Crux

Mathematicorum
Published by the Canadian Mathematical Society.

http://crux.math.ca/

The Back Files

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

Journal title history:

➢ The first 32 issues, from Vol. 1, No. 1 (March 1975) to
Vol. 4, No.2 (February 1978) were published under the
name EUREKA.
➢ Issues from Vol. 4, No. 3 (March 1978) to Vol. 22, No.
8 (December 1996) were published under the name
Crux Mathematicorum.
➢ Issues from Vol 23., No. 1 (February 1997) to Vol. 37,
No. 8 (December 2011) were published under the
name Crux Mathematicorum with Mathematical
Mayhem.
➢ Issues since Vol. 38, No. 1 (January 2012) are
published under the name Crux Mathematicorum.
 ISSN 0705 - 0348

C P UX FATHEfATICQRUM

Vol. 10, No. 9

November 1984

Sponsored by
Carle ton-Ottawa Mathematics Association Mathematique d" Ottawa-Carle ton
Ptafalie par le College Algonquin,- Ottawa
The assistance of the publisher and the support of the Canadian Mathematical Society,
the Carleton University Department of Mathematics and Statistics, the University of Ottawa
Department of Mathematics, and the endorsement of the Ottawa Valley Education Liaison Council
are gratefully acknowledged.
***************************************
CRUX MATHEMATICOROM is a problem-solving journal at the senior secondary and university
undergraduate levels for those who practise or teach aathesatiGs„ Its purpose is primarily
educational, but it serves also those who read it for professional, cultural, or recreational
reasons.
It is published monthly {except July and August} . S a yearly subscription rate for ten
issues is $22 in Canada, $24*50 (or 0SS2O) elsewhere^ Back issuess each $2,25 in Canada,
$2.50 (or US$2) elsewhere. Bound volumes with index: Vols. 1&2 (combined) and each of Vols.
3-9, $17 in Canada, $18.40 (or OS$15) elsewhere. Cheques and money orders, payable to CRUX
MATHEHRTXCOROM, should be sent to the managing editor.
All communications about the content (articles, problems, solutions, etc.5 should be
sent to the editor. All changes of address and inquiries about siabscriptions and back issues
should be sent to the managing editor.
Editor: Leo Sauve, Algonquin College, 140 Main Street, Ottawa, Ontario, Canada K1S 1C2.
Managing Editor: F.G.B. Maskell, Algonquin College, 200 Lees Ave., Ottawa, Ontario,
Canada K1S 0C5.
Typist-compositor: Kuen Huynh.
Second Class Mail Registration No. 5432. Return Postage Guaranteed.

CONTENTS

The Extended Erdos-Mordell I n e q u a l i t y . . . . . Clayton W. Dodge 274

The Olympiad Corner: 59 . . . . . . . . . . . . . M.S. Klamkin 282
The Puzzle Corner . . . . . . . . . . . . . . . . . . . . . . . 290, 292
Problems - Probl§mes 291
Solutions 293

- 273 -
 - 274 -
THE EXTENDED ERDOS-MORDELL INEQUALITY
CLAYTON W. DODGE

Ten years ago The American Mathematical Monthly published the following Prob-
lem E 2462 [81 (i97n) 281] s which is an extension of the earlier Problem 3740
proposed by Paul Erdos [>2 (1935) 396] and first solved by L.J. Mordell [44 (1937)
252-254]:
!f
E 2462. Proposed by Huseyin Demirs Middle East Technical University 5 Ankara^
Turkey.
Let P be a point interior to the triangle AiA 2 A 3 , Denote by R. the distance
from P to the vertex A., and denote by r. the distance from P to the side a. oppo-
^ ^ i
s i t e to A , , The Erdos-Mordell i n e q u a l i t y asserts t h a t
R\ + R2 + R3 ^ 2 ( r i + r 2 + 2*3).

Prove t h a t the above i n e q u a l i t y holds f o r eyery p o i n t P in the plane of AiA 2 A 3 when

we make the i n t e r p r e t a t i o n R. z o always and r. is p o s i t i v e or negative depending
on whether P and A. are on the same side of a . or on opposite s i d e s . "
I t was my pleasure i n 1974 to referee the solutions to t h i s problem. Curiously 9
each of the solvers s t a r t e d w i t h the s o l u t i o n to the o r i g i n a l Erdbs i n e q u a l i t y given
by Kazarinoff i"i] and modified i t f o r the case where v\9 r2, or r 3 is n e g a t i v e .
Each made the same e r r o r , i n v a l i d a t i n g the proof. C u r i o u s l y , Kazarinoff s t a t e d t h a t
his proof "holds even i f P l i e s outside the t r i a n g l e , provided i t remains inside the
c i r c u m c i r c l e " , but the Elementary Problem Department e d i t o r s could not see t h a t such
an extension of the proof was possible w i t h o u t committing the same e r r o r the other
solvers had made. We o u t l i n e K a z a r i n o f f ' s proof and describe the e r r o r . Since we
shall r e l y h e a v i l y on t h i s p r o o f , our o u t l i n e i s q u i t e complete. I t is interesting
to note t h a t , i f Kazarinoff s s statement could have been v e r i f i e d t h e n , a proof would
have been published i n 1975.
In Demir's n o t a t i o n * Kazarinoff l e t P l i e w i t h i n angle Ai and then he r e f l e c t e d
t r i a n g l e AiA 2 A 3 i n the b i s e c t o r AiT of angle Ai i n t o t r i a n g l e A2A2A3s as shown i n
Figure i . Noting t h a t the b i s e c t o r of angle Ai also bisects the angle between the
a l t i t u d e AjD and the circumradius OAi, he used a theorem of Pappus which states t h a t
the area of the parallelogram whose adjacent sides are AiA2' and AiP plus the area of
the parallelogram whose adjacent sides are AiP and A1A3 is equal to the area of the
parallelogram erected on A2A3 whose sides emanating from A2 and A3 are equal , as
v e c t o r s , to A^P. Since A ^ = RX S Kazarinoff obtained the f i r s t o f e q u a l i t i e s (i)f
and the other two are obtained i n the same way when P l i e s w i t h i n angles A2 and A 3 :
 - 275 -

Figure 2
Figure 1

laiEi COS (OA^) = a2r3 + a 3 r 2

cos
\a2R2 (0A2P) = a$ri + a\r^ (1)

\cz3P3 COS (OA3P) = a\r2 + CL2r\.

From this follows 9 when P is an i n t e r i o r point of the triangle^

[f?! + F2 + #3 ^ i?i COS (OAjP) + i?2 COS (0A2P) + #3 COS (0A3P)

fa3,al\„ . tal,a2\ (2)
a3 a2 l aj a 3 ' 2 #2 al 3

2(ri + 2*2 +2*3)

because x f i/x > 2 when x > 0.

Equations ( l ) hold for all locations of P9 provided Demir's sign convention is
observed* Then also the f i r s t two lines of (2) hold. We shall make use of this in
our proofs l a t e r in this paper, so a proof is presented.
Let P l i e outside angle Ai and outside angle A2 but inside angle A3 and inside
the circumcircle of t r i a n g l e A1A2A3 , as shown in Figure 2, Using the notation of
Figure i s we see that the parallelogram on side A^A^ now is the difference between
those on sides AiA£ and AiA3'. Accordingly,
 - 276 -
a\R\ cos (OAjP) = -^2^3 + a^T2 ,
where we take the r. all nonnegative; similarly^
0.282 cos (0A2P) = a^v\ - ai^3,
and we have as before
a3i?3 cos (OA3P) = a\T2 + ^2^1,
since P lies within angle As* Thus equations (l) are true for this case if we
observe Demir's sign convention. That they also hold in other cases is not needed
here. Since the cosines of the angles 0A*P are all s t i l l positive because P lies
inside the circuincircle, the f i r s t two lines of (2) both s t i l l hold. Only the third
line of (2) is in doubt. In fact, Kazarinoff's argument fails at this point 9 as
explained in the next paragraph.
The error in the submitted solutions to Problem E 2M-62 occurred when one of the
r.9 say r 3 , is negative. Then we s t i l l have
a\ a2
— + — > 29
a2 ax

buts since r 3 < o9 the inequality reverses to give

rendering the argument inconclusive. The editors could find no simple remedy for
this flaw since the extended theorem requires that either one or two of the distances
v. be negative. We wrote to those who had submitted solutions, and Leon Bankoff
and I corresponded for perhaps a year in attempting to put together a satisfactory
proof. Over the next nine years I returned to the problem from time to time, fas-
cinated by its challenge.
Two cases were disposed of almost immediately.
Case 1. Point P lies inside the angle vertical to a vertex angle.
For example9 let P lie inside the angle vertical to Ai, as shown in Figure 3.
Then r 2 a n d ^3 are to be taken negative and we must prove that
i?l + i? 2 + i?3 ^ 2 ( r i - r2 - 1*3 ) ,

where the r. have all been taken nonnegative and we have inserted the appropriate
negative signs. Because i?2 and r\ are hypotenuse and leg of a right triangle, we
have
i?2 ^ r\> and similarly i?3 > r\.
Thus
R\ + f?2 + #3 ^ # 2 + i?3 Z r\ + 2»i £ 2 ( P I - V2 - r3 ) . D
 _ 27^ -

A2 F3 A2

Figure 3 Figure 4

Case r. Point P is i n t e r i o r to an angle o f the t r i a n g l e ^ but f a r enough o u t -

side the t r i a n g l e so t h a t a f o o t F. of a distance r. l i e s outside the t r i a n g l e .
As shown i n Figure 4 9 we take P l y i n g w i t h i n angle Ai and f a r enough outside
the t r i a n g l e so t h a t , say f the f o o t F3 o f distance r3 l i e s outside the t r i a n g l e .
Then z»x i s taken n e g a t i v e . Choose p o i n t A2 so t h a t F3 is the midpoint of segment
A2A2, Then PA2 = PA2 and the three distances R. f o r t r i a n g l e AiA 2 A 3 are the same
as those f o r t r i a n g l e AjA^A^ Also r 2 and p 3 remain unchanged, and only r x changes
to r{ . 11% as p i c t u r e d , P l i e s outside t r i a n g l e A X A 2 A 3 , then \v\\ > \r{\ and
-J 3 ! < -r[ since they both must be taken negative e I f P l i e s inside t r i a n g l e A X A 2 A 3 ,
we get -r 2 < o < r{. So i n e i t h e r case f using the appropriate s i g n s we have

Tr{ + r>2 + r 3 > - r i •*- r 2 + r 3 .

I t t h e r e f o r e s u f f i c e s to prove the extended theorem i n the case where a l l three f e e t

F. of the distances r. l i e i n s i d e the t r i a n g l e ' s sides or at i t s v e r t i c e s , and when
t h i s occurs P l i e s inside the c i r c u m c i r c l e , D
A comprehensive computer run showed the theorem
apparently true f o r a l l points i n s i d e the c i r c u m c i r c l e ,
so a l l t h a t remained was to prove the theorem when the
p o i n t P l i e s outside the t r i a n g l e and i n s i d e the c i r -
cumcircle. Moreover, Case 2 e l i m i n a t e s a p o r t i o n of
even t h a t region (when9 say s P l i e s i n s i d e triangle
A 1 A 2 A 3s f o r the o r i g i n a l Erdos i n e q u a l i t y applies to
that t r i a n g l e ) . Let 0 be d i a m e t r i c a l l y opposite ver-
tex Ax on the c i r c u m c i r c l e o f t r i a n g l e AiA 2 A 3 , as
shown i n Figure 5. Without loss of g e n e r a l i t y , we
Figure 5
must prove the theorem whenever P l i e s w i t h i n or on
t r i a n g l e A 2 A 3 D. We may assume t h a t A2 < 90° and A3 < 90c since otherwise the 1ndi-
 - 278 -
cated region is empty". In this regions since r\ is to 6e given a negative sign,
we must prove
j?X + i? 2 + i?3 + 2v\ > 2r2 + 2"3 .

I f Kazarinoff's statement that his proof holds

whenever P l i e s inside the c i r c u m c i r c l e had been
s u b s t a n t i a t e d f then the proof of Problem E 2462
would have been complete a t t h i s p o i n t . The
f o l l o w i n g cases, a l l developed i n the past y e a r ,
do complete the desired proof.
Case 3* Point P l i e s i n t r i a n g l e A2A3D
and at l e a s t one of angles A2 and A3 does not
exceed 30° 8
Referring to Figure 6S l e t A2 ^ 30° 9 so
t h a t s i n A 2 < 1/2. I f /A 3 A 2 P = e 9 then

r3 = i?2 s i n (A 2 +e) 9 r\ = i?2sines

Figure 6
and
sin(A 2 +e) - sine = sin A 2 cose + cos A 2 sine - sine

= sin A 2 cose + sin e(cos A 2 - 1)

< sinA 2 < |.

Now
R2 - PA2 > PA2{2 sin (A 2 +e) - 2 sin e) = 2r 3 - 22^ .
Hence
j?l + F 2 + i?3 + 2ri > r2 + (2r^-2r\) + r2 + 2r\ = 2v2 + 2 r 3 . •

Case 4. Point P l i e s inside the largest angle of the t r i a n g l e .

Let P l i e in triangle A2A3D and suppose Ax > A2 ^ A 3 . Then we have a\ > a2 > a 3
and v\ £ r 2 , and also

a3 a2 a\ a3

Hence, i f for some number N we have u?i = Vr2 + N9 then

N
Uri > Ur2 t
and9 since also
(U-2)ri < (U-2)r2,

we may subtract to get

2ri > 2r 2 + N.
 279
Therefore s since we have 9 by the f i r s t two l i n e s o f (2),

l<*2 ,a3 i fa3 a\. (o.\ a2x

6 K l v z v
a3 a2' a2 a3 a2 <*i' J '
i t f o l lows t h a t

#1 + i?2 + # 3 + 22»i > 22*2 + (™i + -^)p3 > 22*2 + 22*3, •

<-^2 ^ 1
Now only one case remains to be settled, but first we prove a pair of lemmas*
LEMMA l« The function

fix) = 1 - COSx - -Sin (^ - 15 )

has a minimum a t approximately ( 2 9 ° , o . o u M , y - i s i n Ca?-15°)

so fix) > o f o r a l l a: i n the i n t e r v a l
[15°, 90°]. (See Figure 7.)
We have
1 °
fxix) = sina: - - c o s ( x - i 5 ) 9
which vanishes when

tana? = COS 15
y = /(«)
2 - sin15C
.05
t h a t i s , when x « 29.019466 , at which p o i n t
fix) « 0.004419 > o. Since
3C 50 7Q

i " ' ( x ) = cosx + ~ s i n U » i 5 ° ) > o9

Figure 7
the critical point is a minimum. •
LEMMA 2> If 1 < x < 2S then gix) = x + 1/x < 2.5,
%
Clearly g ix) > o in the given interval, so gix) < gi2) = 2,5, •
Our last case takes P inside the triangle A 2 A 3 D of Figure 5, where A 2 > 30°
and A 3 > 30° (by Case 3 ) , and Ax is not the largest angle of the triangle (by Case 4 ) .
We may without loss of generality assume that A 2 is the largest angle. Since we
need not consider A 2 £ 90° (by Figure 5 ) , we have the following case:
Case 5 8 Point P lies inside triangle A 2 A 3 D, and 30° < A 3 < A 2 and A] < A 2 < 90°.
(See Figure 8.)
Let 6 = /0A 3 P. Since A 3 > 30°, we have A2 t A 2 < 150° and A 2 < 75°. So
/A3OA2 < 150° and /OA3A2 "> 15°. From
a« _ sin A 2
and 30° < A 3 < A 2 < 90 ,
a 3 " sin A 3
we get
 - 280 -

Figure 8

£ O 2 < y n 9 0 =

as sin 30
and so9 by Lemma 2S
a a
2 3
— + — < 2.5.
a
^3 2
Now ri z E3 sin (6-15 ), so s by Lemma l,

R3(i - cos <5) > |#3 sin (6-15°) > ir x .

Then
+ 22, + c o s 5 +
R\ * i? 2 + ^ 3 1 = i?l + i?2 ^3 #3 0--COS 5 ) + 2v\

> Ei + i?2 + #3 COS 6 + 2„5PI

> i?! cos (OAiP) + R2 cos (0A2P) + f?3 cos 6 + ( ~ + ~ ) r i

a
^3 2
,a a i , ,ai a .
= ( =32 + ^ i ) r 2 • ( ^ ) r2 3 by ( 2 )
ai a3 ^ a2 ai
> 2 ( p 2 + 2*3 )»

and we are at last finished. The proof of Problem E 2462 is complete.

Finally, we use the extended Erdos-Mordell inequality for triangles to get, as
a corollary 9 a corresponding result for convex quadrilaterals.
Let A1A2A3A^ be a convex quadrilateral, and let P be any point in its plane.
 - 281 -
We set PA. = R. > o and denote
the signed distance between P
and l i n e A.A. by r s the sign
•z- J ij
being determined by Demir s s
convention f o r any t r i a n g l e of
which A.A. i s a s i d e . Thus
(see Figure 9 ) f r 1 2 i s asso-
ciated with triangles kik2k^
and A ^ A i * and has the sane
sign f o r both t r i a n g l e s r e -
gardless of the l o c a t i o n o f
p o i n t P; and s i m i l a r s t a t e -
ments can be made about r 2 3 »
2?3u, and r ^ j . The distance
\r[3\ , on the other hand s i s associated w i t h t r i a n g l e s A1A2A3 and A ^ A ^ ; and i f
P13 is the signed distance associated w i t h t r i a n g l e AiA 2 A 3 § then -P13 i s the signed
distance associated w i t h t r i a n g l e A1A3A4, S i m i l a r l y , i f r2^ corresponds to t r i a n g l e
AiA 2 Ai f , then -v2k corresponds to t r i a n g l e k2k^kh. Our i n e q u a l i t y extended to quad-
r i l a t e r a l s reads as f o l l o w s :
COROLLARY. I f kik2k3\ is a convex q u a d r i l a t e r a l , P i s any p o i n t i n i t s p l a n e ,
and the distances R. and v.. are as defined above s then

3(Ri + R2 + i? 3 + Rk) * ^(ri2 + r23 + 2*3 *t + ***tl ) . ^3^

Proof, We apply the extended Erdos-Mordell i n e q u a l i t y successively to t r i a n g l e s

A 1 A 2 A 3s A1A2ALf9 A ^ A ^ , and k2k3kk:
+
i?l + R2 + i?3 ^ 2 ( r i 2 + 3°23 2?13 ) »

i?! + i? 2 + i?^ ^ 2 O 1 2 + ^2*+ + 3?Ul)»

i?X + i? 3 + i?^ > 2 ( r 3 L , + r i t i - r13),

/? 2 + i?3 + i?i+ > 2 ( r 2 3 + 2*3^ - r 2 t f ) 9

and adding these f o u r i n e q u a l i t i e s y i e l d s (3).

REFERENCE

1, D,K, K a z a r i n o f f , "A Simple Proof of the Erdos-Mordell I n e q u a l i t y f o r T r i -

angles"* Michigan Math. J« , 4- (1957) 97-98,

Mathematics Departments University of Maine, Orono, Maine 04469,

 - 282 -
THE OLYMPIAD CORNER: 59
M.S. KLAMKIN

I present three new problem sets t h i s month. The f i r s t two* which I obtained
through the courtesy o f D i m i t r i s Vathis 9 contain the problems set a t two stages o f
the annual High School Competition of the Greek Mathematical S o c i e t y , The t h i r d
contains a few i n t e r e s t i n g problems from the A p r i l 1984 issue of the Russian j o u r -
nal Kvant. I s o l i c i t from a l l readers elegant solutions to a l l o f these problems.

1984 ANNUAL GREEK HIGH SCHOOL COMPETITION

2nd C l a s s H i g h S c h o o l - March 1 0 , 1984

1, (a) Let A ^ A s A ^ A g be a convex hexagon having i t s opposite sides p a r a l -

nave
lel. Prove t h a t t r i a n g l e s A ^ A s and k2^k^ equal areas.
(b) Consider a convex octagon i n which a l l the angles are equal and the length
of each side is a r a t i o n a l number. Prove t h a t i t s opposite sides are equal and
parallel.

28 An e q u i l a t e r a l t r i a n g l e ABC and an i n t e r i o r p o i n t P are given such t h a t

PA = 59 PB = i+, and PC = 3 . Find the length of a side o f t r i a n g l e ABC.

3, In a given t r i a n g l e ABC, /A = 5TT/8 9 /B = TT/8S and [Z = u/k. Prove t h a t

i t s angle b i s e c t o r CZ5 the median BE$ and the a l t i t u d e AD are concurrent.

i|, (a) Find the real roots of the equation

(x2-x-2)k + Cte+l)1* = (x2+x-l)k .

(b) Find the roots o f the equation

x3 - lax1 + (a2+l)x - 2a + 2 = 0

f o r a l l real a .
(c) Find the range of the f u n c t i o n

J K }
/x^+1 + x + 1

and show t h a t / i s an odd f u n c t i o n .

3rd Class High School - March 10, 1984

1, Prove or disprove that there exists in space a pentagon all of whose sides
are equal and all of whose angles are 90°.
 - 283 -
2 B (a) Find the maximum and the minimum of 8x + 6y - 5 s , where x9y9z are
real and x2 + y2 + z2 - 5.
(b) Express the vector x in terms of the vectors a$ if

a®(x + 5 ) = a2 and t>*x = 0*

I t is assumed that a * o and that the vectors a$9x are coplanar e
38 If G is a multiplicative group and a$b9c are elements of G9 prove:
(a) If b ab = ac, ac = aas and be - cb * then
n, - n n , , , ,n ,n n n(n+l)/2
a 3 = £a e and (o£>) = b a c
for all n e Ne
-1 k
(b) If b ab - a , where £ e tf, then
, - 1 ft, Z n&
b a b - a

f o r a l l Z,n c tf.

Z|,The elements a = ( 1 , 1 , 0 , 0 ) , £ = ( i , o , - i , o ) , a = ( 0 , 0 , 0 , 1 ) , d = ( - 1 , 0 , 1 , 0 ) ,
<g = ( 0 , 1 , 0 , 1 ) , a n d / = ( o 9 o , i , o ) of Rh are given. Let 7 = <a9b,o and
F = <d9e9f> be the subspaces of Rk generated by the vectors a}> 9Q and d9e9f9
respectively,
(a) Prove that the s e t

V + W = {^1+5^2 I xl £
^^ ^2 €
^
is a subspace of tf*.
(b) Find a basis of the space F n w (and denote i t by A).
(c) Find a basis (denoted by B) of the v containing the A, and a basis C of
the fir containing the A as well.
(d) Prove that the set BvC is a basis of the space v+w and t h a t v+w = i?^
(e) Prove t h a t dim (V+W) + dim (VnW) = dim F + dim W,

PROBLEMS FROM KVANT, April 1984

^ 8 5 6 , Proposed by I.Z. Titovich.

(a) Construct a quadrilateral knowing the lengths of i t s sides and
t h a t of the segment joining the midpoints of the diagonals.
(b) Under what conditions does the problem have a solution?

M857 8 Proposed by S. Stadnichenko,

Among the f i r s t 1984 positive integers (from i to 1984) we underline
those which may be represented as the sum of five nonnegative integer powers of 2
 284 -
(i.e., of five not necessarily different numbers 1, 2, 4, 8, . . . ) . Is the set of
underlined numbers larger than that of the nonunderlined ones?

P858 i Proposed by P.B. Gusyatnikov.

The angles a, G, and y of a triangle satisfy the relation

sin 2 a t sin2B = sin y.

(a) Find a?69y if the triangle is isosceles (consider all possible cases).
(b) Can the triangle have only acute angles?
(c) What values can the largest angle of the triangle assume?
F!859 a Proposed by V.P. Pikulin,
Find the least positive number a such that any quadratic trinomial fix)
for which \f(x)\ ^ l whenever o < x < l satisfies !f'(l)i < a.
M860 8 Proposed by Chan Owing.
(a) Let E and 0 be the radius and centre of the circumcircle of triangle
ABCS r and I the radius and centre of its incircle s and K the common point of the
medians of the triangle whose vertices are the points where the incircle touches
the sides of ABC. Prove that I lies on the seament OK and that
1011 |1K| 3i? : r .

(b) Let a,bsa be the lengths of the sides of triangle

ABC, and n ji, sn unit vectors perpendicular to the corre-
a D a
sponding sides of the triangle and directed outward (see
figure). Prove that
a3n + J,3 n + a3^ 125*60,
a b D a

where S is the area of triangle ABC and G is the intersection

point of its medians*
*
I now present solutions to several problems proposed earlier.

J-25 s ri981: 143; 1984: 79] (Corrected) From a list of Russian "Jewish"
problems.
In a convex quadrilateral ABCD S the sides AB and CD are congruent and the mid-
point of diagonals AC and BD are distinct. Prove that the straight line through
these two midpoints makes equal angles with AB and CD.

II. Composite of the (essentially similar) solutions by Chris Fisher (Regina)_,

Lergy F. Meyers (Ohio)s D.J. Bmeenk (The Netherlands)3 Van Sokolowsky (Los Angeles)3
and Esther Szekeres (Australia).
 - 285 -
Solution I by vectors was easy and direct,
but not very elegant. The following synthetic
solution is better.
Let Es G, and F be the respective mid-
points of AC9 BDS and BCf and let line EG in-
tersect AB and CD in H and I, respectively s as
shown in the figure. Then EF || AB9 FG || CD,
and
AB CD
EF = FG.

Therefore /BHE = /FEG = /FGE = /CIG. D

The same proof applies, with very minor modifications, if ABCD is not convex.

8, [1982: 301; 1984: 87] Proposed by John Steinke*

tetermine all six-digit integers n such that n is a perfect square and the
number formed by the last three digit of n exceeds the number formed by the first
three digits of n by l . (n might look like 123124.)
II. Solution by Fviend E, Kierstead^ Jr,s Cuyahoga Fallss Ohio.
There is an old Chinese proverb that when two men a b | N n
are arquing, the one who first resorts to physical vio-
1 143 | 428 183184
lence is the one who first runs out of ideas, Similarly f
11 91 j 727 528529
when a mathematician runs out of ideas he resorts to a
13 77 , 846 715716
computer. Mr. Prielipp's computer took 7 seconds to
77 13 155 24025
solve this problems but i t took a good deal longer to
91 274 75076
write the program and debug 1t. ' I t takes no longer to " •

solve the problem mathematically, and i t gives a lot 143 573

vl
328329

more satisfaction.
Let Nz = n; then N2 - l = loois, where z is a three-digit integer. Let
ab = 1001 = 7«n«i3 and s = xy> Then (N-i)(N+l) = abxy, and we may take N-l = ax
and N+i = by, Then N = ax-tl = by-i9 from which
by = ax + 2.
This equation may be solved for positive integers x and y when a and b are given.
When a - i and b - 143, we obtain

143W 82s y - lw - 4 S whence N = 1001W - 573.

Choosing w so that o < N < 1000 gives N = 428, n - 183184, as shown in the first
line of the table. The rest of the table gives the results corresponding to the
 - 286 -
other possible choices for a and b. If initial zeros are not all owed, we have only
the four solutions n = 183184S 328329, 528529, 715716.
[A similar solution was received from J.T. Groenman, Arnhem9 The Netherlands.]
*
1, C1983; 303] From the 1980 Leningrad High School Olympiad^ Third Round,
If aj?9c9d > o and a-tb+c+d = l, prove that

(Grade 8)
I. Solution by Mark Kantrowitz^ student^ Maimonides School^ BrooklineJ
Massachusetts.
We weaken the hypothesis to a9bfc$d > o and a+b+c+d - l. By the A.M.-G.M.
inequality,

/ ^ I < ^ ^ + 1 = 2a + 1.

with equality if and only if a = o. From this and three similar r e l a t i o n s *

Aa+l + A&+T + /53+T + Ad+l < 2(a+b+c+d) + 4 = 69

and the inequality must be s t r i c t since a$b9csd cannot all be zero.
I I . Solution by M.S.K.
By the power mean inequality
v^to+i t /5F+T + Agj-l + Jtd+i J(^2-hi)+(^-fi)^(Lk?-n)+(4j+i) j -
=
i+ i 4 '
Thus
Az+T + Afc+1 + Aa+1 + A2+1 < 4/2 < 69
with equality if and only if a = b = c = d = 1/4.
2,[1983: 303] From the 1980 Leningrad High School Olympiad^ Third Round.
On the sides AC and BC of a triangle ABC9 points M and Ks respectively 9
are chosen such t h a t
BK«AB = IB2 and AM«AB = IA 2 ,
where I is the incenter of the t r i a n g l e . Prove that the points H9 I , and K are
col l i n e a r . (Grade 8)
Solution by Paul Wagners Chicagos Illinois,
The d i r e c t s i m i l a r i t y of triangles AIB and 1KB follows from the f i r s t part of
the hypothesis, and the d i r e c t s i m i l a r i t y of triangles AIB and AMI follows from
the second p a r t . Therefore
 - 287 -
/MIA + /AIB + /BIK = £TBA - /AIB + /BAI = TT,
and the points Ms I, and K are c o l l i n e a r .

5 , [1983: 304] From the 1980 Leningrad High School Olympics ^cird Round,
A convex quadrilateral is divided by i t s diagonals intc t>ur triangles*
The sum of the squares of the areas of the t r i a n g l e s adjacent tr ^posite sides is
the same. Show that at l e a s t one of the diagonals is bisected a the point of
intersection* (Grade 8)

Solution by Paul Wagner3 Chicago3 Illinois.

Referring to the figure, we have, with square
brackets denoting area,,
A
[AEB] = Jafcsine, [BEC] = i £ e $ i n 6 9
[CED] = Jed s i n e , CDEA] = J d a s i n e .
The hypothesis therefore gives
a2b2 + c2d2 = b2c2 + d2a2s or (a2 - e2)(b2 - d2) = 0
and so a - o or b = d.

8 8 [1983: 304] From the 1980 Leningrad High School Olympian ~*Hrd Bound.
A triangle ABC has sides a,b,c in the usual order. If c':ie A is twice
angle B9 show that a2 = b(b+c). (Grade 9)
Solution by Gali Salvatore3 Perkinss Quebec,
Since A = 2BS we have sin A = 2 s i n B c o s B 9 and so
c2*z2-b2
a = 2b*
2ca
which is equivalent to (a 2 - b(b+c)}(b-e) = 09 If b * c% then c = b(b-bc)« If
b = c, then A = 90° and B = C = 45°, so a2 = 2b2 = b(b-tc) again
The converse s if a2 = b(b+c) then A = 2BS is also t r u e . Se olutions IV and
V to Problem 102 in t h i s journal [1976: 74].

10, [1983: 304] (Corrected) From the 1980 Leningrad High c^ool Olympiad^
Third Round.
Show t h a t 1f, for any value of x in the interval C o . i ] , thf ^ q u a l i t y
I ax2+bx+c\ < 1 is s a t i s f i e d , then | a | + |Z? |+|el < 17. (Grade 9)
[In the original version 17 had been replaced by U
 - 288 -
Solution by Paul Wagner, Chicago, Illinois.
For x - o, i / 2 9 i , we obtain respectively
|c| < 1, |a+22?+4e| < 4S \a+b+c\ < 1.
If we l e t a + 22? + 4c - m and a + b + c = n, then
a = -m + 2n + 2e and £ = m - n - 3c.
Thus
|a | = |-m+2n+2<?| < |m | + 2 |n | + 2 |<? | < 89

\b\ = |n-n-3c?| < |m| + |n I + 3 | c | < 8,

and so \a\ + \b\ + \c\ < 17,
The naximum value 17 is attained for 8a:2 - 8x + l, which satisfies
-i < 8x 2 - 8x + i < 1 when o < x < i.

H , [1983: 304] From £7ie 1980 Leningrad High School Olympiad, Tk>ird Round.
Find two different natural numbers whose arithmetic and geometric means
are two-digit numbers one of which is obtained from the other by interchanging
the d i g i t s . (Grade 9)
Solution by Daniel Ropp, student, Stillman Valley High School, Illinois.
Let a and b be the required natural numbers. Then there are nonzero decimal
d i g i t s c and d such that

^—- = 10c + d and y/ab = lod + o,

and (a+b)/2 > fob implies that c > d. Thus a+b = 2(l0c-kf), ah = (10<i+c)2s and a
and 2? are the roots of
^ 2 - 2(10c+d)x + (lOd'+c)2 = 09
that i s ,
{a, b] = (10c + d ± 3/ll(e 2 -<Z 2 )}. (1)

Now (a-d)ic-d) = i i or 44 since c?2-d2 < 9 2 - i 2 ; and since c+d and c-d have the same
p a r i t y , we must have
(c + d - 11 (c + d = 22
or
Cc - c? = 1 (c - d = 2.
Only the f i r s t system has a solution in d i g i t s a and d: (c9d) = ( 6 , 5 ) , and then
( l ) gives the unique solution {r, b} = (32 9 98}.
 - 289 -
1 2 , C1983: 304] From the 1980 Leningrad High School Olympiad^ Third Round.
We s h a l l c a l l a segment in a convex q u a d r i l a t e r a l a midline if it joins
the midpoints of opposite s i d e s . Show t h a t i f the sum of the midlines o f a
quadrilateral i s equal to i t s semi perimeter, then the q u a d r i l a t e r a l is a paral-
lelogram,, (Grades 9, 10)

Solution by Daniel Ropps studenta Stillman Valley High Schools Illinois.

Let ABCD be a convex q u a d r i l a t e r a l w i t h the stated mid! ine-semi p e r i mater pro-
perty* We use A as o r i g i n o f vectors and use the n o t a t i o n AX = x f o r a l l points X.
The sum of the midlines is

\t f c - d\ + \t * a - t\

and the semi perimeter is

\t\ + |?-5| + \2\ + \cS\

Thus
\b+c~d\ + \d+c-b\ = \b\ + \c~d\ + \d\ + \c-b\. (1)

Nowg by the t r i a n g l e inequality,

-»• -*- -»• - > _ > . _ > .

w i t h e q u a l i t y i f and only i f J = r?(?-?), or AB || CD; and s i m i l a r l y

I?I + \a-b\ * \3+*-S\9

w i t h e q u a l i t y i f and only i f 3 = « ( ? - £ ) , or AD If BC0 Therefore (1) holds j u s t when
ABCD i s a p a r a l l e l o g r a m ,

1 3 , [ 1 9 8 3 : 304] From the 1980 Leningrad High School Olympiad, Third Round,
Are there real numbers a and b such that the function fix) = ax + b
satisfies the inequality
{fix))2 - cosx°fix) < Jsin25?
for all x e [0 S 2TT]? (Grade 10)

Solution by Daniel Ropps student^ Stillman Valley High Schools Illinois,

The inequality of the proposal is equivalent to
tjyf . l+cos^w,,, . f l-cosa^ ^ .
[fix) — ){f(x) + — ^ — - ) < 0S

and t h i s implies

1 - COS a: ^ j?f \ 1 + COS a:

 - 290 -
Setting x - o9 TT, 2TT successively, we obtain
0 < b < 1, ~1 < a-n+b < 0 , 0 < 2cm-+£> < l .

If we sum the first and third of these relations, we get o < 2(air+b) < 2 9 and this
contradicts the second. So there are no real numbers a and b with the desired
property,

15, C1983: 305] From the 1980 Leningrad High School Olympiad^ Third Round,
How many different numbers appear in the sequence
' ±2 ' "22 " 19802j
1980^ • 1980 . .... _1980 j

where the square brackets denote the greatest integer function? (Grade 10)

Solution by Daniel Ropps students Stillman Valley High School» Illinois,

Let Si &nds2 be the subsequences corresponding k = i 9 2,,. es 990 and k = 99i 9 992 9
,.. 3 1980 9 respectively^ of the sequence

S = {[&2/1980] I k = l 9 2 9 8 9 . s 1980}.
Every integer from o to C9902/1980] = 4-95 appears in S\. For if not 9 let n be
a missing integer. Then n < 495 and9 for some k9

k2 „„
< YI < nn+1 (k±±)
< ————
2

1980 *X ~ 1980 •
This implies that (2fe+i)/i980 > l , so k > 990 and n > 495, a contradiction.
No integer appears twice in 5 2 . For suppose m appears twice. Then there is a
positive integer k > 991 such that
^ k2 (k+l)2
m < < m+1
~ 1980 "l980" '

This implies that (2fc+i)/i980 < i and k < 990, a contradiction.

Finally,, observing that [990 2 /i980] < [99i 2 /i980] 9 we conclude that the subse-
quences Si and 5 2 are d i s j o i n t , and therefore the number of different numbers in S
is
1 + 495 + 990 = 1486.

Editor's note. All communications about this column should be sent to pro-
fessor M.S. Klamkin, Department of Mathematics, University of Alberta, Edmonton,
Alberta, Canada T6G 2G1,

THE PUZZLE CORNER

Answer to Puzzle No. 58 [1984; 248]: Epsilon, in slope.
 - 291 -
PROBLEMS — PPOBLEMES
Problem proposals and solutions should be sent to the editor, whose address
appears on the front page of this issue. Proposals should, whenever possible, he
accompanied by a solution, referencesf and other insights which are likely to be of
help to the editor. An asterisk (*) after a number indicates a problem submitted
without a solution*
Original problems are particularly sought, But other interesting problems may
also be acceptable provided they are not too well known and references are given as
to their provenance. Ordinarily, if the originator of a problem can be located, it
should not be submitted by somebody else wizhout his permission.
To facilitate rheir consideration, your solutions, typewritten or neatly hand^
written on signed, separate sheets, should preferably be mailed to the editor before
April 1, 1985, although solutions received after that date will also be considered
until the time when a solution is published.

9818 Proposed by Allan Wm. Johnson Jr«y Washington^ D.C.

Solve the doubly true multiplication

5-ELEVEN = FIFTY5

where FIVE is 9 of course, prime,

982, Proposed by George Tsintsifas3 Thessaloniki3 Greece*

Let P and Q be interior points of triangle k\k2^ • P ° r ^ = i 9 2 s 3 9 let
PA. = x.9 QA. = y., and let the distances from P and 0 to the side opposite A. be
% % is % 1*
p. and q.9 respectively. Prove that

When P = Q, this reduces to the well-known Erdos-Mordell inequality, (See the

a r t i c l e by Clayton W„ Dodge in this journal [1984: 274-281],)

983 8 Proposed by D.J. Smeenks Zalthomr.el3 The Netherlands.

Let A0Alo*„A be an n-simplex in / \
(a) I f m. is the median through A. , prove that

nhn\ = nSk ~ T^

where 5, is the sum of the squares of all the edges meeting in A^, and T^ is the sum
of the squares of all the edges not passing through A-.
(b) Deduce from (a), or otherwise9 that if the medians of the simplex are all
eaual, then the sum of the squares of all the edges meeting in a vertex is the same
for all vertices. Is the converse also true?

984B Proposed by J,C, Fisher and H,N, Guptas University of Begina*

For which k > 3 is ( ) - l a prime power pnl
 - 292 -
985, "Provesec by John J. Martinezs Gonzaga High Schools Washington^ D.C.
Let ABC be a t r i a n a l e with sides a9b9o9 and l e t APSBQ9CR be concurrent
cevians terminatinq in the opposite sides at P9Q9R. We use square brackets to
denote the area of a t r i a n g l e .
(a) If AP,BQ9CR are the internal anale bisectors of the t r i a n g l e , prove that
[PQR1 = CBPR3 if and only if a9b9c9 in some order 9 are in arithmetic progression,
(b) If AP9BQ9CR are the a l t i t u d e s of the triangle and Pfq$? are the sides of
trianale PQR9 Drove that
[PQR] 2pqy
[ABC] ~ aba 8
Q^59 Proposed by Stanley Rabinowitzs Digital Equipment Corp.s Nashua^ New
Fanpskire.
Let
x = Vp + / r + Vq - Jv,

where p,q9r are inteGers and r > o is not a perfect square. If x is r a t i o n a l , prove
that p = q and x is i n t e g r a l .
987, Proposed by Jack Garfunkel3 Flushing3 N.Y,
If t r i a n g l e ABC is acute-angled, prove or disprove that
, * . A . B . C 4,„ . A . B . C,
(a) s i n - - s i n - + s i n - > - { i + sin - s i n - s i n -},
,,s A B C 4r„ . A . B . C,
(b) c o s - + c o s - + c o s - ^ - ^ i i + sin - s i n - s i n - } .

988, Proposed by J.T. Groenman, Arnhems The Netherlands.

Prove that

sec
I TT + I sec
lT = °-
989, Proposed by Kurt Schifflers Schorndorfs Federal Republic of Germany,
Let H be the orthocenter of triangle ABC. Prove that the Euler lines of
triangles ABCS BCH, CAH9 and ABH are all concurrent. In what remarkable point of
triangle ABC do they concur?
990, Proposed by Bob Prielipp3 University of Wisconsin-Oshkosh.
Find all pairs (u,v) of positive integers such that
U3 + (tt+1)3 = I?2.

ft ft ft

THE PUZZLE CORNER

Answer to Puzzle No. 59 [1984: 248]: Isometry (I-some-try),
 - 293 -
S O L U T I O N S

No problem is ever permanently closed. The editor will always be pleased to

consider for publication new solutions or new insights on past problems.

93, C1975: 9 7 ; 1976: 45 9 111] Proposed by E.G. Dworschak, Algonquin College.

Is there a convex polyhedron having exactly seven edges?
IV. Second solution by Charles W. Triggs San Diego, California,
This proposition has been considered by Starke [ i ] , and solutions by Maskell 9
Sauvg, and Trigg have appeared in this journal. An approach differing from these
solutions follows:
If there is a polyhedron having exactly 7 edges, i t s faces cannot all be t r i -
anaular, for each edge of a polyhedron joins two faces, and 7*2/3 is not an integer.
And for every polyhedron having an n-aonal face 9 where n > K the number of edges is
at l e a s t 2n > 8„ For, in addition to the n sides of the n-aonal face 9 the polyhedron
has at l e a s t n additional edges, there being at l e a s t one emanating from each vertex
of the n-gon and terminating at a polyhedral vertex not in the plane of the «~gon„
So there is no polyhedron having exactly 7 edges.

Editorfs comment.
This problem has also appeared in [2] with an outline of solution based on
Euler's formula v-e+f - 2a A Question now naturally a r i s e s : For which n is there
a convex polyhedron having exactly n edges? This problem has also appeared (without
solution) in [2] and in this journal (Problem 121 [1976: 113]) with solutions by
Meyers, Trigg, and Sauve*

REFERENCES

1. E.P. Starke9 "Possible Number of Edqes for a Polyhedron" (Solution of

Problem E 923), American Mathematical Monthly^ 58 (March 195l) 1909
2. Anatole Beck, Michael N. Bleichers and Ebnald WB Crowe, Excursions Into
Mathematics (Experimental Edition), Worth Pub1ishers9 New York, 1967, p. 29.
k k "

346, C1978: 134; 1979: 26] Proposed by Leroy F, Meyers^ The Ohio State Univer-
sity.
It has been conjectured by Erdos that every rational number of the form n/n9
where n is an integer greater than l, can be expressed as the sum of three or fewer
unit fractions (reciprocals of positive integers, also called Egyptian fractions),
not necessarily distinct As a partial verification of the conjecture, show that
at least 23/24 of such numbers have the required expansions.
 - 294 -
I I I . Comment by J.L, Brenner3 Palo Altos California,
The proposal, as c l a r i f i e d in Wilke's solution I [1979: 26-29], in effect asks
for a proof t h a t <S = 1/24 w 0.0417 is an upper bound to the density of integers
n > 3 such that 4/n cannot be expressed as the sum of three d i s t i n c t unit fractions
(Egyptian f r a c t i o n s ) , that i s , such t h a t , for integers xfy5z with i < x < y < z,
the equation
H i l l M,
- = - i- - + - (1)
n x y z
has no solution* In f a c t , Wilke found the b e t t e r upper bound 5 = 1/140 « 0.007i 9
and Maskell (see the e d i t o r ' s comment [1979.- 30]) improved this s t i l l further to
1 9 10

I t is easy to improve even this r e s u l t by extensions of the elementary methods used

by Wilke and Haskell* but to do so would be pointless because i t is already known
that the l e a s t upper bound is 5 = o, that i s § almost every n is representable in
the form ( D . This (and more) was proved in 1970 by Vaughan [4L Moreover, in a
personal communication Koichi Yamamoto stated that he has verified that (1) has a
solution for every n such t h a t 3 < n < i o 8 .
The l i s t of references qiven below augments that given e a r l i e r [1979: 30].
REFERENCES

1. Paul Erdos and Sherman Stein, "Sums of distinct unit fractions", Proc
Amer. Mavk, Soc* , 14 (1963) 126-131, MR 269 71,
2. B.M. Stewart and W,A. Webb, "Sums of fractions with bounded numerators",
Can. J". Math., 18 (1966) 999-1003, MR 33s 7297.
3. William Aa Webb, "On 4/n = 1/x + 1/y + l/s", Proo. Amer. Math, Soc, , 25
(1970) 578-584, MR 419 1639,
4. R.C, Vaughan, "On a problem of Erdos, Straus and Schinzel"* Mathematika,
17 (1970) 193-198. MR 44 9 6600,
58 Li Delang, "On the equation 4/n = l/x + l/y + 1/z", J, of Number Theory,
13 (1981) 485-494,
6, Xun Qian Yang, "A note on 4/n = l/x + l/y + 1/s", Proc. Amer. Math. Soc.t
83 (1982) 496-498,

851, [1983: 177] Proposed by Allan Wm, Johnson Jr.3 Washington^ B.C.
Solve the doubly true decimal alphametic

iwMniiery- iwu * m^ywm

 - 295 -
which means TEN x TEN - TEN = NINETY in Hebrew. [Goyim readers will find it con-
venient to rewrite the alphametic as ABC * ABC - ABC = DECBF.)
This same Hebrew alphametic appeared in the Journal of Recreational Mathematics
15 (1982-1983) 136, proposed by Meir Fader, who asked for a solution in base 7,

Solution by the proposer (revised by the editor).

For typographical convenience we solve the "aoyim" t r a n s l i t e r a t i o n

ABC x ABC - ABC = DECBF, (1)

We note at the outset that A e (l 9 2 9 3}, for DECBF < 98765 implies ABC < 315; and that
(l) implies

100B2 + iooA(2C-l) - 100C + 20B(C-l) + C(C-l) - F = o (mod 103)„ (2)

I t f o l l o w s from ( 2 ) t h a t C(C-i) - F E O (mod 2o) 9 which i s possible only i f

(C,F) € { ( 1 , 0 ) , (3 S 6) S ( 5 9 0 ) 9 ( 7 , 2 ) } ; (3)

and t h a t

B2 + A(2C-D - C + ^ 1 1 + £i^hf = 0 (mod l o ) < (4)

For (C,F) = ( 1 , 0 ) , r e l a t i o n O ) reduces to B2 + A - i = o (mod 1 0 ) . I f A = 3,

there i s no value f o r B; and i f A = 2 , then B = 3 or 7 , but n e i t h e r ABC = 231 nor
271 s a t i s f i e s (l). Proceeding l i k e w i s e w i t h the other possible values o f (C,F) i n
(2)9 we obtain the f o l l o w i n g r e s u l t s :

(C,F) = ( 3 , 6 ) => B e { 0 , 5 } = > no value f o r A,

(C,F) = ( 5 , 0 ) = > B € { l 9 6 } = > no value f o r A,

(C.F) = ( 7 9 2 ) = o j B = 3 = * n o V a l u e f o r A
*
(B = 8 = > A = l .

We f i n d t h a t ABC = 187 s a t i s f i e s ( l ) , and the unique s o l u t i o n i s

187 x 187 - 187 = 34782.

Also solved by CLAYTON W. DODGE, University of Maine at Orono? MEIR FEDER,

Haifa, Israel? RICHARD I, HESSf Rancho Palos Verdes, California? J«AB McCALLUM,
Medicine Hat, Alberta? GLEN E. MILLS, Pensacola Junior College, Florida? KENNETH M.
WILKE, Topeka, Kansas? and ANNELIESE ZIMMERMANN, Bonn, West Germany. Comments were
received from PETER GILBERT and STANLEY RABINOWITZ, Digital Equipment Corp., Nashua,
New Hampshire? and STEVEN KAHAN, Hollis Hills, N„Y.

Editor?s comment,
Rabinowitz stated that, in addition to bases 7 and 10, the Hebrew alphametic
has a unique solution in bases 13 and m , but not in bases 6, 8 f 9, 12, 15, or 16;
and Feder stated Ci] that it has two solutions in base 11. Mills proved that the
 - 296 -
corresponding English alphametic
TEN x TEN - TEN = NINETY (5)
has no solution in base 10 but does have one in base ii s and Feder stated [1] that
the next base in which (5) has a solution is 66.
REFERENCE

l. Solution to Alphametics 1177 and 1178 (proposed by Heir Feder) s Journal of

Recreational Mathematics, 16 (1983-84) 134.
& * *
852 , [1983s 177] Proposed by Jordi dou3 Barcelonas Spain,
Given are three distinct points A S B 5 C on a circle. A point P in the
plane has the property that if the lines PA9PB,PC meet the circle again in A'jB'.C,
respectively, then A S B ! = A ' C
Find the locus of P„
Solution by Jordan Bs Tabovs Sofias Bulgaria,
Let r be the given circle. It is clear that r is part of the required locus,
for P e r implies that A' = B' = Cf = P. We now assume that P t r. Triangles ABP
and ACP are similar to triangles B'A'P and C f A e P 9 respectively^ and so
BA A'B CA A'C*
= =
Pff PC TV
H _ 11 ; therefore
j ^ equivalent
UIC! C I U ! C C to
V j U I V a i C l l b WU

DP AD
PB = AB
PC W (1)

We now consider two cases.

Case 1* AB = AC. Then (l) is satisfied if and only if P e l9 where I is the
line bisecting /BAG. The complete locus in this case is r u Z.
Case 2, AB * AC, Then (l) is satisfied if and only if P e $ 9 where $ is the
circle of Apollonius relative to side BC [i]. This is the circle on diameter MN,
where M and N are the intersections with line BC of the internal and external
bisectors of angle A. The complete locus in this case is r u <£>.
Also solved by J.T. GROENMAN, Arnhem, The Netherlands? GEORGE TSINTSIFAS,
Thessaloniki, Greece,* and the proposer,

REFERENCE

1. Roger A. Johnson, Advanced Euclidean Geometry (Modern Geometry) 9 Dover9

New Yorks i960, p. 295, Theorem 491.
& is ii
 - 297 -
853 , [1983: 178] Proposed by Kenneth S, Will-Cams, Carleton University3 Ottawa,
Let / be a r e a l - v a l u e d f u n c t i o n , defined for a l l x > o9 such t h a t

lim/Gr) and / £^dx (D

X-*®3 Jn x
A)
both e x i s t , the second f o r a l l r > o. Evaluate

x
Jn
7
o
where a > b > o$ and deduce t h e value of
°° ax
ax _ bx
/
x(e +l)(e +1)
(This problem was suggested by Problem A-3 on the 1982 Wil1iam Lowell Putnam
. Mathematical Competition J
I. Solution by Leroy F« Meyerss The Ohio State University.
By the integral property in (i), the integrals

both exist for t > o s and their difference is

f* £^> - fQ>*\te = fat 1&%.

t Jo x
Jbt y

sucn
Let / ( « ) denote t h e l i m i t in ( l ) . Given e > o, t h e r e e x i s t s a real number xQ
that
x > xQ => / ( « ) - e < fix) < f(°°) + £.

If t > XQ/b, then at > bt£Mzi < /so

> XQdy$ and OM±dy < / ££hedy = c/(-)+e).mf.
y y
„- y Jbt Jbt

l\ = liml. = / ( » H n £ . (2)
Since z is arbitrary^ we obtain t-#» v
In particular,, if
1 g e
f(x) = - = . . I9 a- > o s
2(1 + e"*) e* + 1 ^
 - 298 -
then /(«>) = 1/2 and, for every r > o9 the integral in (1) is a proper i n t e g r a l , for
Mm fix)/x = 1/4 by r H 6 p i t a 1 f s r u l e . For this f9 the integral Jx becomes i2. There-
in ^-o
fore
j 2 =/C»).lnf = i-1n|.

I t may be noted that (2) is valid also if b > a > o.

I I . Solution by M.S. Klamkins University of Alberta.

Instead of ( i ) , we assume only t h a t f(°°) exists and t h a t / ! ( x ) is continuous
for x > o. For such a function / , we have immediately, by E l l i o t t ' s extension of
F r u l l a n i ' s Theorem LlL

Tl = (f(oo) . / ( o ) ) - l n | .

In p a r t i c u l a r 9 j j becomes j 2 for fix) = -l/(i+ex)9 and therefore

j 2 = (f (c») -. / ( o ) H n f = (0 + | ) - l n | = § « l n | .

Also solved by W,J, BLUNDON, Memorial University of Newfoundland; RICHARD I.

HESS, Rancho Palos Verdes, California? WALTHER JANOUS, Ursulinengymnasium, Innsbruck,
Austria? and the proposer. One incorrect solution was received.

Editor's comment.
References [2] and [3] were provided by Meyers. In [2], Problems 132 and 134
give (without attribution) Frullani's Theorem and Elliott's extension thereof, respec-
tively. In [3] Elliott's extension is mentioned and misleadingly called Frullani's
Theorem, The most thorough (and accessible) discussion of the Frullani and Elliott
results is probably that in Edwards M ,

REFERENCES

1. B. Williamsons An Elementary Treatise on the Integral Calculus , Longmans,

Green & Co., London, 1941, p. 157.
2. R9 Courant9 Differential and Integral Calculus (translated by E.J, McShane),
Interscience, New York, 1937, Vol. I, p. 561.
3. Bruce C. Berndt, "The Quarterly Reports of S. Ramanujan", American Mathe-
matical Monthly, 90 (1983) 505-516, esp, p, 511,
4. Joseph Edwards, A Treatise on the Integral Calculus, Chelsea, New York,
1955, Vol, II, pp. 337-341, esp, p. 339,
* * ft

854, [1983: 178] Proposed by George Tsintsifas3 Thessaloniki, Greece.

For x.y.z > 0, l e t
 299
^ _ _ _ + ™ _ + —-r-

and
5 = . JH . + _gg. + m ,
(y-Lx)(z+x) (z+y)(x+y) (x+z )(y+z)*
I t is easy to show t h a t A < n < Bs with equality if and only if x = z/ = 2 8
(a) Show that the inequality 4 < 3/4 is "weaker" than 3B > 9/4 in the sense
that

A + 35 > I + i = 3o
When does equality occur?
(b) Show that the inequality M < 3 is "stronger" than 8B > 6 in the sense
that
M^8B<3+6=9e
When does equality occur?
I. Solution by Vedula N9 Murtys Pennsylvania State Universitys Capitol Campus.
We will use the following easily established results:

(y+z )(z+x)(x+y) *
frorr which follows

2 - 8 ( 1 - B ) = 2{£ty-»)2 - $*-*£ - «(*-^ 2 } (1)

and

Observe t h a t 4 < 3/4 £ 5 , mentioned in the proposal, follows from (1) and ( 2 ) .
Proof of (a). This inequality is equivalent to A > 3(i-B) 9 that is* to
yz zx xy 63%/s
(y+z)2 (z-tx)2 (x+2/)2 ~ (y+z)(z+x)(x+y)9
or to
ys(s +x)(x+y) _^ zx(x+y)(y+z ) xi/(y+zHz+x) > ^

If we denote by asb,e the terms on the l e f t side of ( 3 ) , then

abc = x2y2z2(y+z)(z+x)(x+y) > x2y2z2(2v/yz)(2v/zx)(2/xy) = 8x^y3z3

and
a + b + c > zVaEe £ 6xyz»

Thus (3) is establisheds with equality if and only if x = y - z.

 - 300 -
Proof of (b). This inequality is equivalent to (3-4,4) - (2-8(l-B)} > o9 that
i s 9 to
(y-z)2 (z-x)2 (x-y)2 2ix(y-z)2 + y(z-x)2 + z(x-y)2}
(y+z ) 2 TzTxW (x+y)2 (y+z)(z+x)(x+y)
If we multiply this by (y+z )2(z+x)2(x+y)2 > o 9 we obtain an equivalent inequality
f(x,y9z) > o. After some tedious algebra, we find that
f(x9y9z) = (y-z)2 (z-x)2 (x-y)2 > 0a
Thus (4) is estab!ished, with equality if and only if at least two of xsyPz are
equal. (Most of the tedious algebra can be avoided by observing that / is a homo-
geneous and symmetric polynomial of degree 6 in xsysz* and that both / and 3//3a?
vanish when x = y . )

II. Comment by M.S* Klamkin^ University of Alberta*

Let
g(x%y,z9\) = A + X(B-|) - |.

We show that g > o if X > 3 and ^ < o i f X < 2. If X > 3 9 then

A + X(B-|) - | > X + 3(B-~) - I = A + 3B - 3 > 0

by part ( a ) . If X < 2, then

A + X(B-|) - ~ < A + 2(B-|) - | = 4 + 2B - | < 0

by part ( b ) . Either g > o or g < o may occur if 2 < X < 3. For example, if
2 < X < 36i/i80 9 then g(2,i9i9\) > o and #(i 9 2 s 3 9 X) < o.
Putting i t another way, we have shown the following: Of the two inequalities

A + XB > 2.(1 + X) and A + XB < ~(1 + X)9

the first holds for all (x,y,z) if X > 3 S the second holds for all (x,y,z) if X < 2 9
and either one may hold if 2 < X < 3.
Also solved by the COPS of Ottawa? J a T. GROENMAN, Arnhem, The Netherlands? M.S.
KLAMKIN, University of Alberta? VEDULA N. MURTY, Pennsylvania State University,
Capitol Campus (second solution}? and the proposer.
* * *

855 i [1983: 178] Proposed by Christian Friesen3 student. University of New

Brunswick.
Let N be the set of natural numbers. For each n e N9 prove the existence of a
polynomial / (x\ %xl9*.. ,x ) such that the mapping / : N^1 -* N is a bisection.
 - ?01 -
Solutions were received from MICHAEL W. ECKER, Pennsylvania State University,
T
«?orthington Scranton Campus; F.D. HAMMER, Palo Alto, California? WALTHER JANOUSr
Ursulinengymnasium, Innsbruck, Austria? LEROY F. MEYERS, The Ohio State University?
and KENNETH S. WILLIAMS, Carleton University, Ottawa. A joint comment was received
from M.S. KLAMKIN and A. LIU, both from University of Alberta, and GREGG PATRUNO,
student, Princeton University.

Editorfs comment.
This problem has a long history, going back to Cauchy (1821), and there has
been much activity about it and related problems in the more recent literatures as
evidenced by the list of references given below, most of which were found through
information provided by readers.
In response to an anonymous query [l], Lew [2] gave a brief outline of the
history of the problems and he referred to his note r 3] where a much fuller account
is aivens with an abundance of references that augments the list given below. The
information in the next paragraph is taken from [3].
Relative to the polynomial bijections f : / 2 -> N with which we are concerned,
Chowla [4] has displayed one such polynomial for each positive integer nf and Lew
[5] has constructed o(n) essentially distinct such polynomials (i.e., distinct to
within a permutation of the variables}, where e(i) = i,

(c{2\ c ( 3 ) , c(H)9 <?(5), c ( 6 ) , c ( 7 ) , ...) = ( 1 , 39 1 1 , ^ 5 , 197, 9 0 3 , ,,.)

is Seauence lies in Sloane [ 6 ] , and, according to Knuth [-, pp. 2399 534, Ex. 12],
as n -> °°s
o{n) ~ k(3 + 2 / 2 ) " v f 3 / 2 f where k = J / ( 3 / 2 - a ) / ^ « 0.139.
n-1
The constructed polynomials have least and greatest decrees n and 2
Of all our solvers, only Meyers showed how to find a polynomial of least degree
n. He proved that g : iVo -*• N0s where N0 is the set of nonnegative integers and g
is defined by

is a polynomial Injection of degree n a The nontrivial part of the proof is equi-

valent to a well-known problem which can be found in Comtet [ 8 ] , Knuth [ 7 , pp0 729
^88, Ex. 56] 9 and (with solution) in Liu [ 9 ] , Meyers then concluded that fn*. N* + N
is a polynomial bijection of degree n% where

For n - 1 and n = 2S (1) becomes

/ i ( a ? i ) = xi (2)
 - 302 -
and
-* <* * ^ - * + (^1+^2-1)(^1+^2-2) m
ftix\,x1) - x\ + —j -, \*>

respectively. Formula (3), which Lew [3] credits to Cantor (1878), is given in
the more recent references Barnard & Child [10], Mathews [ i l ] 9 and P61ya-Szego [12].
It was conjectured in 1923 by Fueter and Pdlya (reference 6 in [3]) that (3) is
essentially the only polynomial bisection N2 + N. It is known that (3) is essen-
tially the only quadratic polynomial with this property and that, if there is an-
other such polynomial, i t must be at least of fifth degree [2].
We now consider polynomial bisections f .- if1 + N of highest degree 2n~1. For
n = l and n = 2, such polynomials are qiven by (2) and (3), respectively. For
n > 2, Meyers (and all other solvers) argued that the required polynomials can be
defined recursively by
f (Xi 9X? > . . . ,a? ) = f 2(/ * (X\ 9Xo , . . , fX A ) . X ) ,

For if / _ : Nn" -> N is a polynomial bi jection of degree 2n~ 9 then / : nP -* N is

n n
n-l
a polynomial bijection of degree 2
Now a look at some related problems.
Meyers showed how to find a bi jection fm: N -* N9 where N is the set of all
finite sequences (tuples) of positive integers, but he proved that such a bijection
cannot be a polynomial (of finite degree).
Hammer [13] asked if there is a polynomial in two variables with integral coef-
ficients which is a bijection from Z2 onto Z, where Z is the set of all integers,
and if so, how many there are. His questions remain unanswered to this day,
Finally, Lew [3] posed the following hybrid problem: to find a polynomial
/(a?i,...,a; ) which maps f1 surjectively onto N, He writes that this is clearly
impossible when n = l , and i t is always possible when n > 3, by the famous theorem
of Legendre and Gauss: each positive integer is the sum of three triangular numbers.
Hence the two-dimensional case is the sole undecided one*
REFERENCES

1. Anonymous9 "A Problem", The Mathematical Intelligencer* Vol. 5 (1983), No, l ,

p. 63.
2. John S. Lew, Letter to the Editor, ibid., Vol. 5 (1983), No. 3, p. 8,
3
' > "Polynomials in two variables taking distinct integer values at
l a t t i c e - p o i n t s " , Amer. Math. Monthly , 88 (1981) 344-346,
4. Paromita Chowla, "On some polynomials which represent every natural number
exactly once", Norske Vid. Selsk^ Forh. Trondheim, 34 (1961) 8-9, MR 30, 4708.
 - 303 -
5. J.S. Lew, "Polynomial enumeration of multidimensional lattices", Math.
Systems Theory, 12 (1979) 253-270.
6, N.J.A. Sloane, A Handbook of Integer Sequences, Academic Press,, New York,
19^3, p. 107.
78 Donald E. Knuth, The Art of Computer Programming§ Vol. 1; Fundamental
Algorithms, Second Edition, Addison-Wesley, Reading, 1973,
8. Louis Comtet, Advanced Combinatories, D6 Reidel Pub, Co., Dordrecht and
Boston, 1974g p. 75, Exe 10.
9. C»L 8 Liu 9 Introduction vo Combinatorial Mathematics , McGraw-Hill, New York,
1968, p. 55, Problem 2-27,
10. S, Barnard & J.M, Child, Advanced Algebra^ Macmillan, London, 1939S p» 373,
Ex. 109,
11. G.B* Mathews, Theory of Numbers3 Chelsea, New York, 1961, p. 318, Ex. 18,
12. G„ Pdlya and Gs Szego, Problems and Theorems in Analysis, Springer-Verlag,
New York, 1976, Vol. II, pps 151-152, 355, Problem 243,
13. F.D. Hammer, (proposer of) Problem 6028% Amer. Math, Monthly, 82 (1975)
410.
* * a
856, [1983: 179] Proposed by Jack Garfunkels Flushing^ N.Y.
For a triangle ABC with circumradius R and inradius r, let M - (R~2r)/2R,
£n Inequality P > Q involving elements of triangle ABC will be called strong or weak,
respectively, according as
P - Q<M or P - Q > M,
(a) Prove t h a t the f o l l o w i n g i n e q u a l i t y is s t r o n g :

• oA • oB • oC 3
sin2j + smzj + s i n ^ ^ j .

(b) Prove t h a t the f o l l o w i n g i n e q u a l i t y i s weak:

cos2- + cos2| + cos

2 * s1n B s1n c + s1n c s1n A + s1 n A s i n B
" °

Solution by W.J. Blundon3 Memorial University of Newfoundland.

We will use the following well-known r e l a t i o n s , where all sums are cyclic over
A,B9C:

R - 2P * 0 9 Z s i n ^ l = 2S.Z2L. Z C 0 S 2A = ^ r s (1) ,(2)_(3)

and s 2 < 4i?2 + 4i?p + 3r 2 (with e q u a l i t y j u s t when R - 2 r ) , from which f o l l o w s

 - 304 -

SsinBsinC = s - L ^ f - ± j £ s l*-±_£ii. M

For each of the opposed inequalities, i t w i l l be seen that equality holds just
when R = 2PS that i s , just when the triangle is equilateral.
(a) From ( i and ( 2 ) 9

n re^2A 3
2ff-r 3 R-2r ^ R-2r t.

and the given ineauality is strong.

(b) From f : , ( 3 ) 9 and ( 4 ) ,

P - ^ - M = I c o s 2 | - z s i n B sin C - M

4J?
> +P (fl+r) 2 _ R-2v _ G?-2r)G?+r)
U
~ 2R " i?2 2/? 2i?2 ~ '

and the given inequality is weak.

Editor*s comment*
The proposer suggested that readers may find i t interesting to pursue the idea
of classifying known triangle inequalities as strong or weak, using the M given
above or some other measure.
Also solved by CURTIS COOPER, Central Missouri State University at Warrensburg?
the COPS of Ottawa? J.T. GROENMAN, Arnhem, The Netherlands? WALTHER JANOUS, Ursu-
linengymnasium, Innsbruck, Austria? M.S. KLAMKIN, University of Alberta? VEDULA N.
MURTY, Pennsylvania State University, Capitol Campus? BOB PRIELIPP, University of
Wisconsin-Oshkosh? KESIRAJU SATYANARAYANA. Gagan Mahal Colony, Hyderabad, India?
GEORGE TSINTSIFAS, Thessaloniki, Greece? and the proposer.
* * &

857, [1983: 179] Proposed by Leroy F, Meyers3 The Ohio State University,
(a) Given three positive integers, show how to determine algebraically
(rather than by a search) the row ( i f any) of Pascal's triangle in which these
integers occur as consecutive entries,
(b) Given two positive integers, can one similarly determine the row ( i f any)
in which they occur as consecutive entries?
(c)* The positive integer k occurs in the row of Pascal's triangle beginning
with i , k, . . . . For which integers is this the only row in which i t occurs?

Solution by the proposer.

(a) Given three positive integers afi%e, we find necessary conditions for the
existence of a positive integer n such that, for some integer k satisfying i < k < n,
we have
 - 305 -

From the properties of binomial coefficients, ( i ) implies that

a(n-k+l) = bk and b(n-k) = <?(fc+l). (2)

2
Our first necessary condition is b * ca5 for otherwise (2) implies that n - -l.
This condition being satisfied9 we solve equations (2) for n and k9 obtaining
be + 2ca + ab v _ a(b + a)
T-y- —9 k = ,M —-. (3)
ca

So necessary conditions for n to be an answer to our problem are that b2 * oa and

that n and k9 as given by (3), be integers satisfying 1 < k < n.
These conditions, howevers are not sufficients for (2) was obtained by taking
onlv ratios of the binomial coefficients (l). For examples if a = 1 and b - o - 2 S
then (3) yields n = 5 and k - 2$but(;j) = 1 0 ^ 2 , When the necessary conditions are
satisfied9 only one more calculation is needed to determine if the value found for
n is satisfactory: n and the value found for k must satisfy any one of equations (l),
(b) Given two positive integers a and bs we find necessary conditions for the
existence of a positive integer n such that, for some integer k satisfying l < k < n,
we have

(kn_±) = a and (£) = b. W

Two simple cases can be settled instantly: i f a = i , then k = 1 and n = b is satis-
factory; and i f b = i s then k = n and n = a is satisfactory. For arbitrary a and b9
we obtain a(n-k+i) - bk as in ( 2 ) , and sos i f d = gcd (a,b)9 a = a/d, and 8 = b/<f9
we have
a(n-k+l) = B*. (5)

Since gcd (a,3) = l , we must therefore have

k = at . (6)

for some positive integer £, and then, from (5),

n = (a+B)t - 1, (7)

Observe that, for any positive integer t9 (6) and (7) ensure that i < k < n. But
n and fc must also satisfy either one (and hence both) of equations (*o, Since

the positive integer t must therefore satisfy

( a
( ^ * ) = a + £ . (8)
 - 306 -
Our necessary conditions are now a l l i n place: n must be o f the form (7) f o r some
positive integer t sa ti sfyi n a (8).
As i n p a r t ( a ) , these conditions are not s u f f i c i e n t . Having found a p o s i t i v e
integer t s a t i s f y i n g ( 8 ) , we must s t i l l v e r i f y i f the corresponding n and k, as
given by (7) and ( 6 ) , s a t i s f y one o f equations ( 4 ) . F o r t u n a t e l y , only f i n i t e l y
many values o f t need be tested i n (8) because

,(a+B)(£+l)x _ ,(a+B)*w £ (a+TOw £ (a+B)t+a+jx /(a+B)t%

1
a(t+l) j = l at K . ? at+j K ._, Bt+j }
> ( at ]

shows t h a t the f u n c t i o n on the l e f t side o f (8) increases s t r i c t l y w i t h t .

( c ) * I have no idea. Of course 9 a prime number can occur i n only one row.

Also s o l v e d by FRIEND H. KIERSTEAD, J R . , Cuyahoga F a l l s , Ohio.

Editorfs comment,
Kierstead's terse answer t o p a r t ( c ) * was the two-word statement "Almost a l l " .
We await a more d e t a i l e d answer.
.•« A A

858, C1983: 179] Proposed by J9T9 Groenman, Amhem3 The Netherlands*

Let ABC be a t r i a n g l e w i t h sides a9b9c. For n = 0 , 1 , 2 , . , . , l e t P be a
p o i n t i n the plane whose distances d 9d-9d from sides a,b9c satisfy

d , i . d ._ J_ : JL : i.
a b c n «YI n
a b a

(a) A point P beinq given9 show how to construct P „.

(b) Using (a), or otherwise, show how to construct the point P for an arbi-
trary given value of n,

Solution by Roland H. Eddys Memorial University of Newfoundland*

Denote by ^(P) and t(P) the isogonal and isotomic conjugates, respectively, of
a point P in the plane of the given triangle ABC. If (a, 3, y) are the trilinear
coordinates of P with respect to ABC, then [l, p. 159]

g(P) = (i, |. i) and t(P) = C4". zk> T-)-

a a
a 3 Y ^ B ezy
Consequently,

*^CP) = £ . £ . £ ) • (i)
(a) It follows from (l) that P _ = t#(P }. Consequently, P„ being given, we
n~rZ n n
obtain P by c o n s t r u c t i n g the isotomic conjugate o f the isogonal conjugate o f P ,
 - 307 -
(b) For n = o and n - i , we have

P0 = ( l , l , l ) = I 9 the incentre of the triangle

and
Pi = (~s | 9 h = G9 the centroid of the triangle,

and these points are easily constructed. For n > l , we can use the method of part .
(a) and construct successively (and laboriously)

Po* ?2$ Pif 9 ...» P^t i f n is even,

or
P i , P3* Ps, . . . . Pns i f n is odd.

A much more practical method, however, is to calculate the actual distances d $d,9d

of P froma9b9e usina the relationships

rij TJIJ rij 2K
ad 'a - D a-r
b = c aa = 1-n x.l-n 1-n9
a + D + c
where z denotes the area of triangle ABC [i, p, 157]. Any two of these distances
uniquely determines P .
Also solved by KESIRAJU SATYANARAYANA, Gagan Mahal Colony, Hyderabad, India?
and the proposer,

REFERENCE

i. DeM0Y„ Summerville? Analytical Conies* G, Bell and Sons, London, 1961,

859, C1983: 179] Proposed by Vedula N« Murtys Pennsylvania State University^

Capitol Campus.
Let ABC be an acute-angled triangle of type II, that is (see [1982; 64]), such
that A < B < TT/3 < C 9 with circumradius R and inradius r. It is known [1982: 66]
that for such a triangles? > J, where^= r/R% Prove the stronger inequality

> /3 - 1

Solution by the proposer.

We show that the desired result holds even if angle C = 90°, Let s be the semi-
perimeter of the triangle and y = s/j?. The following identity of Bager (p. 10 of
reference 3 in [1982: 68]) is valid for every triangle:

4 cos A cos B cos C = s2* •" - ( |£g2l = ^ - (2+x)2.

 - 308 -
Since our triangle is nonobtuse* we have y2 - (2+rc)2 > o s and so

y > 2 + x. (1)
The result
/3(l+ar) > y (2)

was established in this journal [1982: 64] for type II triangles. Final ly 9 from
(1) and (2) we obtain /3(i+#) > 2+x9 from which follows
/3 - 1 1
x>
Also solved by J«T. GROENMAN, Arnhem, The Netherlands; WALTHER JANOUS, Ursu-
linengymnasium, Innsbruck, Austria? M.S. KLAMKIN, University of Alberta; KESIRAJU
SATYANARAYANA, Gagan Mahal Colony, Hyderabad, India; and GEORGE TSINTSIFAS, Thes-
saloniki, Greece.

860i C1983; 180] Proposed by Anders L'6nnberg3 Mockfjards Sweden.

The sequence is }n=i is defined by

m-1 n-(m-l)
m=l m ,
so that
x
si = i1 = l, s2 = l^-2
= i2_ol = -1, s3 = l 3 -2 2 +3 ! o, Si, -2;J+3Z-41 = -2,

etc. Does e - o ever occur again for some n > 3?

Comment by Stanley Rabinowitz3 Digital Equipment Corp., Nashua, New Hampshire,

The following results suggest that \ s | > Is | for n > 5. If this is true,
then s = o will never occur aaain for any n > 3,
n
^5 = 1 s l 9 = -478198544
S
S = 5 £?20 = -1994889946
[Oops I (Editor)]
s7 = 20 821 = -1669470783

$8
= 28 s22 = 56929813933
s9 = -47 s23 = 615188040196
s = -525 s24 = 3794477505572
10
= -2056 525 = 12028579019537
*11
*12 = -3902 5 2 6 = -50780206473221
s
13 = 9633 s 2 7 = -1172949397924184
s
14 = 129033 s 2 8 = -10766410530764118
s
15 = 664364 S 2 9 = -61183127006113951
= 1837904 s30 = -102718668475675151
*16
sl7 = -2388687 331 = 2573781218763700380
5
18 = -67004697 S32 = 40137777624890418072