THE OLYMPIAD CORNER

65
THE OLYMPIAD CORNER

No. 220
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.
We begin with the problems of the two days of the Turkish Mathemat-
ical Olympiad 1998. Thanks go to Ed Barbeau for collecting these problems
at the IMO in Romania.
VI TURKISH MATHEMATICAL OLYMPIAD
Second Round
First Day - December 11, 1998 (Time: 4.5 hours)
\ \
1. On the base of the isosceles triangle ABC (jABj = jAC j) we
choose a point D such that jBDj : jDC j = 2 : 1 and on [AD] we choose a
\ \
point P such that m(B AC ) = m(B PD).
Prove that m(DPC ) = m(B AC )=2.
2. Prove that
(a + 3b)(b + 4c)(c + 2a) 60abc
for all real numbers 0 a b c.
3. The points of a circle are coloured by three colours. Prove that there
exist innitely many isosceles triangles with vertices on the circle and of the
same colour.
Second Day - December 12, 1998 (Time: 4.5 hours)
4. Determine all positive integers x, n satisfying the equation
x + 3367 = 2n .
3
5. Given the angle XOY , variable points M and N are considered on

the arms [OX ] and [OY ], respectively, so that jOM j + jON j is constant.
Determine the geometric locus of the mid-point of [MN ].
6. Some of the vertices of unit squares of an n n chessboard are
coloured so that any k k square formed by these unit squares on the chess
board has a coloured point on at least one of its sides. If l(n) stands for the
minimum number of coloured points required to satisfy this condition, prove
that
l(n)
!1 n2 = 7 .
nlim
2
66
As a second set for this issue we give the Turkish Team Selection
Examination for the 40th IMO, 1999. Thanks again to Ed Barbeau for col-
lecting them at the IMO in Romania.
TURKISH TEAM SELECTION EXAMINATION

FOR THE 40 IMO
th
First Day - March 20, 1999
(Time: 4.5 hours)
1. Let m n be positive integers and p be a prime. Let p{expansions

of m and n be
m = a + a p + + ar pr ,
0 1
n = b + b p + + bs ps ,
0 1
respectively, where ar , bs 6= 0, for all i 2 f0, 1, : : : , rg and for all

j 2 f0, 1, : : : , sg, we have 0 ai , bj p ; 1.
If ai bi for all i 2 f0, 1, : : : , rg, we write m p n. Prove that
n (=) m n .

p6j m p
2. Let L and N be the mid-points of the diagonals [AC ] and [BD] of

the cyclic quadrilateral ABCD, respectively. If BD is the bisector of the
angle ANC , then prove that AC is the bisector of the angle BLD.
3. Determine all functions f : R ;! R such that the set

f (x) : x 6= 0 and x 2 R
x
is nite, and for all x 2 R
f (x ; 1 ; f (x)) = f (x) ; x ; 1 .
Second Day - March 21, 1999

(Time: 4.5 hours)
4. Let the area and the perimeter of a cyclic quadrilateral C be AC and

PC , respectively. If the area and the perimeter of the quadrilateral which is
tangent to the circumcircle of C at the vertices of C are AT and PT , respec-
A
tively, prove that AT PCT .
C P 2
67
5. Each of A, B, C , D, E and F knows a piece of gossip. They commu-

nicate by telephone via a central switchboard, which can connect only two of
them at a time. During a conversation, each side tells the other everything
he or she knows at that point. Determine the minimum number of calls for
everyone to know all six pieces of gossip.
6. Prove that the plane is not a union of the inner regions of nitely
many parabolas. (The outer region of a parabola is the union of the lines not
intersecting the parabola. The inner region of a parabola is the set of points
of the plane that do not belong to the outer region of the parabola.)
As a nal contest for this issue we give the Final Round of the Japanese
Mathematical Olympiad 1999. Thanks go to Ed Barbeau for collecting this
when he was Canadian Team Leader to the IMO in Romania.
JAPANESE MATHEMATICAL OLYMPIAD 1999
Final Round | February 11, 1999
Duration: 4 hours
1. You can place a stone at each of 1999 1999 squares on a grid pat-
tern. Find the minimum number of stones to satisfy the following condition.
Condition: When an arbitrary blank square is selected, the total number of
stones placed in the corresponding row and column shall be 1999 or more.
2. Let f (x) = x + 17. Prove that for each natural number n, n 2,
3
there is a natural number x, for which f (x) is divisible by 3n but not by

3n .
+1
3. Let 2n + 1 weights (n is a natural number, n 1) satisfy the

following condition.
Condition: If any one weight is excluded, then the remaining 2n weights can
be divided into a pair of n weights that balance each other.
Prove that all the weights are equal in this case.
4. Prove that
f (x) = (x + 1 )(x + 2 )(x + 3 ) (x + n ) + 1
2 2 2 2 2 2 2 2
cannot be expressed as a product of two integral-coecient polynomials with

degree greater than 1.
5. For the convex hexagon ABCDEF having side lengths that are
all 1, nd the maximum value M and minimum value m of three
diagonals AD, BE , and CF and their possible ranges.
68
Next, we turn to readers' comments and solutions to problems of the

13th Iranian Mathematical Olympiad 1995, given [1999 : 456].
1. Find all real numbers a a an satisfying
1 2
n
X n
X n
X
ai = 96 , ai = 144 ,
2
ai = 216 .
3
i=1 i=1 i=1

Solutions by Moubinool Omarjee, Paris, France; and by George
Evagelopoulos, Athens, Greece. Comments by Mohammed Aassila, Stras-
bourg, France; by Pierre Bornsztein, Pontoise, France; and by Edward T.H.
Wang, Wilfrid Laurier University, Waterloo, Ontario.
Wang notes that this is the same as CRUX with MAYHEM problem
#1982, which appeared [1994 : 250] with solution(s) [1995 : 256{257]. Born-
sztein points out the similarity to CRUX with MAYHEM problem #1838,
where a solution in positive integers is required.
2. Points D and E are situated on the sides AB and AC of triangle
ABC in such a way that DE kBC . Let P be an arbitrary point inside the
triangle ABC . Lines PB and PC intersect DE at F and G, respectively.
Let O be the circumcentre of triangle PDG and let O be that of PFE .
Show that AP ? O O .
1 2
1 2
Solution by Toshio Seimiya, Kawasaki, Japan.

A
D F H G E
O2
q
O1
q
;1K ;2
B C
Q
Let ; and ; be the circumcircles of 4PDG and 4PFE , respectively,
so that O and O are centres of ; and ; , respectively. Let Q be the
1 2
intersection of ; and ; other than P . Then O O ? PQ.

1 2 1 2
1 2 1 2
Let H and K be the intersections of AP with DE and BC , respec-

tively.
69
Since DE kBC , we get

DH : BK = AH : AK = HE : KC , so that
DH : HE = BK : KC . (1)
Since FGkBC , we similarly get
FH : HG = BK : KC . (2)
From (1) and (2), we have
DH : HE = FH : HG .
Thus,
DH HG = FH HE .
Note that DH HG and FH HE are the powers of H with respect to ;
and ; , respectively. Hence, H is a point on the radical axis of ; and ; .
1
2 1 2
Since P and Q are intersections of ; and ; , we have that PQ is the

radical axis of ; and ; . Therefore, H is a point on the line PQ, so that A,
1 2
P , Q are collinear.
1 2
Since, PQ ? O O , we have AP ? O O .
1 2 1 2
3. Let P (x) be a polynomial with rational coecients such that

P ; (Q) Q. Show that P is linear.
1
Solutions by Pierre Bornsztein, Pontoise, France; and by Moubinool

Omarjee, Paris, France. We give Bornsztein's argument.
Let P (x) 2 Q[x] such that P ; (Q) Q. It follows that
1
P (Q) Q (since P 2 Q[x]) (1)

and
P (R ; Q) RnQ ( since P ; (Q) Q) .
1
(2)
It is easy to see that P cannot be a constant polynomial.
Since P 6 0, multiply by the denominators of the coecients of P . We
obtain another polynomial with integer coecients, satisfying (1) and (2).
Moreover, if c is the leading coecient of this last polynomial, then,
using x 7! xc and multiplying by cn; (where n is the degree of P ), we
; 1
obtain a monic polynomial, with (1) and (2).

Thus, with no loss of generality, we may suppose that P 2 Z[x], where
P has leading coecient 1 (P is monic), and P satises (1) and (2).
The desired result follows immediately from this claim:
70
Claim: If P 2 Z[x] is a monic polynomial, with degree greater than 1, then

there exists an integer a such that P (x) ; a has a positive real irrational root.
Proof. Let p be a prime such that p > P (1) ; P (0) and greater than the
largest real roots of P (x) ; P (0) ; x.
Let a = p + P (0). Then P (1) ; a = P (1) ; P (0) ; p < 0 and
P (p) ; a = P (p) ; P (0) ; p > 0.
From the Intermediate Value Theorem, it follows that P (x) ; a has a
real root in (1; p), say .
Since P (x) ; a is a monic polynomial with integer coecients, it is well
known that, if is a rational root of P (x) ; a, then divides P (0) ; a = ;p.
Whence, = 1 or = p. Thus, is irrational.
This ends the proof of the claim.
It follows from the claim that if P 2 Z[x] is a monic polynomial satis-
fying (2), then P cannot have a degree greater than 1. That is, P is linear.
4. Let S = fx , x , : : : , xng be an n{element subset of the set
fx 2 R j x 1g. Find the maximum number of elements of the form
1 2
X n
"i xi , "i = 0, 1
i =1
which belong to I , where I varies over all open intervals of length 1,

and S over all n{element subsets.
Solutions by Mohammed Aassila, Strasbourg, France; by Pierre
Bornsztein, Pontoise, France; and by Moubinool Omarjee, Paris, France. We
give Bornsztein's argument.
We will prove that the desired maximum, denoted by M , is
M = nn where bc is the integer part function.

Let S be an n{element subset of (1; +1). Denote by s(A) the sum of the
elements of the subset A of S , with s(;) = 0. Let I be an open interval of
length one.
Claim: If A, B are two subsets of S with A 6 B then s(A) or s(B ) does not
belong to I .
Proof. Since A 6 B there exists x 2 B nA. Thus,
s(B) = x + s(B ; fxg) x + s(A) 1 + s(A) .
Then, s(B ) ; s(A) 1.
It follows that the numbers s(B ) and s(A) cannot belong to a common
open interval of length one.
71
From the claim, it follows immediately that the number of elements of

the form Pni "i xi , with "i = 0, 1, (that is the number of s(A) where A S )
which belong to I is not greater than the maximum number of subsets A ,
=1
A , : : : , of S which may be constructed such that none of the Ai 's is included

1
2
in another.
Such a family of subsets is a Sperner family. It is well known that a
Sperner family has at most nn2 elements (see [1]).
Since it is true for any I , S under the assumptions of the exercise, we
deduce that
M

n .
n (1)
2
Conversely: Let p be an integer such that n < p.

2
For i = 1, : : : , n, let xi = 1 + p i .
1
+
Let A be any n2 {subset of S = fx , : : : , xn g. Then,

j k
1
j k
n < s(A) = j k
n + X 1
p+i
xi 2i A
2 2
< n2 +

X 1
p
xi 2i A
=
j k
n + 1 jnk .
2 p 2
Thus,
s(A) 2 I = n ; n + 1 .
j k j k
2 2
I is clearly an open interval of length one, and I contains all s(A) where
A S with Card(A) = n . 2

Then, at least nn2 elements of the form P "i xi , "i = 0, 1, belong
to I .
It follows that
M

n .
n (2)
2

From (1) and (2), we deduce that M = nn2 , as claimed.
Reference.
[1] K. Engel, \Sperner theory", Cambridge University Press, p. 1{3.
72
5. Does there exist a function f : R ! R that fulls all of the following

conditions:
(a) f (1) = 1
(b) there exists M > 0 such that ;M < f (x) < M
(c) if x 6= 0 then
2
f x + x 2 = f (x ) + f x
1 1
?
Solution by Mohammed Aassila, Strasbourg, France.

Let n be the smallest integer for which f (x) < n for all x 6= 0. Then,
we can nd x 6= 0 such that f (x) n ; 1. Then,
2
f x = f x + x2 ; f (x) < n ; (n ; 1) = 1 ,
1 1
and thus, f ( x ) > ;1. Now, substituting x for x in the original equation,
1 1
we have

(n ; 1) f (x) = f x12 + x ; f x1 < n + 1 .
2 2
Thus, (n ; 1) < n + 1, and thus, n 2 f1, 2g. But, putting x = 1 in the

2
original equation, we get f (2) = 2, and therefore, n > 2, a contradiction.

Comment by Pierre Bornsztein, Pontoise, France. The problem and
solution can be found in \36th International Mathematical Olympiad" pub-
lished by the Canadian Mathematical Society, p. 124.
6. Let a, b, c be positive real numbers. Find all real numbers x, y, z
such that
x+y+z = a+b+c
4xyz ; (a x + b y + c z) = abc .
2 2 2

The second equation is equivalent to
a2 + b2 + c2 + abc
4 = yz zx xy xyz
and also to
4 = x +y +z +x y z ,
2
1
2
1
2
1 1 1 1
where
0 < x = payz < 2 , 0 < y = pbzx < 2 ,
1 1
0 < z = pcxy < 2 . 1

73
Setting x = 2 sin u, 0 < u < , and y = 2 sin v, 0 < v < , we have

1 2 1 2
4 = 4 sin u + 4 sin v + z + 4 sin u sin v z .

2 2 2
1 1
Hence,
z + 2 sin u sin v = 2 cos u cos v ;
1
and then,
z = 2(cos u cos v ; sin u sin v) ; = 2 cos(u + v) .
1
Thus,
a = 2p pyz sin u ,
b = 2pzx sin v ,
c = 2 xy(cos u cos v ; sin u sin v) .
From x + y + z = a + b + c, we get
;p
x cos v ; py cos u + ;px sin v + py sin u ; pz = 0
2 2
which implies
pz = px sin v + py sin u = px y + py x . 1 1
2 2
Therefore,
pz = px pb + py a ,
2 zx 2pyz
and thus, z = a b . Similarly, x = a b , y = c a .
+
2
+
2
+
2
The triple
(x; y; z) = b +2 c ; c +2 a ; a +2 b

is the unique solution.

Comment by Pierre Bornsztein, Pontoise, France. The problem and a
solution are in \36th International Mathematical Olympiad", published by
the Canadian Mathematical Society, p. 122{123.
Next we turn to the February 2000 number of the Corner and solutions
to problems of the Final (Selection) Round of the Estonian Mathematical
Contests 1995-96 given [2000 : 6].
74
1. The numbers x, y and x2 + y2 + 6 are positive integers. Prove that

xy
x2 + y2 + 6 is a perfect cube.
xy
Solutions by Mohammed Aassila, Strasbourg, France; by Pierre Born-
sztein, Courdimanche, France; and by Edward T.H. Wang, Wilfrid Laurier
University, Waterloo, Ontario. We give Wang's write-up and comment.
2 y2 + 6
Suppose x +xy = k for some positive integers x, y and k. We
prove that necessarily k = 8.
Consider
x2 + y 2 + 6 = k (1)
xy
as a Diophantine equation in two variables x and y. Let (a; b) denote the
solution of (1) with the least positive a value. We rst show that a = 1.
Due to symmetry, we may assume that a b. Note rst that a and b must
both be odd since they clearly must have the same parity and if they are both
even, then modulo 4, a + b + 6 2 while kab 0. If a = b, then from
2 2
k = 2 + a2 we deduce immediately that a = 1. If a < b then b a + 1.

6
Since
ka ; b)2 + a2 + 6 = k2 a2 ; 2kab + kab = k
(
(ka ; b)a ka2 ; ab
and
2+ 2+6 2+6
ka ; b = a bb ; b = a b > 0,
we see that (ka ; b; a) is also a solution of (1) in natural numbers. Note2
that
a +6;ab a +6;a(a+1) = 6;a < 0 if a > 6. Thus, ka;b = b < a
2 2 a+6
if a > 6, contradicting the minimality of a. Hence, a 6. It remains to show

that a =6 3, 5. If a = 3, then we get b + 15 = 3kb, which implies that 3
2
divides b and thus, b + 15 6, while 3kb 0 (mod 9). If a = 5, then we

2
get b +31 = 5kb. Since b is an integer and 31 is a prime we deduce from the
2
relation between roots and coecients that 5k = 32, which 2

is impossible.
b
Therefore, we conclude that a = 1, which implies k = b = b + b . +7 7
Hence, b = 7 and k = 8, which is a cube.

Comment. This is a beautiful problem which parallels the following
\infamous" problem of the 29th IMO, supposedly the most dicult IMO
problem ever (with an average score of 0:6 out of 7):
\Suppose a and b are2positive
2
integers such that ab + 1 divides
a +b
a + b . Prove that ab + 1 is a perfect square."
2 2
75
2. Let a, b, c be the sides of a triangle and , , the opposite angles

of the sides, respectively. Prove that if the inradius of the triangle is r, then
a sin + b sin + c sin 9r.
Solutions by Mohammed Aassila, Strasbourg, France; by Michel Bataille,
Rouen, France; by Pierre Bornsztein, Courdimanche, France; by Murray S.
Klamkin, University of Alberta, Edmonton, Alberta; by Hojoo Lee, student,
Kwangwoon University, Kangwon-Do, South Korea; by David Loeer, stu-
dent, Cotham School, Bristol, UK; by Toshio Seimiya, Kawasaki, Japan; and
by Panos E. Tsaoussoglou, Athens, Greece. We rst give Loeer's solution.
First we substitute sin() = 2aR , etc., so that the required inequality
is equivalent to
2 2 2
9r 2aR + 2bR + 2cR , or 18Rr a + b + c .
2 2 2
The product Rr can be neatly expressed by comparing two well-known for-

mulae for the area of the triangle: abc
4R
and rs = r(a +2b + c) . Equating
these gives Rr = 2(a +abcb + c) .
Thus, the required inequality becomes:
9abc
a+b+c a +b +c ,
2 2 2
9abc (a + b + c )(a + b + c) .
2 2 2
However, applying the AM-GM inequality, we see that

2 2 2
(abc) 23 a + b3 + c ;
that is,
3(abc) 23 a + b + c , 2 2 2
and
(abc) 13 a + 3b + c , or 3(abc) 13 a + b + c .
Multiplying these together it follows that 9abc (a + b + c)(a + b + c ), 2 2 2
which we showed to be equivalent to the required inequality.

Now we give Klamkin's generalization and comment.
Since a = 2R sin , etc., where R is the circumradius, the inequality is
now
a + b + c 18Rr .
2 2 2
We prove the stronger inequality

a + b + c xRr ; (2x ; 36)r 18Rr ,
2 2 2 2
where 18 x 24. The right-hand inequality reduces to (x;18)(R;2r)

0
76
which is the well-known Chapple-Euler inequality. For the left-hand

inequality, we use an identity [1; p. 52].
a + b + c = 2(s ; 4Rr ; r ) xRr ; (2x ; 36)r
2 2 2 2 2 2
where s is the semiperimeter. This reduces the inequality to

2s (x + 8)Rr + (38 ; 2x)r .
2 2
Since it is known [1; p. 50] that s 16Rr ; 5r, we must have here that
2
32Rr ; 10r (x + 8)Rr + (38 ; 2x)r

2 2
or that (24 ; x)Rr (48 ; 2x)r and which is valid if x 24 as well.

2
Comment. As a known complementary inequality, we also have

9R a + b + c .
2 2 2 2
For one simple proof of this consider the expansion of (A + B + C ) 0 2
where A, B , C are vectors from the circumcentre to the vertices A, B , C of

the triangle ABC (note that A = R , 2B C = 2R ; a , etc.). Another
2 2 2 2
proof follows from OH = 9R ; (a + b + c ) where O and H are the

2 2 2 2 2
circumcentre and orthocentre of ABC .

Reference:
[1] D.S. Mitrinovic, J.E. Pecaric, V. Volenic, Recent Advances in Geometric
Inequalities, Kluwer, Dordrecht, 1989.
3. Prove that the polynomial Pn(x) = 1 + x + x2 + x3 + + xnn has
no zeros if n is even and has exactly one zero if n is odd.
2 6 !

Rouen, France; by Pierre Bornsztein, Courdimanche, France; and by Murray
S. Klamkin, University of Alberta, Edmonton, Alberta. We give the write-up
of Bataille (with his added supplementary remark).
We shall prove by induction the following property
(
P n has no (real) zeros
(n ) 2
P n has exactly one (real) zero.

2 +1
Clearly ( ) is true since P (x) = 1 and P (x) = 1 + x.

0 0 1
Suppose now that (n ) is true for an integer n 0. We will denote by

xn the unique zero of P n . The continuous function P n has no zeros and
is positive for x 0. Hence, P n (x) > 0 for all x.
2 +1 2
Since the derivative P 0n of P n is P n , the function P n is in-

lim P n (x)= ;1 and x!lim1 P n (x)=+1,
2 +1 2 2 +1
creasing. Furthermore, x!;1
2 +1
2 +1 2 +1
so that P n (x) < 0 for x < xn and P n (x) > 0 for x > xn .
+
2 +1 2 +1
77
Now, using P 0n = P n , we see that P n (x) is a minimum when

x = xn . Hence, for all x
2 +2 2 +1 2 +2
2n+2
P n (x) P n (xn) = P n (xn ) + xnn
2 +2 2 +2 2 +1
(2 + 2)!
= x2nn+2 (
n+1 )2
= (2xnn+ 2)! > 0.
(2 n + 2)!
Thus, P n (x) never takes the value 0.
2 +2
Also, as above, P n is increasing, continuous, and x!;1

2 +3 lim P n (x)= ;1, 2 +3
x!lim1 P n (x)=+1. Hence, P n has a unique zero. This completes the

2 +3 2 +2
induction and shows that property (n ) is true for all non-negative integer n.
+
Remark. As a supplement, we show that nlim !1 xn = ;1.

Consider
n+1 n
P n (;2n;3) =
X
2
(; n;
2 3)
k
=
X
(2n+3) p 2 1
; (2pn++1)!
2
3
.
2 +1
k=0 k !
p=0 (2p)!
We have P n (;2n ; 3) < 0 (since 2n + 3 > 2p + 1 for p = 0; : : : ; n) and

thus, xn > ;2n ; 3. Hence,
2 +1
2n+2
P n (xn ) = P n (xn) ; P n (xn ) = xnn 1 + nxn > 0 ,

2 +3 2 +3 2 +1
(2 + 2)! 2 +3
which implies xn > xn . Therefore, (xn ) is a decreasing sequence of real

numbers and, as such, either limn!1 xn = ;1 or (xn ) converges to a real
+1
number m. Assume that the latter does occur. Note that m xn < 0 for
all n. Since for all n 0 we have P n (x) ex P n (x) for all x 0
2 +1 2
[easy induction], we would have
2n+1
0 exn P n (xn) = P n (xn ) ; (2xnn+ 1)!
2 2 +1
2n+1 2n+1
= ; (2xnn+ 1)! ; (2mn + 1)! .
But then, limn!1 exn = 0 (because nlim m2n+1 = 0) while, by the

!1 (2n + 1)!
!1 e = e 6= 0.
continuity of the exponential function, we must have nlim xn m
!1 xn = ;1.
This contradiction shows that nlim
4. Let H be the orthocentre of an obtuse triangle ABC and A , B ,
C arbitrary points taken on the sides BC , AC , AB, respectively. Prove
1 1
that the tangents drawn from the point H to the circles with diameters AA ,
1
BB , CC are equal.
1
1 1
78
Solutions by Michel Bataille, Rouen, France; and by Toshio Seimiya,

Kawasaki, Japan. We give Seimiya's solution.
B
;2
r r
E t2 r
C1
r
D A r r
H r
B1 t1
r
;1
A1 r
r
t3
F r
;3 r
C
The lines HA, HB , HC meet BC , CA, AB at D, E , F , respectively.
Then, AD ? BC , BE ? AC and CF ? AB .
Let ; , ; , ; be circles with diameters AA , BB , CC , respectively.
1 2 3 1 1 1
Since \ADA = \BEB = CFC = 90 , we have that ; , ; , ;

pass through D, E , F , respectively.
1 1 1 1 2 3
Let t , t , t be the tangent segments from H to ; , ; , ; , respec-

1 2 3 1 2 3
tively.
Since \ADB = \AEB = 90 , A, D, B , E are concyclic, so that
HA HD = HB HE .
Hence, we have
t = HA HD = HB HE = t .
2
1
2
2
Thus, t = t .
1 2
Similarly, we have t = t . Therefore, t = t = t .

1 3 1 2 3
5. Find all functions f : R ! R satisfying the following conditions for

all x 2 R.
(a) f (x) = ;f (;x);
(b) f (x + 1) = f (x) + 1;

(c) f x1 = x12 f (x), if x 6= 0.
79

Rouen, France; and by Pierre Bornsztein, Courdimanche, France. We give the
write-up by Aassila.
Set g(x) = f (x) ; x. Then g satises
(a') g(x) = ;g(;x);
(b') g(x + 1) = g(x);

(c') g x1 = x12 g(x), if x 6= 0.
By (a'){(b') we get g(0) = g(;1) = 0, and for all x 6= 0, ;1 we have

g(x) = g(x + 1) = (x + 1) g x 2
= ;(x + 1) g ; x
1
+1
2 1
+1
= ;(x + 1) g x x

= ;(x + 1) g 1 ; x
2 1
+1
2
+1
2
= ;(x + 1) x x 2 g x x = ;x g 1 + x = ;x g x

2 +1 2 1 2 1
( + 1)
= ;g(x) .
Hence, g(x) 0, and f (x) = x for all x 2 R.
Now we turn to solutions from our readers to problems of the Japan

Mathematical Olympiad, Final Round, 1996 given [2000 : 7].
2. For positive integers m, n with gcd(m; n) = 1, determine the value
gcd(5m + 7m ; 5n + 7n).
Rouen, France; by Pierre Bornsztein, Courdimanche, France; and by George
Evagelopoulos, Athens, Greece. We give Bornsztein's solution.
We will use the following lemma (see [1] for a proof).
Lemma. If a and b are relatively prime integers with a > b, then for every
pair of positive integers m and n we have
(am ; bm ; an ; bn ) = a m;n ; b m;n( ) ( )
where (x; y) denotes the gcd of integers x and y.

Let m, n be positive integers, with (m; n) = 1.
Let Sm = 5m + 7m , Sn = 5n + 7n , S = (Sm ; Sn ) and Ua = 7a ; 5a
for a 2 N .
If m = n = 1, we have S = Sm = Sn = 12.
Since (m; n) = 1 and m, n are playing symmetric parts, we may
suppose that m > n.
80
Let m = a + n. Then, a 2 N and (a; n) = 1.

We have
Sm ; 5a Sn = 7nUa . (1)
Since 5 and 7 are relatively prime, it is easy to see that (S ; 5a ) = 1 and
(S ; 7n) = 1.
Then, from (1), we deduce that S = (Ua ; Sn ).
Case 1. a is odd.
Let Sn = Sl, Ua = S k, where k, l 2 N and (k; l) = 1. Then, using
the binomial expansion
7an = (S l ; 5n )a = S L ; 5an
and
7an = (S k + 5a )n = S K + 5an
where K , L are integers.
Thus,
S (K ; L) = 2 5an .
It follows that S divides 2 5an . Since (S ; 5) = 1, we deduce (using Gauss'
theorem) that S divides 2.
Moreover, Sm , Sn are even. Thus, 2 divides S . Then, S = 2.
Case 2. a is even.
Let a = 2b, where b 2 N and (b; n) = 1.
Moreover, m and n are odd, since m = 2b + n and (m; n) = 1. Then
S divides SnUn = U n = 7 n ; 5 n and S divides Ua = 7 b ; 5 b .
2
2 2 2 2
From the lemma, we deduce that S divides

7 ; 5 = 24 .
2 2
(2)
Since m, n are odd, we have 7m ;1 (mod 8) and 5m 5 (mod 8).
It follows that 4 divides Sm and Sn , but 8 does not divide either Sm
or Sn .
In the same way, Sm Sn 1m + (;1)m 0 (mod 3). Since 3 and
4 are relatively prime, it follows that
Sm Sn 0 (mod 12)
but
Sm 6 0 (mod 24) .
81
From (2), we may conclude that S = 12. Then:

if m, n are odd, we have (Sm ; Sn ) = 12;
if m, n have opposite parities, we have (Sm ; Sn ) = 2.
Reference.
[1] Math. Magazine, Exercise no 1091, vol. 54, no. 2, March 1981, pp. 86{87.
3. Let x be a real number with x > 1 and such that x is not an integer.
Let an = bxn c ; xbxn c (n = 1, 2, 3, : : : ). Prove that the sequence
+1
of numbers fan g is not periodic. (Here byc denotes, as usual, the largest
integer y.)
Solutions by Mohammed Aassila, Strasbourg, France; and by Michel
Bataille, Rouen, France. We give Bataille's solution.
Suppose for purpose of contradiction, that there exists a positive integer
p such that ann pp = ann for all positive integers n. Dening the integer un
by un = bx c ; bx c, we would have un = xun for all n. We now
+
+
+1
distinguish the following mutually exclusive cases:
Case 1. u = bxp c ; bxc 6= 0.
+1
Then, x = uu12 is a rational number that we can also write as x = kl

1
where k, l are coprime integers. Note that k > l > 1 (since x > 1 and
x 62 N). For alln n,n un = xnu so that lnun = knu and, ln being
coprime with k , l divides u . This would mean that u has an innite
+1 1 +1 1
number of divisors (since l > 1), which is clearly impossible.

1 1
Case 2. u = bxp c ; bxc = 0.

1
+1
Then, un = 0 for all n, and an easy induction shows that, for all positive
integers m,
bxmp c = bxc , bxmp c = bx c , : : : , bxmp
+1 +2 2 +p;1 c = bxp; c ) (1)
1
and bxmp c = bxp c .

However, from x < x < < xp < xp , we get
2 +1
bxc bx c bxpc bxp c

2 +1
and since bxp c = bxc, we have bxr c = bxc for r = 1, 2, : : : , p + 1.

+1
Actually, we even have bxn c = bxc for all positive integers n (divide n
!1 x = +1 and thus,
by p and use (1)). Again this is impossible since nlim n
!1bx c = +1.
n
nlim
Thus, in both cases we are led to an impossibility, and fan g cannot be
periodic.
82
p
5. Let q be a real number such that 1 +2 5 < q < 2. When we
represent a positive integer n in binary expansion as
n = 2k + ak; 2k; + + a 2 + a
1
1
1 0
(here ai = 0 or 1), we dene pn by
pn = qk + ak; qk; + + a q + a :
1
1
1 0
Prove that there exist innitely many positive integers k which satisfy
the following condition: There exists no positive integer l such that
p k < pl < p k .
2 2 +1

By induction on n, we can prove that k = qn satises the required
condition, where qn is dened by
8 m
X
>
>
>
>
< qm =
2 2k2
km
=0
X
>
>
>
>
: qm
2 +1 = 2k .
2 +1
k=0
That completes the Corner for this issue of CRUX with MAYHEM. We
are entering Olympiad season. Send me your nice solutions and generaliza-
tions as well as Olympiad Contests.
Professor Toshio Seimiya

Regular readers of this section will be aware of the many beautiful
geometry problems that have been proposed by Professor Toshio Seimiya.
We dedicated some to him in March 2001 [2001 : 114], in celebration of his
90th birthday. Unfortunately, we did not have a photograph available then.
We do now! See page 100.
83
BOOK REVIEWS
JOHN GRANT McLOUGHLIN
Teaching Statistics: Resources for Undergraduate Instructors
edited by Thomas L. Moore, published by the Mathematical Association of
America (MAA Notes Series, #52), 2000,
ISBN 0-88385-162-8, softcover, 222+xii pages, $31.95 (U.S.).
Reviewed by C.L. Kaller, retired Professor of Mathematics, Kelowna, BC.
This instructors' handbook is an eclectic collection of articles (some
published for the rst time, but many reprints of previously published ar-
ticles or commentaries on such articles) on a number of aspects of introduc-
tory undergraduate statistics instruction. It is a compendium of information
intended to be regarded essentially as an instructors' manual for teachers of
statistics courses to undergraduate or even senior secondary school students.
Particular targets of this volume are those teachers of statistics courses who
have limited formal training in that discipline.
The volume consists of articles in six categories:
1. Hortatory Imperatives [of data-based statistical instruction]
2. Teaching with Data [in classroom settings]
3. Established Projects in Active Learning [with usage guidelines]
4. Textbooks [with detailed textbook selection procedures]
5. Technology [resources available for classroom presentations]
6. Assessment [of student achievement].
The motivation behind the choice of articles in each of these categories is rel-
evance to what the editor feels is the current (2000) thinking about just what
constitutes acceptable statistics instruction, particularly when such instruc-
tion is provided by teachers whose formal professional training in statistics
is somewhat limited. The editor points out that the volume is far from an
all-inclusive presentation of resources and ideas on statistics education; it is
his stated hope, however, that the material will provide direction to readers
and to have them keep alert to other instructional resources constantly being
developed.
Laudatory as is the intent of the editor in publishing the contents of
this handbook to help the classroom statistics instructor, this publication is
not without some irritating editorial aws giving this reviewer the distinct
impression that the volume was thrown together in a hasty rush to the print-
ers. Grammar, spelling, consistency and format have not received the careful
editorial attention expected in an MAA publication. Indeed, on the back
cover of the book (as well as elsewhere inside) even the word statistics is
misspelled!
84
Very irritating is the sloppiness in grammar, especially in the use of

the word data which appears inconsistently presented as both a plural and
a singular noun. Indeed, within the same paragraph in the volume, this
word appears as a singular noun in one sentence and plural in another, and
authors inconsistently refer to these data and this data quite interchangeably.
Careful editing of the articles accepted for this publication would at least have
resulted in consistently using (incorrectly, asserts this reviewer) the word
data in singular form if it were impossible to use it correctly as a plural noun.
In the editor's own contribution (on page 31, Implications for Statistics
Teaching) it is stated that \Statistics faculty need to learn new skills. We
now have to grade grammar and writing style as well as statistical thinking."
In a volume intended to give advice and guidance to statistics teachers, it is
regrettable that the editor/author failed to take his own pronouncement to
heart to produce a better quality contribution to the MAA Notes Series.
Gardner's Workout: Training the Mind and Entertaining the Spirit

by Martin Gardner, published by A.K. Peters, 2001.
ISBN 1-56881-120-9, hardcover, 319 + xi pages, $35.00 (U.S.).
Reviewed by Edward J. Barbeau, University of Toronto, Toronto, On-
tario.
Martin Gardner has been part of my mathematical environment since I
discovered his Scientic American columns as an undergraduate and was cap-
tivated by his easy style and elegant problems. Even though he has been suc-
ceeded by others in Scientic American, he has remained very much in view
through appearances in television specials, issues and reissues of his essays,
continuing interest of others in problems that he was the rst to publicize
and articles in a number of journals.
This book is an anthology of 34 articles and 7 reviews that have ap-
peared, mostly during the 1990s, in Quantum, Math Horizons and other
journals of general mathematical interest. An article on Kasparov's defeat
by Deep Blue rst appeared in The Washington Post; four reviews of books
were published in The Los Angeles Times; an essay on the growth of recre-
ational mathematics appeared in Scientic American, although that was a
\heavily revised and cut version" of what appears in this book. Despite the
variety of topics, there is a linkage from many essays to the next that pro-
vides a coherent ow of ideas for the reader who, like myself, progressed
from end to end. Prominent are articles on number play, geometry, graph
theory, articial intelligence, puzzles, games and tricks.
Occasionally Gardner becomes more serious. In a pair of articles on
articial intelligence, Gardner dissents from the views of those who would
see us on the threshold of reproducing human thinking, and puts forward his
belief that the mind is on a much higher plan of complexity and is qualita-
tively dierent from anything that has been produced so far on computers. A
85
review of the 1997 NCTM Yearbook, an algebra text and the PBS videoseries
Life by the Numbers in The New York Review of Books provides the occa-
sion to vent his displeasure at the \fuzzy new math". He skewers it for its
faddishness and dilution with much that is mathematically irrelevant. If the
Yearbook is any guide, reformers seem to be completely ignorant of fasci-
nating and challenging mathematical material available in a number of ne
books published in recent years by established mathematicians. It is hard to
argue with his assessment of the PBS series as \high on special eects, low
on mathematical content". This is an important article, that well deserves
its more permanent place in this volume.
Those who want to enliven the modern school classroom will nd many
riches in this book. Four successive chapters treat, in whole or in part, magic
squares, with some new material. A prize of $100 has not yet been claimed
for a 3 3 magic square whose entries are all squares. There are many
squares that almost do the job, but getting the complete solution is equiv-
alent to nding rational points on certain cubic curves. There are numerous
dissection problems, a genre which I have found encourages my own under-
graduate students to think of geometry in more structural terms. A Quantum
article deals with decomposing both squares and equilateral triangles into
three similar parts, all three of which, two of which and none of which are
congruent. Is it true that there is a unique solution to the triangle problems
with one and no congruent pairs of parts? In the magazine, Cubism for Fun,
Gardner oered a $50 prize for anyone who could, for any integer n exceeding
1, cover the surface of a cube by n congruent polygons without overlapping.
In an addendum, he tells us that this was won by a reader, Anneke Treep,
who without any technical complications applied the right perspective and
degree of imagination.
Other problems, however, are denitely not for the classroom. Con-
sider the problem, due to Gardner himself from the initial chapter of the
book; it asks for the minimum area of a surface placed inside a transparent
cube to render it opaque from any direction. The best answer seems to be
4:2324. \I believe the opaque cube problem to be extremely dicult," he
writes, \it is keeping me awake at nights."A substantial essay treats mini-
mal Steiner trees (spanning trees of minimum length) on a rectangular array
of nodes, and includes a table of the best known results for n n when
2 n 14.
The pace is varied by a discussion of directed graphs to analyze propo-
sitional calculus, and by some word play. Can you provide a square array of
nine letters for which, each row, column and diagonal, spell out a word? Or
provide a chain of words, each altered by a single letter from its predecessor,
that converts BLACK to WHITE?
The eectiveness of this book derives in large part from the passion
with which Gardner shares his mathematical enthusiasm and on the breadth
and erudition of his discussions. This is another winner!
86
On a \Problem of the Month"

Murray S. Klamkin
In the problem of the month [1999 : 106], one was to prove that
p p p p
a + b ; c + b + c ; a + c + a ; b pa + b + pc ,
where a, b, c are sides of a triangle.
It is to be noted that this inequality will follow immediately from the
Majorization Inequality [1]. Here, if A and B are vectors (a ; a ; : : : ; an ),
(b ; b ; : : : ; bn ) where a a an , b b bn, and
1 2
a b , a + a b + b , : : : , a + a + + an; b + b + + bn; ,
1 2 1 2 1 2
a + a + + an = b + b + + bn, we say that A majorizes B and

1 1 1 2 1 2 1 2 1 1 2 1
write it as A B. Then, if F is a convex function,

1 2 1 2
F (a ) + F (a ) + + F (an ) F (b ) + F (b ) + + F (bn ) .
1 2 1 2
If F is concave, the inequality is reversed.

For the triangle inequality, we can assume without loss of generality
that a b c. Then a + b ; c a, (a + b ; c) + (a + c ; b) a + b,
and (a + b ; c) + (a + c ; b) + (b + c ; a) = a + b + c. Therefore, if F is
concave,
F (a + b ; c) + F (b + c ; a) + F (c + a ; b) F (a) + F (b) + F (c)

(for the given inequality, F = px is concave).
As to the substitution a = y + z, b = z + x, c = x + y which was
used in the referred to solution and was called the Ravi Substitution, this
transformation was known and used before he was born. Geometrically, x,
y, z are the lengths which the sides are divided into by the points of tangency
of the incircle. Thus, we have the following implications for any triangle
inequality or identity:
F (a; b; c) 0 (=) F (y + z; z + x; x + y) 0 ,
F (x; y; z) 0 (=) F ((s ; a); (s ; b); (s ; c)) 0
(here s is the semiperimeter). This transformation eliminates the trouble-
some triangle constraints and lets one use all the machinery for a set of three
non-negative numbers.
Copyright
c 2002 Canadian Mathematical Society
87
Another big plus for the Majorization Inequality is that we can obtain
both upper and lower bounds subject to other kinds of constraints. Here are
two examples:
(1) Consider thePbounds on sin a + sin a + + sin an where n 4,
ai 0 and ai = S 2. Since 1 2

; ; ; ; 0; 0; : : : ; 0 (a ; a ; : : : ; an ) S ; S ; : : : ; S ,
2 2 2 2
1 2
n n n
we have
4 sin a + sin a + + sin an n sin Sn .

1 2
(2) Consider the bounds on a + a + + an where ai = S ( n) and

2 2
P 2
the ai 's are positive integers. Since

1 2
(S ; n + 1; 1; 1; ; 1) (a ; a ; ; an) Sn ; Sn ; ; Sn ,

1 2
we have
(S ; n + 1) + n ; 1 a + a + + an n Sn .
2
2 2 2 2
1 2
For many other applications, see [1].

Reference
1. A.W. Marshall, I. Olkin, Inequalities: Theory of Majorization and
Its Applications, Academic Press, NY, 1979.
Murray S. Klamkin
University of Alberta
Edmonton
Alberta
Canada T6G 2G1
88
Substitutions, Inequalities, and History

Shay Gueron
The Non-Ravi Substitution

A solution to the inequality (1996 Asian Pacic Mathematical Compe-
tition)
p
a + b ; c + a ; b + c + ;a + b + c pa + b + pc (1)
p p p
where a, b, c are the sides of a triangle, appeared in Crux [1999 : 106]. It
starts with the substitution a = x + y, b = y + z, c = z + x (x, y, z > 0).
This substitution is referred to as the \Ravi Substitution" and reported to be
known by this name, at least in Canadian IMO circles.
It seems that this awkward credit for the substitution diused to wider
circles. The same inequality (1) appears in a French problem solving book
from 1999 [9, p. 146]. Although the solution proposed in [9] is dierent,
it starts with the same substitution which, amazingly, is called there too
the \Ravi Substitution". Further, [9] includes several other mentions and
applications of the \Ravi Substitution" [9, pp. 130, 146, 147, 155, 237].
In response to Crux [1999 : 106], Klamkin [6] comments telegraphically
that the substitution a = x + y, b = y + z, c = z + x is a classical technique
that was known and used before he (Ravi) was born. It is time to set things
straight.
Although Klamkin gives no reference in his piece, he is right; this sub-
stitution is one of the rst strategies taught in Olympiad training classes,
at least | a la Crux | in Israeli circles. But since even folklore should be
supported by some written evidence, can we nd such support? The answer
is positive. One example appears in Engel's book (an English translation of
previous German versions) [2, p. 178]: the substitution a = x + y, b = y + z,
c = z + x with x, y, z > 0 is advice number 7 in Engel's list of 18 possi-
ble strategies for proving inequalities. It is explained in [2, p. 164] that this
substitution is merely another manifestation of the triangle inequality. The
same explanation appears in more detail in [8, Chapter II, p. 26], a book
devoted to geometric inequalities. It provides some references to papers
where such equivalent forms of the triangle inequality are mentioned (see
pp. 35{36). One of these references, namely [5], is a paper from 1971 |
written by Klamkin.
Consequently, even without digging into earlier references (which are
probably easy to nd) Klamkin's remark is evidently correct.
Copyright
c 2002 Canadian Mathematical Society
89
Karamata and the Majorization Inequality

We continue with a historical mood. Klamkin [6] proposes to solve
(1) by using the Majorization Inequality. This inequality relates to two se-
quences a a an and b b bn and states that:
a + + ai b + + bi for 1 i n ; 1 and a + + an = b + + bn
1 2 1 2
if and only if f (a )+ + f (an ) f (b )+ + f (bn ) for any convex func-

1 1 1 1
tion f (x). Clearly, this inequality includes Jensen's Inequality as a special

1 1
case. The proof of one direction is easy, and the more intricate part can be
proved by applying the Abel Summation Formula
n
X n
X j
X
!
ai bi = ai (bj ; bj ) , +1
i=1 j =1 i=1
setting bn = 0.
+1
The Majorization Inequality is well known, but unfortunately, this

generic name does not reveal its source: this inequality is due to Kara-
mata, 1932 [4], and should therefore be called the Karamata Inequality, as
in [1, pp. 31-32]. It turns out to be a strong tool with various applications,
some of which can be found in [7] and [8, Chapter VIII]). We also note that
before [6], the inequality (1) appeared in [3] (a paper in Hebrew) as an
example (probably well known even before) of a case where the Karamata
Inequality is a useful approach.
Substitution and the Karamata Inequality
We conclude with an example in the spirit of [6], where a substitution
followed by the application of the Karamata Inequality leads to a solution.
This is IMO 2000 Problem 2 (as solved by one of the Israeli contestants, Eran
Assaf; the problem has some half dozen solutions): Let a, b, c be positive
numbers with abc = 1. Prove that

a ; 1 + 1b b ; 1 + 1c c ; 1 + a1 1 . (2)
The substitution a = x=y, b = y=z, c = z=x (x, y, z > 0) converts (2) to
(x ; y + z)(y ; z + x)(z ; x + y) xyz . (3)
Without loss of generality, assume that x y z, so that (y + z ; x),
(z + x ; y) > 0. If x + y ; z 0 (3) follows immediately, so that we may
assume that x + y > z as well.
Since x x + y ; z, x + y (x + y ; z) + (z + x ; y), and
x + y + z = (x + y ; z) + (z + x ; y) + (y + z ; x), we can apply the
Karamata Inequality to the triplets (x; y; z) and (x+y ;z; x+z ;y; z +y ;x),
and obtain (3) by writing
f (x) + f (y) + f (z) f (x + y ; z) + f (z + x ; y) + f (y + z ; x)
90
for the convex function f (x) = ; ln x.

Returning to history, it turns out that inequality (3) to which the IMO
2000 problem is reduced/equivalent, is not really a new one: it is due to
A. Padoa, in 1925 (Period. Mat. (4)5 : 80{85). Moreover, (3) is equivalent to
a (b + c ; a) + b (c + a ; b) + c (a + b ; c) 3abc, which is IMO 1964
2 2 2
Problem 2. Could you guess what substitution is helpful for proving these
inequalities easily?
References
[1] E. Beckenbach and R. Bellman. Inequalities. Springer, Berlin (1965).
[2] A. Engel Problem solving strategies. Springer, Berlin (1997).
[3] S. Gueron and R. Tessler. Majorization and the Karamata Inequality.
Etgar-Gilionot Mathematica 48-49: 4-10 (1999).
[4] J. Karamata. Sur une inegalite relative aux fonctions convexes Publ.
Math. Univ. Belgrade 1:145-148 (1932).
[5] M.S. Klamkin. Duality in triangle inequalities. Notices of the
Amer. Math. Soc. p. 782 (1971).
[6] M.S. Klamkin. On a Problem of the Month. Crux Mathematicorum,
Vol 28, p. 86 (2002) .
[7] A.W. Marshall and I. Olkin. Inequalities: Theory of Majorization and
its Applications. Academic Press, N.Y. (1979).
[8] D.S. Mitrinovic, J.E. Pecaric and V. Volenec. Recent advances in geo-
metric inequalities. Kluwer, London (1989).
[9] T.B. Soulami. Les Olympiades de mathematiques ; Re exes et
strategies. Ellipse, Paris (1999).
Shay Gueron
University of Haifa
Israel
[email protected]
91
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, 977 Orleans Blvd.,
Gloucester, Ontario, Canada. K1C 2Z7 (NEW!). The electronic address is
[email protected]
The Assistant Mayhem Editor is Chris Cappadocia (University of Wa-

terloo). The other sta member is Jimmy Chui (University of Toronto).
MAYHEM PROBLEMS
Envoyez vos propositions et solutions a MATHEMATICAL MAYHEM,
Faculte de mathematiques, Universite de Waterloo, 200 University Avenue
West, Waterloo, ON, N2L 3G1, ou par courriel a
[email protected]
N'oubliez pas d'inclure a toute correspondance votre nom, votre annee sco-
laire, le nom de votre e cole, ainsi que votre ville, province ou e tat et pays.
Nous sommes surtout interesses par les solutions d'etudiants du secondaire.
Veuillez nous transmettre vos solutions aux problemes du present numero
avant le 1er septembre 2002. Les solutions recues apres cette date ne seront
prises en compte que s'il nous reste du temps avant la publication des solu-
tions.
Chaque probleme sera publie dans les deux langues ocielles du Canada
(anglais et francais). Dans les numeros 1, 3, 5 et 7, l'anglais precedera le
francais, et dans les numeros 2, 4, 6 et 8, le francais precedera l'anglais.
Pour e^ tre admissibles au DEFI MAYHEM de ce mois-ci, les solutions doivent
avoir e te postees avant le 1er juillet 2002, cachet de la poste faisant foi.
34. Propose par l'equipe de Mayhem.
Les nombres 1 a 2002 sont e crits au tableau noir et l'on decide de jouer
au jeu suivant :
On lance une piece de monnaie et on eace deux nombres x et y du
tableau. Si l'on tombe sur pile, on e crit x + y au tableau, sinon on e crit
jx ; yj ; on continue le processus jusqu'a ce qu'il ne reste plus qu'un nombre.
Montrer que ce dernier nombre est impair.
.................................................................
92
The numbers 1 to 2002 are written on a blackboard so you decide to

play a fun game. You ip a coin, then erase two numbers, x and y, from the
board. If the coin was heads you write the number x + y on the board, if the
coin was tails you write the number jx ; yj. You continue this process until
only one number remains. Prove that the last number is odd.
On denit deux suites par x = 4732, y = 847, xn = xn yn and +
yn = xnxnyynn . Trouver
1 1 +1 2
2
+1 +
!1 xn
nlim et !1 yn .
nlim
.................................................................
Two sequences are dened by: x = 4732, y = 847, xn = xn yn
+
and yn = xnxn yynn . Find

1 1 +1 2
2
+1 +
!1 xn
nlim and !1 yn .
nlim
Dans un triangle ABC , soit AM la mediane issue du sommet A. Mon-
trer que AM AB AC . +
2
.................................................................
In 4ABC , AM is the median from A. Prove AM AB AC . +
2
37. Propose par J. Walter Lynch, Athens, GA, USA.

Trouver deux entiers positifs dierents, plus petits que 100, et tels que
la somme des chires des deux entiers soit e gale au plus grand et que produit
de leurs chires soit e gal au plus petit.
.................................................................
Find two (dierent) positive integers less than 100 such that the sum of
the digits in both integers is the larger integer and the product of the digits
in both integers is the smaller integer.
Trouver toutes les valeurs de n telles que 1! + 2! + 3! + + n! soit
un carre parfait.
.................................................................
Find all values of n such that 1! + 2! + 3! + + n! is a perfect square.
93
Challenge Board Solutions

Editor: David Savitt, Department of Mathematics, Harvard University,
1 Oxford Street, Cambridge, MA, USA 02138 <[email protected]>
In this issue we present some of the solutions to the Konhauser Prob-
lemfest presented in the April 2001 issue [2001 : 204].
1. Last season, the Minnesota Timberwolves won 5 times as many
games as they lost, in games in which they scored 100 or more points. On the
other hand, in games in which their opponents scored 100 or more points,
the Timberwolves lost 50% more games than they won. Given that there
were exactly 34 games in which either the Timberwolves or their opponents
scored 100 or more points, what was the Timberwolves' win-loss record in
games in which both they and their opponents scored 100 or more points?
Solution: Let the desired win-loss record be a ; b, so that of the games
when both the Timberwolves and their opponents scored at least 100 points,
the Timberwolves won a and their opponents won b. Also, let c be the num-
ber of games in which only the Timberwolves scored at least 100 points and
d be the number of games in which only their opponents scored at least 100
points. Thus, we have
a + c = 5b , (1)
b + d = 1:5a , (2)
a + b + c + d = 34 . (3)
Thus, we get c = 5b ; a, and d = 1:5a ; b from (1) and (2), so that (3) yields
3a + 10b = 68.
Now a and b are non-negative integers, and in particular b 6 and
68 ; 10b is divisible by 3, which is true only for b = 2 and b = 5. If b = 2,
then a = 16, c = 5b ; 1 = ;6 < 0, which is a contradiction. Thus, we must
have b = 5 and a = 6. The Timberwolves were 6 ; 5 in games in which both
they and their opponents scored at least 100 points.
2. Three circles are drawn in chalk on the ground. To begin with, there
is a heap of n pebbles inside one of the circles, and there are \empty heaps"
(containing no pebbles) in the other two circles. Your goal is to move the
entire heap of n pebbles to a dierent circle, using a series of moves of the
following type. For any non-negative integer k, you may move exactly 2k
pebbles from one heap (call it heap A) to another (heap B), provided that
heap B begins with fewer than 2k pebbles, and that after the move, heap A
ends up with fewer than 2k pebbles. Naturally, you want to reach your goal
in as few moves as possible. For what values of n 100 would you need the
largest number of moves?
Solution: This is a close relative of the classic \Towers of Hanoi" prob-
lem. In that problem, a tapering stack of d discs is to be moved from one peg
94
to one of the other two pegs, which are originally empty. No disc is ever al-
lowed to rest on a larger disc, and the discs must be moved one at a time. If bd
is the number of moves required for d discs, one shows that bd = 2bd + 1
+1
(that is, to move the bottom disc, rst move the other discs onto a single
peg, then after the bottom disc is moved, the others are moved back onto
it). However, all we need for our present purpose is that bd > bd .
+1
The connection between the given problem and the \Towers of Hanoi"
problem can be seen by writing n as a sum of distinct powers of 2. Then, n
can be written as such a sum in exactly one way.
The distinct powers of 2 correspond to the discs in the \Towers of
Hanoi" problem, but larger powers of 2 correspond to smaller discs. The
condition for moving 2k pebbles corresponds to the fact that the top disc
must be moved before any of the discs underneath.
Note that d corresponds to the number of distinct powers of 2 that add
up to n (or the number of 1's in the binary expansion of n). So our problem
boils down to: what numbers n 100 have the most digits 1 in their binary
expansion?
The powers of 2 that can be used in the binary expansion of numbers
n 100 are 2 , 2 , 2 , 2 , 2 , 2 , 2 . However, they cannot all be used
6 5 4 3 2 1 0
since their sum is 127 > 100. It is possible to use all but one of them in just
two ways:
2 + 2 + 2 + 2 + 2 + 2 = 95 ,
6 4 3 2 1 0
2 + 2 + 2 + 2 + 2 + 2 = 63 .
5 4 3 2 1 0
Hence, 63 and 95 are the desired values of n.

3. (a) Begin with a string of 10 A's, B's, and C's, for example
A B C C B A B C B A
and underneath, form a new row, of length 1 shorter, as follows: between

two consecutive letters that are dierent, you write the third letter, and be-
tween two letters that are the same, you write that same letter again. Repeat
this process until you have only one letter in the new row. For example, for
the string above, you will now have:
A B C C B A B C B A
C A C A C C A A C
B B B B C B A B
B B B A A C C
B B C A B C
B A B C A
C C A B
C B C
A A
A
95
Prove that the letters at the corners of the resulting triangle are always either
all the same or all dierent.
Solution: Think of the letters A, B, C as representing the numbers 0,
1, 2, respectively, (mod 3). Then, if in some row we have x y
where x and y are integers (mod 3), we get ; x ; y in the next row,
where ;x ; y is computed (mod 3). If the original row corresponds to the
integers x x x (mod 3), then the next rows are:
1 2 10
;x ; x
1 ;x ; x
2 ;x ; x 2 3 3 4
:::
x + 2x + x 1 x + 2x + x
2 3 2 3 4
:::
;x ; 3x ; 3x ; x 1
::: 2 3 4
and we see the pattern (which can be proved by induction): apart from the
signs, the coecients are those of Pascal's triangle, the binomial coe-
cients. In particular, the tenth (bottom) row will consist of the single entry

;x ; 91 x ; 92 x ; ; 98 x ; x
1 2 3 9 10 (mod 3) .
;9 ;9
But since 9 is a power of the prime 3, all binomial coecients , ,
: : : , ; are divisible by 3, so that the entry in the bottom row is equal to
1 2
9
;x ; x (mod 3). Thus, the three corner numbers (x ; x ; ;x ; x )

1
8
10 1 10 1 10
(mod 3) that add to 0 (mod 3) and it follows that they are either all the
same or all dierent.
(b) For which positive integers n (besides 10) is the result from part (a)
true for all strings of n A's, B's, and C's?
Solution: For n = 1, and for n = 3k + 1, k 0. For such an n, the
nth row will have the form:

3 k k
3
3 k
;x ; 1 x ; 2 x ; ; 3k ; 1 xn; ; xn ;x ; xn (mod 3) ,
1 2 3 1 1
because the binomial coecients are divisible by 3.

4. When Mark climbs a staircase, he ascends either 1, 2, or 3 stairsteps
with each stride, but in no particular pattern from one foot to the next. In
how many ways can Mark climb a staircase of 10 steps? (Note that he must
nish on the top step. Two ways are considered the same if the number of
steps for each stride are the same; that is, it does not matter whether he puts
his best or his worst foot forward rst.) Suppose that a spill has occurred on
the 6th step and Mark wants to avoid it. In how many ways can he climb the
staircase without stepping on the 6th step?
Solution: Let an be the number of ways for Mark to climb a staircase of
n steps. Then a = 1, a = 2 and a = 4. For n > 3, consider Mark's last
1 2 3
96
stride. If his last stride was 1{step, then before that he climbed an n ; 1{step
staircase, and there are an; ways in which that can be done. Using a similar
argument for 2{ and 3{step last steps, we get an = an; + an; + an; for
1
n > 3. Using this we can calculate a = 274. Thus there are 274 ways that
1 2 3
10
Mark can climb the stairs before the spill has occurred.
Once the spill has occurred, we can work through the same way. If bn
is the number of ways to get to step n and not step on step 6, then we have:
bn = an for n 5, b = 0. Thus, using the recurrence relation for an we
get b = 0 + 13 + 7 = 20, b = 20 + 0 + 13 = 33, b = 0 + 20 + 33 = 53
6
and b = 20 + 33 + 53 = 106. Thus, there are 106 ways to go up the stairs

7 8 9
10
and avoid the spill.
5. Number the vertices of a cube from 1 to 8. Let A be the 8 8 matrix
whose (i; j ) entry is 1 if the cube has an edge between vertices i and j , and
is 0 otherwise. Find the eigenvalues of A, and describe the corresponding
eigenspaces.
Solution:Let the vertices be numbered as shown.
8 7
6
5
4 3
1 2
Then we have
2 3
0 1 0 1 1 0 0 0
6
6 1 0 1 0 0 1 0 0 7
7
6
6 0 1 0 1 0 0 1 0 7
7
A= 6
6 1 0 1 0 0 0 0 1 7
7
6
6 1 0 0 0 0 1 0 1 7
7
6
6 0 1 0 0 1 0 1 0 7
7
4 0 0 1 0 0 1 0 1 5
0 0 0 1 1 0 1 0
We can nd a few eigenvectors by inspection; for example (1 1 1 1 1 1 1 1)T
is an eigenvector for = 3 and (1 1 1 1 ; 1 ; 1 ; 1 ; 1)T is an
eigenvector for = ;1. From these a pattern starts to emerge. Note that
the entries 1 occur in positions 1, 2, 3, 4 and the entries ;1 occur in positions
5, 6, 7, 8. Now, f1, 2, 3, 4g [ f5, 6, 7, 8g = S [ S is a decomposition of
the vertex set S = f1, 2, 3, 4, 5, 6, 7, 8g into two disjoint parts S , S with
1 2
the property that each vertex in Si is connected to two vertices in Si and one
1 2
97
vertex in Sj for i, j 2 f1; 2g, i 6= j . We now have the observation, whose

proof follows from the denition of A given in the problem:
Let S = S [ S be any decomposition of S = f1, 2, 3, 4, 5, 6,
7, 8g into disjoint subsets S , S such that there exist integers k and l with
1 2
k + l = 3 such that each vertex in Si is connected to k vertices in Si and l

1 2
vertices in Sj for i, j 2 f1, 2g, i 6= j . Then the vector with entries 1 in the
positions corresponding to S and ;1 in the positions corresponding to S
is an eigenvector of A for the eigenvalue k ; l.
1 2
Note that while we could interchange the subsets S and S , that would
1 2
just change the eigenvector to its opposite. Except for this, there are eight
dierent decompositions of S of the desired type, as shown in the table
below.
S
1 S 2 k l Eigenvector
S 3 0 3 (1; 1; 1; 1; 1; 1; 1; 1)T
f1, 2, 3, 4g f5, 6, 7, 8g 2 1 1 (1; 1; 1; 1; ;1; ;1; ;1; ;1)TT
f1, 2, 5, 6g f3, 4, 7, 8g 2 1 1 (1; 1; ;1; ;1; 1; 1; ;1; ;1)T
f1, 4, 5, 8g f2, 3, 6, 7g 2 1 1 (1; ;1; ;1; 1; 1; ;1; ;1; 1)T
f1, 2, 7, 8g f3, 4, 5, 6g 1 2 ;1 (1; 1; ;1; ;1; ;1; ;1; 1; 1)T
f1, 4, 6, 7g f2, 3, 5, 8g 1 2 ;1 (1; ;1; ;1; 1; ;1; 1; 1; ;1)T
f1, 3, 5, 7g f2, 4, 6, 8g 1 2 ;1 (1; ;1; 1; ;1; 1; ;1; 1; ;1)T
f1, 3, 6, 8g f2, 4, 5, 7g 0 3 ;3 (1; ;1; 1; ;1; ;1; 1; ;1; 1)
(Geometrically, the eigenvectors for = 1 in this table can be thought
of as corresponding to pairs of opposite faces of the cube; the eigenvectors
for = ;1 can be thought of as corresponding to pairs of diagonal planes
through the cube.) It is easily checked that the eigenvectors for = 1 (simi-
larly for = ;1) listed above are linearly independent. Therefore, for matrix
A we have eigenvalues = 3, = 1 (with multiplicity 3), = ;1 (with
multiplicity 3) and = ;3, and the eigenspaces are spanned by the vectors
in the table.
6. Let f (x) be a twice-dierentiable function on the open interval (0; 1)
such that
lim f (x) = ;1 , and xlim
x!0+ !;
f (x) = +1 .
1
Show that f 00 (x) takes on both negative and positive values.

Solution: Take an arbitrary number x in the interval (0; 1) and let
f 0 (x ) = m. Suppose that f 00 (x) 0 for all x in (0; 1). Then, in particular,
0
for all x with 0 < x < x we must have f (x) f (x ) + m(x ; x ).

0
0 0 0
But f (x) f (x ) + m(x ; x ) shows that, as x ! 0 , we have +
f (x) f (x ) ; mx , contradicting the given limx! + f (x) = ;1.

0 0
0 0 0
Similarly, if f 00 (x) 0 for all x in (0; 1) then for all x with

x < x < 1 we must have f (x) f (x ) + m(x ; x ), and this contra-
dicts limx! ; f (x) = +1.
0 0 0
1
98
Thus, we must have f 00 (x) > 0 for some x in (0; 1) as well as f 00 (x) < 0
for some x in (0; 1), and we are done.
7. Three stationary sentries are guarding an important public square
which is, in fact, square, with each side measuring 8 rods (recall that one rod
equals 5:5 yards). (If any of the sentries see trouble brewing at any location
on the square, the sentry closest to the trouble spot will immediately cease to
be stationary and dispatch to that location. And like all good sentries, these
three are continually looking in all directions for trouble to occur.) Find the
maximum value of D, so that no matter how the sentries are placed, there
is always some spot in the square that is at least D rods from the closest
sentry.
Solution: Divide the square into rectangles as shown, in such a way
that QY = Y R = 4 and PZ = XY . We claim once this has been done
D = 21 XY is the maximum value of D.
0
S Z R
T Y
P X Q
Proof: One way to place the sentries is at the centres of the rectangles
PXZS , XQY T , TY RZ ; for this placement, every point on the square is at
most D from the sentry at the centre of the rectangle it belongs to. Thus,
a value D > D is not possible. To show that D = D actually works,
0
0 0
we have to show that there is no placement of the sentries for which every
point of the square has distance less than D to the closest sentry. Suppose
we did have such a placement. Since there are four corners P , Q, R, S of
0
the square and only three sentries, at least two corners would be guarded
by the same sentry. Without loss of generality, let P and S be two corners
guarded by the same sentry. Now note that X and Y cannot be guarded by
a single sentry, because by our assumption that sentry would have distance
less than D to both X and Y , so that XY < 2D = XY , a contradiction.
0 0
Similarly, no other two opposite corners of the three smaller rectangles that
we constructed can be guarded by a single sentry. Thus, S , X , Y are all
guarded by dierent sentries, say 1, 2, 3, in that order, and since S and P
are guarded by the same sentry P , Z , Y are also guarded by 1, 2, 3 in that
order. Thus, Q and R must be guarded by 3. But, we will show that the
99
mid-point O of XQ is too far from Z and R to be guarded by 2 and 3, re-

spectively, which will provide a contradiction.
Let PX = a , XQ = 8 ; a. Then
PZ = XY (=) PX + XZ = XQ + QY ,
2 2 2 2
which gives a = 1, so that PX = 1, QX = 7, XO = 27 and

ZO > ZP = 2Dp , showing that O is indeed too far from Z . Thus, the
0
answer is D = 12 65.
0
8. The Union Atlantic Railway is planning a massive project: a railroad

track joining Cambridge, Massachusetts and Northeld, Minnesota. How-
ever, the funding for the project comes from the will of Orson Randolph Kane,
the eccentric founder of the U.A.R., who has specied some strange condi-
tions on the railway; thus the skeptical builders are unsure whether or not it
is possible to build a railway subject to his unusual requirements.
Kane's will insists that there must be exactly 100 stops (each named after
one of his great-grandchildren) between the termini, and he even dictates
precisely what the distance along the track between each of these stops must
be. (Unfortunately, the tables in the will do not list the order in which the
stops are to appear along the railway.) Luckily, it is clear that Kane has put
some thought into these distances; for any three distinct stops, the largest
of the three distances among them is equal to the sum of the smaller two,
which is an obvious necessary condition for the railway to be possible. (Also,
all the given distances are shorter than the distance along a practical route
from Cambridge to Northeld!)
U.A.R.'s engineers have pored over the numbers and noticed that for any
four of Kane's stops, it would be possible to build a railway with these four
stops and the distances between them as Kane species. Prove that, in fact,
it is possible to complete the entire project to Kane's specications.
Solution: Let d(i; j ) be the required distance from stop i to stop j along
the track. Note that if four stops are arranged successfully along the track,
there is a unique greatest distance between two of those four stops: the
one that ends up closest to Cambridge, and the one that ends up closest to
Northeld.
Now for the proof.
The case n = 4 is trivial. Suppose the statement is true for N stops
(N 4) and we have a table of distances d(i; j ) for N + 1 stops satisfying
the given conditions. Find the largest of these distances. This largest dis-
tance can only occur once, because if d(i; j ) = d(k; l) with fi, j g 6= fk, lg, it
would be impossible to arrange stops i, j , k, l (or if two of them were equal,
say j = l, then stops i, j , k, m with m 6= i, j , k) successfully along the track.
Without loss of generality, assume that the greatest distance in the table is
d(1; N + 1). Let a, b be any two distinct stops with 2 a, b N . Then,
100
since stops 1, a, b, N +1 could be arranged successfully, and d(1; N +1) is the

largest distance, we must have either d(1; a) = d(1; b) + d(a; b) or
d(1; b) = d(1; a) + d(a; b) (but we cannot have d(a; b) = d(1; a) + d(1; b)).
Now, ignore stop N + 1 for the moment and arrange stops 1, 2, : : : , N suc-
cessfully along the track, which is possible by the induction hypothesis. Since
we cannot have d(a; b) = d(1; a) + d(1; b) for any a, b 2 f2, 3, : : : , N g,
we have that 1 cannot be between any two other stops in this successful
arrangement; that is, all stops 2, 3, : : : , N must be the same side of stop 1.
Since they all have the proper distances to stop 1 and d(1; N + 1) is larger
than all those distances, we can place stop N + 1 on the track in such a way
that all stops 2, 3, : : : , N are between stop 1 and stop N + 1, that stops 1
and N + 1 are the proper distance apart, and that all stops 1, 2, : : : , N are
still in the same places.
All we now need to do is to check that stop N +1 is the correct distance
d(a; N + 1) to each of the stops a, 2 a N . Then, since d(1; N + 1) is
less than the distance from Cambridge to Northeld, we can arrange for the
stops to follow each other in this order between Cambridge and Northeld,
and we will be done.
Consider any stop a, 2 a N . Since it has the correct distance
d(a; 1) to stop 1, in the above placement its distance to stop N + 1 is given
by d(1; N + 1) ; d(a; 1). But since d(1; N + 1) is the largest of the required
distances between stops 1, a, N + 1, we have
d(1; N + 1) = d(a; 1) + d(a; N + 1) ,
so that
d(1; N + 1) ; d(a; 1) = d(a; N + 1) ,
and we are done.
Professor Toshio Seimiya

101
Problem of the Month

Jimmy Chui, student, University of Toronto
Problem. (a) Given positive numbers a , a , a , , an and the quadratic
n 1 2 3
function f (x) = (x ; ai ) , show that f (x) attains its minimum value at

X 2
i =1
n n n ! 2
n i ai , and prove that i ai n i ai .

1 X X 21 X
=1 =1 =1
(b) The sum of sixteen positive numbers is 100 and the sum of their
squares is 1000. Prove that none of the sixteen numbers is greater than 25.
(1996 Canadian Open, Problem B3)
Solution. (a) The quadratic function f (x) is a parabola, and the graph
y = f (x) opens upward. (The x {coecient is positive.) Hence the vertex
2
of the graph is a minimum point; that is, there is a unique value of x that
minimizes y = f (x), and the point (x; y) is the vertex.
It is known that the x{coordinate of the vertex of the function
f (x) = ax + bx +n c is ;b=n2a. Our given function, after expanding, is
2
f (x) = nx ; 2x X ai + X ai , and the x-coordinate of the vertex is

2 2
n i =1 i =1
n i ai . Hence, for this value of x, the value of f (x) is minimized.

1 X
=1
We also know that f (x) is greater than or equal to 0, since it is the sum
of non-negative squares. Hence, the discriminant must be less than or equal
to 0. (This condition corresponds to f (x) having one or no roots.)
n ! n 2
If we let the discriminant be D, then D=4 =

X
ai ; n X ai 0. 2
i i =1 =1
n 1 n ! 2
Rearranging this inequality gives us the desired result, ai

n i ai .
X X 2
i =1 =1
(b) Let the largest value of the ai 's be b. Consider the 15 ai 's, ex-
cluding b. Then, apply the result in (a) to these 15 numbers. We have
X 1 X
ai ; 15 ai 0, where both summations are taken over the 15
2
2
ai 's excluding b.
X
ai ; 15 1 X a = 1 15 ;1000 ; b ; (100 ; b)
2 n o
2
i 15
2 2
1
= 15 ;16b + 200b + 5000 = ; 15
; 2
8 (b ; 25) (2b + 25) 0 .
Since b is a positive value, the inequality holds true if and only if b 25. In
other words, the largest of the ai 's must not exceed 25, QED.
102
SKOLIAD No. 60
Shawn Godin
Solutions may be sent to Shawn Godin, Cairine Wilson S.S., 975 Orleans
Blvd., Orleans, ON, CANADA, K1C 2Z5, or emailed to
[email protected]
Please include on any correspondence your name, school, grade, city, province
or state and country. We are especially looking for solutions from high school
students. Please send your solutions to the problems in this edition by
1 July 2002. A copy of MATHEMATICAL MAYHEM Vol. 2 will be presented
to the pre-university reader(s) who send in the best set of solutions before
the deadline. The decision of the editor is nal.
This issue's item comes to us from Manitoba. My thanks go to Diane

Dowling at St.Paul's College in Winnipeg for forwarding the material to me.
MANITOBA MATHEMATICAL CONTEST, 2001
For Students in Senior 4
9:00 a.m. { 11:00 a.m. Wednesday, February 21, 2001
Sponsored by
The Actuaries' Club of Winnipeg, The Manitoba Association of Mathematics
Teachers, The Canadian Mathematical Society, and The University of Manitoba
Answer as much as possible. You are not expected to complete the paper. See both sides
of this sheet. Hand calculators are not permitted. Numerical answers only, without explanation,
will not be given full credit.
1. (a) Solve the equation x ; x2 = 9.
6 1
(b) Find real numbers a and b such that (2a )(4b ) = 8 and a+7b = 4.
2. (a) Find all real numbers x such that 25jxj = x + 144. 2
(b) If x is a real number and p 5 p = 7 ; 3px what is the

p
7+3 x
value of x?
3. (a) If two of the roots of x + px + qx + r = 0 are equal in absolute
3 2
value but opposite in sign, prove that pq = r. (p, q and r are real numbers.)
(b) If a, b and c are real numbers such that a + b + c = 14,
c = a + b and ab = 14, nd the numerical value of c.
2 2 2
4. (a) If 0 < < 180 and 2 sin + 3 sin 2 what is the largest
2
possible value of ?
103
(b) Let O be the origin, P the point whose x coordinates are (2; 3)
and F a point on the line whose equation is y = 2 . If PF is perpendicular
to OF nd the coordinates of F .
5. How many consecutive zeros are there at the end of the product of
all the integers from 16 to 100 inclusive?
6. In triangle ABC , \ACB = 135 , CA = 6 and BC = p2. If M is
the mid-point of the side AB , nd the length of CM .
7. A circle of radius 2 has its centre in the rst quadrant and has both
coordinate axes as tangents. Another smaller circle also has both coordinate
axes as tangents and has exactly one point in common with the larger circle.
Find the radius of the smaller circle.
8. A parallelogram has an area of 36 and diagonals whose lengths are
10 and 12. Find the lengths of its sides.
9. a, b, c, d are distinct integers such that (x;a)(x;b)(x;c)(x;d)=4
has an integral root r. Prove that a + b + c + d = 4r.
10. If x, y and z are positive real numbers, prove that
(x + y ; z)(x ; y) + z(x ; z)(y ; z) 0 .
2
Next we present the ocial solutions to the Mandelbrot competitions

from the October issue [2001 : 385].
The Mandelbrot Competition
Division B Round Two Individual Test
December 1997
1. If a group of positive integers has a sum of 8, what is the greatest
product the group can have? (1 point)
Solution.
Clearly we do not want any 1's in our group, since they contribute to
the sum but not the product. Any 5 in the group can be replaced by a 2 and
a 3, which have the same sum but a greater product. Similarly, any 6 can
be replaced by two 3's, and so on, since every number contributes more to
the product if broken down into 2's and 3's. (A 4 can be replaced with two
2's, with no eect on the sum or product.) Hence, our optimal group should
consist of 2's and 3's only. The only such sets of numbers summing to 8 are
2+2+2+2 or 2+3+3. The products in these cases are 16 and 18. Hence,
18 is optimal.
2. There is one two-digit number such that if we add 1 to the number
and reverse the digits of the result, we obtain a divisor of the number. What
is the number? (1 point)
104
Solution.
Let the two-digit number n be t u . If u is not 9, then the num-
ber obtained by adding one and reversing the digits is u + 1 t . Since
this potential divisor; cannot equal n, it must be half of n or less, so that
2(u + 1) t and 2 (u + 1) 10 + t t 10 + u. This restricts us to
the numbers n = 30, 40, 50, 51, 60, 61, 70, 71, 80, 81, 82, 90, 91, or 92. A
quick check shows that none of these numbers works. Hence, u must be 9.
If n = t 9 , then n + 1 = t + 1 0 (unless t = 9), so that the divisor
would be t + 1. But n equals 10(t + 1) ; 1, which cannot be a multiple of
t + 1. Thus t must also be 9, yielding n = 99 , making the divisor equal to
001, or 1.
3. Ten slips of paper, numbered 1 through 10, are placed in a hat.
Three numbers are drawn out, one after another. What is the probability
that the three numbers are drawn in increasing order? (2 points)
Solution.
Let the numbers chosen be A, B , and C . There are six orders in which
the slips can be chosen: ABC , ACB , BAC , BCA, CAB , CBA. Of these
six, only one is in the increasing order we desire. Hence, the probability
is 16 .
4. The three marked angles are right angles. If \a = 20 , then what
is \b? (2 points)
a
Solution.
Note that \b is complementary to an angle which is complementary
to \a; hence, \b = \a, so that \b = 20 .
5. Vicky asks Charlene to identify all non-congruent triangles 4ABC
given:
(a) the value of \A
(b) AB = 10, and
(c) length BC equals either 5 or 15.
105
Charlene responds that there are only two triangles meeting the given con-
ditions. What is the value of \A? (2 points)
Solution.
Consider AB to be a xed segment of length 10. Since BC is either 5
or 15, C must lie on one of the two circles with center B and radii 5 and 15,
as in the diagram. R
S
B A
There are many possible values for \A; the possible positions for point C
are the points where \A intersects the two circles. For some angles, as with
\A = \BAR in the diagram, there is only one possible point C ; for others,
as with \A = \BAT , there are three. The only value which gives exactly
two points C is that which makes the angle tangent to the inner circle, as
with \A = \BAS . Call the point of tangency X . Since AB = 10, BX = 5,
and \BXA is a right angle, we conclude that triangle ABX is a 30 -60 -90
triangle. Hence, \A = 30 .
6. Five pirates nd a cache of ve gold coins. They decide that the
shortest pirate will become bursar and distribute the coins | if half or more
of the pirates (including the bursar) agree to the distribution, it will be ac-
cepted; otherwise, the bursar will walk the plank and the next shortest pirate
will become bursar. This process will continue until a distribution of coins is
agreed upon. If each pirate always acts so as to stay aboard if possible and
maximize his wealth, and would rather see another pirate walk the plank
than not (all else being equal), then how many coins will the shortest pirate
keep for himself? (3 points)
Solution.
Call the pirates p , p , : : : , p , with p the shortest and p the tallest.
Consider what would happen if only p and p remained. Whatever division
1 2 5 1 5
strategy p suggested would hold, since p 's vote alone would constitute half
4 5
the total vote. Thus, p would simply allot himself all the gold. Next consider
4 4
4
106
the situation where three pirates remained. Whatever distribution p chose,

p would have to agree, as long as he got one or more coins. His only alter-
3
native would be to go to the two-pirate situation, in which p gets nothing

5
at all. Hence, p would simply take 4 coins for himself and allot 1 coin to
5
p , getting a majority vote from himself and p . Similarly, with four pirates
3
remaining, p , the bursar, would take 4 coins for himself and allot 1 coin
5 5
to p . Again, since p would have otherwise gotten nothing, he would have

2
to support the plan. With ve pirates, the bursar, p , would allot himself 3
4 4
coins and give one coin each to p and p . Since each of p and p gets more
1
3 5 3 5
than they would get by vetoing the plan, they must support it. The shortest
pirate gets 3 coins.
7. The twelve positive integers a a a have the property
1 2 12
that no three of them can be the side lengths of a non-degenerate triangle.
Find the smallest possible value of aa121 . (3 points)
Solution.
If a, b and c can be sides of a non-degenerate triangle with a b c,
we always must have c < a + b. Hence, if no three of our integers can form
a non-degenerate triangle, we must have ai aj + ak for any three with
i > j , k. Since the numbers are increasing, it suces to show that
a +a a , a +a a , a +a a
1 2 3 2 3 4 3 4 5
and so on. Substituting the rst inequality into the second, we have
a + 2a a .
1 2 4
Substituting this and the second inequality into the third, we get
2a + 3a a .
1 2 5
Substituting into the fth inequality:

3a + 5a a .
1 2 6
We continue in this manner. At each step, we simply increase the coecients

of a and a ; they must increase as Fibonacci numbers, since at each step
1 2
we get the new coecient by adding the previous two. Hence, the nal
inequality will read
55a + 89a a . 1 2 12
To obtain the smallest possible ratio we choose a = a . This simply yields

144a a , or a =a 144. Since this limit can be attained, with
1 2
a = a = 1, a = 2, a = 3, a = 5, : : : , a = 144, we nd that
1 12 12 1
1 2 3 4 5 12
144 is the least value.
107
The Mandelbrot Competition

Division B Round Two Team Test
December 1997
FACTS: A polynomial p(x) of degree n or less is determined by its value
at n + 1 x{coordinates. For n = 1 this is a familiar statement; a line (degree
one polynomial) is determined by two points. Moreover, the value of p(x) at
any other x{value can be computed in a particularly nice way using Lagrange
interpolation, as outlined in the essay An Interpretation of Interpolation.
We will also need a result from linear algebra which states that a system
of n \dierent" linear equations in n variables has exactly one solution. For
example, there is only one choice for x; y, and z which satises the equations
x + y + z = 1, x + 2y + 3z = 4, and x + 4y + 9z = 16.
SETUP: Let p(x) be a degree three polynomial for which we know the
values of p(1), p(2), p(4), and p(8). By the facts section there is exactly one
such polynomial. According to Lagrange interpolation the number p(16) can
be deduced; it equals
p(16) = A p(1) + A p(2) + A p(4) + A p(8) ,
0 1 2 3
for some constants A through A . The goal of this team test will be to
compute the Ai and use them to nd information about p(16) without ever
0 3
nding an explicit formula for p(x).

Problems:
Part i: (4 points) We claim that the Ai can be found by subtracting
x ; (x ; 1)(x ; 2)(x ; 4)(x ; 8) = A x + A x + A x + A .
4
3
3
2
2
1 0
(1)
Implement this claim to compute A through A . 0 3
Solution.
Apparently we have been handed a magic formula which generates the
constants Ai needed for computing p(16). It is referred to frequently in this
solution, so we reproduce it here:
x ; (x ; 1)(x ; 2)(x ; 4)(x ; 8) = A x + A x + A x + A . (2)
4
3
3
2
2
1 0
Rather than marvel at this stroke of good fortune so early on, we set about
computing the values A through A . Carefully multiplying out the left-hand
0 3
side yields
x ; (x ; 1)(x ; 2)(x ; 4)(x ; 8)
4
= x ; (x ; 3x + 2)(x ; 12x + 32)

4 2 2
= x ; (x ; 15x + 70x ; 120x + 64)

4 4 3 2
= 15x ; 70x + 120x ; 64

3 2
= A x +A x +A x+A .
3
3
2
2
1 0
108
Therefore we should use A = 15, A = ;70, A = 120, and A = ;64.

3 2 1 0
Part ii: (4 points) To show that these Ai are in fact the correct numbers
we must show that they correctly predict p(16) for four \dierent" polyno-
mials. We begin with the case p(x) = x. Show that the value of p(16) agrees
with the prediction A p(1) + A p(2) + A p(4) + A p(8). (HINT: try x = 2
0 1 2 3
in (1).)
Solution.
According to Lagrange Interpolation, if p(x) is a polynomial of degree
three or less, then we should be able to predict p(16) based on the val-
ues of p(x) at x = 1, 2, 4 and 8. We will now show that the constants
Ai just computed do the job by showing that p(16) always equals the sum
A p(8) + A p(4) + A p(2) + A p(1). According to the facts section we
need only verify that the Ai work in four dierent cases to know that they
3 2 1 0
will always work. In checking these four cases we employ a somewhat clever
method that never actually uses the numbers calculated in part i, just the
equation (2) that produced them.
First suppose that p(x) = x. Then clearly p(1) = 1, p(2) = 2,
p(4) = 4, and p(8) = 8. Let us check whether or not A p(8) + A p(4) +
A p(2) + A p(1), which is the same as 8A + 4A + 2A + A , correctly
3 2
predicts p(16). Substituting x = 2 into equation (2) gives us

1 0 3 2 1 0
2 ; (2 ; 1)(2 ; 2)(2 ; 4)(2 ; 8) = 8A + 4A + 2A + A .

4
3 2 1 0
The left-hand side equals 16 since the (2 ; 2) factor causes the second term
to vanish. However, the right hand side is our prediction for p(16). Sure
enough, we get p(16) = 16, just as we should for the function p(x) = x.
Part iii: (5 points) Continuing the previous part, show that the Ai
correctly predict p(16) for the three other polynomials p(x) = 1,
p(x) = x and p(x) = x .
2 3
Solution.
In all cases the prediction for p(16) is A p(8) + A p(4) + A p(2) +
A p(1). Continuing our work from part ii we try p(x) = 1, so our prediction
3 2 1
becomes A + A + A + A . Using x = 1 in (2) yields

0
3 2 1 0
1 ; (1 ; 1)(1 ; 2)(1 ; 4)(1 ; 8) = A + A + A + A .

4
3 2 1 0
The left-hand side reduces to 1, so that the prediction is p(16) = 1, which

again is correct. When p(x) = x the prediction for p(16) becomes 2
64A + 16A + 4A + A . This can be found quickly by substituting x = 4

3 2 1 0
into equation (2):
4 ; (4 ; 1)(4 ; 2)(4 ; 4)(4 ; 8) = 64A + 16A + 4A + A .
4
3 2 1 0
The now familiar cancellation occurs on the left hand side, leaving us with a
prediction of 4 for p(16). Since p(x) = x we expect to have p(16) = 16 ,
4 2 2
109
and indeed 16 = (4 ) = 4 . The case of p(x) = x works in exactly

2 2 2 4 3
the same manner, substituting x = 8 into equation (2), so we encourage

the reader to try it as practice. (Naturally teams were expected to show the
details for this case as well in their solutions!)
Notice that we were able to do all of our checking without ever using
the numerical values of A through A . The other more obvious method is
to plug in the values for the Ai and do the arithmetic. However, the slick
0 3
technique can be generalized, while the more routine method cannot.

Part iv: (4 points) Suppose that p(x) is a third degree polynomial with
p(1) = 0, p(2) = 1, and p(4) = 3. What value should p(8) have to guarantee
that p(x) has a root at x = 16?
Solution.
We have now veried that A = 15, A = ;70, A = 120, and
A = ;64 are the correct values needed to interpolate p(16). We are also
3 2 1
told in this problem that p(1) = 0, p(2) = 1, and p(4) = 3. Furthermore,

0
we want p(16) = 0 so that p(x) has a root at x = 16. Substituting all of

these values into our interpolation formula produces
p(16) = A p(8) + A p(4) + A p(2) + A p(1)
==) 0 = 15p(8) ; 70 3 + 120 1 ; 64 0
3 2 1 0
==) 90 = 15p(8)
==) 6 = p(8) .
Thus we need p(8) = 6 to ensure that p(16) = 0 so that there is a root at
x = 16.
Part v: (4 points) Let p(x) be a degree three polynomial with
p(1) = 1, p(2) = 3, p(4) = 9, and p(8) = 27. Calculate p(16). How
close does it come to the natural guess of 81?
Solution.
There are two ways to do this problem|a long way and a short way.
The long way involves substituting all of the values for A through A and
p(1) through p(8) into the interpolation formula shown above and crank-
0 3
ing out the answer. The short way, hinted at by our above work, involves
plugging x = 3 into equation (2). We opt for the short way, obtaining
3 ; (3 ; 1)(3 ; 2)(3 ; 4)(3 ; 8) = 27A + 9A + 3A + A .
4
3 2 1 0
The left-hand side reduces to 81 ; (2)(1)(;1)(;5) = 71, while the expres-

sion on the right is exactly the interpolation formula for p(16). Therefore we
have p(16) = 71, ten less than the intuitive guess of 81.
110
PROBLEMS
Faire parvenir les propositions de problemes et les solutions a Bruce Shawyer,
Department of Mathematics and Statistics, Memorial University of Newfoundland,
St. John's (Terre-Neuve), Canada, A1C 5S7. Les propositions de problemes doivent
e^ tre accompagnees d'une solution ainsi que de references et d'autres indications qui
pourraient e^ tre utiles a la redaction. Si vous envoyez une proposition sans solution,
vous devez justier une solution probable en fournissant susamment d'information.
Un numero suivi d'une asterisque (?) indique que le probleme a e te propose sans
solution.
Nous sollicitons en particulier des problemes originaux. Cependant, d'autres
problemes interessants pourraient e^ tre acceptables s'ils ne sont pas trop connus et si
leur provenance est precisee. Normalement, si l'auteur d'un probleme est connu, il
faut demander sa permission avant de proposer un de ses problemes.
Pour faciliter l'etude de vos propositions, veuillez taper ou e crire a la main
(lisiblement) chaque probleme sur une feuille distincte de format 8 12 "11" ou A4, la
signer et la faire parvenir au redacteur en chef. Les propositions devront lui parvenir
au plus tard le 1er octobre 2002. Vous pouvez aussi les faire parvenir par courriel a
[email protected]. (Nous apprecierions de recevoir les problemes et solu-
tions envoyes par courriel au format LATEX). Les chiers graphiques doivent e^ tre de
format epic ou eps (encapsulated postscript). Les solutions recues apres la date
ci-dessus seront prises en compte s'il reste du temps avant la publication. Veuillez
prendre note que nous n'acceptons pas les propositions par telecopieur.
Chaque probleme sera publie dans les deux langues ocielles du Canada
(anglais et francais). Dans les numeros 1, 3, 5 et 7, l'anglais precedera le francais,
et dans les numeros 2, 4, 6 et 8, le francais precedera l'anglais.
Dans la section des solutions, le probleme sera publie dans la langue de la
principale solution presentee.
2713. Propose par Toshio Seimiya, Kawasaki, Japan.

Soit O un point interieur du triangle ABC et supposons que AO, BO
et CO coupent BC , CA et AB en D, E et F , respectivement. Soit H le pied
de la perpendiculaire a EF passant par D.
Montrer que les pieds des perpendiculaires a AF , FO, OE et EA
passant par H sont sur un même cercle.
.................................................................
Suppose that O is an interior point of 4ABC , and that AO, BO and
CO meet BC , CA and AB at D, E and F , respectively. Let H be the foot
of the perpendicular from D to EF .
Prove that the feet of the perpendiculars from H to AF , FO, OE and
EA are concyclic.
111

Soient respectivement D et E des points sur les côtes AC et AB d'un
triangle ABC , de sorte que \DBC = 2\ABD et \ECB = 2\ACE .
Si BD et CE se coupent en O et si OD = OE , que peut-on dire de ce
triangle ?
.................................................................
Suppose that D and E are points on sides AC and AB , respectively, of
4ABC , such that \DBC = 2\ABD and \ECB = 2\ACE . Suppose
that BD and CE meet at O, and that OD = OE . Characterize 4ABC .
Soit O le centre du cercle inscrit a un quadrilatere convexe ABCD.
Si E et F sont respectivement les centres des cercles inscrits aux triangles
ABC et ADC , montrer que A, O et le centre du cercle circonscrit au triangle
AEF sont sur une droite.
.................................................................
Suppose that the convex quadrilateral ABCD has an incircle with cen-
tre O. Let E and F be the incentres of 4ABC and 4ADC , respectively.
Prove that A, O and the circumcentre of 4AEF are collinear.
On suppose que
1. ABCD est un quadrilatere convexe donne,
2. P est un point sur AD situe au-dela de A de telle sorte que
\APB = \BAC ,
3. Q est un point sur AD situe au-dela de D de telle sorte que
\DQC = \BDC , et
4. AP = DQ.
Que peut-on dire du quadrilatere ABCD ?
.................................................................
Suppose that
1. convex quadrilateral ABCD is given,
2. P is a point of AD produced beyond A such that \APB = \BAC ,
3. Q is a point on AD produced beyond D such that \DQC = \BDC ,
and
4. AP = DQ.
Characterize quadrilateral ABCD.
112
2717. Propose par Mihaly Bencze, Brasov, Romania.

Pour tout triangle ABC , montrer que
8 sin A2 sin B2 sin C2 cos A ;2 B cos B ;2 C cos C ;2 A .

.................................................................
For any triangle ABC , prove that
8 sin A2 sin B2 sin C2 cos A ;2 B cos B ;2 C cos C ;2 A .

2718. Propose par Mihaly Bencze, Brasov, Romania.

Soit Ak 2 Mm (R) tels que Ai Aj = Om , i, j 2 f1, 2, : : : , ng, avec
i < j et xk 2 R , (k = 1, 2, : : : , n). Montrer que
!
n ;
X
det Im + xk Ak + xk Ak2 2

0.
k=1
.................................................................
Let Ak 2 Mm (R) with Ai Aj = Om , i, j 2 f1, 2, : : : , ng, with i < j
and xk 2 R , (k = 1, 2, : : : , n). Prove that
!
n ;
X
det Im + xk Ak + xk Ak2 2

0.
k=1
2719. Propose par Antal E. Fekete, Memorial University, St. John's,
Newfoundland.
Soit k 2 Z et n 2 N. Montrer que, pour n = 1 et n = 2,
1
X ( k + j )n = (;1)n X1 (;k ; 2 + j )n
j =0 j! j =0
j!
et 1 1
(;1)j (k +j !j ) = (;1)n (;1)j (;k ;j2! + j )
X n X n
j =0 j =0
Ces e galites sont-elles vraies ou fausses pour d'autres valeurs entieres
positives de n ?
.................................................................
Let k 2 Z and n 2 N. Show that
1
X (k + j )n = (;1)n X1 (;k ; 2 + j )n
j =0 j! j j!
=0
113
and
1 1
(;1)j (k +j !j ) = (;1)n (;1)j (;k ;j2! + j )
X n X n
j =0 j =0
for n = 1 and n = 2.
Are these equalities true or false for other positive integral values of n?
2720. Propose par Antal E. Fekete, Memorial University, St. John's,
Newfoundland.
Soit k un entier et n un entier non negatif.
1
k + j )n est un multiple entier de e et trouver la somme.
P
(a) Montrer que j!
(
j =0
1
(b) Montrer que (;1)j (k +j !j ) est un multiple entier de 1e et trouver la
P n
j
somme.
=0
.................................................................
Let k be an integer and n be a non-negative integer.
1
k + j )n is an integral multiple of e, and nd the sum.
P
(a) Show that j!
(
j =0
1 n
(b) Show that (;1)j (k +j !j ) is an integral multiple of 1e , and nd the
P
j
sum.
=0
2721. Propose par Vedula N. Murty, Visakhapatnam, India.

On considere l'equation cubique x ; 19x + 30 = 0. Il est facile de
3
verier que les racines de cette e quation sont ;5, 2 et 3. Si l'on essaie de
resoudre l'equation ci-dessus par la methode trigonometrique, on trouve que
les racines sont :
;2 31 cos , ; 2 13 cos , et ; 2 31 cos ,

2 + 4 +
3 3 3
r 32
ou 13 = 19
3
et cos = 15 3
19
.
Sans utiliser de calculatrice, montrer que
;2 13 cos = ;5, ;2 31 cos = 2, et ;2 31 cos = 3 .

2 + 4 +
3 3 3
114
Consider the cubic equation x ; 19x + 30 = 0. It is easily veried

3
that the roots of this equation are ;5, 2 and 3. If one tries to solve the
above equation using trigonometry, the roots come out as
;2 31 cos , ; 2 13 cos , and ; 2 31 cos ,

2 + 4 +
3 3 3
r 3
2
where 31 =
3
and cos = 15 19
19 3
.
Show, without the use of a calculator, that
;2 13 cos = ;5, ;2 31 cos = 2, and ;2 31 cos = 3 .

2 + 4 +
3 3 3
2722. Propose par Vaclav Kone cny, Ferris State University, Big
Rapids, MI, USA.
On considere deux triangles pythagoriciens comme in-
dique dans la gure. Les longueurs de AC , CB , AD
et DE sont des nombres impairs et les longueurs des
côtes superposes, AE et AB , sont des nombres pairs. C
Existe-t-il une conguration semblable telle que la
longueur de CD soit un entier ?
................................................................. A E
B
Consider two Pythagorean triangles as indicated in
the gure. The lengths of AC , CB , AD and DE are D
odd integers and the lengths of the overlapping sides,
AE and AB, are even integers.
Does there exist such a conguration of Pythagorean
triangles such that the length of CD is an integer?
2723. Propose par Walther Janous, Ursulinengymnasium, Innsbruck,
Austria.
Soit n , n , : : : , nk (1 k N ) des entiers non negatifs tels que
n + n + + nk = N . Trouver la valeur minimale de la somme
1 2
k ;nj
1 2
lorsque (a) m = 2 ; (b)? m 3.

P
j =1 m
.................................................................
For 1 k N , let n , n , : : : , nk be non-negative integers such
that n + n + + nk = N . Determine the minimum value of the sum
1 2
k ;nj
1 2
(b)? m 3.
P
m when (a) m = 2;
j =1
115
SOLUTIONS
No problem is ever permanently closed. The editor is always pleased to
consider for publication new solutions or new insights on past problems.
2605. [2001 : 49] Proposed by K.R.S. Sastry, Bangalore, India.

In triangle ABC , with median AD and internal angle bisector BE , we
are given AB = 7, BC = 18 and EA = ED. Find AC .
I. Solution by Peter Y. Woo, Biola University, La Mirada, CA, USA.
Let M , N be feet of the perpendiculars from A, E to BC , respectively.
Let J be on BN such that BJ = BA = 7. Then 4ABE = 4JBE by SAS.
Therefore, EJ = EA = ED. Furthermore, JN = ND = 1 and NC = 10.
Also AE=EC = AB=BC = 7=18. Hence
MC = NC AC
EC = 10 25 = 125 .
18 9
Thus
BM = 18 ; 125 37
9 = 9:
By the Theorem of Pythagoras:
AC ; AB = (MC + AM ) ; (MB + AM ) = MC ; MB .
2 2 2 2 2 2 2 2
Therefore,
2 2
AC = AB + 125
2 2
9 ; 37
9
= 49 + (125 + 37)(125 ; 37) = 49 + 162 88 = 225 ,
81 81
from which we conclude that AC = 15.
A
E
B M JND C
II. Solution by Henry Liu, student, University of Memphis, TN, USA.
Let J lie on side BC with BJ = 7. Then (as in solution I above)
4ABE = 4JBE . Thus, \BAE = \BJE . Also, ED = EA = EJ ,
116
which implies that 4JDE is isosceles, whence \CDE = \BJE = \BAE .

Therefore, 4CDE and 4CAB are similar. Set y = EA = ED. Then
9 = AC ==) AC y = 63
y 7
and AC ; y = 18 ==) AC ; AC y = 162
2
9 AC
==) AC = 162 + 63 = 225
2
==) AC = 15 .
III. Solution by Paul Jereys, student, Berkhamsted Collegiate School,
UK.
Since BE bisects \ABC , we have 7=AE = 18=EC . Thus
AE = 7 and EC = 18 AC: (1)
EC 18 25
Let = \BAC and = \ABC . By the Law of Sines on 4ABE , we have
AE = BE (=) AE sin = BE sin(=2):
sin(=2) sin
Applying the Law of Sines to 4BDE yields:
DE = BE
sin(=2) sin(\BDE ) (=) DE sin(\BDE ) = BE sin(=2) .
Since AE = ED we have sin(\BDE ) = sin . If \BDE = , then 4BAE
is similar to 4BDE . Since BD = 9 =6 7 = BA, we get a contradiction,
whence we have \BDE = 180 ; . Thus BDEA is a cyclic quadrilateral.
Since \BDE = 180 ; , we also have \CDE = , which implies that
4CDE is similar to 4CAB. Therefore, 9=AC = EC=18. From (1), we
have
9 = 18 AC , so that 9 25 = AC . 2
AC 25 18
Thus, AC = 15.

Also solved by SEFKET University of Sarajevo, Sarajevo, Bosnia and
ARSLANAGIC,
Herzegovina; AUSTRIAN IMO TEAM 2001; MICHEL BATAILLE, Rouen, France; FRANCISCO
BELLOT ROSADO, I.B. Emilio Ferrari, and MAR I A ASCENSI ON L OPEZ
CHAMORRO, I.B.
Leopoldo Cano, Valladolid, Spain; CHRISTOPHER J. BRADLEY, Clifton College, Bristol, UK;
SCOTT BROWN, Auburn University, Montgomery, AL; NIKOLAOS DERGIADES, Thessaloniki,
Greece; CHARLES DIMINNIE and KARL HAVLAK, Angelo State University, San Angelo, TX;
C. FESTRAETS-HAMOIR, Brussels, Belgium; VINAYAK GANESHAN, student, University of Wa-
terloo, Waterloo, Ontario; RICHARD I. HESS, Rancho Palos Verdes, CA, USA; JOE HOWARD,
Portales, NM, USA; WALTHER JANOUS, Ursulinengymnasium, Innsbruck, Austria; GEOFFREY

A. KANDALL, Hamden, CT, USA; V ACLAV KONECN Y,
Ferris State University, Big Rapids, MI,
USA; MITKO KUNCHEV, Baba Tonka School of Mathematics, Rousse, Bulgaria; KEE-WAI LAU,
Hong Kong, China; GERRY LEVERSHA, St. Paul's School, London, England; DAVID LOEFFLER,
117
student, Cotham School, Bristol, UK; DAVID E. MANES, SUNY at Oneonta, Oneonta, NY, USA;
REVAI MATH CLUB, Gyor, } Hungary; JOEL SCHLOSBERG, student, New York University, NY,

USA; HEINZ-JURGEN SEIFFERT, Berlin, Germany (2 solutions); ECKARD SPECHT, Otto-von-
Guericke University, Magdeburg, Germany; KENNETH M. WILKE, Topeka, KS, USA; LI ZHOU,
Polk Community College, Winter Haven, FL, USA; and the proposer. There was one incorrect
solution.
Most solvers used a variation of solution III above. Kandall observes \It is easy to show
that if a = 2c, then EA = ED regardless of the length of AC ", where a and c, of course, are
the lengths of the sides opposite A and C , respectively.
q more general problem with BC = a,
Kandall, Kone cny, and Seiert all consider the
AB = c and DE = EA, and conclude that AC = 12 a(a + c).
2607. [2001 : 49] Proposed by Vaclav Konecny, Ferris State Univer-

sity, Big Rapids, MI, USA.
(a) Suppose that q > p are odd primes such that q = pn +1, where n is an
integer greater than 1. Let z be a complex number such that zq = 1.
z p;1 qX;
b j;p 1 czj .
1
Prove that p
z + 1 = j (;1) =1
(b) Suppose that q > 3 is an odd prime such that q = 3n + 2, where n is

an integer greater than 1. Let z be a complex number such that zq = 1.
;
qX
Prove that z ; 1 = (;1)b j;3 3 czj .
3 1
z +1 3
j =1
Solution by Michel Bataille, Rouen, France.
(a) Observing that n is necessarily even, we may write
;1
qX pn
(;1)b j;p 1 c zj = (;1)b j;p 1 c zj
X
j =1 j =1
X p p
X X p X p
= zj ; zj p + zj p ; ; zj n; p.
+ +2 +( 1)
j =1 j =1 j =1 j =1
If z = 1, this sum is p ; p + p ; p + + p ; p = 0. Note that
zp ; 1 = 0 as well.
zp + 1
If z 6= 1, the sum is
z zz ;;11 ; zp zz ;;11 + ; z n; p zz ;;11
p +1
p p ( 1) +1
= zz ;;11 (z) 1 ; zp + z p ; ; z n; p
p
2

( 1)
p 1 (;z )(z q;1 ; 1)

= zz ;;11 (z) ((;;zz))p ;;11 = zzp ; zp ; 1
p pn
+1 z;1 = zp + 1
118
(the latter, using zq; = z1 ). Thus, we have (a). Note that it is not necessary
1
to suppose that p and q are primes | it is sucient to require that p and q

are odd.
(b) This time, n is necessarily odd. We have
;1
qX n+1
(;1)b j;3 3 c zj = (;1)b j;3 3 czj
X
3
S =
j =1 j =1
= ;z ; z + z + z + z ; z + z + z +
2
; 3 4 5
; 6 7 8

; z n; + z n; + z n; 3(
+z n+z n
1) 3( 1)+1 3( 1)+2 3 3 +1
= ;z ; z + z n + z n
2 3 3

+1
+z 1 + z + z 1 ; z + z ; ; z n; .
; 3 2 3 6 3( 2)
3
If z = 1, we have S = 0 = zz3 ;
+1
1
.
If z 6= 1, we have
3 3 n;1
S = ;z ; z + zq; + zq; + zz ;; z ; ;zz3 .
2 2 1 1
1
3 1 (
1+
)
Using zq; = z12 , zq; = z1 and z n; = z15 , we obtain

2 1 3 3

S = zz3 ;
3 1 ;z ( + 1)( z3 + 1) + z3 1 ; 1 = z3 ; 1 .
+1 z 2 z;1 z5 z3 + 1
This completes the proof of (b). Similarly to (a), it is sucent to require that
q is odd.
Also solved by AUSTRIAN IMO{TEAM 2001; VINAYAK GANESHAN, student, University
of Waterloo, Walerloo, Ontario; WALTHER JANOUS, Ursulinengymnasium, Innsbruck, Austria;
GERRY LEVERSHA, St. Paul's School, London, England; HENRY LIU, student, University of
Memphis, Memphis, TN, USA; DAVID LOEFFLER, student, Cotham School, Bristol, UK; JOEL
SCHLOSBERG, student, New York University, NY, USA; HEINZ-JURGEN SEIFFERT, Berlin,
Germany; CHRIS WILDHAGEN, Rotterdam, the Netherlands; and the proposer.
In fact, only Janous and Seiert considered the case: z = 1. The other solvers ignored
it. The editor was generous in not classing their solutions as incomplete.
119
2612?. [2001 : 49] Proposed by Walther Janous, Ursulinengymnas-

ium, Innsbruck, Austria.
Two \Galton"{gures are given as follows:
A B
6
n
? -
n + k (k < n)
(There are n levels in total; there are k levels such that there is no
\intersection" between the levels emanating from A and B .)
Let two balls start at the same time from A and B . Each ball moves
either . or & with probability . 1
2
Determine the probability P (n; k) (1 k < n) such that the two balls
reach the bottom level without colliding.
Solution by Li Zhou, Polk Community College, Winter Haven, FL, USA.
Denote by c(n; k), the number of pairs of colliding paths. Then,
c(n; k) = 0 if k n, c(n; n ; 1) = 1, and c(n; 0) = 2n; 2n; = 2 n; . 1 1 2 2
Also, there are c(n ; 1; k) pairs of colliding paths starting with (.; .),
c(n ; 1; k + 1) with (.; &), c(n ; 1; k ; 1) with (& ; .), and c(n ; 1; k)
with (& ; &). Therefore,
c(n; k) = c(n ; 1; k ; 1) + 2c(n ; 1; k) + c(n ; 1; k + 1) .
n;X k ; 2n ; 2
2n ; 2 2
We claim that c(n; k) = + 2

n;k;1 i i . =1
m

m ; 2
m ; 2
m ; 2
Indeed, using the identity
j = j;2 +2 j;1 + j ,
we see easily that it satises the recursion. Moreover, since

2n ; 2 + 2 nX ; 2n ; 2
2
X n 2n ; 2 2
n;1 i = i = 2 n; , 2 2
i =0 i =0
we see that all three boundary conditions are satised as well. Hence,
nXk; 2n ; 2
P (n; k) = 1 ; c2(n;n;k) = 2 2 ;n;c(n; k) = 2 n1;
n; 2 2 + 1
2 2
i n;k
2
i .
2 2 2
=

Also solve by MANUEL BENITO and EMILIO FERNANDEZ, I.B. Praxedes Mateo Sagasta,
Logro~no, Spain; KEITH EKBLAW, Walla Walla, WA, USA; RICHARD I. HESS, Rancho Palos
120
Verdes, CA, USA; ERIC POSTPISCHIL, Nashua, NH, USA; and JOEL SCHLOSBERG, student,
New York University, NY, USA. There was one incomplete solution.
The submitted solutions varied in length from the above to one of ten pages. Postpischil
noted that, in the CRC Concise Encyclopedia of Mathematics, CRC Press, Boca Raton, FL, USA
(1999), p. 138, Eric Weisstein shows how to write the sum in terms of the Beta Function (also
known as the Eulerian Integral of the Second Kind) and the incomplete Beta Function.
2613. [2001 : 136] Proposed by Murray S. Klamkin, University of

Alberta, Edmonton, Alberta. Dedicated to Toshio Seimiya, and based on his
problem 2515.
In 4ABC , the three cevians AD, BE and CF through a non-exterior
point P are such that AF + BD + CE = s (the semi-perimeter).
Characterize 4ABC for each of the cases when P is (i) the orthocentre, and
(ii) the Lemoine point.
[Ed. The Lemoine point is also known as the symmedian point. See, for ex-
ample, James R. Smart, Modern Geometries, 4th Edition, 1994, Brooks/Cole,
California, USA. p. 161.]
Solution by Theoklitos Paragiou, Limassol, Cyprus, Greece.
In both cases 4ABC is isosceles.
(i) When P is the orthocentre, AD, CF ,2 and 2BE 2are the altitudes.
We have therefore, BD = c cos B = c + 2aa ; b , and likewise,
2 2 2 2 2 2
CD = a + 2bb ; c , and AF = b + c2c ; a . Thus, the following state-
ments are equivalent to the given condition:
AF + BD + CE = s ,
b2 + c2 ; a2 + c2 + a2 ; b2 + a2 + b2 ; c2 = a + b + c ,
2c 2a 2b 2
ab(b ; a ) + bc(c ; b ) + ac(a ; c ) = 0 ,

2 2 2 2 2 2
(a ; b)(b ; c)(c ; a)(a + b + c) = 0 .

The last equation holds if and only if the given triangle is isosceles.
(ii) Standard references tell us that when P is the Lemoine point,
2 2 2
AF = a2cb b2 , BD = b2ac c2 , and CE = c2ba a2 .
+ + +
Thus, the following statements are equivalent to the given condition:

AF + BD + CE = s ,
cb2 + ac2 + ba2 = a + b + c ,
a2 + b2 b2 + c2 c2 + a2 2
(a ; b)(b ; c)(c ; a)[a (b + c) + 2a bc + a(b + 2b c + 2bc2 + c3 ) + bc(b2 + c2 )] = 0 .

3 2 3 2
2(a2 + b2 )(b2 + c2 )(c2 + a2 )
121
Once again, the last equation holds if and only if the given triangle is
isosceles.
Also solved by MICHEL BATAILLE, Rouen, France; FRANCISCO BELLOT ROSADO,
I.B. Emilio Ferrari, Valladolid, Spain; CHRISTOPHER J. BRADLEY, Clifton College, Bris-
tol, UK; JOHN G. HEUVER, Grande Prairie Composite High School, Grande Prairie, Alberta;
WALTHER JANOUS, Ursulinengymnasium, Innsbruck, Austria; HENRY LIU, student, Univer-
sity of Memphis, Memphis, TN, USA; DAVID LOEFFLER, student, Trinity College, Cambridge,
UK; JOEL SCHLOSBERG, student, New York University, NY, USA; D.J. SMEENK, Zaltbom-
mel, the Netherlands; ECKARD SPECHT, Otto-von-Guericke University, Magdeburg, Germany;
PETER Y. WOO, Biola University, La Mirada, CA, USA; and by the proposer.
Most solvers used essentially the same argument as in the featured solution, sometimes
with more detail, sometimes with less. Together with problem 2515 [2000 : 114; 2001 : 144],
we now have established that
AF + BD + CE = s if and only if 4ABC is isosceles
when P is the incentre, the orthocentre, and the Lemoine point. David Loeer continues the
theme with the comment:
In fact, huge numbers of triangle centres may be dealt with in the same way, if you have a
computer algebra system or incredible patience! The Mittenpunkt | Kimberling's X(9) [Clark
Kimberling, Encyclopedia of Triangle Centers, http://cedar.evansville.edu/ck6/tcenters/;
or Math. Mag. 67:3 (June 1994), 163{187] | satises the condition only for isosceles triangles,
although this is slightly more tricky to prove than the above. The same is true of the Spieker
center X(10), all of the power points, and various others such as X(37), X(38), X(39), X(42), and
X(43).
Klamkin wondered whether the same is true of the circumcentre. However, before we
get carried away with wild conjectures note that AF + BD + CE = s for all triangles when
P is either the centroid or the Gergonne point.

Alberta, Edmonton, Alberta. Dedicated to Toshio Seimiya, and suggested by
his problem 2514.
In 4ABC , the two cevians through a non-exterior point P meet
AC and AB at D and E , respectively. Suppose that AE = BD and
AD = CE . Characterize 4ABC for the cases when P is (i) the orthocentre,
(ii) the centroid, and (iii) the Lemoine point.
Solution by Henry Liu, student, University of Memphis, Memphis, TN,
USA.
(i) When P is the orthocentre, BD and CE are the altitudes. By
SAS and the given conditions, 4ADB = 4CEA, so that AB = AC
and consequently, \ABC = \ACB . Further, 4BEC = 4CDB since
they are similar right triangles that share their hypotenuse BC . Thus,
BD = CE = AD. Therefore. 4ABC is isosceles with A = 45 and
B = C = 67:5 .
(ii) When P is the centroid, E and D are mid-points of their respec-
tive sides. By SSS , 4BEC = 4BDC . Since these triangles have BC in
common while the vertices D and E lie on the same side of the line BC ,
the two triangles must coincide (with E and D the same vertex). Since AB
122
contains E and AC contains D, the two sides of 4ABC coincide, and the
triangle is therefore degenerate (with \A = 0 and the vertices B and C
coincident).
(iii) As in problem 2613, when P is the Lemoine point,
2 2
AE = a2cb b2 and EB = a2ca b2 . + +
By Stewart's
Theorem we have
c(CE + AE EB) = a AE + b EB , so 2 2 2

abc a 2 b2 c
that c CE + a2 b2 = a2 b2 and therefore,
2
2 2
+ +

CE = a b 2 2 2 2 a2 + 2b2 ; c2 .
(a2 + b2 )2
Similarly, we have
2
AD = a2bc c2 and BD = a c 2 2 2 2a2 + 2c2 ; b2 .
+ (a2 + c2 )2

AE = BD , so that
The given conditions imply AD
2 2
CE
2
cb2 (a2 + c2 ) = a2 c2 (a2 + b2 )2 2a2 + 2c2 ; b2 . (1)
bc2 (a2 + b2 ) a2 b2 (a2 + c2 )2 2a2 + 2b2 ; c2
Set x = a , y = b , z = c , and (1) reduces to
2 2 2
y (2x + 2y ; z)(x + z) ; z (2x + 2z ; y)(x + y) = 0 .

2 4 2 4
Since y = z satises the equation, we can factor out (y ; z) to get

(y ; z)(y z ; y z + y z + P (x; y; z)) = 0 ,
4 2 3 3 2 4
where P (x; y; z) is a polynomial in x, y, and z having all its terms positive.

Note that the rst three terms in the factor on the right satisfy
y z ; y z + y z = y z ((y ; z) + yz) > 0 .
4 2 3 3 2 4 2 2 2
Since y ; z = (b ; c)(b + c), we see that (1) is equivalent to

(b ; c)Q(a; b; c) = 0 ,
where Q is a polynomial that is positive for all positive values of a, b, and c.
We conclude that b = c, so that 4ABC is isosceles.
It remains to determine the shape of 4ABC . [Here the editor has
replaced Liu's argument with a shorter calculation.]

AE = BD , and set b = c to get
Instead of (1), use CE
2 2
AD
b = a (2a + b ) .
4 2 2 2
123
Since a, b > 0, thispequation implies that b; = 2a. Therefore, 4ABC must

p
satisfy b = c = 2a with \A = cos; . One easily checks that these
1 3
conditions are also sucient.

4
Also solved by CHRISTOPHER J. BRADLEY, Clifton College, Bristol, UK; JOHN

G. HEUVER, Grande Prairie Composite High School, Grande Prairie, Alberta; WALTHER
JANOUS, Ursulinengymnasium, Innsbruck, Austria; DAVID LOEFFLER, student, Trinity Col-
lege, Cambridge, UK; JOEL SCHLOSBERG, student, New York University, NY, USA ((i) and
(ii) only); PETER Y. WOO, Biola University, La Mirada, CA, USA ((i) and (ii) only); and the
proposer.
2615. Proposed by Murray S. Klamkin, University of Alberta,

Edmonton, Alberta.
Suppose that x , x , : : : , xn , are non-negative numbers such that
1 2
(x x ) = n(n2+ 1) ,
X X
x + 2
1 1 2
2
where the sums here and subsequently are symmetric over the subscripts
1, 2, : : : , n.
(a) Determine the maximum of P x . r 1
(b)? Prove or disprove that the minimum of x is n(n2+ 1) .

P
1
Solution by Heinz-Jurgen Seiert, Berlin, Germany.

(a) The given equation,
n
(xj xk ) = n(n2+ 1) ,
X X
xk +
2 2
(1)
k=1 1j<kn
is satised when x = x = ::: = xn = 1, so that the maximum of P x is
greater than or equal to n. Using the trivial inequality 2t ; t 1, t 2 R,
1 2 1
2
we obtain
!2
X 2 n
X n
X X
x 1 = xk = xk + 2
2
xj xk
k=1 k=1 1j<kn
n
X X X
= xk + 2
2
xj xk + (xj xk ) ; (xj xk )
2 2
k=1 1j<kn 1j < kn

= n(n2+ 1) +
X
2xj xk ; (xj xk ) 2
1j<kn
n(n2+ 1) +
X
1 = n , 2
1j<kn
or P
x n. Hence, the maximum of P x is n.
1 1
124
(b) Clearly, if n = 1, then the minimum of x is 1.

P
1
Let n = 2. From the condition x + x + (x x ) = 3, we have

x x pp
2 2 2
1 2
3 < 2, so that (x + x ) = 3 + x x (2 p ; x x ) 3, or
1 2
2
x + x 3. Since an equality x 3 and x = 0, it

1 2 1 2 1 2 1 2
is attained
p when =
follows that the minimum of x is 3.
1 2
P 1 2
Let n = 3. First, suppose that maxfx x p , x x g > 2. If x x > 2,

,x x p
then x + x + x x + x 2px x > 2 2 > 6. Otherwise, we have
1 2 1 3 2 3 1 2
(x +x +x ) = 6+px x (2;x x )+x x (2;x x )+x x (2;x xp) 6,

1 2 3 1 2 1 2
2
or x + x + x 6. Since an equality is attained when x = 6 and

1 2 3 1 2 1 2 1 3 1 3 2 3 2 3
x = x = 0, we conclude that the minimum of P x is p6.

1 2 3 1
2 3 1
We choose
r to disprove for n 4. The equation (1) is satised when
p
x =x =
1 2 2n(n + 1) + 4 ; 2 and x = x = = xn = 0.
1
2 3 4
r q
p
However, x = 2 2n(n + 1) + 4 ; 2 < n(n + 1), because the
P 1
1 2
last inequality easily reduces to n(n + 1) > 16, which is true for n 4.
Also solved by PETER Y. WOO, Biola University, La Mirada, CA, USA. Part (a) only was
solved by the proposer. One solver sent an incomplete solution. Another solver misinterpreted
the condition and solved a dierent problem.

Alberta, Edmonton, Alberta.
A problem in one book was to prove that each edge of an isosceles
tetrahedron is equally inclined to its opposite edge. A problem in another
book was to prove that the three angles formed by the opposite edges of a
tetrahedron cannot be equal unless they are at right angles.
1. Show that only the second result is valid.
2. Show that a tetrahedron which is both isosceles and orthocentric must
be regular.
Solution by Joel Schlosberg, student, New York University, NY, USA.
If the three edges that come out of a single vertex of the tetrahedron
are labelled as the three vectors a, b, c, then the six sides of the tetrahedron
are a, b, c, c ; b, b ; a, a ; c.
If X , Y and Z are the three angles formed by the opposite edges, then
cos X = ajajj(cc ;; bb)j = aja jjc c;;ab jb ,
cos Y = ajbbjja;;bcjc , cos Z = bjc jjcb;;aajc .

125
(The \angle between the opposite segments" is a slightly ambiguous

phrase which could refer to either of the two supplementary angles. How-
ever, this does not aect the problem, since the [absolute values of the]
cosines of two supplementary angles are the same.)
1. A counterexample to the rst result is the tetrahedron with
vertices (0; 0; 0), (0; 2u; 0), (u; u; v), (;u; u; v), where 2u 6= v , so that
2 2
a = (0; 2u; 0), b = (u; u; v) and c = (;u; u; v). A simple calculation shows
that
jc ; bj = j(;2u; 0; 0)j = j(0; 2u; 0)j = jaj ,
ja ; cj = j(u; u; ;v)j = j(u; u; v)j = jbj ,
jb ; aj = j(u; ;u; v)j = j(;u; u; v)j = jcj ,
so that the tetrahedron is isosceles. However,
cos X = aja jjc c;;ab jb = 2u ja;j 2u = 0 ,
2 2
but
cos Y = ajbbjja;;bcjc = 2ujb;j v 6= 0 .
2 2
Therefore cos X 6= cos Y , and these angles are not equal.

To prove the second result, suppose that the three angles formed
by opposite sides are equal. Without loss of generality, assume that
a b a c b c. Then
j cos X j = j cos Y j = j cos Z j ,
and thus,
ac;ab = bc;ab = bc;ac.
jajjc ; bj jbjja ; cj jcjjb ; aj
However, any equation of the form D(x ; y) = E (z ; y) = F (z ; x)
has the unique solution x = y = z. (The equivalent homogeneous linear
system Dx + (E ; D)y ; Ez = 0, Fx ; Ey + (E ; F )z = 0, has the
solution set x = y = z; these are the only solutions, since any solution to a
given n{equation, (n +1){variable system of homogeneous linear equations
is a scalar multiple of any other solution)
Therefore a b = a c = b c, so that a (c ; b) = 0, b (a ; c) = 0 and
c (b ; a) = 0. Hence a ? (c ; b), b ? (a ; c) and c ? (b ; a), implying
that the opposite angles are right angles.
2. If a tetrahedron with sides a, b, c, c ; b, b ; a, a ; c is isosceles,
then jaj = jc ; bj, jbj = ja ; cj and jcj = jb ; aj. Therefore,
jaj = jc ; bj or a a = (c ; b) (c ; b) = c c + b b ; 2b c ,
2 2
126
and thus,
;a a + b b + c c = 2b c .
Similarly
a a ; b b + c c = 2a c and a a + b b ; c c = 2a b .
If the tetrahedron is orthocentric, then a (c;b) = b (a;c) = c (b;a) = 0,
and further, we have a b = a c = b c. Combining this with the previous
equations gives us
;a a + b b + c c = a a ; b b + c c = a a + b b ; c c .
Combining these equations pairwise allows us to deduce that a a = b b =
c c, and hence that jaj = jbj = jcj. Combining this with the equations
jaj = jc ; bj, jbj = ja ; cj and jcj = jb ; aj shows us that all six sides are
equal in length, so that the tetrahedron is regular.
Also solved by JOHN G. HEUVER, Grande Prairie Composite High School, Grande
Prairie, Alberta; PETER Y. WOO, Biola University, La Mirada, CA, USA; and the proposer.
2618 Proposed by Murray S. Klamkin, University of Alberta, Ed-

monton, Alberta.
Determine a geometric problem whose solution is given by the positive
solution of the equation
!
x2 p
1
p
1
p
1
R2 + x2 ; a2 4R2 + x2 ; b2 4R2 + x2 ; c2
3 + +
4
p p p
= 4R2 + x2 ; a2 + 4R2 + x2 ; b2 + 4R2 + x2 ; c2 + a + b + c ,
where a, b, c and R are the sides and circumradius of a given triangle ABC .
Solution by the proposer.
We show that x is the altitude to the face ABC of an orthocentric
tetrahedron of maximum isoperimetric quotient, EV3 , where V and E are
the volume and total edge length of the tetrahedron, respectively.
If PABC is an orthocentric tetrahedron, then P must lie on a line
through H , the orthocentre of ABC , and perpendicular to the plane of
ABC . Then
p p
PA = 4R cos A + x = 4R + x ; a , etc.,
2 2 2 2 2 2

d V
and 3V = x[ABC ]. The given equation corresponds to dx = 0.
E3
That the maximum is unique follows by dividing both sides of the given equa-
tion by x, and then noting that the left hand side is an increasing function
of x, whereas the right hand side is a decreasing one.
No other solutions were received.
127
2620. Proposed by Bill Sands, University of Calgary, Calgary, Al-

berta, dedicated to Murray S. Klamkin, on his 80th birthday.
Three cards are handed to you. On each card are three non-negative
real numbers, written one below the other, so that the sum of the numbers
on each card is 1. You are permitted to put the three cards in any order
you like, then write down the rst number from the rst card, the second
number from the second card, and the third number from the third card.
You add these three numbers together.
Prove that you can always arrange the three cards so that your sum lies
in the interval ; . (Corrected)
1 3
Solution by Michel Bataille, Rouen, France.

2 2
Denote by (a ; a ; a ), (b ; b ; b ), (c ; c ; c ) the triples of numbers

1 2 3 1 2 3 1 2 3
respectively written on each of the cards. By hypothesis, these numbers are
non-negative and
1a + a + a = b + b + b = c + c + c = 1.
2 3 1 2 3 1 2 3
Following the process described in the statement of the problem, we can form
six sums, namely
s = a +b +c ,
1 1 s = a +c +b ,
2 3 2 1 2 3
s = b +a +c ,
3 1 s = b +c +a ,
2 3 4 1 2 3
s = c +a +b ,
5 1 s = c +b +a .
2 3 6 1 2 3
For the purpose of contradiction we will suppose that si 2= ; for i = 1,

1 3
2, : : : , 6. Since s + s + s = 3, one of the numbers s , s , s must be at

2 2
least 1, and one must be at most 1. Say, s 1 and s 1. By supposition,

2 3 6 2 3 6
we even have s < and s > . The latter implies that we cannot have
2 6
1 3
2 6
s > (otherwise s + s > 3); hence s < . Now, s + s < 1, which

2 2
3 1
means that we actually have s > 2, or

3 2 3 6 3 2 2 3
c + b + a > 2. 1 2 (1) 3
Similarly, since s + s + s = 3, two of the numbers s , s , s are less

1 4 5 1 4 5
than (and the third is greater than 2).
1
2
If s < and s < , then b < and a < , which implies b +a < 1.
1 1 1 1
From (1) we have c > 1, contradicting c c + c + c = 1.

1 2 4 2 2 2 3 2 2 3
1 1 1 2 3
If s < and s < , then b < and c < , which implies b + c < 1.
1 1 1 1
From (1) we have the contradiction a > 1.

1 2 5 2 2 2 1 2 2 1
If s < and s < , then a < and c < , which implies a + c < 1.
1 1 1 1
From (1) we have the contradiction b > 1.

4 2 5 2 3 2 1 2 3 1
In the same way it is readily checked that we are also led to a contradic-
tion when the condition (1) is replaced by s > 2 or s > 2. The conclusion 2 3
follows.
128
[Editor's comment: When this problem was originally printed, the in-
terval that appeared was ; . (This problem is impossible since the inter-
1 2
val is too restricted.) A subsequent issue, [2001 : 213], (incorrectly) corrected

3 3

the interval to ; . This new problem can now be proven, but is not as
1 3
sharp as it could be, since we can improve the lower bound on the interval. In
3 2
the next issue, [2001 : 267], the problem was further corrected to the (cor-
rect) interval listed in the problem statement above. Unfortunately, some
solvers missed this last correction, and solved the previously stated (weaker)
problem. As a result, we have split the solvers into two groups: those who
have solved the intended problem, and those who solved the weaker prob-
lem.]
Also solved by CHARLES R. DIMINNIE, Angelo State University, San Angelo, TX, USA;
RICHARD I. HESS, Rancho Palos Verdes, CA, USA; KEE-WAI LAU, Hong Kong; JOEL
SCHLOSBERG, student, New York University, NY, USA; LI ZHOU, Polk Community College,
Winter Haven, FL, USA; and the proposer. The weaker version of the problem was cor-
rectly solved by CHRISTOPHER J. BRADLEY, Clifton College, Bristol, UK; ERIC POSTPISCHIL,
Nashua, NH; and CHRIS WILDHAGEN, Rotterdam, the Netherlands. There was one incorrect
solution.
Both Dimminie and the proposer show that these bounds are the best possible:
; 21 ; 12 ; 0, ; 12 ; 21 ; 0, (0; 0; 1) can give a sum of 12 or 2, and
; 21 ; 12 ; 0, (0; 0; 1), (0; 0; 1) can give a sum of 0 or 32 .
The proposer also asks about generalizing the problem to n cards of n numbers each. He
conjectures that the best intervals for achievable sums appear to be

1 ; 2 ; 1 + 2 for n even, 1 ; 2 ;1+ 2
n n n+1 n + 1 for n odd,
but he has no proof. Perhaps our readers can check these bounds and supply proofs!
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

THE OLYMPIAD CORNER - Canadian Mathematical Society (PDFDrive) PDF

Uploaded by

Copyright:

Available Formats

THE OLYMPIAD CORNER - Canadian Mathematical Society (PDFDrive) PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

THE OLYMPIAD CORNER - Canadian Mathematical Society (PDFDrive) PDF

Uploaded by

Copyright:

Available Formats

65

5. Given the angle XOY , variable points M and N are considered on

TURKISH TEAM SELECTION EXAMINATION

1. Let m  n be positive integers and p be a prime. Let p{expansions

respectively, where ar , bs 6= 0, for all i 2 f0, 1, : : : , rg and for all

2. Let L and N be the mid-points of the diagonals [AC ] and [BD] of

Second Day - March 21, 1999

4. Let the area and the perimeter of a cyclic quadrilateral C be AC and

5. Each of A, B, C , D, E and F knows a piece of gossip. They commu-

there is a natural number x, for which f (x) is divisible by 3n but not by

3. Let 2n + 1 weights (n is a natural number, n  1) satisfy the

cannot be expressed as a product of two integral-coecient polynomials with

Next, we turn to readers' comments and solutions to problems of the

i=1 i=1 i=1

Solution by Toshio Seimiya, Kawasaki, Japan.

intersection of ; and ; other than P . Then O O ? PQ.

Let H and K be the intersections of AP with DE and BC , respec-

Since DE kBC , we get

Since P and Q are intersections of ; and ; , we have that PQ is the

3. Let P (x) be a polynomial with rational coecients such that

Solutions by Pierre Bornsztein, Pontoise, France; and by Moubinool

P (Q)  Q (since P 2 Q[x]) (1)

obtain a monic polynomial, with (1) and (2).

Claim: If P 2 Z[x] is a monic polynomial, with degree greater than 1, then

which belong to I , where I varies over all open intervals of length 1,

From the claim, it follows immediately that the number of elements of

A , : : : , of S which may be constructed such that none of the Ai 's is included

Conversely: Let p be an integer such that n < p.

Let A be any n2 {subset of S = fx , : : : , xn g. Then,

5. Does there exist a function f : R ! R that ful ls all of the following

Solution by Mohammed Aassila, Strasbourg, France.

Thus, (n ; 1) < n + 1, and thus, n 2 f1, 2g. But, putting x = 1 in the

original equation, we get f (2) = 2, and therefore, n > 2, a contradiction.

Solution by Mohammed Aassila, Strasbourg, France.

0 < z = pcxy < 2 . 1

Setting x = 2 sin u, 0 < u <  , and y = 2 sin v, 0 < v <  , we have

4 = 4 sin u + 4 sin v + z + 4 sin u sin v z .

is the unique solution.

1. The numbers x, y and x2 + y2 + 6 are positive integers. Prove that

k = 2 + a2 we deduce immediately that a = 1. If a < b then b  a + 1.

if a > 6, contradicting the minimality of a. Hence, a  6. It remains to show

divides b and thus, b + 15  6, while 3kb  0 (mod 9). If a = 5, then we

relation between roots and coecients that 5k = 32, which 2

Hence, b = 7 and k = 8, which is a cube.

2. Let a, b, c be the sides of a triangle and , , the opposite angles

The product Rr can be neatly expressed by comparing two well-known for-

However, applying the AM-GM inequality, we see that

which we showed to be equivalent to the required inequality.

We prove the stronger inequality

where 18  x  24. The right-hand inequality reduces to (x;18)(R;2r) 

which is the well-known Chapple-Euler inequality. For the left-hand

where s is the semiperimeter. This reduces the inequality to

32Rr ; 10r  (x + 8)Rr + (38 ; 2x)r

or that (24 ; x)Rr  (48 ; 2x)r and which is valid if x  24 as well.

Comment. As a known complementary inequality, we also have

For one simple proof of this consider the expansion of (A + B + C )  0 2

where A, B , C are vectors from the circumcentre to the vertices A, B , C of

proof follows from OH = 9R ; (a + b + c ) where O and H are the

1. Let m n be positive integers and p be a prime. Let p{expansions

3. Let 2n + 1 weights (n is a natural number, n 1) satisfy the

cannot be expressed as a product of two integral-coecient polynomials with

3. Let P (x) be a polynomial with rational coecients such that

P (Q) Q (since P 2 Q[x]) (1)

5. Does there exist a function f : R ! R that fulls all of the following

Setting x = 2 sin u, 0 < u < , and y = 2 sin v, 0 < v < , we have

k = 2 + a2 we deduce immediately that a = 1. If a < b then b a + 1.

if a > 6, contradicting the minimality of a. Hence, a 6. It remains to show

divides b and thus, b + 15 6, while 3kb 0 (mod 9). If a = 5, then we

relation between roots and coecients that 5k = 32, which 2

where 18 x 24. The right-hand inequality reduces to (x;18)(R;2r)

32Rr ; 10r (x + 8)Rr + (38 ; 2x)r

or that (24 ; x)Rr (48 ; 2x)r and which is valid if x 24 as well.

For one simple proof of this consider the expansion of (A + B + C ) 0 2

Clearly ( ) is true since P (x) = 1 and P (x) = 1 + x.

Suppose now that (n ) is true for an integer n 0. We will denote by

Since \ADA = \BEB = CFC = 90 , we have that ; , ; , ;

bxc bx c bxpc bxp c

write it as A B. Then, if F is a convex function,

F (a + b ; c) + F (b + c ; a) + F (c + a ; b) F (a) + F (b) + F (c)

(2) Consider the bounds on a + a + + an where ai = S ( n) and

if and only if f (a )+ + f (an ) f (b )+ + f (bn ) for any convex func-

37. Propose par J. Walter Lynch, Athens, GA, USA.

because the binomial coecients are divisible by 3.

for all x with 0 < x < x we must have f (x) f (x ) + m(x ; x ).

But f (x) f (x ) + m(x ; x ) shows that, as x ! 0 , we have +

f (x) f (x ) ; mx , contradicting the given limx! + f (x) = ;1.

Similarly, if f 00 (x) 0 for all x in (0; 1) then for all x with