INTRODUCTION

The South African Mathematical Society has the responsibility for se-
lecting and training teams to represent South Africa in the annual
Mathematical Induction
International Mathematical Olympiad (IMO). for the
The process of finding a team to go to the IMO is a long one. It begins
with a nationwide Talent Search, in which students are sent sets of
Olympiad Enthusiast
problems to solve. Their submissions are marked and returned with
comments, full solutions and a further set of problems. The principle David Jacobs
behind the Talent Search is straightforward: the more problems you
solve, the higher up the ladder you climb and the closer you get to Mathematical induction is a simple but powerful reasoning tool that
selection.
is useful in solving a wide variety of problems: arithmetical, algebraic
The best students in the Talent Search are invited to attend Mathe- and geometric. The standard school textbook approach to this topic
matical Camps in which specialised problem-solving skills are taught. has been the rather limited use of mathematical induction to prove
The students also write a series of challenging Olympiad-level problem
papers, leading to selection of a team of six to go to the IMO. number-theoretic statements like inequalities and formulae for series.
The approach of most university textbooks is to discuss the principle
The booklets in this series cover topics of particular relevance to Math-
ematical Olympiads. Though their primary purpose is preparing stu- of mathematical induction as part of an axiomatic system for reasoning
dents for the International Mathematical Olympiad, they can with with the natural numbers, focusing on the equivalence of this principle
profit be read by all interested high school students who would like to with other number theoretic statements. For the Olympiad enthusiast,
extend their mathematical horizons beyond the confines of the school or any school student looking for some mathematical enrichment, nei-
syllabus. They can also be used by teachers and university mathemati- ther of these approaches is particularly satisfactory, the former being
cians who are interested in setting up Olympiad training programmes .
and need ideas on topics to cover and sample Olympiad problems. too elementary and limited, the latter being too abstract and technical.
Titles in the series published to date are
In the light of this, my approach is to introduce and discuss math-
1. The Pigeon-hole Principle, by Valentin Goranko ematicai induction largely in the context of problem-solving, and in
2. Topics in Number Theory, by Valentin Goranko particular to discuss problems of the type that appear regularly in
3. Inequalities for the Olympiad Enthusiast, by Graeme West
4. Graph Theory for the Olympiad Enthusiast, ·mathematics competitions. In keeping with this approach, the first
by Gzaeme West section of this booklet starts with a particular problem, and extracts
5. Functional Equations for the Olympiad Enthusiast, from its solution some of the basic essence of the principle of mathemat-
by Graeme West ical induction. The second section introduces mathematical induction
6. Mathematical Induction for the Olympiad Enthusiast, proper, and contains some of the more routine problems in the subject.
by David Jacobs
The third section discusses some of the variants (in fact, most are ac-
Details of the South African Mathematical Society's Mathematical Tal- tually logical equivalents) of the principle of mathematical induction,
ent Search may be obtained by writing to again opting for a problem-solving rather than a theoretical treatment,
Mathematical Talent Search in keeping with the style of this booklet. The last section focuses on
Department of Mathematics and Applied Mathematics some harder problems, where it is less obvious to see exactly where and
University of Cape Town how the principle is used.
7700 RONDEBOSCH
The International Mathematical Olympiad Talent Search is sponsored
by the Old Mutual.
J H Webb
May 1996 1
 1. The Tromino Problem Next consider a 4 x 4 grid with one square shaded. We can divide this
grid into four 2 x 2 grids as shown in Figure 4. One of these contains a
shaded square, and hence this grid can be covered by a tromino. Now
place a tromino in the centre of the 4 x 4 grid as shown in Figure 5.
Notice that the three remaining 2 x 2 grids each have exactly one square
covered, and hence can be covered by trominoes as in Figure 6. Hence
we have solved the 4 x 4 case.

FIGURE 1
m FIGURE 4 FIGURE 5
~ FIGURE 6

Suppose you have an 8 x 8 grid of squares as shown in Figure 1. Pick an

arbitrary square and shade it black. You are left with 63 squares. We
now pose the following question: Is it possible to cover the remaining 63 We use the same trick for the 8 x 8 grid: Divide it into four 4 x 4
squares with 21 shapes ofthe form given in Figure 2 (we shall call these grids, and position a tromino in the centre in such a way that each of
four grids have one square covered (see Figure 7). We can now use our
shapes trominoes), with all remaining squares covered and no overlaps?
procedure for the 4 x 4 grid to cover each of the four grids to yield the
solution shown in Figure 8.

tFEbLB FIGURE 2

To solve this problem, first consider a 2 x 2 grid, and shade an arbitrary

)
square black. We are then left with 3 remaining squares which form
a tromino. Hence it is always possible to cover a 2 x 2 grid with one
square shaded with one tromino (see Figure 3).
FIGURE 7 FIGURE 8

~ Now that we have a solution, let us reflect a bit on some ofthe essence of
the problem and its solution. Notice that this problem co~ceals a more
FIGURE 3
general problem: It is possible to extend this procedure to a 16 x 16
2
3
 Suppose that
grid, or in fact, to any grid with dimensions 2n x 2n, where n E N. (N
is the set of natural numbers, that is, the set {1, 2, 3, ... }). (1) P(1) is true (that is, can be proved),
(2) For each mEN, P(m) :=:::::} P(m + 1) (that is, we can prove
Let P(n) be the statement: "For every 2n x 2n square grid with an P(m + 1) on the assumption that P(m) is true).
arbitrary square shaded, it is possible to cover the rest of the squares
Then all the propositions of the sequence must be true, that is, P( n) is
with trominoes." Notice that the statement P(n) depends on n, a
natural number. What we have shown is that P( n) is true for all true for all n EN.
n EN. We did this as a two step process:
(i) We showed P(1) is true, that is, we could solve the 2 x 2 case.
(ii) We showed, that if P(k) was true for some kEN, then P(k+ 1) Notice that we regard the PMI as an axiom, that is, we accept it as
was true. In our particular case, we showed that if we could valid without proof. In fact, the PMI (or one of its many equivalents)
solve the problem for a 2k x 2k grid, we could solve the problem is regarded as one of the basic axioms in any workable mathematical
for a 2k+ 1 x 2k+1 grid by dividing it up into four 2k x 2k grids.
theory that involves the natural numbers.
Using these assumptions, note that since P(1) is true, P(2) would be
true by (ii). Now, since P(2) is true, by (ii) we have P(3) is true and
so on to conclude that P( n) is true for all n E N. We now give an example of a standard result proved using the PMI.

It is essentially this two-step process that forms the heart of any proof
by mathematical induction. We will formalize this method in the next
section, but first try to tackle the following exercise to get the feel of
this method of proof.
EXAMPLE 2.1. Prove that 12+22+32+. ·+n2 = n(n + 1)(2n + 1)
for each n E N. 6

EXERCISE 1.
Given an n x n grid (n ~ 3) with an arbitrary square shaded, show
SOLUTION. For each n EN, let P(n) be the statement
that it is possibie to coyer up the remaining squares using only 3 x 1
and 2 x 1 rectangles.
12 + 22 + 32 + ... + n2 = n(n + 1)(2n + 1)
6
2. Mathematical Induction Introduced
1 3
P(1) is true since 12 = x: x . Now suppose that for some mEN,
We can now formally state the Principle of Mathematical Induction.
P( m) is true, that is
THE PRINCIPLE OF MATHEMATICAL INDUCTION
(PMI). Consider an infinite sequence of mathematical propositions
12 + 22 + 32 + ... + m2 = m( m + 1)(2m + 1-)
P(1), P(2), P(3), ... 6
5
4
Now But, since mx+mx 2 l (m+ l)x (take m = 1 and x = 0.5 for example),
we cannot conclude P( m + 1) is true. Hence we are stuck!
12 + 22 + 32 + · · · + m 2 + (m + 1) 2
m(m + 1)(2m + 1) ( ) A less obvious approach is to let Q(n) be the stronger statement
= +m+l 2 "(1 + x)n ~ nx + 1". Again Q(l) is true since 1 + x ~ x + 1. Suppose
6
since P( m) is true for some mEN, Q(m) is true. Then

2m3 + 3m 2 + m + 6m 2 + 12m+ 6 (1 + x)m+l = (1 + x)m(l + x)

6 ~ (mx + 1)(1 + x) [P(m) is true]
2m3 + 9m 2 + 13m+ 6
= mx 2 + (m + l)x + 1
6
_ (m + l)(m + 2)(2m + 3) ~(m+l)x+l.
6
_ [m + l]([m + 1] + 1)(2[m + 1] + 1) Hence Q( m + 1) is true, and by the PMI, Q( n) is true for all n E N.
- 6 To complete the problem, one need only notice that nx + 1 > nx for
all n EN. Thus P(n) is true for all n EN. 0
Thus P(m + 1) is true. Hence by the PMI, P(n) is true for all n EN,
which is precisely the statement we wish to prove. 0
EXERCISES 2.
When using the PMI, we call the prooffor step (1) [P(l) is true] the base 1 1 1 1 n
case, and the proof for step (2) [P(m) ==> P(m + 1)] the induction (1) Show t h a t - - + - - + - - + · · · + =--for
1x2 2x3 3x4 n(n + 1) n+ 1
step. In the induction step P(m), the statement we assume true, is allnEN.
called the induction hypothesis. The variable on which the statement (2) Show that 13 + 23 + · · · + n 3 = (1 + 2 + · · · + n) 2 for all n EN.
P depends (in our case n) is called the induction parameter.
(3) Show that 2n+4 ~ (n + 4) 2 for all n EN.
Sometimes the selection of a statement P( n) is not all that obvious, as (4) Show that an - bn is divisible by a - b for all a, b, n E N with
the following example shows. a ::j:. b.
(5) The Fibonacci sequence
EXAMPLE 2.2. Show that, if x is a positive real number, then
(1 + x )n ~ nx for all n E N. F1 = 1, F2 = 1, F3 = 2, F4 = 3, Fs = 5, Fs = 8, ...

is defined as follows: F1 = F2 = 1 and Fn = Fn-2 + Fn-1 for

SOL UTI ON. Let us first try putting P( n) to be the statement
all n ~ 3. Prove that the following statements are true for all
"(1 + x)n ~ nx". P(l) is true since 1 + x ~ x. Now suppose for
some mEN, P(m) is true. Then nEN.

(1 + x)m+l = (1 + x)m(l + x) (a) F1 + F2 + · · · + Fn = Fn+2- 1.

(b) 1Fn+2Fn-(Fn+l) 1=1.
2 r· ,
; .. ! 1 ,
1
~ mx(1 + x) j.
[P(m) is true]
= mx+mx 2 • (c) 5 divides n if and only if 5 divides Fn.

6
7
 (6) Prove that for all n EN, EXAMPLE 3.1. Show that 3n > n 3 for all n EN, n ~ 4.
1 X 3 X 5 X · · · X (2n - 1) 1
--:::------:----:::---"----:::----'- < -
- J2ri'
0

2 X 4 X 6 X · · · X 2n SOL UTI ON. The result is true for n = 4 since 81 > 64. Now suppose,
(7) [ Mathematical Induction, Slinko] Show that for some m ~ 4, 3m > m 3 . Then
1 1 1 1
-
12+ -22+ -32 + · · · + -n2< 2 (a) 3m+l = 3(3m) >3m3 .
for all n EN. -,.., .,,..',
,!r, TIJ
""ci· -s-.~ . . i Also, since m > 3, we have
L-.. .·" .
r o. I 1 I
A
3. Some Variations on Mathematical (b) m3 >3m 2.
Induction
Again, using m > 3, we have m(m- 3) > 1 and hence
In this section, we look at a range of variations on the basic principle
of mathematical induction. In all cases we will give a proof that the (c) m3 ~ m 2 > 3m + 1.
altered principle can be derived from the PMI.
Thus
The first variation is to alter the base case (i.e. P(1)). This can be
formally stated as follows: 3m 3 = m3 + m3 + m3 > m3 + 3m 2 + 3m + 1 = (m + 1)3 .
THE PRINCIPLE OF MATHEMATICAL INDUCTION - using (b) and (c), and hence by (a),
VARIANT 1 (PMil). Consider an infinite sequence of mathemati-
cal propositions 3m+l > (m + 1)3.
P(1), P(2), P(3), ...
Let k E N. Suppose that Thus by the PMil, 3n > n 3 for all n EN, n ~ 4. D
(1) P(k) is true (that is, can be proved),
(2) For each mE N,.m ~ k, P(m) ==> P(m + 1) (that is, we can The PMI can be generalized even further.
prove P(m + 1) on the assumption that P(m) is true).
Then P(n) is true for all n EN, n ~ k.
THE PRINCIPLE OF MATHEMATICAL INDUCTION -
VARIANT 2 (PMI2). Consider an infinite sequence of mathemati-
Proof. For each n E N, let Q( n) = P( n + k - 1). A check shows that cal propositions
the sequence Q(1), Q(2), Q(3), ... satisfies the conditions of the original P(1), P(2), P(3), ...
PMI, and thus Q( n) is true for all n E N, and hence P( n) is true for
all n ~ k. D Let a, dE N. Suppose that
(1) P(a) is true (that is, can be proved),
[A word of warning: The abbreviations PMI and PMil for the various (2) For each mEN, P(a + (m- 1)d) ==> P(a + md) (that is, we
principles of mathematical induction are not standard, and should not can prove P( a + md) on the assumption that P( a + (m - 1)d)
be used in mathematics competitions.] is true).

8 9
 A4r c - - - - - - .A5
Then P(a + (n -l)d) is true for all n EN. _

As B5,,:-------- ~ A6
Proof For each n E N, let Q( n) = P( a+ ( n- 1)d). A check shows that
the sequence Q(l), Q(2), Q(3), ... satisfies the conditions of the original
PMI, and thus Q(n) is true for all n EN, and hence P(a + (n -l)d) is
true for all n EN. 0 A2 86 ~----------J A7

One use, for example, of the PMI2 is to prove a statement P(n) true for A1 As
= =
all even natural numbers, using a 2 and d 2. In order to establish FIGURE 9
the result, we would have to prove that P(2) is true, and for any natural
number m, the truth of P(2m) implies the truth P(2m + 2). Now A1A2 ... Am+2Bm+3Bm+4 ... B2m+2 forms a (2m+ 2)-sided par-
polygon, and hence can be dissected into rhombi. Notice too that

EXAMPLE 3.2. [IMO 1979} A convex polygon with an even number Am+2Am+3Am+4Bm+3,
of sides, all equal to each other and with pairs of opposite sides parallel
is called a parpolygon . Prove that any parpolygon can be dissected into .,
rhombi. B2m+l A2m+2A2m+3B2m+2,
B2m+2A2m+3A2m+4Al
SOLUTION. Let P(n) be the statement "An n-sided parpolygon can are all rhombi, and hence the parpolygon with vertices
be dissected into rhombi". We will use PMI2 with a =
4 and d 2.=
P( 4) is true since a 4-sided parpolygon is automatically a rhombus. A1, A2, ... , A2m+4
Now suppose for some mEN, we have that P(2m+2) is true. Consider
a (2m+ 4)-sided parpolygon with vertices A1, A2, ... , A2m+4· Con- can be dissected into rhombi. Hence by PMI2, the result follows. 0
struct points Bmf.2, Bm+3, ... , B2m+2 inside the parpolygon such that
In all the variants of mathematical induction introduced so far, one of
Am+2Am+3 II Bm+3Am+4 II Bm+4Am+311· · · the induction steps involved assuming a statement is true and proving
· · · II B2m+2A2m+3 II A1A2m+4. that the immediate successive statement could be proven true based
on this assumption. In the next variant on the PMI, we modify this
and condition and have the truth of a statement in a sequence depending
on the truth of all of its predecessors.
Am+2Am+3 = Bm+3Am+4 = Bm+4Am+3 = · · ·
THE PRINCIPLE OF MATHEMATICAL INDUCTION,
· · · = B2m+2A2m+3 = AlA2m+4·
VARIANT 3 (PMI3). Consider an infinite sequence of mathemati-
cal propositions -·
(See Figure 9 for the case m = 2).
P(l), P(2), P(3), ...
10
11
Suppose that Another variant of the PMI is the Principle of the Smallest Number,
(1) P(1) is true (that is, can be proved), which though seemingly obvious is surprisingly useful. It is in fact
(2) For each mEN, [P(1) and P(2) and ... and P(m) :::::::::. P(m+ an equivalent of the PMI, although we will only prove that it is a
1)] (that is, we can prove P(m+ 1) assuming the truth of P(1), consequence of the PMI.
P(2), ... , P(m- 1) and P(m)).
Then all the propositions of the sequence must be true, that is, P( n) is THE PRINCIPLE OF THE SMALLEST NUMBER (PSN).
true for all n EN. Let S be a non-empty set of natural numbers. Then S has a smallest
member, that is, there is an m E S such that for all s E S, m :::; s.
Proof. Let Q(1) be the statement P(1), Q(2) the statement "P(1) and
P(2)", Q(3) the statement "P(1), P(2) and P(3)" and so on. A check Proof. Suppose S is a set of natural numbers with no smallest member.
shows that the sequence Q(1), Q(2), Q(3), ... satisfies the conditions of We will show that S is necessarily empty. Let P( n) be the statement
the original PMI, and thus Q( n) is true for all n E N, and hence P( n) "For all natural numbers k:::; n, k ft S''. Now P(1) is true, for if 1 E S,
is true for all n EN, as for each n EN, Q(n) ==> P(n). 0 1 would be the smallest member as all natural numbers are greater than
or equal to 1. Now suppose P(m) is true for some m E N. Suppose ~
EXAMPLE 3.3. Given any natural number n, define the sequence P(m + 1) were false, that is m + 1 E S. Then, as S has no smallest
member, there is a k E S, such that k < m + 1. But then k :::; m,
S1,S2,S3, ... contradicting P(m). Hence P(m+ 1) must be true. Hence by the PMI,
P( n) is true for all n E N, implying that S is empty. 0
as follows:

sk-1 . One of the most common uses of the PSN is the so-called "method of
s1 =n and, fork> 1, sk = -2- if s~-1 is even. infinite descent". Suppose we have a sequence of mathematical state-
{
sk-1 + 1 if sk-1 is odd. ments
P(1), P(2), P(3), ...
Prove that for all n E N with S1 = n, there is a k E N such that Sk = 1.
that we wish to prove true. If we are able to prove the following: "For
SOL UTI ON. Let P(j) be the statement "For a sequence defined as every k EN such that P(k) is false, there is am EN such that P(m)
above, if there is an i E N with S; = j, there is a k E N such that is false and m < k". Then we may conclude by the PSN that P(n) is
Sk = 1 ". P(1) is true since S 1 =
1. Now suppose there is an m EN true for all n E N. For if not, then the set of all k such that P(k) is
such that for all r :5 m we have that P(r) is true. Now set S1 = m + 1. false is a non-empty set ofthe natural numbers which, by the statement
1 above, has no smallest element, contradicting the PSN. We illustrate
If m + 1 is even, r = m; :=:; m, and hence using the fact that P(r) this method with the following example.
is true, and S 2 = r, we must have that P( m + 1) is true. If m + 1
is odd, then as m 2:: 1, m + 1 2:: 3 (m + 1 is odd), so we have that
2 EXAMPLE 3.4. {USA MATHEMATICAL OLYMPIAD, 1978} An
r = m; :5 m. Hence since S 3 =
r and P(r) is true, we have integer n is called good if we can write
again that P(m + 1) is true. Hence by the PMI3, P(n) is true for all
nEN. 0 a1 + a2 + · · · + ak = n,

12 13
where a1, a2, ... , ak are positive integers (not necessarily distinct) sat- Similarly, if m is odd, we can reach a contradiction by putting p =
isfying m-9 . 1 1 1
1 1 1 - - , and usmg the fact that 2' = '3 + '6 and
2
-+-+···+-=1.
a1 a2 ak
3 + 6+ 2a 1 + 2a2 + · · · + 2ak = 9+ 2p = m.
Given the information that the integers 33 through to 73 are good, prove
every integer greater than 32 is good.
Thus both cases lead to a contradiction, so we must have that S is
SOLUTION. Call an integer bad if it is not good. What we need to empty, which implies all numbers greater than 32 are good. D
show is that there are no bad integers greater than 32. Suppose for a
contradiction, that the set S of all bad integers greater than 32 is not
The last variant of the PMI we introduce is the rather tricky method
empty. By the PSN, it has a smallest member m, say. Note that by
of "Backwards" Induction. Here the il).duction step involves using the
the given information, m ~ 74.
truth of a statement in a sequence to prove its predecessor.
8
Now suppose m is even. Put p = m; . Now p ~ 33, and since m is
THE PRINCIPLE OF BACKWARDS MATHEMATICAL IN-
the smallest integer in S, and p < m, p ~ S, which means p is good.
DUCTION (PBMI). Consider an infinite sequence of mathematical
So choose a 1 , ... , ak such that
propositions
1 1 1 P(1), P(2), P(3), ...
-+-+
a1 a2
.. ·+-=1
ak Suppose that
and (1) For some infinite subsetS ofN, P(s) is true for all s E S.
a1 + a2 + · · · + ak = p. (2) For each m EN, P(m + 1) ==} P(m) (that is, we can prove
P(m) on the assumption that P(m + 1) is true).
Notice that
Then all the propositions of the sequence must be true, that is, P(n) is
true for all n E N.
4 ;!- 4 + 2a1 + 2a2 + · · · + 2ak = 8 + 2p = m
and Proof. LetS= {x1,x2, .. . ,xn, .. .}, and without loss of generality as-
1 1 1 1 1 sume that x 1 < x 2 < x 3 < . . . . We can rearrange the sequence of
-+-+-+-+
4 4 2al 2a2
.. ·+-
2ak
statements as follows:
1 1 1 1 1 1
=-+-+-(-+-+
4 4 2 a1 a2
.. ·+-)
ak
P(xl), P(x1- 1), P(x1- 2), ... , P(1), P(x2), P(x2- 1), ...
1 1 1 ... , P(x1 + 1), P(x3), ...
=- +- +-
4 4 2
=1. Now let Q(1) = P(x 1), Q(2) = P(x 1 - 1), Q(3) = P(x1- 2), and in
general let Q( n) be the nth statement in the above seq_yence. A check
But this shows that m is good, a contradiction. shows that the sequence Q(1), Q(2), Q(3), ... satisfies the conditions of

14 15

_....
the original PMI, and thus Q( n) is true for all n E r:J, and hence P( n) Now suppose P(2m) is true for some mE f:J. Then
is true for all n E f:J. D
Xt + X2 + ... X2m+I
2m+l
= ~(Xl + X2 + · · · + X2m X2m+1 + X2m+2 + · · · + X2m+I)
The PBMI is usually used as follows: One of the variants of the PMI 2 2m + 2m
is used to prove part ( 1) for some set S (popular choices are multiples > \ /( Xt + X2 ~~ · · + X 2m) X ( X2m+l +X 2m;~+ · · · + X2m+I )
or powers of a fixed number), and then part (2) is proved directly.
since P(2) is true

~ y(xtX2 ... X2m)~ X (x2m+tX2m+2 ... X2m+I)~

EXAMPLE 3.5. Prove that ( since P(2m) is true
= (XtX2 ... X2m+I) ,m'+i .
Xt + x2 + ... + Xn ~ (XtX2 .. . xn)~
n Hence P(2m+l) is true, and thus P(s) is true for all s E {2, 4, 8, 16, ... }.

Now suppose for some m E r:J, P(m + 1)

is true. Given positive real
for all n E f:J, where Xt, x2, ... , Xn are all positive -real numbers. This Xt +
X2 + · · · + Xm .
result is known as the arithmetic-geometric mean inequality. numbers Xt, x2, ... , Xm, put Xm+l = . Smce Xm+l
m
is the mean of Xt, x2, ... , Xm, the mean of Xt, x2, ... , Xm is the same as
the mean of Xt, X2, ... , Xm+l, that is,

Xt + X2 + · · · + Xm _ Xt + X2 + · · · + Xm+l
So LUTI0 N. Let P ( n )b e "x1+x2+···+xn
~
(
n . We
XtX2 .. . Xn )'" m - m+1
n
show first by the PMI that the result is true for all P(s ), where s E
{2, 4, 8, 16, ... }.· Thus

Xt + X2 + · · · + Xm+l
Xm+t = m+1
Firstly P(2) is true, since
~ (x1x2 ... Xm+t) m~ 1 using P(m + 1)
~ (XtX2 ... Xm) m~I (xm+t) m~I.
(xt - x2) 2 ~ 0
==::} (x1) 2 + 2XtX2 + (x2) 2 ~ 4XtX2 Thus
m+l >
(Xm+l )1--' _ (XtX2 · · · Xm )-'
m+I,
==::} (xt + x2) 2 ~ 4xtx2
that is,
==::} Xt ~ X2 ~ yfXtX2.
(xm+t)m'+l ~ (XtX2 ... Xm)m~l,
16 17
and hence
other but do not overlap. Prove that these counters can be
Xm+l=
Xt + X2 + · · · + Xm .l.. painted with four colours so that no counters of the same colour
m ->(XtX2···Xm)m. are touching.
Thus P(m) is true, and hence by PBMI the result follows. 0 (9) A real-valued function f(x) is called convex if for any a, b real
a+b f(a)+f(b) .
numbers, f ( --) ~ (see Ftgure 10 for a graph-
2 2
ical idea of the notion of a convex function). Prove that for
EXERCISES 3. any convex function f(x) and any n E N, and real numbers
(1) Recall the Fibonacci sequence F 1 , F 2 , F3, ... introduced in ex- Xt, X2, ... 'Xn,
ercises 2 (5). Prove that for n, mEN, n > 1 we have
J(l + X2 + · · · + Xn) ~ f(xt) + f(x2) + · · · + f(xn).
Fn+m = Fn-lFm + FnFm+l· n n
(2) [Math. Digest, Jan. 1984] Prove that for all n ~ 4, it is possible (This result is known as Jensen's Inequality.)
to dissect any non-equilateral triangle into n isosceles triangles.
(3) [ Mathematical Induction, Slinko] On each planet in a plan- ~

etary system consisting of an odd number of planets there is a !:1:-t(x.)

astronomer observing the nearest planet. The distances be-
tween each pair of planets are all different. Prove that at least
one planet is not observed by an astronomer. H~)+·HI:.)
a.
(4) A sequence a 1 ,a 2,a 3 , ... is defined as follows: a 1 = -2, H~)
a2 =-16 and an+2 = Ban+l - 15an for n E N. Show that
an = 3" - 5" for all n E N.
(5) [ Mathematical Induction, Slinko] Suppose for a positive 0.. O..+b
real number a, a+ ~ is an integer. Show that a" + _!__ is an a
a a"
integer tor all n. E N. FIGURE 10
(6) [Tournament of Towns, 1979] Consider all the possible subsets
of the set { 1, 2, ... , n} which do not contain any consecutive (10) Suppose I have an unlimited supply of coins of denominations a
numbers. For each of these subsets, the product of their ele- cents and b cents, where a and b are natural numbers with high-
ments is squared. Prove that the sum of all of these squared est common factor 1. Show that I can always pay an amount
products is (n + 1)!- 1. of n cents, where n is any integer greater that ab- a- b.
(7) [ Mathematical Induction, Slinko] Prove that an arbitrary (11) [Moscow Mathematical Olympiad, 1979] The area of the union
sum of n cents can be paid by 3 cent and 5 cent coins, where n of several circles (that is, the area of covered by all points lying
is an integer greater than 7. '\ inside at least one of the circles) equals 1. Prove that it is
(8) [Mathematical Induction, Slinko] Circular counters of equal possible to choose several of them that do not intersect each
diameter are lying on a table so that some of them touch each other and whose total area is greater than ~.
18
19
4. Harder Induction Problems
Suppose all the numbers in the collection are odd. Then the collection
In this last section we will look at some rather difficult problems that
X1 + 1 X2 + 1 X2n+l +1
can be solved using mathematical induction. Sometimes the choice --2-,--2-, ... , 2
of the induction parameter can be quite subtle, as the next example
shows. has property P and its maximum value is less than or equal to m.
. Xi + 1
Hence by our assumptwn all the - --'s are equal, and hence all x;'s
2
EXAMPLE 4.1. A collection of natural numbers is said to have are equal.
property P if the following holds: If any member of the collection
is removed, the remaining numbers may be divided into two groups Thus by the PMI3, the result holds. 0
with equal sums. Prove that any collection of 2n + 1 natural numbers
x 1 , x 2 , ... , x 2 n+l with property P, necessarily has all members equal to
each other.
Another complication is that a problem may have two or more varying
parameters. One approach is to first use the PMI on one of these pa-
rameters, and then apply the PMI to the other. Occasionally, though,
SOL UTI ON. We use the PMI3 on the maximum value of the two parameters can be combined into one, as in the next problem.

X1, X2, ... , X2n+l·

EXAMPLE 4.2. [Mathematical Induction, Slinkoj Let x1, ... , Xm
and Yl, ... , Yn be natural numbers such that
If the maximum value of x 1 , x 2 , ... , x 2 n+l is 1, then all members of
the set must be equal to 1, and the result holds. Suppose the result
holds for all sets of 2n + 1 natural numbers whose maximum value is X1 + · · · + Xm = Yl + · · · + Yn < mn.
less than or equal tom. Let x 1 , x 2 , ... , x 2 n+l be a collection of natural
Prove that it is possible to cross out from the equation at least two terms
numbers with maximum value m + 1 that satisfies property P. Let
(but not all of them) such that the equation remains true.
S = X1 + X2 + · · · + X2n+l·
First note that 'the numbers in the collection are either all even or all SOL UTI ON. We will apply the PMil to s = m + n. As
odd: For, if we pick i,) such that i,j E {1, 2, ... , 2n + 1}, then there
are natural numbers A; and Aj such that S x; + 2A;= =
Xj + 2Aj by m ~ x1 + · · · + Xm < mn,
property P, that is x; and Xj must have the same parity.
we conclude that n > 1 and similarly m > 1. Hence the minimum value
Suppose all the numbers in the collection are even. Then the collection of s =4, when m n = =
2. For this case the result holds trivially as
=
the only possible equations are 1 + 1 1 + 1 and 1 + 2 1 + 2. Now, =
X1 X2 X2n+l suppose the result holds for some s = k E N, k ~ 4. Let
2'2'" .. '2 \
has property P and its maximum value is less than or equal to m. Hence
X1 + · · · + Xm = Yl + · · · + Yn < mn,
by our assumption all the ~i 's are equal, and thus all x;'s are equal. where m + n = k + 1. We may suppose, without lo~s of generality,
that x 1 and Y1 are the largest numbers on the left-hand and right-hand
20
21
sides respectively of the above equation. If x 1 = Yl, we are done as we (3) [Chinese Mathematical Olympiad, 1979] 2n balls lie in several
can cross both these terms out. Hence, we may suppose without loss boxes. Consider the following procedure: select two boxes, and
of generality that x1 > Yl. Then consider the equation if the first one contains p balls and the other contains q balls
(p 2: q), to remove q balls from the first box and to put them
(xl - yt) + X2 + · · · + Xm = Y2 + · · · + Yn · in the the other box. Prove, that by repeating this procedure
several times, it is possible to collect all the balls in a single
In order to use our assumption for s = k, we need to show that y2 + box.
M
· · · + Yn < m(n- 1): If M = Yl + · · · + Yn, then Yl 2:-
n
and (4) Prove that for every rational number a and for all n E I'!, the
1
. -1 +-+···+-=a 1 h as on Iy fi mte
. I y many pos1t1ve
..
equatwn
n-1 X! X2 Xn
Y2 + · · · + Yn :::; M--
n
< m(n- 1). integer solutions.
(5) [All Union Mathematical Olympiad, 1966] The numbers
Since m + n- 1 = k, using the assumption above we can cancel at least
two terms from the equation al,a2, ... ,an

satisfy the conditions:

(x1- Yl) + X2 + · · · + Xm = Y2 + · · · + Yn·
(We regard ( x 1 - yt) as one term.) Hence we can cancel out at least two 0 :::; a1 :::; a2 :::; 2al
terms from the original equation, and hence by the PMil, our result a2 :::; a3 :::; 2a2
holds. 0

The last set of exercises contains some fairly hard problems of the sort an-1 :::; an:::; 2an-1·
that appear in international mathematics competitions. Most of them Prove that it is possible to choose the signs in the sum
use a combination of the ideas that have been developed in this booklet,
together with a few more clever tricks that require some innovative s = ±a1 ± a2 ±···±an
thought.
in such a way that the condition 0 :::; s :::; a 1 is satisfied.
(6) [ Mathematical Induction, Slinko] Let x 1 , x 2 , ... , x 2 .. be a
EXERCISES 4. sequence of non-negative integers. Transform this sequence into
(1) [Kurschak Mathematics Competition, 1932] Prove that if a, b another sequence
and n are positive integers and b is divisible by an, then
(a + 1)b - 1 is divisible by an+l. lx1- x2l, lx2- x31, ... , lx2"- x1l·

(2) [A// Union Mathematical Olympiad, 1971] Prove that for any Prove that after several such transformations a sequence of only
positive integer n there exists a positive integer which is divis- zeroes is obtained.
ible by 2n and whose decimal representation consists only of (7) Consider the following problem: Given a circle_,. choose any n
digits 1 and 2. distinct points on the circumference, and draw in all possible
22 23
 chords with end-points any of these points. Let Sn be the max-
imum amount of regions into which such a procedure can divide 5. Hints and Tips to the Exercises
a circle. Our aim will be to find a formula for Sn.
Section 2.
(a) Find S2, S3, S 4 , S 5 . Can you detect any pattern? Now try
s6 and see what happens! . n(n+1)
(2) FirSt show 1 + 2 + · · · + n = .
(b) Show that for a set { 1, 2, 3, ... , n}, it is possible to choose r (4) Use am+l- bm1-l = a(am- bm) +(a- b)bm.
I (5) (b) For the induction step, show
different numbers from the set in '( n. )' ways (note r ~ n,
r. n- r.
and by convention we let 0! = 1). We will use the notation Fm+3Fm+l- (Fm+2) 2 = (Fm+t) 2 - Fm+2Fm.

(c) Let P( n) be the statement

(~)
n!
- r!(n- r)!"
Fn+S = 5Fn+l + 3Fn.
(6) Let P(n) be the statement
(c) Suppose a circle has n chords, no 3 concurrent, with m
internal intersection points. Show that the circle is divided into 1x3x5x···x(2n-1)
______ :..__....:_ < - -1 .
n + m + 1 regions. 2 X 4 X 6 X ··· X 2n - y'2n +1
(d) Using all of the above, show that Sn = 1 + G)
+ (~).
(7) Let P ()b
"1
n e - + -221 + -321 + · · · + -n21 -< 2 - -n1, .
(8) [Polynomials, E. J. Barbeau] Let x and y be natural numbers 12
such that
l(y- x)y-x 2 = 1.
1
Section 3.
Show that y- x, x and y are consecutive terms in the Fibonacci (1) Use the PMI3 on parameter m.
sequence jntroduced in exercises 2 (5). (2) For the base case, show first that every right-angled triangle
can be divided into two isosceles triangles, and proceed using
the PMil.
(3) Use PMI2 with a= 3 and d = 2, with the number of planets as
the induction parameter. Also, notice that for the two planets
closest together, the astronomers observe each other.
(4) Use the PMI3 with parameter n.
(5) Let P(n) be the statement: "There is an integer kn such that
1 1
\ (a+- )n =an+-+ kn" -and use the PMI3.
a an
(6) Use PMI3. In the induction step, when considering all possible
subsets of { 1, 2, ... , m, m + 1}, consider those sets. that contain
m + 1 and those sets that do not contain m + 1.
24
25

J ~
(8) Let P( n) be the statement "Let x and y be distinct natural 6. Answers to the Exercises
numbers such that
r
2
Section 1. The cases for the 3 x 3 grid can be solved as in Figure
l(y-x)y-x 1=1, { 11.

with y :S Fn+l· Then y - x, x and y are consecutive terms in

the Fibonacci sequence." Use exercise 2 no 5(b).

FIGURE 11

Suppose we can solve any m x m grid. Consider an (m+ 1) x (m + 1)

grid with one square shaded. We can always rotate the grid so that
the bottom row and the rightmost column do not contain the shaded
square. Now solve the problem for them x m grid obtained by deleting
the bottom row and rightmost column of the (m + 1) x (m + 1) grid.
With that done, we can solve the (m + 1) x (m + 1) grid as in Figure
12 if m is even, and as in figure 13 if m is odd.

FIGURE 12
I I
FIGURE
lui
13
n
I

28 29
 Section 2. 2 2
1 1 -for all n EN: For n = 1, 13 = 1 (1+1)
4
is true. Now suppose
(1) For n = 1, - - = - - is true. Now suppose
1x2 1+1 m (m + 1)
2 2
13 + 23 + 33 + . · · + m3 = for some m E N. Then
4
1 1 1 1 m
- - + - - + - - + .. ·+ =--
1x2 2x3 3x4 m( m + 1) m+ 1 13 + 23 + 33 + · · · + m3 + (m + 1)3
for some m E N. Then m2(m + 1)2 ( 1)3
= + m+
4 .
1
- - + ... +
1 1
+ --.,------ (m + 1) 2(m 2 + 4m + 4)
1x 2 m(m + 1) (m + 1)(m + 2) 4
m 1 (m + 1) 2 (m + 2) 2
= _m_+_1 + (m + 1)(m + 2) 4
m 2 +2m+ 1
Thus the result follows by the PMI.
(m+1)(m+2)
m+1
Combining the two results, we have, for all n E N,
m+2

Hence the result holds by the PMI. n 2 (n + 1) 2 ·

13 +2 3 +···+n 3 = =(1+2+···+n) 2.
4
(2) We first use the PMI to show 1 + 2 + 3 + · · · + n = n(n + 1)
2
1(1 + 1) . (3) For n = 1, 21+ 4 = 32 2:: (1 + 4) 2 = 25 is true. Now suppose
for all n EN: For n = 1, we have 1 = ts true. Now
suppose 2 2m+4 2:: (m + 4) 2 for some mEN. We have

2m+5 = 2(2m+4)
1 + 2 + 3 + .. , + m = m( m + 1)
2 2:: 2(m + 4) 2
for some m E N. Then = 2m2 + 16m+ 32
> m2 +10m+ 25
1+2+3+· .. +m+(m+1) = m(m+ 1) +(m+ 1) = (m+ 1)(m+2)
2 2 , =(m+5) 2.

so the result follows by the PMI. Thus by the PMI, the result follows.
1
(4) For n = 1, a - b is obviously divisible by a- b. Now suppose
1
We now use the PMI again to show
I
am - bm is divisible by a- b. Hence a(am - bm) is divisible
by a- b. Also (a- b)bm is clearly divisible by a- b, hence
2
13 +23 +33+···+n3= n (n+l)2 a( am - bm) + (a- b)bm = am+l - bm+l is divisib1e by a- b and
4 the result holds by the PMI.
30
31

---
(5) (a) For n = 1, F1 = 1 and F3 = 2 imply F1 = F3- 1. Now
suppose F1 + F2 + · · · + Fm = Fm+2- 1 for some mEN. Then Fm+4 = Fm+3 + Fm+2
= Fm+2 + Fm+l + Fm+2
F1 + F2 + · · · + Fm + Fm+l = Fm+l + 2Fm+2
=Fm+2-1+Fm+l = Fm+l + 2(Fm+l + Fm)
= Fm+3 -1 = 3Fm+l + 2Fm.
since Fm+3 = Fm+l + Fm+2·
Using this, the induction hypothesis and another application of
Thus by the PMI, the result follows. the rule above, we have

(b) For n = 1, we have jF3F1-(F2) 2I = 12-11 = 1 is true. Now

suppose, for some mEN, 1Fm+2Fm- (FmH) 21= 1. Then Fcm+1)+5 = Fm+6
= Fm+5 + Fm+4
1Fm+3Fm+l- (Fm+2) 21 = 5Fm+l + 3Fm + 3Fm+l + 2Fm
= I(Fm+2 + Fm+l)Fm+l- (Fm+2) 21 = 3Fm+l + 5(Fm+l + Fm)
since Fm+3 = Fm+2 + Fm+l = 3Fm+l + 5Fm+2·
= 1Fm+2(Fm+l- Fm+2) + (Fm+t) 21
= 1Fm+2(Fm+l- (Fm+l + Fm)) + (FmH) 21 Thus P( m + 1) is true. Hence by the PMI, P( n) is true for all
since Fm+2 = Fm+l + Fm nEN.
= 1- Fm+2Fm + (Fm+t) 21
= 1Fm+2Fm- (Fm+t) 21 To see how this helps to answer the question, notice that for
=1. each n E N, Fn and Fn+5 have the same remainder when divided
by 5. By looking at the first 5 Fibonnaci numbers 1, 1, 2, 3, 5,
Hence by the PMI, the result follows. only the fifth number is divisible by 5. Hence, using an inductive
argument again, we can conclude that n is divisible by 5 if and
(c) Let P(n) be the statement "Fn+5 = 5Fn+l + 3Fn". P(1) only if Fn is divisible by 5.
is true since Fs = =8 5F2 + 3F1 . Now suppose P(m) is true (6) Let P(n) be the statement
for some mEN, Fm+5 = 5Fm+l + 3Fm. Now, using repeated
applications of the rule F~c = F~c-i + F~c- 2 for k 2: 3, we have
_1_x_3_x_5_x_··_·_x_:(,_2_n_-_1--"-) < __1 .
/ 2 X 4 X 6 X · · · X 2n - V2n + 1

It is a fairly trivial observation that P(1) is true: Now suppose

33
32
 for some mEN, P(m) is true. Then To see that this implies P( m + 1) is true, notice that

111 1 1 1 1
1 X 3 X 5 X · · · X (2m- 1) X 2m+ 1 12 +
22 + 32 + ... + m 2 + (m + 1) 2 ::;
2 -m+ (m + 1)2
2 X 4 X 6 X · · · X 2m X 2(m + 1)
since P(m) is true
1 2m+ 1
< X --:------:-
m 2 + m+ 1
v'2m+1 2(m+1) =2--,---~
v'2m + 1 m(m + 1) 2
2m+2 m 2 +m
< 2- ----;------,-,:-;o-

./ 2m+ 1 m(m + 1) 2
- V4m +8m+4
2
=2--1-.

::;
V 2m+1
4m 2 + 8m + 3
m+1

Thus by the PMI, P(n) is true for all n E N,and the result
2m+ 1 follows since since 2 - .!_ < 2 for all n E N. .
(2m+ 1)(2m + 3) n
1
Section 3.
v2m+3
1 (1) We use the PMI3 on the parameter m. For m = 1, Fn+ 1 =
y2[m+ 1] + 1 Fn-1 + Fn = Fn-1F1 + FnF2 is true. For m = 2, Fn+2 =
Fn+1 + Fn = Fn + Fn-1 + Fn = Fn-1F2 + FnFa. Now suppose
the result is true for all m::; k, for some kEN ~ith k > 1. Then
Hence P(m + 1) is true, and by the PMI, P(n) is true for all Fn+k+1 = Fn+k + Fn+k-1 = Fn-1Fk + FnFk+1 + Fn-1Fk-1 +
n EN. To complete the problem, one need only note that FnFk = Fn-1(Fk + Fk-1) + Fn(Fk+1 + Fk) = Fn-1Fk+1 +
FnFk+2· Hence the result is true form= k + 1, and hence by
the PMI3, the result follows.
I 1 1
-===<-
v'2n + 1- ffn (2) We use PMil with k = 4 and parameter the number of isosceles
triangles. It is always possible to dissect a right-angled triangle
into two isosceles triangles, using the dissection given in Figure
for all n E N, so the original statement is true for all n E N. 14.
( 7) Let P( n ) b e " -1 + -1 + -1 +. · · · + -1 < - -1, . P( ) 1s
. true as
2 1
12 22 3 2 n2 - n

IS,
1
.1 2 ::; 2-f. Now suppose P(m) is true for some mEN, that
/
1 1 1 1 1
- + -2+ - +
2 · · · + -2 < 2 - - FIGURE 14
12 2 3 m m
34 35
,,

to each other observe each other's planets, and hence the third
Hence, for any non- equilateral triangle, if we drop a perpendic- planet remains unobserved. Now suppose, for some m EN, the
ular from the vertex opposite the longest side, we can dissect result is true for 2m+ 1 planets. Consider a 2m + 3 planetary
each of the two resulting right-angled triangles into two isosceles system. Again, pick the two planets closest to each other, and
triangles, yielding a dissection of the triangle into four isosceles notice that the two astronomers observe each other's planets.
triangles (see Figure 15). This leaves 2m + 1 planets. If an astronomer on any of these
planets is observing one of the two planets mentioned before,
:'' then one of the other 2m+ 1 planets will be unobserved. If
nobody does this, then we can consider the 2m + 1 planets
I

jl
i' as a planetary system all on its own, and use our assumption
to conclude there must be a unobserved planet among them.
Hence the result is true for 2m + 3 planets. Hence, by the
PMI2, the result holds.
w (4) We use PMI3. The result is clearly true for n = 1 and 2.
Suppose for some k > 1, am= 3m+ 5m for all m :S k. Then,
FIGURE 15
am+l = 8am - 15am-l
Now suppose that we can dissect any non-equilateral trian- = 8(3m - 5m)- 15(3m- 1 + 5m- 1)
gle into m isosceles triangles. Now, given any triangle ABC
with shortest side AB say, label the point D on BC such that
= 3m- 1(8 X 3 -15)- 5m- 1(8 X 5 -15)
BD = AB (see Figure 16). Now L.C is the largest angle as it = 3m+1 - 5m+1.
is opposite the shortest side AB, and hence it cannot be 60° as
ABC is not equilateral. Thus triangle ADC is non-equilateral. Hence, by the PMI3, the result is true.
We apply the induction hypothesis and dissect ADC into m
isosceles triangles, thereby dissecting ABC into m + 1 isosceles
triangles. Hence by the PMil, the result follows.

8~ FIGURE
D
16
c
/

(3) We use PMI2 with a = 3, d = 2 on the number of planets.

For three planets, the two observers on the two planets closest
37
36
 (5) Let a+~a be an integer. Let P(n) be the statement "There is sum of two parts, namely, those sets that contain m + 1, and
those sets that do not. Firstly, the subsets containing m + 1
an integer kn such that (a+ ~ )n = an + J_ + kn". P(1) is and at least one other member, since it cannot contain m, will
a an
! I
trivially true, and P(2) is true as contribute ( m + 1)2Sm-1 to the sum. Secondly, subsets that do
not contain m + 1 will contribute Sm to the sum. Lastly, the
1 1 singleton set { m + 1} contributes ( m + 1) 2 to the sum. Hence
(a+- )2 = a2 + 2 + 2.
a a 2
Sm+1 = Sm + (m + 1) 2Sm-1 + (m + 1)
I
I'
Now suppose, for some m > 1, P(k) is true for all k ~ m. Then, = (m + 1)!- 1 + (m + 1) 2(m!- 1 + 1)
I since P(m) and P(m- 1) are true,
I
I
= (m + 1)!- 1 + (m + 1)((m + 1)!)
1 =(m+2){m+1)!-1
I (a+ -)m+1
a
'I =(m+2)!-l.
1 1
I =(a+-)m(a+-)
a a Thus P(m + 1) is true, and by the PMI3, the result follows.
1 1 (7) Let S be a subset of natural numbers defined as follows: x E S
=(am+-+km)(a+-)
am a if and only if x 2: 7 and x cents cannot be paid out out using
I 1 1 1 only 3 and 5 cent coins. We need to show that S is empty.
= am+1 + - - +am-1 + - - +km(a+ -)
am+1 am-1 a Suppose not, then by the PSN, S has a smallest member, say
I
1)m-1 - km-1 + km{a+-. m. Now since 8 = 3+5, 9 = 3+3+3, 10 = 5+5, 8, 9 and 10 are
ij!
=a m+l + - 1 {
am+1 + a+-
1)
a a not inS, hence m-3 > 7. Now, suppose we can pay out m-3
II . 1. .
.N ow, smce a+ - IS an mteger,
cents. Then, adding a 2 cent coin, we would be able to pay out
m cents, a contradiction. Hence m - 3 E S, contradicting m is
II a
the smallest member. Hence S must be empty, and the result
follows.
=(a+~ )m-l- km-1 + km(a + ~)
II
I
km+1
!I a a {8) We use the PSN. Suppose there is an arrangement of counters
II that requires more than 4 colours. Let S be the set of all natu-
is an integer, and hence P(m + 1) is true. Hence by the PMI3, ral numbers n such that there is an arrangement of n counters
the result follows. that requires more-than 4 colours. By the PSN, S has a smallest
1
{6) Let Sn be the sum of the squares of the products of the numbers member, say m. Notice that m > 4. Now consider an arrange-
in all possible subsets of the set { 1, 2, ... , n} that do not contain ment of m counters requiring more than 4 colours. Consider the
any consecutive numbers. Let P(n) be the statement "Sn = centre points of the m counters, and all possible straight lines
(n+ 1)! -1". We use PMI3. P{1) is true since S 1 = 12 = 2! -1. joining two centre points. A subset of these straight lines will
P{2) is true, for the only valid subsets are {1} and {2}, and form a polygon which contains all the other lines. At least one
s2 = 12 + 22 = 3! - 1. / of the vertices of this polygon must have angle less than 180°.
Call this vertex 0. Now if the counter with centre 0 is touched
Now suppose for some mEN, m > 1, we have that P(m) and by two other counters with centres P and Q (see Figure 17),
P(m - 1) are true. We derive the formula for Sm+1 as the then as PQ 2: PO and PQ 2: QO, L.POQ 2: 60'6 , and thus the

38 39
 ,,

counter with centre 0 can only be touched by at most 2 other Hence the result holds for n = 2m.
counters.
Now let m E !'!, and suppose the result holds for n = m + 1.
We show the result holds for n = m. Let x1, x2, ... , Xm be any
X1 + X2 + · · ·+ Xm
real numbers. Put Xm+t = . Now
m
f( X1 + X2 + · · · + Xm)
m
-j(
_ X1 + X2 + · · · + X m+l)
FIGURE 17 m+1
< f(xt) + · · · + J(xm) + J(Xl + · · · + Xm)
m
If we remove this counter, by the mininimality of m, we can
m+1
colour the arrangement using only four colours. But then we using the n = m + 1 case, from which
can colour the original arrangement using only four colours,
by colouring the counter at 0 with a colour not used by any J(l + X2 ~- · · + Xm) ~ f(xt) + f(x2) + · · · + f(xm).
counter touching it. But this contradicts m E S, hence S must
be empty and the result follows.
Hence, by the PBMI, the result holds.
(9) We use the PBMI. For n = 1, the result is trivially true, and
(10) We use PBMI. Let P( n) be the statement "We can pay n cents
for n = 2 the result follows from the convexity of f.
using only coins of denominations a and b cents". We wish to
show that P(n) is true for all n > ab- a- b. Firstly note that
We next show that if the statement is true for some n = m, P(s) is true for all s E {a,2a,3a, ... }. Now suppose, for some
then it is true for all n = 2m, and hence the statement is true m > ab- a- b, we have that P(m + 1) is true, that is, there
!:' for all powers of 2. Suppose the result is true for some n = m, are non-negative integers x, y such that ax+ by= m + 1. Now
and let Xt, x2, ... , X2m be real numbers. Now m+1 2: ab-a-b+2, and hence ax+by-1 2: ab-a-b+l. Since
I~
rl
J( Xt + x2+ · · · + X 2m)
the highest common factor of a and b is 1, there are integers
rl, p and q such that pa + qb = 1. (This is a standard result
2m in number theory - see Topics in Number Theory by Valentin
~ ~ (f(Xl + · · · + Xm) + f(Xm+l + · · · + X2m )) Goranko, which is Book number 2 in this series.) Hence ax +
{ 2 m m by- pa- qb +a+ b 2: ab + 1, which in turn gives
II using the convexity of f
x-p+1 q-y-1>1+_.!_>1.
< ~ (f(xt) + · · · + f(xm) + _f(-'--x_m-'-+-'-1)_+_·_·_·+__
f(_x_2m-'-)) b - ab
I
-2 m m
I using the case n = m Hence, there must be an integer t such that
I f(xt) + J(x2) + · · · + f(x2m) x-p+1 q-y-1
IIi 2m -.->t>
I
40 41
I!
I
 Now q- y- 1 < at implies at- q + y 2:: 0, and x- p + 1 > bt
implies x - bt - p 2:: 0. Thus

a(x- bt- p) + b(at- q + y)

=ax+by-(ap+bq) 3r (7\
:I
= (m + 1)- 1
= m.
\cjc
II Thus P(m) is true. Hence, by PBMI, P(n) is true for all n >
ab- a- b.
il
(11) Let P(n) be the statement "If the union of n circles in the FIGURE 18
plane has area A, it is possible to choose from among them
I
several pairwise non-intersecting circles, the sum of whose areas
with area of their union Aa, and Aa 2:: ~2 Now Sa U{C}
is greater than or equal to :". We use PMI3.
possesses the required properties to make P( m + 1) true, as C
does not intersect any of the circles in Sa, and the combined
P(1) is trivially true. union of their areas is greater than or equal to

Now suppose for some mEN, P(k) was true for all k :Sm. Let
S be a set of m + 1 circles with area of their union A. Let C be -A2
9
+ A 1 >- A- 9A1
9
+ A 1 = -.
A
9
the circle of largest radius in S, and denote its radius by r and
its area by A 1 . If A 1 2:: :, then C has the required property, Hence, by the PMI3, the result holds.
A
,,,i I hence we may assume that A1 < 9.
Section 4.
Consider a circle C 1 of radius 3r, concentric with C. The radius
l
,I
i
of each circle· in S is no greater than r, and thus if such a circle (1) We use the PMI on n, starting ·from n = 0. For n = 0 it is
::j intersects C, it lies completely in C1. (See Figure 18.) Let 5 1 true, since (a+ 1)b - 1 is divisible by a. Now suppose that
:i
be the set of all circles that intersect C (including C). The area the proposition is true for some n = k, that is, for all positive
of their union must be less than 9A 1 < A. But then there is a integers a and b such that b is divisible by ak, we have (a+ 1)b -1
circle in S that is not in S1, and hence the set S2 = S \ S1 is is divisible by ak+ 1 . We will show that if a and b1 are positive
I: not empty, and A 2 is the area of the union of circles in 5 2 , we integers with b1 divisible by ak+ 1 , then (a+ 1)h - 1 is divisible
have A2 2:: A - 9A1. by ak+2.

Now suppose 5 2 contains k circles. Since C fl. 5 2 , k :S m,

and since P(k) is assumed true, we may suppose there is a Let b1 be divisible by ak+ 1 . Let b = ~.
a
Then, since b is divisible
subset Sa of S2 consisting of pairwise non-intersecting circles by ak, and thus by our assumption, (a + 1)b - 1 is divisible by

I 42 43

,,
' I
 ak+l. Observe that total number of balls in red boxes is also even. Hence there must
be an even amount of red boxes. Pair these boxes arbitrarily,
(a+l)b'-1 and perform one move on each pair of boxes. In this way we
=(a+1)ab_1 obtain a placement of balls such that each box contains an even
number of balls. It is possible now to group the balls in each
II' = [(a+ 1)b- 1]((a + 1)(a-l)b +(a+ 1)(a-2)b + ... + 1]. box in turn in pairs, call each such pair a two-ball. We thus
'
obtain 2k two-balls, and notice that if applying moves for single
It thus remains to show that the second factor in the last ex- balls has the same effect as applying the move for two-balls.
pression is divisible by a. This is true as we can rewrite this We thus can use our assumption to solve the problem for the
factor as two-balls, and hence by the PMI the result holds.
((a+ 1)(a-l)b- 1] +[(a+ 1)(a-2)b- 1] + ... (4) We proceed by induction on n. The result is clearly true if
n = 1. Suppose it is true for n = k. Consider the equation
i
···+((a+1)b-1]+a.
1 1 1 1
-+ -+ ···+-+--=a.
This proves the required result. Xn X2 Xk Xk+l

(2) We prove a slightly stronger result: For each n E N, there is We may suppose without loss of generality that
an n-digit number divisible by 2n whose decimal representation X1 ~ X2 ~ · · · ~ Xk+l·
consists only of the digits 1 and 2. We proceed by induction on
n. Then
1 1 1 1 k+1
a=-+-+···+-+--:::;--,
For n = 1, the number 2 satisfies the above properties. Now Xk X2 Xk Xk+l Xk+l
suppose the result is true for some n = k, that is, there is an 1
k-digit number, say x, such that x = 2k x y for some y E N, and hence Xk+l:::; k + . Since Xk+l EN, there are only finitely
a
and the decimal representation of x consists only of 1 's and many possible values for Xk+l· But then the equation
2's. Consider the two (k + 10)-digit numbers z 1 = 10k + x 1 1 1 1
and z 2 = 2 x 10k + x. Notice that their decimal representations -+-+···+-=a---
Xn X2 Xk Xk+l
consist of 1 'sand 2's only. We will prove one of them is divisible
by 2k+1. To show this, notice first that since 5k is odd and 2 x 5k has only finitely many solutions x 1 , ... , x 2, for each possible
is even, one of the integers 5k + y and 2 x 5k + y is even. Hence .value of Xn+l, using our assumption. Hence the result follows.
one of 2k(5k + y) = z1 and 2k(2 x 5k + y) = z2 is divisible by (5) We proceed by induction on n. For n = 1, the result is triv-
2k+ 1 , and this completes the proof. ially true. Suppose that the result holds for some n = k. Let
i (3) We prove the result using the PMI on n. For n = 1 the stated b1 , b2, ... , bk+l be any numbers satisfying the inequalities
I proposition is trivially true. Now suppose that the statement
0 :::; bl :::; b2 :::; 2bl
is true for some n = k, and consider an arbitrary placement
of 2k+ 1 balls in the boxes. Colour each box containing an odd b2 :::; b3 :::; 2b2
number of balls red, and each box containing an even number of
balls blue. Notice that since the total number of balls is even,
and the total number of balls in blue boxes is even, thus the bk :::; bk+l :::; 2bk.

44 45
 _••.•••,A-

We need to show that the signs in the sum sL = ±b1 ± · · · ± bk+ 1 The parity function and the operation *obey the following rules
can be chosen in such a way such that 0 ~ s1 ~ b1. Letting for all non-negative integers a, b and c:

a~ = b2,
p((a *b) H)= p(a * (h c))
p( a * b) = p( h a)
a~= b3,
p( a * b) = p( a) * p( b)
p(a *a)= 0
a~_ 1 = bk, p(p( a)) = p( a).
I
a~= bk+l,
s' = ±b2 ± · · · ± bk+l Before we consider the actual problem, consider an infinite se-
=±a~± ... a~, quence of non-negative integers

we can apply our assumption, deducing that the signs in s' can XI, X2, X3, ... 'Xm, ...

II be chosen in such a way that 0 ~ s' ~ ai = b2.

We can transform this sequence into another infinite sequence
1/:
I
Compare s' and b1 . There are two cases:
X1 * X2, X2 * X3, X3 * X4, .. ·, Xm * Xm+l, · · ·
Ill
~I s' ~ b1: Then s 1 = b1 - s' has the required property, for We make the following claim: For any n E N, after 2n such
transformations of the sequence
0 ~ s1 = b1 - s' ~ b1.
Xl,X2,X3,···,Xm,···

we get a sequence
s' > b1: Then s1 = -b1 + s' has the required property, for
II'/I Y1 , Y2, Y3, · · · , Ym , · · ·
I, 0 ~ s1 = s' - b1 ~ b2 - b1 ~ 2bl - b1 = b1.
such that
Thus 0 ~ s1 ~ b1 in both cases, and hence the result is true for
n = k + 1 as well. Hence by the PMI, the result follows. p(yl) = p(x1 * x2"+1)
(6) We start by introducing two bits of notation. Given non-negative p(y2) = p(x1 * x2"+2)
integers a and b, we denote by a*b the number la-bl, and define
the parity function
P(Ym) = p(xl * X2"+m)
1 if a is odd
p(a) = { 0 if a is even.

47
46

,I

L
~~ --
,I

,II
We prove this claim by induction on n. so the result is true for n = k + 1, and hence by the PMI the
i claim is true.
1: For n = 1, after 2 transformations we see, for each m E N,
Now to return to the original problem. Suppose I have a se-
Ym = (xm * Xm+d * (xm+l * Xm+2), and hence, applying the
laws above we have p(ym) = p(xm * X2+m)· quence of 2n non-negative integers
!I
X1, X2, ... , X2n,
Now suppose the result is true for some n k. Consider an
infinite sequence and apply the transformation stated in the problem 2n times,
by the claim above and the wrap-around effect of the transfor-
Xl, X2, X3, ... 'Xm, ... mation, we will obtain a sequence

and suppose after 2k+l transformations it yields the sequence Yl, Y2, · · ·, Y2n

Zl' Z2' Z3' ... ' Zm' ... such that form= 1, ... , n, p(ym) = p(xm) * p(xm) = 0, that is
a sequence with each member even.
We can divide the process into two parts: First apply 2k trans-
I formations to the sequence We finish off the problem by applying the PMI3, but this time
on the largest number in the sequence. Suppose we have a
xl,x2,X3,···,Xm,··· sequence
j:l X1, X2, ... , X2n,
,,I to get a sequence, say, and the largest number in the sequence is 1. Then the sequence
~1 1 consists only of O's and 1's, and hence after 2n transformations
1 Y1 , Y2 , Y3 , · · · , Ym ,··· the sequence can consist only of even numbers, and it is not
: ]
hard to see that these must all be O's. Now suppose the result
,II and then apply another 2k transformations to this sequence to is true for all sequences with largest number in the sequence at
d get the final sequence
most k, and consider a sequence

Zl,z2,Z3,···,Zm,··· X1, X2, ... , X2n

Using our assumption that the result is true for n = k twice, whose largest member is is k + 1. After 2n transformations we
and applying the rules above, we have for any mEN, will be left with a sequence

p(zm) = P(Ym * Y2k+m) Yt, Y2, · · ·, Y2n

= P(Ym) * P(Y2k+m) whose members are all even, and less than or equal to k + 1.
= [p(xm) * p(X2k+m)] * [p(x2k+m) * p(X2k+2k+m)] Now the sequence
= p(xm * X2k+2k+m) Yt Y2 Y2n
= p(xm * X2k+l+m), 2'2'"""'2
iII 48
49
 ~

has largest member less than k + 1, and hence by our assump- Hence the formula holds for n = k + 1 and the result holds by
tion after a number of transformations will reach a sequence of the PMI.
1:
O's. Hence the sequence Yl, Y2, ... , Y2n will reach a sequence of
I'' O's, and thus the original sequence will reach a sequence of O's. (c) We use the PMI on n. With 1 chord, there can be no
,,I Hence the result is true for all sequences with largest number intersection points (that is, m = 0) and it will cut the circle
!'
k + 1, and thus by the PMI3, the result is true. into 1 + 0 + 1 == 2 parts trivially, so we have that the formula
(7) (a) 52= 2, 53= 4, 54= 8, 55 = 16 but 5s = 31. holds for n = 1.
II

,I~ (b) Notice first that there is only one way of choosing 0 items
from a set, and the formula reflects this. So we need only prove
Now suppose the formula holds for some n = k. Consider a
circle containing k + 1 chords and m intersection points. Pick
I! I
the formula for values of r greater than 0. To do this, we use any chord from the circle, and let p be the number of intersec-
the PMI on n. For the set {1}, it is possible to choose 1 item tion points this chord makes with other chords. Now erase this
II from this set in only one way, and G) = 1, hence the formula chord from the circle. We are left with k chords and m - p
I is true for n = 1. intersection points. By our assumption, there are k + m- p + 1

'1 ~
regions that the circle is divided into. Now replace the erased
Now suppose the formula is true for some n = k. Consider chord. Since it cuts the other chords p times it will divide p+ 1
the amount of ways of picking r different numbers from the set regions into two new regions, that is p +1 regions will be in-
ll1
{ 1, 2, ... , k, k + 1}. There are two possible cases: either your troduced. Hence the total number of regions will be given by
''II
choice of r numbers contains k + 1 or it does not. The number k + m- p + 1 + p + 1 = (k + 1) + m + 1 (see Figure 19), and
',11 I' I hence the formula holds for n = k + 1 as well. Hence by the
of ways of choosing r numbers which contain k + 1 is the same
PMI, the formula is true for all natural numbers n.

If
1/;1 as choosing r - 1 numbers from the set { 1, 2, ... , k}, and the
number of ways of choosing r numbers not containing k + 1 is
equal to the number of ways of choosing r numbers from the
I
I
set { 1, 2, ... , k}. By our assumption, these numbers are equal
to c~l) and (~) respectively, and hence the number of ways of
choosing r numbers from the set { 1, 2, ... , k, k + 1} is
II''
k ) (k) k! k!
IIIII, ( r-1 + r =(r-1)!(k-r+1)!+r!(k-r)!
FIGURE 19
_ k! x r + k! x (k- r + 1)
r!(k-r+1)!
(d) If we choose n points on the circumference of a circle, and
k! X (k + 1)
draw in all possible chords, a maximum amount of regions will
r!((k + 1)- r)!
result if no three chords are concurrent. Now the amount of pos-
(k + 1)! sible chords is given by (~),for each chord is uniquely specified
r!((k + 1)- r)! by its two end-points. Now if no three chords are concurrent,
'I:II,I
= e~ 1). each possible pair of distinct chords not sharing an endpoint
will give rise to a unique internal intersection point. Since each
I,
'I 50 51

I
 ~

chord is uniquely determined by its two endpoints, the number CASE 1 - x ~ Fk+l: In this case, let a = y- x and b = x.
of such pairs of chords will be the number .of ways of choosing Notice that
4 distinct end-points from the possible n points, or using (b),
(~). Hence by (c), the maximum number of regions Sn is given l(b- a)b- a 21 = l(2x- y)x- (y- x) 21
by (~) + (~) + 1.
= l2x 2 - xy- y2 - x 2 + 2xyl
(8) Firstly, notice that y ~ x. To see this, suppose y < x. Then
y- x < 0 ===> (y- x)y ~ -1. Also, -x 2 ~ -1, hence = lx2 + xy - Y21
(y- x)y- x 2 ~ -2, contradicting I(Y- x)y- x 21= 1. =lx2-y(y-x)l
=1.
Secondly, notice that y- x ~ x. To see this, suppose y- x > x.
I Then (y-x)y-x 2 > xy-x 2 = x(y-x) > x 2 ~ 1, contradicting Hence, since P(k) is true, 'we may assume b- a, a and b are
I i(y-x)y-x21=1. consecutive terms of the Fibonacci sequence, hence a = y - x,
b = x and a + b = y are consecutive terms of the Fibonacci
Let P( n) be the statement "Let x and y be distinct natural sequence as well.
i
numbers such that
CASE 2- x > Fk+ 1 : In this case, -x ~ -Fk+l, hence since
l(y- x)y- x 2
1 = 1, y ~ Fk+2, we have y- x ~ Fk+ 2 - Fk+ 1 = Fk. Put b = y- x
and a= 2x- y. Notice that
withy~ Fn+l· Then y-x, x andy are consecutive terms in the 2
Fibonacci sequence." (Recall Fk is the kth Fibonnaci number l(b- a)b- a 2 1:;::: I(Y- x- 2x + y)(y- x)- (2x- y) 1
2
introduced in exercises 2(5).) We will show by a combination = l2y 2 - 5xy + 3x 2 - 4x 2 + 4xy- y 1
of the PMil and PMI3 that P(n) is true for all n ~ 2, hence =ly2-xy-x21
proving the original problem.
= i(y-x)y-x21
Firstly P(2) is true, since then, if y ~ F 3 = 2, since x and y =1.
I are distinct the only possible result is that y = 2, x = 1 and
y- x = 1, yielding the desired result. Hence, since P(k- 1) is true, we have b- a, a and b are
I! consecutive terms of the Fibonacci sequence. The sequence
Now suppose for some k EN, P(m) is true for all m ~ k. Let will continue as follows: b - a, a, b, a+ b, a + 2b,.... Hence
x and y be distinct natural numbers such that b = y- x, a+ b = x, a+ 2b = y are consecutive terms of the
Fibonacci sequence.
I (y - x )y - x 2 1 = 1, Since in both cases we have proved that P(k + 1) is true, the
result holds by the PMil and PMI3.
withy~ Fk+2·

We consider two possible cases:

52 53

..