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 I fevrier
Volume 18 #2 1992

CONTENTS I TABLE PES MATIERES

The Olympiad Corner: No. 132 R.E. Woodrow 33

Problems: 1711-1720 43

Solutions: 1508, 1531, 1601-1612 45

Canadian Mathematical Society 1^^ Soci&e' mathgmatique du Canada

& _
 Founding Editors: Leopold Sauve, Frederick G.B. Maskell
Editors-in-Chief: G.W. Sands and R.E. Woodrow
Managing Editor: G.P. Wright
EDITORIAL BOARD
G.W. Sands (Calgary)
R.E. Woodrow (Calgary)
G.P. Wright (Ottawa)
R. Guy (Calgary)
C. Fisher (Regina)
D. Hanson (Regina)
A. Liu (Alberta)
R. Nowakowski (Dalhousie)

GENERAL INFORMATION

Crux Mathematicorum is a problem-solving journal at the senior secondary and university under-
graduate levels for those who practice or teach mathematics. Its purpose is primarily educational
but it serves also those who read it for professional, cultural or recreational reasons.

Problem proposals, solutions and short notes intended for publications should be sent to the Editors-
in-Chief:

G.W. Sands and R.E. Woodrow

Department of Mathematics & Statistics
University of Calgary
Calgary, Alberta, Canada, T2N 1N4

SUBSCRIPTION INFORMATION

Crux is published monthly (except July and August). The 1991 subscription rate for ten issues is
$ 20.00 for members of the Canadian Mathematical Society and $40.00, for non-members. Back
issues: $4.00 each. Bound Volumes with index: volumes 1 & 2 (combined) and each of 3, 7, 8 &
9: $10.00 (Volumes 4, 5, 6 Sz 10 are out-of-print). All prices are in Canadian dollars. Cheques and
money orders, payable to the CANADIAN MATHEMATICAL SOCIETY, should be sent to the
Managing Editor:

Graham P. Wright
Canadian Mathematical Society
577 King Edward
Ottawa, Ontario, Canada KIN 6N5

ACKNOWLEDGEMENTS

The support of the Department of Mathematics and Statistics of the University of Calgary and of
the Department of Mathematics of the University of Ottawa is gratefully acknowledged.

© Canadian Mathematical Society, 1992

Published by the Canadian Mathematical Society
Printed at Ottawa Laser Copy

ISSN 0705-0348 Publications Mail Registration Number 5432

 33

T H E OLYMPIAD C O R N E R
No. 132
R.E. W O O B R O W

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>
This month we give a selection of problems used to select and train I.M.O. teams
in three countries. The first problems were some used in the competition to determine the
Irish I.M.O. team for 1990. Thanks go to Andy Liu, University of Alberta, for forwarding
these problems to me.

S E L E C T I O N Q U E S T I O N S F O R T H E 1990 I R I S H I . M . O . T E A M
1 . Find all pairs of integers (x,y) such that ys — x3 = 91.
2* Observe that, when the first digit of x = 714285 is moved to the end, we get
y = 142857 and y = x/5. Find the smallest positive integer u such that if v is obtained
from u by moving the first digit of u to the end, then v = u/2.
3 . Let 2 , 3 , 5 , 6 , 7 , 8 , 1 0 , 1 1 , 1 2 , 1 3 , 1 4 , 1 5 , 1 7 , • • • be the sequence of non-squares (i.e.,
the sequence obtained from the natural numbers by deleting 1 = I 2 , 4 = 2 2 , 9 = 3 2 ,
16 = 4 2 , etc.). Prove that the nth term of the sequence is
1 '
n n
+ o+ v
(Note that, for x a real number, [x] denotes the greatest integer z with z < x. Thus, for
example, [16/7] = 2.)
4 . Let n > 3 be a natural number. Prove that
1 1 1 1
33 43 n 3
12

5 . Let t be a real number and let

an = 2 cos

(n = 1 , 2 , 3 , . . . ) . Let 61 be the product at.. .an. Find a formula for bn which does not
involve a product of n terms and deduce that
2cost + l
lim bn =
n-*oo

The next set of three problems are from the First Test of the Third Chinese National
Mathematics Training Camp Examination of April 1988. Thanks go to Edward T.H. Wang ?
Wilfrid Laurier University, Waterloo, Ontario, who translated and sent them to me.

T V ^ Printed on recycled paper

Imprime sur papier recycle
34

1988 C H I N E S E O L Y M P I A D T R A I N I N G C A M P
Test I

1 . Prove that
xyz(x + y + z + \/x2 + y2 + z2) 3 + \/3
2 2 2
(x + y + z )(xy + yz + zx) ~~ 9
for all positive real numbers x, y and z with equality holding if and only if x = y = z.
2* Determine the smallest value of the natural number n > 3 with the property
that whenever the set Sn = {3,4,... ,n} is partitioned into the union of two subsets, at
least one of the subsets contains three numbers a, b and c (not necessarily distinct) such
that ah = c. [Compare with Problem 11 by Morocco, [1988: 226] — E.T.H.W.].
3 . A pharmacist has a number of ingredients some of which are "strong". Using
these ingredients he is to make 68 different medicines such that each medicine contains 5
different ingredients, at least one of which is "strong" and furthermore, for any 3 ingredients
chosen, there is exactly one medicine containing them. Prove or disprove that one of the
68 medicines must contain at least 4 "strong" ingredients.
*

The last test questions for this number are the two problems given the last day of
the training camp of the then Soviet Union held in January 1991 in preparation for the
contest in Sweden. These were forwarded from Zun Shan to Andy Liu who translated
them from Chinese for use in the Corner.

T R A I N I N G T E S T O F T H E 1991 U.S.S.R. I.M.O. T E A M

Last Day — 4 Hours

1 . Let a\ = 1 and

2 4an
ls a
Prove that J2J-\ positive integer for n > 1.
2 . Let n be a positive integer and Sn be the set of all permutations of { 1 , 2 , . . . , n}.
For aeSn let f(a) = £?=i K - <K0I- P r o v e t h a t

* * *

Now we turn to answers for Klamkin's Five Quickies given last issue. Thanks to
Murray S. Klamkin, University of Alberta, for the problems and their solutions.
 35

FIVE KLAMKIN QUICKIES

1 . Determine the extreme values of ri/hi + r2/h2 + r3/h3 + r4/h4 where /i l5 /i 2 , h3j
h4 are the four altitudes of a given tetrahedron T and ri, r 2 , r35 r4 are the corresponding
signed perpendicular distances from any point in the space of T to the faces.
Solution. If the face areas and volume of the tetrahedron are F\, F2j F3, F4, and V
respectively, then
riFi + r2F2 + r3F3 + r4F4 = 3 V,
and hiFi = ^2^2 = ^3-^3 = h4F4 = ZV. Now eliminating the J^'s, we get
^1/^1 + r2jh2 + r3/h3 + r4/h4 = 1 (a constant).

2* Determine the minimum value of the product

P = (1 + Xl + yt)(l + X2 + y2)...(l + xn + yn)

where a?,-, j/* > 0, and x ^ • • • xn = S/1J/2 • • • yn = aH -

Solution. More generally, consider
P = (1 + a-i + j / ! + • • • + wi)(l + x2 + y2 + h w2)... (1 + xn + Vn + • • • + wn)

where x\X2 ... xn = £ n , 2/13/2 • • - Vn = *T? • • • 5 ^1^2 -. - ^n = ^ n ? and a;,-, y»,..., u;,- > 0. Then
by Holder's inequality,

p i / n > { 1 + n * J / B + n y? /B +• • • + n «.•/B}
or
P > ( l + £ + i7 + ...+u>) n -
In this case £ = 7) = a, so
P > (l + 2a) n .

3® Prove that if F(x^y^z) is a concave function of x, t/, z, then {F(x,y,z)}~~2 is a

convex function of x, t/, z.
Solution. More generally G(P) is a convex function where G is a convex decreasing
function. By convexity of (7,
\G{F(xuyuz1)} + (1 - X)G{F(x2,y2jz2)} > G{XF(x1,yljz1) + (1 - A)F(z 2 ,y 2 ,s 3 )}.

By concavity of F ,

\F(xuyuzl) + (l-\)F(x2,y2,z2)<F([\x^
Finally, since G is decreasing,

XGiFfay^Z!)} + (1 - A)G{F(x a ,y 2 ,«i)} >

^ ( [ A ^ + (1 - A)z2], [At/: + (1 - A)j/2], [Azx + (1 - X)z2})}.
36

More generally and more precisely, we have the following known result: if F{X) is
a concave function of X = (£i,£2> • • • ? #n) &nd G(y) is a convex decreasing function of y
where y is a real variable and the domain of G contains the range of i*1, then G{F(X)} is
a convex function of X.
4 . If a, 6, c are sides of a given triangle of perimeter p, determine the maximum
values of
(i) ( a - 6 ) 2 + ( 6 - c ) 2 + ( c - a ) 2 ,
(ii) | a - & | + | 6 - c | + | c - a | ,
(iii) |a - &||& - c| + |& - c\\c — a| + |c — a||a — 6|.
Solution, (i) (a - 6)2 + (6 - c) 2 + (c - a) 2 = 2 ( £ a2 - £ 6c) < fcp2-
Let c = 0, so that & > 1/2. We now show that k = 1/2 suffices. Here,

2 Q > 2 - £ M < ^ ( « + & + c)2

reduces to
26c + 2ca + 2a6 - a2 - 62 - c2 > 0.

The LHS is 16 times the square of the area of a triangle of sides y^a, \fb, \/C or

(y/a + Vb + y/c){y/a + Vb - V^)(v^ - Vb + y/c)(-y/a + Vb + y/c).

There is equality iff the triangle is degenerate with one side 0.
(ii) \a - 6| + |6 - c\ + \c - a\ < kp.
Letting c = 0, k > 1. To show that k = 1 suffices, assume that a > 6 > c, so that

\a - 6| + |6 - c\ + \c - a\ = 2a - 2c < a + 6 + c

and there is equality iff c = 0.

(iii) \a - 6| |6 - c\ + \b - c| |c — a| + |c — a| |a - 6| < kp2.
Letting c = 0, k > 1/4. To show that k = 1/4 suffices, let a = y + z, b = z + x, c = x + y
where z > y > x > 0. Our inequality then becomes

k ~ y\ \z - y\ + \y - A \z - ^1 + \z - ^1 k - y\ ^ fa + 2/ + ^)2
or
# 2 — y2 + £2 + J/£ — 3zx + xy < x2 + y2 + z2 + 2yz + 2zx + 2xy
or
2j/2 + 5z# + Ixy + lyz > 0.
There is equality iff x = y = 0 or equivalently, a = 6, and c = 0.
5 . If A, I?, C are three dihedral angles of a trihedral angle, show that sin A, sin £?,
sin C satisfy the triangle inequality.
 37

Solution, Let a, 6, c be the face angles of the trihedral angle opposite to A, J3, C
respectively. Since
sin a _ sin 6 _ sine
sin A sin B sin C
by the Law of Sines for spherical triangles, it suffices to show that sin 6 + sine > sin a, or
. 1 1 1 1
2 sin™ (6 + c)cos~(6 — c) > 2 sin -a cos - a ,
for any labelling of the angles. We now use the following properties of a, 6, c:
(i) they satisfy the triangle inequality, (ii) 0 < a + 6 + c<27r.
Hence, cos \{b — c) > cos | a . To complete the proof, we show that

• 1/L X • !

sin-(6 + c) > s i n - a .

This follows immediately if6 + c < 7 r ; i f 6 + c>7r, then h + c aN

sin-(6 + c) = sin i TT — ~(6 + c) }• > sin~a 1 since IT — — > -
sin — (6 + c) = sin \ TT — -(6 + c) > > sin - a I since ic

Comment: More generally, if ai, a 2 , . . . , a n are the sides of a spherical n-gon (con-
vex), it then follows by induction over n that

sinai + sina 2 H + sina n > 2sina l ? i = 1,2,... ,ra.

It also follows by induction that

| sin ail + I sina 2 l + h |sina n | > |sin(ai + a2 + • • • + a n )|

for any angles ai, a 2 , . . . , an.

Now we turn to solutions to problems from the October 1990 number of Crux, and
the 25th Spanish Mathematics Olympiad, First Round, 1988.
I* [1990: 225] 25th Spanish Mathematics Olympiad, 1988.
Let n be an even number which is divisible by a prime bigger than y/n. Show that
n and n 3 cannot be expressed in the form 1 + (2/ + 1)(2/ + 3), i.e., as one more than the
product of two consecutive odd numbers, but that n2 and n 4 can be so expressed.
Solutions by Seung-Jin Bang, Seoul, Republic of Korea; 0. Johnson, student, King
Edward School, Birmingham, England; Stewart Metchette, Culver City, California; Bob
Prielipp, University of Wisconsin-Oshkosh; and by Edward T.H. Wang, Wilfrid Laurier
University, Waterloo, Ontario. [We give Wang's solution.]
Call a natural number 'expressible5 if it is of the form 1 + (21 + 1)(2I + 3) for some
/. If n is even, then n 2 = 1 + (n - l)(n + 1) = 1 + (21 + 1)(2? + 3) where 1 = (n - 2)/2,
showing that n2 is expressible. Since n2 is even when n is even, n4 is also expressible. On
38

the other hand suppose n were expressible. Then n = 4/2 + 8/ + 4 = 4(7 + l) 2 showing that
the largest prime divisor of n can not exceed max{! + 1,2} which is a contradiction since
%Jri = 2(1 + 1). Now suppose n 3 were expressible. Then n 3 = (2(/ + I)) 2 , implying that n3
is a perfect 6th power. Hence the largest prime divisor of n3 (and thus the largest prime
divisor of n) can not exceed (n 3 ) 1 / 6 or ^/n, which is again a contradiction.
[Editor's Note. Both Johnson and Metchette point out the generalization to other
powers of n.]
3 . [1990: 225] 25th Spanish Mathematics Olympiad, 1988.
The natural numbers 1,2,..., n2 are arranged to form a n n x n matrix

1 2 ... n
n + 1 n + 2 . . . 2ra
A =

n2

A sequence ai, <Z2,... of elements of A is chosen as follows. The first element a\ is chosen
at random and the row and column containing it are deleted. As long as elements remain,
the next element is chosen at random from among the elements that remain, and its row
and column are deleted. The process continues until only one element is left. Calculate
the sum of this last number and all the numbers previously chosen. Show that this sum is
independent of the choices made.
Solutions by 0. Johnson, student, King Edward School, Birmingham, England;
and by Edward T.H. Wang, Wilfrid Laurier University, and Wan-Di Wei, University of
Waterloo, Waterloo, Ontario.
The method of selection ensures that we select exactly one element from each
row and exactly one element from each column. Since the (i,j) entry of the matrix is
(i — l)n + j for i, j = 1,2,..., n, the sum S of the n numbers must be YA=I(^ — l ) n + a%
where ai, a 2 , . . . , an is a permutation of 1,2,... 9 n. Thus

<? „ f « U + TT4 " ( " - 1)* +, n(n + 1) = n(n 2 + l)

6 = n 2^{i - 1) + 2 J 3 = o 9 9 '
1=1 j=l L L L

5 . [1990: 226] 25th Spanish Mathematics Olympiad, 1988.

Let ABCD be a square and let E be a point inside the square such that AECD is
isosceles with IC = ID = 15°. What can one say about A ABE?
Solution by Seung-Jin Bang, Seoul, Republic of Korea.
We may assume that ABCD is a unit square. Let F and G be the points of
intersection with AB and CD respectively, of the line through E perpendicular to those
sides. The length of the segment EG is \ tan 15° = (2 - \/3)/2. It follows that the length
of the segment EF is y/Z/2. It turns out that tan LEAF = v^3, that is LEAF = lEBF =
60°. Thus A ABE is equilateral.
 39

7. [1990: 226] 25th Spanish Mathematics Olympiad, 1988.

Find the maximum value of the function

f(X)=n i* - *i
jfe=0

for x in the closed interval [3,4].

Solutions by Seung-Jin Bang, Seoul, Republic of Korea; and by Edward T.H. Wang,
Wilfrid Laurier University, and Wan-Di Wei, University of Waterloo, Waterloo, Ontario.
[We give the solution of Wang and Wei.J
Clearly f(x) = x(x-l)(x-2)(x-3)(4-x)(b-x)(6-x)(7-x). By the arithmetic-
geometric-mean inequality we have

x(7-x)<Q\ ( a ; -i)(6-x)<(|)2,

(*-2)(5-*)<(!)', (x-3)(4-x)<Q\

with equality holding in any (and hence all) of these inequalities if and only if x = 7/2.
Thus the maximum value of f(x) is 3 2 • 5 2 • 72 • 2""8 or 11025/256 attained uniquely when
x = 7/2.
In exactly the same way one can show that if a is a natural number then for x in
[a, a + 1] the maximum value of f(x) = OibLo1 \x "" ^1 *s ((%a + l)!!/2° +1 ) 2 attained uniquely
at x = (2a + l)/2.
8 . [1990: 226] 25th Spanish Mathematics Olympiad, 1988.
Let m be odd. Show that for each integer n > 2, the sum of the mth powers of the
numbers less than n that are relatively prime to n is a multiple of n.
Solutions by Bob Prielipp, University of Wisconsin-Oshkosh; and by Edward T.H.
Wangf Wilfrid Laurier University, Waterloo, Ontario.
Let S = {k : 1 < k < n such that (k9n) = 1}. Then clearly k £ S if£ n — k £ S,
Furthermore, for k G 5, k ^ n — k. Otherwise n = 2k would imply that (n, fc) = Jfc, a
contradiction since n > 2 implies k = 1. Hence the elements of S can be grouped into
disjoint pairs {k,n — k}. The result now follows from the fact that a + b is a factor of
am + bm when m is odd.
We have no solutions on file for the other even numbered exercises from the com-
petition. Here is an opportunity, and a challenge to provide nice solutions!
*

The remaining solutions that we give this month are to problems from the November
1990 number of the Corner. We first give solutions to some of the problems from the 1990
Asian Pacific Mathematical Olympiad. Here we give only solutions that differ from the
official solutions published in the solutions manual for that contest.
40

2 . [1990: 257] 1990 Asian Pacific Mathematical Olympiad.

Let a i , a 2 , . . . , a n be positive real numbers, and let Sk be the sum of products of
a i , a 2 , . . . , an taken k at a time. Show that

SA.k > ( j :ai«2 . . . a , n for fc = 1 , 2 , . . . , n — 1.

Solution by George Evagelopoulos, Athens, Greece.

There are m products of the at- taken A: at a time. Any given a?- will appear in
exactly u l j ) of these products, since once a,- is chosen there are u l j ) ways of choosing the
other factors of the product. Now the arithmetic and geometric mean of the m products
are related by
Sk */(:)
>
ft) L i=l

by the arithmetic-geometric means inequality. Since MM = j u _ J) ?

n \ V71
& >

Hence
*-k
k/n 2 / n N
n ' n ^
SkSn-k >
,n — ki »=i ; n«.-
*=i

as desired.
5 . [1990: 258] 1550 Asian Pacific Mathematical Olympiad.
Show that for every integer n > 6, there exists a convex hexagon which can be
dissected into exactly n congruent triangles.
Solution by George Evagelopoulos, Athens, Greece.
Consider an isosceles triangle S with IB = IC and IA < 60°. One can form a
parallelogram T by juxtaposing AC and BA as shown.
A AB C

We can align n copies of T to form a parallelogram using 2n triangles congruent to S:

 41

A C

B C
AZ__A/ A
A hexagon can now be formed in two ways: (i) add one copy of S at each end, identifying
AC and AB in each case;
A CB

CB A
or (ii) add one copy of S at the left as in (i) and two copies of S at the right, as shown.
A CAA

CB AB CB
In both cases, IA < 60° ensures that all six angles of the hexagon are less than 180° so
each hexagon is convex. Using either (i) or (ii) one can build convex hexagons that can be
decomposed into exactly n congruent triangles for n = 4 , 5 , 6 , . . . .
[Editor's remark: the official solution obtains right triangles.]
*

The last two solutions are to the first two problems of the Fourth Nordic Mathe-
matical Olympiad held April 5, 1990.
1 . [1990: 258] Fourth Nordic Mathematical Olympiad.
Let m, n and p be positive odd integers. Show that the number
(n-l)P
km
E
is divisible by n.
Solutions by Seung-Jin Bang, Seoul, Republic of Korea; and by Edward T.H. Wang,
Wilfrid Laurier University, Waterloo, Ontario.
Since x + y is a factor of xm + ym for all odd positive integers m, we have for each
k = 1,2,..., (n-l)*72 that fcm + [ ( n - l ) * - ( f c - l ) ] m is divisible byfc+ [ ( n - l ) P - ( f c - l ) ] =
(n - l)p + 1. Now since (n - l)p + 1 = n? - fyn*-1 + • • • + (/^n — 1 + 1 = 0 mod n, we
conclude that
(n-iy (n-l)P/2
k =
J2 ™ E (*m + l(n - lY - (* - l)]m) = ° mod n-
42

2» [1990: 258] Fourth Nordic Mathematical Olympiad.

Let oj, c&2,..., an be real numbers. Show that

*la{ + a 3 + . . . + a 3 < ^/ a 2 + a 2 + . . . + a 2 .

When does equality hold?

Solution(s) by David Vaughan and Edward T.H. Wang, Wilfrid Laurier University,
Waterloo, Ontario.
Since the LHS of the given inequality does not decrease if any at- is replaced by |a t |
we may assume, without loss of generality, that af- > 0 for alH = 1,2,..., n.
We give two solutions, the first of which establishes a more general result.
Solution I. The result is a special case of the following:
Theorem. If at- > 0 for i = 1,2,..., n and 0 < p < q then

(al + al + -.- + al)1'i<(ap1 + ap2 + --- + aZ)1/'>

with equality holding if and only if either p = q or at most one a,- ^ 0.

Proof. The claimed inequality is equivalent to

a! + --- + a « < K + «5 + --- + <)' /p (1)

which we shall prove by induction on n. The case n = 1 is trivial. For n = 2, we need to
show that
al+al<(al + aW*. (2)
If ai = 0, there is nothing to prove. Assume ax ^ 0. Then (2) is equivalent to

Consider /(a?) = (1 + a?*)'/* - (1 + x«) for x > 0. Then

/ ' ( s ) = ^ (1 + a ^ ' / p ) " 1 • px*-1 - qxq~l = 9Z P_1 ((1 + a ^ ' / p ) " 1 - a*-") > 0
P
since (1 + X P ^ / P ) " 1 > ( X P ^ / P ) " 1 = X«~P. Since /(0) = 0, we conclude that f(x) > 0 and
thus (2) follows. It is easily checked that (2) is strict if both 01,02 > 0.
Now suppose that (1) holds for some n > 2. Then using the induction hypothesis
and (2) we have, with y = (a^-\ h <)1/p,

= y9 + 4+i < (yp + <+i)9/p = (a? + ••• + < + < +1 ) 9/p ,

completing the induction. The last inequality is easily seen to be strict when at least two
a t 's are nonzero by rearranging the terms, if necessary, so that an+\ is one of these nonzero
 43

terms. For the proposed problem, this means that equality holds iff either p = q or at
most one a,- ^ 0. •
[Editor's Note, This result is known as cc Jensen ? s Theorem 5 '. See eq. 1.4.1, p. 4 of Hardy,
Littlewood, and Polya, Inequalities.]
Solution II. We use the Cauchy-Schwarz Inequality
M i + a2b2 + • • • + anbnf < (a? + a\ + • • • + a2n){b\ + b\ + • • • + 6?).
Setting hi = a?, then we have

(<*! + «*! + ••• + a n ) 2 < (a? + ^2 + • • • + «»)(«i + aj + • • • + aj)

<(a? + a» + ... + aJ)3 (3)
since clearly af + a\ + • • * + a\ < (a\ + a | + • • • + a^) 2 . Since equality holds in (3) if and
only if either n = 1 or a f - ^ 0 for at most one i, we see that, for the proposed problem,
equality holds iff either af* = 0 for all i or n = 1 and at > 0.
* * *

That's the Corner for this month. Contest season is upon us. Send me your regional
contests and your nice solutions.

sfc sfs sfc % sfe

PROBLEMS
Problem proposals and solutions should be sent to B. Sands, Department of
Mathematics and Statistics, University of Calgary, Calgary, Alberta, Canada T2N 1N^.
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,
1992, although solutions received after that date will also be considered until the time when
a solution is published.

1 7 1 1 * . Combination of independent proposals by Berta T. Freitag, Roanoke,

Virginia, and by Walther Janous, Ursulinengymnasium, Innsbruck, Austria.
Let 0 < r < 1, let gi, #2> • - • 9 #& be fixed positive numbers, and define

= X
%n+k+l 2^ n+i

for n > 0. Show that the sequence {xn} converges and determine its limit.
44

1712. Proposed by Murray S. Klamkin, University of Alberta.

Determine the minimum value of
lesin^A^sin^Jg^sin^C^) + 1
tan(A/2) tan( J3/2) tan(C/2)
where A, J3, C are the angles of a triangle.
1713. Proposed by Jeremy Bern, student, Ithaca H.S., Ithaca, N.Y.
For a fixed positive integer n? let K be the area of the region

in the complex plane. Prove that K > 7r(lln 2 + 1)/12.

1714. Proposed by Toshio Seimiya, Kawasaki, Japan.
Let P and Q be two points lying in the interior of IB AC of A ABC, such that the
line PQ is the perpendicular bisector of BC, and such that IABP + 1ACQ = 180°. Prove
that lBAP = lCAQ.
1715. Proposed by Seung-Jin Bang, Seoul, Republic of Korea.
Evaluate the sum

t0kl\2l 3! + +l ij
(n-k)\)
for n > 2.
1716. Proposed by Jordi Dou, Barcelona, Spain.
Equilateral triangles A B C , B'CA, CAB are erected outward on the sides of triangle
ABC. Let 0 be the circumcircle of A'B'C* and let A", £?", Cn be the other intersections
of 0 with the lines A1 A, B'B, C'C, respectively. Prove that AA" + BB" + CCn = AA!. [It
is known that AA^BB1 and CC are concurrent; e.g., see [1991: 308].]
1717. Proposed by Edward T.H. Wang, Wilfrid Laurier University, Waterloo,
Ontario.
For each positive integer n, let f(n) denote the number of ordered pairs (x,y) of
nonnegative integers such that n = x2 — y2. For example, /(9) = 2 as 9 = 3 2 — 02 = 52 — 42
are the only representations. Find a formula for f(n).
1718. Proposed by Juan Bosco Romero Mdrquez, Universidad de Valladolid,
Spain.
Let C be a central conic with foci Fi and F 2 , and let X and Y be the points where
the tangent to C at a point P on C meets the axes (extended) of C. Prove that

PX-PY = PF1 • PF2.

 45

1 7 1 9 * . Proposed by B. Lindros, Le Colisee, Quebec.

Let k and / be two positive real numbers less than 1, and consider a square of side
1 whose sides are horizontal and vertical. A horizontal line of length k is drawn at random
inside the square (the line cannot stick outside the square). Independently, a vertical line
of length I is drawn at random inside the square. What is the probability that the two
lines intersect?
1 7 2 0 . Proposed by P. Penning, Delft, The Netherlands.
The osculating circle at point P (not a vertex) of a conic intersects the conic in one
other point Q. Find a simple construction for Q, given the conic itself, its axes and the
tangent at P .

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 5 0 8 . [1990: 20; 1991: 89] Proposed by Edward T.H. Wang, Wilfrid Laurier
University, Waterloo, Ontario.
Let a < b < c be the lengths of the sides of a right triangle. Find the largest
constant K such that

a2(b + c) + b2(c + a) + c2(a + b) > Kabc

holds for all right triangles and determine when equality holds. It is known that the
inequality holds when K = 6 (problem 351 of the College Math. Journal; solution on p.
259 of Volume 20, 1989).
II. Generalization by Murray S. Klamkin, University of Alberta.
Instead of right triangles, we consider all triangles of fixed largest angle C = 20 and
determine the maximum K. Here 30° < 0 < 90°, and without loss of generality we can
take c = 1. Then
T. . lJ(b+l) + P(l + a) + a + b\ . ( ^ ^ l 1 a 6N
p
max K = mm ( —L—— ^
ab
•jL———- = mm a + ^H
J I
I-T+T + ~
a b b a ,
subject to
1 =a2 + b2-2abcos20. (1)
We will show that the minimum is taken on when a == 6. Clearly a/b + b/a is
minimized when a = b. Since

a + b+- + \ = (a + b)(l + l-),

a b \ abj
it now suffices to show that
^ = <« + »),(1 + 3 )
46

takes on its minimum for a = b. Letting

— A — a^ _ s * n A sin B _ cos(P — A) — cos(B + A)

c2 sin 2 C 2sin2<7
we find that
1 + cosC 1
2
0 < x < 2sin C 2(1 -cos20) '
where the upper bound occurs for A = B, i.e., for a = 6. Also,

l< l <i.
4 - 2(1 - c o s 2 0 ) ~
It follows from (1) that

(a + b)2 = 1 + 2a6(l + cos 20) = 1 + 4x cos2 0.

Hence
P = ( 1 + 4z cos2 0) (1 + -) = 1 + - + - 12 + 4 fx- + -) cos2 0 + 8 cos2 0.
\ xj x x \ xj
Since # < 1, P is minimized when # is a maximum, that is, when a = 6.
Thus from (1),
1 = 2a2 - 2a2 cos 20 = 4a2 sin2 0
or a = 1/(2sin 0), so

max.: # = 2a + -2a
i - + 2 = 2 + 4sin0 + - r ^ = ( 2 > / s i n 0 - - 7 ! = ) + 6 .
sm0 \ Vsin0/
This means that maxl^ is an increasing function of 0 in the interval 30° < 0 < 90°. Thus
from 1/2 < sin 0 < 1,

min maxi^ = 6 and max max/{ = 7.

3O°<0<9O° 3Q°<0<9O°

Thus for all angles C between 60° and 180° there is only a spread of 1 in max/{.

1 5 3 1 . [1990: 108; 1991: 156] Proposed by J.T. Groenman, Arnhem, The Nether-
lands.
Prove that
v+w be w+u ca u+v ab ^ A, , x
+ _ Z 7+ — > 4 ( a + 6 + c),
u s—a v s—b w s—c
where a,6,c,s are the sides and semiperimeter of a triangle, and u,v,w are positive real
numbers. (Compare with Crux 1212 [1988: 115].)
 47

IV. Comment by Murray S. Klamkin, University of Alberta.

At the end of his solution [1991: 160], Walther Janous leaves open the problem of
proving or disproving that

)/<*(* - a) + 1/6(5 - b) + yjc(s - c) > 2sJr/R . (1)

We prove a generalization of (1) using results in my solution [1991: 157]. We have that for

[tf-^a* + ¥ + <*)]lt* >a+b+c> ^/uF^S , (2)

where F is the area of the triangle. Now if a, 6, c are sides of a triangle, then so are
Ja(s — a), Jb(s — 6), Jc{s — c), and the area of this triangle is F/2 [see VII.5.2, page 113
of D.S. Mitrinovic, J.E. Pecaric, and V. VoJenec, Recent Advances in Geometric Inequali-
ties]. Hence a dual inequality to (2) is

[&-\[a(3 - a)Y'2 + [b(s - b)]*'2 + [c(s - c)]*'2)]1'* > JtFyfi . (3)

Now
\/6Fyfi > 2s^f7[R ,
as this is equivalent (via F = rs) to 3>/3-R > 2s, a known inequality corresponding to the
fact that the maximum perimeter triangle inscribed in a given circle is the equilateral one.
Consequently (3) becomes

[«(* - a)f/ 2 + [b(s - b)}^ + [c(s - c)K'/2 > 3 ( f y ^ ) ' (4)

Inequality (1) is just the special case of (4) when p

* * *

1 6 0 1 . [1991: 13] Proposed by Toshio Seimii,a, Kawasaki, Japan.

ABC is a right-angled triangle with the right angle at A. Let D be the foot of the
perpendicular from A to BC, and let E and F he the intersections of the bisector of IB
with AD and AC respectively. Prove that "DC > 2EF.
I. Solution by the Con Amort Problem Group, Royal Danish School of Educational
Studies, Copenhagen.
From the figure (where M is the midpoint of AC),
IFEA = LDEB = IAFB **
and thus
EF/2 _ AF .D
?
AF ~~ BF
i.e. (E
4(AF)2
2EF =
BF
Also
F M
48

DC AC (Acy
so DC =
AC ~ BC ' BC
Hence (since BF is the bisector of IB)
DC (AC)2 • BF (AC)2 BA _ (AC)2 AF
>
2EF ~ BC • 4(AF)2 4(AF)2 ' BC ~ 4(AF)2 ' FC
(AM)2 _ (AM)2 _ (AM)2
AF • FC (AM - x)(AM + x) (AM)2 - x*
>1,
so DC > 2EF.
II. Solution by Charles H. Jepsen, Grinnell College, Grinnell, Iowa.
In the figure, APQ 1 BF, RQS\\BF and
ST 1 BC. Since BF bisects IB, AP = PQ,
thus RS = 2EF. We finish the proof by showing
DC > RS. From

area (ADC) > area (ADTS) = area (ARS)

we have

\ DC - AD >\ RS • AQ >\ RS • AD,

L u JL

and so Z)C > RS.

III. Solution by Giannis G. Kalogerakis, Canea, Crete, Greece
We know that DC/AD = AD/BD, i.e.,

Also, the triangles A 5 F and 5£?JD are similar,

hence BF/BE = AB/BD. Thus we find

BE-AB-BE- BD
EF = BF-BE =
BD
We therefore have to show that

(AD)2 2BE(AB - BD)

BD BD
or
(AB)2 - (BD)2 > 2BE(AB - BD),
which simplifies to AB + BD > 2BE. But if BG is the median (G the midpoint of AD),
then
AB + BD > 2BG > 2BE.
 49

IV. Generalization by Shailesh Shirali, Rishi Valley School, India.

Actually we need only the fact that I BAG — lADB; the right angle is not really
necessary.
To start with, note that

(BF)2 < BA • BC]

for if K is defined to be the point where the

ray BF meets the circumcircle of A ABC,
then triangles ABF and KBC are similar,
so AB/BF = KBjBC and the stated result follows from BF < BK.
Now draw FP\\AD with P on BC; P is symmetric to A with respect to BF. From
the similarity of triangles ABF and DBE, we find that AAEF is isosceles, so AE = AF.
Also AF = FP by symmetry, so AE is equal and parallel to FP and therefore AEPF is a
rhombus, with PE\\CA. Extend PE to meet AB at Q. From the various pairs of similar
triangles in the figure we have
EF BF EF BF
-QA=m> PC = BC> DP = QA (by symmetry),

and so

or EF < {DC) 12.

Also solved by MIGUEL AMENGUAL COVAS, Cala Figuera, Mallorca, Spain;
SEFKET ARSLANAGIC, Trebinje, Yugoslavia (three solutions); SAM BAETHGE,
Science Academy, Austin, Texas; SEUNG-JIN BANG, Seoul, Republic of Korea;
FRANCISCO BELLOT ROSADO, LB. Emilio Ferrari, and MARIA ASCENSION LOPEZ
CHAMORRO, LB. Leopoldo Cano, Valladolid, Spain; ILIYA BLUSKOV, Technical
University, Gabrovo, Bulgaria; JORDI DOU, Barcelona, Spain; HANS ENGELHA UPT,
Franz-Ludwig-Gymnasium, Bamberg, Germany; C. FESTRAETS-HAMOIR, Brussels,
Belgium; RICHARD I. HESS, Rancho Palos Verdes, California; JEFF HIGHAM, student,
University of Toronto; L.J. HUT, Groningen, The Netherlands; WALTHER JANOUS,
Ursulinengymnasium, Innsbruck, Austria; DAG JONSSON, Uppsala, Sweden; MARCIN
E. KUCZMA, Warszawa, Poland; KEE-WAI LAU, Hong Kong; P. PENNING, Delft,
The Netherlands; D.J. SMEENK, Zaltbommel, The Netherlands; CHRIS WILDHAGEN,
Rotterdam, The Netherlands; and the proposer.
The variety of solutions submitted for this problem was quite astonishing; above
are four of the best. Shirali was the only reader to notice the more general result given
in solution IV. Readers may enjoy discovering whether the other given solutions can be
adapted to yield the same thing.
50

1 6 0 2 . [1991: 14] Proposed by Marcin E. Kuczma, Warszawa, Poland.

Suppose xi,x2,...,xn € [0,1] and £" = 1 xi = m + r where m is an integer and
r G [0,1). Prove that
n
]T^x2 < ra + r 2 .

I. Solution by Richard L Hess, Rancho Polos Verdes, California.

Consider maximizing J2^i xi subject to ]^JL1 xt- = m + r. For xt- + Xj = c,

x? + x) = a?? + (c - x^)2 = 2x? - 2xiC + c2

is maximized when

{ c or 0 if c < 1,

1 or c - 1 if c > 1
[i.e., at an "endpoint", since the graph of y = 2x —2xc+c2 is a parabola opening upwards].
2

This means that for maximum ]C?=i xh a * m o s t one of the Xj's can be different from 0
or 1, allowing one to maximize J27=i xl ^Y letting rn of the X{ be 1, one X{ be r, and the
remaining n — m — 1 of the xt- be 0. Thus
n
2
^x < ra + r 2 ,

with equality holding in the above case.

II. Generalization by Walther Janous, Ursulinengymnasium, Innsbruck, Austria.
We prove
THEOREM. Let q be a positive real number and / : [0,g] —» R be a convex
function. Furthermore, let x i , . . . , xn G [0, g] (where n > 2) be such that X^LX #* = m j + r,
where m is an integer and 0 < r < q. Then

£ / ( * , ) < (n - m - l)/(0) + m/(g) + / ( r ) .

1= 1

We first prove the following:

LEMMA. Let / : [a, /?] —> R be a convex function and a, 6, c, d G [a, /?] such that
a < c,d < b and a + 6 = c + d. Then

f(a) + f(b)>f(c) + M.
Proof. If equality occurs at least once in a < c, d < 6, the claimed inequality is clear
(being also an equality). Thus let a < c, d < b. The convexity of / yields

m - /(a) ^ /(&) - /(q) ^ /(i) - /(^)

c —a ~~ b—a b—d

Comparing the first and third terms of this inequality, we get the claimed one. •
 51

Proof of theorem. We proceed by Induction. If n = 25 then either

(I) m = 09 and then x\ + £2 = 0 + r where 0 < a?i, x 2 < r,
or
(ii) m = 1, and then #1 + £2 = 9 + ^? where r < xly x2 < q;
and In both cases we're done by the lemma. [A third case, namely m = 2, xx = x2 = g,
r = 0, is easily taken care of. —Ed.]
Let us assume the validity of the theorem up to n. Let Yl?=i x% = mq + ^ and put
^n+i = 3?- We now apply the Induction hypothesis to J2?=i xi = m 9 + r — #.
(7a^e (i): 0 < x < r. Then 0 < r — a? < g, so

t=i t=i
< (n _ m _ i)/( 0 ) + m /( g ) + /( r _ x) + /(*)
< ( n - m ) / ( 0 ) + m/(9) + /(r)
(where we have used the lemma for f(r — x) + /(#) < /(0) + / ( r ) ) .
Case (II): r < x < q. Then
n
^ Xi = (m — l)g + q + r — x9 0 < g + r - x < 9,

and thus
n-fl n

1=1 1=1

< (n _ m )/( 0 ) + ( m - l)f(q) + f(q + r - x) + f(x)

< (n - m)/(0) + ro/(«r) + / ( r )
(where we have used the lemma for f(q + r — x) + f(x) < f(r) + f(q)). Done! •
Choosing q = 1 and /(x) = x 2 we get the original problem.
Also solved by MARGHERITA BARILE, student, Universitd degli Studi di
Genova, Italy; MURRAY S. KLAMKIN, University of Alberta; KEE-WAI LAU, Hong
Kong; PAVLOS MARAGOUDAKIS, student, University of Athens, Greece; BEATRIZ
MARGOLIS, Paris, France; JEAN-MARIE MONIER, Lyon, France; CORY C. PYE,
student, Memorial University of Newfoundland, St. John's; DAVID C. VAUGHAN,
Wilfrid Laurier University, Waterloo, Ontario; CHRIS WILDHAGEN, Rotterdam, The
Netherlands; and the proposer.
The method of choice for most solvers was induction on n. However, the proposer's
solution was somewhat like Hess's (solution I). Vaughan gave a generalization weaker than
that of Janous. Klamkin proved essentially the same theorem as Janous, but did it by
applying the majorization inequality to
(xu x2,...,xn)^<(qJq,...,q,r,Q,..., 0)
(in which there are m q's and n — m — 1 \9s).
^K ^js Jyfc ^j^ 2|C
52

1603. [1991: 14] Proposed by Clifford Gardner, Austin^ Texas, and Jack
Garfunkel, Flushing, N.Y.
Given is a sequence Fi, F2,... of con-
centric circles of increasing and unbounded
radii and a triangle AiBiCi inscribed in
IV Rays AiBijBiCijCiAi are extended
to intersect T2 at J?2,C2,A2, respectively.
Similarly, AA3B3C3 is formed in T^ from
AA2B2C21 and so on. Prove that AAnBnCn
tends to the equilateral as n —* 00, in the
sense that the angles of AAnBnCn all tend
to 60°.
Solution by Marcin E. Kuczma, Warszawa, Poland,
Let triangle AnBnCn have angles Ani J3 n ,
Cn and let

dn = max{|A n - Bnl \Bn - C n |, \Cn - An\}.

We must show lim^oo dn = 0.

Denote by O the common center of all
circles and by Rn the radius of the nth circle,
and write tn = Rn/Rn+i. Since the radii are Un+1
unbounded, the infinite product t^ts . . . is equal
Cn+l
toO.
Let Kn, Lnj Mn be the midpoints of BnCn,
CnAnj AnBn, respectively. Then

OKn = Rn cos(lKnOCn) = Rn cos An

and
OKn = i?n+i cos(lKnOCn+i).
Hence £KnOCn+i = fn(An), where
fn(x) = arccos(i n cos#), 0 < x < n. (1)
Likewise, LLnOAn+\ = fn(Bn) and LMnOBn+i = / n (C n )-
Note the angle equality
Z # n O C n + 1 + Z C n + 1 0 4 n + 1 = / / ^ O C n + lCnOLn + Z£ n 0A n + 1 ,

i.e.
fn(An) + 2 £ n + 1 = A n + Bn + fn(Bn). (2)
(The argument has to be slightly modified if AAnBnCn is obtuse-angled, but the formula
(2) is valid in that case, as well; to avoid the need of considering this as a special case we
could work with oriented angles.)
 53

Now we rewrite equation (2) together with its two cyclic counterparts in the form
22?n+i = An + Bn + fn(Bn) — / n (i4„),
2Cn+l = Bn + Cn + fn(Cn) — fn(Bn),
2A n+1 - Cn + An + / n (A n ) - / n (C„).
It is easily seen by differentiating (1) that \fn(x)\ <tnj thus

\fn(x)-fn(y)\ <tn\*-y\

for all x}y [by the mean value theorem]; i.e., fn is contractive with Lipschitz constant tn.
So we get
2\Bn+1 - Cn+xl < K - C n | + \fn(Bn) - fn{Cn)\ + \fn(An) - fn(Bn)\
< \An - Cn\ + tn\Bn - Cn\ + tn\An - Bn\
and similarly
2|C n+ i — An+i\ < \Bn — An\ + tn\Cn — An\ + tn\Bn — C„|,
2|A n+1 - Bn+l\ < \Cn - Bn\ + tn\An - Bn\ + tn\Cn - An\.
Since the configuration is cyclic, we may assume that Bn is the middle angle, i.e.,
either An > Bn > Cn or An < Bn < C„ holds, so that \An -Bn\ + \Bn - Cn\ = \An - Cn\ =
dn. Then
2|J5n+i — Cn+i\ < \An — Cn\ + tn\An — Cn\ = (1 + tn)dn,
2\Cn+1 - An+1\ < \Bn - An\ + tn\Cn - An\ + \Bn - Cn\ = (1 + tn)dn,
2\An+1 - Bn+1\ < \Cn - Bn\ + \An - Bn\ + tn\Cn - An\ = (1 + tn)dn,
and consequently
<*»+i < ^ (l + i » K . (3)
Since the product FI^Li ^n is zero, so is FI£Li(l + *n)/2. [iftftior's rao£e. Helpful
colleague Len Bos supplied the following proof of this. One easily checks that

i ± £ < *1/3 for 6<x<l,

2 ~ - - »
for some positive constant 8 < 1. Then if infinitely many of the <n's are less than S,

while if only finitely many of the fn5s are less than 5,

(n*. \ 1/3

=<>;
tn>S J
in either case II£Li(l + *n)/2 = 0.]
54

From (3),
dn < di U —-— ,
and therefore limn-+oo dn = 0.
Also solved (the same way) by the proposers. One other reader sent in a partial
solution in which he showed that dn (in the above proof) decreases, but didn't show that it
went to zero.

1 6 0 4 . [1991: 14] Proposed by K.R.S. Sastry, Addis Ababa, Ethiopia.

Ever active Pythagoras recently took a stroll along a street where only Pythagoreans
lived. He was happy to notice that the houses on the left side were numbered by squares of
consecutive natural numbers while the houses on the right were numbered by fourth powers
of consecutive natural numbers, both starting from 1. Each side had the same (reasonably
large) number of houses. At some point he noticed a visitor. "It is awesome!" said the
visitor on encountering Pythagoras. "Never did I see houses numbered this way." In a
short discussion that followed, the visitor heard strange things about numbers. And when
it was time to part, Pythagoras asked "How many houses did you see on each side of the
street?" and soon realized that counting was an art that the visitor had never mastered.
"Giving answers to my questions is not my habit", smilingly Pythagoras continued. "Go
to a Crux problem solver, give the clue that the sum of the house numbers on one side is
a square multiple of the corresponding sum on the other side and seek help."
Solution by Hans Engelhaupt, Franz-Ludwig-Gymnasium, Bamberg; Germany.
Let n be the number of houses on either side. The sums of house numbers on the
left and right sides are respectively

" .2 _ n(n + l)(2n + 1) A .4 _ n(n + l)(2n + l)(3re2 + 3n - 1)

2^ _ _ and 2^ - - -^ ,
so their ratio being a square implies
3n2 + 3n - 1 2
= q
— 5 -
Thus
- 3 + y/2T+ 60?2 - 3 + V21 + 15r2
n = =
6 6
with r = 2g. We want solutions of the Pell equation

z2 - 15r2 = 21 (1)

from which we will then get n = (z — 3)/6 (when this is an integer). Equation (1) has
smallest solutions
(*i,r!) = (6,l) and (s 2 ,r 2 ) = (9,2), (2)
 55

and the Pell equation u2 — lbv2 = 1 has smallest solution (u, v) = (4,1). Therefore solutions
of (1) are given by

(:::) • (J ? ) (::) •
or
zi+2 = 4zi + 15r,-, r,-+2 = *,- + 4r,-.
Starting with the solutions (2), one gets

zk rk n = (zk - 3)/6
6 1 —
9 2 1
39 10 6
66 17 —
306 79 —
519 134 86
2409 622 401
4086 1055 —
18966 4897 —
32169 8306 5361

Also solved by SAM BAETHGE, Science Academy, Austin, Texas; C. FESTRAETS-

HAMOIR, Brussels, Belgium; RICHARD I. HESS, Rancho Palos Verdes, California;
WALTHER JANOUS, Ursulinengymnasium, Innsbruck, Austria; SIDNEY KRAVITZ,
Dover, New Jersey; J.A. MCCALLUM, Medicine Hat, Alberta; P. PENNING, Delft, The
Netherlands; BOB PRIELIPP, University of Wisconsin-Oshkosh; R.P. SEALY, Mount
Allison University, Sackville, New Brunswick; D.J. SMEENK, Zaltbommel, The Nether-
lands; KENNETH M. WILKE, Topeka, Kansas; and the proposer.
Most solvers guessed that either 86 or 401 was the intended answer. Hess chose 86
because, as he says, "Pythagoreans are probably a rather selective group".
Janous wonders for which positive integer values of k and I the quotient
n In

|=1 / |=1

is a polynomial with rational coefficients. The example (&, /) = (4? 2) occurs in this problem.

1 8 0 5 . [1991: 14] Proposed by M.S. Klamkin and Andy Liu, University of Alberta.
ADB and AEC are isosceles right triangles, right-angled at D and E respectively,
described outside A ABC. F is the midpoint of BC. Prove that DFE is an isosceles
right-angled triangle.
56

Solution by Leon Bankoff, Los Angeles, Califc

Let Gj H be the midpoints of AB} AC
respectively. It is easily seen that DG = AG =
HF and that HE = AH = GF. Also5 LDGF =
IF HE since the extensions of their corresponding ^
sides are mutually perpendicular. Hence triangles
DGF and FiLE are congruent with DF = F E .
Furthermore, since corresponding sides of triangles
DGF and .FJETE are mutually perpendicular, DF
is perpendicular to FE. Thus D/JiF is a right-
angled isosceles triangle.
Also solved by MIGUEL AMENGUAL COVAS, Gala Figuera, Mallorca, Spain;
SEFKET ARSLANAGIC, Trebinje, Yugoslavia; SAM BAETHGE, Science Academy,
Austin, Texas; SEUNG-JIN BANG, Seoul, Republic of Korea; FRANCISCO BELLOT
ROSADO, LB. Emilio Ferrari, Valladolid, Spain; ILIYA BLUSKOV, Technical Univer-
sity, Gabrovo, Bulgaria (three solutions); JASON COLWELL, student, Edmonton; JORDI
DOU, Barcelona, Spain; HANS ENGELHAUPT, Franz-Ludwig-Gymnasium, Bamberg,
Germany; C FESTRAETS-HAMOIR, Brussels, Belgium; HERTA T. FREITAG,
Roanoke, Virginia; RICHARD L HESS, Rancho Palos Verdes, California; JEFF
HIGHAM, student, University of Toronto; L.J. HUT, Groningen, The Netherlands;
WALTHER JANOUS, Ursulinengymnasium, Innsbru.k, Austria; CHARLES H. JEPSEN,
Grinnell College, Grinnell, Iowa; DAG JONSSON, Uppsala, Sweden; GIANNIS G.
KALOGERAKIS, Canea, Crete, Greece; MARGIN E. KUCZMA, Warszawa, Poland;
KEE-WAI LAU, Hong Kong; JEAN-MARIE MONIER, Lyon, France; P. PENNING,
Delft, The Netherlands; JOHN RAUSEN, New York, N.Y. (two solutions); MARIA
MERCEDES SANCHEZ BENITO, LB. Luis Bunuel, Madrid, Spain; SHAILESH
SHIRALI, Rishi Valley School, India; DJ. SMEENK, Zaltbommel, The Netherlands;
CHRIS WILDHAGEN, Rotterdam, The Netherlands; KENNETH M. WILKE, Topeka,
Kansas; and the proposers (three solutions).
Janous, Rausen, and the proposers observed that the result also holds if the triangles
ADB and AEC are constructed inwardly on AABC. Janous and the proposers further
show that if ADB and AEC are similar and isosceles but not right-angled, then they will
not be similar to ADFE.
Wilke points out the similar problem Crux 540—see especially the published solution
on [1981: 127]. (Interestingly, Crux 540 was proposed by our featured solver Leon Bankoff!)
Bellot reduced the problem to exercise 24(c), pages 39 and 96-97 of LM. Yaglom, Geometric
Transformations I, NML 8, M.A.A., 1962.
 57

1 6 0 6 * . [1991: 14] Proposed by Walther Janous, Ursulinengymnasium, Innsbruck,

Austria.
For integers n > k > 1 and real x, 0 < x < 1, prove or disprove that

Solution by G.P. Henderson, Campbellcroft, Ontario.

We will prove that the inequality is true. We have

n-j

>|(i-iy(;pa-^
2
S( 1 -5)(^ 1 -'*"'
where the last step follows because

IV
S 1
('-*) -p ^O'1"
as is easily proved by induction on j .
Also solved (almost the same way) by SEUNG-JIN BANG, Seoul, Republic of Korea;
and MANUEL BENITO, LB. Sagasta, Logrono, Spain.
The problem came from a statistical question put to the proposer by one of his
colleagues.

1 6 0 7 . [1991: 15] Proposed by Peter Hurthig, Columbia College, Burnaby, B.C.

Find a triangle such that the length of one of its internal angle bisectors (measured
from the vertex to the opposite side) equals the length of the external bisector of one of
the other angles.
58

I. Solution by Andy Liu, University of Alberta.

We shall find a triangle ABC such that the
exterior bisector BE\ of I ABC and the interior
bisector AD of I BAG are equal in length to each
other and to AB. Extend AB to P . Let I BAD =
x. Then LOAD = a?, / B E i C = 2a? and LEXBC =
Z£ii3P = 4a?. Also, ZABD = ZADB = 90°-a?/2.
It follows that

180° = iBiBP+lExBC+lABD = 8a?+90°-x/2.

Hence x = 12°, and we have

Z£AC = 24°, M 5 C = 84°, LACB = 72°.

II. Solution by Shailesh Shirali, Rishi Valley School, India.
Let the lengths of the internal and external angle bisectors associated with an angle
X be denoted txy Tx, respectively. Then
a simple instance where the equality tA =
TB holds is obtained by placing an additional
restriction, namely that angle A equals angle
B. Then, with tA = tB = ~BE, TB = B ^ 5
and I ABE = /? as shown in the diagram, we
have 0 + 2/3 = TT/4 and so /? = TT/12. Thus
triangle ABC with angles
7T 2fl"
(A, 5 , C) = ( | , £ , y ) = (30°, 30M20 0 )
satisfies the relation tA = TB.
Proceeding more generally, let us note that by elementary trigonometry,

2 r
^M^) )' *
Then the relation £4 = TB implies that
c — a.
1

b[(b + c)2- [fr2 - (c - a)2]

(6 + c)2
^M^htfcy^H
This can be substantially simplified to yield
(c - a) 2

ba2 - (b2 + c2 + Zbc)a + c2{b + c) = 0.

This is a quadratic equation in a (if wefixb and c). Without loss, we may put 6 = 1 and
then the above equation yields
( c + l)2 + c _^( c 2 + c + 3 ) 2 _ 8

a— (1)
 59

If we denote the function on the right side of (1) by /(c), then the following relations are
easily verified:
(i) c > 0 implies 0 < /(c) < c and also /(c) + 1 > c;
(ii) c> >/3 - 1 implies f(c) + c > 1, while the triangle inequality fails for c < V^ - 1 ;
(iii) (/(c) ? 1, c) is a feasible triple for the sides of a triangle for any c> \/Z-l.
This procedure obviously gives us infinitely many triangles that have the required property.
For instance, c = 1 gives the triple

M,c)=p^,i,i)
and c = 2 gives

(..M-i'ii^.M
(a,6,c) = f-

The solution listed at the beginning corresponds to c = y/3.

III. Solution by Jordi Don, Barcelona, Spain.
[Dou first gave the 30°, 30°, 120° example in Solution II. — Ed.]
We have the following interesting result: if the interior bisector AD is equal to the
exterior bisector BE\} then IA < <p, where (p is approximately 30.214335°.
To establish this result, consider triangle
ABC with AD = 1. Then
s i n(C
+ A/2)
sm(<7 + A)

and so

no _ AB sin A 8JnAsin(C + A/2)

1 l ? h { }
"" sin((C - A) 12) sm{C + A) sin((C - A)/2)
The condition AD = BEi is F(i4, (7) = 1. For a fixed value of A, the function F(A, C) =
*U(C) is defined for values of C with A < C < 180° - A. FA(C) goes to oo as C —-» A
and as (7 —-* 180° — A. Since FA(C) > 0, ^ ( C O has an absolute minimum at C = 74,
say. When FA(*1A) = 1? then A = (p. With formula (2) and a hand calculator, one obtains
for <p the value given above. (I don?t like it that (p is so close to 7r/6 but not exactly!)
We see that if A < ip, there exist two triangles ABC such that AD = BEX = 1. If
A = <p, there exists a single triangle ABC such that AD = J3I?i = 1. In this case, angle (7
is 74 « 113.46°. If A, 5 , C > y> there exists no triangle ABC which satisfies the condition.
Also solved by FRANCISCO BELLOT ROSADO, LB. Emilio Ferrari, Valladolid,
Spain; C FESTRAETS-HAMOIR, Brussels, Belgium; L.J. HUT, Groningen, The
Netherlands; WALTHER JANOUS, Ursulinengymnasium, Innsbruck, Austria; MARGIN
E. KUCZMA, Warszawa, Poland; P. PENNING, Delft, The Netherlands; and the pro-
poser. There were two incorrect solutions sent in.
60

The solutions were about evenly split between those containing the condition on
the sides a, 6, c (Solution IIJ and those containing the condition on the angles A>B,C
(Solution IIIJ. Most solvers, including the proposer, found the 30°,30 o ,120 o triangle given
in Solutions II and III. The proposer also spotted the 24°?84°,720 triangle given by Liu in
Solution L Are there any other solutions with all angles integral?
Kuczma showed that for every a G (2 — \/3? 1) there is a unique triangle of semi-
perimeter 1 (and side opposite A of length a) such that the internal bisector of angle A
equals the external bisector of angle B. Festraets-Hamoir noted that the angles of such a
triangle ABC satisfy the nice equation

cos(A/2)sin(5/2) = cos 2 (C/2).

The problem was suggested by the result (due to Bottema) that there are nonisosceles
triangles with two external angle bisectors of equal length.

1 6 0 8 . [1991: 15] Proposed by Seung-Jin Bang, Seoul, Republic of Korea.

Suppose n and r are nonnegative integers such that no number of the form n2 + r —
k(k + 1), fc = 1,2,..., equals —1 or a positive composite number. Show that 4n2 + 4r + 1
is 1, 9, or prime.
Solution by Marcin E. Kuczma, Warszawa, Poland.
Assume that m = 4n2 + 4r + 1 is neither 1 nor prime. Then it has an odd prime
divisor p = 2k + 1, p < ^Jm (k > 1).
Case (i): p < %Jm. Then

2 | i n , _ 4 n 2 + 4 r + l - ( 4 P + 4 & + l) m-p2
n' + r-k(k + l) = —^ = —j—

is a positive integer, divisible by p. In view of the assumption it must be equal to p. So

n2 + r-k(k + l) = 2A? + 1,
i.e.
n2 + r - (k + l)(k + 2) = 2k + 1 - 2(fc + 1) = - 1 ,
in contradiction to the other condition of the problem.
Case (ii): p = y^m. Then

4n2 + 4r + l = (2fc + l) 2

whence
n2 + r-k(k- 1) = 2fc,
a contradiction again, unless k = 1, in which case 4n2 + 4r + 1 = 9, as desired.
Also solved by MARGHERITA BARILE, student, Universita degli Studi di Genova,
Italy; WALTHER JANOUS, Ursulinengymnasium, Innsbruck, Austria; KEE-WAI LAU,
Hong Kong; and the proposer. One partial solution was received.
 61

Janous calls the problem "lovely", but notes that a simpler formulation is possible
merely by replacing n2 + r by N, say. He then asks for which integers N > 0 the condition
of the problem holds — only finitely many?
The problem is a generalization of a problem in the 1990 Korean Math Olympiad,
* * * * *

1609. [1991: 15] Proposed by John G. Beuver, Grande Prairie Composite H.S.,
Grande Prairie, Alberta.
P is a point in the interior of a tetrahedron ABCD of volume V, and Faj F^ FCJ Fd
are the areas of the faces opposite vertices A, J3? C, D, respectively. Prove that

PA • Fa + PB * Fb + PC * Fc + PD - Fd > 9V.

Solution by Iliya Bluskov, Technical University, Gabrovo, Bulgaria.

Let haj hhj hci hd be the altitudes from vertices A, £?5 C, /?, and let da, d\,,rfc,dd
be the distances from P to the faces opposite vertices A, B^ C? £), respectively. Then

PA > ha — daj etc.

Since /&0.F0 = 3F ? etc., we get (with sums cyclic)

T,PA'F«^ E*a^a - £<U^ = 4 • W - W = 9F5

with equality when P is the intersection of the altitudes.
Also solved by CON AMORE PROBLEM GROUP, Royal Danish School of Educa-
tional Studies, Copenhagen; C. FESTRAETS-BAMOIR, Brussels, Belgium; WALTBER
JANOUS, Ursulinengymnasium, Innsbruck, Austria; MURRAYS. KLAMKIN, University
of Alberta; MARGIN E. KUCZMA, Warszawa, Poland; TOSHIO SEIMIYA, Kawasaki,
Japan; SBAILESH SHIRALI, Rishi Valley School, India; and the proposer.
Janous notes that an analogous inequality is known for n-dimensional simplices:

where F{ is the (n — 1)-dimensional area of the face opposite vertex A{, and R{ = A{P. See
item 2.29; pp. 494-5 of D,S* Mitrinovic et al, Recent Advances in Geometric Inequalities.
Klamkin gave a generalization,
% * $ * Jjc

1610. [1991: 15] Proposed by P. Penning, Delft, The Netherlands.

Consider the multiplication dx ddx ddd, where d < b — 1 is a nonzero digit in base
6, and the product (base b) has six digits,, all less than b — 1 as well. Suppose that, when
d and the digits of the product are all increased by 1, the multiplication is still true. Find
the lowest base b in which this can happen.
62

Solution by Sam Baethge, Science Academy, Austin, Texas.

We have (in base 6)
d 3 ( i ) ( i i ) ( i n ) +111111 = {d + i ) 3 ( i ) ( i i ) ( n i ) ,
or
1001 = 11(3J2 + 3J + 1).
Then
b3 + l = (b+l)(U2 + 3d+l)J
2 2
b -b+l = 3d + 3d+l,
and so
6(6 - 1) = 3d(d + 1).
This equation is satisfied by d = 1, 6 = 3 which we reject since the product [1 • 11 • 111 =
I22I3] is less than 6 digits. The lowest base which satisfies the problem is b = 10, and then
d = 5, where
5(55)(555) = 152625, 6(66)(666) = 263736.
Also solved by DUANE M. BROLINE, Eastern Illinois University, Charleston;
KEIR COLBO, student, Memorial University of Newfoundland, St. John's; HANS
ENGELHAUPT, Franz-Ludwig-Gymnasium, Bamberg, Germany; RICHARD I. HESS,
Rancho Palos Verdes, California; STEWART METCHETTE, Culver City, California;
SHAILESH SHIRALI, Rishi Valley School, India; and the proposer. Two incorrect
solutions were sent in.

1 6 1 1 . [1991: 43] Proposed by George Tsintsifas, Thessaloniki, Greece.

Let ABC be a triangle with angles A, J3, C (measured in radians), sides a, 6, c, and
semiperimeter s. Prove that

Solution by Ian Goldberg, student, University of Toronto Schools.

First, for positive a?, t/, the function /(#,2/) = x/y is linear or concave up in each
variable. Therefore, by Jensen's inequality,
#1/2/1 + #2/2/2 + #3/2/3 (xi + x2 + x3)/3
—
3 (2/1 + 2/2 + jfa)/3
or
£ l , £2 , f^3 > o (x\ + #2 + X^\ /-v
2/i 2/2 2/3 ~~ V 2/1 + 2/2 + 2/3 /
Let 2/1 = A, 2/2 = 1?, 2/3 = C\ then this becomes
#1 Xo #3 ^ 3 , x ,^x
^ • + ^ H - - £ 7 > - ( x i + Z2 + s 3 ). (2)
 63

Now for part (i), let xx = b + c - a, x2 = c + a - 6, x3 = a + b - c, so that

xi + X2 + X3 = 2s, and the result follows.
For part (ii), let
b+ c — a c+ a— b a+ b— c
Xi = , x2 = ——7 , x3 = ;
a 0 c
then by (1)
Y > , > 3 ((b + c-a) + (c + a-b) + (a + b~c)\ = $
fei ' ~ V a+ b+ c J '
and the result follows by (2).
We can also solve Crux 1637 [1991: 114], by the same proposer, the same way. Let
xi = sin B + sin C, x2 = sin C + sin A, x3 = sin A + sin B. For 0 < a < 7r/2, sin a > 2a/7r;
thus
3 9
£ * . - = 2(sinA + sinB + sinC) > 2 • - (A + B + C) = 4,
,=1 *
and so from (2)
sin 5 + sin C ^ 3 ^ ^12
A -^'-V
which Is Crux 1637.
Also solved by $EFKET ARSLANAGIC, Trebinje, Yugoslavia; C. FESTRAETS-
BAMOIR, Brussels, Belgium; STEPBEN D. HNIDEI, Windsor, Ontario; WALTBER
JANOUS, Ursulinengymnasium, Innsbruck, Austria; MARCIN E. KUCZMA, Warszawa,
Poland; PAVLOS MARAGOUDAKIS, student, University of Athens, Greece; VEDULA
N. MURTY, Penn State Barrisburg; and the proposer.
Janous showed more generally that for A > 1, fi > 0,
^(b + c-a)x 3^ A+1 (2^) A
*-* A»> ~ 7T»

and

These can be proved as above, using f{x,y) = xxjy^. Be also pointed out the related in-
equality 4*2, P®ge 168 of D*S. Mitrinovic et al, Recent Advances in Geometric Inequalities.

1 6 1 2 * . [1991: 43] Proposed by Walther Janous, Ursulinengymnasium, Innsbruck,

Austria.
Let x^yjZ be positive real numbers. Show that

*—f z -\- x

where the sum is cyclic over x, y, z} and determine when equality holds.
64

Solution by Margherita Barile, student, Universitd degli Studi di Genova, Italy.

Let
5 - 3/ 2 ~s 2 z2-y2 x2-z2
z+x x +y y+z
Since the sum is cyclic over x} y, z we only need to consider the following two cases.
Case 1: x < y < z. We have to prove that
z2 — y2 y2 — x2 z2 — x 2
a? + y z + x ~~ y + z '
But x + y < J/ + z and # + # < y + z, so that
z2 — y2 y2 — x2 z2 — y2 + y2 — x2 __ z2 — x2
x+y z + x ~~ y+z y+z
Case 2: x < z < y. The result is equivalent to
y2 — x2 y2 — z2 z2 - x2
z + x -~ x + y y+z
which is true because z + x < x + y and z + x < y + z.
Equality holds if and only if x = y = z.
Also solved by H.L. ABBOTT, University of Alberta; HAYO AHLBURG, Benidorm,
Spain; BENO ARBEL, Tel Aviv University, Israel; SEFKET ARSLANAGIC, Trebinje,
Yugoslavia; SEUNG-JIN BANG, Seoul, Republic of Korea; MANUEL BENITO, L B.
Sagasta, Logrono, Spain; ILIYA BLUSKOV, Technical University, Gabrovo, Bulgaria;
DANIEL BROWN, North York, Ontario; IAN GOLDBERG, student, University of
Toronto Schools; JEFF HIGHAM, student, University of Toronto; STEVE HNIDEI,
Windsor, Ontario; PETER HURTHIG, Columbia College, Burnaby, B.C.; L.J. HUT,
Groningen, The Netherlands; MURRAY S. KLAMKIN, University of Alberta; MARGIN
E. KUCZMA, Warszawa, Poland; ANDY LIU, University of Alberta; PAVLOS
MARAGOUDAKIS, student, University of Athens, Greece; BEATRIZ MARGOLIS, Paris,
France; VEDULA N. MURTY, Penn State Harrisburg; M. PARMENTER, Memorial
University of Newfoundland, St. John's; TOSHIO SEIMIYA, Kawasaki, Japan; CHRIS
WILDHAGEN, Rotterdam, The Netherlands; and the proposer. There was one anony-
mous solution sent in. Two other readers submitted incorrect solutions due to considering
only one of the above two cases.
The proposer proved more generally that for any A > 0 ;

EyA"gA>o
^ Z+ X
for all positive x, y, z, which can be shown as above. Even more generally, Klamkin replaced
y2 by F(y), etc., where F(t) is a nondecreasing function oft for t > 0, to obtain
^ F(y) - F(x)
^ z+x
for all positive x,y,z.
 • • • • BOUND VOLUMES * * • *

I • • • VOLUMES RELIES • • • |

THE FOLLOWING BOUND VOLUMES OF/ CHACUN DES VOLUMES SUIVANTS DE

CRUX MA THEM A T/CORUM

ARE A VAILABLE AT $10.00 PER VOLUME / EST DISPONIBLE A 10$

1 & 2 (Combined / ensemble), 3, 7, 8 and 9

PLEASE SEND CHEQUES MADE PAYABLE TO

VEUILLEZ ACHEMINER VOS CHEQUES LIBELLES AU NOM DE

CANADIAN MA THEMA TICAL SOCIETY / SOCIETE MA THEMA TIQUE DU CANADA

577 King Edward

Ottawa, Ontario
Canada K1N 6N5

Volumes: x $10.00 = $ _^

(Canadian residents, please add GST / Residents canadiens, veuillez ajouter la TPS)

MAILING ADDRESS / ADRESSE POSTALE |

|| Name/ Norn:

Address/ Adresse:

City, Province / Ville, Province: ||

Postal Code, Country / Code postal, pays:

 PUBLIC A Tl O N S
The Canadian Mathematical Society
577 King Edward, Ottawa, Ontario K1N 6N5
is pleased to announce the availability of the following publications:

1001 Problems in High School Mathematics

(Collected and edited by E.J. Barbeau, M.S.
Klamkin and W.O.J. Moser.)

Book I Problems 1-100 and Solutions 1-50, 58 pages OUT OF PRINT

Book II Problems 51-200 and Solutions 51-150, 85 pages ($7.50, CMS Members: $6.00)

Book III Problems 151-300 and Solutions 151-350, 95 pages OUT OF PRINT

Book IV Problems 251-400 and Solutions 251-350, 115 pages ($7.50, CMS Members: $6.00)

I BookV Problems 351-500 and Solutions 351-450, 86 pages ($7.50, CMS Members: $6.00) ||

The Canadian Mathematical Olympiads

(1969-1978) and (1979-1985)

Problems set in the first ten Olympiads (1969-1978) together w i t h suggested

solution. Edited by E.J. Barbeau and W . O . J . Moser. 89 pages.
($7.50, CMS Members: $6.00)

Problems set in the Olympiads (1979-1985) together with suggested solution.

Edited by C M . Reis and S.Z. Ditor. 84 pages.
($7.50, CMS Members: $6.00)

(The Set of 2 books is available at $12.00)

(Canadian residents, please add GST)

Prices are in Canadian dollars and include handling charges.

Information on other CMS publications can be obtained
by writing to the address given above.