0 and din, m) = di0,0) = 1, so din, m) =

/ \2
 24th BMO, 1987B 141

(ii) Now suppose that d(n,m) = whenever n + m < K, and try to prove
that when n + m = K the same result holds (use the recurrence in the
question, and the induction hypothesis that d(n - 1, m) = ( n ~ 1 ) and
/ i \2 3 4 \ m )
din - \,m - 1)= x J ).

[An alternative to this ‘direct proof by induction’ is to observe that the

number triangle d(n,m), n >ra 3* 0, certainly agrees with j on the two

sloping edges of l’s. So if you simply check that 8(n, m) = | ^ j satisfies the
same recurrence

m8(n, m) = m8(n - 1, ra) + (2 n — m)8(n - 1, m — 1) for 0 < m < n,

then the two number triangles d(n, m) and 8{n, m) must be identical. All of
this should leave you with the question of whether there is some way of
proving ‘that all the d(n,m) are integers’ without calculating the exact value
of every d(n, m) first.]

6. Show that, if x and y are real numbers such that lx2 + 3xy + 3y2 = 1,
then the least positive value of (x2 +y2)/y is

(1) Don’t panic! The last question on each paper is meant to sort out the
sheep from the goats. But that doesn’t mean that it is necessarily terribly
hard. There may well be fancy ways of tackling questions of this kind, but if
you stick to simple-minded methods then there is really only one way to
begin. Suppose that you wanted to show that, for all values of t, t2 — At ^ —4.
Any inequality like this can be rearranged into the form ‘new LHS > O’. And
the new LHS is then a perfect square. What is it the square of?
(2) In easy exercises (such as proving that t2 — At + 4 ^ 0 for all t) the LHS
factorizes very simply as a single perfect square. Life is not always so kind.
However, if you can prove that ‘t2 - At ^ -4 for all t\ then you can certainly
prove that ‘t2 - At ^ -5 for all t\ The second inequality may at first seem
harder, but to show that t2 - At + 5 ^ 0 for all t, it is enough to write the LHS
as the sum of two squares: namely (t - 2)2 + l2.
(3) Now use this simple idea to tackle the original problem. To find the least
positive value of (x2 +y2)/y you only have to worry about certain values of
y. Which ones? Rearrange the required inequality ‘(x2 +y2)/y ^ j into the
form ‘new LHS > O’.
(4) Somehow you have to use the supplementary relation lx2 + 3xy + 3y2 = 1
to simplify the new LHS. (That is, you have to prove that for points above
the x-axis on the curve with equation lx2 + 3xy + 3y2 = 1, the value of
(x2 +y2)/y - | is never negative.)
 Hints and outline solutions
142

Unfortunately, the new LHS has an awkward y in the denominator; and

there is no easy way to use an expression such as lx2 + 3,xy + 3y of degree 2
to simplify a single y of degree 1. The trouble is that the expression (x +y )/y
has a numerator of degree 2 and a denominator of degree 1, so the whole
expression (x2 + y2)/y =x2/y + y behaves in some ways as if it had degree
2-1 = 1. This makes it difficult to write (x2 +y2)/y in terms of an expres¬
sion such as lx2 + 3xy + 3y2 of degree 2.
The simplest potential escape from this dilemma is to square our original
inequality: for y > 0, (x2 +y2)/y > \ is equivalent to [(x2 +yz)/y]2 > This
approach may not work, but it is definitely the first thing to try. Rearrange
the inequality ‘[(x2 +y2)/y]2 > j’ into the form new LHS ^ 0 , getting rid of
denominators. You must now use the supplementary relation 1 — 7x + 3xy +
3y2 to simplify the LHS, with a view to proving that the LHS is >0.
(5) There are many possible ways of using the supplementary relation 1 =
lx2 + 3xy + 3y2 to rewrite the numerator. Most of these produce an expres¬
sion which does not simplify in any obvious way. One natural substitution is
to replace the coefficient T’ of y2 in the expression ‘4(x2 + y2)2 — 1-y2 by
‘lx2 + 3xy + 3y2’. This may look as though it could only make things worse;
but try it and see what happens. Then (remembering the point made in (2)
above) try to finish off the solution.
(6) To complete the solution successfully you have to notice slightly more
than was explicitly stated in (2). In proving 4x4 + x2y2 + y4 — 3xy3 > 0, you
would like to write the LHS as a sum of squares. However, since there exist
values x = x0 and y=y0 satisfying lx2 + 3xy + 3y2 = 1 which make the
LHS = 0, the squares have to be chosen carefully: there is no slack. Find x0
and y0.
It follows that each of the squares that you use to rewrite the LHS must
vanish when x=x0 and y=y0. In particular, since y0 = 2x0, this suggests
that each square should involve (2x -y)2. Expand (4x2 -y2)2 + 3y2(2x -y)2.
How does this solve the problem?
[Alternatively, let x/y = k. Then y2 = l/(lk2 + 3k + 3) and the inequality
to be proved is just 4A:4 + k2 - 3k + 1 > 0. The solution can then be com¬
pleted either by factorizing and completing the square, or by using calculus.]

23rd British Mathematical Olympiad, 1987A

1. (a) Find, with proof, all integer solutions of

a3 + fr3 = 9.

(b) Find, with proof, all integer solutions of

35x3 -I- 66x2y + 42xy2 + 9y3 = 9.

 23rd BMO, 1987A 143

(1) Part (a) looks reasonably easy; you can probably see two obvious solu¬
tions, a = 2, b = 1 and a = 1, b = 2. Moreover, it doesn’t look as if there
could be any others although you will have to prove this later. How about
part (b)? That 9 on the RHS looks suspiciously like the 9 in part (a). Could
the LHS in part (b) possibly be simpler than it looks? Could it even be exactly
the same as the LHS in part (a), but heavily disguised: that is, of the form
a1 2 3 + b3 = 9, with

(*) a = ax + (3y, b = yx+8y?

If so, you would have to have a3 + y3 = 35, 03 + 83 =_. There may be

many solutions (since a, (3, y, and 8 could be negative). However, it is
tempting to try a = 2, y =_ (or a = 3, y =__) and 0 = 1, 8 =_ (or
0 = 2, 8 =_). One of these works. Which one is it? So once you know all
integer solutions a, b to part (a), you can substitute these values of a, b in (*)
and find all integer solutions x, y to part (b).
(2) So perhaps the ‘easy-looking’ part (a) is, in some sense, the worst bit!
Suppose that a3 + b3 = 9, with a and b both positive. Show that 0 < a3 and
b3 < 9, and that there are just two solutions.
(3) But what if b (say) is negative (b = —0) and a (say) is positive. Then

a3 — 03 = 9 witha,0>O.

Can two positive cubes a3 and 03 differ by 9? Explain. Hence complete the
solution to part (a).

2. In a triangle ABC, ABAC = 100° and AB =AC. A point D is chosen

on the side ^1C so that LABD = ACBD. Prove that AD + DB = BC.

(1) Pictures are much easier to understand than all those words and letters!
So draw a diagram and mark the information given on it.
(2) The most striking thing is that you seem to know so many angles. Indeed,
there are so many angles that one is inclined to overlook crucial facts, such as
AB=AC and AABD = ACBD. Still, when one knows lots of angles, one
obvious strategy is to use trigonometry to find AD, DB, and BC, and then
use trig identities (cos(90 — a) = sin a, sin 2a = 2 sin a cos a, cos a —
cos 0 = ...) to try to prove that AD + DB = BC. Trigonometry tends to be a
little easier if you can work with right-angled triangles. Is there a natural way
of introducing right-angled triangles, given that BD bisects the angle Z. ABC2
(3) An angle bisector is equidistant from the two legs of the angle that it
bisects. Hence D is equidistant from BC and from BA (extended). Drop
perpendiculars DE from D to BC, and DF from D to BA extended. Use the
fact that DE = DF to find x and y such that
AD =x-BD, BC =y-BD.

Then use trig identities to show that AD + DB = BC.

144 Hints and outline solutions

(4) You should have managed to make this first approach work, but it is a
little messy. What is worse, it does not really explain why AD + DB is equal
to BC. When the relationship to be proved is as simple as AD + DB = BC,
one feels that there should be a simple geometrical proof, although one can
never be sure! (The trouble with simple proofs is that they are only simple
after you have found them, and it can take an awful lot of searching before
you hit on the right idea—the key that unlocks the whole puzzle.) The
important things to notice here are that:

• the segments AD and DB are not in a line, so they are difficult to add
together in a simple-minded way;
• neither of these segments lies along BC (so it looks as though you will
definitely have to move something);
• the two 20° angles at B are equal.
Rotate triangle BA(D)C through 20° to triangle BA'(D')C' with A' lying on
BD. Where does D' lie? What do you know about BD'2 So what must you
prove about D'Cl
(5) By now your diagram may well be rather cluttered, but you should stiii be
able to work out almost any angle you want. Join AA'. What can you say
about the three points A, A', and £)'? What can you say about triangle
ADD'l What can you say about triangle DD'Cl How does this help you solve
the original problem?

3. Find, with proof, the value of the limit as n —> °° of the quotient

(1) This looks horrid! There’s no way in which you can think about an
expression like this until you can begin to see what it really means. Where on
earth could it have come from? Forget about the exact form of the expression

for the moment. Suppose that the numerator were E ln\-2r. Would that
r=0'V
be better? Of course it would: the binomial theorem says that

(!+*)"= EteW;
r=0 V ’

••• E (”)-2r= (1 + 2)n -3".

r= 0 ' '
2n . .
(2) This tells you the value of £ 2n ) -2*. However, the expression in the
j=o ' '
original problem has two awkward features:
 23rd BMO, 1987A 145

® Only the even binomial coefficients | ^) occur in the numerator, with

the odd ones occurring in the denominator.

® The binomial coefficient ^ occurs with 2r, not with 22r as you might
expect. ' ’

The second of these is perhaps the easiest to understand. You expect |j

to occur as the coefficient of x2r. If 2r=x2r, what must the value of x be?
t3) This suggests that the whole problem has something to do with the
binomial expansion of something like (1 + f22)2n. But you still have to explain

why there are no V2s around, and why the odd terms L 2n „ 2r are in the
\2r+l)
denominator. (These two puzzles are clearly related! After all, you would not

expect to find any 72s with the even terms Suppose that

| 52 | 2” j 52 | 2^+ ) j
1 2r = t. The value of t actually depends on n; but

you are interested in the value as n —>°°. Suppose that you cross-muItiply and
shift all the terms to the LHS. What value of t would allow you to simplify
the LHS nicely using the Binomial Theorem? And what does (1 — f2)2n tend
to as n -»00? Now use this idea to solve the original problem.

4. Let P(x) be any polynomial with integer coefficients such that

P(21) = 17, P(32) = -247, P(37) = 33.

Prove that if P(N) = N + 51, for some integer N, then N = 26.

(1) This whole question looks very strange. What do we know about polyno¬
mials? Not much—except for the *e*ai**e* theorem. Well, can we use the
*e*ai**e* theorem? In its simplest form the theorem states that if Q(x) is a
polynomial and a is a real number, then
Q(a) = 0 <=> Q(x) = (x - a)-R(x).
Most introductory texts leave it at that, but the usual proof actually proves a
little more than this.
‘If Q(x) = anxn +an_lXn~l + ... +alX + a0,

then Q(a) = anan + an_1an~1 + ... +axa + a0.

.-. <2(x) - Q(a) =an(xn - an) + a„_1(xn_1 -an~1) + ... +ax(x-a).
Now x' — a‘ = (x — a)(x'_1 +x‘-2a + ... +a'~1) for each i.
Q(x) - Q(a) = (x-a)X (some polynomial R(x)),
where the coefficients of R(x) are integer combinations of the
coefficients of Q(x) and powers of a. Hence if a is an integer and
Q(x) has integer coefficients, then R(x) has integer coefficients too!’
146 Hints and outline solutions

(2) It may not be clear at first how to use the Remainder Theorem with the
polynomial P(x) itself. The Remainder Theorem applies to poly¬
nomials Q(x) and numbers a for which Q(a) = 0, whereas Question 4
requires us to work with the condition ‘P(N) = N + 51 for some N . There is
an easy way out of this. You have to define a new polynomial Q(x) (closely
related to P(x) so that P(N) = N+ 51 <=> Q(N) = 0. Now work with this
new polynomial Q(x). Find <2(21) =-, Q(32) =-, and Q(37) —

(3) The fact that Q(21) = Q(37) * 0 suggests that we may have chosen to
work with the wrong polynomial Q(x). If we let Q*(x) = Q(x) + 55, then
<2*(21) = <2*(37) = 0; so by the Remainder Theorem x - 21 and x - 37 are
both factors of £>*(x) and

<2*(x) = (x — 21)(x — 37) -i?(x),

where R{x) still has integer coefficients. Moreover,

P(N) = iV + 51 ~ <2(AO = 0 ~ £*(AO =-

Can you see how this helps you to solve the original problem? (When you
have succeeded, you might like to reflect on what role the information
‘P(32) = -247’ has played in your solution.)

5. A line parallel to the side BC of an acute angled triangle ABC cuts the
side AB at F and the side AC at E. Prove that the circles on BE and
CF as diameters intersect on the altitude of the triangle drawn from A
perpendicular to BC.

(1) Draw a diagram and mark all the information given.

(2) In mathematics one should always be on the lookout for tell-tale clues,
which look as though they are going to be important in the eventual solution
of a problem. In this particular problem the only obvious hint of this kind is
the fact that altitudes and semicircles both involve *i*** a***e*. Of course
this may turn out to be irrelevant, but you should bear it in mind when
looking for a way into the problem. (School mathematics tends to overempha¬
size the idea that every problem includes a more or less obvious clue as to
how to solve it. Real mathematical problems are much more interesting than
this, and the important ideas generally emerge gradually as you begin to get
to grips with the problem.) Suppose that the two circles on BE and CF as
diameters intersect at P and Q. Then P is one of the two points for which
ABPE — A CPF = 90°. We have to show that these conditions force P to lie
on the altitude from A to BC. In other words, we have to show that AP is
perpendicular to BC. There may be a clever way of doing this, but it looks as
if we have not yet made enough use of the other information in the question
 23rd BMO, 1987A 147

(for example, FE parallel to BC). In geometry questions, once one under¬

stands roughly what the diagram should look like, but still cannot quite see
how to proceed, it is often a good idea to draw a more accurate diagram.
Draw an accurate diagram with BC and FE horizontal and the altitude from
A to BC vertical.
(3) Solving a mathematical problem is a process of pulling yourself up by
your own bootstraps—one small step at a time. You should not expect to see
straight through a good problem in one move. Instead, you should be
permanently on the lookout for small steps that seem to lead in the right
direction. Whatever your accurate diagram may suggest, it isn’t at all obvious
why P and Q should lie exactly on the altitude from A to BC. However, if
you stare at your diagram (and think!) you should be able to see something
which is definitely interesting. The common chord PQ of the two circles has
to be ‘vertical’, that is, parallel to the altitude from A to BC. Can you see
why? (The common chord of two intersecting circles is always perpendicular
to the line joining their centres. So all you have to prove is that the line
joining the midpoints of BE and CF has to be parallel to BC.)
(4) This certainly feels like progress. But the final step is still elusive. You
now know that the altitude through A and the common chord PQ are
parallel. It remains to be proved that they are not only parallel, but identical!
In other words, you have to show that the point A has to lie on the common
chord PQ. In solving any mathematical problem there comes a point at which
one simply has to know something, or work it out from scratch. You want to
know that A lies on the common chord PQ of our two circles. It is therefore
hard to see how you can avoid the question

‘What condition guarantees that a point A lies on the

common chord of two intersecting circles? ’

If you don’t already know the answer, the next hint should help you work it
out for yourself.
(5) Suppose that the point A lies outside a given circle, and that the line
AXY cuts the circle in X and Y, while AT is tangent to the circle at T. Show
that AATX=/-TYX. Conclude that triangles ATX and AYT are similar.
Hence show that AX-AY = AT2. Deduce that a point A outside two inter¬
secting circles lies on their common chord precisely when the lengths of the
tangents from A to the two circles are equal. Try to see how this might help
to solve the original problem. (If you succeed, then fine. If not, read on.)
(6) Your diagram does not contain a ‘tangent AT’ from A to the circle with
diameter BE. Moreover, drawing one in would seem to make things worse.
Thus the condition for A to lie on a common chord has to be used
intelligently. Perhaps the easiest way is to forget about ‘tangents’, and to use
the full force of the result in (5). The most obvious line from A which cuts
the circle with diameter BE is AB; if this cuts the circle again at X, then
148 Hints and outline solutions

AX-AB = AT2 (even though we do not know where T is). Similarly, the most
obvious line from A which cuts the circle with diameter CF is AC; if this
cuts the circle again at Y, then you need only show that

AX-AB =AY-AC.

(For this, first observe that triangles ABC and AFE are *i*i*a* (Why?), so
AB/AC =AF/AE. Then combine this with the fact that AF/AY = AE/AX
—from the similar right-angled triangles AFY and AEX.)

6. If x, y, and z are positive real numbers, find, with proof, the maximum
value of the expression
xyz
(1 +x)(x +y)(y +z)(z + 16)

(1) Each of the three factors x, y, and z in the numerator occurs in two
separate factors in the denominator. So if you try to make the whole
expression larger by increasing t^e numerator, then as x, for example, tends
to infinity, the whole expression actually tends to zero! If, on the other hand,
you try to make the whole expression larger by making the denominator
small, then it is not enough to make just one of the variables small since, for
example, x and y occur together in the same bracket; you have to make both
x and y (or both y and z) small at the same time. But then the numerator
contains the product x-y (or y-z), so the whole expression actually becomes
very small! Thus it looks as though the values of x, y, and z which give the
expression its maximum value will be neither very large nor very small. So try
some values which are neither very large nor very small. When x = y = z = 1,
the expression has the value l/(2-2-2* 17). What values does the expression
have when x=y=z = 2? Is this larger or smaller than 1/(2-2-2-17)? What
if x = 4, y = 8, and z = 16? Can you produce an even larger value?
(2) The previous step may give you a feeling for the approximate size of the
maximum value (rather small!). But it proves nothing. If you have never seen
a problem like this before, it may not be at all clear what to do next. So you
really have no alternative but to try something simple-minded and see what
happens.
One awkward feature of the given expression is that it doesn’t appear to
simplify very easily. But suppose that you turn the whole expression upside
down! You will then have to look, not for the maximum value of the original
expression, but for the *i*i*u* value of

(1 + *)(* +y)(y +z)(z + 16)

xyz

At this point you must suppress any misguided urge to multiply out and
 23rd BMO, 1987A 149

collect up terms. Factorized form is generally far more useful, and multiplying
out is often the worst possible thing to do with such an expression. However,
you could get rid of the denominator by incorporating the ‘x’ in the second
bracket, the y in the third bracket, and the ‘z’ in the last bracket. Then all
four brackets have the form (1 + something). You must therefore find the
minimum value of

(1 + *)(1 +y/x)( 1 +z/y)( 1 + 16/z).

If you now put x = a, y/x = (3, z/y = y, and 16/z = 8, then the expression
takes the more symmetrical form

(l + a)(l + /3)(l + y)(l + 5).

However, a, (3, y, and 8 are not ‘independent’ variables, since cr(3-y-8 =

(3) You may still not know how to solve the problem, but at least it is now in
a sufficiently nice form that you can look at some similar (but easier) cases to
try to get some ideas. For example, you could start by trying to find the
minimum value of (1 + a)(l + (3) given that a- f3 = 4.
Your answer to this easier problem should suggest a likely answer to the
original problem (namely ^-). But your method of solving the easy problem
will probably not work for the original problem. So take a closer look at the
easy problem. First observe that (l + aXl + /3)=l+(a + /3) + a/3. So if the
value of a-/3 is given, then you are really trying to minimize the value of
a + (3. This is precisely the famous (and possibly familiar) problem of finding
which rectangle with a prescribed area (a-jS) has the smallest possible
*e*i*e*e* (2(a + f3)). The minimum occurs when a = [3; that is, for a
square. Can you use the same approach to find the minimum value of
(1 + a)(l + /3Xl + y) given that a- (3 - y = 8?
(4) Another way of getting a feeling for the problem is to think in terms of
graphs. Let fix) =x2 + bx + c be a quadratic with negative roots x = -a and
x = —j8. Suppose that the constant term c = a{3 is fixed but that a and f3
can vary. Sketch the graph of y =f{x). As a and (3 vary (but c = a(3 remains
fixed), when does the value /(l) take its minimum value? (That is, you want
to minimize 1 + (a + c/a) + c as a varies.) What has this to do with the first
easy case in (3) above?
Let g(x)=x3 + px2 + qX + r be a cubic with three negative roots x = -a,
x= -(3, and x = — y. Suppose that the constant term r = a-(3- y is fixed but
that a, (3, and y can vary. Sketch the graph of y = g(x). As a, (3, and y vary,
when does the value g(l) take its minimum value? What has this to do with
the second easy case in (3) above?
This graph idea may suggest a possible approach using calculus. (However,
this approach is messy and slightly subtle.)
(5) For those who prefer simple algebra, the missing idea here, as in many
150 Hints and outline solutions

similar problems, is the ‘AM-GM inequality’. (See ‘A little useful mathemat¬

ics’ (Section 2.6) for a discussion of this result.) Use the AM-GM inequality
for two variables to prove that if c is a constant and and x2 satisfy
xx ■x2 = c2, then

(1 +;c1)(l +x2) > (1 + c)2,

with equality only if xx=x2 = c.

Use the AM-GM inequality for three variables to prove that if c is a
constant and xu x2, and x3 satisfy x1x2x3 = c3, then

(1 -I-jcj)(1 +*2)(1 + x3) > (1 + c)3,

with equality only if x3 =x2 =x3 = c.

Finally, use the AM-GM inequality for four variables to solve the original
problem.

7. Let n and k be arbitrary positive integers. Prove that there exists a

positive integer x such that \x(x + 1) — k is divisible by 2”.

(1) You may recognize ^x(x + 1) as the xth triangular number Tx, where

rx = 1+ 2 + 3 +_+x = jx(x + 1).

It is not clear whether this interpretation will help. But it seems that you have
to show that

‘for any given 2", there is a triangular number Tx = \x(x + 1) to go with

each k, so that jx(x + l) — k is exactly divisible by 2"’.

Is there another way of expressing this mouthful? What does it mean to say
that l\x(x + \)-k is exactly divisible by 2"’? What does that ‘minus k’
mean?

T- k is exactly divisible by t' means

‘T- k = tx something (with no remainder)’,
that is, T = t X something + k ’.

If you think carefully about this last version, you will realize that k is the
remainder when we divide T by t. So your problem boils down to showing
that

‘given any 2",

when you divide successive triangular numbers Tx = \x(x + 1) by 2”,

you get all possible remainders k ’.

 23rd BMO, 1987A 151

Explain why you only have to worry about values of k < 2". (Note that when
k = 2" the ‘remainder’ would usually be taken to be 0. If we stick to positive
k, as suggested in the question, then we cannot allow k — 0 as a remainder,
but must take k = 2n in this case.)
(2) It is not clear where to go from here, but you are now well-placed to do a
little experimenting, keeping an eye open for anything unexpected. Complete
the following tables. Do you notice anything remarkable?

Let n = 1 k 1 2

first Tx with Tx - k divisible by 21 Tx = 1 r3 = 6

Let n = 2 k 1 2 3 4

first Tx with Tx — k divisible by 22 T, = l Tj = 6 r2 = 3 T7 = 28

Let n = 3 k 1 2 3 4 5 6 7 8
bT

f-r

o
r—1

t-H

first Tx with Tx — k divisible by 23

II

II
£

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

first Tx with Tx — k divisible by 24

(3) The optimist might have hoped to obtain all possible remainders
1,2,...,2" using only the first 2" triangular numbers. This idea breaks down
right at the start:

When n = 1 T1 = 1 and T2 = 3 both correspond to remainder k = 1. We

need to go on to T3 = 6 to get remainder k = 2 (or, if you
prefer, k = 0).

When n = 2 Tx = 1, T2 = 3, T3 = 6 all leave different remainders when

divided by 22. But T4 = 10 leaves the same remainder as T3.
We need to go on to T7 = 28 to get remainder k = 4 (or, if you
prefer, k = 0).

When n = 3 we have to go all the way up to T15 = 120 to get remainder

k = 8 (or, if you prefer, k = 0).

There really are many things that you should have noticed about the tables in
(2). What is the largest value of x (or the largest triangular number Tx) that
you need to use when n = 4? Which remainder does it correspond to? What
would you expect to be the largest value of x (or triangular number Tx) you
need to use when n = 5? Which remainder k do you expect will need this
largest triangular number? Check your guesses by completing a similar table
for n = 5.
152 Hints and outline solutions

(4) Once you notice what seems to be a general pattern you must formulate
precisely what you believe to be true, and then try to prove that things really
do work in the way you think they do. You know that you certainly cannot
obtain all the remainders k = 1,2,...,2" by looking at the first 2n triangular
numbers only. But it does look as though you can obtain every remainder by
using at most **i*e as many triangular numbers as this absolute minimum. In
fact, it looks as though the first 2n+1 - 1 triangular numbers will always be
enough. If you looked more carefully at the ‘repeats’ in your tables, you may
have noticed that each remainder k occurs exactly twice—and from two very
closely related triangular numbers! Suppose that \x(x + 1) and \y(y + 1)
leave the same remainder k when divided by 2". What does this tell you
about x and y? (Factorize x(x + 1)-y(y + 1).) How does this solve the
original problem?

22nd British Mathematical Olympiad, 1986

1. Reduce the fraction N/D to its lowest terms when

N =2 244 851485 148 514 627, D = 8118 811881188 118 000.

(1) You may be tempted to dismiss this problem as ‘mere arithmetic’. It is

certainly true that, although the problem is harder than it looks, elementary
methods should succeed. However, even the most simple-minded approaches
need to exploit one or two simple mathematical ideas (such as noticing that
both N and D are divisible by *i*e, and that any number which consists of
several repetitions of the same string of digits has an obvious factor).
Whether you succeed in solving the problem this way, or get stuck, you
should not be satisfied until you discover a more mathematical approach.
(2) So even if you think you know the answer, go back to the beginning and
think things through again, step by step. Of the two numbers N and D, one
looks far more promising as a starting point. Which one is it?
(3) The first thing to notice about D is that it has a factor A = 1000. In
contrast, the numerator is divisible neither by 2, nor by 5. Hence no part of A
can ever cancel. Thus, even in lowest terms, the denominator will still be a
multiple of 1000.
If you factorize D=AxD', then D' has an obvious four-digit factor
B =_. And when you write

D=AXBXD",

then D" consists entirely of zeros and ones. Moreover, just as the repeated
sequence of digits ‘8118’ in D means that D has a factor ‘8118’, so the
repeated sequence TO 001’ in D" means that D" factorizes as 10001 X_.
 22nd BMO, 1986 153

Most interesting of all is the fact that, in this factorization of D", the first
factor 10001 = 101 2 * 4 + 1, while the second factor can be written as_.
(4) There is not much point in trying to break down these factors of D
further at this stage for it is not yet clear which bits of the denominator D
have a chance of cancelling with the numerator N.
So turn your attention to N. The number N has something of the same
structure as D, with three repeated blocks in the middle. (The repeating
block 8514 stands out, but there is in fact another four-digit block that
repeats three times. What is it?) You would like to exploit these repeating
blocks. But how? There are two linked observations which might show how:

(a) First, if the numerator and denominator of the fraction N/D in lowest
terms are going to be reasonably small, then it is essential that most, or
perhaps all, of the factors 108 + 1 and 104 + 1 in D should cancel with N.
Such factors arise (as you have seen) when a number has a block of digits
which repeats after the appropriate number of steps (either eight or four).
As it stands, N does not quite have this structure.

(b) However, although the numerator N appears to have a slightly different

structure from that of D, that is only because the first three digits ‘224’
and the last three digits ‘627’ mess things up. If you increase the number
formed by the three digits on the left-hand end, and decrease the number
formed by the three digits on the right-hand end by the same amount k,
then the number N is changed by k X_.

Notice that the second factor here is divisible by 108 + 1! Moreover, there is
one particular choice of k, namely k =_, that produces a number

N' = N + kX (1016 - 1)

which has exactly the same structure as D (so that N' can then be factorized
in exactly the same way as D was in (3) above). What is this magic value of
k! Use this to complete the solution very elegantly.

2. A circle S of radius R has two parallel tangents, tx and t2. A circle 5,

of radius rx touches S and tx; a circle S2 of radius r2 touches S and t2;
Sj also touches S2. All of the contacts are external. Calculate R in
terms of rx and r2.

(1) To get started you must draw a good diagram, showing all the information
given, and marking in any points or lines which are likely to be important
(even though they may not be explicitly mentioned in the problem).
(2) The fact that you are asked to ‘calculate R in terms of rx and r2 may
seem slightly strange (since R is given first, the circle of radius rx is
constructed next, and R and rx together determine r2—\so you might expect
154 Hints and outline solutions

to calculate r2 in terms of R and rx). Nevertheless, you should certainly

expect there to be some kind of formula relating the three quantities R, rx,
and r2; and such a formula can presumably then be written in the form
‘R =_’•
(3) The question is about three *i***e*. To specify a *i***e it is no good
just marking the circumference: so you should have naturally marked the
*£***5^ and also any interesting *a*ii for each circle. Moreover, if the
*e***e* of the circles are important, you should give them names: let C be
the centre of the circle S, and let C, be the centre of circle St (i = 1,2). Since
the question is about circles which are tangent to each other, you should
naturally have drawn in the lines joining the centres of touching circles. And
since the circles are tangent to two lines, you should mark the radii from the
centres to the four points of tangency (and label the points so that you can
refer to them: let P, be the point at which S meets t, (i = 1,2)). No other
information is given, so it remains to interpret the diagram in some way
which captures the fact that the three circles are tangent to each other and to
the two parallel lines.
(4) The fact that the two lines (f, and '2) are *e**e**i*u*a to the radii you
have just marked should suggest that ****a*o*a*’ theorem may come in
handy. Calculate the distance dx from Cx to the line PXP2, and the distance
d2 from C2 to the line PXP2. Next, calculate the distance from Cx to the line
through C2 parallel to t2, and hence the distance d from C2 to the line
through Cx parallel to PXP2. Finally, express the obvious relation between dx,
d2, and d as an equation in R, rx, and r2.
(5) The relation between R, rx, and r2 is so unexpected and so beautiful that
is is natural to ask whether there is not some simpler reason why the three
quantities are related in this way. There are certainly many other
solutions—for example, using angles and the cosine rule. You might also
expect such a simple formula to have other interesting consequences. For
example, consider the following. Let X) be the point at which St touches tt,
and let Y,C, meet S, again at Yj (i = 1,2). What can you prove about the
three points C, Yx, and Y2? And what interesting point lies on the line XxX2l

3. Prove that if m, n, and r are positive integers and

1 + m + n'J 3 = (2 + V,3)2r~l,

then m is a perfect square.

(1) At first sight, there is no obvious reason whatsoever why the number m
should always be a square, so it is natural to begin by calculating the first few
odd powers: (2 + V3)1 =_, so m = (_)2 is a square; (2 + V3)3 =
_, so m = (_)2 is a square; (2 + V3)5 = (2 + V'3)3(2 + \/3)2 =_,
so m = (_)2 is a square; (2 + V3)7 = (2 + v/3)5(2 + v^)2 =_, so
 22nd BMO, 1986 155

m ~ (-)2 is a square. Such calculations may convince you that there is

indeed something to prove; but the numbers quickly become uncomfortably
large, and do not immediately suggest any obvious pattern. It may even be
worth pointing out that the actual numbers which drop out of such calcula¬
tions can often prevent you from seeing how to solve the problem: for
example, the sequence 1,5,19,71,... is so unexpected that it is tempting to
think that the fact that m is always a square may be purely accidental.
Instead of calculating more numbers, what you need is some insight into the
way in which the value of m is determined in general.
(While you are still feeling your way into the problem, it is natural to ask
why the question is restricted to odd powers (2 + V,3)2r~1. In such cases it is
always worth checking a few examples to see whether even powers really do
behave differently: (2 + V3)0 =1, so m = 0, which is a square; (2 +V3)2 =
-, so m =_, which is not a square; (2 + \/3)4 = (7 + 4\/3)2 =
_, so m =_, which is not a square.)
(2) It is natural to hope that calculating the odd powers with r= 1, r = 2,
r = 3, and r = 4 will not only show that m is a square, but provide a clue as to
what m is the square of when r = 5, r = 6, and so on. Sadly, that is rather
unlikely in this case! So you have no choice but to try to think about the
general case. Choose a general value of r and let

(2 + ^3)2r_1 = a +hv^3.

You want to prove that a — 1 is always a square. Use the *i*o*ia* theorem
to write down the sum of all the terms in the expansion of (2 + v/3)2r_1 which
do not include f3, and which therefore have sum equal to a. All of these
terms except one involve an e*e* power of ^3; and so are multiples of 3; the
one remaining term is a power of 2, so is not a multiple of 3. Hence a is
never a multiple of 3. It follows that, for each choice of r, either a + 1 or
a — 1 is always a multiple of 3. Is it always a + 1 that is a multiple of 3? Or is
it always a — 1? Or is it sometimes one and sometimes the other? Prove your
claim.
(3) To solve any good Olympiad problem you need an idea. ^3 is defined to
be the positive square root of 3; that is, ^3 is the positive root of the equation
x2 = 3. In multiplying out the expression (2 + f3)2r~\ you repeatedly used the
fact that (/3)2 = 3 to simplify and collect up terms to obtain the two
coefficients a and b. The equation x2 = 3 has another root, namely _,
which behaves exactly like f3 (in that (_)2 = 3). This means that the two
expressions (2 + v/3)2r_1 and (2 - V'3)2r~1 are so similar that, if you let

(2 — f3)2r~X =a' +b'f3,

you should be able to say straight away (without doing any messy calculations)
how a and a' are related, and how b and b' are related. Use this, together
156 Hints and outline solutions

with the observation that (2 + >/3) • (2 ^3) —-, to prove that

a1 2 3 — 3 b2 = 1.
(4) Given what you found at the end of (2), it should be almost impossible to
resist the temptation to take the 1 to the other side and to rewrite this
equation in the form (_X_) = 3b~. From (2) you know that a + 1 is
always a multiple of 3; it follows that a — 1 and (a + l)/3 are integers, and
that their product is a perfect square. All you need to finish off the question
is to prove that HCP(a - 1, (a + l)/3) =_ (see ‘A little useful mathe¬
matics’, Section (4.3.3)).
[Alternatively, you might spot a recurrence relation for the sequence
A1 = 1, A2 = 5, A3 = 19, A4 = 71,..., which appears to determine how each
term can be calculated from the previous two terms. You then have two
separate sequences:

• the original sequence at, a2, a3,..., where ar is the term independent of
V3 in the expansion of (2 + \/3)2r_1, and
• the sequence Ar, defined by a simple recurrence relation.

You may then be able to (i) prove a recurrence for the sequence (ar), and (ii)
to prove (by induction) that aT — 1 = A2r for every r > 1.]

4. Find, with proof, the largest real number K (independent of a, b, and

c) such that the inequality

a2 + b2 + c2 > K(a + b + c)2

holds for the lengths a, b, and c of the sides of any obtuse-angled

triangle.

(1) It is not easy to grasp what exactly is being asked. If you only had to find
a value of K such that a2 + b2 + c2 > K(a + b + c)2 always holds, then you
could choose K — 0 (or any negative value of K). Thus, it is important to find
the best possible value of K—which in this case means the largest possible K.
But that is not all, for the numbers a, b, and c are not just any old numbers!
You have to find the largest possible value of K such that the inequality holds
whenever the numbers a, b, and c are the sides of an obtuse-angled triangle.
(2) Explain why K must be < 1.
(3) For the proof of (2) you don’t need to worry about obtuse-angled
triangles; you only need the fact that a, b, and c are *o*i*i*e. To understand
the problem as given, it may help to start by finding what the best possible
value of K would be if a, b, and c were not restricted to being the sides of an
obtuse-angled triangle, but could be any three positive real numbers. The
answer to this question exploits the standard facts that (a - b)2, (b - c)2, and
(c - a)2 are all > 0; hence their sum (a - b)2 + (b - c)2 + (c - a)2 ^ 0 (with
 22nd BMO, 1986 157

equality holding precisely when a = b = c). Rearranging this gives a2 + b2 +

c > (a + b + c)2/3, with equality holding precisely when a = b =c. This has
two interesting consequences.
(3A) The first consequence may be slightly unexpected. You have proved
that, if a, b, and c are positive, then a2 + b2 + c2 > (a + b + c)2/3. It follows
that a2 + b2 + c2 > K(a + b + c)2 whenever K < |, but not when K=\
(since the inequality' with K = } becomes an equality whenever a = b — c).
Hence there is no latest value of K such that ‘a2 + b2 + c2 > K(a + b + c)2’
holds for all triples a, b, and c of positive real numbers!
(3B) The second consequence is that, if a, b, and c are the lengths of the
sides of an obtuse-angled triangle, then you know that a = b = c is impossible
(Why?), so the strict inequality a2 + b2 + c2 > + b + c)2 certainly holds
for all such triples a, b, and c. Hence the best possible value of K is > j.
(4) You now know that the required value K satisfies 1. Suppose
that you choose your favourite obtuse-angled triangle, and calculate the ratio
(a2 +b2 + c2)

(a + b + c)2
Tou have to find the largest value of K for which this ratio is >K for
all obtuse-angled triangles. Hence the ratio must be >K for each individual
obtuse-angled triangle. So the ratio that you obtain for any particular
obtuse-angled triangle provides an upper bound for the required value of
K. For example, if you take the triangle with sides of length 1, 1, and V3
(having angles 30°, 30°, and 120°), then the ratio is equal to 5/(2 + V3)2 =
5(2 — \/3)2 = 35 — 20^3. Prove (without using a calculator) that 35 — 20^3 >
Conclude that 35 — 20V3
(5) It should now be clear what the problem is asking, even if you still do not
know how to solve it. However, you have already seen the possible advantage
of rearranging the given inequality with all the a’s, b’s, and c’s on one side,
with the elusive ‘K’ on the other. You have to find the largest value of K
such that
(a2 + b2+c2)
(*) -y- >K
(a + b + c)
holds whenever a, b, and c are the sides of an obtuse-angled triangle. Once
the LHS is in this form, one of the first things you might notice is that when a
given obtuse-angled triangle is enlarged—say with scale factor A—the ex¬
pression on the LHS is unchanged! So, given any obtuse-angled triangle, you
may enlarge (or shrink) the triangle in any way that happens to be conve¬
nient. In particular, you may enlarge the given triangle so that the perimeter
a + b + c is as simple as it could possibly be; namely, equal to_. The LHS
of the inequality (*) then becomes much simpler, and all you have to find is
the largest possible value of K, such that
a2+ b2+c2>K
158 Hints and outline solutions

holds whenever a, b, and c are the sides of an obtuse-angled triangle with

perimeter a + b + c =_.
(6) In any obtuse-angled triangle, one side is definitely longer than the other
two: let c be the length of the longest side. It is then natural to use the
*o*i*e rule to obtain a formula expressing c2 in terms of a, b, and the angle
C. Do this, and use this expression for c2 to obtain an expression for
‘a2 + b2 + c2’.
At the same time, you should not forget that c = 1 — (a + b). Combine this
with the expression for c2 given by the cosine rule to prove the important
relation
2 (a + b) — 2ab(\ + cos C) = 1.

It is much harder to handle the algebra if all three variables a, b, and c are
allowed to vary at the same time. One way of simplifying things which is often
useful is to concentrate on one permissible value of c at a time; that is, to
keep c fixed for the moment. Then a + b = 1 — c is also fixed; and the
equation that you have just proved then shows that 2ab(l + cosC) is also
fixed. Since you are trying to find a universal lower bound for the expression
la2 + b2 + c2’, it is natural, for each value of c, to rewrite this expression in a
form which suggests a natural lower bound: for example, involving squares
(which are always > 0), and terms depending only on a + b and 2ab(\ + cos C)
(which remain fixed when c is fixed). Use c = 1 — (a + b) to show that

(a+b)2 (a-bf 2
a2 +b2 + c2 ---1---h (a+b) — 2ab(l + cos C).

It follows, for each permissible value of c, that

2 ,2 2 3 (a+b)2 3 (a+bf
a2 + b2 + c2>---2ab{\ + cos C) =---+ 1-2 (a+b),
^ z
with equality holding if and only if a = b (that is, if and only if the
obtuse-angled triangle is i*o**e*e*). Hence, for a given value of c, the best
possible lower bound K(c) is 2a2 + 2a2{\ - cos C) = |(1 - c)2 + c2.
It remains to find the best value of K such that 2a2 + 2u2(l - cos C) > K
whenever the angle C is o**u*e. For this, it is sufficient to concentrate on
the case which gives the smallest value of a2 + b2 + c2 for each c; namely,
when a=b (so that the triangle is isosceles). Show that c>y/2-a, whence
c >-> giving K(c) = |(1 — c)2 + c2 >_, and this is the required
value of K (since c can be arbitrarily close to ^2 - 1).
[This problem is genuinely hard! I have tried to avoid presenting a slick,
but completely unmotivated, solution. There are other approaches, but they
all have to face the subtlety that the extremal example (an isosceles right-
angled triangle) is not obtuse-angled’, and so does not belong to the given
domain. One way of avoiding the cosine rule is to let c (>a,b) be fixed.
 22nd BMO, 1986 159

Then a2+ b2 = [(a + b)2 + {a - b)2]/2 >(a+ b)22/2 = (1 - c)2/2, since a +

b + c =_. Hence a2 + b2 + c2 ^ c2 + (1 - c)2 /2, with equality if and only
if a = b.]

5. Find, with proof, the number of permutations

a\, a2, a3,...,an of 1,2,3,...,/*

such that

ar<ar+2 forl<r<n-2 and ar<ar+3 forl<r<rc-3.

(1) A reasonable way of trying to obtain a feeling for what the question
means is to work with small values of n in turn, and to list, and hence count,
all permutations satisfying the given conditions. Let Pn denote the number of
permutations of 1,2,...,/* which satisfy the given conditions.
When n = 1 or n = 2 the conditions do not apply (Why not?), so Pn counts
all permutations: hence P1 = 1 and P2 =_. When n = 3, the second condi¬
tion does not apply (Why not?), and the first condition requires only that
ax < a3; hence, either a3 = 2 (so ax =_, and there is just one such permuta¬
tion), or a3 = 3 (in which case we must have either ax =_ or ax =_, so
there are exactly_ such permutations). Thus P3 =_. Now calculate P4.
(2) While it can be helpful to work your way into a problem like this, there
are two major snags. The most frequent (but least significant) snag, is that it
is easy to make a mistake! The data that you generate will then mislead
rather than help. (This is one instance of the infamous ‘GIGO’ (Garbage In,
Garbage Out) principle: if your basic data is flawed, your conclusions are
doomed.) The more serious snag is that you have to solve the problem for all
n, and, although you may sometimes be lucky, you should not assume that the
numbers you calculate for small values of n will necessarily give you a clue as
to what happens in general. In this instance, for example, P4 does not have
the ‘obvious’ value which may be suggested by the values of Pv P2, and P3.
But if you work out P4 and P5, you should come up with another idea. That
will at least give you something to aim at.
(3) But whether or not you think you know what the answer is, you still have
to find some way of counting which will give the value of P„ for every n> 1.
One natural approach in problems such as this one is to try to find a
*e*u**e**e *e*a*io*; that is, you want to express the value of Pn in terms
of P„_j, P„_2, and so on.
One thing that you should have noticed when calculating P3, P4, and P5
was that the largest available number n always has to go in one of two places.
What are these two places? Use the given conditions ‘ar < ar+2’ and car < ar+f
to prove that this restriction always applies.
(4) The hint in (3) should have been enough to show the way; for once you
know that the number n has to go in either the very last position or the
160 Hints nnd outline solutions

last-but-one position, the stage is set for a recurrence relation. Suppose that
n goes in the very last position; then the first n- 1 terms form a permutation
of the numbers which satisfies the standard conditions ‘ar <
ar+2 and ‘ar <ar+3\ and there are exactly _ such permutations. Suppose,
on the other hand, that n goes in the last-but-one position; then the standard
conditions guarantee that the last position must be occupied by_, so the
last two positions are uniquely determined and the first n — 2 terms form a
permutation of 1,2,..., n — 2 which satisfies the standard conditions, so there
are exactly _ such permutations. Hence Pn=Pn_i + Pn-2- Finally, since
the first two terms are the second and third *i*o*a**i numbers, it follows
that Pn = Fn + j for all n > 1.

6. AB, AC, and AD are three edges of a cube. AC is produced to E so

that AE = 2 AC, and AD is produced to F so that AF = 3AD. Prove
that the area of the section of the cube by any plane parallel to BCD is
equal to the area of the section of tetrahedron ABEF by the same
plane.

(1) Let the cube have side 1, and let A',

B', C', and D' be the vertices of the
cube diametrically opposite A, B, C, and
D respectively. Since the edges AB, AC,
and AD belong to both the cube and the
tetrahedron ABEF, it is natural to take
A as ‘origin’, and to position the cube in
the positive octant, with AB, AC, and
AD as x-, y-, and z-axes.
(2) The plane BCD has equation x +
y + z =_. Each plane parallel to BCD
has an equation of the same form, x +
y + z = c, with different constants c on
the RHS.
When c < 0, such a plane cuts neither
the cube nor the tetrahedron ABEF. And when 0 < c < 1, the plane x + y +
z — c cuts both the cube and the tetrahedron in exactly the same triangle
(parallel to BCD, and with area c2-area(5CD)). Hence the two cross-
sections have equal area whenever c < 1.
The plane x+y + z = 3 cuts the cube in the single point _, and cuts the
tetrahedron ABEF in the single point_. When c > 3, the plane x+y + z = c
cuts neither the cube nor the tetrahedron.
You therefore need only consider the remaining cases in which 1 < c < 3.
(3) The plane x + y + z — 2 cuts the cube in the **ia***e _, and cuts
the tetrahedron ABEF in the **ia***e PEQ, where P is the midpoint of BF
and Q is the midpoint of DF. Hence, when c ^ 2, the only part of the cube
that need concern you is the tetrahedron B'C'D'A', and the only part of the
 22nd BMO, 1986 161

tetrahedron ABEF that need concern you is the tetrahedron PEQF. How¬
ever, the bases B'C'D' and PEQ of these two tetrahedra lie in the same
plane x + y + z = 2; and the two tetrahedra have equal ‘height’ perpendicular
to this plane (since the parallel plane x + y + z = 3 passes through both
apexes A’ and F). Show that the two bases B'C'D' and PEQ have equal
areas. Hence prove that, for each value of c, 2 < c < 3, the cross-sections
produced by the plane x +y + z = c with the tetrahedra B'C'D'A' and
PEQF have equal areas.
(4) It remains to sort out what happens for values of c between 1 and 2. Let
c — 1 +1, where 0 < t < 1.
(4A) Consider first the cross-section produced by the intersection of the
plane x+y+z=\+t with the cube. This cross-section is equal to the
triangle BCD when t =_, and is equal to the triangle B'C'D' when
t= . For values of t between 0 and 1, the plane cuts the cube between
BCD and B'C'D' in a *e*a*o*, having successive sides parallel to B'C',
DB, C’D', BC, D'B', and CD. Hence the angles of the hexagon are all equal

If the plane x +y + z = \ +1 cuts DC' at X and DB' at Y, then DX =

DY =_. Hence XY =_. Similarly, if the same plane cuts C'B at Z,
then C'X=C'Z =_, so XZ =_. Since BZ =_, it follows that
the next side of the hexagon has the same length as _, and that the edge
lengths of the cross-section are alter¬
nately t)l2 and (1 — fV2. Hence the
hexagon is obtained from an equilateral
triangle LMN of side XZ + 2 XY =
(1 + f)V2 by chopping off three congruent
corners of side length XY =_. If you
wish, it is now a simple matter to calcu¬
late the area of this cross-section (al¬
though it may be better to wait and see
whether this is strictly necessary).
(4B) Consider next the cross-section produced by the intersection of the
plane x +y + z = l +1 with the tetrahedron ABEF. When this
cross-section is a *ua**i*a*e*a* with successive sides parallel to BD, DC,
CB, and PE. Moreover:
(i) the plane cuts the face AEF in pre¬
cisely the segment MN;
(ii) the plane cuts the face ABF in a
segment NS, where S lies on NL\
(iii) the plane cuts the face ABE in a
segment MR, where R lies on ML.
Thus you only need to calculate the
lengths NS and MR.
162 Hints and outline solutions

In the plane y = 0, the line NL has equation * + z =_. Write down the
equation of the line BF. Hence find the x co-ordinate x(5) of the point S,
and show that NS =*(£)•/2.
In the plane z = 0, the line ML has equation x+y =_. Write down the
equation of the line BE. Hence find the x co-ordinate x(R) of the point R,
and show that MR = *(/?)•/2.
Finally, show that LR = 2t-^2, LS = 31^2/2, so that the triangle LRS has
area equal to 3 X area( AXYN).
[Alternatively, you might prefer to work out the 3-D co-ordinates of L, R,
and S, and use these to calculate area( A LRS).]

21st British Mathematical Olympiad, 1985

1. Circles S1 and S2 both touch a straight line p at the point P. All points
of S2, except P, are in the interior of Sv The straight line q: (i) is
perpendicular to p\ (ii) touches S2 at R; (iii) cuts p at L; and (iv) cuts
Sj at N and M, whc re M is between L and R.
(a) Prove that RP bisects A MPN.
(b) If MP bisects A RPL, find, with proof, the ratio of the areas of the
circles Sj and S2.

(1) Draw a decent diagram.

(2) The question is all about *i***e* (Sj and S2) and *a**e*** (p and q),
so you should have already started to apply the two most basic facts that you
know:

(a) the two tangents LP and LR from L to S2 must have the same *e****
(so triangle LPR is i*o**e*e*); and

(b) the ‘alternate segment theorem’ (which states that the angle between a
tangent and a chord of the same circle is equal to the angle subtended in
the opposite segment).

The first of these facts shows that A LPR = A LRP. And if we denote the
point at which PN cuts S2 by X, then the second fact shows that A NRX =
A_ and that ALPM = A_. You should now be able to prove that RP
bisects A MPN (combine the above observations with the fact that A LRP is
an exterior angle of triangle RPN).
[Alternatively, if 02 is the centre of S2, then PLR02 is a **ua*e, so PR
bisects LLP02. Also ALPM = A PNL (by (b)), and AP7VL = LNP02 (since
LN and P02 are parallel). Hence A MPR = A RPN.]
(3) If you reflect on what you have just proved, it should be clear that part (a)
of the problem does not use the given information that ‘<7 is perpendicular to
p’- You should therefore expect this to play an important role in part (b).
 21st BMO, 1985 163

(4) Now suppose that MP also bisects Z.RPL. You have to calculate the ratio
of the areas of 5t and S2■ In other words, you have to find the ratio of the
**ua*e of the radii of the two circles, so you need to mark the *e***e* Ox
and 02 of the two circles S1 and S2. These two centres lie on the perpendicu¬
lar to the tangent p through the point P.
What can you say about the quadrilateral 02RLP1 What does this tell you
about /-LPR1 What can you say about Z.MPN? And what does this tell you
about /2MOlNl So what can you conclude about the lines 02L and O^Ml
Finally, what does this prove about the ratio 01M:02R—and hence about
(01M)2 :(02R)21

2. Given any three numbers a, b, and c between 0 and 1, prove that not
all of the expressions a( 1 — b), b( 1 — c), and c(l — a) can be greater
than

(1) At first sight, the wording may seem strange: you have to ‘prove that not
all of the expressions a( 1 - b), b( 1 - c), and c(l - a) can be greater than \\
but you have no idea which one is supposed to have this property! However,
since you are only told that ‘a, b, and c lie between 0 and 1’, there is no
reason why any particular one of the three expressions should be < if a = {
and b = c = \, then b( 1 - c) < whereas if a = b=\ and c = \, then it is
a( 1 - b) that is <\.
(2) There are several approaches which work here, but they all come down
to the same basic fact: if ‘a lies between 0 and T, then you should know (or
be able to prove for yourself) a fundamental inequality of the form
‘a( 1 - a) <_[Given any two positive real numbers x and y, the AM-GM
inequality guarantees that -Jxy < (x + y)/2; now put x = a and y = (1 — a).]
(3) You have to prove that not all of a( 1 - b), b( 1 - c), and c(l - a) can be
greater than \ at the same time. (You have already seen that each one can be
>i provided that the other two are small.) One way to prove that something
must happen is to show that it could not possibly be false. It therefore makes
sense to turn the question around and to ask whether all three of these
expressions could possibly be > \ at the same time. If that were to happen,
then the **o*u** of the three expressions
a(l-b)-ba-c)-dl-a)
would be >(^)3. Now rearrange the six factors and use the inequality you
proved in (2) to show that the above product is always <(|)3. Hence complete
the solution.

3. Given any two non-negative integers n and m with n>m, prove that
164 Hints and outline solutions

(1) If you want to solve any mathemati¬ 1

cal problem, you need to know a thing or 1 1
two to get started—although you can 2 1
usually manage with much less than you 3 3 1
think! The binomial coefficients on the 6 4 1
LHS of the identity to be proved corre¬ 10 10 5 1
spond to entries in Pascal’s triangle with
fixed second ‘co-ordinate’ m. Thus these
entries lie on a ‘diagonal’ which starts at ^ j and which slopes down to the
left—like the diagonal picked out in bold in the figure here. Notice that
1 + 2 + 3 =_, and that 14-2 + 3 + 4 =_. This should suggest that it
might be helpful to begin by proving the identity

Then rewrite the LHS of the identity to be proved in the form

+ ("41) + ('!m2) + -+(m)

+ ("42) + -+(m)
*(
and use (*) over and over again to obtain a sum like the LHS of (*), but with
n and m replaced by n + 1 and m + 1. Hence complete the solution.
(2) Another approach is to notice that the RHS of the identity to be proved
counts the number of subsets of the set {1,2,...,« +2} which have ra + 2
elements. The first term of the LHS clearly counts the number of such
subsets which include both n + 1 and n + 2. Explain why the second term on
the LHS counts those subsets which contain n and just one of n + 1 and
n + 2. Continue in this way to explain how the LHS of the required identity
accounts for all possible subsets of the set {1,2,...,n + 2} which have m + 2
elements.

4. The sequence /„ is defined by /0 = 1 and f1 = c, where c is a positive

integer, and for all n > 1 by

fn = 2/„_i —fn-2 + 2.
Prove that, for each k > 0, there exists h such that fkfk+i =fh.
 21st BMO, 1985 165

(1) Later questions on the BMO tend to be harder than earlier questions! As
it stands, this one may seem thoroughly opaque. You are not told the value of
c, the recurrence relation is unfamiliar; and you are asked to prove that, ‘for
each k ^ 0, there exists an h’ without having any idea what the value of h
may be.
(2) One way to escape from this feeling of helplessness is to choose a simple
value of c and to calculate a few terms of the sequence, while keeping your
wits about you to try to see what is going on. The simplest ‘positive integer’ to
try for c is presumably f1 = c =_. So try this value for c, and calculate the
first 18 or so values:

/0 = 1, h=c = \, f2 = 3, /3 = 7, /4 = 13,
-^5 21, /6 = 31, f7 —-, /g =-, fg=-,
f 10 — > fll — ! f\2 = > /l3 =-5 /l4 = 5
fl5~-> /16 = > /l7 = > /l8 =-

Now look at fx-f2 =_=/?, f2 'fj =_= /?. Remember that you are
trying to predict the value of h for each value of k\ so after calculating the
first few products, risk a guess about what seems to be going on and make a
prediction before you calculate the next product /3-/4 =_=/,.
(3) Now repeat all of this with c = 2, and try to understand how the new
value of c affects the value of h. When you think you understand, predict
what you expect to happen when c = 3, and check a few cases.
(4) You are probably now in a position to formulate a conjecture which
expresses the h in lfk-fk+\ =fh’ in terms of k and c. If you try to prove this
conjecture directly using induction, you may have problems: while it is easy to
check your formula for k = 0, the induction step is not at all easy. You may
assume that fk_ x ■fk has the form that you expect, and then use the given
recurrence multiplied by fk,

fk 'fk + l = ^fk ~ fk- 1 'fk + 2/yt,

to try to prove that the same formula holds for fk-fk+v but it is not clear
what to do next—unless you could somehow discover a formula for each termfk
in terms of k and c. (There may be other ways to solve the problem, but this
would seem to be the most direct approach.)

(a) When c = 1, you should notice that each fk is slightly less than k2; it is
then not hard to guess the formula fk = k2 -_.

(b) When c = 2, things are even easier, since then each fk =_.

(c) When c = 3, the previous two cases soon suggest trying fk =_,
and this should lead you to conjecture a general formula for fk in terms
of k and c.
This formula can be proved relatively easily (by induction) using the given
166 Hints and outline solutions

recurrence. Your conjectured value of h can then be checked by direct

algebraic computation.

5. A circular hoop of radius 4 cm is held fixed in a horizontal plane. A

cylinder with radius 4 cm and length 6 cm rests on the hoop with its axis
horizontal, and with each of its two circular ends touching the hoop at
two points. The cylinder is free to move subject to the condition that
each of its circular ends always touches the hoop at two points. Find,
with proof, the locus of the centre of one of the cylinder’s circular ends.

(1) Draw a diagram showing the initial position of the hoop and the cylinder
(that is, with the axis of the cylinder horizontal). Then draw diagrams showing
possible subsequent positions. What is the maximum possible angle of tilt of
the axis of the cylinder?
(2) Let C be the centre of one of the cylinder’s circular ends. You have to
find the locus of C. Finding loci, or equations of curves and surfaces, does
not normally require anything more sophisticated than Pythagoras’ Theorem.
In this case, the problem is made harder by the fact that you are not told
what point to use as origin, or what lines to use as axes. It is also unclear
what sort of locus to expect.
Any acceptable position of the cylinder can be rotated around the vertical
line through the centre of the hoop, so the required locus will be a three-
dimensional *u**a*e. Given the rotational symmetry of this surface, it is
natural to choose the vertical line through the centre of the hoop as the
z-axis. Then, for each fixed value of z, the locus of C will be a *i***e.
Hence, it will suffice if you can describe the locus in any fixed vertical plane
containing the z-axis (since you can then rotate this part of the locus around
the z-axis). So, fix a y direction (to the right, say), let C be the centre of the
cylinder’s right-hand circular end, and restrict C to move in the y-z plane.
In the absence of any other clues, it makes sense to choose the centre O of
the hoop as origin of co-ordinates, with the radius to the right as the y-axis,
meeting the hoop at the point Y:

(a) What are the y and z co-ordinates of the point Y?

(b) Find the co-ordinates (y, z) of C when the axis of the cylinder is
horizontal.

(c) Now tilt the cylinder upwards so that the right-hand circular end touches
the hoop in two ‘coincident’ points (at Y = (4,0)). Find the co-ordinates
of C. (This is completely elementary. However, you will need a combina¬
tion of (i) persistence, and (ii) faith that 3-D diagrams can almost always
be analyzed using only Pythagoras and similar right-angled triangles.)

(d) Finally, tilt the cylinder downwards until the left-hand circular end
 21st BMO, 1985 16 7

touches the hoop in two coincident points (at (-4,0)). Find the co¬
ordinates of C. (This should be straightforward once you have done part
(c). But be prepared for a surprise!)

(3) For a general solution you could find the co-ordinates (y, z) of the point
C when the axis of the cylinder is inclined at angle 6 to the horizontal, and
hence find the equation of the locus. An alternative approach is to think
about the results of your calculations in parts (b)-(d) of (2) above. What is
the most likely possibility for a curve which passes through the point (3,3),
and which passes through the two points (§, ff) and (§, §)? The symmetry
of these three points about the line z = 3 should lead you to check the
distances of these three points from the point (_,_). You should now
have a fairly clear idea of what the locus of C must be, and can now try to
show that a general point on this curve between the two extreme points found
in parts (c) and (d) of (2) is indeed the position of C for some inclination of
the cylinder’s axis.
[Alternatively, you may now realize that it suffices to explain why the
midpoint of the axis of the cylinder remains fixed throughout the motion!]

6. Show that the equation x1 2 +y2 * =z5 +z has infinitely many solutions in
positive integers x, y, and z having no common factor greater than 1.

(1) What is the ‘simplest’ possible triple x, y, z (having no common factors)

which satisfies x2+y2=z5+z? What is the next simplest solution? Make a
list of fifth powers and use it to find the next simplest solution.
(2) It makes sense to interpret ‘simplest’ as meaning ‘with smallest possible
value of z’. z = 1 certainly works (with x =_ and y =_); z = 2 also
works (with x =_ and y =_). And though z = 3 and z = 4 do not
correspond to solutions, you probably discovered that z = 5 does work (with
x =_ and y =_).
At this stage all that you know is that z = 1, z = 2, and z = 5 correspond to
solutions, but z = 3 and z = 4 do not. For a quick solution, you need to notice
two things. Observe first that the simplest solution of all is a little special
(since any triple with x = y = z would normally contradict the requirement
that x, y, and z have no common factors). Thus we should concentrate on
the fact that z = 2 and z = 5 work (and on the fact that z = 3 and z = 4 do
not). Notice next that the values of z which work—namely 2 and 5—are
precisely those which can themselves be written as the sum of two non-zero
squares (and the values which do not work namely 3 and 4 are precisely
those which cannot be written as the sum of two non-zero squares). Once you
have made this jump, the rest is fairly straightforward:

(a) Show that, if a, b, c, and d are integers, then the product (a2 + b2)(c2 +
d2) can always be rewritten as the sum of two squares. (If_z = a+b\ and
w = c + di are complex numbers, then (zz)-(ww) = (zw)-(zw).)
168 Hints and outline solutions

(b) Use (a) to show that whenever z can be written as the sum of two squares
so can z5+z. Finally, use the fact that infinitely' many integers (for
example, all integers of the form n2 + 1) can be written as the sum of two
squares to complete the solution.

20th British Mathematical Olympiad 1984 ,

1. P, Q, and R are arbitrary points on the sides BC, CA, and AB
respectively of triangle ABC. Prove that the triangle the vertices of
which are the centres of the circles AQR, BRP, and CPQ is similar to
triangle ABC.

(1) It may take a moment or two to realize that when the question refers to
‘the circle AQR’ it can only mean one thing—namely, the circle through A,
Q, and R\ that is, the circumcircle of the triangle AQR. Thus the centre of
‘circle AQR’ must be the circumcentre of triangle AQR (so you will need to
know how to construct the circumcentre of a triangle). Now that you under¬
stand the question, you should draw an accurate diagram using ruler and
compasses. (A good diagram can give clues about how to proceed—and how
not to proceed! But for that it is important to make sure that ABC is a
general triangle, and that the points P, Q, and R are positioned carefully to
avoid ‘accidents’.)
(2) One of the simplest of all problem-solving principles is Name and
Conquer. To refer to something, it must have a name: and it often helps to
choose a good name. You have to prove that the triangle the vertices of
which are the circumcentres of AQR, BRP, and CPQ is ‘similar to triangle
ABC\ Part of your problem is to decide which circumcentre corresponds to
A, which to B, and which to C. In the absence of other clues, it seems likely
that A will correspond to the circumcentre nearest to A: so let A' denote
the circumcentre of AQR, B' the circumcentre of BRP, and C' the circum¬
centre of CPQ. Your immediate task is to find some reason why the angles at
A and at A' are equal (the reason will probably carry over to explain why the
angles at B and B' are equal). Before proceeding, draw the three circles
AQR (centre A'), BRP (centre B'), and CPQ (centre C') as accurately as
you can.
(3) The first two circles meet at R and at a second point X (say). The second
two circles meet at P and at a second point X' (say). The first and third
circles meet at Q and at a second point X” (say). Your diagram may lead you
to suspect something about the three points X, X', and X"\
The quadrilateral ARXQ is ****i*, so /LRXQ = 180° - LA- similarly
Z. RXP = 180° — Z. J5. Hence

Z. PXQ = 360° - (180° - z. A) — (180° ~/_B)=_

 20th BMO, 1984 169

It follows that quadrilateral CPXQ is ****i*, so X' = X" = X. The rest is

easy: A A' = AB'A'X + AXA'C' = \ARA'X + jAXA'Q = {ARA'Q =
A A.

2. Let an be the number of binomial coefficients ^ j (0 < r < n) which

leave remainder 1 on division by 3, and let bn be the number which

leave remainder 2. Prove that an >bn for all positive integers n.

l n=0
71= 1
(1) The nth row of Pascal’s triangle l l n—2
contains n + 1 numbers—namely l 2 l
1 3 3
(o)’(i)*•••»(”)• Each leaves remain-
14 6 4
der 0, 1, or 2 on division by 3. If we • • •
are only interested in these remainders,
it makes sense to replace each entry
in Pascal’s triangle by its residue 1
(mod 3). Using the basic recurrence 1 1
= (mod3)’ one 1 2 1

10 0 1
can generate the triangle very quickly.
Generate the next ten or eleven rows of 110 11

the triangle on the right, keeping an eye 12 112 1

open for any striking patterns which might 1 0 0 2 0 0 1
give you a clue as to why the required
110 2 2 0 11
result ‘an > b„’ should be true for all n >

(2) The result ‘an > bn’ is certainly true for n = 0,1,2. One of the things that
should have struck you in (1) is the way in which the next three rows
(n = 3,4,5) of Pascal’s triangle (mod 3) seem to be built out of two copies of
the first three rows with an inverted triangle of 0’s in between. Since there
are more Is than 2s in rows 0, 1, 2, it follows that for each n — 3,4,5 the
difference an - bn will be exactly twice the corresponding difference in row
n-3. The rows for n = 6,7,8 show that there is rather more to the problem
than this! Nevertheless, it is important to formulate and prove the general
fact underlying the above observation.

(a) Let n = 3k. Prove that the coefficient of tr in the expansion of (1 +t)n
is divisible by 3 for all r, satisfying [Write (1+U3 =
([1 + r]3)3*-1 = (l + 3t + 3t2 +13)3"~' = (1 + t3)3 + • • • • Then use induc¬
tion on k.]
170 Hints and outline solutions

(b) Conclude that the rows n = 3k,3k + 1,...,3* + 3k — 1 consist precisely of

one copy of rows 0 to (3k - 1) on the extreme left and another on the
extreme right, with an inverted triangle of Os in between. Hence prove
that an > bn whenever n has the form 3k + i (0 < i < 3k) for some k ^ 0.

(3) It remains to prove that an > bn when n has the form 2 • 3* + i (0 < i < 3*).
From what you have just proved, you know that the (2 -3* — l)th row consists
of two end-to-end copies of ‘121212...121’. Hence the (2-3fc)th row has
the form
‘100000...00200...000001’.

This then guarantees that the next 3* rows will consist of one copy of the first
3* rows on the extreme left, another copy of the first 3k rows on the extreme
right, and a third triangle in the middle (with its apex at the ‘2’ in the centre
of the (2 - 3*)th row), these three sub-triangles being separated by inverted
triangles of zeros. The first three rows look like this:

100000.00200 .000001
1100000. .. .002200. ...0000011
12100000. ..0021200. ..00000121

The central triangle contains lots of 2s, and at first sight it may not be clear
how to prove that each row will still contain more Is than 2s.
Fortunately, the central triangle is really very simple, for it is generated in
exactly the same way as the left-hand and right-hand triangles—except that it
starts with ‘0,2, 0’ instead of ‘0,1, 0’ in the top row. If we use the same Pascal’s
triangle recurrence, and simply multiply the initial row by some constant,
then each subsequent row will be equal to the usual row multiplied by the
same constant. Thus, whenever the (usual) triangle on the extreme left has a
1, the corresponding entry in the central triangle will be a _, and
whenever the (usual) triangle on the extreme left has a 2, the corresponding
entry in the central triangle will be a _. Thus the number of Is in the
left-hand triangle is equal to the number of 2s in the central triangle, and
the number of 2s in the left-hand triangle is equal to the number of Is in the
central triangle. Hence these two triangles ‘cancel each other out’—so the
difference an — bn when n = 2-3k + i (0 < i < 3*) is controlled by the corre¬
sponding difference in the triangle on the extreme right (namely a, — b{,
which we already know is positive).
(4) The above approach stems from ‘looking for obvious patterns’ in Pascal’s
triangle (mod 3). It is in fact slightly easier to prove the result directly by
induction three rows at a time. The result is certainly true when n = 0,1,2.
Suppose that it is true for all rows up to row 3m - 1. Use (l+r)3m =
(1 +13, + 3(t + t2))m to show that, if the rath row has the form T,p,q,..., 1\
then the (3ra)th row has the form ‘1,0,0,/?,0,0,g,0,0,...,0,0,1’, whence
a3m > t>2m- When you use this row to construct the (3ra + l)th row you find
 20th BMO, 1984 171

that a 3m+ 1 - auu b- + 1

and u3m Constructing the
SO fl3m + l > ^3m + l
(3 m + 2)th row shows that a 3m + 2 = 2am+bm and b3m + 2
-» S0 a3m+2 ~
^3m + 2 =am~ - > 0.
(5) Look at the actual values of un — bn (they are all *o*e** of **o). If you
think carefully about the above proofs, you should be able to find a formula
for an — bn (the exponent is related to the base 3 representation of n).

3. (a) Prove that, for all positive integers m,

/ 1W 3 \l 5 / 2m — 1 \
2-- 2-- 2-- ... 2-
m I
Urn!. .
\ m l\ rn) \ \ m )
(b) Prove that, if a, b, c, d, and e are positive real numbers, then

/a\4 /b\4 / c \4 (d\ 1 e \4 b c d e a

+

+
+

+
\V
■1

1
- +
\b ) \c1 \d) U / ' a ) a b c d e

(1) Parts (a) and (b) look totally unrelated. Nevertheless, it makes sense to
start with part (a) (if only because the first part of such a question is generally
easier than the second). How many factors are there in the product m! on the
RHS? And how many factors are there in the product on the LHS? It may
seem optimistic, but one reason why ‘LHS < RHS’ might be true is if each
factor on the LHS happens to be less than or equal to the corresponding
factor on the RHS! (Note that 2 X 8 < 3 X 7 even though 8 > 7, so you have
no right to expect this to work—but it just might!)
1 2m — 1
(a) Prove that 2-< m and 2-< 1 for all m > 1.
m m
2i + 1
(b) Is it true that 2-< m - i for all i (0 < i < m)?
m
[Alternatively, reversing the order of the factors on the LHS and simplify¬
ing gives

1 3 (2m — 1)
<L2m'‘<L2-3-...-m=m!]
m m m

(2) The methods outlined in (1) should have convinced you that the inequal¬
ity in part (a) of the problem is rather weak. Show that

/ r \I (2m -r) \
12-11 2-I < 1 for each r.

Hence prove that the LHS of part (a) is < 1.

172 Hints and outline solutions

(3) Part (b) of the problem looks more challenging. If you knew a thing or
two about inequalities, you would probably not be reading this in search of
help; so we shall proceed slowly. The symmetry of the two sides should
suggest a family of instances when the inequality becomes an equality
(although it is less clear that this is the only way in which the two sides can be
equal). However, this should convince you that the actual values of a, b, c, d,
and e are irrelevant: it is the *a*io* that matter. It therefore makes sense to
simplify the notation slightly by writing a/b = v, b/c = w, c/d=x, d/e=y,
and e/a = z, and to rewrite the inequality to be proved in terms of u, w, x, y,
and z. Do this. (Note that you now have the extra piece of information that
vwxyz =_.)
(4) Thus you have to prove an inequality involving five variables v, w, x, y,
and z and fourth powers. One way of getting to grips with such a problem is
to consider the corresponding problem for two and three variables first. This
may prove nothing, but it can be a source of ideas:

(a) Write down the corresponding two-variable inequality (involving v and w

only) and use the fact that uw =_to show that the inequality is in fact
an equality in this case.

(b) Write down the corresponding three-variable inequality in v, w, and x

(involving second powers on the LHS, and with v, w, and x satisfying
vwx = 1). Use vwx = 1 to simplify the RHS. At bottom every inequality
comes down to proving that some expression is non-negative. In this case
you have to prove that v2 + w2 + x2 - uw — wx — xv ^ 0. The form of the
terms on the LHS should suggest that you should try to write the LHS as
a sum of perfect squares. (This is easiest if you first multiply both sides
by 2.)

(5) Now return to the five-variable problem. You have to prove that

11111
y4 + w4 +x4 +y4 +z4 + — + - + - +
vwxyz

The first move is to use vwxyz = 1 to rewrite the RHS as wxyz +_+
_+_+_. Now take these terms of the LHS, multiply the whole
inequality by 2, and try to write the new LHS as a sum of perfect squares.
(Since the problem with five variables is more complicated than that with
three variables, you should not be surprised if you need an intermediate step:
the obvious first move of writing five squares such as (v2 — w2)2, introduces
new terms (‘—2v2w2\ and so on). If you ‘cancel’ each such term (by adding
‘2v2w2\ and so on), you obtain five new terms; these and the five old terms
‘-(2wxyz + 2xyzw + 2yzvw + 2zvwx + 2vwxy)' can then be rewritten (as in
(4)) as the sum of five more perfect squares.)
[Alternatively, the fact that the LHS is a sum, together with the symmetry
 20th BMO, 1984 173

of the terms should suggest using the AM-GM inequality—although the fact
that there are *i*e terms and that the exponent is *ou* argues strongly
against a simple-minded application. The fourth powers suggest taking fourth
roots, taking the terms on the LHS four at a time we obtain five different
inequalities. Adding these inequalities gives the required result.]

4. Let N be a positive integer. Determine, with proof, the number of

solutions x in the interval l<x<A of the equation x2-[x2] =
U-[x])2.

(1) Problems which involve an unfamiliar notation are often much easier
than they seem. However, you may have to do some simple-minded calcula¬
tions with small numbers just to get used to things.
(2) The variable x in the question can be any real number in the interval
1 <x <7V. The expressions [x] and [x2] are easy to understand when x (and
x2) are i**e*e**, so it is natural to think about this case first:

(a) Suppose that x is an integer. What is the value of x- [x]? What is the
value of x2 - [x2]? Does the equation x2 - [x2] = (x - [x])2 hold in this
case?

(b) Now suppose that x is not an integer, and let [x] = n be the largest
integer <x. Then x = n + r, where 0 < r < 1. Calculate the value of the
RHS (x — [x])2 (in terms of r). Calculate x2 (in terms of n and r). What
value must [x2] take if the equation x2 - [x2] = (x — [x])2 is to be
satisfied?

(c) Suppose that n — 1; for how many values of r (with 0 < r < 1) is 2nr an
integer? Suppose that n = 2; for how many values of r (with 0 < r < 1) is
2nr an integer? Suppose that n = 3; for how many values of r (with
0<r<l) is 2nr an integer? In general, given any integer n, for how
many values of r (with 0 < r < 1) is 2 nr an integer?

(d) You now know that solutions x of x2 - [x2] = (x - [x])2, with 1 <x < N,
are of two kinds: integer solutions, of which there are exactly_; and
non-integer solutions of the form n + r (1 ^ n < N, 0 < r < 1), of which
there are exactly 1 + 3 + 5 + ... +_. Use this to obtain your
answer in closed form.

5. A plane cuts a right circular cone with vertex V in an ellipse E, and

meets the axis of the cone at C. The point A is an extremity of the
major axis of E. Prove that the area of the curved surface of the slant
cone with V as vertex and E as base is (VA/AC) X (area of E).
174 Hints and outline solutions

(1) No matter how you solve this problem you are going to have to struggle to
picture things in three dimensions. So start by drawing as good a diagram as
you can manage.
(2) One approach would be to try to calculate the curved surface area of the
cone and the area of E directly. But this is not strictly necessary: all you have
to do is to show that the *a*io of these two areas is equal to_. Imagine a
very thin (slightly curvy) triangle VPQ, with P and Q on the ellipse either side
of, and equidistant from, the point A. Then PQ is *e**e**i*u*a* to VA, so
the triangle VPQ has area (PQ-VA)/2. Also, PQ is perpendicular to CA, so
the triangle CPQ has area _. Thus the required ratio for the areas of the
cone and the ellipse is in fact equal to the ratio of the areas of these two
small triangles!
(3) Let ECVA = 6 and L.VCA = a. Express VA/CA in terms of sin 6 and
sin a. Let A' be the end of the major axis of E opposite A, and let P' and
Q' be points equidistant either side of A' (with P'Q' very small). Show that
the ratio of the areas of the two triangles VP'Q' and CP'Q' is VA'/CA', and
show that this is equal to VA/CA. (Observe that you were only told that ‘ A is
an extremity of the major axis of E'\ you were not told which extremity.)
(4) The previous two steps suggest an unexpected reason why the ratio of the
two areas (of the cone and the ellipse) may be equal to VA/CA. Suppose you
could show that for any two ‘nearby’ points P and Q on the ellipse, the ratio
of the areas of VPQ and CPQ is exactly VA/CA. The result would then
follow by decomposing the curved surface of the cone and the surface of the
ellipse into a sequence of matching triangles. (This argument can be made
rigorous using calculus.)
Let P and Q be two nearby points on the ellipse and let X be the
midpoint of PQ. The calculation when X is neither at A nor at A' is a little
more involved, since PQ is then perpendicular neither to VX nor to CX. Let
A' be the lower of A and A'.
Cut the cone perpendicular to the axis
through the point A' to create a circular
base. By scaling, we may assume that the
base has radius 1.

(a) Show that the angle between the plane

of E and the horizontal is 6. When a
region in this plane is projected onto the
base plane, its area is multiplied by a
feed constant. What is that constant?

(b) Express the slant height VA' in terms of

6. Let the axis of the cone meet the
base at the point C', and let the generators VP and VQ of the cone meet
the base circle in P' and Q' respectively. If jLP'C'Q' = y, prove that
CPVQ = y-sin d. Conclude that the area of triangle VPQ is equal to
[VX2 - y sin 0]/2.
 20th BMO, 1984 175

(c) Let Z VCX = a'. Calculate the length of the projection of CX on to the
base plane. Hence show that the area of the projection of triangle CPQ
on to the base plane is (CYsin a')2{y/2). Now use the result of (a) to
find the area of triangle CPQ.

(d) Hence show that area(LP0/area(CP<2) = sin a/sin 0 = VA/CA, and so

complete the solution.

(1) You are given positive integers a and m and an integer x such that
m | a(ax — 1). Note first that hcf(o, ax — 1) =_. Thus m factorizes as
m=p-q, where p\a and q\(ax-l) (and hcf(p,q) = 1).
You have to show that there exists an integer y such that both y(ay - 1)
and a(ay - 1) are divisible by m. Now, since hcf(ra, a) = p, m | a(ay - 1) if and
only if q | (ay - 1) —in which case q = hcf(m, ay - 1). Hence, m divides both
a(ay - 1) and y(ay - 1) if and only if p \y and q \ (ay - 1).
(2) You are told that q\(ax- 1). Thus q\(ay- 1) if and only if q \ [(ay - 1) -
(ax — 1)] =-; that is, if and only if q | (y —x). Thus you must prove that
there exists a y of the form kq+x which is also a multiple of p. To prove
this, you must use the fact that hcf(/>, q) —_. Euclid’s algorithm guarantees
that there exist integers u and u such that u-p + wq = 1, and hence
xu -p + xu • q = x. Thus y = xu-p = (-xu)-q + x has the required form and is
visibly divisible by p as required. This completes the solution.
(3) A more direct approach (which produces a particular value of y) is to
observe that ay2 —y = (a2y - a)(y/a). Thus it makes sense to try to choose
y = za to be a multiple of a—since you can then forget one of the two messy
conditions and concentrate on choosing y so that m \ (a2y - a). That is, you
must choose z such that ‘m \ (a2x — a) (given) implies m I (a3z — a)’. Show
that m | (a2x — a) implies that m \ (a3x2 - a). Conclude that you may choose
z =_.

7. The quadrilateral ABCD has an inscribed circle. To the side AB we

associate the expression uAB = p^sin A DAB) + p 2(sin/LAB C), where
px and p2 are the lengths of the perpendiculars from A and B
respectively to the opposite side CD. Define uBC, uCD, and uDA
similarly, using in each case the perpendiculars to the opposite side.
Show that uAB = uBC = uCD = uDA.

(1) Let A denote A DAB, and so on. The definition of uAB introduces the
perpendicular distances px and p2, so it is natural to begin by expressing
176 Hints and outline solutions

each of these unfamiliar lengths in terms of something more familiar: for

example, using the angle at D we have sin D =px/AD, so p1 = AD sin D;
similarly, p2 = BC sin C. Hence uAB =AD sin D sin A + BC sin C sin B,
(2) You are told that the quadrilateral ABCD ‘has an inscribed circle’. Thus
the side AD consists of a tangent from A and a tangent from D and the side
BC consists of a tangent from B and a tangent from C. If tA denotes the
length of the tangent from the point A to the inscribed circle, and so on, then

(*) uab= + ^>)sin jD sin A + (tB + tc)sin C sin B.

Since the RHS is unchanged when we swap A and D and swap B and C,
uab = ucd■ Similarly, uBC = uAD.
(3) Thus it remains to prove that tiAB = uAD. Since later questions tend to be
more demanding than earlier questions, you should expect this to be harder.
However, it is not all that hard. The expressions for uAB and uAD involve a
curious mixture of lengths and trig functions. In (2) you managed to make
partial progress despite this. Further exploration suggests one is unlikely to
prove uAB = uAD in the same way, so you must be prepared to use some
trigonometry to rewrite uAB and uAD in the hope that they will become
visibly equal!

(a) Let O be the centre of the inscribed circle. Use the fact that AO bisects
A BAD to express tA in terms of r and A/2. Do the same for tD, tB, and
tc-
(b) Now substitute in the RHS of equation (*) above and use the half-angle
formula for ‘sin’ to prove that

uab = 4r cos(y4/2)cos(D/2)(sin(D/2)cos(^l/2) + cos(D/2)sin(yl/2))

+ 4rcos(5/2)cos(C/2)(sin(5/2)cos(C/2) + cos(5/2)sin(C/2))

= 4/-[cos(v4/2)cos(£)/2)sin((y4 +D)/2)

+ cos(5/2)cos(C/2)sin((5 + C)/2)].

(c) Now use the fact that (A + D)/2 = 180° - (B + C)/2 to rewrite this as

uab = 4r[cos(yf/2)cos(D/2)sin((5 + C)/2)

+ cos(5/2)cos(C/2)sin((y4 +D)/2)].

Finally, expand sin((5 + C)/2) and sin((^ +D)/2) and rearrange to obtain
an expression which is visibly equal to uAD.
(4) Alternatively, you might like to show that pl = r(l + cos D +
sin D cot(A/2)). Hence prove that uAB = r(sin A + sin B + sin C + sin D),
which does not depend on ‘AB\
 19th BMO, 1983 177

19th British Mathematical Olympiad 1983 ,

l. In triangle ABC with circumcentre O, AB=AC, D is the midpoint of
AB, and E is the centroid of triangle ACD. Prove that OE is
perpendicular to CD.

(1) There are three possible approaches here. If one could see a geometric
reason why OE should be perpendicular to CD, then a purely geometric
solution might turn out to be simplest. In the absence of an obvious
geometric reason it is tempting to try using vectors, or even co-ordinate
geometry:

(a) For a vector approach, let O be the origin, and let a, b, c, d, and e be the
vectors for the points A, B, C, D, and E respectively. You should know a
standard result which expresses e in terms of a, c, and d. You should also
be able to write d in terms of a and b, and so express CD in terms of
a, b, and c. Hence prove that the dot product 12OE CD = 3a a + b b-
4c • c + 4a • b - 4a • c. Finally, explain why a a = b•b = c c, and use the
fact that A ABC is isoceles to explain why a (b - c) = 0. Hence complete
the solution.
(b) For a co-ordinate approach you have to choose an origin. One good
choice is to use the midpoint M of BC as origin, with MC as x-axis and
MA as y-axis. Then A = (a,0), C = (0,c), B = (0, -c), D = (_,_), and
E = (_,_). The circumcentre O lies on the y-axis, so has co-ordinates
(0, y) for some y. Now use AO = CO to express y in terms of a and c:
y =_• Finally, calculate the gradients of OE and of CD to complete
the solution.
(2) You should suspect that, if only you knew a little more, then you would
understand that the result is no accident! Let F be the midpoint of AC and
G be the centroid of Ayl.BC. Then G lies one third of the way up MA; in
particular, D, G, and C are collinear. Moreover, DF is parallel to BC, and E
lies on DF; hence GO is perpendicular to DE.
(a) Prove that FE/FD = FG/FB. Conclude that GE is parallel to BD—and
hence perpendicular to DO.
(b) Hence show that O is the orthocentre of A DEG. Why does this solve
the problem?

2. The Fibonacci sequence {Fn} is defined by

F,-l, F2= 1, F.-F._,+Fn.2 (n > 2).

Prove that there are unique integers a, b, and m such that 0 < a < m,
0 <b <m, and Fn - anbn is divisible by m, for all positive integers n.
178 Hints and outline solutions

(1) Any problem with four parameters (n, a, b, and m) and a sequence (Fn)
can be a little off-putting at first. It is hard to know how to begin. However, if
‘Fn - anbn is to be divisible by m for all positive integers n’, then it must
certainly work when n = 1,2, or 3. So why not start by investigating what this
tells you about a, b, and ml

(a) m divides Fx — a ■ 1 -b if and only if m \ ab — 1. Use this and the fact that
m must divide F2 - a -2-b2 to prove that m \ 2b - 1.

(b) Use (a) and the fact that m must divide F3 — a-3-b3 to prove that
m | b2 + b — 2. Factorize b2 + b — 2, and use the fact that m \ 2b — 1, to
conclude that m\b + 2. Hence m must divide 2(b + 2) - (2b - 1) =_.

(c) Since we are asked to find integers m (and a) satisfying m > a > 0, this
leaves only one possible value for ra—namely m =_. Since m > b > 0
and m\b + 2, there is only possible value of b—namely b =_. And
since m\ab — 1, there is only one possible value of a—namely a =_.

(d) It remains to prove (by induction, using the given recurrence) that
Fn = 2n-3n (mod5) for all n > 1.

3. Given any real number a ¥= —1, the sequence xv x2,x3,... is defined by

xx=a, and xn + l=x2+xn for all n ^ 1.

Let yn = 1/(1 +xn). Let Sn be the sum, and let Pn be the product, of
the first n terms of the sequence yv y2,
y3,... . Prove that aSn + Pn = 1,
for all n.

(1) Like the previous question, this one introduces so much notation that it is
hard to tell whether it is really complicated, or just something simple wrapped
up to look complicated.

(a) Check that Sj = 1 /(a + 1) = Px. Hence verify that aSx + Px = 1.

(b) Suppose that aSn +Pn = 1 for some n> 1. Show that this would imply
that aSn + j + Pn + j = 1 provided that Pn=a/(xn + 1).

(c) Check that Px = a/x2. Prove (by induction on n) that Pn = a/xn + l for all
n $51. Hence solve the problem.
 19th BMO, 1983 179

4. The two cylindrical surfaces

x2+z2 = a2, z> 0, \y\^a,

and
y2+z2=a2, z> 0, \x\ <a,

intersect. Together with the plane z = 0 they enclose a dome-like shape

which we shall call a cupola. The cupola is placed on top of a vertical
tower of height h, the horizontal cross-section of which is a square of
side 2a. Find the shortest distance over the surface of the cupola and
tower from the highest point of the cupola to a corner of the base of
the tower.

(1) The main difficulty in solving three-dimensional problems is often the

first step of seeing exactly what is going on. So draw the best diagram you
can, to get a feeling for the solid which is the intersection of these two
(half-)cylinders.
(2) Each (half-)cylinder has diameter la and length 2a, so the base of their
intersection is a square of side _ (just the right size to fit on top of the
vertical tower). The uppermost point of their intersection is at the point at
which the horizontal upper ‘generators’ of the two cylinders cross. Four
(curved) lines run from this highest point to the four corners of the square
base: these are the lines where the two cylinders cut into each other. Hence
these curved lines lie on the surface of both cylinders; the rest of the surface
of the intersection of the two cylinders lies inside one of the two
cylinders. Thus the surface of the solid of intersection
A
consists of four ‘curvy-triangular’ sections—each be¬
ing part of the surface of one of the two cylinders.
Unlike a sphere, the surface of a cylinder can be laid
out flat. Thus each of the cupola’s four curvy triangles
can be ‘unrolled’ so that it is a direct continuation of
the vertical wall of the tower to which it is joined. The
shortest distance from the apex A to the corner C of
the base is therefore given by drawing the straight
line AC joining the two points on this ‘flattened’
picture. The flattened curvy triangle has height_
(since its altitude was originally one quarter of a circle of radius a), so the
‘tower plus curvey triangle’ has height AM = h + Tra/2. Now use AC2 =
AM2 + MC2 to calculate AC.

5. Given ten points inside a circle of diameter 5, prove that the distance
between some pair of the given points must be less than 2.
180 Hints and outline solutions

(1) This may look tantilisingly familiar, yet different! If you have ever seen
anything like it before, you will expect to have to use the pigeonhole principle'.
if N + 1 points are arranged in N boxes, then some box must contain at least
two points. In this case you are told that there are ten points in a circle of
diameter 5: so if you cut up the circle into nine parts, some part must contain
at least two points. Unfortunately, this will only solve the problem if you
choose the parts so that they all have ‘diameter’ < 2 (since then the distance
between the two points which end up in the same part will be less than 2).
(2) Unfortunately, the most obvious ways of cutting a circle of diameter 5
into nine pieces don’t work. This doesn’t mean that the idea is useless—only
that you must use it more imaginatively. The ‘obvious’ ways of cutting a circle
into nine pieces tend to assume that all the pieces are more or less the same
shape. For example, if you treat the circle as a cake and cut it into nine equal
‘slices’, you face the problem that each slice has ‘diameter’ § (equal to the
radius of the circle), and § (> 2) is too large.
(3) You need an idea. Suppose that you start by cutting out a circle of radius
r = __ from the centre of the large circle. If two points were to end up inside
this small circle, the distance between them would certainly be less than 2r
(so you want 2r < 2). You now need to cut up the outer ring (between this
small circle and the large circle) into eight pieces, and to check that each of
these pieces has ‘diameter’ less than 2. Checking the diameter of each piece
is easiest if all the pieces have exactly the same shape; so if you want an easy
life, there is really only one way to cut up the outer ring!

6. Consider the equation

(*) v/(2 p + 1 -x1 2 (**)) + A3 x+p + 4) = Ax2 + 9x + 3p + 9)

in which x and p are real numbers and the square roots are to be real
(and non-negative). Show that if x and p satisfy (*), then

(x2 +x -p)(x2 + 8x + 2p + 9) = 0.

Hence find all real numbers p for which (*) has just one solution x.

(1) It is always worth hesitating briefly before charging ahead and squaring
both sides of an equation of the form {a + yfb ={c, since this can some¬
times lead one into an algebraic quagmire. However, in this instance you
should be able to see that the RHS of the resulting equation

(**) 2({a -y/b) = c - b - a

simplifies rather nicely.

 18th BMO, 1982 181

(a) Square both sides of the given equation and rearrange it into the form

(b) Now square both sides of (.»), take all the terms to one side and factorize
into the required form.

(2) You now know that, if x satisfies the original equation, then one of the
two factors (x + x-/?)or(x2 + 8x + 2/? + 9) must be zero: for each given
value of p this gives *ou* possible values of x. Unfortunately, some of these
pairs p, x do not solve the original equation, since squaring introduces
spurious solutions. However, you have no real choice but to suppose that p is
given, and to examine the four possible values of x in detail.

(a) Suppose that p is given. Write down the two roots xx and x2 of
* + x-p = 0, and the two roots x3 and x4 of x2 + 8x + 2p + 9 = 0.

(b) If x satisfies x2 + 8x + 2p + 9 = 0, then x2 = -(8x + 2p + 9). Use this to

show that 2p + 1 - x2 = -2(x + 2)2 < 0, and that 3x + p + 4 =
-(jXx + l)2 < 0, and that x2 + 9x + 3p + 9 = (2x + 3)2 ^ 0. Hence show
that, no matter what the value of p may be, x3 and x4 can never be
solutions of tiie original equation.

(c) Let Xj be the larger of xx and x2. If x satisfies x2+x-p = 0, then x is

real if and only if p >_. Moreover, p =x2 +x; use this to show that
2p + l-x2 = (x + l)2^0, that 3x +/? + 4 = (x + 2)2 > 0, and that x2 +
9x + 3p + 9 = (2x + 3)2 > 0. Hence, given any value of p ^ —(|), xx is
a solution of the original equation provided that \xx + 1| + \xx +2\ =
\2xx + 31, and x2 is a solution of the original equation provided that
|x2 + 1| + |x2 + 2| = |2x2 + 3|.

(d) Show that |xj + 1| + |xa + 2| = |2xj + 3| for each p > —(|).

(e) Show that |x2 + 1| + |x2 + 2\ — |2x2 + 3| provided that p>2, or —(|) <
p < 0.
(f) Conclude that, if p is given, then the original equation has a unique
solution x if and only if either x=Xj =x2 and p = -(}), or x=xx and
0 < p < 2.

18th British Mathematical Olympiad, 1982

1. The convex quadrilateral PQRS has area A; O is a point inside PQRS.

Prove that if
2 A = OP2 + OQ2 + OR2 + OS2,

then PQRS is a square with O as its centre.

182 Hints and outline solutions

(1) Faced with an unusual formula like this one, it is all too easy to lose sight
of the information that is given and to go round and round in circles. The
RHS could arise in a number of ways (from Pythagoras, from the cosine rule,
or even from squaring algebraic expressions involving OP, OQ, and so on).
You are given two major clues.

(a) First, you are told that the RHS is equal to ‘2A’. Thus you should
perhaps begin by finding a suitable expression for the area of the
quadrilateral that involves the four lengths OP, OQ, OR, and OS. Do this
(there is only one real choice).

(b) Second, you are told to prove that the given formula implies that the
quadrilateral must be a square with O at its centre. Thus, part of your
conclusion must be that OP = OQ = OR = OS\ that is, that the differ¬
ences OP - OQ, OQ - OR, OR - OS, and OS - OP are all equal to_.
Combined with the form of the RHS, this suggests that it might be
enough to prove that

(*) (OP - OQ)2 + (OQ - OR)2 + (OR - OS)2 + (OS - OP)2 = 0.

(2) You know that the LHS of (*) is >0. You also know from the formula in
the question that, when you expand the terms on the LHS of (*), the sum of
the squared terms 2(OP2 + OQ2 + OR2 + OS2) is equal to _. Now com¬
bine the expression for A from (lXa), and the fact that 0 < sin 6 < 1 when¬
ever 0 < 6< 180°, to deduce that the sum of the mixed terms 2OP OQ +
20Q 0R + 2OR OS + 2OS OP in (*) is >_. Conclude that the LHS of
(*) is also <0. Use this to complete the solution.

2. When written in base 2, a multiple of 17 contains exactly three digits 1.

Prove that it contains at least six digits 0, and that if it contains exactly
seven digits 0, then it is even.

(1) Let N be such an integer. Then N = 2a + 2b + 2C, with a> b> c; N is

divisible by 17 if and only if N' =N/(2C) = 2° + 2b + 1 is divisible by 17
(where a' = a — c, b' = b — c).

(a) How do powers of 2 behave (mod 17)? Write out the sequence of
remainders (mod 17) for 2° = l,2\22,... . Explain why 28+l has the same
remainder as 2'.

(b) Show that the only way in which 2° + 2b + 1 (with a' > b' > 0) can be a
multiple of 17 is if the three remainders ‘2a', 2b', 1 (mod 17)’ are (i) ‘8, 8,
V or (ii) T, —2, V (or ‘—2, 1, T). In the first case, a' = 8k + 3 and
b' = 8 m + 3 (with k > m > 0), so N' has at least 12 digits, and hence at
least nine zeros. In the second case one of a' and b' has the form 8k
with k > 1, so N' has at least *i*e digits, and hence at least *i* zeros.
 18 th BMO, 1982 183

(c) Conclude that if TV = 2a + 2b + 2C has exactly seven digits zero, then N'
cannot be of the form (bXi). Thus N must be even, with N' = N/2 of the
form (bXii).

(2) Alternatively, you may observe that 17 = 10001 (base 2). Let N = 17m. If
m < 16, then m =wxyz (base 2) must have four or fewer binary digits; but
then 17m = wxyzwxyz (base 2) would have an even number of l’s. Thus
m > 16, so 17m >-has at least *i*e digits—exactly three of which are l’s;
hence 17m has at least *i* zeros. This idea can be sharpened to complete the
second part.

111 1
3. If sn = 1 + 2+^ + ^ + -- - + ~ and n > 2, prove that

n(n + l)a — n < sn < n — (n - 1 )nb,

where a and b are given in terms of n by an = 1 and b(n - 1) = -1.

(1) Some problems are excellent Olympiad problems because there are many
possible approaches, all of which require the solver to demonstrate ingenuity.
Other problems challenge the solver to find the one approach that is likely to
work. This problem is of the latter kind—although you should not need
divine inspiration to find the right method. The problem involves inequalities,
and the key ingredient sn is, by definition, a *u*. If you consider the first
inequality to be proved, namely

n(n + 1— n < sn,

the main term on the LHS is a **o*u**. Moreover, the product on the LHS
involves nth roots (since a — 1 /n), and you have to prove that this product is
less than the sum sn + n. All of this should suggest strongly the need to use
the _-_ inequality.
(2) The fact that the term (n + l)a involves nth roots suggests further that
the sum sn + n should be rewritten as a sum of n (rather than n + 1) terms;
the only obvious way of doing this is to share out the extra term n equally
between the n terms of the form 1 /r, so that each becomes (1 /r) + 1 =_.
And the idea of using the AM-GM inequality suggests that the factor n in
the product on the LHS should be taken to be the other side so that
(sn + n)/n becomes the arithmetic mean of these n terms.
Everything now falls miraculously into place. The n terms in the sum are
all different, so the AM-GM inequality becomes a strict inequality; and the n
rewritten terms (1 +r)/r in the sum are such that their product cancels
beautifully, giving the required inequality.
184 Hints and outline solutions

(3) Now do something similar for the sum ‘(n - sn)/(n - 1)’ and for the
product nb to prove the second inequality.

4. For each choice of real number ux, a sequence uvu2,u3,... is defined

by the recurrence relation u3n = un_1 + |f (n > 2). By considering the
curve x3 =y + |§, or otherwise, describe, with proof, the behaviour of
un as n tends to infinity.

(1) You have to ‘describe the behaviour of the sequence {«„} as n tends to
infinity’. ‘Behaviour’ here presumably means: Does the sequence tend to a
(finite) limit? Does it diverge to +°° or to — °°? Or does it wobble about? In
general, one expects the answer to depend on the starting value uv For
example, the sequence defined by the recurrence relation xn = 2/xn_x con¬
verges to v^2 if “i>0, and converges to —y/2 if u\ < 0-
(2) In this problem the recurrence relation u3n = un_ j + is such that it is
difficult to investigate exactly what happens for particular starting values (no
calculators!). One way to get started is to suppose that (for some choice of
Uj) the sequence uvu2,u3,... tends to a limit, x say, and to try to calculate
the possible values of x.

(a) If the sequence converges to x, then, as n increases, both un and un_x

tend to x. Thus x satisfies the equation x3 = x+ |f. Use the Remainder
Theorem to find one root a of this equation. Hence factorize the
expression f{x)=x3 — x— |§ as a product of one linear factor and one
quadratic factor. Then find the two other roots f3 and /3' (>3 < 0 < /3').
Hence there are at most three different limits for the sequence {«„}.

(b) Calculate the first four values ux, u2, u3, and u4 of the sequence:

(i) when u1 = a; (ii) when ux = (3; (iii) when ux = fi'.

Conclude that the limits a, (3, and /3' can all occur, so there are exactly
three different possible limits depending on the starting value ux.

(c) Try to obtain a rough idea of how the sequence uv u2, u3,... behaves (i)
when w, = -2, (ii) when ux = -1, (iii) when ux = 0, (iv) when ux = 1, and
(v) when ux = 2.

(3) You should now be in a position to use ‘the curve *3 =y + ||’ to give a
complete analysis of how the sequence behaves.

(a) Make an accurate sketch of the curve x3 =y + |f. On the same pair of
axes, draw the line y = x.

(b) Choose an initial value ux for y and draw the horizontal line y = ux until
it cuts the curve (at x = u2). Then draw the vertical line x = u2 until it
 18th BMO, 1982 185

cuts the line y = x (at the point (_,_)). Then draw the horizontal
line y — u2 until it cuts the curve (at x —_). Continue in this fashion.
(This procedure is very like the Newton-Raphson procedure for finding
approximate roots of an equation.)

(c) What happens if your initial value ux is less than y8 ? What happens if
0 < ui < “? what happens if a<u1<f3’l What happens if f3' < ux?

(4) You should now think that you know the answer to the problem. So all
that remains is to find some way of presenting the proof.

(a) Suppose that ux < (3. Prove: (i) un < (3, for all n> 1; (ii) un_1<un for all
n > 1. Thus {un} is an increasing sequence, bounded above by (3, and so
must converge to some real number < f3. (To conclude that the sequence
converges to /3, you must appeal to what you proved in (2Xa); namely,
that there are only three possible limiting values for {un}, and that only
one of these is < (3.)

(b) Suppose that /3<ux<a. Prove: (i) (3<un<a, for all n> 1; (ii) un_x <
u . . for all n > 1. Conclude that {«n} converges to a.

(c) Now do something similar when a <ux < (3', and when /3' < ux.

5. A right circular cone stands on a horizontal base of radius r. Its vertex

V is at a distance / from each point on the perimeter of the base. A
plane section of the cone is an ellipse with lowest point L and highest
point H. On the curved surface of the cone, to one side of the plane
VLH, two routes, Rx and R2, from L to H are marked: Rx follows the
semi-perimeter of the ellipse, while R2 is the route of shortest length.
Find the condition that Rx and R2 intersect between L and H.

(1) Draw a decent diagram of the cone and the ellipse, marking the points V,
L, and H. Draw another diagram of the cone cut along VH and laid flat,
marking the points V, L, and H and the line VL. Mark the shortest route R2
(— HL) on the ‘flattened’ cone.
The particular diagrams that you have drawn may tempt you to jump to the
conclusion that the two paths Rx and R2 can never intersect! However,
diagrams can mislead. There are a number of parameters here: the inclina¬
tion of the plane can vary—thereby changing the relative positions of H and
L; and the angle of the flattened circular sector—after cutting the cone along
VH—can vary from almost 0° to nearly 360°.
(2) What angle does the semi-perimeter Rx of the ellipse make at the point
H with the generator VH? What angle does Rx make at L with the
generator VL?
Suppose that the plane section is perpendicular to the axis of the cone.
186 Hints and outline solutions

Then all points of Rl are equidistant from V. Hence, in.the diagram of the
‘flattened’ cone, R1 is the circular arc LH centred at V, so the paths Rl and
R2 do not intersect between L and H.
Thus you may assume that the plane is genuinely tilted and that L is below
H. Let X be an arbitrary point on the semi-perimeter Rx of the ellipse. As X
moves from H to L along Rv what happens to the length VX?
(3) On the flattened cone, the ellipse Rx sets out from L perpendicular to
the line VL] that is, Rx starts out from L on the opposite side of LH (= R2)
from V.
(a) Suppose that Rl also sets out from H on the opposite side of LH from
V. Then the two paths Rx and R2 must cross an e*e* number of times. A
complete solution should prove that, in this case, the two paths do not
cross at all. However, this is not obvious and depends on a tricky
calculation! (Let O be the centre of the base; let VL, VH, and VX
meet the base of the cone, at L', H', and X'. Take OV and OH' as
the z- and y-axes, and let C be the point on the circular base such
that OC is the x-axis. Thus the co-ordinates of X' in terms of
LCOX' = </> are (rcos </>, rsin <£,0). Show tha: the co-ordinates of the
point X (in terms of $ and the distance px of X from the z-axis OV) are
(pcos 4>, psin cf), OV - pcot a), where sina = r/7. Write VX = R and
find the polar equation of Rj in the plane of the flattened out cone.
Hence show that Rx can have at most one point of inflexion, and so
cannot cross LH more than once.)
(b) Suppose next that Rx sets out from H on the same side of LH as V; that
is, suppose that in the flattened cone, Z.VHL> 90°. Then Rx must at
some point cross the line segment R2. In the flattened cone show that
L.HVL = Ttr/l\ hence rewrite the condition L VHL > 90° as an inequality
VH < VL cosinr/l).

6. Prove that the number of sequences ax, a2,..., an, with each a, = 0 or 1

and containing exactly m occurrences of ‘01’, is | 2m+ \ )'

(1) Combinatorics problems can often be solved very simply, but the easy
solution almost always comes after one has struggled to find a simple-minded
but messy solution!
The most simple-minded approach here is to count the number of se¬
quences for each possible position of the first occurrence of ‘01’, and then
add the results.
(a) Let /(n; m) denote the number of possible sequences of length n with
exactly m occurrences of ‘01’. If the first ‘01’ occurs in the first two
 18th BMO, 1982 187

positions, then there must be exactly _ occurrences of ‘01’ in the

remaining - positions, so there are exactly f(n - 2; m - 1) possible
sequences of this type. (Strictly speaking, to prepare for the general case,
you should say that there are exactly /(0; 0) ways of filling the first 0
positions and f(n - 2; m - 1) ways of filling in the last n - 2 positions,
and hence /(0; 0)/(n - 2; m - 1) ways altogether—where one clearly
needs the convention /(0; 0) = 1.)
If the first occurrence of ‘01’ is in the second and third positions, then
there are exactly /(1; 0) ways of filling the first position, and /(_;_)
ways of filling the last n- 3 positions, and hence /(l; 0)/( ; )
ways altogether.

(b) Repeat this for the general case (in which the first occurrence of ‘01’ is in
the ith and (i + l)th positions). Hence derive the recurrence

n-2
f(n;m) £ /(i;0)/(n -1 - 2;m - 1).
i=0

(c) Finally, check that f(n; 0) = n + 1 = (” | 1 j for every n and then use
induction to change the RHS into a familiar expression involving bino¬
mial coefficients (see 1985, Question 3).

(2) A variation on (1) is to partition the collection of all possible sequences

into just two types: (a) those that end with 0, and (b) those that end with 1.
Let A(n; m) be the number of sequences of type (a) which have length n with
m 01’s, and let B(n; m) be the number of sequences of type (b). Thus

/(n; m) =A(n; m) + B(n;m).

(i) Explain why A(n; m) =/(n - 1; m).

(ii) Show that B(rv, m) — f{n — 2; m — 1) + B(n — 1; m).
(iii) Guess an expression for B(n; m) =_ (as a binomial coefficient), and

prove both B{n\ m) =_ and f(n; m) = j 2m + l) s*mu^taneous^y by

induction.
(3) An alternative approach is to reformulate the question in some way to
explain the simple form of the answer 12m+ \)' sequence av a2,...,an
has not only n ‘positions’, but also n + 1 ‘gaps’ between terms—including the
two ends. Thus it looks as though you have to explain why the number of
sequences with exactly m 01’s is the same as the number of ways of choosing
2m + 1 of these ‘gaps’.
Once you have had this idea, it should not take long to realize that each
188 Hints and outline solutions

occurrence of ‘01’ involves a jump up (from 0 to 1), and that before the next
occurrence of ‘01’ the terms of the sequence must jump back down (from 1 to
0). Thus each sequence with exactly m occurrences of ‘01’ is uniquely
determined by specifying the 2m + 1 gaps (chosen from the n + 1 possible
gaps) at which the sequence ‘jumps’—with the sequence remaining constant
between successive selected gaps: there are exactly j 2m + \) suc^ se0uences
—provided that one specifies that all sequences are deemed to start with a
string of 0’s unless the very first ‘gap’ (to the left of the first term) is among
those chosen, in which case the sequence jumps from 0 to 1 before it starts,
and so starts with a string of l’s.

17th British Mathematical Olympiad, 1981

1. The point H' is the orthocentre of triangle ABC. The midpoints of BC,
CA, and AB are A', B', and C' respectively. A circle with centre H
cuts the sides of triangle A'B'C' (produced if necessary) in six points:
Dj and D2 on B'C'; E1 and E2 on C'A'; and Fx and F2 on A'B'.
Prove that ADX =AD2 = BEl - BE2 = CF1 = CF2.

(1) This question involves so many points and lines that it is crucial to draw a
decent diagram—using ruler and compasses (and a sharp pencil).
(2) Half of what you have to prove is straightforward. Let AH meet B'C' at
the point L. Since H is the centre of the circle, HDl = HD2 and LDX = LD2.
Moreover, the line DXD2 = B'C' is *a*a**e* to BC, and hence
*e**e**i*u*a* to AH. Thus ADl =AD2 (Why?). Similarly, BEX = BE-, and
CF1 = CF2.
(3) It remains to show that ADx — BEV This is not obvious, and requires a
calculation (using Pythagoras, or vectors).
(a) Let BH meet CA' at M. Then AL2 + LC'2 = AC'2 = BC'2 = BM2 +
MC'2 (since C' is the *i**oi** of AB).
(b) Hi} + LD\ = HD\ = HE2 = HM2 + ME2 (since H is the *e***e of the
circle through Dx and Ex).

(c) HL2 + LC'2 = HC'2 = HM2 + MC'2 (since ^HMC' = Z.HLC' = 90°).
(d) Adding (a) and (b) and subtracting (c) gives

(AL2 + LC'2) + (HL2 + LD2) - (HL2 + LC'2)

= (BM2+MC'2) + (HM2+MEl) - (HM2 + MC'2).

.-. AD2 =AL2 + LD2 = BM2 + ME2 = BE2.

 17th BMO, 1981 189

A similar calculation shows that BE1 = CFV

(4) Alternatively, let H be the origin of (vector) co-ordinates, and let a b
and c be the vectors for points A, B, and C. Then a-(b - c) = 0 (since AH is
*e**e**i*u*a* to BC); similarly, b(c - a) = 0, so a b = b e = ca = k (say).
Now let a', b\ and c' be the vectors for A', B', and C'. Then b' =_and
=-• Since Dl lies on B'C', its vector d, = tb' + (1 - t)c'. Moreover, D,
ies on die circle centre H—of radius r (say); thus d, • d,=r2. Now show that
\ADl\ - (dj - a)-(d! - a) = r2 - k. Since this depends only on r and k, the
result follows.

2. Given positive integers m and n, Sm is equal to the sum of m terms of

the series

(n + 1) - (n + 1X« + 3) + (n + lXn + 2Xn + 4)

— (n + 1X« + 2Xn + 3Xn + 5) + ,

the terms of which alternate in sign, with each term (after the first)
equal to the product of consecutive integers with the last but one
integer omitted. Prove that Sm is divisible by m\, but not necessarily by
m\{n + 1).

(1) This looks horrible, but it is only a simple question dressed up to look
difficult. Throughout, we assume that n is fixed.

(a) Clearly, Sj = n + 1. Calculate S2 (in fully simplified—that is,

factorized—form). Show that S2/2! is odd when n is odd, so that S2 is
not divisible by 2!(n + 1) when n is odd. (This observation answers the
second part of the question.)

(b) Now calculate S3 (in fully simplified form).

(c) State what you expect the simplified form of S4 to be. Then do the
necessary calculation to check that your guess is correct.

(d) You now want to prove that Sm =(-l)m+1(« + 1 )(n + 2)•••(« + m). This
is clearly true when m = 1. And if it is true for some m> 1, then

Sm + j = Sm + ( — l)m^ 2{n + l)(n + 2)•••(« + m)(n + m + 2)

= (—1 )m + 2(n + 1 )(n + 2) ••• (n + m)[— 1 + (n + m + 2)]

3. Let a, b, and c be any positive numbers. Prove that:

(a) a3 + b3 + c3 > a2b + b2c + c2a;
(b) abc > (a + b — cXb + c — a)(c + a — b).
190 Hints and outline solutions

(1) Harmless looking inequalities like the one in part (a) can be maddeningly
elusive: there are always so many more ways to fail than to succeed. The
simplest principle for inequalities depends on the fact that squares are
non-negative; yet here we have cubes (see (3) below). The LHS suggests using
the AM-GM inequality; but the form of the RHS means that one cannot use
this in the ‘obvious’ way (see (4) below).
(2) You may use the cyclic symmetry of the two sides to assume (without loss
of generality) that c (say) is the smallest of the three variables. However, it is
not at all clear that you can also assume that a ^ b (or b ^ a!).
One thing that you should certainly notice, and bear in mind all the time, is
that the inequality becomes an equality when a =_=_. Another is
the fact that the RHS and LHS are both homogeneous of the same degree
(that is, all terms are of the same degree); so you are free to scale the
variables to make one of them—c say—equal to 1.

(i) Put c = 1, b = 1 +y, and a = 1 +x (x,y > 0). Then expand the two sides
and collect up terms to show that the inequality to be proved is equiva¬
lent to proving ‘(x-y)2 +x3 + y3 +x2 + y2 -x2y > O’.

(ii) Observe that, if x >y, then x3 - x2y > 0, and if y >x, then y3 -x2y > 0.
Hence complete the proof of (a).

(3) Alternatively, since you know that a, b, and c are *o*i*i*e, the fact that
‘squares are ^ 0’ implies that a{a — b)2 + b(b — c)2 + c(c — a)2 ^ 0. Unfortu¬
nately, the LHS of this inequality involves unwanted terms such as ab2.
However, you should be able to find a positive integer k such that
(ka + bXa - b)2 does not involve ab2. Then add three such terms to com¬
plete the solution.
(4) A third solution to (a) comes from pursuing the AM-GM idea more
tenaciously. The LHS certainly invites you to use the AM-GM inequality:
(x+y + z)/3 > /xyz. Since each application of AM-GM will give a single
term on the RHS, you must clearly use this idea ***ee times—first to obtain
a2b on the RHS, then to get the other two terms. Which three cubes x, y,
and z should you multiply together if their cube root is to equal a2b (that is,
the product of the three cubes must be a6b3)? Now rewrite the LHS in the
form (a3 +a3 + b3)/3 + (b3 + b3 + c3)/3 + (c3 + c3 + a3)/3 and complete
the proof.
(5) One approach to part (b) is to multiply out the three brackets on
the RHS and collect all terms on the LHS to show that the problem is
equivalent to proving that ‘3abc + a3 + b3 + c3 — (a2b + b2c + c2a) — (azc +
b2a + c2b) O’. One can make this approach work, but it has to be handled
carefully. (It is tempting to apply part (a) to show that the LHS is >3 abc +
a3 + fr3 + c3 - (a2c + b2a + c2b), and then to try to show that this simpler
expression is ^0: unfortunately, this expression can be negative!)
(6) A much nicer approach is to assume (without loss of generality) that
 17th BMO, 1981 191

a>b and a > c, and to observe that then a+ b-c> 0 and c + a-b>0. If
b + c - a < 0, then the RHS is also < 0, so the inequality holds. Thus you may
assume that b + c-a> 0, and let a + b - c =x, c + a - b =y, and b + c-
a~Z.' Then (;c+>’)/2 =-, (y+z)/2 =_, and (z+x)/2 =_, and
t“^.inequality to be proved reduces to ((jc +y)/2)-((y + z)/2)-((z +x)/2) >
yfxy y/yz • yfzx =xyz, which follows from three applications of the AM-GM
inequality.

4. Given n points in space such that no plane passes through any four of
them, let S be the set of all tetrahedra the vertices of which are four of
the n points. Prove that a plane which does not pass through any of the
n points cannot cut more than n2(n - 2)2/64 of the tetrahedra of S in
quadrilateral cross-sections.

(1) It is often true that the worse a problem seems, the simpler it will be once
you have made the effort to understand what is is saying. Imagine four points
forming a tetrahedron in three dimensions. A plane which cuts off just one
corner of the tetrahedron creates a **ia**u*a* cross-section. What can you
say about a plane which cuts the tetrahedron in a quadrilateral cross-section?
Where must the plane go?
(2) This (and one simple inequality) is all you need to solve the problem.
(a) Any plane which does not pass through any of the n points separates the
n points into two sets—with i (say) on one side and_on the other
side. Such a plane cuts a tetrahedron formed by four of the n points in a
quadrilateral cross-section precisely when **o of the four points are on
one side of the plane, and the other **o are on the other side. Since
there are
a)
_ ways of selecting two points on one side and
(V)
of selecting two points on the other side of the plane, the number of
ways

tetrahedra cut by this plane in quadrilateral cross-sections is exactly

_X_.

(b) Now i(n — i) < ([/' + (n - i')]/2)2 = (n/2)2 (by the AM-GM inequality);
similarly, (i - l)(n - i — 1) <_ . Hence | ^) (n 2 * j <

5. Find, with proof, the smallest possible value of |12m - 5"|, where m and
n are positive integers.

(1) This problem is unusual and may throw you at first; but it is an excellent
challenge.
192 Hints and outline solutions

(a) When m=n = 1, |12m-5"| =_. Thus you know that the smallest
possible value is at most _.
(b) Since 12 is e*e* and 5 is o**, the difference |12m - 5"| is always o**. This
reduces the possible smallest values to just *ou*; namely _, _, —,
and 7.
(c) Extending the argument in (b), you know that 5" is always a multiple of
_, whereas 12m is never a multiple of _. Hence the difference
|12m - 5"| can never be a multiple of . Similarly, 12m is always a
multiple of 3, but 5” is never a multiple of 3, so |12m - 5"| is never a
multiple of 3.
(d) Hence the smallest possible value of |12m — 5"| is either _ or 7. You
therefore have to decide whether you should try to find values m and n
with |12m - 5" | = 1, or try to prove that none exist. Unless you can spot
suitable values quickly, it makes sense to try to prove that none exist—for
in the process of trying to prove that no values exist, you will find out
more and more about what properties such values would have if they did
exist.
(e) Suppose that |12m - 5"| = 1. Then 5" = ±1 (mod 12), and 12m =
±l(mod5). Show that 5” = -1 (mod 12) has no solutions, and that 5" s=
1 (mod 12) precisely when n is e*e*. Show also that 12m = +l(mod5) if
and only if m is e*e*. Thus m = 2i and n = 2j for some positive integers
i and j. But then ±1 = 12m - 5” = 122' - 52' = (12')2 - (5')2 is a differ¬
ence of two **ua*e*. However, the only squares which differ by 1 are_
and _. Use this to conclude that |12m - 5"| # 1.

6. Given distinct non-zero integers a, (1 < i < n), let p, be the product of
all the factors (a, — a,), (a, — a2),...,(a, — an) except for the zero factor
(u, — a,). Prove that if k is a non-negative integer, £"=itff/P; is an
integer.

(1) This looks even worse than Question 2 above. But stay cool, calm and
collected! To show that the given expression ‘is an integer’, you clearly have
to put everything over a common denominator and explain why the denomi¬
nator D must divide the numerator N. This does not look inviting, but it is
not as bad as it may seem. However, it does require a willingness to treat the
symbols u, as ‘algebraic indeterminates’ or variables (rather than as integers
with particular values), and to treat the expressions for D and N in the
question as polynomials in these variables (rather than as integers), with
‘division’ meaning ‘polynomial division’.
(2Xa) Suppose that you have to write the sum in the question as a rational
expression with a common denominator. The simplest denominator
to choose is just D = plp2 ••• pn. (You could try taking
 16th BMO, 1980 193

"\[{S D Pi Pi Pn) 5 since every bracket (a, — al) occurs ex¬

actly twice, with opposite signs; however, square roots are algebraically
messy.) The first two terms in the numerator N when the expression
)/Pi\ is written with common denominator D are

(*) ^PiPiPa "• Pn) + ai(P\PzPA ••'/?„) + ... •

Write out the next two terms.

(b) Each factor in the product pj has the form (aj — cij) for some j # 1; in
particular, each factor involves ax, and one of these factors is (al — a2).
Similarly, each factor in the product p2 involves a2—and one of these
factors is (a2 — a2). No other product pk involves the factor (ax — a2).
(c) Clearly, each term after the first two in (*) is divisible by the product
Pi Pi> an^ hence by (at — a2)2. On the other hand, the sum of the first
two terms in (*) can be written as

P3P- ■■■ Pn^iPi+aiPi)

and

aklpi + ak2pl

= (a2 - ax)[a\(a2 - a3) ••• (a2 - an) - ak2(ax - a3) ••• (ax - an)].

The expression in square brackets on the RHS vanishes if ax = a2;

thus if you view this expression as a polynomial in ax, then it must
have (flj — a2) as a factor. Hence the numerator N is divisible by
(ax - a2)2.

(d) Similarly, N is divisible by (a, — ay)2 for each pair i, j. Hence, if you
treat the symbols ava2,...,an as independent variables, and D as a
polynomial in these n variables, then D divides N (as polynomials). If
you then substitute the given integer values of ava2,...,an in the
quotient N/D you obtain the (integer) quotient—hence proving that
N/D is an integer.

16th British Mathematical Olympiad, 1980

1. Prove that the equation xn +yn =zn, where n is an integer > 1, has no
solution in integers x, y, and z with 0 <x < n and 0 <y < n.

(1) This is a very special case of ‘Fermat’s Last Theorem’. The restrictions
‘0 Cx < n, 0 <y < n’ must be important, but at first sight it is not at all clear
194 Hints and outline solutions

how they are meant to help. However, you should eventually stumble on the
following idea.

(*) Suppose (without loss of generality) that x <y < n, with n > 2.
Then y<z,soz^y + l.

Hence

xn +yn + yn + n-yn~l < {y + 1)” <zn,

so there are no solutions.

2. Find a set S of seven consecutive positive integers for which a

polynomial P(x) of degree 5 exists with the following properties:
(a) all of the coefficients in P(x) are integers;
(b) P(n) = n for five numbers ne5, including the least and the
greatest;
(c) P(n) = 0 for some n e S.

(1) You are given certain properties of the polynomial P(x), but to deter¬
mine a polynomial you would prefer to know its *oo**. Thus it makes sense
to switch attention to the polynomial Q(x) = P(x) —x. This new polynomial
Q(x) has integer roots a, (3, y, 8, and s, where a</3<y<8<e=a + 6,
and satisfies Q(a) =_, where a is one of the two integers a and a'
different from (3, y, and 8 which lie between a and e. Hence

Q(x) =A(x - a)(x — /3)(x — y)(x— 8)(x- e),

so Q(a) —A(a — a)(a — (3)(a — y)(a — S)(a — e) = —a.

Since a lies between a and e = a + 6, the possible values of the factors

a - a, a - /3, a - y, a - 8, and a - e are highly restricted (being distinct
integers between —5 and +5). For example, if a = a + 3 and a' = a + 1, the
second condition becomes Q(a) =A - 3* 1 •(—IX—2X—3) = —a. Thus you may
choose >1 = 1 and a = 18, whence S = {15,16,17,18,19,20,21} and P(x) =
Q(x)+x is a monic polynomial. [Alternatively, you may prefer to choose
A = 18 and a = 1, whence S = {-2, -1,0,1,2,3,4}.]

3. Given a semi-circular region with diameter AB, P and Q are two points
on the diameter AB, and R and S are two points on the semi-circular
arc such that PQRS is a square. C is a point on the semi-circular arc
such that the areas of the triangle ABC and the square PQRS are
equal. Prove that a straight line passing through one of the points R
and S and through one of the points A and B cuts a side of the square
at the incentre of the triangle.
 16th BMO, 1980 195

(1) Let the radius of the circle be 1; and let Q lie between P and B.

(a) Use Pythagoras to find the side length of the square PQRS inscribed in
the semicircle.

(b) Let CX be the perpendicular from C to AB. Equating areas shows that
(2/i/5 )2 = CX. Thus C lies between B and R (or between A and 5).

(c) BS passes through the incentre of triangle ABC if and only if BS is the
*i*e**o* of L ABC. Thus it suffices to show that the chord AS has the
same length as SC. Use Pythagoras to show that each of these chords
has length yjl - (2/yf$).

4. Find the set of real numbers a0 for which the infinite sequence {«„} of
real numbers defined by an+1 = 2" — 3an (n > 0) is strictly increasing.

(1) This is considerably tougher than the previous three questions. One
approach is to find a closed formula for the nth term an. While it is clear that
this will involve powers of 2 and 3, a beginner could be excused for not
guessing that the general term an can be expressed in the form an = A-2n +
B(-3)n.
(a) Use the recurrence relation to show that, if an =A -2” + B-(-3)n, then
A=_.
(b) Use Oj = 2° — 3a0 to show that, if an=A-2n + B-(-3)n, then B =_.

(c) Prove (by induction) that a„ = A ■ 2n + B • (-3)" for these values of A and
B. Conclude that the sequence {an} is increasing if and only if a0 =_.

(2) A more direct approach is to grind out successive values and to look for a
closed form which depends directly on a0. When doing this it is important not
to multiply everything out, but to preserve the form of each term so that you
can see what exactly is going on.

(a) flj = 2° - 3\; a2 = 21 - 3(2° - 3a0) = [21-3° - 31-2°] + 32a0. Write out
similar expressions for a3 and a4.

(b) Write out the corresponding expression for an+1.

(c) The constant term ‘2"-3° — 2"-1 -31 + ... +(—1)"3"-20, is a GP with
common ratio r =_. Write down the formula for its sum. Simplify your
expression as much as possible. Hence write down a closed form expres¬
sion for the nth term
196 Hints and outline solutions

(d) Suppose that an_1 < an < an+1. Use your closed form expression for an to
show that, if n is odd, the inequality an_1 <an implies that

2” + 3" 2n~1 - 3n~1

5 5
ao <
3"_1-4

<an + 1 implies that

2” + 3'n 2n+1 _ 3"+1

5 5
a0>
3”-4

As n tends to infinity, the powers of 3 dominate, so the RHS of each

inequality tends to _. Thus there is only one possible value of a0
which gives rise to an increasing sequence {a„}; namely, a0 =_. Check
this starting value to show that it does indeed generate an increasing
sequence.

5. In a party of ten people, you are told that among any three people there
are at least two who do not know each other. Prove that the party
contains a set of four people, none of whom knows the other three.

(1) Denote the ten people by the vertices of a network, or graph, joining two
people precisely when they are acquainted. Given any vertex x, the first
condition guarantees that none of the vertices joined to x can be joined to
each other. If some vertex x were joined to four or more other vertices, you
could choose any four of the vertices joined to x to obtain ‘four persons none
of whom knows the other three’.
(2) Thus you may assume that each person knows at most three people. The
quickest way to finish the solution is then to observe that, given any vertex x,
there are at least _vertices not joined to x. If we choose six vertices not
joined to x, then by a well known result, among these six vertices there are
either three vertices forming a triangle (that is, ‘all of whom know each
other’), or three vertices with no edges between them (that is, ‘no two of
whom know each other’). The first possibility is ruled out by the first
condition in the question. Hence there must exist three vertices u, v, and w,
not joined to x, no two of which are joined to each other—giving a set
[u, v, w, x) of ‘four persons, none of whom knows the other three’.

15th British Mathematical Olympiad 1979 ,

1. Find all triangles ABC for which AB+AC = 2 and AD + BC = i/5,
where AD is the altitude through A, meeting BC at the point D.
 15th BMO, 1979 197

(1) One of the difficulties in this problem is the fact that so many things are
varying. So why not start by fixing the points B and C, and hence the length
BC = a (where a <AB + AC = 2). The second condition then becomes

AD = J5 — a,

and so A lies on a line *a*a**e* to BC and distance v^5" — a away from it.
Since B and C are fixed, the first condition now says that A lies on an
e**i**e with B and C as *o*i.

(a) Use the given information to write down the length _ of the major
axis of the ellipse.

(b) Show (using Pythagoras, or the eccentricity of the ellipse) that the minor

axis has length 2yJ12 - (a/2)2. Conclude that AD < yj l2 ~(a/2)2.

(c) Finally, use the condition AD = \/5 - a to show that (a - (4yT/5))2 < 0.

Hence deduce that there is exactly one triangle satisfying the conditions in
the question.

2. From a point O in three-dimensional space three given rays, OA, OB,

and OC, emerge, with ZBOC = a, ZCOA = (3, and ZAOB = y,
0 < a, (3,y < n. Prove that, given 2s > 0, there are unique points X, Y,
and Z on OA, OB, and OC respectively such that the triangles YOZ,
ZOX, and XOY have the same perimeter 2s. Express OX in terms of s,
sin(a/2), sin( /3/2), and sin(y/2).

It helps to draw a diagram that looks

three-dimensional and that helps you see
what is going on. What exactly must you
do to prove that ‘given 2s > 0, there are
unique points X, Y, and Z, and so on?
Presumably you have to take arbitrary
points X, Y, and Z; find expressions for
the perimeters of triangles YOZ, ZOX,
and XOY; and then equate these expres¬
sions and show that there is a unique
‘solution’. Since XY can be expressed in terms of OX and OY (and y), a
good first move (in the spirit of Name and Conquer) is to let OX = x, OY = y,
and OZ = z.

(a) Find an expression for XY2 in terms of x, y, and y.

(b) Suppose that triangle OXY has perimeter 2s. Find another expression for
XY2.
(c) Equate the two expressions, and show that (5 — x/s —y) =xy sin2(y/2).
198 Hints and outline solutions

(d) Write down the corresponding expressions for (s—yXs—z) and for
(5-zX^-x). Multiply the three expressions together and take the
(positive) square root of both sides to obtain an equation for
(s -xXs -yXs -z). (When taking square roots you should explain why
each of (s—x), (s-y), and (s - z) is positive, and also why sin(a/2),
sin( >3/2), and sin(y/2) are all positive.) Finally, combine this equation
with your equation for (s -yXs - z) to obtain an equation for (5 - x) in
terms of x, sin(a/2), sin(/3/2), and sin(y/2); hence show that ‘given
2s > 0, there is a unique’ value of x (and similarly a unique value of y
and z), so there are unique points X, Y, and Z as required.

3. S = {al, a2,..., an) is a set of distinct positive odd integers. The

differences |a, — a -| (1 < i <j ^ n) are all distinct. Prove that

E a, > \n(n2 + 2).

/= 1

(1) Order the integers so that al<a2< ... <an. You are told that the
integers are all o**, so you can be sure that each of the differences |a, - a;| is
e*e*. Moreover, there are exactly _ different pairs at, (1 <1 <7 <n).

Since you are told that no two differences are equal, these j ^ j even integers
must all be different, so the largest difference an — ax must be greater than or

equal to the I”)1*1 even *nte§er __ Hence

an>a1+ n(n - 1) > 1 + n(n - 1).

(2) You now know that ax > 1 and an > n(n - 1) + 1. You need a lower
bound for each at. Since the differences |a, - a-| are all different,

a2 > ax + (a2 - ax) > 1 + 2

(since a2 - ax > 2). Similarly,

a3 > ax + (a2 - ax) + (a3 - a2) ^ 1 + (2 + 4),

since a2 ~ ax and a3 - a2 are positive, different, and even. In general,

ar>ax + (a2 - ax) + ... +(ar-ar_x) > 1 + (2 + 4 + ... +2(r- 1)),

= 1 + r(r — 1).
Hence
n n

Y, ai> Y [1 + <(* - 1 )]>n + E i2 E *=

1= 1 1= 1 i= 1 1= 1
 15th BMO, 1979 199

4. The function / is defined on the rational numbers and takes only

rational values. For all rationals x and y,

^ f(x+f(y))=f(x)f(y)

Prove that / is a constant function.

(1) Observe first that, if /(*) = c is a constant function, then the LHS of the
given functional equation (*) is equal to _, while the RHS is equal to
- . Only two rational numbers c satisfy c = c2, namely, c =_ and
c = — • Thus y°u have to prove that the given functional equation, implies
either (a) fix) = 0 for all rational x or (b) fix) = 1 for all rational x.
(2) Suppose that fiy) = 0 for some rational y. Use (*) to show that then
fix) = 0 for all rational x, so (a) holds.
(3) Thus you may assume from now on that fiy) # 0 for all y. In particular,
/(0) = c # 0. It seems reasonable to try to prove first that c = 1, and then that
‘fix) = 1 for all rational x\
(i) Put x = 0 and y — 0 in (*) to find the value of /(c) in terms of c. Then
put x = c and y — 0 to find the value of /(2c). Use a similar method to
find fikc) ik e N).
(ii) You are told that ‘/ takes only rational values’. Hence c —p/q, for some
peZ, and some q <= N, so qc = p e Z. Thus /(p) =fiqc) = cq+1.
(iii) Put x — — c and y = 0 in (*) to find the value of /(—c).
(iv) Put x = 0 and y = -c in (*) to find the value of /(1). Then put x = 1 and
y=—c to find the value of /(2). Let n e N and suppose that you have
already proved that fik) = c for all integers k, -n<k^n; put x = ±n
and y = -c to show that fi~n) - c and fin + 1) = c.
(v) Combine (ii) and (iv) to show that either c =_, or q is e*e* and
c =_. Then use (iii) to show that c = — 1 is impossible.
(vi) Finally, let r = m/n, with meZ and neN. Suppose that fir) = d =
p'/q' g Q. Imitate your method in (i) to show that fid) = d, and that
fikd) = dk for all k e fU Conclude that 1 =fip') —fiq'-d) = dq , so that
d = ± 1. Then imitate your method in (iii) to show that fi~d)= l/d.
Hence show that d= 1.

5. If ft is a positive integer, denote by pin) the number of ways of

expressing n as the sum of one or more positive integers. Thus pi4) = 5,
as there are five different ways of expressing 4 in terms of positive
integers; namely,

1 + 1 + 1 + 1, 1 + 1 + 2, 1+3, 2 + 2, and 4.

Prove that pin + 1) - 2pin) + pin - 1) ^ 0 for each n > 1.

200 Hints and outline solutions

(1) Write out the pi3) = 3 ways of expressing n = 3 as a sum. Do the same
for n = 4 and n = 5. You have to find a reason which explains why ‘p(5) -
2p(4) + p(3) > 0’ and which works equally well for other values of n.
(2) It is tempting to try to partition the set of expressions for n + 1 into three
disjoint sets—two of which are easy to count and have sizes pin) and
pin) - pin — 1), in which case the third set need not be counted at all. The
following two observations make this approach look even more promising.

(a) Explain why there are exactly pin) ways of writing n + 1 as a sum in
which the smallest summand is a T.

(b) Use (a) to explain why the number of ways of writing n as a sum in which
the smallest summand is ‘>2’ is pin) — pin — 1).

Unfortunately, it is less clear how to fit these two tantalising facts together to
complete the solution. (Each expression for n as a sum can be extended to an
expression for n + 1 by adding a 1 at the front. Similarly, each expression for
n as a sum with smallest summand >2 can be extended to an exprssion for
n + 1 by adding a 1 at the front. Unfortunately, the second set of expressions
for n + 1 is a subset of the first, so all the expressions in (b) have been
counted twice, while expressions which do not start with aT have not been
counted at all.)
(3) If you look carefully at the failed idea in (2), you may see a way out. Let
pxin) denote the number of expressions for n with smallest summand T, and
let p2in) denote the number of expressions for n with smallest summand > 2.

(a) Explain why p^in)=pin - 1). Hence explain why the number of expres¬
sions for n + 1 that have at least two summands equal to 1 is equal to
pin - 1).

(b) Explain why p2in) = pin) - pin - 1). Explain why this is equal to the
number of expressions for n + 1 that have exactly one summand equal
to 1.

(c) Finally, if you take each expression for n which has smallest summand
>2 and add one to its largest summand, you obtain exactly p2in) =
pin) -pin - 1) expressions for n + 1 in which the smallest summand is
>2.

(d) The construction in (c) gives some, but by no means all, of the expres¬
sions for n + 1 with smallest summand > 2: for example, when n = 3, it
misses 2 + 2. However, the three sets counted in (a), (b), and (c) are
disjoint, so this gives the required inequality in the form

pin + 1) >pin - 1) + 2ipin) -pin - 1)).

 15th BMO, 1979 201

(4) Alternatively, you may notice that the inequality to be proved can be
written in the more natural form:

pin + 1) -pin) >pin) -pin - 1).

Thus you only need to show that, if qin) = pin) - pin - 1), then qin + 1) >
qin). By (2Xa), qin) is precisely the number of ways of writing n as a sum
without using any 1 s. To prove the inequality qin + 1) ^ qin) you only have
to show that there are at least as many ways of expressing n + 1 without using
any l’s as there are ways of expressing n without using any l’s. This follows
directly from (3Xc) (since each such sum for n can be changed into a sum for
n + 1 by increasing the largest term by 1).

6. Prove that there are no prime numbers in the infinite sequence

10001,100010001,1000100010001,....

(lXa) Any factorization of 10001 =aXb with a > b corresponds to writing

10001 as the difference of two squares [ia + b)/2]2 - [ia - b)/2]2.
Moreover, since a and b are both odd and congruent (mod 4), ia + b)/2
is odd and ia - b)/2 is even. Checking 1012,1032,1052,... one soon
finds that 10001 = 1052 -_2, giving the factorization 10001 =
_X_. Hence the first term is not prime.

(b) The second term has digit sum equal to _, is therefore divisible by
_, and so is also not a prime.

(c) The third term is visibly divisible by 10001. More generally, the
imn - l)th term in the sequence is divisible by the (m - l)th term,
and so is not prime (if m > 2).

(2) This feels like progress. But it leaves the (still infinite!) subsequence of
terms involving p l’s, where p is a prime ^5. So start again from scratch and
think about the form of the terms under scrutiny.

(a) Let un denote the nth term (with n + 1 l’s). Write the fourth term as a
sum of powers of 10: T + ... ’. Then use the formula for the sum of a GP
(with first term 1 and common ratio r =_) to obtain a closed formula
for the fourth term. Factorize the numerator as a difference of two
squares, and so explain why u4 is not prime.

(b) Use the same method to show that, in general, = [(10" + 1)4 — 1]/
[104 — 1]. Factorize the numerator as a difference of two squares. Hence
show that, when n > 2, un is never prime.
202 Hints and outline solutions

14th British Mathematical Olympiad, 1978

1. Determine, with proof, the point P inside a given triangle ABC for
which the product PLPMPN is a maximum, where L, M, and N are
the feet of the perpendiculars from P to BC, CA, and AB respectively.

(1) It is rather hard to see how the product PL PM PN varies as P

varies—there are just too many degrees of freedom! (This remains a problem
even though the question implicitly suggests that the product will attain its
maximum at some relatively familiar point P.) In such cases a good strategy is
to treat one of the three variables, say PN, as fixed, and then:

(a) first maximize the value of the product for each fixed value of PN;

(b) then find the maximum of all these maxima as the value of PN varies.

(2) Draw a triangle ABC, mark a particular point P' inside A ABC, and
draw in the three perpendiculars P'L', P M', and P'N'. Let P be an
unknown point in the triangle, and let PN be the perpendicular from P to
AB.

(a) Draw the locus of points P for which PN = P'N'. Let this line cut CA in
the point A', and cut CB in the point B'.

(b) Fix the value of PN and try to locate the point P on A'B' for which
PL PM is a maximum. Use the fact that ALB'P =_to express PL in
terms of PB' and AB; similarly, express PM in terms of PA' and A A.
Use the fact that A A and AB are fixed (once A ABC is given) to
conclude that PL • PM is a maximum precisely when PB 'PA' is a
maximum. Finally, use the fact that, when PN is fixed, PA' + PB' =_
is *o***a** to conclude that the maximum is attained at the *i**oi** of
A'B'.

(c) Now let PN vary. For each value of PN the maximum value of the
product PL PM PN is attained at the midpoint of A'B'. These points
all lie on the line joining C to the midpoint of AB. Hence the required
point P lies on the *e*ia* from C to AB. Similarly, P must lie on the
median from A to BC. Hence the maximum value of the product
PL PM PN is attained when P is at the *e***oi* of triangle ABC.

(3) Alternatively, A = area( ABC) = \PL •a + \PM-b + \PN-c. Hence A ^

l(abc)l/i[PL-PM-PN]l/3 (by AM-GM), with equality if and only if
area(FSC) = area(PG4) = area(/M5) = A/3. Since A is fixed, it follows that
PL PM PN achieves its maximum value (2A/3)3-(l/abc) when P is at the
centroid.
 14th BMO, 1978 203

2. Prove that there is no proper fraction m/n with denominator n < 100,
the decimal expansion of which contains the block of consecutive digits
‘167’, in that order.

(1) What a curious question! While there

? ? 16 7
may be some clever way to solve it, the i ,
way it is stated leaves one with little
option but to try to extract some mathematically usable information from the
long-division process of dividing n into m. Suppose that the usual process of
dividing n into m were to give rise at some stage to the sequence ‘167’. You
must show that n must then be greater than 100. Suppose that the remain¬
ders at the three crucial stages are a, b, and c as shown. Then 10a = n +_,
10b = 6n +_, and 10c = ln+d (where d is the remainder at the next
stage, so 0 < d < n).

(a) Substitute for b (in terms of n and a) from the first equation into the
second. Then substitute for c from this new second equation into the
third to show that 1000a = 16 7n + d.

(b) Show that, if n < 100, then a <

167 a 168
(c) Use 0 <:d <n to show -< — <
1000 n 1000
(d) The most simple-minded (yet still fairly efficient) way to finish the job is
to use the fact that | ^ to exclude all the possible values of a,
starting with a = 16. If a = 16, then

16 16 ^ 167 16 ^ 168 16
96 ~ (6 X 16) < 100 < n < 1000 < 95 ’

so that 95 <n< 96, which is impossible. The other values of a may be

excluded one at a time in the same way. However, the calculation for
a = 16 effectively eliminates them all: since, if a = i < 16, then

i 167 a i 168 16 _i_

67 < 1000 <n~n< 1000 < 95 6i - 1’

so that 6/ — 1 < n < 6i, which is impossible.

(2) An alternative to (lXc), (d) is to notice that the equation 1000a = 167n + d
implies that -2a = d(modl67). Thus 167 Id + 2a, whereas 0 <d + 2a < 132
(since a < 16, and d is a remainder on division by n < 100, so d < 100).
(3) You might like to refine the above methods to prove that the smallest
204 Hints and outline solutions

possible n the decimal of which contains the sequence ‘167’ is n = 131 (with
m — 22).

3. Show that there is one and only one sequence {un} of integers such that

Mj = 1, u1<u2, and u3n + 1 = un_lun + l for all n > 1.

(1) To determine that there is ‘one and only one sequence {«„}’ you have to
show two things: first, that any two such sequences must be identical; and,
second, that there is at least one sequence with the given properties.

(a) Let u2 = a > 1. Find expressions for u3 and u4. Explain why the sequence
{«„} is completely determined once u2 is given.

(b) You want {un} to be a sequence of i**e*e**. Thus you may assume that
a is an integer. Once u2 = a is chosen to be an integer, u3 is automati¬
cally an integer. Look carefully at your expression for «4. Show that for
«4 to be an integer a must divide _. Hence conclude that a =_
(since U2 > Mj). Thus any sequence of integers with ux = 1 and satisfying
the given recurrence relation has to have u2 = a =_, and so is
uniquely determined.

(2) It remains to prove that the sequence obtained when a = 2 is in fact a

sequence of integers. If you are tempted to calculate successive terms in the
hope of spotting some pattern, you will soon have second thoughts! Thus the
fact that always divides u3n + 1 must probably be proved algebraically.
While there is endless scope here for going round and round in algebraic
circles, the necessary calculation is not that hard.

(a) Direct calculation shows that ux = 1, u2 = 2, u3 =_, and u4 =_.

(b) Suppose that you know that all terms up to (and including) un are
integers, for some n> 4; in particular, (u3n_x + l)/un_2 is an integer.
You have to prove that un+x is an integer; that is, you have to prove that
the integer u3n + 1 is divisible by the integer un_v

3 , , 1 + 1)' , 1 (^-1 + 1)
+ 1 = -4- + 1 = --— + 1
U n-2
(Un-lUn-3 “ 1)

(M«- 1 + D + (m„_iW„_3 ~ 1)
(m„_1Mb_3- 1)

(“l + 1)(w„_1m„_3 - 1) = (u3n_1 + l)3 + (un_xun_ 1).

All terms in this final equation are integers (by induction). Moreover, the
 14th BMO, 1978 205

RHS is an exact multiple of un_x (Why?), whereas the second factor on

the LHS is relatively prime to un_x (Why?). Since the RHS (and hence
the LHS) is divisible by un_v it follows that un_ x must divide the first
factor ul + 1 on the LHS. Hence, un + l = (u\ 4- \)/un_x is an integer.

4. An altitude of a tetrahedron is a straight line through a vertex which is

perpendicular to the opposite face. Prove that the four altitudes of a
tetrahedron are concurrent if and only if each edge of the tetrahedron
is perpendicular to its opposite edge.

(1) You have two things to prove.

(a) First, when the altitudes are concurrent, you have to prove that each pair
of opposite sides are ‘perpendicular’ to one another.

(b) Second, if each pair of opposite sides are ‘perpendicular’ to one another,
you have to show that the altitudes are concurrent.

(2) Let the tetrahedron be ABCD. The repeated mention of ‘altitudes’ and
‘perpendicular’ tend to suggest trying a vector approach. Suppose first that
the four altitudes (from A to BCD, from B to CDA, from C to DAB, and
from D to ABC) all cross at the single point O. Take O as the origin, and let
a, b, c, and d be position vectors of A, B, C, and D.

(a) The altitude OA is perpendicular to the plane BCD and hence to any
vector in that plane. Conclude that a • (b — c) = a • (c - d) = a • (d - b) = 0,
and hence that a b = a- c = a-d.

(b) Repeat this argument for the other altitudes.

(c) Conclude that (d - a)-(c - b) = 0, so that AD is perpendicular to BC.

Show, similarly, that AB is perpendicular to CD, and that ,4C is
perpendicular to BD.
(3) Now suppose that AD is perpendicular to BC, that AB is perpendicular
to CD, and that AC is perpendicular to BD, and try to prove that the four
altitudes are concurrent. Here you may choose between a qualitative argu¬
ment ((a), (b)) and an explicit calculation ((c), (d)).

(a) Let A' be the foot of the perpendicular from A on to the opposite face
BCD. Define B',C', and D' similarly. Since AB is perpendicular to CD,
the perpendicular projection BA’ of BA on to the plane BCD will also be
perpendicular to CD. Thus the plane ABA' is perpendicular to CD, and
so contains the altitude BB' (since BB' is the intersection of the plane
ABA' and the plane through B perpendicular to AC). Hence the plane
ABA' is the same as the plane BAB', and contains both AA' and BB'.
Explain why these two lines cannot be parallel, and hence conclude that
they cross—at some point O.
206 Hints and outline solutions

(b) Similarly, CC' meets AA'. Moreover, the point of intersection X of CC'
and AA' lies on AA'—and hence in the plane ABA', and also on
CC'—and hence in the plane CBC'. Thus X lies on the intersection of
these two planes—namely on BB', as well as on AA'. Conclude that
X = O. Hence complete the solution.
(c) If you prefer an explicit calculation, take the point A' as the origin of
co-ordinates. Deduce that a-b = a- c = a- d = 0. Then use the assumption
(d - a)• (b - c) = 0 and so on to conclude that db=dc=bc=fc, say.

(d) Finally, show that there exists a value of t such that ta lies not only on
the altitude AA' but also on the other three altitudes. For example, for
ta to lie on BB' it must satisfy (b — ta)-(d - c) = 0, (b — ta)-(c — a) = 0,
and (b — ta)-(a — d) = 0. Show that the first is automatically satisfied, and
that the second holds precisely when t — —(k/( a-a)). Then show that this
value satisfies the third condition. Conclude that, since the value of t is
independent of c and d, the point ta will also lie on the other two
altitudes.

5. Inside a cube of side 15 units there are 11000 given points. Prove that
there exists a sphere of unit radius containing at least six of the given
points.

(1) It looks very much as though you will need to use the pigeonhole principle.
If so, then you clearly have to cover the cube with sufficiently many spheres
‘of unit radius’, so as to guarantee that some sphere will contain more than
five of the 11000 points. If there were >2200 unit spheres, then it is
conceivable that each might contain no more than five of the points. Thus
you need to cover the cube with fewer than 2200 unit spheres.
(2) You will find these large numbers impossible to handle unless you keep
things extremely simple.

(a) What is the simplest way of cutting up a cube into lots of identical pieces
with small ‘diameter’?

(b) Suppose that you cut the large cube into lots of identical small cubes,
each of side s, say. What is the radius of the smallest sphere containing
one of these small cubes of side 5? If this circumscribing sphere is to have
radius 1, what is the largest possible value of si

(3) If this approach is to work, you must:

(a) first cut up the 15 by 15 by 15 cube into identical smaller cubes with side
length <2//3;

(b) then check that the number of small cubes is <2200—whence at least
one of them will contain more than 5 of the 11000 points.
 13th BMO, 1977 207

(0 Show that 15/(2/^3”) < 13.

(ii) Check that 133 < 2200.

(iii) Explain how this solves the problem.

6. Show that if n is a non-zero integer, 2 cos n6 is a polynomial of degree

n in 2 cos 0. Hence or otherwise, prove that if k is rational, then cos ktr
is either equal to one of the numbers 0, + j, or +1, or is irrational.

(1) The first part is a straightforward induction proof. When n = 0,

2cos(0- 0) =p0i2cos 0), where p0(x) = 2 is a polynomial of degree 0. When
n = 1, 2cos nd =/?1(2cos 9), where pfx) =x is a monic polynomial of degree
1. Suppose that 2cos n9 — pni2cos 9), where pn(x) is a monic polynomial of
degree n with integer coefficients. Use

cos[(n + 1)0] + cos[(/i — 1)0] = 2 cos_-cos_

to show that ph+lix) is the monic polynomial of degree n + 1 (with integer

coefficients) given by

pn + i(x) =xpn(x) -pn_1(x).

(2) Now suppose that k = p/q is rational (peZ,?eW,^2). Let x =

2cos ktr. You have to prove that either x e {-2, -1,0,1,2}, or x is irrational.

(a) Show that pqix) = 2 cos_= ±2.

(b) Suppose that x = r/s is rational (/•eZ.jeN,hcf(r,s) = 1). The leading

term of pq(x) is then xq = (r/s)q - rq/sq, with hcfU17, sq) =_. Deduce
that pq(x) = N/sq for some integer N with hcf(N, sq) = 1. Finally, use (a)
to conclude that s — 1. Hence complete the solution.

13th British Mathematical Olympiad, 1977

1. A non-negative integer f(n) is assigned to each positive integer n in

such a way that the following conditions are satisfied:

(a) f(mn) = f(m) +f(n), for all positive integers m, n:

(b) f(n) = 0 whenever the units digit of n (in base 10) is a ‘3’; and

(c) /(10) = 0.

Prove that fin) = 0, for all positive integers n.

208 Hints and outline solutions

(1) It is instructive (although not strictly necessary) to begin by considering

all values of n < 10. You already know that /(3) = 0 and that /(10) = 0.

(a) Use 7(ran) = fim) + /(«), for all m,n> V to prove that /(1) = 0.

(b) Use /(10) = 0 to prove that /(2) = 0 and /(5) = 0.

(c) Use /(2) = 0 to prove that /(4) = 0.

(d) Use /(2) =/(3) = 0 to prove that /(6) = 0.

(e) Prove that /(8) = 0 and /(9) = 0.

(f) Now you know that /(9) = 0, use the fact that /(63) = 0 to prove that
/(7) = 0.
(2) You are now ready to prove the ‘fin) = 0’ for all n ^ 1 by induction on n.
The statement is certainly true when n < 10. Suppose that for some n > 10
you already know that f(k) = 0 for all k < n.

(a) If n + 1 can be factorized—say, n + 1 = a -b, with 1 < a, b <n + 1—use

the fact that fia) =/(b) = 0 to prove that fin + 1) = 0.

(b) Suppose that n + 1 is prime. Then, since n + 1 > 7, the units digit of
n + 1 is (i) 1, or (ii) 3, or (iii) _, or (iv) __. Hence some multiple
kin + 1) has units digit equal to 3 (in case (i) take k = 3; in case (ii)
take k= 1; in case (iii) take k = 9; in case (iv) take k = l); hence
fikin + 1)) =_. Use this to prove that fin + 1) = 0.

2. The sides BC, CA, and AB of a triangle touch a circle at X, Y, and Z

respectively. Prove that the centre of the circle lies on the straight line
through the midpoints of BC and of AX.

(1) The centre of the inscribed circle is precisely where the angle bisectors of
Z A and Z B meet. Let the bisector of Z A meet BC at D. Let a, b, and c be
the position vectors of the three vertices, and let a, b, and c be the lengths of
the three sides inot the magnitudes of a, b, and c). Let D have position
vector d.

(a) Show that BD : DC =AB : AC (the ‘Angle Bisector Theorem’). Use this
to show that d = (bb + cc)/ib + c).

(b) Now let the bisector of ZB in AABD meet AD at I (the incentre of

A ABC). Let I have position vector i. Use BD = DC-iAB/AC) =
ia - BD)-ic/b) to find BD. Then use the same method as in (a) to show
that i = (aa + bb + cc)/ia + b + c).

(2) To simplify matters slightly, choose the origin at the midpoint L of BC.
Then c = -b, so i = (aa + ib — c)b)/(a + b + c).
 13th BMO, 1977 209

(a) Show that BX = (a+c- b)/2. (Use BX = BZ = u, CX=CY=v, and

AY = AZ = w, and solve u + v = a, u + w = b, and w + u — c to find u.)

(b) Let X have position vector x. Find the magnitude of b (in terms of a\).
Hence show that x + ((a + c — b)/2)-((2/a)b) = b. Simplify to find x (in
terms of b and a, b, and c).

(c) Let the midpoint T of AX have position vector t. Show that t =

|(a + x) = (|)a +_b. Show that (2a/{a + b + c))t = i. Conclude that
L, I, and T are collinear.

(3) If ‘the side BC’ refers to the doubly infinite line BC (rather than the line
segment BC), then the circle may be an escribed circle. You should check that
a similar calculation works in this case.

3. (a) Prove that if x, y, and z are non-negative real numbers, then

x(x —yXx — z) +y(y —zXy ~x) +z(z -x\z -y) ^ 0.

(b) Hence or otherwise, show that, for all real numbers a, b, and c,

a6 + b6 + c6 + 3a2b2c2 > 2(b3c3 + c3a3 + a3b3).

(1) The simplest, but least elegant, solution to (a) is probably to use the
symmetry of the LHS to observe that one may assume z >y >x (>0).

(a) Show that x(x - yXx - z) > 0.

(b) Check that z(z -x)(z -y) > 0 >y(y - z\y -x).

(c) Use z^y >0 (and hence z — x — x to show that

z(z —x)(z —y) +y(y -z)(y-x) = (z-y)[z(z-x)-y(y-x)]> 0.

(d) Combine (a) and (c) to prove the required inequality.

(Note that (6) in the outline solution to 1981 Question 3(b) shows that the
two questions are exactly equivalent.)
(2) Since the terms on the LHS of (a) are of degree 3, it is natural in part (b)
to start by putting x — a2, y —_, and z =_ in the inequality of part (a) to
obtain an inequality involving a6 + b6 + cb + 3a2b2c2.

(a) Write out the resulting inequality.

(b) Thus, to prove the inequality in part (b) it suffices to prove

a\b2 + c2) + b4(c2 + a2) + c4(a2 + b2) >2(b3c3 + c3a3 + a3b3).

Write a4(b2 + c2) + bA(c2 + a2) + c4(a2 + b2) — 2(b3c3 + c3a3 + a3b3) as
the sum of three perfect squares, and so complete the solution.
210 Hints and outline solutions

4. The equation x3+qx + r=0, where W 0, has roots u, v, and w.

Express the roots of the equation r2x3 + q3x + q3 = 0 in terms of u, u,
and w. Show that if u, v, and w are real, then this latter equation has
no root in the interval —1 <x <3.

(1) This question suggests that, though the two equations look different, they
are in fact closely related. If the equations are related in a simple way (say
with x in the original equation corresponding to kx in the second), then they
should at least have the same *o***a** term.

(a) Multiply the second equation by a constant factor to give them both the
same constant term r.

(b) Then identify the change of variable y = kx which turns the second
equation into y3+qy + r = 0. Hence express the roots of the second
equation in terms of u, v, w, q, and r.

(2) The question asks you to ‘express the roots of the second equation in
terms of m, v, and w’ only. Therefore you have to find an expression for q/r
in terms of u, v, and w. So far you have made no use of the standard
relations between the roots u, v, and w and the coefficients 1, 0, q, and r.

(a) Use the fact that the coefficient of x2 is zero to write down a relation
which u, v, and w must satisfy.

(b) Use the known values of the other coefficients in the original equation to
write down relations of the form uv + vw + wu =_ and uuw =_.
Divide the first of these two relations by the second and simplify to obtain
an expression for q/r in terms of u, u, and w. Hence express the three
roots of r2x3 + q3x + q3 = 0 in terms of u, u, and w only.

(3) Now suppose that u, v, and w are all real. You have to find some
(preferably simple) way to use this additional information to prove that none
of the roots of the second equation lie in the interval -1 <x < 3.

(a) If u, v, and w are all real, explain why f(x)=x3 +qx + r must have a
maximum at x= -]/(-q/3) and a minimum at * = ]/(-q/3) . Con¬
clude that q < 0.

(b) Explain why the maximum must be above the y-axis and the minimum
must be below the y-axis. Try to simplify the two conditions
Y(- ]/(~q/3)) > 0’ and ‘f(sJ(-q/3)) < O’. This leads to rather messy
algebra. You may have to pursue this; but it is worth looking for a simpler
alternative. In particular, you should be aware of the fact that you have
still not made use of the important relation (2Xa)!

(c) What does the given information W0’ tell you about the roots u, v,
and w?
 13th BMO, 1977 211

(d) If u, v, and w are all real, how many of them can be negative? (Use the
relation that you found in (2Xa).) Conclude that you may assume that u is
negative, that w is positive, and that u<u <w, with u * 0.

(e) Suppose that u = -(v + w) is the only negative root. Then r> 0 (since
uvw —-). Hence qu/r is *o*i*i*e, whereas the other two roots qv/r
and qw/r of the second equation are *e*a*i*e, with qw/r < qu/r.
Moreover,

(q/t) ■ u (1 /u 1/u -\- 1/w) • u = — 1 + (y + w)/u + (y + w)/w

= 1 + w/v + v/w >_

(since ((v/w) - l)2 ^ 0 can be rearranged to show that w/v + v/w ^ 2).
Similarly,

qw/r < qv/r = -(1/u + 1/v + l/w)-v

— — 1 + (l/(v + w) — 1 /w)-v < -1 + 0

(since v + w > v > 0 implies that (1 /(v + w) - 1 /v) < 0).

(f) The case in which u and v are both negative requires only a slight
modification of the above argument. (However, this is not strictly neces¬
sary, since if u, v, and w are roots of x3 + qx + r = 0 with u and v
negative, then -u, -v, and —w are roots of x3 + qx - r = 0, and both
equations give rise to exactly the same ‘second’ equation r2x3+q3x +
q3 - 0.)

5. The regular pentagon AXA2A3A4A5 has sides of length 2a. For each
i = 1,2,..., 5, Kt is the sphere with centre At and radius a. The spheres
Kx,K2,...,K5 are all touched externally by each of the spheres P1 and
P2—also of radius a. Determine, with proof, whether or not Px and P2
have a common point.

(1) This question is slightly ambiguous. Imagine that the regular pentagon
AxA2A3A4A5 lies in a horizontal plane II. If the spheres Kv K2, K3, K4,
and K5 are touched externally by each of two spheres Px and P2, then one of
these spheres—Pt, say—must touch the spheres Kt above the plane II and
the other—P2, say—must touch the spheres Kt below the plane II.
If the spheres Px and P2 were solid, then they could not interpenetrate
each other, so the only way in which they could ‘have a common point’ would
be if the lowest point X of Px (and hence the highest point of P2) were to lie
exactly at the circumcentre C of the pentagon A1A2A3A4A5. Let O, be the
centre of Pt. If X = C, then OxC = OxX =_ (given); since AxOx =_
(as the two spheres Kx and Px touch), and Z.AxCOx =_, this would imply
212 Hints and outline solutions

that the circumradius r=AxC of the pentagon AXA2A3A4A5 is equal to

(y[3)a, which is easily seen to be false (for example, since sin 36° # 1/ V3).
(2) One suspects that candidates were supposed to consider the spheres Px
and P2 as loci which may interpenetrate. The question then becomes
whether the lowest point X of the sphere Px lies above or below (or in) the
plane Id: that is, whether OxX > OxC or OxX < OxC (or OxX= OxC). One
way to decide is to use the same right-angled triangle AxCOx to calculate
COx = \/(4a2 - p2). You must then decide whether (i) yj(4a2 — p2) <a or
(ii) y/(Aa2 - p2) > a (or yJ(Aa2 - p2) = a).

(a) Let M be the midpoint of AXA2. Use A A XMC to show that p =

a/sin 36°.

(b) Show that (i) holds if sin 36° <1/^/3 , and that (ii) holds if sin 36° > 1/ V-T .

(c) Use the geometry of the regular pentagon to show that sin 36° =
(1 + /5 )/4. Hence solve the problem.

6. The polynomial 26(x+x2 + x3 + ... +xn), where n > 1, is to be decom¬

posed into a sum of polynomials, not necessarily all different. Each of
these polynomials is to be of the form axx + a2x2 + a3x3 + ... +anxn,
where each at is one of the numbers 1,2,3,..., n and no two a, are
equal. Find all the values of n for which this decomposition is possible.

(1) The wording here is very strange; but once you manage to understand
what it is saying, the problem is actually rather easy. You have to find all n
for which the given polynomial

26(x tx2 +x3 + ... -t-jc”)

can be written as a sum of other polynomials of the form

(*) axx + a2x2 + a3x3 + ... +anxn,

in which the coefficients a, are just the integers 1,2,3,...,n in some order.

(a) You are not told exactly what these other polynomials look like. However,
you know that the largest coefficient in (*) is always _, and that each
of the coefficients in the original polynomial =_. Deduce that n <

(b) Explain why n = 26 is impossible. Show that n = 25 is possible.

(2) This certainly feels like progress, but you need another idea if you are to
avoid the mess of trying to deal with each of the remaining possible values of
n in turn.
 12th BMO, 1976 213

(a) What is the sum of all the coefficients in (*)? If you were to add up k
polynomials of this form, what would be the sum of all the coefficients in
the resulting polynomial?

(b) Suppose that it is possible to find k polynomials like the one in (*), the
sum of which is equal to the polynomial 26(x +x1 2 + x3 + ... +xn). Show
that kn(rt + l)/2 = 26n. Use the fact that n > 1 and n < (from (1)) to
conclude that n = (with k = ), or n = (with k = ), or n = 25
(with k = 2).

(c) You already know that n = 25 works. Now take the other two possibilities
in turn and show that they also work.

12th British Mathematical Olympiad, 1976

1. Find, with proof, the length d of the shortest straight line which bisects
the area of an arbitrary given triangle, but which does not pass through
any of the vertices. Express d in terms of the area A of the triangle and
one of its angles. Show that there is always a shorter line (not straight)
which bisects the area of the given triangle.

(1) Let the triangle be ABC. The median joining A to the midpoint L of BC
splits the triangle into two parts of equal area (equal bases, same height). The
same is true for the other two medians. To confront a common misconcep¬
tion, it is instructive to begin by proving that these are the only lines through
the centroid G that bisect the area of the triangle!

(a) Suppose that y (¥-A) is a point on AB and that Z is a point on BC such

that Y, G, and Z are collinear. Let u, v, w, and x denote the areas of
AYG, ABZC, ZGL, and BLGY. Show that YZ bisects the area of the
triangle if and only if u = w. Let CN be the median from C to AB; use
GC = 2 • GN to deduce that GZ = 2 ■ GY.
(b) Hence prove that Z = C. (Suppose that Z ¥= C. Show that A GNY and
AGCZ would then be similar. Conclude that LGYN = ZGZC, whence
YN would have to be parallel to CZ.)
(2) Choose points P on AB and Q on AC such that PQ bisects the area of
A ABC. Let AP and AQ have lengths p and q respectively. Let PQ=x.

(a) Use the cosine rule in AAPQ to express x2 in terms of p, q, and LA.
You want to find the minimum possible value of x2.

(b) Express the area of AAPQ in terms of p, q, and LA. Put this equal to
A/2, and hence show that, as P and Q vary (with PQ still bisecting the
area of A ABC), the product pq remains constant.
214 Hints and outline solutions

(c) Combine (b) and (a) to deduce that x2 is a minimum precisely when
p2 + q2 is a minimum. Next explain why p2 + q2 is a minimum precisely
when (p — q)2 is minimum. Hence prove that, if ^(A/sin A) is less
than or equal to both b and c, then the minimum occurs when
p = q = }/(.A/sin A).
(d) Show that when p = q, x2 = 2p2(l — cos A) = 4p2 sin2(v4/2). Conclude
that the required length d is equal to the minimum of 2A tan(^l/2),
/2Atan(fl/2), and ^/2Atan(C/2) —and so occurs for the smallest of
the three angles A, B, and C. You should check that if a < b and a < c,
then y^A/sin A) < b and ^(A/sin A) < c. (The strict inequality here
implies that the shortest straight area-bisector is never a median.) This
completes the first part of the question.

(3) To find a ‘shorter (not straight) line’ which bisects the area of AABC,
suppose that ABAC = 6 is the smallest angle of AABC.

(a) Choose X on BC such that AX bisects ABAC and apply the same
reasoning as in (2) to AABX and AAXC separately The shortest line
bisecting the area of AABX has length dx and the shortest line bisecting
the area of AAXC has length d2. Prove that dx+ d2< d. Use the result
of (2Xc) to explain why the cut of length dx has an endpoint in common
with the cut of length d2.

(b) If you find proving dx+ d2<d slightly messy, you may prefer the follow¬
ing argument. If you bisect A BAX and AXAC and repeat the above
procedure, you obtain a (not straight) line which consists of four straight
segments, and which again bisects the area of the triangle. Repeating this
over and over again gives a sequence of polygonal lines, which bisect the
area of AABC, and which approximate more and more closely to an arc
of a *i***e with centre A, radius r= yXA/0) (where 0 is measured in
radians), and length r- 0 = /K0. Show that VA0 < ^2A tan(0/2) . Hence
complete the second part of the question.

2. Prove that if x, y, and z are positive real numbers, then

x y z 3
-1-1-> —.
y +z z+x x+y 2

(1) There are many approaches here that work (and even more that simply
go round in circles). Two things are worth noticing. First, you would prefer
not to have those mixed sums ‘y +z’ and so on in the denominator. Second,
the LHS of the inequality to be proved is unchanged when you replace x, y,
 12th BMO, 1976 215

and z by Ax, Ay, and Az for any A > 0; you may therefore choose a scaling
factor to suit yourself.

(a) Multiply all three variables by ‘1 /(x+y+z)’, to obtain new positive

variables which will also be called x, y, and z(!)—with sum x + y + z =

(b) Rewrite the LHS in the form ‘[x/(l — x)] + Then write x/(l —x) =
— 1 +- and show that the inequality to be proved is equivalent to

1 1 1
1 -x 1 -y + 1-2 3
3 > 2'
(c) The LHS now invites you to use the AM-GM inequality. It would
therefore suffice to prove that

7_1_ 3
V (1 x)(l —y)(l -z) ^ 2 '

(d) Notice that (1 -x) + (1 -y) + (1 -z) =_, so (by the AM-GM inequal¬

ity) t/(l x)(l —y)(l -z) <_, which completes the solution.
[A different way of getting rid of the mixed sums in the denominators is
to put u =y + z, v — z + x, and w —x + y. Then x = (v + w — u)/2 and so
on, so

x y z 1 (v u w v u w
I H-= — ( —I-1-1-1-1-3
y+z z+x x+y 2\u v v w w u
$5 j(2 + 2 + 2 — 3).]

(2) Alternatively, let X = (x+y + z)/(y + z), Y = (x +y + z)/(z +x), and

Z = (x +y +z)/(x +y). Then the harmonic mean H of X, Y, and Z satisfies
3
" = T—i—“-
— + — + —
X Y Z

The standard harmonic mean inequality says that H is less than or equal to
the arithmetic mean; hence f <(X+Y+ Z)/3, so X+Y+Z^l, which is
precisely the inequality to be proved.
(3) An even simplier method is to multiply out and show (using x, y, z > 0)
that the inequality to be proved is equivalent to

2(x3 + y3 +z3) >x2y +y2z +z2x + xy2 +yz2 +zx2.

Now use 1981 Question 3(a) (twice).

216 Hints and outline solutions

3. SVS2,...,S50 are subsets of a finite set E. Each subset contains more

than half the elements of E. Show that it is possible to find a subset F
of E having not more than five elements, such that each St (1 < i 50)
has an element in common with F.

(1) Suppose that the set E has n elements. Count the number of pairs (x,i)
in the set ^ = (U,i):x£5,} in two different ways.

(a) Suppose that, on average, each element x belongs to A of the sets St.
There are exactly _ possible first co-ordinates x, and each element x
occurs, on average, in _ pairs (x,i). Hence there are n-A such pairs.

(b) On the other hand, there are_possible second co-ordinates i, and each
set St has more than n/2 elements; hence there are >50-(n/2) pairs.

(c) Conclude that A > 25, and hence that some element xx must belong to
at least 26 of the sets St.

(2) Removing 26 of the sets S, which contain the element xx leaves exactly
_ sets Sj. As before, count pairs (x,y) in two ways, where x e Sj and Sj is
one of the remaining 24 sets. Conclude that some element x2 must belong to
at least 13 of the remaining sets Sj.
(3) Removing 13 of the remaining sets that contain the element x2 leaves
exactly _ sets. The same argument shows that some element x3 must
belong to at least six of the remaining sets.
(4) Removing six of the remaining sets that contain x3 leaves just _ sets.
The same argument then shows that some element x4 must belong to at least
three of the remaining sets. Finally, each of the two other sets contains more
than half the elements in E, so they have at least one element x5 in common.
The set F = {xv x2, x3, x4, x5} then does the trick.

4. Prove that if n is a non-negative integer, then 19 X 8" + 17 is not a

prime number.

(1) Let 19-8" -f 17 = un.

(a) If n = 0, then u0 = 19 • 8° + 17 =_. Factorize this!

(b) If n = 1, then ux = 19-85 + 17 =_. Factorize this!

(c) If n = 2, then u2 = 19 • 81 2 + 17 =_. Factorize this!

(2) There is little point doing further computations until you have an idea to
test (and if the idea is a mathematical one, there may well be better ways to
test it than using brute force computation). The values of u0 and ux are
misleading in that they are both **ua*e*, whereas u2 is not. However, it is
 12th BMO, 1976 217

hard not to notice that the only small prime factor of u2 is_, and that this
is also a factor of u0.

(a) Show that, if n is e*e*, then 8" =_(mod3). Hence show that
19 - 8" + 17 is never prime when n is even.

(b) In the same (optimistic) spirit, the only small prime factor of ux is_.
Show that, if n is o**, then 8" = +_(mod 13). Hence show that
19-8" + 17 is never prime when n is an odd number of the form _.

(c) You are clearly going to have to work a little harder when n = 4k + 3.
However, if you consider which small primes p might divide u3 =
19-8" + 17, then it should not take you long to exclude p = 2 and p = 3,
and hence to discover that u3 is a multiple of . You should then be
able to prove that 19-8" + 17 is a multiple of_ whenever n has the
form 4k + 3.

5. Prove that, if a and /3 are real numbers, and r and n are positive
integers with r odd and r < n, then

0-l)/2

(r—l)/2
n \\ r—t
= L (aj8)'(a + /3) r-2t
r=0
r-t \ t

(1) The easiest approach (that is, one which allows most students to get
started) almost never leads to the easiest, or most satisfying, solution. In this
problem, perhaps the most straightforward approach (although one which will
test your ability to manipulate algebraic expressions correctly) is simply to
find the coefficient of as (say) on each side, and then to show that
corresponding coefficients are equal. (A neater method is to try to identity
what it is that both sides are equal to: see (3).)

(2) (a) For s <(r— l)/2, the coefficient of as on the LHS is (r ” 5 )/3r_s>
and the coefficient of otr~s is _.

(b) The coefficient of as on the RHS is more complicated, in that there is

a contribution from each ‘term’ in the sum. Let s<(r- l)/2. Then
a5 = a‘- as~‘ for each t, 0 < t < 5. Check that the coefficient of a5 on
the RHS is
218 Hints and outline solutions

(c) Show that

n r — 2t
r— s—t

(d) Show that E(^)(r”^s) = (r”5)- (Suppose that you are given a

set N of size n and a fixed subset S qN of size s. What does the RHS
of this identity count? Now explain why the LHS counts the same
collection of sets in a different way.) Hence solve the problem.

(3) An alternative approach comes from observing that | n( j is the coefficient

of a' in (1 + a)”, and that ( ” J is the coefficient of /3r_' in (1 + /3)'1.

Hence the rth summand on the LHS of the identity in the question consists
of precisely those terms in the expansion of (1 + a)"(l + /3)" which have
degree r. Another way of calculating the terms of degree r in the expansion
is to multiply (1 + aXl +/3) first and then to expand the product (1 +
(a + /3) + a/3)". In multiplying out the n brackets (1 + (a + /3) + a/3), for
each term one may choose either a T, or an ‘a + /3’, or an ‘a/3’ from each
bracket. Choosing ‘a/3’ from t brackets contributes a factor (a/3)' of degree
2t\ thus, to land up with a term of degree r, it remains to choose ‘a + /3’ from
exactly r - 2t brackets (and ‘1’ from the remaining n - r + t brackets). There

are | n ^ j j ^ ^ ] wa^s
= +1 c^oos*nS n ~ r + t brackets from which

to take a T; there are then |r ^ j ways of choosing t of the remaining r-1

brackets from which to take ‘a/3’ (with ‘a + /3’ taken from the other r —2t
brackets). Hence the rth summand on the RHS also gives the terms of degree
r in the expansion of (1 + (a + /3) + a/3)”, so the two sides are equal.

6. A sphere with centre O and radius r is cut by a horizontal plane

distance r/2 above O in a circle K. The part of the sphere above the
plane is removed and replaced by a right circular cone having K as its
base and having its vertex V at a distance 2r vertically above O. Q is a
point on the sphere on the same horizontal level as O. The plane OVQ
cuts the circle K in two points X and Y, of which Y is the further from
Q. P is a point of the cone lying on VY, the position of which can be
determined by the fact that the shortest path from P to Q over the
surfaces of cone and sphere cuts the circle K at an angle of 45°. Prove
that VP = y!3tM1 + l/t/5).

(1) Draw a decent diagram (more or less to scale). Your diagram should
suggest something slightly unexpected. Show that VY is tangent to the sphere
at y.
 UthBMO, 1975 219

(2) Position V directly above the ‘north pole’. Then K becomes a circle of
latitude. Let PQ cut the circle K at T.

(a) Explain why QXT is not a ‘spherical triangle’.

(b) Let the circle of longitude through T meet the equator at Z. Check that
QZT is a spherical triangle. Let q, z, and t denote the angles subtended
by the arcs ZT, TQ, and QZ at the centre of the sphere. Use the given
spherical triangle formulae to show that /2-sinr = sinz and that
(v3 /2)cos t = cos z. Conclude that sin t = 1/ /5 .

(c) Check that one half of the great circle through Q and T starts at Q,
curves up to T, and rises further above the circle of latitude K before
cutting K again at a point T', and then dropping to Q' opposite Q on
the sphere. Conclude that the angle QOZ is acute. Hence show that the
arc YT on the circle K has length (/3/2)H> - arcsin(l//5)). (This tells
you exactly where the point T is on the base K of the cone.)

(3) Slit the cone along VY and lay it flat.

(a) Show that VY =

(b) Use the known length of the arc YT on the circle K to calculate LYVT.
Then use the fact that Z. VTP = tt/4 to find L VPT.

(c) Let 26 = (arcsin(l//^)). Use the double angle formulae to show that
cos20 = (5 + 2/5 )/10 and sin2# = (5 - 2/5 )/10.

(d) Use the (ordinary) sine rule in A VPT to show that

/3,
VP =
(5 + 2/5) (5-2/5)
+
10 ~lo

Hence complete the solution.

11th British Mathematical Olympiad 1975 ,

The 1975 paper (like the 1965-74 papers) differs in several ways from later
papers. Only three hours were allowed. There were more questions—
eight—than on later papers, and many of the problems are shorter and
perhaps more standard than on later papers.
Moreover, the original rubric—‘Do as much as you can’—suggests that
candidates were not expected to complete all the questions.
220 Hints and outline solutions

1. Given that x is a positive integer, find (with proof) all solutions of

[3/f] + $2] + ... +[\/U3 - 1) ] = 400.

3 3 3
(a) Check that [\/T] = [v/2]= ...[/7] =_. Hence these first seven terms
contribute a total of_ to the LHS of the equation (*).

(b) Find the largest N such that [i/8 ] = [t/9 ] = ...[/N] = 2. Hence calculate
how much these terms contribute to the LHS of the equation (*).

(c) Prove that [yfy] = k precisely when A:3 < {k + l)3. Calculate the exact
contribution which the _ terms with k = 3 make to the LHS of (*).
Do the same for k = 4. Hence find the unique value of x which satisfies
(*).

2. The first n prime numbers 2,3,5,, pn are partitioned into two

disjoint sets A and B. The primes in A are a1,a2,...,ah, and the
primes in B are b2,..., bk, where h + k = n. The two products

PI«?' and HI bt‘

i= i i = i

are formed, where the a, and /3, are any positive integers. If d divides
the difference of these two products, prove that either d = 1 or d > pn.

Denote the two products by n and n\ Suppose that d ¥= 1 divides n — n\

Let p be the smallest prime factor of d.

(a) Explain why p I (n — II').

(b) Suppose that p </?„. Then p = p, (some i, 1 < i < n) is one of the first n
primes, and so occurs (to a positive power) in one of the two products;
say, p | n (so that p e A). Now use (a) to deduce that pin', and hence
that p e B—contrary to the given fact that A and B are disjoint.

(c) Conclude that p>pn. Since d >p, the result follows.

3. Use the pigeonhole principle to solve the following problem. Given a

point O in the plane, the disc S with centre O and radius 1 is defined
as the set of all points P in the plane such that |OP| < 1, where OP is
the distance of P from O. Prove that if S contains seven points such
that the distance from any one of the seven points to any other is ^ 1,
then one of the seven points must be at O.
 11th BMO, 1975 221

Let S be the disc \z\ < 1 in the complex plane. Partition S into seven parts:

{0} ,and 5^ = {2G5;z#0, k(ir/ 3) ^ arg(z) < (k + 1)(tt/3)} (0 < k < 5).

Suppose none of the seven given points is equal to 0. Then one of the six sets
Si must contain (at least) two of the points. Prove that these two points will
then be at a distance < 1 from each other.

4. In a triangle ABC, three parallel lines AD, BE, and CF are drawn,
meeting the sides BC, CA, and AB in D, E, and F respectively. The
points P, Q, and R are collinear and divide AD, BE, and CF
respectively in the same ratio k: 1. Find the value of k.

(1) Let a, b, c, d, e, f, p, q, and r be the position vectors of A, B, C, D, E, F,

P, Q, and R respectively, and let x = d - a be a vector in the direction of the
three parallel lines. Choose l such that d = tb + (1 -1)c. We show that k=\.

(a) Use the fact that AP : PD = k : 1 to show that p = [a + ktb +

k( 1 — t)c\/(k + 1). (Note that k # — 1, since k = — 1 would correspond in
some sense to P being a ‘point at infinity’.)

(b) The general point on the line BE has vector b + sx = -sa + (st + l)b +
5(1 - Oc, and the general point on the line AC has vector ra + (1 - r)c.
Deduce that e = (l/f)a + (1 — (1/t))c, and hence that q = [(k/t)a + b +
k( 1 - (l/0)c]/U + 1).

(c) Do the same with CF andAB to show that f=[l/(l— r)]a + [-r/(l— r)]b,
and hence that r = [(&/( 1 - r))a - (kt/( 1 - t))b + c\/(k + 1).

(d) The vectors a, b, and c are linearly dependent. Hence, to obtain a

necessary and sufficient condition for p, q, and r to be collinear (that is,
for p — q and q — r to be linearly dependent) it is convenient to take the
origin at A. Put a = 0 and show that p — q is then a multiple of q - r if
and only if 2A:2 4- A: — 1=0. Hence conclude that k =

(e) Let ABC be equilateral and let D be the midpoint of BC. Locate P, Q,
and R in this case and check that they are collinear.

(2) You might like to find a direct proof of the converse of what is proved in
(1): namely, if parallel lines AD, BE, and CF are drawn through the vertices
of triangle ABC, meeting the opposite sides at D, E, and F, and if P, Q, and
R are the points of trisection of the segments AD, BE, and CF (with
PD = 2 AP, and so on), then P, Q, and R are collinear.
222 Hints and outline solutions

5. Let m be a fixed positive integer. You are given that

(20m ) +(2r )cos9 + (22m )Cos2e+ -+[2m)COs2me

- (2 cos \ d )2m cos m d,

where there are 2m + 1 terms on the LHS; the value of each side of
this identity is defined to be fm(d). The function gm(d) is defined by

£m(0)=(2^) + (2™)cos20 + (2™)cos40 + ...+(^)cos2 mO.

Given that there is no rational k for which a = krr, find the values of a
for which limm_ Jgm(a)/fm(a)\ =

(1) You need to find some connection between fm and gm. Since gm(6) is
equal to fm(Q) without the ‘odd’ terms (in d, 3d, and so on), this suggests
looking at

fm(0+ 77 ) = 1 + |2^ jcos(0+ 77) + |^|cOS2(6+ 77)

+ ... + cos2ra(0 + 77)

= 1
(2rH+(2r) cos 2 6— ... +cos2 md,

in which all the ‘odd’ terms have negative coefficients. Hence

+/m(0+

2m
d+ 77\\2m
2cosl —II cosra0+ 2cos cos m(d+ tt)

d + 77 ’
(2) Show that cos = —sin I — I. Conclude that

a 2m

2cos| — jj cos ma ± ^2 sin ^ — j cos ma

SmM_
2m
/m(«)
2 cos ■ cos m a
(!))
a 2m

1 ± Itanl —
 11th BMO, 1975 223

Show that this converges (to_) as m tends to °o if and only if |tan(a/2)| <
, that is, if and only if 2mr— (7t/2) < a < 2nir + (7r/2) for some «eZ.
(3) How does this calculation depend on the given fact that ‘there is no
rational k for which a = kTr’2

6. Prove that, if n is a positive integer greater than 1, and x >y > 1, then
Xn+1~l y"+ 1 — 1

xixn~l - 1) > y(_y”~1 - 1) ‘

(1) There is no obvious way of using standard results about inequalities here.

(a) Let fix) — (xn +1 — l)/(xn — x). Compute the numerator g(x) of the
derivative fix) = g(x)/(xn -x)2.

(b) Show that gix) = ix - l\x2n~l + x2n~2 + ... +x" + 1 - in - \)xn +

xn~1 + ... +x + 1).

(c) Use x > 1 to conclude that fix) is an increasing function for values of x
relevant to the problem. Hence complete the solution.

(2) Alternatively, it is enough to show that S =yixn+1 - lXy"~1 - 1) -

xixn~ 1 — lXy"+1 — 1) > 0. Show that the expression S can be factorized as
ix-y)ixy- IX...). Finally, observe that the final bracket can be written as
the sum of [n/2] positive terms

(x"-1 - l)(y"_1 - 1) +xyixn~3 - l)(y"~3 - 1)

+ x2y2ixn~5 — l)(y"-5 — 1) + ... .

7. Prove that there is only one set of real numbers xl,x2,...,xn such that

(1 -xf)2 + (*J -x2)2 + ... +ixn_1 -xn)2 +x2n = -

(lXa) Let u0 = 1 —xv ux =x: —x2,...,un_l =xn_x —xn, un =xn. Show that
£”=0 ui = 1- Use this to rewrite the given equality in the form

ul + u\ + ... +u2n = ——iu0 + ux + ... +unf.

(b) Show this implies niuf + u\ + ... +«^) — 2EI<;-utUj = 0. Conclude

that 'Li<jiui - Uj)2 = 0. Hence complete the solution.
224 Hints and outline solutions

(2) Alternatively, use the Cauchy-Schwarz inequality ((X"=0 fl,A)2 <

(E"=0 u,2XE"=0 b2), with equality if and only if u, = Abt for all i) with a, = ut
and bt =_ for all i. Hence prove that £”=0 ^ l/(« + 1). Conclude that
there is only one set of real numbers x1,x2,...,xn satisfying £”=0w2 =
1 /(n + 1).

8. The interior of a wine glass is a right circular cone. The glass is half
filled with water and is then slowly tilted so that the water reaches a
point P on the rim. If the glass is further tilted (so that water spills out),
what fraction of the conical interior is occupied by water when the
horizontal plane of the water level bisects the generator of the cone
furthest from P?

Let & be the cone, A its apex, and M the midpoint of the generator of the
cone ‘furthest from P\ Let II be the plane through P and M representing
the final water level.
(a) Shift the cone (and the plane II) so that A is at the origin, and the axis of
the cone goes up the positive z-axis. Any stretch in the z-dhection
changes & into another right circular cone, shifts M to the new midpoint
of the ‘generator furthest from P’, and changes all volumes by the same
scale factor. All possible configurations can be obtained in this way, and
the answer to the problem is the same for all of them. Thus it suffices to
solve the problem for your favourite cone
(b) Assume that /-MAP is tt/2, that AM = 1 (so AP =_). Then PM =
_. Let X lie on PM, with AX on altitude of AAPM. Use similar
triangles to show that AX - _.

(c) Let C be the centre of the circular ‘base’ of the cone. Find the radius CP
of the base. Hence show that the volume of the whole cone is 2^2 • 7t/3.
(d) It remains to find the volume of the water—that is, the volume of the
cone between the plane II and the apex A. The plane n cuts the cone g7
in an e**i**e E, with major axis PM of length 2a = yf5 . Thus you have
to find the volume of a pyramid with base E and height AX = 2/\[5 . If
you could find the length 2b of the minor axis, then you could use the
formula ‘7Tab' to find the area of the base E. The volume of water would
then be equal to Tab)-AX.
(e) Show that b = 1. (The equation of the cone is x2 + y2 =z2. If CP points
in the ^-direction, then the equation of the plane n is 3z-x = 2\/2.
Substituting for z gives the equation of the ellipse E as 8x2 - 4^/2x +
9y2 = 8. The minor axis runs in the y-direction, and y2 = f(2 + y/2x -
2x2). Since -2<*<1, x+x2^}, so 2-x-x2<f. Hence y2 < 1.)
Hence show that, in its final position, the volume of water is exactly 1 /2^2
of the volume of the glass.
 Appendix
The International Mathematical
Olympiad:
UK teams and results, 1967-1996

The problems in this book are intended to challenge and stimulate large
numbers of interested young mathematicians in their last years at school.
However, they also constitute the first round of the selection process for the
team which has represented the UK at the International Mathematical
Olympiad each year since 1967 (with the exception of 1980, when the IMO
was cancelled, and two teams took part in alternative events). Indeed, the
prospect of being invited to attend the IMO in 1967 or 1968 was perhaps the
main reason for starting the British Mathematical Olympiad in 1965. It
therefore seems appropriate to end with a brief record of UK participation in
the IMO.
The IMO began in 1959 in Romania, and was at first restricted to Eastern
European countries. Starting in the mid-1960s, countries from the West were
gradually invited to take part. Teams from different countries gather each
July to sit two four and a half hour papers, each with just three problems.
Until 1981 countries could send a team of up to eight students. By 1982 the
number of teams had increased substantially, and in that year countries were
invited to send teams of up to four students. Since 1983 countries have been
invited to send up to six students. The event has continued to grow: in 1967,
13 countries took part; in 1982, 30 countries took part; in 1996, 75 countries
took part.
The IMO is officially an individual competition. The medals are officially
known as ‘first’ (1), ‘second’ (2), and ‘third’ (3) prizes, but are often referred
to as ‘gold’, ‘silver’, and ‘bronze’ medals. The proportions of each have varied
slightly from year to year: at most one half of contestants receive a prize, with
the ratio gold: silver: bronze being approximately 1:2:3.
For each year we list the host country for the IMO; the UK team Leader
and Deputy Leader; the UK team’s position, score, and medal tally; and the
team members, their medals, and their schools.
226 Appendix

1967 YUGOSLAVIA Team Leader: R. C. Lyness Deputy: N. Routledge

Position 4/13; Score 231/336; Medals 1G, 2S, 4B
Michael J. Cullen (3) Winchester C Simon P. Norton (1) Eton C
Anthony L. Davies (3) Manchester GS Patrick J. Phair (2) Winchester C
David W. Garland (3) Eton C George Cameron KEVI GS, Stratford
Smith (3)
Robert Hill Manchester GS Malcolm J. Manchester GS
Williamson (2)

1968 USSR Team Leader: N. Routledge Deputy: David Monk

Position 4/12; Score 263/320; Medals 3G, 2S, 2B
Clifford C. Cocks (2) Manchester GS William Porterfield (1) Westminsters
Elwyn H. Davies (3) Manchester GS Michael Proctor Shrewsbury S
Noel Leaver (3) Burnley GS John Scholes (2) Winchester C
Simon P. Norton (1) Eton C Malcolm J. Manchester GS
Williamson (1)

1969 ROMANIA Team Leader: F. R. Watson Deputy: L. Beeson

Position 5/14; Score 193/320; Medals 1G, IS, IB
David J. Aldous (2) Exeter S John F. Segal Dulwich C
Patrick M. Bennett RGS, Newcastle Peter D. Smith (3) Manchester GS
A. J. Mclsaac Charterhouse Alan G. Trangmar Dulwich C
Simon P. Norton (1) Eton C Nick S. Wedd Kelly C, Tavistock

1970 HUNGARY Team Leader: B. Thwaites Deputy: Margaret Hayman

Position 6/14; Score 180/320; Medals 1G, OS, 6B
Charies J. K. Batty (3) Rugby S David Grubb Greenford GS
Stuart Bell Dartford GS John Proctor QMGS, Walsall
Daniel Jeremy M. Edwards RGS, Newcastle John F. Segal Dulwich C
Stephen B. Furber Manchester GS Bernard W. City of London S
Silverman (1)

1971 CZECHOSLOVAKIA Team Leader: Frank Budden Deputy: Peter Reynolds

Position 5/15; Score 110/320; Medals 0G, IS, 4B
David J. Allwright (3) Rugby S David J. Jackson (2) Perse S
Stuart Bell Dartford GS Nicholas S. Manton Dulwich C
Daniel Jeremy M. Edwards RGS, Newcastle Angus H. Rodgers (3) Royal Belfast
Acad. Inst.
Christopher R. Hills (3) Dulwich C Colin W. Vout (3) Dulwich C

1972 POLAND Team Leader: R. C. Lyness Deputy: Margaret Brown

Position 5/14; Score 179/320; Medals 0G, 2S, 4B
David J. Allwright (2) Rugby S David J. Jackson (2) Perse S
G. Mungo Carstairs George Peter Jackson RGS, Newcastle
Watsons C
David J. Goto (3) St. Pauls S Andrew L. James (3) Sherborne S
Ian J. Holyer (3) St. Benedicts S James E. Macey (3) Nottingham HS

1973 USSR Team Leader: Frank Budden Deputy: R. W. Payne

Position 5/16; Score 164/320; Medals 1G, OS, 5B
Michael D. Beasley (3) Kingston GS David Pritchard (3) QMGS, Walsall
David J. Goto (1) St. Pauls S A. J. Scholl (3) Worth S
Guy Herzmark (3) Highgate S Malcolm K. Sparrow Millfield S
John Hurley Ernest Bailey GS Richard J. Treitel (3) Eton C

1974 GDR Team Leader: R. C. Lyness Deputy: David Monk

Position 9/18; Score 188/320; Medals 0G, IS, 3B
Michael P. Allen Woking County Michael A. Gray Perse S
Boys S
Andrew B. Apps (2) Kings S, Richard C. Mason Manchester GS
Canterbury
Michael D. Beasley (3) Kingston GS David J. Seal Winchester C
John E. Cremona (3) Perse S A. J. Wassermann (3) RGS, Newcastle
 Appendix 227

1975 BULGARIA Team Leader: R. C. Lyness Deputy: John Durran

Position 5/17; Score 239/320; Medals 2G, 2S, 3B
Pelham M. Barton (3) Mill Hill S John R. Rickard (1) City of London S
David G. Frankis (2) RGS, Newcastle David J. Seal (2) Winchester C
John Peter Hartley (3) Manchester GS Paul Verschueren Manchester GS
Jonathan J. Hitchcock (1) Kingston GS David S. Walker (3) Manchester GS

1976 AUSTRIA Team Leader: R. C. Lyness Deputy: Colin Goldsmith

Position 2/19; Score 214/320; Medals 2G, 4S, IB
Patrick L. H. Brooke (3) Winchester C Jonathan Kingston GS
Hitchcock (2)
Douglas F. Easton (2) St. Edmunds C, Richard C. Mason (1) Manchester GS
Ware
Simon C. Farmbrough (2) Oundle S John R. Rickard (1) City of London S
David G. Frankis (2) RGS, Newcastle Alex J. Ryba Manchester GS

1977 YUGOSLAVIA Team Leader: R. C. Lyness Deputy: Terry Heard

Position 3/21; Score 190/320; Medals 1G, 3S, 3B
Richard E. Borcherds (2) KES, Brian A. King (3) St. Pauls S
Birmingham
Andrew G. Buchanan (2) Edinburgh Acad. Richard Wyggeston SFC
Pennington (3)
Alan G. Bustany Sullivan Upper S John R. Rickard (1) City of London S
Philip E. Gibbs (2) Currie HS Paul M. Stickland Dulwich C

1978 ROMANIA Team Leader: R. C. Lyness Deputy: John Hersee

Position 3/17; Score 201/320; Medals 1G, 2S, 2B
Richard E. Borcherds (1) KES, Philip E. Gibbs (2) Currie HS
Birmingham
Alan J. Dix (2) Howardian HS, Martin P. Gilchrist St. Albans S
Cardiff
Richard M. Durbin (3) Highgate S Richard Wyggeston SFC
Pennington (3)
Clive A. Frostick Dulwich C Peter P. Taylor Dulwich C

1979 UK Team Leader: Colin Goldsmith Deputy: Terry Heard

Position 4/23; Score 218/320; Medals 0G, 4S, 4B
Nigel Boston Manchester GS W. D. K. Green St. Albans S
Henry G. Bottomley Westminster S Colin H. A. Hogben Kent C
Richard M. Durbin Highgate S Paul Taylor (3) RGS, H. Wycombe
Clive A. Frostick Dulwich C Nigel C. Westbury Canford S

1980 MONGOLIA (cancelled)

[1980A FINLAND Team Leader: David Cundy Deputy: J Hersee

KCS, Wimbledon Colin H. A. Hogben Kent C
Graham M. Clemow
Kingston GS Jane Mills Leighton Park S
Thomas Davy
Perse S Malcolm Riley Manchester GS
Johnathan M. Edwards
Bishop Veseys S Christopher P. Smith St. Johns C,
David J. Harvey
Southsea

1980B LUXEMBOURG Team Leader: David Monk Deputy: Graham Howlett

Westminster S John Lister St. Albans S
Henry G. Bottomley
St. Albans S Fergus R. Mclnnes Royal HS,
W. D. K. Green
Edinburgh
City of London S Jeremy C. Rickard City of London S
Peter B. Kronheimer
Alleynes S, Stone Richard L. Taylor Magdalen C ]
Bernard Leek

1981 USA Team Leader: R. C. Lyness Deputy: John Hersee

Position 3/27; Score 301 /336; Medals 3G, 4S, IB
Ranelagh S Imre B. Leader (2) St. Pauls S
Mark David Bennet (1)
Eton C Stephen J. Kings S,
W. Tim Gowers (1)
Montgomery-Smith (2) Peterborough
Rugby S Jeremy C. Rickard (1) City of London S
Ian R. Jackson (2)
City of London S S. Paul Smith (3) RGS, Newcastle
Peter B. Kronheimer (2)
228 Appendix

1982 HUNGARY Team Leader: R. C. Lyness Deputy: David Monk

Position 10/30; Score 103/168; Medals 0G, OS, 4B
Paul N. Balister (3) KCS, Wimbledon Piers J. Coxon (3) Dulwich C
D. A. Chalcraft (3) Latymer Upper S William Sutherland (3) Monmouth S

1983 FRANCE Team Leader: R. C. Lyness Deputy: David Monk

Position 11/32; Score 121/252; Medals 0G, 3S, IB
Paul N. Balister (2) KCS, Wimbledon Alexander S. Clark (2) Charterhouse
Richard S. Biswas (2) Leighton Park S Ian J. Leary Ormskirk GS
Mark V. Bravington City of London S Alison McDonald (3) South Park SFC

1984 CZECHOSLOVAKIA Team Leader: David Monk Deputy: Frank Budden

Position 6/34; Score 169/252; Medals 1G, 3S, IB
Paul N. Balister (1) KCS, Wimbledon Marcus Moore (2) Manchester GS
Richard S. Biswas (2) Leighton Park S Matthew J. Millfield S
Richards (2)
Michael Harrison Loughborough GS Ian D. B. Stark (3) Winchester C

1985 FINLAND Team Leader: R. C. Lyness Deputy: David Monk

Position 10/38; Score 121/252; Medals 0G, 2S, 3B
Chris Kilgour Kings S, Matthew J. Richards Millfield S
Gloucester
John Longley (3) Portsmouth GS Alex Selby (3) City of London S
Marcus Moore (3) Manchester GS Ian D. B. Stark Winchester C

1986 POLAND Team Leader: David Monk Deputy: Paul Woodruff

Position 11/37; Score 141/252; Medals 0G, 2S, 3B
James Angus Wells Cath. S John Longley (3) Portsmouth GS
Kevin M. Buzzard (3) RGS, H. Wycombe Tom Roughan (3) Leeds GS
Dominic Joyce (2) QE Hospital Andrew Smith (2) City of London S

1987 CUBA Team Leader: R. C. Lyness Deputy: Terry Heard

Position 10/42; Score 182/252; Medals 1G, 2S, 2B
James Angus City of London S George Russell (2) KES,
Birmingham
Kevin M. Buzzard (1) RGS, H. Wycombe Andrew Smith (2) City of London S
Gareth McCaughan (3) Lincoln Christs Gerard Thompson (3) St. Georges C,
Hospital S Weybridge

1988 AUSTRALIA Team Leader: David Monk Deputy: Paul Woodruff

Position 11/49; Score 121/252; Medals 0G, 3S, 2B
Colin Bell (3) Trinity S, Chris R. Nash (3) KES,
Croydon Birmingham
Malcolm J. Law (2) KES, Oliver M. Riordan (2) St. Pauls S
Birmingham
Gareth McCaughan (2) Lincoln Christs Joshua R. X. Ross City of London S
Hospital S

1989 FRG Team Leader: David Monk Deputy: Paul Woodruff

Position 20/50; Score 122/252; Medals 0G, 2S, IB
Katherine Christie Portsmouth GS Chris R. Nash (2) KES,
Birmingham
Vin de Silva (3) Dulwich C Oliver M. Riordan (2) St. Pauls S
Clive R. Jones Dulwich C John Simcox Chase HS,
Malvern

1990 CHINA Team Leader: Peter Shiu Deputy: Paul Woodruff

Position 10/54; Score 139/252; Medals 2G, OS, 2B
Vin de Silva (1) Dulwich C Tom Leinster Lancing C
Michael Fryers (3) Altrincham GS Oliver M. Riordan (1) St. Pauls S
Alan Iwi Westminster S Amites Sarkar (3) Winchester C
 Appendix 229

1991 SWEDEN Team Leader: Tony Gardiner Deputy: Paul Woodruff

Position 17/56; Score 142/252; Medals 1G, OS, 2B
Michael Fryers (1) Altrincham GS Luke T. Pebody Rugby S
Oliver T. Johnson KES, Adam P. Shepherd KES,
Birmingham Birmingham
Robin Michaels (3) Haberdashers Steven P. Wilcox (3) Portsmouth GS
Askes S

1992 RUSSIA Team Leader: Tony Gardiner Deputy: Christopher Bradley

Position 5/58; Score 168/252; Medals 2G, 2S, 2B
Oliver T. Johnson (3) KES, Karen M. Page (3) S. Bromsgove HS
Birmingham
Eva R. Myers (1) Streatham Hill Luke T. Pebody (2) Rugby S
& Clapham HS
Robin Michaels (2) Haberdashers Mark J. Walters (1) WealdS
Askes S

1993 TURKEY Team Leader: Adam McBride Deputy: Christopher Bradley

Position 14/73; Score 118/252; Medals OG, 3S, 3B
Tom Fisher (2) Exeter S Alex Paseau (3) St. Pauls S
Alistair Flutter (3) Hills Road SFC Luke T. Pebody (2) Rugby S
Catriona Maclean (2) Harrogate GS Chuan-Tse Teo (3) Dulwich C

1994 HONGKONG Team Leader: Tony Gardiner Deputy: Vin de Silva

Position 7/69; Score 206/252; Medals 2G, 2S, 2B
Ed Crane (3) Colchester RGS Catriona Maclean (1) Harrogate GS
Matthew Fayers (3) Wilsons S Joseph Myers (1) Rutlish S
Ben Green (2) Fairfield GS Jacob Shapiro (2) Westminster S

1995 CANADA Team Leader: Tony Gardiner Deputy: Christopher Bradley

Position 10/73; Score 180/252; Medals 2G, IS, 3B
Ed Crane (1) Colchester RGS Peter Keevash (3) Leeds GS
Matthew Fayers (3) Wilsons S Joseph Myers (1) Rutlish S
Ben Green (2) Fairfield GS Louisa Orton (3) Northgate HS

1996 INDIA Team Leader: Adam McBride Deputy: Philip Coggins

Position 5/75; Score 161/252; Medals 2G, 4S
David Bibby (1) Ysgol Rhiwabon John Haslegrave (2) King Henry
VIII S
Michael Ching (1) Oundle S Hugh Robinson (2) King Henry
VIII S
Toby Gee (2) John of Gaunt S Paul Russell (2) St. Brides HS
 DATE DUE / DATE DE RETOUR

SEP 1 i 199#

CARR MCLEAN
38-297
Mathematical Olympiad competitions started in Hungary at the end of the
nineteenth century; national Olympiads are now held in over a hundred
countries, and there are numerous international events. Olympiads chal¬
lenge able secondary school pupils to develop their mathematical skills by
solving problems. Olympiad problems are unpredictable and have no obvi¬
ous starting point; although their solution may require little more than ordi¬
nary school mathematics, they are definitely ‘problems’ rather than routine
exercises, and this can make them seem rather hard. The Mathematical
Olympiad Handbook introduces readers to these challenging problems and
aims to convince them that Olympiads are not just for a select minority.

The book contains problems from the first 32 British Mathematical

Olympiad (BMO) papers 1965-96 and gives hints and outline solutions to
each problem from 1975 onwards. An overview is given of the basic math¬
ematical skills needed, and a list of books for further reading is provided.
Working through die exercises provides a valuable source of extension and
enrichment for all pupils and adults interested in mathematics.

Tony Gardiner has been involved in die British Mathematical Olympiad

competitions for several years. He was leader of the UK International
Mathematical Olympiad team from 1990 to 1995 and is currently Vice-
President of the World Federation of National Mathematics Competitions.
In recognition of his work since 1988 in strengthening national competitions
within the UK he received the Paul Erdos National Award in 1995.

ALSO PUBLISHED BY OXFORD UNIVERSITY PRESS

Discovering mathematics
A. Gardiner

What to solve?
Judita Cofman

Numbers and shapes revisited

Judita Cofman

Mathematics masterclasses: stretching the imagination

Michael Sewell (ed.)

OXFORD UNIVERSITY PRESS