225

THE ACADEMY CORNER

No. 41
Bruce Shawyer
All communications about this column should be sent to Bruce
Shawyer, Department of Mathematics and Statistics, Memorial University
of Newfoundland, St. John's, Newfoundland, Canada. A1C 5S7

In this issue, we present some solutions to the First Day problems of the
7th International Mathematics Competition for University Students, held at
University College, London, UK, on 26 July and 31 July 2000 [2000 : 385].
The International Mathematics Competition for
University Students
First Day Problems, 26 July 2000
1. Is it true that if f : [0; 1] ! [0; 1] is
(a) monotone increasing
(b) monotone decreasing
then there exists an x 2 [0; 1] for which f (x) = x?
Solution by Michel Bataille, Rouen, France.
(a) True. Let A = fx 2 [0; 1] : f (x)  xg. Then A is non-empty (since
0 2 A) and bounded above by 1. It follows that A has a least upper bound
a satisfying a 2 [0; 1]. We will show that f (a) = a.
For all x 2 A, we have f (x)  f (a) (since x  a and f is increasing).
Hence, x  f (a).
Thus, f (a) is an upper bound of A, and, consequently, a  f (a). (1)
Observing that f (a) 2 [0; 1], we have that (1) yields f (a)  f (f (a)),
which shows that f (a) 2 A. Hence, f (a)  a. This and (1) give f (a) = a,
as required.
(b) False. Take f de ned by

f (x) =
(
; x2 + 1 if x 2 [0; 12 ) and
; x2 + 12 if x 2 [ 12 ; 1]
Then f : [0; 1] ! [0; 1] is monotone decreasing, and the equation f (x) = x
has no solution.
226

2. Let p(x) = x5 + x and q (x) = x5 + x2 . Find all pairs (w;z ) of complex

numbers with w 6= z for which p(w) = p(z ) and q (w) = q (z ).
Ocial solution.
Let P (x; y ) =
p(x) ; p(y) = x4 + x3 y + x2y2 + xy3 + y4 + 1,
x;y
and Q(x; y ) =
q ( x ) ; q (y)
x ; y = x + x y + x y + xy + y + x + y.
4 3 2 2 3 4

We need those pairs which satisfy P (w;z ) = Q(w; z ) = 0.

From P ; Q = 0, we have w + z = 1. Let c = wz . A short calculation
shows that c2 ; 3c + 2 = 0, which has solutions c = 1 and c = 2.
From the system w + z = 1, wz = c, we obtain the following pairs:
p
1  3i ; 1  3i
p ! 1  p7i 1  p7i !
2 2 and ; .
2 2
Also solved by MICHEL BATAILLE, Rouen, France.
3. Suppose that A and B are square matrices of the same size with
rank (AB ; BA) = 1 .
Show that (AB ; BA)2 = 0.
Solution by Michel Bataille, Rouen, France.
Let C be a square matrix with rank(C ) = 1. Then, the columns of C
are of the form u1 V , u2 V , : : : , un V , where V is a non-zero column vector
and the uk are non-zero scalars. Let U be the column vector with the uk as
its entries. Then, C = V U T , where U T denotes the transpose of U .
Observe that U T V = tr(C ) (where tr(C ) is the trace of the matrix C ).
Thus, we have
C 2 = ;V U T
;V U T
= V ;U T V
U T = ;U T V
V U T = tr(C )C .
Applying this result to C = AB ; BA, we obtain that (AB ; BA)2 =
tr(AB ; BA)(AB ; BA) = 0, since tr(AB ;BA) = tr(AB );tr(BA) = 0.
4. (a) Show that, if fxk g is a decreasing sequence of positive numbers,
then
Xn ! 21 Xn xk
x2k  p .
k=1 k=1 k
(b) Show that there is a constant C such that, if fxk g is a decreasing
sequence of positive numbers, then
1 1 ! 12
1
p1
X X X
m x2k C xk .
m=1 k=m k=1
 227

Ocial solution.
(a)
!2
n n X
n n i n n
pxii pxiip
xj  pxii pxij  pxii i pxii =
X X X X X X
=
j x2i .
i=1 i=1 j =1 i=1 j =1 i=1 i=1
(b) Using a generalization of (a), we obtain
1 1 ! 21 1 1
p1 p1m pi ;xmi + 1
X X X X
m x2
k 
m=1 k=m m=1 i=m
1 i
pmpi 1; m + 1 .
X X
= xi
i=1 m=1
i i+1
1
pxpidx
X Z
By checking that sup p p
i m=1 m i ; m + 1
 + 1 ; x = ,
0
the result is complete.
5. Let R be a ring of characteristic zero (not necessarily commutative). Let
e, f and g be idempotent elements of R satisfying e + f + g = 0. Show
that e = f = g = 0.
(R is of characteristic zero means that, if a 2 R and n is a positive
integer, then na 6= 0 unless a = 0. An idempotent x is an element
satisfying x = x2 .)
Solution by Michel Bataille, Rouen, France.
From eg + fg + g 2 = 0g = 0 = g 0 = ge + gf + g 2 and g 2 = g , we
obtain eg + fg = ;g = ge + gf . (1)
Similarly, ef + gf = ;f = fe + fg (1')
and fe + ge = ;e = ef + eg. (1")
Developing 0 = (e + f + g ) , and using (1), we readily obtain
2
ef + fe = 2g. (2)
It follows from (2) that ef + efe = 2eg and efe + fe = 2ge, so that, by
di erencing and using (1"), we have ef ; fe = 2(eg ; ge) = ;2(ef ; fe).
Hence, 3(ef ; fe) = 0 and so, ef ; fe = 0.
Thus, we have ef = fe, and, with (2), we get 2ef = 2g , or
2(ef ; g) = 0. It follows that ef = fe = g , and, similarly, fg = gf = e
and eg = ge = f .
Now, efg = e(fg ) = e2 = e and efg = (ef )g = g 2 = g , so that
g = e. Similarly, f = g, and we conclude that e = f = g, and further,
e = f = g = 0, since 0 = e + f + g = 3e = 3f = 3g.
228

6. Let f : R ! (0; 1) be an increasing di erentiable function for which

!1 f (x) = 1 and f is bounded.
0
xlim Z x
Let F (x) = f (t) dt. De ne the sequence fang inductively by
0
a0 = 1 , an+1 = an + f (1a ) , (1)
n
and the sequence fbng simply by bn = F ;1 (n).
!1(an ; bn ) = 0. [Ed. The original said 1, but the
Prove that nlim
solution proves 0.]
Ocial solution. From the conditions, it is clear that F is increasing
!1 bn = 1.
and that nlim
By Lagrange's Theorem and the recursion in (1), for all non-negative
integers k, there exists a real number  2 (ak ; ak+1 ) such that
F (ak+1) ; F (ak) = f ()(ak+1 ; ak ) = ff((a)) . (2)
k
By the monotonicity, f (ak)  f ()  f (ak+1). Thus,
1  F (ak+1) ; F (ak)  ff(a(ak+1) ) = 1 + f (ak+1f ()a;)f (ak ) . (3)
k k
Summing (3) over k and substituting F (bn) = n, we have
F (bn) < n + F (a0)  F (an)
;1
nX
 F (bn) + F (a0) + f (ak+1f ()a;)f (ak) . k
(4)
k=0
!1 an = 1.
From the rst two inequalities, we already have an > bn and nlim
Let  > 0. Choose an integer K such that f (aK ) > 2 . If n is su-
ciently large, we have
;1 f (a
nX
k+1) ; f (ak ) =  ; 1 f (a
KX !
F (a0) + F (a0) + k+1 ) ; f (ak)
k=0 f (ak) k=0 f (ak)
nX;1 f (ak+1) ; f (ak)
+ f (ak)
k=K
nX;1
< O (1) + f (a1 ) (f (ak+1) ; f (ak))
K k=K
< O (1) + 2 (f (an) ; f (aK ))
< f (an) . (5)
 229

Inequalities (4) and (5) together imply that for any positive , if n is
suciently large, we have
F (an) ; F (bn) < f (an) .
Again, by Lagrange's Theorem, there is a real number  2 (bn ; an ) such
that
F (an) ; F (bn) = f ( )(an ; bn) > f (bn)(an ; bn) . (6)
Thus,
f (bn)(an ; bn) < f (an) . (7)
Let B be an upper bound for f 0 . Apply f (an) < f (bn) + B (an ; bn)
in (7):
f (bn)(an ; bn) <  (f (bn) + B(an ; bn)) ,
(f (bn) ; B) (an ; bn) < f (bn) .
!1 f (bn) = 1, we see that the rst factor is positive, and
Because of nlim
we have
an ; bn <  f (bf ()b;n)B < 2
n
for suciently large n.
Thus, for arbitrary positive , we have proved that 0 < an ; bn < 2
if n is suciently large, and the proof is complete.

An IMO Training Problem

Given 4A0 B0C0 , construct the First Gunther 3{Triangle 4A1 B1 C1 as
follows:
A1 lies on A0B0 and is such that A0A1 = 13 A0B0, B1 lies on B0C0
and is such that B0 B1 = 31 B0 C0 and C1 lies on C0A0 and is such that
C0C1 = 31 C0A0.
We can clearly iterate this construction.
Show that the Third Gunther 3{Triangle is similar to the First Gunther
3{Triangle.
In a similar way, we can de ne the First Gunther n{Triangle.
Is there a K th Gunther n{Triangle that is similar to the First Gunther
n{Triangle?
230

THE OLYMPIAD CORNER

No. 214
R.E. Woodrow
All communications about this column should be sent to Professor R.E.
Woodrow, Department of Mathematics and Statistics, University of Calgary,
Calgary, Alberta, Canada. T2N 1N4.
Murray Klamkin has written to point out that \Quickies" did not orig-
inate with him. They rst appeared in the March 1950 issue of Mathematics
Magazine. Quickies originated with the late Charles W. Trigg, who was then
Problems Editor. In those days, many Quickies were published by the late
Leo Moser as well as by Murray. But most probably it is true that Murray
has published more Quickies than anyone else.
We start this number with an additional set of ve Klamkin Quickies.
Thanks go to Murray Klamkin, University of Alberta. Try them before looking
ahead to the solutions!
FIVE MORE KLAMKIN QUICKIES
1. Determine the maximum value of
S = 4(a4+b4+c4+d4);(a2bc+b2cd+c2 da+d2 ab);(a2b+b2c+c2d+d2 a) ,
where 1  a, b, c, d  0.
2. If a, b, c, d are > 0, prove or disprove the two inequalities:
(i) c + bcd + cda + dab  a + b + c + d,
ab
(ii) a2 b + b2c + c2 d + d2 a  abc + bcd + cda + dab.
3. Determine all the points P (x; y;z), if any, such that all the points of
tangency of the enveloping (tangent) cone from P to the ellipsoid
x22 + y22 + z22 = 1 (a > b > c), are coplanar.
a b c
4. Determine whether or not there exists a set of 777 distinct positive
integers such that for every seven of them, their product is divisible by their
sum.
5. If R is any non-negative rational approximation to p5, determine
an always better rational approximation.

Next we give the problems of the 28th Austrian Mathematics Olympiad

1997, Final Round Advanced Level. Thanks go to Walther Janous, Ursulinen-
gymnasium, Innsbruck, Austria for sending them to us.
 231

28 AUSTRIAN MATHEMATICS OLYMPIAD 1997

th

Final Round Advanced Level

First Day | June 4, 1997 (Time: 4 hours)
1. Let a be a xed whole number.
Determine all solutions x, y , z in whole numbers to the system of equa-
tions
5x + (a + 2)y + (a + 2)z = a ,
(2a + 4)x + (a2 + 3)y + (2a + 2)z = 3a ; 1 ,
(2a + 4)x + (2a + 2)y + (a2 + 3)z = a + 1 .
2. Let K be a positive whole number. The sequence fan : n  1g is
de ned by a1 = 1 and an is the nth natural number greater than an;1 which
is congruent to n modulo K .
(a) Determine an explicit formula for an .
(b) What is the result if K = 2?
3. We are given a triangle ABC . On side AC a point P is chosen. On
the production of ray CB (beyond B ) there lies the point Y which subtends
equal angles with AP and PC , respectively.
On side BC , point Q is chosen. On the production of ray AC (beyond
C ) there is point X , subtending equal angles with BQ and QC , respectively.
Furthermore, R is the point of intersection of lines Y P and XB ,
S is the point of intersection of lines XQ and Y A, and D is the point of
intersection of lines XB and Y A.
Prove: PQRS is a parallelogram if and only if ACBD is inscribable.
Second Day | June 5, 1997 (Time: 4 hours)
4. Determine all quadruples (a; b; c; d) of real numbers satisfying the
equation
256a3 b3c3d3 = (a6 + b2 + c2 + d2)(a2 + b6 + c2 + d2 )
(a2 + b2 + c6 + d2 )(a2 + b2 + c2 + d6) .
5. We de ne the following operation which will be applied to a row of
bars being situated side-by-side on positions 1, : : : , N :
Each bar situated at an odd numbered position is left as is, while each
bar at an even numbered position is replaced by two bars. After that, all bars
will be put side-by-side in such a way that all bars form a new row and are
situated (side-by-side) on positions 1, : : : , M .
From an initial number a0 > 0 of bars there originates (by successive
application of the above-de ned operation) a sequence, fan : n  0g of
232

natural numbers, where an is the number of bars after having applied the
operation n times.
(a) Prove that for all n > 0 we have an 6= 1997.
(b) Determine the natural numbers that can only occur as a0 or a1 .
6. Let n be a xed natural number. Determine all polynomials
x2 + ax + b, where a2  4b, such that x2 + ax + b divides x2n + axn + b.

Next we turn to the Icelandic Olympiad of 1995{1996. Thanks go to

Mohammed Aassila, Strasbourg, France for sending this to us.

I SLENZKA STAEROFR
 
OIKEPPNI

FRAMHALDSSKOLANEMA 1995{1996

Urslitakeppni
Laugardagur 23. mars 1996 kl. 10{14

1. Calculate the area of the region in the plane determined by the

inequality
jxj + jyj + jx + yj  2 .
2. Suppose that a, b and c are the three roots of the polynomial
p(x) = x3 ; 19x2 + 26x ; 2. Calculate
1 + 1 + 1.
a b c
3. A collection of 52 integers is given. Show that amongst these num-
bers it is possible to nd two such that 100 divides either their sum or their
di erence.
4. (i) Show that the sum of the digits of every integer multiple of 99,
from 1
99 up to and including 100
99, is 18.
(ii) Show that the sum of the digits of every integer multiple of the
number 10n ; 1, from 1
(10n ; 1) up to and including 10n
(10n ; 1),
is n
9.
5. The sequence fang is de ned by a1 = 1 and, for n  1,
an+1 = 1 +ana
n .
n
Find a1996.
 233

6. In a square bookcase two identical books are placed as shown in the

gure. Suppose the height of the bookcase is 1. How thick are the books?

As a third set of Olympiad problems we give the Second Round of the

1997 Iranian Mathematical Olympiad. Thanks go to Mohammed Aassila,
Strasbourg, France.
1997 IRANIAN MATHEMATICAL OLYMPIAD
Second Round
Time: 2  4 hours
1. Suppose that S is a nite set of real numbers with the property
that any two distinct elements of S will form an arithmetic progression with
another element of S . Give an example of such a set with 5 elements and
prove that no such set exists with at least 6 elements.
2. Suppose that ten points are given in the plane such that any ve of
them contain four points which are concyclic. What is the largest number N
for which we can correctly say: \At least N of the ten points lie on a circle"?
(4  N  10.)
3. Suppose that ; is a semi-circle with centre O and diameter AB.
Assume that M is a point on the extension of AB such that MA > MB .
A line through M intersects ; at C and D such that MC > MD. Circum-
circles of the triangles AOC and BOD will intersect at points O and K .
Prove that OK ? MK .
4. Find all functions f : N ! Nnf1g such that for all n 2 Nnf0g we
have,
f (n + 1) + f (n + 3) = f (n + 5)f (n + 7) ; 1375 .
5. Suppose that ABC is an acute triangle with AC < AB and the
points O, H , and P are circumcentre, orthocentre, and the foot of the
altitude drawn, from C on AB , respectively. The line perpendicular
to OP at P intersects the line AC at Q. Prove that \PHQ = \BAC .
234

6. Suppose that A is a symmetric (0; 1)-matrix such that all of its

diagonal entries are 1. Prove that there exist 0  i1 < i2 <


< ik  n
such that Ai1 + Ai2 +


+ Aik  (1; 1; : : : ; 1) (mod 2), where Ai is the
ith row of A.
As a nal set for this number we give the problems of the Final Round of
the 1997 Iranian Mathematical Olympiad. Again, thanks go to Mohammed
Aassila, Strasbourg, France.

1997 IRANIAN MATHEMATICAL OLYMPIAD

Final Round
Time: 2  4 hours

1. Let n be a positive integer. Prove that there exist polynomials f (x)

and g (x) with integer coecients such that,
f (x)(x + 1)2n + g(x)(x2n + 1) = 2 .
2. Suppose that f : R ! R has the following properties:
(a) 8x 2 R, f (x)  1
(b) 8x 2 R, f x + 42
; 13
+ f (x) = f ;x + 1
+ f ;x + 1
.
6 7
Prove that f is periodic; that is, there exists a non-zero real number T such
that for every real number x, we have f (x + T ) = f (x).
3. Suppose that w1, w2, : : : , wk are distinct real numbers with a non-
zero
Pk sum. Prove that there exist integer numbers n1, n2 , : : : , nk such that
i=1Pniwi > 0 and for any non-identity permutation  on f1, 2, : : : , kg we
have ki=1 ni w(i) < 0.
4. Suppose that P is a variable point on arc BC of the circumcircle of
triangle ABC , and let I1 , I2 be the incentre of the triangles PAB and PAC ,
respectively. Prove that,
(a) The circumcircle of PI1I2 passes through a xed point.
(b) The circle with diameter I1 I2 passes through a xed point.
(c) The mid-point of I1 I2 lies on a xed circle.
5. Suppose that f : R+ ! R+ is a decreasing continuous function that
ful ls the following condition for all x, y 2 R+ :
f (x + y) + f (f (x) + f (y)) = f ;f (x + f (y)) + f (y + f (x))
.
Prove that f (x) = f ;1 (x).
 235

6. A one story building consists of a nite number of rooms which have

been separated by walls. There are one or more doors on some of these walls
which can be used to travel in this building. Two of the rooms are marked
by S and E . An individual begins walking from S and wants to reach to E .
By a program P = (Pi)i2I , we mean a sequence of R's and L's. The
individual uses the program as follows: after passing through the nth door,
he chooses the rightmost or the leftmost door, meaning that Pn is R or L.
For the rooms with one door, any symbol means selecting the door that he
had just passed. Notice that the person stops as soon as he reaches E .
Prove that there exists a program P (possibly in nite) with the property
that no matter how the structure of the building is, one can reach from S to
E by following it. [Editor's comment: one has to assume that there is a way
of getting from any room to any other room.]

Now we give Klamkin's solutions to the ve Quickies given at the

beginning of this Corner.
SOLUTIONS TO FIVE MORE KLAMKIN QUICKIES
1. S  4(a2 +2b22 + c22+2 d2) ;2 (2a2b22c2 2+ b2c2d2 + c2d2a2 + d2 a2b2)
;(a b + b c + c d + d a ).
Since the expression on the right hand side is linear in a2 , b2, c2 ,
and d2 , it takes on its maximum at the endpoints 0, 1 for each variable.
By inspection, Smax = 9 and is taken on for a = b = c = 1 and d = 0.
2. Neither inequality is valid.
(i) Just consider the case: b = 2, c = 5, d = 1 and a is very large.
(ii) Just consider the case: a = 2, b = 1, c = 8 and d is very small.
3. Consider the ane transformation x0 = ax , y0 = yb , z0 = zc which
takes the ellipsoid into a sphere. Under this transformation, lines go to lines,
planes go to planes, and tangency is preserved. Consequently, any envelop-
ing cone of the ellipsoid goes into an enveloping cone of the sphere and which
by symmetry is a right circular one and its points of tangency are a circle
(coplanar) of the sphere. Thus, P can be any exterior point of the ellipsoid.
4. Just take any 777 distinct positive integers and multiply each one
by the product of the sums of every 7 of them.
5. Assuming the better approximation has the form aR +b
cR+d where a, b,
c, d are rational, we must satisfy
aR + b ; p5 < jR ; p5j . (1)
cR + d
236

p
If R ! 5, the left hand side must ! 0. Thus, we must have acp5+
p
5+b = 5, p
d
so that d = a and b = 5c. Then substituting
p these values in (1) and dividing
both sides by the common factor jR ; 5j, we get
jcR + aj > jcp5 ; aj
and which can easily be satis ed by letting a = 2 and c = 1. Finally, our
better approximation is 2RR+2
+5 .

Before turning to readers' solutions, an apology. Somehow, my

ling system misplaced a group of solutions sent in by Miguel Amengual
Covas, Cala Figuera, Mallorca, Spain. He sent in solutions to problems 1, 3 of
the XXXIX Republic Competition of Mathematics in Macedonia
[1999 : 196, 2001 : 99], [1999 : 196, 2001 : 101], to problem 1, 2 of Class III
[1999 : 197, 2001 : 105{106], and to problem 2 of Class IV [1999 : 197, 2001
: 107{108].

Next we turn to readers' solutions to problems from the September

1999 Corner, and the Georg Mohr Konkurrencen I Matematik 1996 [1999 :
261{262].

GEORG MOHR KONKURRENCEN I

MATEMATIK 1996
January 11, 9{13

Only writing and drawing materials are allowed.

1. \C in 4ABC is a right angle and the legs BC and AC are both
of length 1. For an arbitrary point P on the leg BC construct points Q,
respectively, R, on the hypotenuse, respectively, on the other leg, such that
PQ is parallel to AC and QR is parallel to BC . This divides the triangle
into three parts.
Determine positions of the point P on BC such that the rectangular
part has greater area than each of the other two parts.
 237

Solution by Pierre Bornsztein, Pontoise, France.

A
x
R Q

1;x

C x P 1;x B
Denote x = CP , x 2 (0; 1). Then PB = 1 ; x, and, from Thales'
Theorem,
p p
QB = (1 ; x) 2 , QA = x 2 , RC = 1 ; x; AR = x .
Thus,
[RQPC ] = x(1 ; x) , [PBQ] = 12 (1 ; x)2 , [AQR] = 12 x2 .
It remains to solve (
x(1 ; x) > 12 (1 ; x)2 ,
x(1 ; x) > 12 x2 ,
which is equivalent to (
x > 13 ,
x < 23 .
Thus, we will have the desired result if and only if PC = x with x 2 13 ; 23 .
;

2. Determine all triples (x; y;z), satisfying

xy = z , (1)
xz = y , (2)
yz = x . (3)
Solution by Pierre Bornsztein, Pontoise, France.
If (x; y;z ) is a solution, then, multiplying, we have (xyz )2 = xyz .
Thus, xyz = 0 or xyz = 1.
Case 1. If xyz = 0 then, for example, z = 0.
From (2) we get y = 0, and from (3) we obtain x = 0. Conversely, it is
easy to see that (0; 0; 0) is a solution.
238

Case 2. If xyz = 1 then z = xy1 .

From (1), we deduce z 2 = 1. Thus, z 2 f;1, 1g.
In the same way, x 2 f;1, 1g and y 2 f;1, 1g. Moreover, since
xyz = 1, the number of ;1's in (x; y; z) is even. This leads to
(;1; ;1; 1) , (;1; 1; ;1) , (1; ;1; ;1) , (1; 1; 1) .
Conversely, it is easy to see that these triples are solutions. Then,
S = f(0; 0; 0), (1; 1; 1), (;1; ;1; 1), (;1; 1; ;1), (1; ;1; ;1)g .
3. This year's idea for a gift is from \BabyMath", namely a series of
9 coloured plastic gures of decreasing sizes, alternating cube, sphere, cube,
sphere, etc. Each gure may be opened and the succeeding one may be placed
inside, tting exactly. The largest and the smallest gures are both cubes.
Determine the ratio between their side-lengths.
Solution by Pierre Bornsztein, Pontoise, France.

a
2

R
d a
2
a
2
If a sphere with radius R is circumscribed to a cube with edge a then
the sphere and the cube have the same centre, and the vertices of the cube
are points of the sphere.
From Pythagoras' Theorem:
R2 = d2 + a4 = 34a .
2 2

Thus,
p3
R = 2 a. (1)
 239

If a sphere with radius R is inscribed in a cube with edge b, then the

sphere and the cube have the same centre, and the centres of the sides of the
cube are points of the sphere. Then
R = 2b . (2)
From (1) and (2), it follows that the ratio between the side-lengths of
the \outside cube" and the \inside cube" is
b = p3 .
a
Since there are 5 cubes, the ratios between the side-lengths of the largest
p 4
and the smallest gures is 3 = 9.
4. n is a positive integer. It is known that the last but one digit in the
decimal expression of n2 is 7. What is the last digit?
Solutions by Michel Bataille, Rouen, France; and by Pierre Bornsztein,
Pontoise, France. We give Bataille's solution.
We prove that this last digit is 6.
Write n as 10a + b where a is a non-negative integer and b 2 f0, 1, 2,
3, 4, 5, 6, 7, 8, 9g. Then n2 = 100a2 + 20ab + b2 and the last two digits of
n2 are also those of 20ab + b2.
But 20ab = 10  2ab ends with the digits 00 or 20 or 40 or 60 or 80.
Since it is given that the penultimate digit of 20ab + b2 is 7, we see that the
penultimate digit of b2 must be odd; this can occur only when b = 4 or b = 6,
and then b2 = 16 or b2 = 36. Adding these two values to any of the integers
00, 20, 40, 60, 80, we obtain only one result whose last but one digit is 7,
and it is 76. This completes the proof.
5. In a ballroom 7 gentlemen, A, B, C , D, E, F and G are sitting
opposite 7 ladies a, b, c, d, e, f and g in arbitrary order. When the gentle-
men walk across the dance oor to ask each of their ladies for a dance, one
observes that at least two gentlemen walk distances of equal length. Is that
always the case?
The gure shows an example. In this example Bb = Ee and Dd = Cc.
A
s
B
s
C
s
D s
E
s
F
s
G
s

s s s s s s s

f d b g c e a
240

Solutions by Pierre Bornsztein, Pontoise, France; and by Edward T.H.

Wang, Wilfrid Laurier University, Waterloo, Ontario. We give Wang's solu-
tion.
Note rst that the observation is only correct if we assume that the 7
gentlemen are \evenly" spaced.
We show that in general, if there are n gentlemen and n ladies, then
the same conclusion holds when n  2, 3 (mod 4).
Suppose the n ladies are situated at (k; 0) and the n gentlemen, at
(k; 1); k = 1, 2, : : : , n. Suppose that the gentleman at (k; 1) walks a dis-
tance of dk to the lady at (ak ; 0), ak 2 f1, 2, : :P: ,nng. Then (a1 ; a2 ; : : : ; an )
is a permutation of (1; 2; : : : ; n) and thus, k=1 (ak ; k) = 0. Since
dk = (1 + (ak ; k)2)1=2, we have d2k = 1 + (ak ; k)2. We show that
the values of the dk 's cannot all be distinct.
Note that ak ;k 2 f0, 1, 2, : : : , (n;1)g. Suppose to the contrary
that (ak ; k)2 6= (aj ; j )2 for all j 6= k, j , k = 1, 2, : : : , n. Then we have
fjak ; kj : k = 1, 2, : : : , ng = f0, 1, 2, : : : , n ; 1g and thus,
n ;1
nX
i = n(n2; 1) .
X
jak ; kj = (1)
k=1 k=0
On the other hand, since jtj ; t must be even for all integers t, we have
n
X n
X
jak ; kj = 2d + (ak ; k) = 2d , for some integer d . (2)
k=1 k=1
Comparing (1) and (2) we get n(n ; 1) = 4d which implies that
n  0, 1 (mod 4). Therefore, if n  2, 3 (mod 4) then we must have
dj = dk for some j 6= k.
Editor's question. If n is congruent to either 0 or 1, is it always possible
to arrange the n gentlemen and the n ladies in a way such that the distance
are all di erent?

Next we turn to solutions to problems of the St. Petersburg City

Mathematical Olympiad, Third Round, 1996 [1999 : 262].
ST. PETERSBURG CITY MATHEMATICAL OLYMPIAD
Third Round { February 25, 1996
11th Grade (Time: 4 hours)
1. Serge was solving the equation f (19x ; 96=x) = 0 and found 11
di erent solutions. Prove that if he tried hard he would be able to nd at
least one more solution.
 241

Solutions by Mohammed Aassila, Strasbourg, France; by Michel

Bataille, Rouen, France; by Pierre Bornsztein, Pontoise, France; and by
Edward T.H. Wang, Wilfrid Laurier University, Waterloo, Ontario. We give
Wang's write-up.
Note rst that x = 0 is not a solution. If r 6= 0 is a solution, then so is
t = ; 1996r , since
   
96 96
f 19t ; t = f ; r + 19r = 0 .
Since r = ;1996r is impossible, di erent r's will correspond to di erent t's.
Therefore, the number of solutions must be even (if it is nite) and the con-
clusion easily follows.
2. The numbers 1, 2, : : : , 2n are divided into two groups of n numbers.
Prove that pairwise sums of numbers in each group (sums of the form a + a
included) have identical sets of remainders on division by 2n.
Solution by Pierre Bornsztein, Pontoise, France.
Let G1 , G2 be the two groups, and let S1 = fp 2 f1, 2, : : : , 2ng :
there exist a, b 2 G1 for which a + b  p (mod 2n)g. S2 is de ned in the
same way. We must prove that S1 = S2.
Let p 2 f1, : : : , 2ng.
We will have a + b  p (mod 2n), with a, b 2 f1, : : : , 2ng, if and only
if a + b = p or a + b = 2n + p.
Case 1. If p is odd, then p = 2k + 1 for some integer k such that
0  k < n.
Let a, b 2 f1, : : : , 2ng, with a  b.
We will have a + b = p if and only if (a; b) is one of the k pairs
(1; 2k) , (2; 2k ; 1) , : : : , (k; k + 1) .
We will have a + b = 2n + p if and only if (a; b) is one of the n ; k
pairs
(2k + 1; 2n) , (2k + 2; 2n ; 1) , : : : , (n + k; n + k + 1) .
Then the numbers 1, 2, : : : , 2n are divided into n pairs to give the
remainder p.
If p 2 S1 , then G1 contains at least one of these pairs. We have to
choose at most n ; 2 numbers to complete the group G1. It cannot be done
if we want G1 to contain at least one of the members of each pair. Then at
least one pair is included in G2 . Thus, p 2 S2 .
242

Case 2. If p is even, then p = 2k for some integer k such that

1  k  n.
As above the numbers 1, 2, : : : , 2n are divided into n + 1 pairs to give
the remainder p. The pairs are
(1; 2k ; 1) , (2; 2k ; 2) , : : : , (k ; 1; k + 1) , (k;k)
and
(2k; 2n) , (2k +1; 2n ; 1) , : : : , (n + k ; 1;n + k +1) , (n + k;n + k) .
Then we have n ; 1 pairs and two \isolated" numbers (k and n + k).
If p 2 S1 and if at least one of the two \isolated" numbers is not in G1,
then k 2 G2 or n + k 2 G2 . Thus, p 2 S2 .
If p 2 S1 and if both of the two \isolated" numbers are in G1, then G1
contains n ; 2 other numbers. Thus, G1 cannot contain at least one of the
numbers from each of the n ; 1 pairs.
It follows that at least one of the pairs is included in G2 . So, we deduce
that p 2 S2 .
Then, in each case, if p 2 S1 then p 2 S2 . That is,
S1  S2 .
By symmetry, we have S2  S1 . Thus,
S1 = S2 .
3. No three diagonals of a convex 1996{gon meet in one point. Prove
that the number of the triangles lying in the interior of the 1996{gon and
having sides on its diagonals is divisible by 11.
Solutions by Mohammed Aassila, Strasbourg, France; and by Pierre
Bornsztein, Pontoise, France. We give the solution of Aassila.
If ABCDEF is a 6{gon such that no three diagonals meet in one point,
then the triangle formed by AD, BE and CF is the only one having sides ; on


its diagonals. Hence, the number of such triangles for a 1996{gon is 1996 6 ,
but since 1991 = 11  181, we deduce the desired result.
4. Points A0 and C 00 are taken on the diagonal BD of0 a parallelogram
ABCD so that AA 0 kCC . Point K lies on the segment A C , the line AK
meets the line C 0 C at the point L. A line parallel to BC is drawn through
K , and a line parallel to BD is drawn through C . These two lines meet at
point M . Prove that the points D, M , L are collinear.
 243

Solutionsby Mohammed Aassila, Strasbourg, France; by Michel Bataille,

Rouen, France; by Pierre Bornsztein, Pontoise, France; and by ToshioSeimiya,
Kawasaki, Japan. We give the solution of Seimiya.
A D

C0

A0 K M
B C

Let O be the intersection of AC and BD. Since ABCD is a parallelo-

gram we have AO = OC .
Since AA0 kCC 0 , we obtain A0 O : OC 0 = AO : OC = 1 : 1, so that
A0 O = OC 0. Thus, AA0 CC 0 is a parallelogram, and further, AC 0 kA0C .
Since ADkKM , C 0 DkCM , AC 0 kKC , and AC 0 6= KC , we have AK , C 0 C ,
and DM are concurrent at L.
Therefore, D, M , L are collinear.

That completes this issue of the Corner. We will continue with these
problems in the next number. Please send me Olympiad Contests and your
nice solutions and generalizations.
244

BOOK REVIEWS
ALAN LAW
Mathematical Olympiads: Problems and Solutions from Around the World
1998{1999,
edited by Titu Andreescu and Zuming Feng,
published by the Mathematical Association of America, 1999,
ISBN 0-88385-803-7, softcover, 280 pages, $28.50 (U.S.).
Reviewed by Christopher G. Small, University of Waterloo, Waterloo,
Ontario.
This book represents a continuation of the book Mathematical Con-
tests 1997-1998: Olympiad Problems and Solutions from around the World.
The authors have collected olympiad problems from the national contests of
18 di erent countries (including Canada), together with seven regional con-
tests from 1998 and the national contests of 22 countries and eight regional
contests from 1999. Problems from 1998 are published with solutions, but
solutions for 1999 problems are not included. The volume comes with a
brief glossary of basic mathematical identities and de nitions, and an index
of problems classi ed lexicographically by subject area, country of origin, and
year. Both problems and solutions are presented with a uni ed notation.
There are many reasons why people involved in mathematics contests
should want a book of this kind. Having schmoozed at the IMO several times,
I have found that the task of gathering national contest problems is not easy.
Hauling my mathematical loot back home, I am forced to contemplate my
linguistic inadequacies. (Perhaps you know that \cung" is Vietnamese for
\arc," but I did not. Unfortunately, more or less knowing | as I do now
| that \cung" means \arc" provides no assistance with the host of other
words I need to know.) Fortunately, many countries translate their prob-
lems into English, as they know full well that their English is better than our
Vietnamese/Farsi/whatever. However, the translation process is often im-
perfect. Consider the following recent problem translated from a contest in
a Spanish-speaking country:
A sequence is de ned as a1 = 3, y an+1 = an + a2n . Determine
the last two digits of a2000.
A student could be forgiven for asking what the value of y is. Occasionally,
the translation into English su ers the dubious honour of being too good.
For example, another recent contest used the English word \wherefrom,"
which is perfectly good English, but should probably be avoided in contest
problems in our postliterate world.
On reading through the problem collections and the various addenda,
I had only one criticism. The authors have chosen to continue the common
 245

tendency to describe convex functions as \concave up" and concave functions

as \concave down." If it is unnecessary for us to use such terms in research
papers, it is surely also unnecessary in the school system.
Anyway, hats o to Titu Andreescu and Zuming Feng for putting to-
gether this interesting collection of the best from around the world. The
problems are written in plain and simple English without any of the transla-
tion e ects that I have mentioned above. From each volume of problems we
can celebrate the vigour of mathematical culture in many di erent countries.
Now where's that Russian problem I was working on : : : ?

Another Maze in Three Dimensions

Izador Hafner
Here is another maze, this time, on a dodecahedron, given as an unfolded
plane plan. Can you solve it?

Izidor Hafner
Faculty of Electrical Engineering
University of Ljubljana
Trzaska 25
1000 Ljubljana, Slovenia
246

Another Look at the Volume of a

Tetrahedron
Nairi M. Sedrakyan

In November 1999's CRUX with MAYHEM [1999 : 422-425], Murat

Aygen proved that for the tetrahedron ABCD with
a = BC , b = CA, c = AB, a1 = AD, b1 = BD, c1 = CD,
the volume V and the circumradius R are related by
6V R = K ,
where K is the area of the associated triangle with sides aa1 , bb1, and cc1 .
Here is an alternative proof.
Solution. First, we will prove a lemma.
Lemma. Given a triangle DMN , choose on the rays DM and DN
points M1 and N1 such that DM1 =
1 and DN = 1 .
DM 1 DN
Then, we claim that
\DM1 N1 = \DNM , \DN1 M1 = \DMN (1)

and M1N1 = DMMN
DN .

Proof. Indeed, we have
DM1 = DN1 = 1 . From this, it follows
DN DM DM
DN
MN
that 4M1 DN1  4NDM , from which we nd 1 1 =
1
MN DM
DN . This
implies (1).
Draw a diameter DS of the sphere and consider points A1 , B1 , C1 , S1
contained by the rays DA, DB , DC and DS , correspondingly, where
DA1 = DA 1 , DB = 1 , DC = 1 , DS = 1 .
1 DB 1 DC 1 DS
According to the lemma, we have \DS1 A1 = \DAS = 900, \DS1 B1 =
\DBS = 900 and \DS1 C1 = \DCS = 900 . Hence, DS1 is the altitude of
the pyramid DA1 B1 C1 .
Copyright c 2001 Canadian Mathematical Society
 247

Let V1 denote the volume of the pyramid DA1 B1C1 . Denote the area
of the triangle XY Z by [XY Z ]. Denote by (X; MNK ), the distance of
the point X from the plane (MNK ). Now, according to the lemma, we have
A1B1 = DAAB = cc1 ,

DB a1 b1c1
B1C1 = a aab 1c , A1C1 = a bbb 1c .
1 1 1 1 1 1
Hence,
[A1 B1C1] = K a2b12 c2 ,
1 1 1
where K is the area of the triangle with sides aa1 , bb1, cc1 , and
V1 = 13 [A1B1C1]
DS1 = 61R
a2K . (2)
1 b21 c21
On the other hand,
V1 = 31 [DB1C1](A1; DB1C1) = DB1
DC1
DA1 = 1 . (3)
V 1 [DBC ](A; DBC )
3 DB
DC DA a21b21c21
From (2) and (3), we nd 6V R = K .
See [1] and [2] for other versions of the formula for a tetrahedron's
volume in terms of its edges. In [1], the authors traced their formula back to
Euler, and provided the references [3] pages 285-289, and [4] pages 124-125.
Reference [2] attributes the determinant version to Cayley and Menger, but
provides no reference. See also the discussion of Crux problem 930 [1985 :
162] for further references and related information.
References
1. Keven L. Dove and John S. Sumner, Tetrahedra with integer edges and
integer volume, Math. Mag. 65:2 (April, 1992) 104-111.
2. Ira Rosenholtz, Problem Q890, Math. Mag. 72:2 (April, 1999) 149.
3. H. Dorrie, 100 Great Problems of Elementary Mathematics, Dover, New
York, 1965.
4. D.M.Y. Sommerville, An Introduction to the Geometry of N {Dimen-
sions, Dover, New York, 1958.
Nairi M. Sedrakyan
A. Shahinyan High School of Physics and Mathematics,
Azatutyan av, 2-nd side street, 9
375037, Yerevan, Armenia
[email protected]
248

THE SKOLIAD CORNER

No. 54
R.E. Woodrow
Starting with the September number of Crux Mathematicorum with
Mathematical Mayhem, the Skoliad Corner will be moving to the care of
Shawn Godin. He has some excellent ideas for involving school-age students
and their teachers more fully.
To close my role with the Skoliad, we give the problems and solutions
of the rst round of the Alberta High School Mathematics Competition Prize
Examination written in November, 2000.

ALBERTA HIGH SCHOOL MATHEMATICS

COMPETITION 2000-2001
Part I
November 20, 2000

1. Amy has 58 coins totalling one dollar. They are all pennies, nickels
and dimes. The number of nickels she has is
(a) 2 (b) 3 (c) 4 (d) 5 (e) 6
2. Nima runs for 5 kilometres at 10 kilometres per hour, followed by
10 kilometres at 5 kilometres per hour. Her average speed in kilometres per
hour for the whole trip is
(a) 6 (b) 6:5 (c) 7 (d) 7:5 (e) 8
3. The sum of the digits of the number 2200052004 is
(a) 8 (b) 9 (c) 11 (d) 13 (e) 14
4. A bug crawls 1 centimetre north, 2 centimetres west, 3 centimetres
south, 4 centimetres east, 5 centimetres north and so on, at 1 centimetre per
second. Each segment is 1 centimetre longer than the preceding one, and at
the end of a segment, the bug makes a left turn. The direction in which the
bug is facing 1 minute after the start is
(a) north (b) west (c) south (d) east (e) changing
5. The average of fteen di erent positive integers is 13. The maximum
value for the second largest of these integers is
(a) 28 (b) 46 (c) 52 (d) 90 (e) none of these
 249

6. A polynomial p(x) has four terms, so that it is of the form

p(x) = axk + bx` + cxm + dxn , where k, `, m and n are four di erent
non-negative integers and a, b, c and d are non-zero real numbers. The
largest number of terms the polynomial (p(x))2 can have is
(a) 6 (b) 8 (c) 10 (d) 12 (e) 16
7. If x2 ; x ; 1 = 0, then the value of x3 ; 2x + 1 is
p p
(a) 1;2 5 (b) 0 (c) 1+2 5 (d) 2 (e) not uniquely determined
8. A thin rod 6 metres long has one end on the level ground and is lean-
ing against a sphere on the ground, 4 metres in diameter. The rod touches
the sphere at a point 1 metre from the top end of the rod. The distance, in
metres, between the bottom of the sphere and the bottom of the rod is
(a) 3
p
(b) 2 5 (c) 5 (d) 6 (e) none of these
9. David went to a casino with $64 and played a certain game six times.
Each time, he bet half the amount of money he had at the time. The chance
of winning this game was 21 , and David won three of the six games, though
it was not recorded which three they were. In each of those three games, he
won an amount of money equal to his bet. After these six games, the amount
of money David had was
dependent on
(a) $27 (b) $32 (c) $64 (d) $128 (e) which games
David lost
10. The number of integers n which satisfy (n ; n ; 1)n+2 = 1 is
2
(a) 1 (b) 2 (c) 3 (d) 4 (e) 5
11. In triangle ABC , \BCA = 90. Points E and F lie on the
hypotenuse AB such that AE = AC and BF = BC . If \ECF = x, then
(a) x = 30 (b) 30 < x < 45 (c) x = 45

(d) 45 < x < 60  (e) 60 
12. The integer part of a number is the largest integer not exceeding
the number, and the fractional part of a number is the di erence between it
and its integer part. For instance, the integer part of 53 is 1 and the fractional
part is 32 . Suppose that the product of the integer parts of two positive ratio-
nal numbers is 5, and the product of their fractional parts is 14 . Their product
is
(a) 54 (b) 214 (c) 132 (d) 334 (e) not uniquely determined
13. The sum of the positive divisors of 120 is 360. The sum of the
reciprocals of the positive divisors of 120 is
(a) 2 (b) 3 (c) 4 (d) 6 (e) not uniquely determined
250

14. a ; b b ; c c ; a (a ; b)(b ; c)(c ; a)

The expression
a + b + b + c + c + a + (a + b)(b + c)(c + a)
simpli es to
(a) 1 (b) a + b + c (c) a+1b+c (d) 2((aa+;bb)()(bb+;cc)()(cc+;aa)) (e) none of these
15. A certain positive integer has m digits when written in base 3 and
m + 1 digits when written in base 2. The largest possible value of m is
(a) 1 (b) 2 (c) 3 (d) 4 (e) 5
16. ABCD is a square. A line through B intersects the extension of
CD at E, the side AD at F and the diagonal AC at G. If BG = 3 and
GF = 1, then the length of FE is
(a) 4 (b) 6 (c) 8 (d) 10 (e) 12

Now for the solutions. Hope you did not peek!

ALBERTA HIGH SCHOOL MATHEMATICS
COMPETITION 2000-2001
Solutions to Part I
1. (e) The number of pennies Amy has must be a multiple of 5. If she
has 55 pennies, then the maximum value of the other 3 coins is 30 cents, and
the total is at most 85 cents. If she has 45 pennies, then the minimum value
of the other 13 coins is 65 cents, and the total is at least 120 cents. If she
has fewer pennies, the total will be higher. Hence, she has 50 pennies. If
the other 8 coins are all nickels, the total will be 10 cents short. Hence, two
of the nickels must be changed into dimes.
2. (a) Nima covers 15 kilometres in 2:5 hours.
3. (d) The number is equal to 102000 times 54 = 625.
4. (c) Since 1 + 2 +


+ 10 = 55 and 1 + 2 +


+ 11 = 66, the bug
is on its eleventh segment 1 minute after the start.
5. (e) The minimum sum of the smallest thirteen of the numbers is
1 + 2 +


+ 13 = 91. Hence, the maximum sum of the largest two is
13
15 ; 91 = 104. It follows that the maximum value of the second largest
number is 51.
6. (c) When expanded, (p(x))2 becomes
a2 x2k + b2kx+2`` + c2x2km++ d2x2n k+n
+2abx + 2acx + 2adx + 2bcx`+m + 2bdx`+n + 2cdxm+n .
m
By choosing k = 1, ` = 10, m = 100 and n = 1000, all the terms in the
above expansion are distinct.
 251

72. (d) Long division3 yields x3 ; 2x + 1 = (x2 ; x ; 1)(x + 1) + 2.

Since x ; x ; 1 = 0, x ; 2x + 1 = 2.
8. (c) In the vertical plane containing the rod and the centre of the
sphere, the cross-section of the sphere is a circle. From the bottom of the
rod, two equal tangents can be drawn to this circle, one running along the
rod and the other connecting the bottom of the sphere and the bottom of the
rod.
9. (a) After each game in which he lost, David had half as much money
as he had before. After each game in which he won, David had one and a
half times as much money as he had before. No matter which three games
he won, he would have $64( 21 )3 ( 23 )3 = $27 at the end.
10. (d) There are three cases. First, n + 2 = 0, while n2 ; n ; 1 6= 0.
This yields n = ;2. Second, n ; n ; 1 = ;1 while n + 2 is a non-zero
2
even integer. This yields n = 0. Finally, n2 ; n ; 1 = 1 while n + 2 is any
non-zero integer. This yields n = ;1 and n = 2.
11. (c) Let \ACF = y and \BCE = z, as illustrated in the diagram.
Then x + y + z = 90 . Since AC = AE , \AEC = \FEC = x + y .
Similarly, \EFC = x + z . It follows that x + (x + y ) + (x + z ) = 180, so
that 2x = 90 .
A
F

E
y
xz
C B
12. (e) The integer parts of these two rational numbers must be 5 and
1, but there are many possible choices for their fractional parts, such as 31
and 34 . Now (5 + 13 )(1 + 43 ) = 283 , while (5 + 43 )(1 + 13 ) = 233 .
13. (b) When the reciprocal of a divisor of 120 is multiplied by 120, the
product is also a divisor of 120. When the sum of the reciprocals of all the
divisors of 120 is multiplied by 120, the product is the sum of all the divisors
of 120, which is given to be 360.
14. (e) We have aa +; bb + bb +; cc = (a2+b(ba)(;b +c) c) , while
c ; a 1 + (a ; b)(b ; c)  = c ; a
2b(c + a) = 2b(c ; a) .
c+a (a + b)(b + c) c + a (a + b)(b + c) (a + b)(b + c)
Hence, the original expression simpli es to 0.
252

15. (d) The chart shows the numbers of digits of a positive integer n
in bases 2 and 3. Since the number of digits in base 2 increases every second
integer, while that in base 3 every third, the largest number with one digit
more in base 2 than in base 3 is 9.
n 1 2 3 4 5 6 7 8 9 10 11
Base 2 1 2 2 3 3 4 4 5 5 6 6
Base 3 1 1 2 2 2 3 3 3 4 4 4
16. (c) In the diagram, triangles BCG and FAG are similar, as are
triangles ABF and DEF .
EF DF BC ; AF BG
Hence,
BF = AF = AF = FG ; 1 = 2.
B A
G
F

C D E

I will be forwarding my Skoliad les to Shawn. Please send him your

school contest materials and ideas.
 253

MATHEMATICAL MAYHEM
Mathematical Mayhem began in 1988 as a Mathematical Journal for and by
High School and University Students. It continues, with the same emphasis,
as an integral part of Crux Mathematicorum with Mathematical Mayhem.
All material intended for inclusion in this section should be sent to
Mathematical Mayhem, Cairine Wilson Secondary School, 975 Orleans Blvd.,
Gloucester, Ontario, Canada. K1C 2Z5 (NEW!). The electronic address is
NEW! [email protected] NEW!
The Assistant Mayhem Editor is Chris Cappadocia (University of Wa-
terloo). The rest of the sta consists of Adrian Chan (Harvard University),
Jimmy Chui (University of Toronto), Donny Cheung (University of Waterloo),
and David Savitt (Harvard University).

Polya's Paragon
Shawn Godin
Sometimes a very simple observation can go a long way. Let us start with a
problem that, hopefully by the end of the article, you will be set to solve.
Problem. Given any 52 integers, show that there exist two of them
whose sum, or di erence, is divisible by 100.
If you have not encountered a problem like this before, chances are you
have no idea of how to start. Worry not! The Dirichlet Box Principle (or as it
is more colourfully called, the Pigeonhole Principle) will give us the necessary
method.
The Pigeonhole Principle. If you have n pigeonholes and n + 1 pigeons
to go into them, then one of the pigeonholes must contain at least 2 pigeons.
The idea behind the Principle is simple and obvious, but very powerful.
If we set up our pigeonholes and pigeons carefully we can go a long way.
Example 1. Show that if 13 people are gathered together, at least two
are born on the same month.
Solution. Here the months are our pigeonholes and the people are the
pigeons, and the result follows directly from the Principle.
Example 2. Show that, of any 5 points chosen inside an equilateral
triangle with side 2, there are two whose distance apart is at most 1 unit.
Solution. Here we can break the original triangle up into 4 smaller
equilateral triangles of side length 1 by joining the mid-points of the sides
254

of the original triangle (see gure). Now the 4 triangles are our pigeonholes
and the points are the pigeons. So at least two of the points are in the same
triangle. Since the furthest apart that these two points can be is 1 unit (if the
points are at the vertices), the result follows.

Example 3. In a class of 25 students, show that there is a group of at

least 4 who were born on the same day of the week.
Solution. To solve this we need the generalized version of the Principle.
The Pigeonhole Principle. If you have n pigeonholes and mn+1 pigeons
to go into them, then one of the pigeonholes must contain at least m + 1
pigeons.
Again, the idea is simple but powerful. Now if we use the days of the
week as our 7 pigeonholes, then we have 25 > 3  7 + 1 students (pigeons),
and our result follows again.
Now, hopefully, you are ready to attack the original problem. Choose
your pigeonholes wisely! Below are a couple of other problems to try.
1. Prove that if any number of people are gathered together, at least 2
people know the same number of people. (We can assume that if A knows
B, then B knows A.)
2. Prove that, of any 5 points chosen within a square of side length 1,
there are two whose distance apart is at most p12 .
3. Prove that, in any group of 6 people, there is a group of 3 who know
each other or a group of 3 who are mutual strangers. (We can assume that if
A knows B, then B knows A.)
4. Seventeen people correspond by mail with one another | each one
with all the rest. In their letters, only three di erent topics are discussed.
Each pair of correspondents deals with only one of these topics. Prove that
there are at least three people who write to each other about the same topic.
(IMO 1964)
 255

Problem of the Month

Jimmy Chui, student, University of Toronto
Problem. A pair of telephone poles d metres apart is supported by two cables
which run from the top of each pole to the bottom of the other. The poles
are 4 m and 6 m tall. Determine the height above the ground of the point, T ,
where the two cables intersect. What happens to this height as d increases?
(1997 Euclid, Problem 7b)
S

Q
T

P O R

Solution. Let the height of the poles be a and b. Let the distance PO
be c. Let h be the height above ground of the point, T .
Since 4QPR and 4TOR are similar, a : d = h : (d ; c), or equiva-
lently, d ; c = dha.
Since 4SRP and 4TOP are similar, b : d = h : c. This may be
written as c = dhb .
;

Adding these equations (eliminating c), we have d = dh a1 + 1b .
Solving for h yields h = aab+b .
Hence, if a = 4 and b = 6, then h = 125 (all values in metres).
Thus, the height of T is 125 metres, regardless of d.
Note: The fact that h does not change when d changes does make sense
intuitively. View the collection of lines as two rigid poles and three elastic
strings, and imagine pulling outward on the two poles (and increasing d in
the process). The horizontal measurements change, but the vertical mea-
surements stay constant, due to the rigidness of the two poles.
256

Mayhem Problems
The Mayhem Problems editors are:
Adrian Chan Mayhem High School Problems Editor,
Donny Cheung Mayhem Advanced Problems Editor,
David Savitt Mayhem Challenge Board Problems Editor.
Note that all correspondence should be sent to the appropriate editor |
see the relevant section. In this issue, you will nd only solutions | the
next issue will feature only problems.
We warmly welcome proposals for problems and solutions. With the
schedule of eight issues per year, we request that solutions from the previous
issue be submitted in time for issue 4 of 2002.

High School Solutions

Editor: Adrian Chan, 1195 Harvard Yard Mail Center, Cambridge, MA,
USA 02138-7501 <[email protected]>
H281. Correction. Proposed by Jose Luis Daz, Universitat
Politecnica de Catalunya, Terrassa, Spain.
Suppose the monic polynomial A(z ) = nk=0 ak z k can be factored into
P

(z ; z1 )(z ; z2)


(z ; zn ), where z1 , z2 , : : : , zn are positive real numbers

and n is an even number. Prove that a1 an;1  n2a0 .
H267. Comment by Edward T.H. Wang, Wilfrid Laurier University,
Waterloo, Ontario, and Ravi Vakil, MIT, Cambridge, MA, USA.
Without the additional note a, b, c are real numbers, you get an addi-
tional pretty solution, if a, b, c are vertices of an equilateral triangle centered
at the origin in the Argand plane.
Editors note: Two typos, introduced by the rookie editor, appear in
the original solution: x = ac++db should be x = ba++dc and the conclusion should
have t is a non-zero real number.
 257

H269. Find the lengths of the sides of a triangle with 20, 28, and 35
as the lengths of its altitudes.
Solution by Michel Bataille, Rouen, France; and Andrei Simion,
student, Brooklyn Technical High School, Brooklyn NY, USA.
Let a, b, c be the sides corresponding to the altitudes 20, 28, 35,
respectively, and let A be the area of the triangle.
Thus, 2A = 20a = 28b = 35c and we get, b = 57a and c = 47a .
If s is the semi-perimeter of the triangle, then s = 12 (a + b + c) = 87a .
By Heron's Formula, we have
p
s
8a   a   3a   4a  = 4a2 6 .
p
A = s(s ; a)(s ; b)(s ; c) = 7 7 7 7 49
But, recall that A = 10a.
p p
245 6 , b = 175 6 , c = 35 6 .
p
Hence, a =
12 12 3
H270. Find all triangles ABC that satisfy
sin(A ; B) + sin(B ; C ) + sin(C ; A) = 0 .
Solution by Michel Bataille, Rouen, France; Andrei Simion, student,
Brooklyn Technical HS, Brooklyn NY, USA; and Edward T.H. Wang, Wilfrid
Laurier University, Waterloo, Ontario.
Using elementary trigonometric formulae, we get:

sin (A ; B)+ sin(B ; C) + sin (C ; A)

A
= 2 sin 2 cos; C A ; 2 B + C 
C ; A
+ 2 sin 2 cos 2
 
C ; A 

2
= 2 sin 2

A ; C 
cos

A ; 2 B + C  
C ; A 

2  + cos  2
= ;4 sin A ;2 C sin C ;2 B sin A ;2 B .
  

Hence, our given equation becomes,

sin A ;2 C sin C ;2 B sin A ;2 B = 0 .
     

From this, we can conclude that A = B , or B = C , or C = A. This

implies that ABC must be an isosceles (or equilateral) triangle.
Also solved by HOJOO LEE, student, Kwangwoon University, Kangwon-Do, South Korea.
258

H271. Proposed by Hojoo Lee, student, Kwangwoon University,

Kangwon-Do, South Korea.
Let bxc denote the greatest integer less than or equal to x.
Let n = b1=(a ; bac)c for some positive real number a.
Show that b(n + 1)ac  1 (mod (n + 1)).
Solution by Michel Bataille, Rouen, France; and Edward T.H. Wang,
Wilfrid Laurier University, Waterloo, Ontario.
For n to be well de ned, we assume that a is not an integer. From the
properties of the oor function, we get:
bac < a < bac + 1 and n  a;b1 ac < n + 1. Since a ; bac > 0, this
yields n(a ;bac)  1 < (n + 1)(a ; bac) and 1 + (n + 1)bac < (n + 1)a 
1 + nbac + a. Using the fact that a < bac + 1, we get 1 + (n + 1)bac <
(n + 1)a < 2 + (n + 1)bac.
It follows that b(n + 1)ac = 1 + (n + 1)bac, meaning that
b(n + 1)ac  1 (mod (n + 1)).
H272. Proposed by Hojoo Lee, student, Kwangwoon University,
Kangwon-Do, South Korea.
Let fa1 , a2 , a3 , : : : , an g be a set of real numbers. Let s = a1 + a2 +
X n X n



+ an . Show that (ak ; ai)(s ; ai)  0 .
i=1 k=1
Solution by Michel Bataille, Rouen, France; Vedula N. Murty, Dover,
PA, USA; Andrei Simion, Brooklyn Technical HS, Brooklyn NY, USA; and
Edward T.H. Wang, Wilfrid Laurier University, Waterloo, Ontario.
n X
X n n
X n
X
(ak ; ai)(s ; ai) = (s ; ai ) (ak ; ai )
i=1 k=1 i=1 k=1
Xn
= (s ; ai )(s ; nai)
i=1
Xn ;
= s2 ; (n + 1)(ai)(s) + n(ai)2

i=1
n
X
= n s2 ; (n + 1) s2 + n
;
;

(ai )2
i=1
!2
n
X n
X
= n (ai )2 ; ai
i=1 i=1
n ! n ! n !2
X X X
= 12 (ai )2 ; ai 0
i=1 i=1 i=1
by the Cauchy-Schwarz Inequality.
 259

Advanced Solutions
Editor: Donny Cheung, c/o Conrad Grebel College, University of
Waterloo, Waterloo, Ontario, Canada. N2L 3G6 <[email protected]>
A245. Show that a polygon with xed side lengths has maximal area
when it can be inscribed in a circle.
Solution.
Consider the polygon with the given xed side lengths which is inscribed
in a circle, and attach to each side the corresponding segment of the circum-
circle (the region bounded by the side and the arc of the circle between the
two endpoints of the side). Note that the area of the polygon is exactly the
area of the segments subtracted from the area of the circle.
Consider the polygon with its sides in a di erent con guration, but still
with the segments from before attached. The area of the polygon is still
at most the area bounded by the circular arcs minus the xed area of the
segments, but the area bounded by the circular arcs has the same perimeter
as the original circumscribing circle, and thus cannot have a larger area. This
means that the polygon's area is also bounded by the area of the original
polygon, inscribed in the circle.
A246. Proposed by Mohammed Aassila, CRM, Universite de
Montreal, Montreal, Quebec.
Given a triangle with angles A, B , C , circumradius R, and inradius r,
prove that
1 + Rr  sin A2 + sin B2 + sin C2  17 +
12 6R
r .
Solution by Vedula
P N.AMurty, Dover, PA, USA.
The inequality sin  1 + r is well known, and a simple proof
P 2 R ; B+C
; B ;C

of this is 2 P
cos A = (cos B + cos C ) = P
2 cos 2 cos 2 
P A P A
2 sin 2 (where each sum is cyclic). Thus, sin 2  cos A = 1 +
P

4 sin A2 sin B2 sin C2 = 1 + Rr .

For
P
sin A  17 + r , we use the known inequality 2 P sin B sin C 
2 12 6R 2 2
P
cos A and the equation P sin2 A2 = 21 P(1 ; cos A) = 32 ; 12 P cos A to
get
sin A + sin B + sin C  3 + 1 cos A .
 2 X
2 2 2 2 2
r
Using the formula cos A = 1 + R gives us the inequality
P

A
sin 2 + sin 2 + sin 2  2 + 2 1 + R = 2 + x2 ,
B C 2

3 1 
r 

where 0 < x = Rr  12 .
260
q
Thus, sin A2 + sin B2 + sin C2  2 + x2 .
;1)2 , so that 2 + x  ; 17 + 1 x
2 , and the result
Finally, 0  (2x144 2 12 6
follows.
Also solved by Kee-Wai Lau, Hong Kong, China.
A248.
(a) Prove that in every sequence of 79 consecutive positive integers written
in the decimal system, there is a positive integer whose sum of digits is
divisible by 13.
(1997 Baltic Way)
(b) Give a sequence of 78 consecutive positive integers each with a sum of
digits not divisible by 13.
Solution by Michel Bataille, Rouen, France.
(a) We will denote by [ak ak;1 : : : a1 a0 ] the integer a0 + 10a1 +


+
10k;1ak;1 + 10kak . We will say that such an integer is good when
ak + ak;1 +


+ a1 + a0 is divisible by 13.
Consider a list L of ten consecutive integers of the form:
[ak ak;1


a10] ,[ak ak;1


a1 1] , : : : ,[ak ak;1


a1 9] .
If the sum ak + ak;1 +


+ a1 is congruent modulo 13 to one of the
integers 13, 12, : : : , 5, 4, then clearly L contains a good integer.
Now, we observe that a sequence of 79 consecutive positive integers
necessarily includes 7 lists such as L; moreover, there must be 4 con-
secutive lists out of these 7 which are between two integers of the form
[bkbk;1


b200] and [bkbk;1


b299], say:
L1 [bkbk;1


b2b10], : : : , [bk bk;1


b2b19]
L2 [bkbk;1


b2(b1 + 1)0], : : : , [bkbk;1


b2(b1 + 1)9]
L3 [bkbk;1


b2(b1 + 2)0], : : : , [bkbk;1


b2(b1 + 2)9]
L4 [bkbk;1


b2(b1 + 3)0], : : : , [bkbk;1


b2(b1 + 3)9]
where b1 2 f0, 1, 2, 3, 4, 5, 6g.
Then, either L1 contains a good integer, or bk + bb;1 +


+ b2 + b1  3,
2, or 1 (mod 13), in which case L2 , L3 , or L4 contains a good integer,
respectively.
(b) When there are only 78 integers, we have only 6 lists such as L with 3 on
either side of a multiple of 100 in order for the above to fail. For exam-
ple, the 78 positive integers from 9; 999; 999; 961 to 10; 000; 000; 038
inclusively do not have a sum of digits divisible by 13.
 261

Challenge Board Solutions

Editor: David Savitt, Department of Mathematics, Harvard University,
1 Oxford Street, Cambridge, MA, USA 02138 <[email protected]>
C93. Let H be a subset of the positive integers with the property that
if x, y 2 H then x + y 2 H . De ne the gap sequence GH of H to be the
set of positive integers not contained in H .
(a) If GH is a nite set, prove that the arithmetic mean of the integers
in GH is less than or equal to the number of elements in GH .
(b) Determine all sets H for which equality holds in part (a).
Solution. We remark, before we begin, that a subset of the positive
integers with the property that if x, y 2 H then x + y 2 H is called a
semigroup.
(a) We proceed by induction on the size g of GH . Since GH is non-
empty if and only if 1 2 GH , the case g = 1 is clear. For larger g , the key
observation is:
Lemma. The largest element of GH is at most 2g ; 1.
Proof: For an integer n  2g , consider the pairs of integers
(1; n ; 1); : : : ; (g;n ; g). If n 2 GH , then H has exactly g ; 1 other gaps,
and so for at least one of the pairs of integers (1;n ; 1), : : : , (g;n ; g ),
both integers in the pair are contained in H . Since n is the sum of this pair
of integers, n is in H , and so is not in GH .
Next, we observe that if x is the largest element of GH , then
H 0 = H [ fxg is also a semigroup, and GH0 has size g ; 1. By the
induction hypothesis,
1 X
g ; 1 y2GH0 y  g ; 1 .
Since x  2g ; 1, we get
X
y  (g ; 1)2 + 2g ; 1 = g2 ,
y2GH
and the desired result follows by induction.
(b) Equality holds in (a) exactly when x = 2g ; 1 and the elements
of GH 0 have average size g ; 1. Inductively, it follows that equality holds
whenever GH has the form f1, 3, : : : , 2g ; 1g; that is, whenever H has the
form
feven integers less than 2g and all integers at least 2gg .
262

C94. Proposed by Edward Crane and Russell Mann, graduate stu-

dents, Harvard University Cambridge, MA, USA.
Suppose that V is a k{dimensional vector subspace of the Euclidean
space Rn which is de ned by linear equations with coecients in Q. Let 
be the lattice in V given by the intersection of V with the lattice Zn in Rn ,
and let ? be the lattice given by the intersection of the perpendicular vector
space V ? with Zn . Show that the (k{dimensional) volume of  is equal to
the ((n ; k){dimensional) volume of ? .
Solution by Mark Dickinson, graduate student, Harvard University,
Cambridge, MA, USA.
Let us say that a sublattice of Zn is primitive if it arises as the inter-
section with Zn of some vector subspace of Rn . Any sublattice of Zn can be
represented by a matrix with integer entries whose rows or columns give a
basis for the lattice. The following lemma uses this to give a characterization
of the primitive lattices.
We recall the following terminology: If a1 , : : : , am are vectors (of the
same length), then their Z{span (respectively, their R{span) is the collection
of all vectors of the form X
ci
ai
i
with ci 2 Z (respectively, with ci 2 R).
Lemma. Suppose that A is an m  n matrix with integer entries, and
write  for the Z{span of the rows of A. Then  is a primitive sublattice of
Zn having the rows of A as a Z{basis, if and only if A can be completed, by
adding extra rows, to an n  n matrix B with integer entries such that B is
invertible and B ;1 has integer entries.
Proof. First suppose that A can indeed be completed to some such B .
Take V to be the R{span of the rows of A. We claim that  is the intersection
of V with Zn . Since B ;1 exists and has integer entries, every element of Zn
is an element of the Z{span of the rows of B , namely, v = (vB ;1 )B . In
fact, since the rows of B are linearly independent, every element of Zn can
be written uniquely as a combination of the rows of B . But every element
of V is an element of the R{span of the rows of A, and so by the above
uniqueness, an element in both V and Zn is in the Z{span of the rows of A.
Hence V \ Zn  , and since the reverse inclusion is clear, this half of the
lemma follows.
Conversely, suppose that  is primitive and that the rows of A form
a basis for , so are linearly independent. Then by elementary row and
column operations we can reduce A to a matrix of the form (D 0) where D
is a diagonal m  m matrix with positive entries d1 , : : : , dm . It follows that
by row operations alone, we can reduce A to a matrix A0 in which all entries
of the ith row are divisible by di . (To see this, one may easily check that
the property of being reducible by elementary row operations to a matrix
in which all the entries of the ith row are divisible by di is preserved under
 263

both elementary row and column operations. Since (D 0) has this property,
so does A.)
The rows of A0 give a new basis of . Now if any of the di were larger
than 1 we would be able to divide the corresponding row of A0 by di and
take the Z{span of the rows to obtain a sublattice of Zn which is contained
in V and strictly contains ; but this would violate the assumption that 
is primitive. Hence all the di are equal to 1, and by augmenting the matrix
(D 0) to the identity matrix and then undoing the row and column operations
that gave (D 0) from A we obtain the desired matrix B extending A.
With this lemma in hand, we now suppose that the rows of A are the
basis of a primitive lattice , and extend A as above to obtain the matrix B .
Consider the rightmost n ; m columns of the inverse of B . By the lemma,
these columns give a basis of a primitive lattice of rank n ; m, and from the
equation B
B ;1 = In we see that this lattice is perpendicular to , hence
must be the lattice ? .
Recall that the square of the volume of a lattice  is equal to the deter-
minant of A
AT for any matrix A whose rows give a basis of . Therefore,
since B has determinant 1, in order to complete the solution it is enough
to prove the following:
Proposition. Suppose B is an invertible n  n real matrix, and suppose
that A is the m  n given by the rst m rows of B while C is the n  (n ; m)
matrix given by the last n ; m columns of B ;1 . Then
det(A
AT ) = det(C T
C ) = det(B )2 .
Proof. We can always write B = L
Q where L is a lower-triangular
matrix and Q is an orthogonal matrix (that is, QT
Q = Q
QT = In ). Writing
L in the form  
L = LL02 L01
with L0 an m  m matrix and L1 an (n ; m)  (n ; m) matrix, we have
 
A = (L0 0)Q and C = QT L0;1 ,
1
from which
det(A
AT ) = det(L0
LT0 ) = det(L0)2
and similarly det(C T C ) = det(L1 );2 . Since we also know that
det(B ) = det(L0 ) det(L1), the proposition follows.
264

Real Roots of Cubic Polynomials

Keith A. Brandt and Joseph C. Roma

Abstract: Elementary calculus is used to nd conditions that indicate

when a cubic polynomial will have all real roots. This note serves as an ex-
ample of the power of calculus and gives some insight into a famous problem
from classical algebra.
In this note, we study the roots (counting multiplicities) of polynomials
with real coecients. The following statement is an immediate consequence
of the quadratic formula: f (x) = ax2 + bx + c has two real roots if and
only if b2 ; 4ac  0. We use elementary concepts from calculus to de-
rive an analogous statement for cubic polynomials. We have not found this
problem in standard calculus texts, although speci c cases are studied in [2]
(see pages 130-131). The results in this note are often presented along with
Cardan's formulas for the cubic ([1], p. 9). Our approach shows that some
important information regarding the nature of the roots can be found without
deriving Cardan's formulas.
Let f (x) be a cubic polynomial. We can use a substitution ([1], p. 9)
to eliminate the square term, so write f (x) = x3 + bx + c. First note that
for f (x) to have more than one real root, it is necessary for f 0(x) = 3x2 + b
to have a real root. This happens if and only if b  0. This condition ensures
that the graph of y = f (x) \bends" suciently.
Condition 1: If f (x) has three real roots, then b  0.
Condition 1 above is not, however, sucient for f (x) to have three
real roots. The size of the coecient c will determine whether the graph of
y = f (x) will touch the x{axis more than once. To nd the3 allowable values
for c, assume b  0 and consider the function g (x) = x + bx. If b = 0,
then necessarily, c = 0. If b < 0, the maximum and minimum values of g (x)
determine the values of c that guarantee f (x) has three real roots. They are
easy to calculate:

q  q 
max(g) = g ; ;3b = ;32b ;3b ,
q  q
min(g) = g ;b = 23b ;3b .
3

Copyright c 2001 Canadian Mathematical Society

 265

y = g(x) ; min(g)

q ;b q ;b - y = g(x)
; 3 3

y = g(x) ; max(g)

Figure 1. Finding the allowable values for c.

We must3 then2 have ; max(g )  c  ; min(g ) (see gure 1) which
b + c  0.
simpli es to 27 4
b3 + c2  0.
Condition 2: If f (x) has three real roots, then 27 4
Note that Condition 2 implies Condition 1. Furthermore, it is easy to
see that Condition 2 implies that f (x) has three real roots. Hence we have
the following ([1], p. 9):
Theorem: f (x) = x3 + bx + c has three real roots if and only if 27b3 + c2  0.
4
When the inequality is strict, f (x) has three distinct roots. If b = 0,
then c = 0 so f (x) has a root of multiplicity three at x =q 0. If b < 0 and
b3 c2 ;b
27 + 4 = 0,then f (x) has a double root at one of x =  3 .
We give the corresponding statements for the more general cubic poly-
nomial f (x) = x3 + ax2 + bx + c. They can be derived using the approach
above | with more technical calculations | or via the substitution x = y ; a3 .
Condition 1: If f (x) has three real roots, then 3b ; a2  0.
Condition 2: If f (x) has three real roots, then
; 2
2 2 3
2
27 (3b ; a ) + c ; 3 ab + 27 a  0.
2 3 ; 1
Not surprisingly, the expressions in these conditions have a striking
resemblance to those in Cardan's formulas.
Example 1: Let f (x) = x3 + 2x2 + 3x + 4. Since 3b ; a2 = 5, f (x)
fails Condition 1 and thus has a non-real root.
Example 2: Let f (x) = x3 + 3x2 + 2x + 1. Since 3b ; a2 = ;3, f (x)
passes Condition 1; however, the quantity in Condition 2 equals 27 23 . Thus
f (x) fails Condition 2 and has a non-real root.
Example 3: Let f (x) = x3 ; 3x2 ; 3x + 1. The quantity in Condition 2
equals ;16. Hence f (x) has three real roots.
266

Conditions 1 and 2 can be used to gain some information about higher

degree polynomials. We use a lemma (implicitly used for Condition 1), whose
proof follows from Rolle's Theorem and basic properties of repeated roots.
Lemma: Let f (x) be a polynomial of degree n. If f (x) has n real roots,
then f 0(x) has n ; 1 real roots.
To use the lemma, let f (x) = xn + dn;1 xn;1 + dn;2 xn;2 +


+
d1 x + d0, and consider the cubic polynomial k(x) = f (n;3)(x). If k(x)
has a non-real root, then f (x) has a non-real root. Therefore, by applying
Conditions 1 and 2 to k(x), we can nd conditions necessary for f (x) to have
all real roots. For example, Condition 1 applied to k(x) yields the following:
Condition 3: If f (x) has all real roots, then 2ndn;2 ; (n ; 1)d2n;1  0.
Loosely speaking, Condition 3 says that if dn;2 is large enough (relative
to dn;1 ), then f (x) has a non-real root.
Example 4: Let f (x) = x4 + 4x3 + 8x2 + 3x ; 1. The expression in
Condition 3 equals 16, so f (x) has a non-real root. Note that the constant
term can be chosen so that f (x) has no real roots. The point is that the
coecients d3 = 4 and d2 = 8 do not allow y = f (x) to bend enough to
have four real roots.
Example 5: Let f (x) = x5 + 2x3 ; 4x2 + 1. Since d4 = 0 and d3 is
positive, Condition 3 is not satis ed. Thus f (x) has a non-real root.
In the spirit of the Rational Roots Theorem and Descartes' Rule of
Signs, the conditions studied here give more insight into the connections
between the roots of a polynomial and its coecients. Moreover, these con-
ditions follow from elementary calculus and are easily programmed into a
computer or programmable calculator.
Exercises
1. Derive Condition 2 for the general case.
2. Prove the lemma.
3. Derive Condition 3.
4. Derive a weaker version of Condition 3 that does not involve n.
5. Apply Condition 2 to k(x) to derive Condition 4 for higher degree polynomials.
References
[1] Beyer, William H., ed., CRC Standard Mathematical Tables, CRC Press, Boca
Raton, FL, USA (1981).
[2] Kodaira, Kunihiko, ed., Basic Analysis: Japanese Grade 11, American Mathemat-
ical Society, Providence, RI, USA (1996).
Keith A. Brandt Joseph C. Roma
Department of Mathematics, Computer Science, and Physics
Rockhurst University
Kansas City, MO 64110, USA
[email protected] [email protected]
 267

PROBLEMS
Problem proposals and solutions should be sent to Bruce Shawyer, Department
of Mathematics and Statistics, Memorial University of Newfoundland, St. John's,
Newfoundland, Canada. A1C 5S7. Proposals should be accompanied by a solution,
together with references and other insights which are likely to be of help to the editor.
When a proposal is submitted without a solution, the proposer must include sucient
information on why a solution is likely. An asterisk (?) after a number indicates that
a problem was proposed without a solution.
In particular, original problems are solicited. However, other interesting prob-
lems may also be acceptable provided that they are not too well known, and refer-
ences are given as to their provenance. Ordinarily, if the originator of a problem can
be located, it should not be submitted without the originator's permission.
To facilitate their consideration, please send your proposals and solutions
on signed and separate standard 8 12 "11" or A4 sheets of paper. These may be
typewritten or neatly hand-written, and should be mailed to the Editor-in-Chief,
to arrive no later than 1 December 2001. They may also be sent by email to
[email protected]. (It would be appreciated if email proposals and solu-
tions were written in LATEX). Graphics les should be in LATEX format, or encapsulated
postscript. Solutions received after the above date will also be considered if there
is sucient time before the date of publication. Please note that we do not accept
submissions sent by FAX.

Addendum | 2338. [1998 : 234, 1999 : 243] Peter Y. Woo, Biola Uni-
versity, La Mirada, CA, USA, points out that problem 2338 was originated by
Wu WeiChao, Honan Normal University, China, in 1995. Janous had stated
that he remembered the problem, but could not give a reference. Perhaps
this is it?
Correction | 2598. [2000 : 499] D.J. Smeenk, Zaltbommel, the Neth-
erlands, points out that part (d) should read AD < a.
Clari cation | 2602?. [2001 : 48] In response to a query, Walther
Janous, Ursulinengymnasium, Innsbruck, Austria, states that his intention
was that it should be shown that no member of Q(6; 3; ;2) is a square num-
ber.
Correction | 2620. [2001 : 138, 213] The Editor regrets that he has to
state that the interval in this problem is, in fact, [1=2; 3=2], not [1=3; 3=2],
as stated in the April 2001 correction.

2637 . Proposed by Toshio Seimiya, Kawasaki, Japan.

Suppose that ABC is an isosceles triangle with AB = AC . Let D be
a point on side AB , and let E be a point on AC produced beyond C . The
line DE meets BC at P . The incircle of 4ADE touches DE at Q.
Prove that BP
PC  DQ
QE , and that equality holds if and only if
BD = CE.
268

2638 . Proposed by Toshio Seimiya, Kawasaki, Japan.

Suppose that ABC is an acute angled triangle with AB 6= AC , and that
H and G are the orthocentre and centroid of 4ABC respectively. Suppose
further that
1 + 1 = 2 , where [PQR] denotes the area
[HAB] [HAC ] [HBC ]
of 4PQR.
Prove that \AGH = 90 .
2639 . Proposed by Toshio Seimiya, Kawasaki, Japan.
Suppose that A is a point outside a circle ;. The two tangents through
A to ; touch ; at B and C . A variable tangent to ; meets AB and AC at P
and Q respectively. The line through P parallel to AC meets BC at R.
Prove that the line QR passes through a xed point.
2640 . Proposed by Toshio Seimiya, Kawasaki, Japan.
Suppose that ABC is an acute angled triangle with \BAC = 45 . Let
O be the circumcentre of 4ABC , and let D and E be the intersections of
BO and CO with AC and AB respectively. Suppose that P and Q are points
on BC such that OP kAB and OQkAC .
Prove that OD + OE = 2PQ.
p
2641 . Proposed by G. Tsintsifas, Thessaloniki, Greece.
Let H be a centrosymmetric convex hexagon, with area h, and let P be
its minimal circumscribed parallelogram, with area p.
Prove that 3p  4h.
2642 . Proposed by Christopher J. Bradley, Clifton College, Bristol,
UK. 1
Suppose that k is a positive integer. Prove that
X n ;+ 2k
= 1 .
n=0 2n+1 n+k+1 k
2643 . Proposed by Ma Jesus  Villar Rubio, Instituto Torres Quevedo
Santander, Spain.
It is known that ifpa quadrilateral has sides a, b, c and d, then its area
is less than or equal to (s ; a)(s ; b)(s ; c)(s ; d), where s is the semi-
perimeter (2s = a + b + c + d). What happens for polygons with more than
four sides?
2644 . Proposed by D.J. Smeenk, Zaltbommel, the Netherlands.
Find a closed form expression for the sum of the rst n terms of the
series
1 + 2 + 4 + 4 + 8 + 8 + 8 + 8 + 16+ 16 + 16 + 16 + 16 + 16 + 16 + 16+ : : : ,
where 1 occurs once and, for k  1, 2k occurs 2k;1 times.
 269

2645 . Proposed by Hojoo Lee, student, Kwangwoon University,

Kangwon-Do, South Korea.
Suppose that a, b and c are positive real numbers. Prove that
2 ;a3 + b3 + c3
+ 9(a + b + c)2  33 .
abc a2 + b2 + c2
2646 . Proposed by Hojoo Lee, student, Kwangwoon University,
Kangwon-Do, South Korea.
Suppose that p 2 N0 = f0, 1, 2, : : : g.
Show that there exists a function f : N0 ! N0 such that, for all n 2 N0,
we have f (f (n)) + f (n) = 2n + p if and only if 3jp.
2647 . Proposed by Michel Bataille, Rouen, France.
Let O, H and R denote the circumcentre, the orthocentre and the
circumradius of 4ABC , and let ; be the circle with centre O and radius
 = OH . The tangents to ; at its points of intersection with the rays [OA),
[OB) and [OC ) form a triangle.
Express the circumradius of this triangle as a function of R and .
2648 . Proposed by Michel Bataille, Rouen, France.
Find the smallest positive integer n such that the rightmost digits of
52001+n reproduce the digits of 52001. (Here \digit" means \decimal digit",
and the order of the digits in 52001+n and 52001 must be the same.)
2649 Proposed by Vaclav Konecny, Ferris State University, Big
Rapids, MI, USA.
Solve the equations:
p3
sin(2x) sin(4x) sin(8x) = 8 , (1)
cos(2x) cos(4x) cos(8x) = 81 . (2)
See problem 2486 [1999 : 431, 2000 : 510].
2650 Proposed by Juan-Bosco Romero Marquez, Universidad de Val-
ladolid, Valladolid, Spain.
In 4ABC , let a denote the side BC , and ha , the corresponding al-
titude. Let r and R be the radii of the inscribed and circumscribed circles,
respectively. Prove that ra < ha R.
270

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

2525 April [2000 : 177] . Proposed by Antreas P. Hatzipolakis,

Athens, Greece, and Paul Yiu, Florida Atlantic University, Boca Raton, FL,
USA.
A A
In 4ABC , we have B = 135 ; and C = 45 ; . Show that
2 2
(a) the centre V of the nine-point circle of 4ABC lies on the side BC ;
(b) if A = 60, then AV bisects angle A.
I. Solution by Nikolaos Dergiades, Thessaloniki, Greece.
Let R be the circumradius of 4ABC , M be the mid-point of BC , and
D be the foot of the altitude from A to BC .
R
(a) Since the nine-point circle passes through D and M and its radius is ,
2
to prove that V lies on BC , it is sucient to prove that DM = R. Since
B = 135 ; A2 and C = 45 ; A2 , then 2C = 90 ; A and B = 90 + C .
Hence, \BAD = C and
DB = AB sin C = 2R sin C sin C = R(1 ; cos(2C ))
= R ; R sin A = R ; a2 .
a a
Thus, DM = DB + BM = R ; + = R, and V lies on BC .
2 2
(b) If A = 60 then C = 15 , and from triangle ABD
AD = AB cos C = 2R sin C cos C = R sin(2C ) = R2 = DV ,
which means that \DAV = 45, so that \BAV = 30 and AV is the
bisector of angle A.
II. Solution by David Loeer, student, Cotham School, Bristol, UK.
Kimberling tells us that the trilinear coordinates of the nine-point
centre has coordinates cos(B ; C ) : cos(C ; A) : cos(A ; B ). Substituting
the given values, we have V represented by



cos 90 : cos 45 ; 3 A 
: cos

3 A ; 135


2 2
 3 A 
3 A 
= 0 : cos 45 ; 2 : cos 2 ; 135 .

 271

Since the trilinear coordinates of a point are proportional to the distances of

the point from the corresponding sides BC , CA, AB of the triangle, it is
immediate that the nine-point centre must lie on BC (as the rst coordinate
is zero).
Furthermore, if A = 60 then V is represented by 0 : cos(;45) :
cos(;45); therefore V is equidistant from AB and AC , which implies that
it lies on the bisector of angle A, as claimed.
Editor's comments. Several solvers pointed out that the condition in part (a)
is both necessary and sucient. Although Loeer did not say so, his proof
is clearly reversible: when V lies on BC , its rst coordinate is 0, so that
B ; C = 90 and, therefore, B = 135 ; A2 and C = 45 ; A2 . Janous,
whose solution is essentially the same as Loeer's, suggested that we all
agree to some simple way of referring to Clark Kimberling's encyclopedic list
of important points, somewhat like the way we refer to the \Bottema Bible."
I propose that we simply say \Kimberling says so" (like our Pythagorean
colleagues use of ipse dixit). I feared that \Kimberling's Koran" might o end
some readers, and this editor does not know any suitable k-words. Readers
may have better ideas. For those who have not seen the list, one can access
it electronically,
http://cedar.evansville.edu/ck6/encyclopedia/
(Loeer referred to the Mathworld site, http://Mathworld.wolfram.com/,
but it is presently shut down due to an ugly copyright dispute.)
or hard-copily,
Central Points and Central Lines in the Plane of a Triangle, Mathematics
Magazine, 67:3 (1994), 163-187. (This is an early version with 101 points
listed.)
Triangle Centers and Central Triangles, Congressus Numerantium 129, Util-
itas Mathematica Publishing Inc., University of Manitoba, 1998. (Expanded
version with many further points listed.)
Also solved by AUSTRIAN IMO TEAM 2000; MICHEL BATAILLE, Rouen, France;
MANUEL BENITO and EMILIO FERNANDEZ,  I.B. Praxedes Mateo Sagasta, Logro~no, Spain;
CHRISTOPHER J. BRADLEY, Clifton College, Bristol, UK; WALTHER JANOUS, Ursulinengym-
nasium, Innsbruck, Austria; KEE-WAI LAU, Hong Kong; GERRY LEVERSHA, St. Paul's School,
London, England; HENRY LIU, student, Trinity College, Cambridge, UK; TOSHIO SEIMIYA,
Kawasaki, Japan; ACHILLEAS SINEFAKOPOULOS, student, University of Athens, Greece
(2 solutions); D.J. SMEENK, Zaltbommel, the Netherlands; PETER Y. WOO, Biola University,
La Mirada, CA, USA; and the proposers.

2526. [2000 : 177] Proposed by K.R.S. Sastry, Bangalore, India.

In a triangle, prove that an internal angle bisector trisects an altitude
if and only if the bisected angle has the measure =3 or 2=3.
Preliminary comments. To make sense of the possibility that the
bisected angle be obtuse, it is necessary to rephrase the problem:
272

Prove that the internal angle bisector of one angle of a triangle

divides an altitude in the ratio 1 : 2 if and only if the bisected
 2
angle measures or .
3 3
Several readers point out that the word trisect (in the original statement)
implies division into three equal parts, which eliminates the obtuse case.
Solution by Jeremy Young, student, University of Cambridge, Cam-
bridge, UK.
In triangle ABC let BN be the altitude from B to AC , and let P be
the point where the bisector of \CAB meets BN . We consider two cases.
(i) A  .

2
A
BN = AN tan A = 21AN tan 2
; tan2 A and PN = AN tan A2
2

imply that
PN = 1 ; tan2 A2 .
BN 2
PN = 2 or 1 .
PN is divided internally in the ratio 1 : 2 if and only if BN
A
 
4 3 3
In the former case 1 ; tan 2
2 = 3 has no solution. In the latter case
1 ; tan2 A2 = 32 implies A2 = 6 (since A > 0), so that A = 3 .
 


Conversely, if A = then
PN = 1 .
3 BN 3
(ii) A > .

2
BN = AN tan( ; A) = ;AN tan A and PN = AN tan A 2
PN tan2 A ; 1
2
imply that
BN = 2 .
Since BP  PN , we know that BN is divided externally in the
ratio 1 : 2 if and only if
BP = 2. In this case PN = BN , so that
A  ; 1 = 2 and (since
 PNA  A  2
tan2 2 2 < 2 ), therefore 2 = 3 . Thus, A = 3 .
Finally, it is clear that the argument can again be reversed so that A =
2
3
is also sucient.
Also solved by the AUSTRIAN IMO TEAM 2000; MICHEL BATAILLE, Rouen, France;

MANUEL BENITO and EMILIO FERNANDEZ, I.B. Praxedes Mateo Sagasta, Logro~no, Spain;
CHRISTOPHER J. BRADLEY, Clifton College, Bristol, UK; JONATHAN CAMPBELL, student,
 273

Chapel Hill High School, Chapel Hill, NC; NIKOLAOS DERGIADES, Thessaloniki, Greece; JOHN
G. HEUVER, Grande Prairie Composite High School, Grande Prairie, Alberta; OLEG IVRII, stu-
dent, Cummer Valley Middle School, Toronto; WALTHER JANOUS, Ursulinengymnasium, Inns-
bruck, Austria; MURRAY S. KLAMKIN, University of Alberta, Edmonton, Alberta; V ACLAV 
KONECN  Y,
 Ferris State University, Big Rapids, MI, USA; GERRY LEVERSHA, St. Paul's School,
London, England; MICHAEL PARMENTER, Memorial University of Newfoundland, St. John's,
Newfoundland; D.J. SMEENK, Zaltbommel, the Netherlands; PANOS E. TSAOUSSOGLOU,
Athens, Greece; KENNETH M. WILKE, Topeka, KS, USA; PETER Y. WOO, Biola University,
La Mirada, CA, USA; and the proposer.

2527. [2000 : 177] Proposed by K.R.S. Sastry, Bangalore, India.

Let AD, BE and CF be concurrent cevians of 4ABC . Assume that:
(a) AD is a median; (b) BE bisects \ABC ; (c) BE bisects AD.
Prove that CF > BE .
I. Solution by the Austrian IMO team 2000.
Let P be the point of intersection of lines AD, BE , CF . Since BP
is both median and angle bisector in 4ABD, the triangle is isosceles with
BD = BA, and \BDP must be acute. In 4CDP and 4BDP , we have
CD = BD; they share the common side DP , and \BDP being acute
implies its supplement \CDP is obtuse. Therefore
CP > BP . (1)
Since AD, BE , and CF are concurrent, Ceva's Theorem says that
BF
AE
CD = 1; since CD = 1, AE = FA so that EF and BC
FA EC DB DB EC BF
are parallel. This implies (because P is the common point of BE and CF )
that 4CBP  4FEP .
Hence from (1), FP > EP , and CF = CP + PF > BP + PE = BE ,
as desired.
II. Independent identical solutions by Manuel Benito and Emilio
Fernandez, I.B. Praxedes Mateo Sagasta, Logro~no, Spain and Kee-Wai Lau,
Hong Kong.
In a Cartesian system assign (without loss of generality) the coordinates
A(;1; 0) and D(1; 0). The line BE bisects \ABD by condition (b) and
passes through O(0; 0) by (c), so that by the bisector theorem applied to
4ABD, BE is the y{axis and B has coordinates B(0;b). Because D(1; 0)
 C =(2; ;b). Moreover E , the intersection
is the mid-point of BC , we have
b
of AC with the y {axis, is E 0; ; , and F (where CO intersects AB ) is
3

2 b 
F ; 3 ; 3 . We have, nally,

CF = 43 4 + b2 > 34 jbj = BE .
p
274

Also solved by MICHEL BATAILLE, Rouen, France; CHRISTOPHER J. BRADLEY, Clifton

College, Bristol, UK; NIKOLAOS DERGIADES, Thessaloniki, Greece; RICHARD B. EDEN, Ate-
neo de Manila University, Manila, the Philippines; OLEG IVRII, student, Cummer Valley Mid-
dle School, Toronto; WALTHER JANOUS, Ursulinengymnasium, Innsbruck, Austria; MURRAY
S. KLAMKIN, University of Alberta, Edmonton, Alberta; GERRY LEVERSHA, St. Paul's School,

London, England; HENRY LIU, student, Trinity College Cambridge, England; HEINZ-JURGEN
SEIFFERT, Berlin, Germany; TOSHIO SEIMIYA, Kawasaki, Japan; D.J. SMEENK, Zaltbommel,
the Netherlands; PETER Y. WOO, Biola University, La Mirada, CA, USA (2 solutions); and the
proposer.
As a by-product of his solution Sei ert obtained CF > 43 AB > BE . These inequalities
are clear from solution II above.

2531. [2000 : 178] Proposed by G. Tsintsifas, Thessaloniki, Greece.

Let F be a convex plane set and AB its diameter. The points A and B
divide the perimeter of F into two parts, L1 and L2 , say. Prove that
1 < L1 <  ; 1 .
 ; 1 L2
While a number of solvers proved the inequality stated in the problem,
others established a sharper set of bounds. We present two solutions.
I. Solution by Murray S. Klamkin, University of Alberta, Edmonton,
Alberta.
It is known [1] that D  L  2D, where L and D are the perime-
ter and diameter of a convex plane set. Hence, for each of the two regions
divided by AB
D  L1 + D  2D and D  L2 + D  2D , so that
 ; 1  LD1  1 and 1  LD   ;1 1 .
2
Then, by multiplying the latter inequalities, we get
 ; 1 > LL1 >  ;1 1
2
(the equality signs do not hold here).
1. T. Bonnesen, W. Fenchel, Theory of Convex Bodies, BCS Associates,
Moscow, Idaho, 1987, p. 88.
II. Solution by Walther Janous, Ursulinengymnasium, Innsbruck,
Austria.
In what follows, we show the sharper inequality
3  L1  2 .
2 L2 3
 275

It is enough to show the right-hand inequality, for the other side then
follows by taking reciprocals of the rst one. We thus let L1  L2 . We have
d := AB  L2 .
Then the \L1 {part" of the boundary @F has to
C be contained in the region determined by the
d-circles about A and B. Because of the convexity
of F , it follows that:
L1  AC + CB = 2d 6 + 2d = 2d .
6 3
A
d L 2d
B Therefore, L1  3d = 3
2
2
(which is less than  ; 1).
Also solved by AUSTRIAN IMO TEAM 2000; MICHEL BATAILLE, Rouen, France;
NIKOLAOS DERGIADES, Thessaloniki, Greece; J. SUCK, Essen, Germany; PETER Y. WOO, Biola
University, La Mirada, CA, USA; and the proposer.

2533. [2000 : 178] Proposed by K.R.S. Sastry, Bangalore, India.

In the integer sided 4ABC , let e denote the length of the segment of
the Euler line between the orthocentre and the circumcentre.
Prove that 4ABC is right-angled if and only if e equals one half of the
length of one of the sides of 4ABC .
Compare problem 2433. [1999 : 173, 2000 : 187]
Solution by Walther Janous, Ursulinengymnasium, Innsbruck, Austria.
c
To show, without loss of generality: OH = if and only if 4ABC is
2
right-angled.
One direction is clear | if C is a right angle then OH = R = .
c
2
c
For the other direction, we assume that OH = . Using a familiar
2
formula [found in books that deal with the Euler line, such as Nathan
Altshiller Court's College Geometry, formula 203, page 102, or Coxeter and
Greitzer, Geometry Revisited, exercise 1.7.2, pages 20 and 157], we see that
OH 2 = 9R2 ; a2 ; b2 ; c2 = c4 .
2

Next use R2 =
a2b2c2 (where F is the area of 4ABC ), and multiply through
16F 2
by 4
16F 2, where (by Heron's formula)
16F 2 = 2a2b2 + 2b2 c2 + 2c2a2 ; a4 ; b4 ; c4 ,
276

and factor to get

 
;
c2 ; a2 ; b2

5c4 ; ;a2 + b2
c2 ; 4 ;a2 ; b2
2 = 0 .
If the rst factor is zero, then the triangle has a right angle at C , as desired.
If not, then
c2 = 10 1 a2 + b2  (a2 + b2)2 + 80 (a2 ; b2)2 .
 q
(1)


For c to be an integer, we must have 81(a2 ; b2)2 + 4a2 b2 = z 2 for some

integer z ; that is,
(3a)2 ; (3b)2 2 + [2ab]2 = z 2 .
 

Comparing this to the formula for generating Pythagorean triples (namely

[a2 ; b2 ]2 + [2ab]2 = [a2 + b2]2), we infer a = b. When a = b, formula (1)
gives c2 = (2a2 2a2 )=10. Since c is non-zero (in a non-degenerate triangle),
we conclude that if the triangle were not a right triangle then
c2 = 25a ,
2

which is impossible for integers a and c.

Also solved by the proposer. There was one incomplete submission.

2534. [2000 : 178] Proposed by Walther Janous, Ursulinengymnas-

ium, Innsbruck, Austria.
Suppose that a is an integer and x and y natural numbers. De ne
za(x; y) = x + y + a .
2 2
xy
1. Show that there exist in nitely many values of a such that za (x; y ) is
an integer for in nitely many pairs (x; y ) 2 N2 .
2.? Is the set E (a) of integers za (x; y ) as obtained above necessarily
in nite? If the answer is \no", determine those a's which determine
nite sets E (a).
I. Solution by Manuel Benito and Emilio Fernandez, I.B. Praxedes
Mateo Sagasta, Logro~no, Spain.
1. For example, for any pair of natural numbers  and d, we have
z;d2 (d; d) =  d +dd2 ; d =  .
2 2 2 2
 277

2. For the rst part, the answer is \no", since for a = 0, for example,
we have
z0(x; y) = x + y .
2 2
xy
Let gcd(x; y ) = d, and thus x = x1 d and y = y1 d with x1 , y1 relatively
prime natural numbers. We then have that
z0(x; y) = x1x+y y1 .
2 2
1 1
But gcd(x21 + y12; x1 ) = gcd(x21 + y12; y1) = 1, so that for z0 (x; y ) to be a nat-
ural number it must necessarily be true that x1 = y1 = 1. Thus z0 (x; y ) = 2,
and we have shown that E (0) = f2g.
For the second part, we have already shown in 1, that if a = ;d2 , where
d is any positive integer, then E(a) is an in nite set; in fact E(a) = N. We
will prove now that, for every other integer value of a, E (a) shall be a nite
set, with cardinality greater than or equal to 1.
In the following proposition we prove that jE (a)j  a + 2 when
a 2 N [ f0g.
Proposition 1. Let a 2 N [ f0g be given. If x +xyy +a = where x; y
2 2

and are natural numbers, then  a + 2.

Proof. Suppose that, for a certain xed value of y 2 N, there are
x and , natural numbers, such that x2 +xyy2+a = or, what is the same,
y2 + a = x( y ; x). Both x and y ; x being necessarily positive, the
maximum value of the sum x + ( y ; x) = y is attained when x = 1 and
y ; x = y2 + a, or2when x = y2 + a and y ; x = 1. The said maximum
value of y is thus y + a + 1. Since y was xed, the maximum value of is
y2 + a + 1 = y + a + 1 .
m (y ) = y y
But the maximum integer value of the function B : N ! R given by
B(y) = y + a+1y is B (1) = B (a + 1) = a + 2, as is easily seen (if y 2 N and
y > a + 1 then y + a+1y 62 N), and this completes the proof.
Lastly, the following proposition proves that E (a) is also a nite set
when a < 0 and a is not of the form ;d2 for d 2 N.
Proposition 2. If the positive integer a is not a perfect square, then
E(;a) is a nite set.
Proof. We suppose that, for some natural numbers x and y , the number
= (x2 + y2 ; a)=(xy) is an integer.
Consider rst the possibility that y 2 ; a p< 0. This gives a nite number
of possible values for y (namely 1, 2, : : : , b ac), and since x must divide
278

y2 ; a 6= 0, x can take on only a nite number of values for each particular

value of y . Consequently, the number of values that takes on is nite.
It remains only to examine the possibility that y 2 ; a > 0. By reasoning
as in the proof of Proposition 1, for each xed value of y , the maximum value
of y shall be y 2 ; a + 1, and the maximum value of is
 (y ) = y ; a + 1 = y ; a ; 1 .
2
(1)
y y
But the maximum integer p value of the function (1), since y goes over all
natural numbers between d a e and a ; 1, is  (a ; 1) = a ; 2 [since  (y )
is increasing]. Consequently, the number of values possibly taken by is
also nite in this case.
II. Partial solution by Edward T.H. Wang, Wilfrid Laurier University,
Waterloo, Ontario.
1. Let Sa denote the set of all (x; y ) 2 N2 such that za (x; y ) is an
integer. We shall show that Sa is in nite for all non-negative integers a.
This is obvious for a = 0 since z0(x; x) = 2 for all x 2 N. Hence, we
assume that a > 0. Note rst that (1; 1) 2 Sa since za (1; 1) = a + 2 is an
integer. Next suppose (b; c) 2 Sa . Then b2 + c2 + a = kbc for some integer
k = za(b;c). Hence [since b2 + a = c(kb ; c) means that kb ; c > 0],
za(kb ; c; b) = (kb ;b(ckb )2 + b2 + a = k2 b2 ; 2kbc + (b2 + c2 + a)
; c) b(kb ; c)
= kb(bkb;;kbc
2 2
c) = k .
This shows that (kb ; c; b) 2 Sa whenever (b;c) 2 Sa ; in fact
za(kb ; c; b) = za(b; c) = k.
Using this observation, we can construct, for all a > 0, an in nite
sequence of solutions (xn ; yn ) 2 Sa as follows. Let (x1 ; y1) = (1; 1)
[for which za (1; 1) = a + 2], and for n  1 de ne
(xn+1 ; yn+1) = ((a + 2)xn ; yn ; xn) .
Then (xn ; yn ) 2 Sa for all n as above, provided that xn > 0 for all n.
We now show by induction that the sequence fxn g is strictly increasing. It
would then follow that xn > 0 and all the (xn ; yn)'s are distinct. Clearly
x2 = (a + 2)x1 ; y1 = a + 1 > 1 = x1. Suppose that xn+1 > xn for some
n  1. Then
xn+2 = (a + 2)xn+1 ; yn+1 = (a + 2)xn+1 ; xn  2xn+1 ; xn > xn+1;
completing the induction and the proof.
Editorial note. This shows that any non-negative integer value of a will
answer part 1. In Solution I it was shown that any a which is the negative
of a perfect square will work too, but that other negative values of a will not.
 279

2. (Partial solution) No, the set E (a) could be nite. In fact we shall
show that E (6) = .
[Editorial comment. Here Wang speci ed a certain nite set of positive in-
tegers. We withhold this information at this time because it forms part of a
solution by Wang of Problem 1 from the Estonian Math Contest 1995{1996,
which appeared in the Olympiad Corner [2000 : 6] and which readers may
still want to work on. This problem and 2534 are of course closely related.
In fact Wang states his \hunch" that 2534 and the Estonian Math Olympiad
problem have the same \origin".]
Part 1 was also solved by CHRISTOPHER J. BRADLEY, Clifton College, Bristol, UK; and

by HEINZ-JURGEN SEIFFERT, Berlin, Germany. Bradley gave the same general result as in
Solution II. Sei ert solved this part as in Solution I.

2535?. [2000 : 179] Proposed by Walther Janous, Ursulinengym-

nasium, Innsbruck, Austria.
1. Prove that neither of the integers a(n) = 3n2 + 3n + 1 and
b(n) = n2 + 3n + 3 (n  1) has a divisor k such that k  2 (mod 3).
2. Prove or disprove that both of the sequences fa(n)g and fb(n)g (n  1)
contain in nitely many primes.
Solution of part 1 compiled from received solutions.
Supposing that a(n) or b(n) has a divisor k  2 (mod 3), then it has a
prime divisor p  2 (mod 3). Since a(n) and b(n) are odd, p  5 (mod 6).
Now
12a(n) = 12(3n2 + 3n + 1) = (6n + 3)2 + 3
and
4b(n) = 4(n2 + 3n + 3) = (2n + 3)2 + 3 ,
so that pja(n) or pjb(n) implies that p divides an integer of the form N 2 +3;
that is, N 2  ;3 (mod p); that is, ;3 is a quadratic residue of p. But it
is known that ;3 is a quadratic residue of a prime p > 3 if and only if
p  1 (mod 6) (see for instance #5 of Problems 9.3 in David Burton's
Elementary Number Theory), which is a contradiction.
Part 1 solved by CHRISTOPHER J. BRADLEY, Clifton College, Bristol, UK; MANUEL
BENITO and EMILIO FERNANDEZ,  I.B. Praxedes Mateo Sagasta, Logro~no, Spain; KENNETH
M. WILKE, Topeka, KS, USA; and JEREMY YOUNG, student, University of Cambridge, Cam-
bridge, UK.
Most solvers used quadratic reciprocity to prove that ;3 is a quadratic nonresidue of
any prime p  5 (mod 6).
Part 2 was not attempted by any solvers. In fact this part should probably not have
been included, as there is not a single quadratic polynomial which has been proven to generate
in nitely many primes. Although most everyone believes that there are in nitely many primes
of the form n2 + 1, for instance, nobody has the slightest idea how to prove it.
280

2537. [2000 : 179] Proposed by Aram Tangboondouangjit, Carnegie

Mellon University, Pittsburgh, PA, USA.  
Find the exact value of cot

 
+ cot 2 ; cot 3 .
 

7 7 7
I. Solution by David Doster, Choate Rosemary Hall, Wallingford, CT,
USA (slighty modi ed by the editor). p
Let S denote the given expression. We show that S = 7. We let
 = 7 and use the following facts freely throughout: for all integers k,
cos(k) = ; cos (7 ; k) , sin(k) = sin (7 ; k) , and in particular,
;
;

sin 3 = sin 4 = 2 sin2 cos 2.

We have
cos  ; cos 3 + cos 2 = sin 3 cos  ; cos 3 sin  + cos 2
S = sin  sin3 sin2 sin  sin3 sin2
= sinsin2  cos 2 1 sin2 cos 2
 

 sin 3 + sin2 = sin  sin3 + 2 cos 

= sin1  2 cos1 2 + 2cos

2  = 1  cos  + cos2 2 
cos  2 sin  cos  cos 2
= 2 cos  + 2 cos2 2 = 2 cos  + (1 + cos 4)
2 sin 2 cos 2 sin4
; cos 3 + 1 = T .
= 2 cos  sin 3
Now,
T 2 = 4 cos  + cos 3 + 1 ; 4sincos  cos 3 + 4 cos  ; 2 cos 3
2 2
2 3
= (4+4 cos 2)+(1+cos6)+22;sin 4(cos4 +cos 2)+8 cos  ; 4 cos 3
2 3
= 7 + (cos 6 + 8 cos ) ; 4(cos 4 + cos 3)
1 ; cos 6
7 ; 7 cos 6 
= 1 ; cos 6 = 7 .

Since 0 <  < 3 < , clearly T > 0, and thus, it follows that
2
S = p7.
II. Solution by Peter Y. Woo, Biola University, La Mirada, CA, USA.
Let = 7 and let Y denote the answer. Let ABCDE be a side of a
triangle AEF in which \BAF = ; AB = BF = FE = 1; CE = CF and
FD ? AE. Let x = BE. Then BD = DE = x2 . (See gure.)
 281

F
2
1 1
2 2
A 1 B CD 1 -E
 x x
-
Since \FBE = \FEB = \CFE = 2 , 4FBE is similar to 4CFE .
Hence, \FCE = 3 and CE =
CE = FE = 1 , which implies
1 x FE BE x
CD = x ; 2 .
x2
Let h = FD. Then h2 + = 1 and
4
Y = cot( ) + cot(2 ) ; cot(3 ) = AD
h + DE ; CD ,
h h
or

hY = 1 + x ; x ; 2 = 1 + 32x ; x1 .
x
 
1 (1)
Since 4FBC is similar to 4AEF , we have
BC = BC FE 1
BF = AE = AE = 1 + x .
1

Hence, x = BE = BC + CE =
1 + 1 , which simpli es to
1+x x
x3 + x2 = 2x + 1 . (2)
1 2
x = x + x ; 2. Substituting into (1), we get
From (2), we have

hY = 1 + 32x ; (x2 + x ; 2) = 3 + x2 ; x2 or 2hY = 6 + x ; 2x2 .

Hence,
4h2 Y 2 = ;6 + x ; 2x2
2 = 36 + 12x ; 23x2 ; 4x3 + 4x4 . (3)
From (2) we have x4 = ;x3 + 2x2 + x. Substituting into (3), we get

4h2Y 2 = 36 + 16x ; 15x2 ; 8x3 . (4)

282

Again, from (2), we have x3 = ;x2 + 2x + 1. Substituting into (4), we get

(surprise!!) 4h2 Y 2 = 28 ; 7x2 = 28 1 ;
x2  = 28h2 or Y 2 = 7 .


4
p
Since, clearly, Y > 0, we conclude that Y = 7.

Also solved by SEFKET ARSLANAGIC,  University of Sarajevo, Sarajevo, Bosnia and
Herzegovina; MICHEL BATAILLE, Rouen, France; BRIAN D. BEASLEY, Presbyterian Col-
lege, Clinton, SC, USA; MANUEL BENITO and EMILIO FERNANDEZ,  I.B. Praxedes Mateo
Sagasta, Logro~no, Spain; CHRISTOPHER J. BRADLEY, Clifton College, Bristol, UK; CHARLES R.
DIMINNIE, Angelo State University, San Angelo, TX, USA; RICHARD B. EDEN, Ateneo
de Manila University, Philippines; RICHARD I. HESS, Rancho Palos Verdes, CA, USA;
JOE HOWARD, New Mexico Highlands University, Las Vegas, NM, USA; WALTHER JANOUS,
Ursulinengymnasium, Innsbruck, Austria; KEE-WAI LAU, Hong Kong, China; HENRY LIU, stu-
dent, Trinity College, Cambridge, England; GOTTFRIED PERZ, Pestalozzigymnasium, Graz, Aus-

tria; JUAN-BOSCO ROMERO M ARQUEZ, Universidad de Valladolid, Valladolid, Spain; HEINZ-

JURGEN SEIFFERT, Berlin, Germany; ACHILLEAS SINEFAKOPOULOS, student, University of
Athens, Greece; D.J. SMEENK, Zaltbommel, the Netherlands; RICHARD TOD, Forest of Dean,
England, UK; JEREMY YOUNG, student, University of Cambridge, Cambridge, UK; and the
proposer. There was also one partially incorrect solution.
Most of the submitted solutions use either the expansion formulae for sin 7, cos 7
and/or tan 7, with or without derivations; or De Moivre's Theorem and Euler's formula; or
various other known results. Solution I above is one of the very few exceptions and is self-
contained. On the other hand, the proof given in Solution II is the only one based on geometry.
Sinefakopoulos pointed out that this problem is equivalent to problem #2547. Tod com-
mented that it was part of a Cambridge University Scholarship and Exhibitions Examination in
Mathematics which took place in the nineteen fties.

2538. [2000 : 179] Proposed by Edward T.H. Wang, Wilfrid Laurier

University, Waterloo, Ontario.
On a recent calculus test, students were asked to compute the arc length
of a curve represented by a certain function f (x), for x = a to x = b,
a < b. One of the students, a Mr. Fluke, simply calculated f 0(b) ; f 0(a),
and obtained the correct answer.
Determine all real functions, f (x), di erentiable on some open inter-
val I , such that, for all a, b satisfying (a; b)  I , the arc length of the curve
y = f (x), from x = a to x = b is equal to f 0(b) ; f 0(a).
Solution by David Loeer, student, Cotham School, Bristol, UK.
We must have
Z bq b
1 + (f 0(x))2 dx = f 0(x) a .
a
Since this must be true for all a, b in I , and the left hand side possesses a
derivative, the right hand side must also. Then, we may di erentiate with
respect to b: q
1 + (f 0(b))2 = f 00 (b) .
 283

for simplicity of notation, that g (x) = f 0 (x). Thus, we must solve

Suppose p
0
g (x) = 1 + g(x)2, which we do as follows:
dg = p1 + g2 , p
dg = dx ,
dx Z 1 + g2
p
dg = dx , Z

1 + g2
sinh;1 (g) = x + c , g(x) = sinh(x + c) .
Since g (x) = f 0(x), p
it follows that f (x) = cosh(x + c) + d. (No special
integrals exist, since 1 + g 2 cannot be zero; thus, we have not \lost" any
solutions by dividing by it.)
Also solved by MIGUEL CARRI ON  ALVAREZ,
 Universidad Complutense de Madrid,
Spain; MICHEL BATAILLE, Rouen, France; NIKOLAOS DERGIADES, Thessaloniki, Greece; KEE-
WAI LAU, Hong Kong; GERRY LEVERSHA, St. Paul's School, London, England; DAVID VELLA,
Skidmore College, Saratoga Springs, New York; ALBERT WHITE, St. Bonaventure University,
St. Bonaventure, New York; KENNETH M. WILKE, Topeka, KS, USA; and the proposer. There
was one incomplete and one incorrect solution.

Note: Alvarez also makes the following observation:
The hyperbolic cosine is known as the `catenary' in mechanics because it is the shape
adopted by an inextensible string (`catena' is Latin for `chain') of constant density per unit
length, with xed endpoints and subject only to its own weight. The equilibrium condition
for a string is that the tension at any two points cancel the weight of the string between the
two points. Since the tensions at points x = a; b are parallel to (;1; ;f 0(a)) and (1; f 0 (b))
respectively and the weight of the string is proportional to the length of the string and directed
along (0; ;1), the equilibrium condition is equivalent to the condition that the length of the
string is proportional to f 0 (b) ; f 0(a).

2539. [2000 : 236, 372] Proposed by Hojoo Lee, student, Kwang-

woon University, Kangwon-Do, South Korea, adapted by the editor.
Let ABCD be a convex quadrilateral with vertices oriented in
the clockwise sense. Let X and Y be interior points on AD and BC ,
respectively. Suppose that P is a point between X and Y such that
\AXP = \BY P = \APB =  and \CPD =  ;  for some .
(a) Prove that AD
BC  4 PX
PY .
(b)? Find the case(s) of equality.
Solution by Toshio Seimiya, Kawasaki, Japan.
(a) The triangles APX and PBY are similar, because
\AXP = \PY B = 
and
\APX = \BPX ; \APB = (\PBY + \PY B ) ; \APB = \PBY .
284

B
A

 P
X   Y
;
D

Then AX : PY = PX : BY , so that
AX
BY = PX
PY . (1)
Since \DXP = \PY C = \DPC =  ; , we similarly obtain
DX
CY = PX
PY . (2)
By the AM{GM Inequality,
p
AD = AX + DX  2 AX
DX
and
p
BC = BY + CY  2 BY
CY .
Multiplying these two inequalities, we get
p
AD
BC  4 AX
BY
DX
CY ,
and from (1) and (2),
p
AD
BC  4 PX 2
PY 2 = 4PX
PY .
(b) Equality holds when AX = DX and BY = CY . We can show that
there is a quadrilateral ABCD having all the required properties and such
that AX = DX and BY = CY .
 285

E A
 B

X 

D P 

Y

C
Start with any triangle PCB . Let Y be the mid-point of the side CB .
Extend Y P and construct the point E on the line Y P (so that the order of
points is Y , P , E ) and a point A such that
\APE = \PBC
and
\AEP = \PCB .
Thus, 4AEP is similar to 4PCB . Therefore, if X is the mid-point of PE ,
we have
\PXA = \BY P
and
\PAX = \BPY .
Consequently,
\APB =  ; \APE ; \BPY
=  ; \PBC ; \BPY
= \BY P .

Thus, \APB = \BY P = \PXA. Let \PXA = , and let D be the

point symmetric to point A with respect to point X . We can now show that
the quadrilateral ABCD has all of the required properties. We only need a
proof that \DPC =  ; . Since 4AEP is similar to 4PCB , we obtain
\XAE = \Y PC .
Since 4AXE and 4DXP are congruent, we have
\XEA = \XPD .
286

Consequently,
\DPC =  ; \Y PC ; \XPD
=  ; \XAE ; \XEA
= \AXE
=  ; \PXA
=  ;,

which completes the proof.


Also solved by SEFKET ARSLANAGIC, University of Sarajevo, Sarajevo, Bosnia and
Herzegovina; and MIHAI CIPU, IMAR, Bucharest, Romania.

2540. [2000 : 236] Proposed by Paul Yiu, Florida Atlantic University,

Boca Raton, FL, USA.
Given an equilateral triangle ABC , let P and Z be points on the incircle
such that P is the mid-point of AZ and BZ < CZ . The segment CZ and
the extension of BZ meet the incircle again at X and Y respectively. Show
that:
1. triangle XY Z is equilateral;
2. the points A, X and Y are collinear; and
3. each of the segments XA, Y B and ZC is divided in the golden ratio
by the incircle.
Solution by Toshio Seimiya, Kawasaki, Japan.
1. Let I be the incentre of 4ABC . The incircle of 4ABC touches
BC , CA, and AB at D, E, and F , respectively. Then D, E, and F are mid-
points of BC , CA, and AB , respectively. (See the diagram below.) Since P
is the mid-point of AZ , we have FP k BZ and PE k ZC . Hence,
\BZC = \FPE = \FEC = 180 ; \AEF = 120 .
Since I is the incentre of the equilateral triangle ABC , we have
\BIC = 120 = \BZC .
Therefore, B , C , I , Z are concyclic, implying that
\IZX = \ICB = 30 , \IZC = \IBC = 30 .
Since IX = IZ = IY , we get
\IXZ = \IZX = 30 , \IY Z = \IZY = 30 .
Thus, we have 4IZX  = 4IZY , so that ZX = ZY . Since \BZC = 120,
we also have that \XZY = 60 . Hence, 4XY Z is equilateral.
 287

X
F P E
I
Z

Y
B D C

2. Since \IXB = 30 = \IAB , we see that A, B , I , and X are

concyclic. Hence, \AXB = \AIB = 120 .
Thus, \AXB + \BXY = 120 + 60 = 180 . Therefore, A, X , and
Y are collinear.
3. Since 4XY Z and 4DEF are equilateral triangles which are
inscribed in the same circle, they are congruent. Hence,
XY = Y Z = ZX = DE = EF = FD = 21 BC = 21 CA = 12 AB .
Thus, XY 2 = AE 2 = AX
AY . Similarly, Y Z 2 = CD2 = CY
CZ , and
ZX 2 = BF 2 = BZ
BX . Therefore, each of the segments Y A, XB, and
ZC is divided in a golden ratio by the incircle.
Also solved by MICHEL BATAILLE, Rouen, France; CHRISTOPHER J. BRADLEY, Clifton
College, Bristol, UK; NIKOLAOS DERGIADES, Thessaloniki, Greece; WALTHER JANOUS, Ursu-
linengymnasium, Innsbruck, Austria; KEE-WAI LAU, Hong Kong; HOJOO LEE, student, Kwang-
woon University, Kangwon-Do, South Korea; GERRY LEVERSHA, St. Paul's School, London,
England; HENRY LIU, student, Trinity College Cambridge, UK; D.J. SMEENK, Zaltbommel, the
Netherlands; JEREMY YOUNG, student, Nottingham High School, Nottingham, UK; and the
proposer. There was one incorrect solution.

2541. [2000 : 237] Proposed by Edward T.H. Wang, Wilfrid Laurier

University, Waterloo, Ontario.
Show that, for all natural numbers n and k  2,
X n kn;2i ;k2 ; 4
i
i 2i
is divisible by 2n;1 .
288

I. Solution by Gerry Leversha, St. Paul's School, London, England.

Let an denote
h the
p given sum.
n Note prst, using
n the
i Binomial Expansion,
that an = 2 k + k ; 4 + k ; k ; 4 . Now this expression
1 2 2
is recognisable as the solution to a second-degree linear recurrence rela-
tion. The auxiliary equation is x2 ; 2kx + 4 = 0, so that the relation is
an+2 = 2kann+1 ; 4an, and the initial conditions are a0 = 1 and a1 = k. To
prove that 2 ;1 jan for all n  1, we proceed by induction. The case n = 1
is trivial and a2 = 2k2 ; 4, which is divisible by 2. Suppose 2r;1jar and
2r jar+1 for some r  1. Then ar+2 = 2k
A
2r ; 4B
2r;1 = (Ak ; B )2r+1
for some integers A and B , completing the induction.
II. Solution by the proposer.
 p  
Let x = 21 k + k2 ; 4 . Then x;1 = 12 k ; k2 ; 4 , and,
p 

a
as in I, xn + x;n = n;n 1 .
2
It therefore sucies to show that xn + x;n is an integer for all natural
numbers n. This is clearly true for n = 1. Suppose xm + x;m is an integer
for all m = 1, 2, : : : , n ; 1 for some n  2. Then
 
xn+1 + x;(n+1) = (x + x;1 )(xn + x;n ) ; xn;1 + x;(n;1) ,
which is an integer by the induction hypothesis.
Also solved by MICHEL BATAILLE, Rouen, France; DAVID DOSTER, Choate Rosemary
Hall, Wallingford, CT, USA; WALTHER JANOUS, Ursulinengymnasium, Innsbruck, Austria;
KEE-WAI LAU, Hong Kong; HENRY LIU, student, Trinity College, Cambridge, England;
DAVID LOEFFLER, student, Cotham School, Bristol, UK; HEINZ-JURGEN  SEIFFERT, Berlin,
Germany; and PETER Y. WOO, Biola University, La Mirada, CA, USA.
Most of the submitted solutions are virtually the same as solution I above.

Crux Mathematicorum
Founding Editors / Redacteurs-fondateurs: Leopold Sauve & Frederick G.B. Maskell
Editors emeriti / Redacteur-emeriti: G.W. Sands, R.E. Woodrow, Bruce L.R. Shawyer
Mathematical Mayhem
Founding Editors / Redacteurs-fondateurs: Patrick Surry & Ravi Vakil
Editors emeriti / Redacteurs-emeriti: Philip Jong, Je Higham,
J.P. Grossman, Andre Chang, Naoki Sato, Cyrus Hsia