Crux

Mathematicorum
Published by the Canadian Mathematical Society.

http://crux.math.ca/

The Back Files

The CMS is pleased to offer free access to its back file of all
issues of Crux as a service for the greater mathematical
community in Canada and beyond.

Journal title history:

➢ The first 32 issues, from Vol. 1, No. 1 (March 1975) to
Vol. 4, No.2 (February 1978) were published under the
name EUREKA.
➢ Issues from Vol. 4, No. 3 (March 1978) to Vol. 22, No.
8 (December 1996) were published under the name
Crux Mathematicorum.
➢ Issues from Vol 23., No. 1 (February 1997) to Vol. 37,
No. 8 (December 2011) were published under the
name Crux Mathematicorum with Mathematical
Mayhem.
➢ Issues since Vol. 38, No. 1 (January 2012) are
published under the name Crux Mathematicorum.
 CRUX MATHEMATICORUM
February
Volume 20 02 fevrier
1994

CONTENTS I TABLE PES MATIERES

A Corollary to Gauss's Lemma Andy Liu 31

Congruences Identifying the Primes Richard J. Mcintosh 33

The Olympiad Corner: No. 152 R.E. Woodrow 35

Problems: 1890, 1911-1920 47

Solutions: 1695, 1816, 1817, 1821-1825,1827, 1829, 1830 49

Canadian Mathematical Society

Q> Soci6t£ mathgmatique du Canada
 Founding Editors / R&lacteurs-fondateurs: Leopold Sauve, Frederick G.B. Maskell
Editors-in-Chief / R&Iacteurs en chef: G.W. Sands, R.E. Woodrow
Managing Editor / R£dacteur-g£rant: G.P. Wright

EDITORIAL BOARD / CONSEEL DE REDACTION

G.W, Sands (Calgary)
R.E. Woodrow (Calgary)
G.P. Wright (Ottawa)
R. De Peiza (East York Board of Education)
R. Guy (Calgary)
C. Fisher (Regina)
D. Hanson (Regina)
A. Liu (Alberta)
R. Nowakowski (Dalhousie)
J. Totten (Cariboo College)
E. Wang (Wilfrid Laurier)

GENERAL INFORMATION RENSEIGNEMENTS GEN&RAUX

Crux Mathematicorum is a problem-solving journal at the Crux Mathematicorum est nine publication de resolution de
senior secondary and university undergraduate levels for problemes de niveau secondaire et de premier cycle
those who practice or teach mathematics. Its purpose is universitaire. Bien que principalement de nature educative,
primarily educational but it serves also those who read it for elle sert aussi a ceux qui la lisent pour des raisons
professional, cultural or recreational reasons. professionnelles, culturelles ou recreative.

Problem proposals, solutions and short notes intended for Les propositions de problemes, solutions et courts articles
publication should be sent to the appropriate member of the a publier doivent etre envoyes au membre approprie du
Editorial Board as detailed on the inside back cover. conseil de redaction tel qu'indique sur la couverture arriere.

SUBSCRIPTION INFORMATION RENSEIGNEMENTS SUR L'ABONNEMENT

Crux Mathematicorum is published monthly (except July Crux Mathematicorum est publie mensuellement (sauf juillet
and August). The subscription rates for ten issues are et aout). Les tarifs d'abonnement pour dix numeros
detailed on the inside back cover. Information on Crux figurent sur la couverture arriere. On peut egalement y
Mathematicorum back issues is also provided on the inside retrouver de plus amples renseignements sur les volumes
back cover. anterieurs de Crux Mathematicorum

ACKNOWLEDGEMENTS REMERCIEMENTS

The support of the Department of Mathematics & Statistics Nous rendons hommage a l'appui du departement de
of the University of Calgary and of the Department of mathematique et statistique de FUniversite de Calgary et du
Mathematics of the University of Ottawa is gratefully departement de mathematiques de TUniversite d'Ottawa.
acknowledged.

® Canadian Mathematical Society / Societe mathematique du Canada 1994

Printed at / imprime a: Synergy Print and Copy, Ottawa.

ISSN 0705-0348 Publication Mail Registration No. 5432

 31

A COROLLARY T O GAUSS'S L E M M A
Andy Liu

Let p be a prime and a be an integer not divisible by p. We say that a is a quadratic

residue of jp if the congruence x2 = a (mod p) has solutions, and a quadratic non-residue if
no solutions exist.
Gauss's Lemma provides a necessary and sufficient condition which distinguishes
the quadratic residues from the non-residues of an odd prime. Let the integers ak, 1 <
k <{p — l)/2, he reduced modulo p so that \ak\ < (p — l ) / 2 . Then a is a quadratic residue
°fp if and onfy if th>e number of these integers which are negative is even.
E X A M P L E 1: Let a = 3 and p = 17. Then (p - l ) / 2 = 8. The integers
3,6,9,12,15,18,21 and 24 reduce modulo 17 to 3,6, —8, —5, —2,1,4 and 7 respectively.
Since an odd number of them are negative, 3 is a quadratic non-residue of 17.
When p and a are much larger, Gauss's Lemma is not very convenient to use. Thus
its principal application lies in proving the most practical Quadratic Reciprocity Law.
Nevertheless, it may pay dividends to work through a more involved case using Gauss's
Lemma.
E X A M P L E 2: Let a = 19 and p = 461. We have to reduce (p - l ) / 2 = 230
integers. However, it is not necessary to handle them one at a time. We have 19 x 1 = 19
(mod 461) and 19 x 12 = 228 (mod 461). Hence all of 19A;, 1 < k < 12, are positive. Now
19 x 13 = -214 (mod 461) and 19 x 24 ~ - 5 (mod 461). Hence all 19fc, 13 < k < 24,
turn negative. Thereafter, we have alternate blocks of 12 positive and 12 negative integers,
until the pattern is broken at 19 x 97 = -—1 (mod 461). This pattern is disrupted one more
time, at 19 x 194 = —2 (mod 461). Of the 228 integers in blocks of 12, an even number
of them are negative. The two extra integers —1 and —2 are both negative, so that the
overall total remains even. By Gauss's Lemma, 19 is a quadratic residue of 461.
We generalize the above example to the following result.
C O R O L L A R Y T O GAUSS'S L E M M A : Let p be an odd prime and a be an
integer not divisible by p. Let (p — l)/2 = aq + r, 0 < r < a — 1. Let s = 0 if r = 0. If
r > 0, let 6 be the inverse modulo p of a. Let the integers 6i, 1 < i < r, be reduced modulo
p so that \bi\ < (p—1)/2. Let s be the number of these integers which are negative. Then a
is a quadratic residue of p if and only if s = q[a/2] (mod 2), where [x] denotes the greatest
integer < x.

Before we prove this result, let us use it to work out Examples 1 and 2 again.
E X A M P L E 1: We have 8 = 3 x 2 + 2. Hence q[a/2] = 2 x 1 is even. Now the
inverse modulo 17 of 3 is 6. From 6 x 1 = 6 (mod 17) and 6 x 2 = - 5 (mod 17), 5 = 1.
Hence 3 is a quadratic non-residue of 17.
E X A M P L E 2: We have 230 = 19 x 12 + 2. Hence q[a/2] = 12 x 9 is even. Now
the inverse modulo 461 of 19 is 364. Prom 364 x 1 = - 9 7 (mod 461) and 364 x 2 == - 1 9 4
(mod 461), 5 = 2. Hence 19 is a quadratic residue of 461.
32

PROOF OF COROLLARY: Consider the integers afc, 1 < k < (p - l ) / 2 ,

reduced modulo p so that \ak\ < (p — l ) / 2 . Since (p —1)/2 = aq + r, these integers can be
divided into a blocks of q or q + 1 consecutive integers, such that all integers in the same
block have the same sign. Moreover, there are exactly r blocks of size g + 1 , and the blocks
alternate in signs, starting with a block of positive integers.
If r = 0, the total number of negative integers is exactly q[a/2]. By Gauss's Lemma,
a is a quadratic residue of p if and only if q[a/2] is even, or q[a/2] = 0 = 5 (mod 2).
Suppose r > 0. Consider the r blocks of integers of size q + 1 and the integer with
the smallest absolute value in each block. These r absolute values must be 1,2, • • •, r. The
number of negative integers not including any from these r is q[a/2] as before. It remains
to count the number of negative integers among these r.
For 1 < i < r, we have ak{ = dti (mod p) for some integers fct-, where 1 < k{ <
(p — l ) / 2 . Multiplying by 6, we have ±A:t- = bi (mod p). Note that | ± fct| < (p — l ) / 2 , and
±ki has the same sign as ± i . It follows that the number of negative integers among the
r in question is equal to the number s of negative integers among 6i, 1 < i < r, reduced
modulo p so that \bi\ < (p — l)/2.
It now follows from Gauss's Lemma that a is a quadratic residue of p if and only if
s = q[a/2] (mod 2). This completes the proof of the corollary. •
This result is particularly useful when a is fixed and p varies. We conclude with a
further application.
E X A M P L E 3: We wish to determine the odd primes p for which a = 3 is a
quadratic residue. Note that [3/2] = 1. We consider four cases.
Let p = 12k + 1. Then (p - l ) / 2 = 6fc, q = 2k, r = 0 and 5 = 0. Since 2k = 0
(mod 2), 3 is a quadratic residue of p.
Let p = 12fc + 5. Then (p - l ) / 2 = Qk + 2, q = 2fc and r = 2. The inverse modulo
p of 3 is 4fc + 2. Prom (4A: + 2) • 1 = 4k + 2 (mod p) and (4fc + 2) • 2 = -(4fc + 1) (mod p),
5 = 1. Since 2k ^ 1 (mod 2), 3 is a quadratic non-residue of p.
Let p = 12fc + 7. Then (p - l ) / 2 = 6k + 3, g = 2k + 1 and r = 0 and s = 0. Since
2k + 1 ^= 0 (mod 2), 3 is a quadratic non-residue of p.
Let p = 12k + 11. Then (p - l ) / 2 = Qk + 5, q = 2fc + 1 and r = 2. The inverse
modulo p of 3 is 4k + 4. Prom (4k + 4) • 1 = 4fc + 4 (mod p) and (4fc + 4) • 2 = -(4fc + 3)
(mod p), 5 = 1. Since 2k + 1 = 1 (mod 2), 3 is a quadratic residue of p.
In summary, a = 3 is a quadratic residue of an odd prime p if and only if p = 12A:± 1.

Department of Mathematics
The University of Alberta
Edmonton, Alberta, Canada T6G 2G1
 33

CONGRUENCES IDENTIFYING THE PRIMES

Richard J. Mcintosh

Gauss (Disquisitiones Arithmeticae, 1801, art. 329) wrote:

The problem of distinguishing prime numbers from composite numbers ...is known to be
one of the most important and useful in arithmetic. ... The dignity of the science itself
seems to require that every possible means be explored for the solution of a problem so
elegant and so celebrated.
Wilson's Theorem, which was actually discovered by Leibniz and first proved by
Lagrange, states that if p is prime then (p — 1)! = —1 (mod p). It is easy to see that the
converse of Wilson's Theorem also holds. Thus, Wilson's Theorem can be used to identify
the primes: an integer n > 2 is prime if and only if (n — 1)! = —1 (mod n). Another
congruence identifying the primes, communicated to me in 1980 by Professor James P.
Jones of the University of Calgary, is the following.

T H E O R E M !• An integer n > 2 is prime if and only if

(n + l)(2n + l)(3n + 1) • • • ((n - l)n + 1) = 0 (mod (n - 1)!).

Before proving this theorem let us examine the divisibility of products of inte-
gers in arithmetic progressions by factorials. It is easy to see that the product n(n + 1)
(n + 2) • • • (n + h — 1) is divisible by k because one of the numbers in this product is divisible
by k. It is much more difficult to show that this product is divisible byfc!,unless one makes
the clever observation that the binomial coefficient

n + k - l\ n(n + l)(n + 2) - - - (n + A: - 1)
k J ~ kl
is an integer. We claim that this product of consecutive integers can be replaced by a
product of integers in arithmetic progression. In particular, we have

T H E O R E M 2.1fn is relatively prime to k\ and a is any integer, then the product

a(n + a)(2n + a) - • • ((k — l)n + a) is divisible by kl.

We will need the following lemma.

L e m m a . / / n is relatively prime to m and t is any integer, then the set {£, n + 1 ,

2n + 1 9 . . . , (m — l)n + 1 } forms a complete set of residues modulo m.

Proof of Lemma. It suffices to show that for each b 6 {0,1,2,..., m - 1} the

congruence xn +1 = 6 (mod m) has a solution. Since n is relatively prime to m, n has an
inverse modulo m, and thus x = (6 — t)^*"1 (mod m). •
34

Proof of Theorem 2. By the fundamental theorem of arithmetic it suffices to show

that for each prime p the multiplicity to which p divides the product a(n + a)(2n + a) • • •
((k — l)n + a) is greater than or equal to the multiplicity to which p divides fc!.
Let px < fc, where p is prime and i > 1. We construct [k/pl] sets of p% elements,
where [x] denotes the greatest integer < x. The first set consists of the first p% terms of our
product, the second set consists of the next p% terms, and so on. We ignore any numbers
left over at the end. Since n is relatively prime to &!, and thus relatively prime to p\ by
our lemma each set forms a complete set of residues modulo p\ and hence exactly one
element in each set will be divisible by jp\ Therefore the number of terms in the product
a(n + a)(2n + a) • • • ((k — l)n + a) that are divisible by p% is > [k/p*]. Recall that some
of the ignored terms at the end may be divisible by p\ However, the number of terms in
the set { 1 , 2 , 3 , . . . , A;} that are divisible by p% is exactly [k/p*]. This completes the proof
of Theorem 2. •

Returning to Theorem 1, it follows that if n is prime, then n is relatively prime to

(n — 1)!, and thus by Theorem 2, with a replaced by n + 1 and k replaced by n — 1, we
have
(n + l)(2ra + l)(3n + 1) • - • ((n - l)n + 1) = 0 (mod (n - 1)!).
Conversely, if n is not prime, then n has a prime factor p and p divides (n — 1)!. Since
n + 1,2ra + 1,3n + 1 , . . . , (n — l)n + 1 are congruent to 1 modulo p, it follows that p, and
hence (n — 1)!, does not divide the product (n + l)(2n + l)(3n + 1) • • • ((n — l)n + 1 ) . This
completes the proof of Theorem 1. •

We end with a conjectured congruence for primes, this time involving binomial
coefficients. It is not difficult to show that y^Zi) — 1 (mod n) for all primes n. This
congruence is also satisfied by squares of odd primes and cubes of primes > 5. The only
other solutions found so far are n = 29-937, 787-2543,69239-231433,168434 and 21246794.
In 1819 Babbage observed that the stronger congruence y^Zi) = 1 (mod n 2 ) holds for all
primes n > 3, and Wolstenholme, in 1862, proved that \**Zi) = ^ (mod n 3 ) for all primes
n > 5. (See Dickson, History of the Theory of Numbers, vol. 1, p. 271.) I conjecture that
the only composite solutions of
(2n - 1\ =
1 (mod n 2 )

2
U-i/
are of the type n = p , where p is a prime satisfying

( * _ - / ) . 1 (mod / ) ;

the only primes p < 108 satisfying this last congruence are 16843 and 2124679. Jones has
conjectured that

n > 5 is prime if and only if

* ICT--11)1 = 1 (mod n ).
 35

(See Guy, Unsolved Problems in Number Theory, problem B31, p. 47, and Ribenboim, The
Book of Prime Number Records, 2ed., p. 21.) I have verified this for all n < 3 X 107. I've
also shown that if n = p2 satisfies the congruence modulo n 3 , then n = p3 satisfies the
congruence modulo n 2 . Thus, the first conjecture implies the second. The details of these
computations and conjectures should soon be published elsewhere in a research paper titled
"On primality and Wolstenholme's Theorem".

Department of Mathematics and Statistics

University of Regina
Regina, Saskatchewan
Canada S4S 0A2
e-mail: [email protected]

THE OLYMPIAD CORNER

No. 152
R.E. WOODROW

All communications about this column should be sent to Professor R.E. Woodrow,
Department of Mathematics and Statistics, The University of Calgary, Calgary, Alberta,
Canada, T2N 1N4.

We begin this number with a "pre-Olympiad" set, the 2nd U.K. Schools Mathe-
matical Challenge which was written February 2nd, 1989. This is a multiple choice format
for which the students must not have been over 15 years of age by August 31, 1989. My
thanks go to Georg Gunther, Sir Wilfred Grenfell College, Corner Brook, Newfoundland,
for collecting the set.

2nd U.K. S C H O O L S M A T H E M A T I C A L C H A L L E N G E
2nd February 1989—Time allowed: 1 hour

1 . If the following fractions are written in order of size, which will be in the middle?

A. - B. — C. 31% D . 0.03 E . 0.303

3 10
2* I start counting at 19 and go on to 89, taking one second to say each number.
How long do I take altogether?
A* 1 min 10 sec B . 1 min 29 sec C. 1 min 11 sec D . 1 min 19 sec E* 1 min exactly
36

3 . The diagram shows a regular pentagon with

two of its diagonals. If all the diagonals are drawn in,
into how many areas will the pentagon be divided?
&
A. 4 B. 8 C. 11 D . 10 E. 5

4 . The names of the whole numbers from one to twelve are written down in the
order they occur in a dictionary. What is the fourth number on the list?
A. four B. five C. six D . seven E. nine
5 . T h e graph shows how the weight of a letter ™&*
(including the envelope) varies with the number of sheets 8
3J
2
~F
/
of paper used. What is the weight of a single sheet of | /
paper? ^
A. 5g B . lOg C. 15g D . 20g E . 25g
Y
1 2 3 4

6 . In a magic square, the three numbers in each

row, in each column, and in each diagonal add up to the [l3
same number. When the magic square shown here is 10
completed which of the following numbers is not used?
Li. jrl
A- 6 B. 8 C. 12 D . 14 E . 15

7 . Which of these numbers is the average (mean) of the other four?

A. 28 B . 30 C. 26 D . 37 E . 29
8 . The square has area 36cm2. What is the area
(in cm2) of its inscribed circle?
A. 6TT B. 9TT C. 12TT D. 36TT E. 81TT n
9 . An ant is crawling in a straight line from one corner of the table to the opposite
LJ
corner when he bumps into a one centimetre cube of sugar. Instead of crawling round it,
or eating his way through it, he climbs straight up and over it before continuing on his
intended route. How much does this detour add to the expected length of his journey?
A. 1cm B . 2cm C. 3cm D . 4cm E. Can't be sure

1 0 . When the 8th of December 1988 is written in abbreviated form 8/11/88, we

see that 8 x 11 = 88. How many such dates will there be in 1990?
A. 5 B. 4 C. 3 D. 2 E. 1

1 1 . I want to read the coded message shown

here. I know that each number stands for a letter, but I 26 5 12 18 23
have lost the 'key'. All I can remember is that 26 stands
26 25 24 22
for 'A' and that 5 stands for 4 V\ What does 23 stand
for?
A. B. C. D. E.
 37

1 2 . One million is approximately the number of

A* grains of sand on a beach B . seconds in a day C, people in England and Wales
D . grains of sugar in a cupful E* blades of grass on a football pitch

1 3 . The Kryptor Faction required competitors

to choose one of these five jigsaw pieces and to fit it into
the hole on the right without turning the piece over.
Which is the correct piece?
A. B. C. D.

c ^

14* Baby's nearly 1 now. We've worked out how to weigh her, but nurse and I
still have trouble measuring her height. She just will not stand up straight against our
measuring chart. In fact she can't stand up at all yet! So we measure her upside down.
Last year nurse held Baby's feet, keeping them level with the 140cm mark, while I read off
the mark level with the top of Baby's head: 97cm. This year it was my turn to hold the
feet. Being taller than nurse I held them against the 150cm mark while nurse crawled on
the floor to read the mark level with the top of Baby's head: 84cm. How many centimetres
has Baby grown in her first year?
A. 13 B . 237 C. 53 D. 23 E. 66
1 5 . In the diagram the lengths SP, SQ and SR are
equal and the angle SRQ is x°. What is the size (in degrees)
of angle PQR1
A- 90 B . 2x C. 3x D . 180 -x E . 180 -2x P
1 6 . This grid can be filled up using only the num- 13 4 5
bers 1, 2, 3, 4, 5 so that each number appears just once in a 2
row, once in each column, and once in each diagonal. Which
number goes in the centre square?
A. 1 B. 2 C. 3 D. 4 E. 5 JJ

1 7 . A car with five tyres (four road tyres and a spare) travelled 30,000 km. All
five tyres were used equally. How many kilometres wear did each tyre receive?
A. 6,000 B- 7,500 C. 24,000 D . 30,000 E. 150,000

1 8 . When a 16 by 9 rectangle is cut as shown here

the pieces can be rearranged to make a square of perimeter
A. 32 B . 36 C. 40 D . 48 E. 51
38

1 9 . Our school dinners offer the same basic choice each day: Starters: soup or
fruit juice; Main course: meat, fish, salad, chicken or curry; Afters: crumble, cake or
sponge. I change my choice of course each day, trying each option in turn, going back to
soup after fruit juice, to meat after curry, and to crumble after sponge. Today, to celebrate
the 2nd SMC, I shall sit down to Soup, Meat and Crumble. How many school dinners will
I have eaten before I next sit down to the same combination?
A. 3 B. 5 C. 10 D . 15 E. 30

2 0 . A knitted scarf uses three balls of wool. I start the day with 6 balls of wool
and knit s scarves. How many balls of wool do I have at the end of the day?
A. 3 ( 6 - 5 ) B. 6 - 3 - 5 C. 6 - 3 5 D . 365 E. 5 + 36

2 1 . A stone is dropped from the top of a high tower. Which of the following graphs
shows how its speed changes with time up to the moment when it hits the ground?
• speed •speed speed speed

B. C. D. E.
2 2 . When Ann stands on Ben's shoulders she can just see over a wall. When
Ben stands on Con's shoulders he can see nothing but bricks. When Con stands on Den's
shoulders she can see over easily. Who are the tallest and the shortest?
A. Ann k, Ben B . Den & Con C. Den k, Ben D . Ann &; Con E. Can't be sure

2 3 * What fraction of the area of the regular

hexagon is the shaded triangle?
A A B- I °* 8 D -5- E. i
IJm
12

2 4 . A crossnumber is like a crossword ex- Across Down

cept that the answers are numbers, with one digit 1. See 3 down 2. Square
in each square. What is the sum of all the digits 3. Cube 3. 4 times answer
in the solution to this crossnumber? 4. 5 times answer to 1 across
A. 15 B . 17 C. 23 D . 26 to 3 down 1 2

E . Can't be sure Kill

3

4 Hi
2 5 . How many patches (pentagons and hexagons) were
stitched together to make this football?
A. 15 B . 19 C. 30 D . 32 E . 34
 39

As an Olympiad level contest this issue we give the Final Round of the Czechoslovak
Mathematical Olympiad for 1992. As there is no longer a country of that name, this also
marled the last. Thanks again go to Georg Gunther who collected this contest (and others)
while Canadian I.M.O. Team leader in Moscow.

CZECHOSLOVAK MATHEMATICAL OLYMPIAD

Final Round 1992

1* Let p = (ai, a 2 , . . . , ai?) be any permutation of numbers 1,2,..., 17. Let kp

denote the greatest index k for which the inequality

fli + a2 H \- ®k < fljb+i + Q>k+2 H 1- &17

holds. Find the greatest and the smallest possible value of kp and find the sum of all
numbers kp corresponding to all different permutations p.
2 . Let a, 6, c, d, e, / be the lengths of edges of a given tetrahedron and S be its
surface area. Prove that

S <^(a2 + b2 + c2 + d2 + e2 + f).
6

3« Find all natural numbers n which satisfy equalities

S(n) = S(2n) = S(3n) = ... = S(n2)

if S(x) denotes the sum of digits of the number x (in decimal).

4« Find all solutions of the equation

cos 12# = 5 sin 3x + 9 tan 2 x + cot2 x.

5* A function / is defined on the interval (0,1) by

f(x) = { x,

i D+i
for x irrational

*
h

£ w ^ p^ y coprime, positive and p < q.

Find the maximum value of / on the interval (7/8,8/9).
6* In a plane the acute triangle ABC is given. Its altitude through vertex B
intersects the circle with diameter AC in points P , Q and the altitude through point C
intersects the circle with diameter AB in points M, N. Prove that all the points M, N,
JP9 Q lie on the same circle.
* * *
40

First an apology. I did not list D.J. Smeenk, Zaltbommel, The Netherlands, when I
gave the solution to problem 4 of the 1990 Dutch Mathematical Olympiad, Second Round
[1994: 13]. His solution was inadvertently stuck with solutions he sent in to problems from
the 1991 numbers of Crux which I am saving to use later on.
# # *

Now we turn to the remaining solutions sent in to problems in the December 1992
number of Crux. These are problems of the Vietnamese National Olympiad in Mathematics
For Secondary Schools, 1991 [1992: 296-297].
1 . Determine all functions / : H —> R such that

\f(*v) + \f0») ~ /(*)/(»*) > \

holds for arbitrary x,y,z G H.
Solutions by Seung-Jin Bang, Albany, California; by Beatriz Margolis, Paris, Prance;
by Pavlos Maragoudakis, Pireas, Greece; by Gottfried Perz, Pestalozzigymnasium, Graz,
Austria; and by Edward T.H. Wang, Wilfrid Laurier University, Waterloo, Ontario.
Clearly the constant function 1/2 satisfies the given equality. We show this is the
only possibility.
Setting x = y = z = 0 one gets

|/(0) + |/(o) - /2(0) > i

so [/(0) - 1/2]2 < 0 and /(0) = 1/2. Similarly / ( l ) = 1/2.
Now setting y = z = 0 one obtains

|/(0) + |/(0) - /(*)/«>) > I ,

giving f(x) < 1/2.
From setting y = z = 1,

\m+\s(x) - Hx). /(i) > I

and f(x) > 1/2. Thus f(x) = 1/2 for all x € 1R.
2 . Let k be an odd integer (k > 1). For every positive integer n, denote by f(n)
the greatest non-negative integer such that (kn — 1) | 2^nK Find a formula for / ( n ) in
terms of k and n.
Comment and solution by Seung-Jin Bang, Albany, California.
It is clear that (kn - l)|2^(n> should be replaced by 2f^\{kn - 1).
Let n = 2 s m (where m is an odd integer). Since

kvm _ ! = ( f c 2- _ 1)((jbse*)««-i. + . . . + k2' + 1) = (k2' - 1) • (odd integer)

we have / ( n ) = /(2 s ).
 41

Let k = 2pl + 1 = 2gm — 1, where /, m are odd integers. Since

(2r + l ) 2 + 1 = ! ( ' ; (2r)*" + 2 = 2- (odd integer)

for t = 1,2,... we have

k2' - 1 = (fc - l)(fc + l)(fc2 + l)(fc22 + 1 ) . . . (k2"1 + 1)

= (k - l)(k + l)2 a ~ 1 • (odd integer)

and/(2 s ) =p + q + s-l.
Answer: f(n) = a(k — 1) + a(k + 1) + a(n) — 1, where a(a) is the greatest integer
b such that 2b\a.
5 . Let a triangle ABC with centre G be inscribed in a circle of radius R. Medians
from vertices A, B, C meet the circle at D, E, F, respectively. Prove the inequalities

3^ 1 1 1 r / 1 1 1 \
R " GD + GE +
GF ~ V3
\AB +
BC +
CA~) '

Solution by Pavlos Maragoudakis, Pireas, Greece.

Let O be the centre and R the radius of the
circumscribed circle. Then AG - GD = R2 — OG2 so
1 AG
GD R - DB2
2

Therefore

GD ZRt-OG2'
Similarly

n and
GE ZRt-OG2 GF SR2-OG2'
So
1 1 1 2 ra + rb + rc
+ + (1)
GD GE GF 3* R2-G02
By Leibnitz's Theorem GO2 = R2 - (1/9)(a 2 + b2 + c2). Now (1) becomes

1 , 1 , 1 „r 0 + r 6 + r c
GD GE GF a2 + b2 + c2

Finally it is enough to prove

3 Rri + rb + rc
(2)
R~ at + P + c2 \a b c/
42

(i) We have a2 + b2 = 2r2 + <?/2 so

a2 + 62 + c2 = 2r2 + ^ > 2 ^ ^ .

So {c?+lP+<?)/c > 2>/3r c . Similarly (<?+&+<?)/a > 2 ^ ^ and (a 2 +6 2 +c 2 )/6 > 2y/Zrh.
By addition (a2 + b2 + c 2 )(l/a + 1/6 + 1/c) > 2\/3(r a + rb + r c ). This gives

(ii) We shall use the inequality

4flra > 62 + c2. (3)

Now we have the following:

2
. «22&
2 + 2c2 - a2 ^ ,, 22 ,2, ,2 /a6c\2262 + 2 c 2 - a 2 /l2 2.2
(3) <*=>• 4i? > (b + c ) <*=> ( — J > (b2 + c 2 ) 2 .

This is equivalent to

4 a W ( 2 6 2 + 2c2 - a2) > (a + 6 + c)(a + c- b)(a + b-c)(b + c- a)(b2 + c 2 ) 2

or

8aW + 8 a W - 4 a W > [(b + c)2 - a2][a2 - (b - c)2](b2 + c?)2

= [2a2(62 + c 2 ) - o 4 - ( 6 2 - c 2 ) 2 ] ( 6 2 + c 2 ) 2
= 2a2(62 + c 2 ) 3 - a4(b2 + c 2 ) 2 - (ft2 + c 2 ) 2 (6 2 - c 2 ) 2 .

This becomes
2 a W + 2 a W - 2a266 - 2a2c6 + (b2 + <?)2{b2 - c 2 ) 2 + a4(62 - c 2 ) 2 > 0

or
2a2[64(c2 - b2) + c\b2 - c2)] + [{b2 + c 2 ) 2 + o4](62 - c 2 ) 2 > 0
which becomes
-2a 2 (6 2 - <?f(b2 + (?) + [(b2 + c 2 ) 2 + a4](b2 - c 2 ) 2 > 0

whence
(b2 - c2)2[(62 + c 2 ) 2 + a4 - 2a2(62 + c2)] > 0
which is (62 - c2)2((62 + c2) - a 2 ) 2 > 0 and that is true!
Similarly 4-Rr& > a2 + c2 and 4i?rc > a2 + b2. On addition 4i?(ra + r& + rc) >
2(a 2 + 62 + c 2 )so
L <er* + r1> + rc
fl - a2 + 62 + c2
as required to complete the solution.
 6* Let x, y, z be positive real numbers with x > y > z. Prove

z x y

Solution by Pavlos Maragoudakis, Pireas, Greece.

z x y
just in case

£f<^£> + itpl > ^ (£z£) = U _ y) + £ t o _ ,,.

z y
1/ \ z / a? ar
The problem is equivalent to showing

or

\ z z z xJ \x x x yJ
This becomes
(y-z)(x-y)(x + y) (y-z)y2(x-z) > (x-y)(y - z)(y +z) (x-y)z2(y-x)
z xz "~ x xy
Now this is
*2(x _ „,\2
(s - y)(y - z) (£±g - £±£) + y2(y - ^ g - g) + *2(*" y)2 > o.
V z x / arz xy
a:?/
Finally we obtain

( g - y ) ( y - * ) ( g - * ) ( 3 + y + *) y2(y-^)(^~^) *2(ar-y)2>0
arz arz xy ~~

which is true since x >y > z. Note that equality holds just if rr = t/ = z.
Alternate solution (using calculus) by Seung-Jin Bang, Albany, California; and by
Edward T.H. Wang, Wilfrid Laurier University, Waterloo, Ontario.
The given inequality is equivalent to

x2y + — + - > x2 + y2 + 1 for x > y > 1.

x y
Let f(x) = a;3y + y2 + x2/y — a:3 — (y2 + l)a: (for y fixed).
44

Then f'(x) = 3x2t/ + (2x)/y - Zx* - (y2 + 1) and /"(a) = 6xy + 2/y - 6x =
6x(y-l) + 2/y>0. From /'(y) = (3y2-y - l ) ( y - 1) = (y-i)((2y + l)(y- 1) + */2) > 0
we have /'(x) > 0 (for x >y).
Since /(y) = y2(t/ — l ) 2 > 0 we conclude that /(#) > 0 for x > y, which completes
the proof. Note that the equality holds only if x = y = z.
9 . Given is a sequence of positive real numbers a?i, #2, • • • ? #n> • • • defined by the
formula: #i = 1, x 2 = 9? #3 = 9, x 4 = 1,

#n+4 = v^n#n+i#n+2#n+3 if n > 1.

Prove that this sequence is convergent and find its limit.
Solutions by Seung-Jin Bang, Albany, California; and by Pavlos Maragoudakis,
Pireas, Greece. We give the solution of Maragoudakis.
From the formula we have x\ = 1, x^ = 9, x$ = 9, #4 = 1, #5 = 3 4 , x& = 3 4 ,
x7 = 3 1 0 / 3 .
First we shall prove by induction that 1 < xn < 3 4 for all n > 1. It is enough to
prove that 1 < x* < 3 4 , Vfc < n, for all n > 1.
For n < 7, the statement is true.
Now suppose that

1 < xk < 3 4 , VA; < n where n > 1. (1)

We argue that 1 < xk < 3 4 , VA; < n + 1. It is enough to prove that 1 < xn < 3 4 . From the
first remark we can assume n > 8. Then
n
#n = - ^ X n _ 4 • ar n _ 3 • £ n _ 2 • # n - l •

By (1), 1 < 3 4 . So

l < x n < tt-^/3I6 = 316/(n-"4).

Also n>8=> 16/(n — 4) < 4 so 1 < x n < 3 4 . This completes the induction.
Now for n > 5

1 < Xn = «-yXn-4 • X n _ 3 • *n-2 • *»-l < 3 1 6 / ( n ~ 4 ) .

But linin^oo 316/(n"4) = 1 so limn^ooXn = 1.

* * *

Next an update on one of the errors that I missed, and for which a corrected solution
by Stan Wagon was published earlier. Another faithful reader spotted the earlier flaw, gave
us a solution and a reference.
8 . [1992: 196; 1993: 228, 260] Proposed by the U.S.S.R.
Let an be the last nonzero digit in the decimal representation of the number n\.
Does the sequence become periodic after a finite number of terms?
 45

Comment by Waldemar Pompe, student, University of Warsaw, Poland.

The problem has appeared earlier- I found it in a Russian book "Foreign Mathe-
matical Olympiads," published in 1987!
* * #

Next we give a nice geometric solution to a problem of the Canadian Mathematical

Olympiad. The official solution we published involved more trigonometry.
3 . [1993: 159, 196] 1993 Canadian Mathematical Olympiad.
In triangle ABC, the medians to the sides AB and AC are perpendicular. Prove
that cot B + cot C> §.
Solution by Waldemar Pompe, student, University of Warsaw, Poland.
Let AH be the altitude, G be the centroid and M the midpoint of BC of a given
triangle. Since M is a circumcentre of ABCG, we have MB = MC = MG = x and
AG = 2x. Therefore
„ BH CH BC ^ BC 2x 2
cot B + cot C = 1- - — = > = — = - .
AH AH AH - AM 3x 3
Equality holds only if AB = AC.
This solution also works when one of the angles B or C is obtuse.
# * *

We finish this month's Corner with three comments/solutions by Murray Klamkin

regarding problems from the 1991 numbers of the Corner.
2* [1986: 202; 1991: 200] 1986 USAMO Training Session.
Determine the maximum value of
x3 + yz + zz - x2y - y2z - z2x
for 0 < x,y, z < 1.
Comment by Murray S. Klamkin, University of Alberta.
Here we give a generalization with a simpler solution.
We find the maximum value over x, y, z of
S = a(xp + yp + zp)~ b(xqyr + yqzT + zqxr) - c{xqyr + yqzr + zqxr)
where a, 6, c > 0, jp > g > r > 0, and 0 < X, y, 2 < 1.
Clearly,
5 < a(s* + yq + zq) - (6 + c)(arV +y*** + 2%«).
Since the latter right hand expression is linear in xq, yq, zq, it takes on its maximum value
at the end points xq, yq, zq = 0 or 1. Hence,
Smax = max{3(a — 6 — c), 2a — 6 — c, a}.
We can extend the result further by having different suitable exponents, e.g., xPl +yP2+zP3
instead of xp + yp + zp where p\ > p2 > jps > <?, etc. Also, one can increase the number of
variables.
46

3 . [1989: 65; 1991: 8] 1987 Annual Greek High School Competition.

Let A be an n x n matrix such that A2 — 3A + 21 = 0 where J is the identity matrix.
Prove that for all natural numbers &,
A2k _ yk + yAk + 2kj = 0

Comment by Murray S. Klamkin, University of Alberta.

More generally,

if n ( A — art-7) = 0, then ]\{Ami - x^I) = 0

where the products are over t = 1 to n, the rat's are positive integers and the a?;'s are
scalars. This follows immediately from the factorization

(Ami - x?*i)=n(^ - ^ v )
where the product is over 1 to mz-, a? is a primitive mt-th root of 1 and that the factors on
the right hand side commute, i.e., (A — xI)(A — yl) = (A — yI)(A — #J).
3 . [1991: 2; 1992: 102] Celebration of Chinese New Year Contest
Let f(x) = x" + x 98 + x97 H h x2 + x +1. Determine the remainder when f(x100)
is divided by f(x).
Comment by Murray S. Klamkin, University of Alberta.
A generalization with a simpler proof than the previous one is as follows:
Let f(x) = 1 + x H h x71"1. Then the remainder when f(xmn) is divided by f(x)
is n where m is any positive integer. Here, f{x) = (xn —1)/(# — 1) and we rewrite f(xmn)
in the form

f(xmn) = (armn<n-1> - 1) + (a;mn(n-2) - l) + . . . + ( x m n - 1) + n.

Then since xfcn — 1 is divisible by xn — 1, the remainder is immediately n.

The case when n = 5 with similar proof appears in a problem by Norman Anning,
School Science and Math., 54 (1954), 576.
* * *

That completes this number of the Corner. The Olympiad season approaches. Send
me your national/regional contests, and your nice solutions. Also please send any suitable
pre-Olympiad material.
 47

PROBLEMS
Problem proposals and solutions should be sent to B. Sands, Department of
Mathematics and Statistics, University of Calgary, Calgary, Alberta, Canada T2N 1N4-
Proposals should, whenever possible, be accompanied by a solution, references, and other
insights which are likely to be of help to the editor. An asterisk (*) after a number indicates
a problem submitted without a solution.
Original problems are particularly sought But other interesting problems may also
be acceptable provided they are not too well known and references are given as to their
provenance. Ordinarily, if the originator of a problem can be located, it should not be
submitted by somebody else without permission.
To facilitate their consideration, your solutions, typewritten or neatly handwritten
on signed, separate sheets, should preferably be mailed to the editor before S e p t e m b e r 1,
1994, although solutions received after that date will also be considered until the time when
a solution is published.

1 8 9 0 . [1993: 265] (corrected) Proposed by Federico Ardila, student, Colegio San

Carlos, Bogota, Colombia.
Let n be a positive integer and let
to A rC If

f(x) = anxn + an-i^"" 1 + g(x) = — xn H

h aix + a 0 , xn~l + h —x + — ,
an an__i a\ aQ
where k and the a t 's are positive real numbers. Prove that

f(g(l))g(f(l)) > 4k.

When does equality hold?

[Editor's note. The error in the previous statement of this problem was due to the
editor's overzealous "simplification" of the proposer's original version. Sorry about that.]
1911. Proposed by Marcin E. Kuczma, Warszawa, Poland.
Let B and E be two opposite vertices of the regular icosahedron. Consider the
random walk over the edge-skeleton of the icosahedron, beginning at B. Each time he
arrives at a vertex, the walker continues the walk along any one of the five edges emanating
from that vertex, with equal probability of each choice. The walk ends when the walker
reaches vertex E. Find the expected length of the walk (in number of edges).
1912. Proposed by Toshio Seimiya, Kawasaki, Japan.
ABC is a triangle with AB ^ AC. Similar triangles ABD and ACE are drawn
outwardly on the sides AB and AC of AABC, so that LABD = LACE and IB AD =
LCAE. CD and BE meet AB and AC at P and Q respectively. Prove that AP = AQ if
and only if
[ABD] • [ACE] = [ABC]2,
where [XY"Z] denotes the area of triangle XYZ. (This problem is an extension of Crux
1537 [1991: 182].)
48

1913. Proposed by N. Kildonan, Winnipeg, Manitoba.

I was at a restaurant for lunch the other day. The bill came, and I wanted to give
the waiter a whole number of dollars, with the difference between what I give him and the
bill being the tip. I always like to tip between 10 and 15 percent of the bill. But if I gave
him a certain number of dollars, the tip would have been less than 10% of the bill, and if
instead I gave him one dollar more, the tip would have been more than 15% of the bill.
What was the largest possible amount of the bill? [Editor's note to non-North American
readers: your answer should be in dollars and cents, where there are (reasonably enough)
100 cents in a dollar.]
1914. Proposed by K.R.S. Sastry, Addis Ababa, Ethiopia.
Let A1A2... An be a regular n-gon, with Mi, M 2 , . . . , Mn the midpoints of the sides.
Let P be a point in the plane of the n-gon. Prove that

] T PMi > cos(180°/n) JT PA.

1915. Proposed by Richard K. Guy, University of Calgary.

Is (*J ever relatively prime to ("J for 0 < r < s < n? (This is not a new problem.
Its history will be revealed when a solution is published.)
1916. Proposed by G. Tsintsifas, Thessaloniki, Greece.
Let A\ A2... An+i be a regular simplex inscribed in a unit sphere in n-dimensional
space, and let P be a point on the sphere. Prove that

n
i=i

1917. Proposed by Joaquin Gomez Rey, LB. Luis Bunuel, Alcorcon, Madrid,
Spain.
For any positive integer n, evaluate an/bn where

a =
- S,an22lh' 6 = tan2
" n i *r?7-
1918. Proposed by D.J. Smeenk, Zaltbommel, The Netherlands.
ABC is a triangle with circumcentre O and incentre i", and K, L, M are the
midpoints of BC, CA, AB respectively. Let E and F be the feet of the altitudes from B
and C respectively.
2 2 2
(a) If OK = OL + OM , show that £", F and O are collinear, and determine all
possible values of IB AC.
(b) If instead OK = OL + OM, show that £*, F and I are collinear, and determine
all possible values of IB AC.
 49

1919. Proposed by H.N. Gupta, University of Regina.

In Lotto 6/49, six balls are randomly drawn (without replacement) from a bin
holding balls numbered from 1 to 49. Find the expected value of the fcth lowest number
drawn, for each k = 1,2,..., 6.
1920. Proposed by Walther Janous, Ursulinengymnasium, Innsbruck, Austria.
Let a, 6, c be the sides of a triangle.
(a) Prove that, for any 0 < A < 2,

1 ^ (a + b)(b + c)(c + a)
2
(1 + A) (Xa + b + c)(a + Xb + c)(a + b + Ac) V2 + A) '
and that both bounds are best possible.
(b)* What are the bounds for A > 2?

SOLUTIONS
No problem is ever permanently closed. The editor will always be pleased to consider
for publication new solutions or new insights on past problems.

1 6 9 5 . [1991: 301; 1992: 287] Proposed by Seung-Jin Bang, Seoul, Republic of

Korea.
Let p(x) = ao + a\x + a2x2 + &3#3 + &4^4 + ahx^ w
ith a® > 0 and
a0 + a2 a2 + a4 a4
«o + ^ - +- T - + y<0.

Prove that there exists at least one zero of p(x) in the interval (—1,1).
III. Comment by Jun-hua Huang, The 4th Middle School ofNanxian, Hunan, China.
Here is an answer to the editor's question appearing on [1992: 288]: the interval
(—3/Vl4,3/%/l4) 5 shown on [1992: 287] to contain a zero of p{x), can not be further
shortened.
We take a\ = as = a& = 0, ao = 1 and

14 a
196 +35s+ e2
a2 = 5 44 —
e' 9e
where e > 0 is a constant. It is easy to verify that
a0 + a2 a2 + a4 a4
«o + - ^ — + - r ~ + y<0,

because this is equivalent to 35ao + 14a2 + 9a4 < 0, which is true. Also p(—x) = p(x),
and we now prove that p(x) > 0 for all 0 < x < 3/V14 + e. £>inee e > 0 is arbitrary, this
establishes the result.
50

Letting t = x2, we have p(x) = f(t) = a4t2 + a2t + 1 where 0 < t < 9/(14 + s).
Since /(0) = 1 > 0 and

/ 9 N 9 /(14 + g)» 14 196 + 35 g + e 2 \

7
U 4 + J " (14 + e)2 V 9 +
e^*
+
^ £ )
9 2
- /(14 + g) \ _ 9 /7 19e £2\
2 21 +
" ( 1 4 + e) V 9 V~(14 + e)2^9 9 + 9j > 0
'

and since a^ < 0 [so / is a parabola opening downward], hence

/(t)>ndn{/(0) f /( n ^)}>0
for 0 < t < 9/(14+ e). So p(x) > 0 for all 0 < x < 3/y/uTe.

1 8 1 6 . [1993: 49] Proposed by Marcin E. Kuczma, Warszawa, Poland.

Given a finite set S of n + 1 points in the plane, with two distinguished points B
and E in 5 , consider all polygonal paths V = PQP\ . . . Pn whose vertices are all points of
<S, in any order except that PQ = B and Pn = E. For such a path V define l(V) to be the
length of V and

a ( P ) = ^ ( ^ , ^ + i ) ,

where 0(v,w) is the angle between the vectors v and w, 0 < 0(v,w) < 7r. Prove or
disprove that the minimum values of l(V) and of a(V) are attained for the same path V.
Solution by the proposer.
The statement is false; here is a counterexample. Let n = 5, and let
S = {B,E,P,Q,R,S} where P and Q trisect BE and PQRS is a square. Then of
the two paths BPQSRE and BPSRQE, one has a(V) greater and the other has l(V)
greater.

B P Q E B P Q E

[Editor's note. In fact it is clear that BPSRQE has the minimum l(P) over all
paths V, while

a(BPQSRE) = ^ + ^ + J = ^<2TT = a{BPSRQE).

So as the proposer says, the shortest path need not be the "least curved" one.]
 51

_ 1 8 1 7 . [1993: 50] Proposed by K.R.S. Sastry, Addis Ababa, Ethiopia.

For a natural number n, d(n) denotes the number of positive integer divisors of
n (including n itself) and <f>(n) denotes the number of positive integers less than n and
relatively prime to n. Find all positive integers n so that d(n) + 0(n) = n.
Composite solution by Nick Lord, Tonbridge School, Kent, England, and Edward
T.H. Wang, Wilfrid Laurier University, Waterloo, Ontario.
We show that
M \ , x( \ \ n if n = 6,8,9,
d(n) + 0(n) = < i V
v
' [ » + l if n = 1,4 or a prime,
and d(n) + (f>(n) < n otherwise.
Clearly d(l)+(f>(l) = 1+1 = 2 and if n is a prime, then d(ri)+<f>(ri) = 2+n—1 = n + 1 .
We assume then n > 4 and is a composite. Since 1 is the only positive integer counted by
both d(n) and 0(n), we have d(n) + <f>(n) < n if there exist distinct a, 6 G {2,3, • • •, n — 1}
which are neither divisors of n nor relatively prime to n. If n > 10 is even, then we can
take a = n — 4, 6 = n — 2 since b > a > n/2 implies neither of them are divisors of n.
If n > 16 is odd, then it has a prime factor p with p < yfri < n/4 and hence we can
take a = 2/?, b = 4p. Directly checking the remaining cases when n = 4,6,8,9 and 15
reveals that d(4) + <f>(4) = 3 + 2 = 5, d(6) + 0(6) = 4 + 2 = 6, d(8) + 0(8) = 4 + 4 = 8,
d(9) + 0(9) = 3 + 6 = 9, and d(15) + 0(15) = 4 + 8 = 12.
Also solved by H.L. ABBOTT, University of Alberta; FEDERICO ARDILA,
student, Colegio San Carlos, Bogota, Colombia; SEUNG-JIN BANG, Albany, California;
CHARLES R. DIMINNIE, St Bonaventure University, St Bonaventure, New York;
HANS ENGELHAUPT, Franz-Ludwig-Gymnasium, Bamberg, Germany; RICHARD L
HESS, Rancho Polos Verdes, California; JOHN G. HEUVER, Grande Prairie Composite
H.S., Grande Prairie, Alberta; WALTHER JANOUS, Ursulinengymnasium, Innsbruck,
Austria; KEE-WAI LAU, Hong Kong; PAVLOS MARAGOUDAKIS, Pireas, Greece;
HENRY J. RICARDO, Tappan, New York; SHAILESH SHIRALI, Rishi Valley School,
India; CHRIS WILDHAGEN, Rotterdam, The Netherlands; and the proposer. Three other
readers submitted the correct solution without proof, and one reader sent in an incorrect
solution.
Actually, strictly according to the definition given in the problem, 0(1) = 0 (there
are no positive integers less than 1!), so n = 1 should also be a solution. However, readers
apparently realized that 0(1) = 1 is the correct definition.
* * * * #

1 8 2 1 . [1993: 77] Proposed by Gerd Baron, Technische Universitdt, Vienna, and

Walther Janous, Ursulinengymnasium, Innsbruck, Austria.
Determine all pairs (a, 6) of nonnegative real numbers such that the functional
equation
f(f(x))+f(x)=ax +b
has a unique continuous solution / : H -+ R.
52

Solution by the 1993 U.K. I.M.O. team.

The required pairs are precisely: all (2,6) where 6 > 0.
There exists a linear solution f(x) = ex + d precisely when c(cx + d) + d+ (cx + d) =
ax + 6 for all x\ i.e., when
c(c+l) = a (1)
and
(c + 2)d = 6. (2)
For any value of a > 0, equation (1) yields two possible values of c. If c ^ —2, then each
value of c corresponds, via equation (2), to a unique value of d, namely d = b/(c + 2).
Hence for any a > 0, if c ^ —2, there are at least two linear (hence continuous) functions
satisfying the given functional equation. Thus this equation can have a unique continuous
solution only for values of a which give rise to c = — 2 — that is, when a = 2. We therefore
assume from now on that, for some b > 0, / satisfies

/ ( / ( * ) ) + /(*) = 2x + b for all a:. (3)

If b = 0, then we get two possible linear functions, namely f(x) = — 2x and /(#) = x.
Thus we need only consider the case where 6 ^ 0 : c = — 2 is then impossible, so c = 1,
and there is a unique linear function, f(x) = # + 6/3. We prove that, for each b > 0, this
is the only continuous function satisfying (3).
Let / be any such function. Since 6 ^ 0 , (3) shows that for every #, f(x) ^ x, so the
graph of / lies wholly above or wholly below y = x. Since b > 0, we cannot have f(x) < x
for all x; hence f(x) > x for all x. By (3), / must be one-one. Since / is continuous,
/ is either increasing or decreasing, and must in fact be increasing (otherwise the graphs
y = f(x) and y = x would cross).
Lemma. For all x, we have
(i) x + 6/4 < f(x) <x + 6/2;
(ii) more generally, for all integers n>\,

x+
[~T~J <f(x)<x + b
[-—3—I-
Proof. We prove part (i); the induction step for part (ii) is almost identical. We
know that f(x) > x for all x. Suppose /(#o) ^ #o + 6/2 for some XQ. Then

f(f(x0)) + f(x0) > f(xQ) +x0 + 6/2 > 2XQ + 6,

contradicting (3). Hence x < f(x) < x + 6/2. If f(xo) < XQ + 6/4 for some rco, then

f(f(x0)) + f(x0) < f{x0) + 6/2 + x0 + 6/4 < 2x0 + 6,

which also contradicts (3). •

Taking limits in part (ii) shows that the only possible / is f(x) = x + 6/3.
 53

Also solved by MARGIN E. KUCZMA, Warszawa, Poland; WALDEMAR POMPE,

student, University of Warsaw, Poland; ACHILLEAS SINEFAKOPOULOS, student,
University of Athens, Greece; A. N. 'T WOORD, Eindhoven University of Technology,
Eindhoven, The Netherlands; and the proposer. Three other readers gave the correct set of
pairs (a, 6) without a complete proof
Sinefakopoulos's solution was the same as the above one,

1 8 2 2 * [1993: 77] Proposed by Toshio Seimiya, Kawasaki, Japan.

AB and AC are tangent to a circle F at B and C respectively. Let D be a point on
AB produced beyond J?, and let P be the second intersection of F with the circumcircle
of AACD. Let Q be the foot of the perpendicular from B to CD. Prove that LDPQ =
2LADC.
Combination of solutions of Waldemar Pompe, student, University of Warsaw,
Poland, and the proposer.
Let E be the second intersection of F „
with the line CD. Then

LAPD = LACD = LCPE

since AC is tangent to F at C, so

LEPD = LCPA = LCD A.

This means that the circumcircle of ADEP is

tangent to AD at D. Let PE and AD meet at
M. Then

(DM)2 = ME*MP= (MB)2

since MB is tangent to F at B, which gives us DM = MB. Therefore M is the midpoint
of BD and is also the circumcentre of ADBQ. This means that

LDPE = LEDM = LDQM.

Thus D, M, Q, P are concyclic. Hence

LDPQ = LDPE + LMPQ = ZADC + LADC = 2ZADC.

-Afeo solved by CHRISTOPHER J. BRADLEY, Clifton College, Bristol, U.K.;

P. PENNING, Delft, The Netherlands; and A. N. T WOORD, Eindhoven University
of Technology, Eindhoven, The Netherlands.
The solutions of Pompe and the proposer were almost identical.
54

1 8 2 3 . [1993: 77] Proposed by G.P. Henderson, Campbellcroft, Ontario.

A rectangular box is to be decorated with a ribbon that goes across the faces and
makes various angles with the edges. If possible, the points where the ribbon crosses the
edges are chosen so that the length of the closed path is a local minimum. This will ensure
that the ribbon can be tightened and tied without slipping off. Is there always a minimal
path that crosses all six faces just once?
Solution by Marcin E. Kuczma, Warszawa, Poland.
Let the box have size a x b x c, where a = max{a, 6,c}. Unfold the surface and
place it on the coordinate plane as shown:
A
Q

.{ c
1
f
• /
D

al

V A
B

A \y ¥

a
* > ^
b
,
c
• >*
V
a
'
> »

Segment PQ illustrates the ribbon; rectilinearity is imposed by the minimality of

its length. The labelling (a, 6, c) may always be chosen so that the ribbon crosses the faces
in the order as in the diagram.
Let P = (p,0), 0 < p < a. Then Q = (a + 6 + c — p,a + b + c) and hence the line
PQ is given by the equation y = x — p. The existence of the minimal path crossing all
faces just once is equivalent to the requirement that (by a suitable choice of p) the points
A and C should lie above line PQ and points B and D should lie below it.
As A = (a,c), B = (a + 6,c), C = (a + 6,a + c), D = (a + b + c,a + c), this
requirement yields the system of inequalities

a—p<c<a + b — p<a + c<a + b + c — p;

equivalently,

p > max{a — c, b — c} = a — c, p < min{a + b — c, 6} = 6.

Thus p exists if and only if a — c < b; that is, when a, 6, c are the lengths of sides of a
triangle!
Also solved by JORDI DOU, Barcelona, Spain; HANS BNGELHAUPT, Franz-
Ludwig-Gymnasium, Bamberg, Germany; RICHARD I. HESS, Rancho Palos Verdes,
California; P. PENNING, Delft, The Netherlands; and the proposer.
 55

Dou 's solution, a seven-page tour de force of case analysis and three-dimensional
diagrams, includes examples of minimal paths which do not cross the faces of a box in the
above manner! Rather the ribbon traverses a pair of opposite faces from one side to the
opposite side, like so: . —1— —-p .

However, it seems that no box can be decorated this way that cannot also be decorated the
other way.

1 8 2 4 . [1993: 77] Proposed by George Tsintsifas, Thessaloniki, Greece.

Let ABC be a triangle and M a point in its plane. We consider the circles with
diameters AM, BM, CM and the circle containing and internally tangent to these three
circles. Show that the radius P of this large circle satisfies P > 3r, where r is the inradius
of AABC.
Correction by Toshio Seimiya, Kawasaki, Japan.
The conclusion P > 3r is not right, the correct inequality must be P > 2r.
We denote the large circle of radius
P by F. Let O be the center of I\ As the
circle with diameter AM is contained in T5
A lies on or within T. Therefore P > AO.
Similarly we get P > BO and P > CO.
Hence we have

3P>AO + BO + CO. (1)

Using item 12.14 of Bottema et al? Geometric

Inequalities, we get

AO + BO + CO> 6r. (2)

Prom (1) and (2) we have 3 F > 6r, thus P > 2r. Equality holds when AABC is equilateral
and M is its circumcenter.
Counterexample also found by FEDERICO ARDILA, student, Colegio San Carlos,
Bogota, Colombia; CHRISTOPHER J. BRADLEY, Clifton College, Bristol, U.K.; and
WALDEMAR POMPE, student, University of Warsaw, Poland.
The proposer's solution contained a minor arithmetical error which, if corrected,
leads to the result P > 2r.
56

1 8 2 5 . [1993: 77] Proposed by Marcin E. Kuczma, Warszawa, Poland.

Suppose that the real polynomial x4 + axz + bx2 + ex + d has four positive roots.
Prove that abc > a2d + 5c2.
I. Solution by Christopher J. Bradley, Clifton College, Bristol, U.K.
Given that the real polynomial has four positive roots it must factorize over the
reals into the form
(x2 -Ax + B){x2 -Cx + D)
where
A2>4B>0 and C2 > W > 0, (1)
and we may take A, C > 0. Comparing coefficients gives

a = - ( A + C), b = B + D + AC, c = -(AD + BC), d = BD.

Prom (1), evidently

B2 + D2 + A2D + C2B >B2 + D2 + WD > 10BD.

Hence
ACB2 + ACD2 + A3CD + AC*B > 10ABCD.
Also from (1),
A2BC2 > AB2C2 and A2C2D > AA2D2.
Putting these results together,

ACB2 + ACD2 + ASCD + AC3B + A2D2 + B2C2 + A2BC2 + A2C2D

> bA2D2 + 10ABCD + 5B2C2. (2)

After some elementary algebra this gives abc—a2d > 5c2. [Editor's note. For instance, the
left side of (2) is

(A + C)(B2C + AD2 + A2CD + ABC2)

= (A + C)[AD(D + AC) + BC(B + AC)]
= {A + C)[{B + D + AC){AD + BC) - BAD - DBC]
= (A + C)(B + D + AC)(AD + BC) ~(A + C)2BD
= abc — a2d,

while the right side is clearly 5c2.]

II. Solution by Walther Janous, Ursulinengymnasium, Innsbruck, Austria.
In familiar notation, —a, 6, —c, d are respectively the elementary symmetric func-
tions 5i, 52, 53, 54 of the four roots. Using the also familiar notation 5& = u ) p * with (for
us) n = 4, we get 5i = 4pi, 52 = 6^2? £3 = 4^3, 54 = #4, and thus the claimed inequality
is equivalent to
6P1P2P3 > P2IPA + 5pl- (1)
 57

Now the Newton-Maclaurin inequalities state that

r < s => Pr-iPs < PrPs-i (2)

(see e.g. pp. 95-97 of D.S. Mitrinovic, Analytic Inequalities, Blerlin, 1968). Putting r = 2
and s = 4 in (2) we get pip4 < p2pz, and (1) would follow from the sharper inequality
6P1P2P3 > P1P2PZ + $P% i-e.
P1P2 > Pz- (3)

[Editor's comments. Janous then gives a proof of (3) using Lagrange multipliers.
However, this is just the special case r = 1, s = 3 of (2), where we define po = 1.
Of course, (3) is true for any value of n, but Janous then suggests a different
generalization of (3). Writing it as Si £2 > 6S3 and then in terms of the four positive
roots (denoted a, 6, c, d by Janous with the apologetic remark "hopefully causing no
confusion!"), he gets
(£a)(£ab)>6J2abc
(with the sums symmetric over a, 6, c, d), which he then expands into

Y^a2(b + c + d)+3j^abc>6j2abc
or
i£a2(& + c + d ) > 5 > c .
Thus he asks whether, for any n > 2 and ai, 02,.. •, a n > 0 with ^27=1 a* == ^?

2
^ £ a r ( S - a , ) > n a , £ ± . (4)

Equality holds when n = 2, and n = 4 is the case handled above. Can any reader settle
(4) in general?]
Also solved by FEDERICO ARDILA, student, Colegio San Carlos, Bogota,
Colombia; ED BARBEAU, University of Toronto; TIM CROSS, Wolverley High School,
Kidderminster, U.K.; HANS ENGELHAUPT, Pranz-Ludwig-Gymnasium, Bamberg,
Germany; RICHARD I. HESS, Rancho Palos Verdes, California; KEE-WAILAU, Hong
Kong; and the proposer.
Barbeau's book Polynomials (Springer-Verlag, 1989J contains a similar problem (#7
on page 225j. With the same assumption as Crux 1825, it could be written: if c + d = 0,
then a + b > 80.
* * * * *

1 8 2 7 . [1993: 78] Proposed by Sefket Arslanagic, Trebinje, Yugoslavia, and DM.

Milosevic, Pranjani, Yugoslavia.
Let a, 6, c be the sides, A,B, C the angles (measured in radians), and s the semi-
perimeter of a triangle.
58

(i) Prove that

^ be 12s

where the sums here and below are cyclic.

(ii)* It follows easily from the proof of Crux 1611 (see [1992: 62] and the correction
on [1993: 79]) that also

^ A ~ 7T '
Do the two summations above compare in general?
I. Solution to part (i) by Kee- Wai Lau, Hong Kong.
Let R and r be the circumradius and inradius of the triangle. It is well known that

abc = ARrs, tan — = , a = 2iJsin A.

2 s-a
Hence
w
^ A{s-a) ^ Asm A ^ A
By Cauchy's inequality,

^sec2(^/2) ( S s e c ( ^ / 2 ) ) 2 _ (£sec(A/2)) 2

Since secx is convex for 0 < x < 7r/2,

Y, sec(A/2) > 3 sec (A +

^ + C
) = 2v ^ . (3)

The inequality of the problem now follows from (1), (2) and (3).
II. Solution to part (ii) by Richard I. Hess, Rancho Polos Verdes, California.
Let

For a = 5, 6 = 6, c = 10 we get A » .38976, B « .47345, C » 2.27838, and so

5i - 5i « 77.792 - 77.561 > 0.

For a = 6, b = 7, c = 10 we get A « .6315, B « .75976, C » 1.75033, and so

Si - S2 « 53.700 - 55.406 < 0.

Therefore the two summations don't compare in general.

Part (i) also solved by FEDERICO ARDILA, student, Colegio San Carlos, Bogota,
Colombia; TIM CROSS, Wolverley High School, Kidderminster, U.K.; and the proposers.
 59

* * * * *

1 8 2 9 . [1993: 78] Proposed by C.J. Bradley, Clifton College, Bristol, U.K.

The quadrilateral ABCD is inscribed in a circle with centre O. AD and BC meet
at P. L and M are the midpoints of AD and BC respectively. Q and R are the feet of
perpendiculars from O and P respectively to LM. Prove that LQ = RM.
I. Solution by Jordi Dou, Barcelona, Spain.
A quick glimpse at Figure 1 shows the solution. Let UJ be the circle through P ,
L, 0 , M with centre J, the midpoint of OP. [Note that LOLP = ZOMP = 90°, so u
exists.—Ed] The secants PR and OQ are equidistant from / [since they are parallel], and
are perpendicular to the chord LM. Therefore the segments LQ and MR of the chord are
equal.

II. Solution by John G. Heuver, Grande Prairie Composite H.S., Grande Prairie,
Alberta.
See Figure 2. Let H be the orthocenter of ALMP. [Note H lies on PR.] Then
OM\\LH and OL\\MH since OM ± BP and OL ± AP. Prom this follows that in
parallelogram LOMH we have AHMR and AOLQ are congruent, and we conclude that
LQ = MR.
Also solved by FEDERICO ARDILA, student, Colegio San Carlos, Bogota,
Colombia; SEUNG-JIN BANG, Albany, California; P. PENNING, Delft, The Netherlands;
WALDEMAR POMPE, student, University of Warsaw, Poland; TOSHIO SEIMIYA,
Kawasaki, Japan; D.J. SMEENK, Zaltbommel, The Netherlands; A. N. T WOORD,
Eindhoven University of Technology, Eindhoven, The Netherlands; and the proposer. One
incorrect solution was received.
The solutions of Pompe, Seimiya and Smeenk were all similar to Solution I.
# * 5fS * #
60

1 8 3 0 . [1993: 79] Proposed by P. Tsaoussoglou, Athens, Greece.

I f a > 6 > c > 0 and a - 1 + ft-1 + c - 1 = 1, prove that
4_ 1 1 4
c2 +
(a - 6)6 + (6 - c)c - 3 '

Solution by Sefket Arslanagic, Nyborg, Denmark.

From a > 6 > 0 and 6 > c > 0, by the Arithmetic-Mean Geometric-Mean inequality
we have
a = (a - 6) + 6 > 2yj{a - 6)6 and 6 = (6 - c) + c > 2<J(b - c)c,
or
1 ^ 4 1 ^ 4
(a-6)6"a2 (6-c)^62-
Now, we have

c2 + (a - 6)6 + ( 6 - c ) c " \a2 + b2 + c2)' ()

Equality in (1) holds if and only if a = 26 = 4c (i.e., o = 7, 6 = 7/2, c = 7/4 because

G -i + &-i + c - i = i). Since

2
(1-lY
\a b)
(1-lY
\b c)
(±--\
\c a)
implies that
1 1 1 1/1 1 1\2 1
+ + > + + (2)
^ P 5 3la 6 c) = 3
(strict inequality holds here because a > 6 > c > 0), it follows from (1) that
_4_ 1 1 4
2 + + >
c (a-6)6 (6-c)c 3 '
i.e., the given inequality is correct, but only strict inequality holds.
[Editor's note. As many solvers pointed out, (2) also follows by Cauchy's inequality.]
Also solved (often the same way) by FEDERICO ARDILA, student, Colegio San
Carlos, Bogota, Colombia; FRANCISCO BELLOT ROSADO, LB. Emilio Ferrari,
Valladolid, Spain; MIGUEL ANGEL CABEZON OCHOA, Logrono, Spain; TIM CROSS,
Wolverley High School, Kidderminster, U.K.; MARCIN E. KUCZMA, Warszawa, Poland;
KEE-WAI LAU, Hong Kong; HENRY J. RICARDO, Medgar Evers College, Brooklyn,
N.Y.; and the proposer. Two other readers sent in incomplete solutions.
Most solvers noted that the inequality is not sharp. Richard I. Hess, Rancho Polos
Verdes, California, and Kuczma separately calculated that the minimum value of
4 1 1
2 + 7 rrr +
c (a-6)6 (6-c)c
under the given conditions is about 1.5434, occurring at the uninteresting values a w 4.724,
6 « 3.227, c « 2.09.
 j EDITORIAL BOARD / CONSEIL DE REDACTION J
Dr. G.W. Sands, Editor-in-Chief Department of Mathematics and Statistics, University of Calgary,
1 Calgary, Alberta T2N 1N4
Dr. R.E. Woodrow, Editor-in-Chief Department of Mathematics and Statistics, University of Calgary,
[ Calgary, Alberta T2N 1N4 |

Dr. G.P. Wright, Managing Editor Department of Mathematics, University of Ottawa,

Ottawa, Ontario KIN 6N5 |

Dr. R. DePeiza East York Board of Education, 840 Coxwell Avenue,

East York, Ontario M4C 2V3 |

Dr. R. Guy Department of Mathematics and Statistics, University of Calgary,

Calgary, Alberta T2N 1N4 |

Dr. C. Fisher Department of Mathematics, University of Regina,

Regina, Saskatchewan S4S 0A2 |

Dr. D. Hanson (Articles) Department of Mathematics, University of Regina,

Regina, Saskatchewan S4S 0A2

Dr. A. Liu (Book Reviews) Department of Mathematics, University of Alberta,

Edmonton, Alberta T6G 2G1 |

Dr. R. Nowakowski Department of Math., Stats and Computing Science, Dalhousie

University, Halifax, Nova Scotia B3H 3J5 |

Dr. J. Totten Department of Math. & Stat., Cariboo College, Kamloops,

British Columbia V2C 5N3 |

Dr. E. Wang Department of Mathematics, Wilfrid Laurier University

Waterloo, Ontario N2L 3C5

Short articles intended for publication should be sent to Dr. Hanson, contest problem sets and solutions to Olympiad Corner
problems should be sent to Dr. Woodrow and other problems and solutions to Dr. Sands.

BACK ISSUES AND SUBSCRIPTION INFORMATION

INFORMATION SUR L'ABONNEMENT ET LES NUMEROS PRECEDENTS

1994 (Volume 20) SUBSCRIPTIONS / ABONNEMENTS 1994 (volume 20)

(10 issues per volume / 10 num6ros par volume)
Regular / R<§gulier $46.00
CMS Members / membres de la SMC $23.00

These subscription rates include shipping and handling

Les frais de manutentions sont inclus dans ces tarifs
CANAD/AN ADDRESSES ADD 7% GST/ ADRESSES CANADIENNES, AJOUTER 7% DE TPSJ

BACK ISSUES / NUMEROS ANTERIEURS

Volume: $50.00 Issue / Num6ro: $ 6.00

BOUND VOLUMES / VOLUMES RELIES

(available / disponible: 1&2, 3, 1, 8, 9): $10.00 each/chacun

FOR BACK ISSUES AND BOUND VOLUMES ONLY

POUR NUMEROS ANTERIEURS ET VOLUMES RELIES SEULEMENT
Canadian addresses, please add 7% GST (#R118 833 979).
Les adresses canadiennes, veuillez ajouter 7% de TPS (#R118 833 979).
 CANADIAN MATHEMATICAL SOCIETY
SOCIETE MATHEMATIQUE DU CANADA
577 King Edward
P.O. Box 450, Station A
Ottawa, Ontario K1N 6N5 CANADA
Tel: (613) 564-2223, Fax: (613) 565-1539

ORDER FORM / BON DE COMMANDE

CRUX MATHEMATICORUM

Subscription / Abonnement __ X $46.00 (1)

Back issues / Numgros antgrieurs _ X $ 6.00 = (2)

_ x $50.00 = (3)

Bound volumes / Volumes relics . __ x $10.00 = (4)

, ( 1 + 2 + 3 + 4)

Canadian addresses must add GST: 7% of total of lines 1 to 4 (# R118 833 979)
Les adresses canadiennes: veuillez ajouter 7% de TPS sur la somme
des lignes 1 a 4 (# R118 833 979)

TOTAL $

• Cheque enclosed / cheque inclus • Invoice / Facture

(micro-encoded cheques only
cheques encodes seulement)

• Please charge / Portez a mon compte • VISA • MasterCard

Account no. / Numero de compte Expiry date / date d'expiration

Signature

MAILING ADDRESS /ADRESSE POSTALE

Name / Nom:

Address / Adresse:

City, Province / Ville, Province:

Postal Code, Country / Code postal, pays: