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.
* ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft
ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft ft

ISSN 0705 - 0348

CRUX I ^ A T H E M A T I C 0 R U f*

Vol. 7 9 No. 9
November 1981
Sponsored by
Carleton-Ottawa Mathematics Association Mathematique d'Ottawa-Caxleton
Publie par le Colleqe Algonquin
The assistance of the publisher and the support of the Canadian Mathematical Olympiad
Committee, the Carle ton University Mathematics Department, the Ottawa Valley Education liaison
Council, and the University of Ottawa Mathematics Department are gratefully acknowledged.
ft * ft ft * * ft Aftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftft•«ftft
ftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftft* ••• * *

CRUX MATHEMATICORUM is a problem-solving journal at the senior secondary and university

undergraduate levels for those who practise or teach mathematics. 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). The yearly subscription rate fcr ten
issues is $12.00. Back issues: $1.20 each. Bound volumes with index; Vols. 1&2 (combined),
$12.00; Vols. 3-6? $12.00 each. Cheques and money orders, payable to CRUX MATHEMATICCPUM
(in US funds from outside Canada), should be sent to the managing editor.
All communications about the content of the magazine (articles, problems, solutions, etc.)
should be sent to the editor. All changes of address and inquiries about subscriptions and
back issues should be sent to the managing editor.
Editor: Leo Sauve, Architecture Department, Algczqvii*. College, 2S1 Echo Drive, Ottv ,
Ontario, KlS 1N3.
Managing Editor: F.G.B. Haskell, Mathematics Department, Algonquin College, 200 Lees Ave.,
Ottawa, Ontario, KlS 0C5.
Typist-compositor: Joanne Rossignol
ft * *

CONTENTS

Pandiagonal Magic Square Equations . . . . Allan Wm. Johnson Jr. 258

The Olympiad Corner: 29 . . . . M.S. Klamkin 267
Postcript to "Three More Proofs of Routh's Theorem" M.S. Klamkin 273
27I+
Problems - ProblSmes
276
Mama-Thematics .• • •
Solutions . . - 277
29
The Puzzle Corner - °

- 257 -
 - 258 -
PANDIAGQNAl MAGIC SQUARE EQUATIONS
ALLAN WM. JOHNSON JR.

A magic square of order n is an mn block of numbers whose rows, columns,

and main diagonals all add up to the same constant, called the magic sum. If
we represent the numbers in a magic square by the algebraic symbols ASB,C3...
and use the symbol z for the magic sum, then A+B+c+D - zs E+F+G+H = z, etc., are
the algebraic equations for the magic square

A B C V !
E F G E
(l)
I J K L

M N 0 P

A magic square equation can be represented as a square array consisting of

the coefficients of the symbols in the equation. For the magic square (l), the
equation A+B+c+V = l becomes the coefficient array

1 1 1 1

M l

Here an extra block is attached to the lower right-hand corner to hold the coef-
ficient of the magic sum.
This coefficient array notation is introduced not only because it is yery
compact but also because it allows magic square equations to be easily manipulated
and provides a convenient mechanism to derive some remarkable relationships among
the numbers in a magic square. For example, the square equation at the top of the
next page constitutes a quick proof that the four numbers in the center of a
fourth-order magic square add up to the magic sum.
Pandiagonal magic squares add up to the magic sum along the broken diagonals
as well as the main diagonals. When broken diagonal equations are included in the
coefficient array, it is not always clear which diagonal is meant, so broken dia-
 259 -

1
\ 1
\ 1 1 -1-1-1-1
1 1 1 1 2 2
— — j — — +
L ——i—4 +
1 1 1 1 2 2
1 1 2 1 1 2 -1 -1-1-1-2 2

gonals are indicated with circled numbers and arrows, as in the following proof
that if the magic square (l) is pandiagonal, then GiJ+2K+L+0 = (3/2)z:

K X 6 6
X 8 4

6 6 4 4 I I
6 6 8

6 6 12 H 4 8 | 4 8

(2)

-2 -2 -8 -?4 -l* -4

~2 -M -2 -4
+
-4
=
8

-2 -2 -4 8 16 8

-4

d
~2 -2 -4 -4 -8 8 12 |
1
The last square in (2) contains four numbers in a diamond-like configuration.
It turns out that every even-order pandiagonal magic square has an equation whose
coefficients are arranged in the shape of a diamond. For a pandiagonal magic
square of order 2n9 this diamond equation is formed as follows:
(a) Number the rows in the (n+i)-st column from the bottom up as follows:

1, 2, 3, ..., rc-1, n, rc-1, ..., 3, 2, 1.

(b) Number the columns in the (w-ii)-st row from right to left as follows;
1, 2, 3, ..., rc-1, n9 n-l9 ..., 3 S 2 9 1.
(c) Between numbers that are equal, copy the number along the broken dia-
gonal that connects the equal numbers.
The coefficient of the magic sum for this diamond is (2n2+l)/6. For example,
the diamond equations for pandiagonal magic squares of orders 6 and 8 are
 - 260 -

1 2 1

1 1 2 3 2 1

1 2 1 1 2 3 4 3 2 1

1 2 3 2 1 1 2 3 2 1

1 2 1 1 2 1

1 19/6 1 11/2

Before proving the existence of this diamond for all even-order pandiagonal
magic squares, we develop some interesting consequences of the diamond.
THEOREM 1. 2 divides the magic sum of every pandiagonal magic square of
order 2n that is composed of integers.
Proof. The diamond represents a linear combination of magic square cells
adding up to (2n2+l)/6 magic sums. The linear combination is an integer because
the cells are integers. Hence (2n2+l)z/s must be an integer, which is possible
only if 21E.
THEOREM 2. 6 divides the magic sum of every pandiagonal magic square of
order 6k that is composed of integers.
Proof. As in the proof of Theorem 1, (2n2+l)E/6 must be an integer. If
n = 3fc-l or n = 3&+1, then 3|(2n 2 +l), but 3j(2n 2 +l) when n = 3k.
THEOREM 3. Removing a main diagonal from a pandiagonal magic square of order
2n leaves behind two triangles of numbers, each of which sums up to (2n-l)/2 magic
sums.
Proof. We prove this theorem for order 6 by a method that generalizes to any
even order, A pandiagonal magic square keeps its pandiagonal property when a col-
umn of the magic square is relocated from the right side to the left side, an
operation equivalent to moving a column of the coefficient array from the left
side to the right side. Move a column of the diamond's coefficient array and
subtract the result from the diamond. The result is shown at the top of the
next page.
Rotating a pandiagonal magic square 90° counterclockwise is equivalent to
rotating the coefficient array 90° clockwise. Move half the columns of (3) from
 - 261 -

1 i

1 2 1 i 2 1

1 2 3 2 1 1 2 3 2 1

1 2 1 i 2 1

1 19/6 1 19/6

-1 1

~i -1 1 1
(3)
-i -i »1 1 1 1

-l -1 1 1
the right side to the left side, then rotate
90° and add: -1 1 0

1 I i 1 1

-1 i i 1

l I 1 I I I -1 J -l i

1 1 1 -l-i-l -i

—t 1—\- -i -i -1
i i -i -1
-1 -i -i -1 -1
i -i

1 i i 1 1

1 i i 1 -1

1 i i -1 -1

1 i -1 ~1 -1

! i -i -1 -1 -1

-i -i -1 -1 -1
LI
 - 262 -
This shows that the two triangles of numbers have equal sums. Because the two
triangles together add up to 2n-l magic sums, it follows that each triangle is
composed of numbers summing up to (2n-l)/2 magic sums. D
A construction method is employed to prove that a diamond equation exists
for all pandiagonal magic squares of even order. For one of order 2n9 we select-
ively multiply the 2n row equations and the 2rc column equations by the algebraic
unknowns FQJ RU F7> •••* R anc* suni the resulting equations to form a coefficient
array, to which are added the 2n northeast/southwest (NE/SW) diagonal equations
and the 2n northwest/southeast (NW/SE) diagonal equations after they have been
selectively multiplied by the algebraic unknowns F 0 J V\9 vls ...., V . The se-
lective multiplication produces a coefficient array with the following appearance:
(a) VQ occurs in the NE/SW and NW/SE broken diagonals starting at row 1/
column (n+l).
(b) v occurs in the NW/SE main diagonal and in the NE/SW broken diagonal
starting at row 1/column l.
(c) For i = 2,3,.. .,n, V ±_. occurs in the NE/SW and NW/SE broken diagonals
starting at row 1/column i and at row l/column (2n+2-t).
(d) F occurs in the top row (row ij and in the leftmost column (column l ) .
(e) /?o occurs in row (n+l) and in column (n+l).
(f) For i = 2,3 n, F . . occurs in row i9 in row (2n+2~i)s in column i9
•* n+1-^ '
and in column (2n+2--£).
The resulting order-8 coefficient array is shown at the top of the next page.
The coefficient array thus constructed has symmetrical properties facilitating
the writing of algebraic equations in the 2n+2 unknowns VQS V\, ..., v 9 F0, j?^ ...^
F s whose values can then be determined in a way that turns the coefficient array
into the diamond equation. The (2n-i)x(2n-l) square in the lower right-hand corner
of the 2rt*2n coefficient array has cells that are symmetrical with respect to its
center column, to its center row, and to each of its main diagonals. The (n+l)*(n+l)
square in the upper left-hand corner of the 2n*2n coefficient array has cells that
are symmetrical with respect to its NW/SE main diagonal. When the cell in the upper
left-hand corner is eliminated, the cells in the reduced top row (of the 2n*2n
coefficient array) are symmetrical with respect to its center cell, and the cells
in the reduced leftmost column are also symmetrical with respect to its center
cell. These symmetries mean that we know what is in every cell of the 2n*2n
coefficient array when we know what is in the cells located on or above the NW/SE
main diagonal of the (n+l)x(n-fl) square in the upper left-hand corner of the
2nx2n coefficient array.
 263
fi)
^
1
^ 3
^ JS. A A A A^ 1
RxRh RXRk
V2V2
R2Ri^
F3F3
Ri+Rk 1 #3^ RQRU *3*<f j

V3V3 V}V3 V0V2 V0V2 V2Vh

— i? 3
R3Rk *3*3 R2R3 1 #1#3 RQR3 RXR3 1
#2#3 #3*3

1 v2v2 1 vxv3 V0Vk VXV3 V2V2 VXV3 V0Vh ^ 3 1

R2Ri> R2R3 R2R2 RXR2 1 #0#2 RXR2 #2#2 i?2i?3 j

1 V\V\ V0V2 V\V3 V2Vh ^ 3 V2Vh VXV3 VQV2 I

R\Rh RlR3 RXR2 R1R1 R1R1 RlRl #1#3 j
(4)
v0v0 ^ 1 V2V2 V3V3 VhVh ^ 3 V2V2 vxvx j
RORM R0R3 R0R2 RGRl RQRO R0RX R0R2 #0*3 j
1 V\V\ ! VQV2 V\V3 j v2vh v3v3 V2Vk I ^ 3 | y0F2 j
\R\Rh | R1R3 ; RXR2 RXRX RXR2 | #1*3 j
^ 2 ^ 3 V\V3 1 V2V2 ^ 3 V0Vk ^ 3 1
R2Rk RiR$ R2R2 RXR2 R0R2 i?ji?2 1 R0R7 R2R3 j

V3V3 v2vh VXV3 VQV2 V0V2 vxv3 F 2 ^tf

#3^ J #3*3 R2R3 j RJR3 | #0*3 j #1#3 R2R3 1 R3R3 J
. .1
t t t
*3 /?, #0 *1 R2 *q

To treat the cells located on or above

the NW/SE main diagonal of the (n+l)x(n+l) A /o

square„ we first separate these cells into 1/3

three groups: 2
(a) Group 1 contains the cells in the
top two rows.
(b) Group 2 contains the cells that are
on or to the left of the NE/SW main diagonal.
(c) Group 3 contains the cells located
to the right of the NE/SW main diagonal.
The diagram on the right shows the three
cell groups for the order-8 coefficient array.

For a coefficient array like (M) to eo,ual the diamond equation, the following
equations for the cells in the top row of group 1 must be true:
V + V + R + R = 0, m = 0,1,... ,n. (5)
m m m n

For the second row of cells in group 1, the equations are

 - 264 -
+ +
Vl ^+l *m+Vl = 0
* "*1.2,....n-l (6)
and
Pi + F"i + i?o + R x = 5? * 0 . (7)

To derive values of V\>Vzs**.9V SRQ,R\***.,R in terms of x and V 0 , we subtract

corresponding equations of (6) from ( 5 ) , thereby obtaining
(V
m"Vm 1 } - (V
m+l~
V
J + (
V*„ 1} =
°' W =
*.2....,n-l. (8)
in nt~i tn+± m n n-±
Adding up these equations gives
iVl.VQ) . (Vn-Vn^) + (n-lHR^B^) - 0. (9)

Subtracting equation (5) a t w = n-l from equation (5) at m = n yields

V - V . = -10? - R , ) . (10)
n n-1 n n-1
Subtracting equation (7) from equation (5) at m = 0 gives

7x - F0 = fe + HRn - V l } ' (11)

On substituting equations d o ) and (11) into ( 9 ) , we obtain

R - E * = -£ • (12)
n n-1 2n
Substituting (12) into (11) yields

Substituting equation (12) into (8) and then solving (8) for 7 23 P r 3 J ... J 7 give

Vm = V, + M ^ f r , OT s 0,1,. ...n. (13)

Equation (5) at m = n shows that R = -V and so, by (13),

n n

Finally, we obtain R by substituting equations (13) and (l«+) into (5):

m
v - T/ .. {^2~2y77(2n-m)}x m -n 1 *. n^
=
#OT ~ M ) + tf^ •» m 0,1,...,n. (15)
Having determined values for V and R , we turn to cells of group 2 and
m m
group 3 to see what their values are.
The NE/SW broken diagonals starting at row l/column (n+i-i) separate the
cells of group 2 into subsets
V + V R + R
i U2k+ Uk n-k> * = <M.•••.*; * =0,1,2.....
 - 265 -
Substituting equations (13) and (15) shows that this expression is identically
zero.
The NE/SW broken diagonals starting at row (t+l)/column ( m l ) separate the
cells of group 3 into subsets

D
i = Vi + V
U2k + R
k + R
n-i-k> t--1'2 " ; * = 0.1.2.....

Substituting equations (13) and (15) shows that V. = lx.

This completes the proof that a diamond equation exists for pandiagonal magic
squares of all even orders. To determine how many magic sums the diamond equation
represents, we add up the multipliers that were selectively applied to the magic
square's diagonal equations before they were summed to obtain the diamond equation,

2V0 + Wi+v2+...+vn_±) + 2Vn = 4nF0 + ( 2 n " 1 ) e 2 n " > "— ;

and we also add up the multipliers used on the magic square's row and column
equations,

Hence the diamond equation equals

(2n2+l)x 1c sumSi (16)
6

The values of V and R computed by equations (13) and (15) contain fractions.
m xn
To obtain integral values for v and R , we set x = 4n and, to further simplify the
3
m m
resulting equations, we put v0 = o. This reduces (13), (15), and (16) to
V = m(2n-m)9
771
m = 0,1,. ..,«
R - n2 - 2m(2n-m),
m
mKZ1 +1
^ > magic sums,

which, for n = 2, yields 12 magic sums as in (2),

Does a diamond-like equation exist for all pandiagonal magic squares of odd
order? The answer is yes, and the construction is similar to that for squares of
even order. We illustrate the construction for pandiagonal magic squares of order
7 and then give the equations for v and R , the verification of which we leave
* m m
to the reader. (The reader should now t u r n to the diagram a t the top of the next
page.)
For pandiagonal magic squares of order 2 n + l , the diamond equals n(n+l)x/e
magic sums and, f o r m = o , l , . . . , « . V and i? are given by the equations
m m
 266
® @ ® ® ® ® ® ®
\ A A X> sS S>
1
s> ' .
V2V3 ViV2 i ^ 1 vxv2 V2V3
R2R3 R2R3 •*3
#3^3 #1*3 #0#3 #3^3

1 V\V2 ^ 3 F0^2 vxvx v0v2 V0V3 VXV2

•i?2
1 *2#3 i? 2 i?2 /?l/? 2 R0R2 RXR2 R2R2 R2R3

V0VX v0v2 VXV3 v2v2 ^1* 7 3 VQV2 nvx •R X

*1*3 RXR2 RXRX R0RX RXRX RXR2 ^1^3 J
v2v2 V3V3 V2V2 VXVX F 0 7 o |
•*o
#0^3 R0R2 R0RX R0RX RQR2

V0VZ ViV3 V2V2 VQV2 Wl 1

fll#3 J? 1*2 RXRX R0RX RXR2 #1*3

nv2 F0F3 v0vz VXVX v0v2 1 v0v3 vxv2] #2

#2#3 #2#2 RXR2 RQR2 * 1 * 2 ! R2R2 R2R3

1 F2F3 ^ 2 ! V0VX VQVO j VXV2 | v2v3 •/?3

#3^3 ! *2#3 #0#3 #1*3 #2#3 #3^3

t t I f t
#3 i? 2 ^1 *1 i?2 #3

k{n-k)x
V0 +
'2fc 2n+l
0,1,2,.
(fc+l)(n-fe)a?
2k+l 2n+l

R
m = ~F° + 57^711
The reader is also invited to prove these interesting consequences of the
diamond:
THEOREM 4. 3 divides the magic sum of every pandiagonal magic square of
order 3(2&+t) that is composed of integers,
THEOREM 5. Removing a main diagonal from a pandiagonal magic square of
order ?rc+l leaves two triangles of numberss each of which sums up to n magic sums.
THEOREM 6. In every pandiagonal magic square of order 2n+ls Bn of the (2n+l)2
2
cells can each be expressed as a linear combination of the remaining (2n-l) cells.
For n > 1, 2n+l of the (2n-l)2 cells can be located in the fourth row, with the
rest of the (2n-l)2 cells located below the fourth row in a cell rectangle of
dimensions 2n-3 rows by 2n columns.
We end with the analogue of Theorem 6 for magic squares of even order.
THEOREM 7. In every pandiagonal magic square of order 2n3 8n-5 of the
(2n)2 cells can each be expressed as a linear combination of the remaining
 - 267 -
(2n-2)2fl cells. Two of the (2n-2)2+l cells can he located in the first column
at rows 3 and 4, with the rest of the (2rc-2)2+i cells located to the right of
the first column and below row 3 in a cell rectangle of dimensions 2n-3 rows by
2n-l columns,

524 S. Court House Road, Apt. 301r Arlington, Virginia 22204.

* ft it

THE OLYMPIAD CORNER: 29

M.S. KLAMKIN
First a correction. In the 1981 International Mathematical Olympiad, U.S.
team member James R. Roche won a Second Prize Award, not a Third Prize Award as
reported earlier in this column [1981: 222].
I give two 1981 Olympiad problem sets this month, one from Hungary and one
from England. I invite readers to send me, for possible publication in this
column, copies of other recent competitions, as well as elegant solutions to the
problems proposed in the present and in past columns.
I am grateful to L» Csirmaz for the English version of the Hungarian problems
given below. (I am not sure what First, Second, and Third Versions imply, and I
do not know the time allotted for these problems.)
1981 HUNGARIAN MATHEMATICAL OLYMPIAD (SECOND ROUND)

First Version

1, For which natural numbers n is 2 8 + 2 1 1 + 2 n a perfect square?

2, The real numbers x and y satisfy o < x < 1 and x+y - l. Determine the
maximum and minimum values of the expression
1+x2 , 1+y2
>y~ 4- 77 • U
x y
1-hc l+y
3, A frustum of a certain triangular pyramid has lower base of area A ,
upper base of area B (where B < A)9 and the sum of the areas of its
lateral faces is P. The frustum is such that it can be divided, by a plane par-
allel to the bases, into two smaller frustums in each of which a sphere can be
inscribed. Prove that

p = (/A + JB){¥A + VB)2 .

Second Version

1, Make pairs from the medians of the faces of a tetrahedron in such a way
 - 268 -
that medians starting from the same midpoint of an edge form a pair. Suppose
that in each pair the two medians have equal lengths. How many different lengths
of these medians can there be?
2, Let
„,(x) N fsinTTx, if x < o ,,.
a d t . = fcos TT#. if re < 1
f = W-D+l, I f ^ O " *<x) [gix-lUl. If x i i.
Solve the equation fix) = gix).
3, Denote by f i k ) the number of zeros in the decimal representation of the
natural number h* Compute
n
A
10
where n = 10 -1.
Third Version

1, Six points are given on a circle. Choosing any three of them (this can
be done in 20 ways), the orthocentre of the triangle determined by these
three points is connected by a straight line to the centroid of the triangle de-
termined by the remaining three points. Prove that these 20 lines all go through
a fixed point.
2, Let n be a positive integer, and let f i n ) denote the number of triplets
consisting of three different positive integers the sum of which is exactly
n. (Two triplets are considered to be identical if they differ only by the order of
their elements.) For which n is f(n) an even number?
3, Construct (and prove your result) a polynomial P(x) with integral coef-
ficients such that
IPC*) - 0.5| < J ^ ;

for every real number x in the interval Co.19, 0.813.

I am grateful to Willie S.M. Yong for sending me the problems of the 1981
British Mathematical Olympiad which appear below.
BRITISH MATHEMATICAL OLYMPIAD
19 March 1981
Time allowed: 3$ hours

1, H is the orthocentre of triangle ABC. The midpoints of BC,CA,AB are

A ' J B ' J C , respectively. A circle with centre H cuts the sides of triangle
A'B'C (produced if necessary) in six points: dx ,D2 on B ' C ; Ex ,E2 on C'A'; and
 - 269 -
F l f F a on A ' B \

Prove that AD! = AD 2 = BE! = BE 2 = CFX = CF 2 ,

2, Given are the positive integers m and n. s is the sum of m terms of

m

the series

(n+1) - (n+l<)(«+3) + (n+l)(n+2)(n+U) - (n+l)(n+2)(n+3)(w+5) + ...,

where the terms alternate in sign and each, after the first, is the product of
consecutive integers with the last but one omitted.
Prove that s is divisible by ml but not necessarily by m\(n+l).
3, Given that a,b,o are positive numbers, prove that
(i) a3 + b3 + a3 > b2o + a2a + a2b;

(i i) abo > (b+c~a) (<?+a-b )(a+b-c),

4, n points are given in general position in space (i.e., no four are

coplanar), and s is the set of all tetrahedra whose vertices are four
of the n points.
Prove that, if a plane does not pass through any of the n points, then that
plane cannot cut more than n2(n-2)2/e^ of the tetrahedra of S in quadrilateral
cross sections.
5, Find the smallest possible value of I12777 - 5 n |, where m and n are positive
integers, and prove your result.
6, Given that a\, a2s ...s a are distinct nonzero integers and that

p. - n (a. -a.)s i = 1,2,...,«,

v 3
•* j=i
J=l
j*i
prove that \ a./p. is an integer for every nonnegative integer k.
1=1 J»J

I now present solutions to some problems published earlier in this column.

I would appreciate receiving from readers solutions to the problems which still
remain unsolved, or more elegant solutions to those already solved. Two of the
solutions are by Noam D. Elkies who, as a member of the winning U.S.A. team at the
1981 International Mathematical Olympiad, achieved a perfect score; so these pro-
blems were duck soup to him. He is m years old.

J—22 • C1981i 143] Can a spatial figure have exactly six axes of symmetry?
Solution by Noam D, Elkies, student, Stuyvesant H.S., New York, N.Y.
Yes: a regular pentagonal prism.
 - 270 -
J-23. £1981: 143] Three given circles, 0\ , 0 2 » 0 3 , intersect pairwise:
0\ and o2 at points A and B s o2 and 0 3 at points C and D, and o3 and
Oi at points E and F. Prove that the straight lines AB, CD, and EF intersect at
a point.
Solution.
This is a well-known theorem: "The radical axes of three circles, taken in
pairs, are concurrent. For, the point in which any two radical axes intersect has
equal power with regard to all three circles, and therefore lies on the third
radical axis. The theorem is evidently still valid in the various special cases,
namely if one or more of the circles be null, or if two of them be concentric, or
if their centers be collinear." [R.A. Johnson, Advanced Euclidean Geometry, Dover,
New York, i960, p. 32.]

J-24. £1981: 143] A point P is selected in the base BCD of a given tetra-
hedron A-BCD (not necessarily regular) and lines are drawn through
it parallel to the edges AB, AC, AD, intersecting the faces of the tetrahedron in
other points U, V, W. Find the point P of base BCD for which the volume of
tetrahedron P-UVW is a maximum.

Solution.
We give a very direct vectorial solution. A being the origin of vectors, let
the position vectors of the points B,C,D,P be £.,£,3,p., respectively. Then there
are nonnegative scalars x3y,z such that

p = xb + yc + zds x + y + z - 1,
and it follows that the points U,V,W have the position vectors

u ~ yc + zd, v = zd + xb, w - xb \ yc.

Using square brackets to denote the volume of a tetrahedron, we now have
[P-UVW] = ||Ptl*PVxpfc| = ±\xt-yc*zZ\ = a^/aCA-BCD],

By the A.M.-G.M. inequality, xyz is a maximum when x - y - z - 1/3. Thus the

desired point P is the centroid of triangle BCD.

J-27, £1981: 1^3] Solve the inequality 2xy In (x/y) < x2-y2.
Solution.
It is clear that x and y must have the same sign. If we set t = x/y, then
the given inequality is equivalent to

Fit) = t - j - 2lnt > o.

 - 271 -
2
Since F'(t) = (1-1/t) z o and F(l) = o, the inequality holds if and only if
t - x/y > 1.

J-28, C1981: 143] The lengths of the sides of a convex quadrilateral are,
in order, as b, cs d and its area is s. Prove that 2S < aa + bd.
Solution.
We have 2S = ef sine, where e and / are the lengths of the diagonals and e is
the angle between them. Then, by Ptolemy's inequality, we have

2S - ef sin 8 < ef < ac -f bd,

and the desired inequality is established.

There is equality if and only if 8 = 90° and the quadrilateral is cyclic.

J-29, C1981: 143] The base ABC of a pyramid P-ABC is an equilateral triangle.
If the angles PAB, PBC 5 and PCA are all congruent, prove that P-ABC is
regular.

Solution by Noam D. Elkies, students Stuyvesant H«S.S New York, N.I.

First we clarify a slight ambiguity in the proposal. We are asked to show
that P-ABC is a regular pyramid, not necessarily a regular tetrahedron. A regular
pyramid is one whose base is a regular polygon and whose lateral faces make equal
angles with the base (or having the foot of its altitude at the center of the base).

Suppose the pyramid is cut open and

laid flat, with elements labeled as shown
in the figure. It suffices to show that
a = b - c. Without loss of generality, we
may assume that a < b < c. Then triangle PBC
shows that a < e. On the other hand, a comparison
of triangles PCB and PCA shows that e < a. So a =
and b = c. Now triangles PCA and PAB are congruent,
so 8 = e = Y» and a - b - c follows.

J-32, C1981: 143] What conditions must be

satisfied by the coefficients u3v}w of the
polynomial
£3 ~ UX1 + VX - W (1 )

in order that line segments whose lengths are roots of the polynomial can form a
triangle.
 - 272 -
Solution,
Let a9b3c be the roots of (1). Then a,b3o are the lengths of the sides of
a triangle if and only if (i) they are all real, (ii) they are all positive, and
(iii) they satisfy the triangle inequalities.
(i) The roots are all real if and only if the discriminant A of (1) is non-
positive, that is, if and only if [l, pp. 179-180]
27A = 21w2 - IBuvw + *4tf3 + Hwu3 - u2V2 < 0. (2)

(ii) When (2) is satisfied, the roots are all positive if and only if
u > 0, V > 0, W > 0. (3)

(iii) When (2) and (3) are satisfied, the triangle inequalities hold if and
only if
(a+h+o)(b+c-a)(c+a-h)(a+b-c) = 2lb2e2 - lah > 0. (4)

Since u = la, v = tbc, w = abe3 and [1, p. 95J

lb2c2 = v2 - 2wu} lah = uh - Hu7v + 2v2 + ^wu,

condition V\) is equivalent to

^u2v - uh - Stou > 0. (5)

Thus (2), (3), and (5) are the required necessary and sufficient conditions.
REFERENCE
1. S. Barnard and J.M. Child, Higher Algebra, Macmillan, London, 1964.

J-34, C1981: 3 W ABC is a triangle of perimeter p. The tangent to its

incircle which is parallel to BC meets AB in E and AC in F. Among
all triangles of perimeter p, is there one for which EF is of maximum length?

Solution.
We use customary notation for triangle ABC: a>b,c3s for the sides and semi-
perimeter, h for the altitude from A, r for the inradius, and K for the area.
From similar triangles AEF and ABC, we have
EF a_ .
h -2r h '
a a
hence
=
« < * -2P) = t e - ^ ) =«<£2SLl.
EF
h a 2KX a s ' s
a
Since aHe-a) - s - p/2 is constant, a(s-a) takes on its maximum value when
a - s-a, that is, when a = s/2 = p/4, and then EF = p/8. This maximum is attained
for the infinite class of triangles for which a = p/M- and b+c = 3p/<+.
 - 273 -
J-37i C1981: 144] If ayb3c,d a r e , i n order, the sides of a convex quadri-
l a t e r a l 'and s is i t s area, prove that
b d
S * {{^)( -f-)(^-)( -f-)f.
Solution,
It follows from Bretschneider's formula that, for sides of given lengths, the
area S is a maximum when the quadrilateral is cyclic, in which case we have
Smax = y/Ts~a)(s-b)(s-c)(s-d),

where s is the semiperimeter. Hence it suffices to show that

(s-a)(s-b)(s~c)(s-d) < (s-^-)(s~^)(s-^)(s-^).

This follows from the multiplication of the four obvious inequalities

(s-a)(s-b) < («-*f-)\

(s-b)(s-e) < (s-~~)2,

(s-r)(s-d) < {s-C-f-)\

(s~d)(s-a) < (s-^-)2.

There is equality if and only if a - b = o - d9 that is, if and only if the given
quadrilateral is a square. D
More generally, one can show that
(s-a\ ){s-a2){s~a^ )(s-a^ ) < (s-b\ )(s~b2 )(s-b3 )(s-bu ),

where s = la J 2 and
t
= + W
h V Uia? + W
U2°* + W
i+3ai» ^ = 1.2,3 .U,

in which w. , = w., zw. = 1, and w. > o. (This and other extensions are to appear
in a joint paper by M.S. Klamkin and Clarke Carroll, Australian National University.)
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.
& & J'C

POSTSCRIPT TO "THREE MORE PROOFS OF ROUTH'S THEOREM"

I have just come across a direct vectorial proof of Routh's Theorem which is
essentially the same as one of the proofs that A. Liu and I gave recently in this
journal [1981: 199-203]. It can be found in A.S.B. Holland, "Concurrencies and
Areas in a Triangle," Elemente der MathemaKk 22 (1967) 49-55.
M.S. KLAMKIN
 - 271\ -

PROBLENS--PROBLEMES
Problem proposals and solutions should be sent to the editor, whose address
appears on the front page of this issue. Proposals should, whenever possible, be
accompanied by a solution, references, and other insights which are likely to be
of help to the editor. An asterisk (*) after a number indicates a problem submit-
ted without a solution.
Original problems are particularly sought. But other interesting problems
may also be acceptable provided they are not too well known and references are
given as to their provenance. Ordinarily, if the originator of a problem can be
located, it should not be submitted by somebody else without his permission.
To facilitate their consideration, your solutions, typewritten or neatly hand-
written on signed, separate sheets, should preferably be mailed to the editor be-
fore March 1, 1982, although solutions received after that date will also be con-
sidered until the time when a solution is published.

660, ri98l: 20M] Correction: The displayed equation should read

DL + EM + FN = e tan | .

6811 Proposed by J.A.H. Hunter, Toronto, Ontario.

Of all the girls that are so smart,

There's none like pretty Sally.
She is the darling of my heart,
And she lives in our alley.
And when our alley the sunlight dapples,
My darling SALLY
SELLS
RIPE .
W¥HS
(With apologies to Henry Carey (c. 1687 ^ 1743).)

682, Proposed by Robert C. Lyness, Southwold, Suffolk, England.

Triangle ABC is acute-angled and Ai is its orthic triangle (its vertices
are the feet of the altitudes of triangle ABC). A 2 is the triangular hull of the
three excircles of triangle ABC (that is, its sides are the external common tan-
gents of the three pairs of excircles that are not sides of triangle ABC).
Prove that the area of triangle A 2 is at least loo times the area of triangle hx.

683, Proposed by Kaidy Tan, Fukien Teachers University, Fooehow, China.

Triangle ABC has AB > AC, and the internal bisector of angle A meets
BC at T. Let P be any point other than T on line AT, and suppose lines BP,CP
intersect lines AC,AB in D,E, respectively. Prove that BD > CE or BD < CE according
as P lies on the same side or on the opposite side of BC as A.
 - 275 -
r i
6841 Proposed by George Tsintsifas, i hessalonikis Greece.
Let 0 be the o r i g i n of the l a t t i c e plane, and l e t \Mpsq) be a l a t t i c e
p o i n t w i t h r e l a t i v e l y prime p o s i t i v e coordinates ( w i t h q > 1 ) . For i = i , 2 , . . . ^ - l ,
l e t P. and Q^ be the l a t t i c e p o i n t s , both w i t h ordinate i , t h a t are r e s p e c t i v e l y
the l e f t and r i g h t endpoints of the h o r i z o n t a l u n i t segment i n t e r s e c t i n g OM. F i n a l -
l y , l e t P.O. n OM = M..
^ ^ %
(a) Calculate
q-1
sx - 1 ™v
(b) Find the minimum value of PTff. f o r l < i < q-l.

(c) Show t h a t F I T + P FT = l, i < s < q-l.

1 H
' 8 s q~s q-s

(d) Calculate p-^

*2 = I - ^ •
(e) Show t h a t the area o f a simple t r i a n g l e is \. (A simple triangle i s one
whose v e r t i c e s are l a t t i c e points and which has no other l a t t i c e p o i n t i n i t s i n -
t e r i o r or on i t s perimeter.)

6851 Proposed by J.T. Groenman, Arnhem, The Netherlands,

Given i s a t r i a n g l e ABC w i t h i n t e r n a l angle bisectors t ^ t ^ t meeting
a3bsc i n U,V,W, r e s p e c t i v e l y ; and medians m 3m,sm meeting a,b3c i n L,M,N, respec-
tively. Let

m n t , = P, nu n t = Q, m n t = R.
a b b e ' c a

Crux 588 [J980: 3171 asks f o r a proof o f the e q u a l i t y

AP BQ CR _ Q

pTqffW = 8 '
Establish here the inequality
AR,BP,CQ >

with equality if and only if the triangle is equilateral.

686. Proposed by Charles W, Trigg5 San Pi ego3 California,

Without using calculus, analytic geometry, or trigonometry, find the
area of the region which is common to the four quadrants that have the vertices of
a square as centers and a side of the square as a common radius.
TA solution using analytic geometry appears in School Science and Mathematics ,
78 (April 1978) 355.3
 - 276 -
687. Proposed jointly by J. Cliris Fishers University of Regina; and Roger
Servranckx, University of Saskatchewan at Saskatoon.
(a) Show that there exists a number y such that the equation

x
c
c

has three solutions whenever o < a < y.

(b) How many solutions does the equation

cx
c
c

have when there are 2n c's in the ladder and o < a < yl
6881 Proposed by Robert A. Stumps Hopewell, Virginia.
Let o denote a binary operation on the set of all real numbers such that,
for all real numbers a9b9o9
(i) 0°a = -a; (ii) a°(b°c) = c°(b°a).
Show that ao(boa) = (a°b)°(-c).

6891 Proposed by Jack Garfunkels Flushing, N.I.

denote the lengths of the medians to sides asb9c, respective-
Let m^m^jTn^
ly, of triangle ABC, and let M SM.SM denote the lengths of these medians extended
to the circumcircle of the triangle. Prove that
Ma fma + Mbu/mub + Mc fmc > 4.

6901 Propose par Eippolyte Charless Waterloos Quebec.

n gtant un e n t i e r p o s i t i f donng, s o i t
n+k trix, , sn+k ,nS/ nXn+fc , . sn-lf n x
S = n
k ~ 4)(w-D + \2nn-2) - . . . + (-1) (n-1)-
Calculer s~ pour k = 0 , 1 , 2 , 3 .

67^i C1981: 2391 (Correction) The second sentence should read: The i n c i r c l e
o f the medial t r i a n g l e touches i t s sides i n R, S, T (R being on B ' C ,
etc.).
ft ft ft

MAMA-THEMATICS
Frau Hilbert, to her son: "David! Stop staring into space!"
Napier's mother: "Make no bones about it, John, computation is as easy as
rolling off a log."

ALAN WAYNE, Holiday* Florida

 - 277 -
SOLUTIONS
No problem is ever permanently closed. The editor will always be pleased to
consider for publication new solutions or new insights on past problems.

492. £1979: 291; 1980: 291; 1981: 50, 117] Proposed by Dan Pedoes University
of Minnesota.
(a) A segment AB and a rusty compass of span r > 1AB are given. Show how
to find the vertex C of an equilateral triangle ABC using, as few times as possible,
the rusty compass only.
(b) Is the construction possible when r < |AB?

III. Comment on part (b) by Stanley Rabinowitzs Digital Equipment Corp.,

Merrimackj New Hampshire.
We consider here the possibility of finding the third vertex C of an equila-
teral triangle ABC, given only the two points A and B (not the entire segment AB)
and a rusty compass of span r < |AB. The circles with centers A and B do not
intersect; hence no new "distinguished" points can be constructed and, in attempt-
ing to solve the problem, we will have to pick "arbitrary" points in the plane
and use them as centers to draw additional circles. We will show that the problem
can always be solved in practice. Our technique will be based on the following
lemma, which was suggested by a problem of Bankoff [l]. (The notation A v B = C
used in the lemma means that ABC is an equilateral triangle with the vertices
labeled A,B,C in counterclockwise order J
LEMMA. Let A, B, P be any three distinct points in the plane. If

PVA = Q, B V P = R, R v A = S, Q V S = C,

then B V A = C. p
Proof. We assume that the points are / T ^ ^ ^ ^
imbedded in the complex plane and, to yS 1\ ^s"*"\^^
simplify the notation, we identify each yS 1\ ^^-^>^^ B

point with i t s a f f i x (see Figure l ) . A *^~~ " IT " " ~/

f t is known (see, e . g . , C23) that i f / \ ^ N . 1\ /
2 Q
X v Y = Z, then Z = -Xo> - Yo> , where / ^ \ ^ \ /
u = e2^/d. Easy calculations (using / y\\ \ /
w = 1 and 1 +• a) + cu2 = o) now give
3
I jf ^\\s
I jS ^Jf ^ Figure 1
2
Q = -Pw - Aw , / X ^^"^\
R = -Bto - Pw2, 1 / ^ ^ ^ \
 - 278 -
S = -Rw - Aw 2 = P + (B-A)co2,
C = -Qo) - SOJ2 = ~Bo) - Aw 2 ,

and B v A = C follows. •
How does this help with our problem? Let us say that we can reproduce a
distance d if, given two points X and Y that are d units apart, we can construct
the point Z = Y v X with our rusty compass. Prior work on this problem [1980: 291]
shows that we can reproduce any distance d < 2r.
Let A and B be two points such that AB > 2r. If we can pick a point P in the
plane such that none of the lengths PA, PB, AR, and QS (with P,Q,R,S as in the
lemma) exceeds 2r, then we can construct Q, R, S, and C in succession and thereby
reproduce AB. For example, Figure 1 shows how to reproduce AB with a rusty compass
of span r = (l/3)AB, since none of the distances PA, PB, AR, and QS exceeds (2/3)AB.
Suppose we could prove the following theorem for some fixed G > 0:
THEOREM T . If we can reproduce any distance that does not exceed 6S then we
can reproduce the distance (l+e)6.
Then it would follow that we could reproduce any distance. For we already
know we can reproduce any distance d z 2r3 and any distance D > 2r can be built
up from some distance d << 2r by a finite number, n9 of successive applications of
Theorem T . (Pick some n > log. (D/2r).)

But how do we pick the point P given A and B? If we C*=AvB

were fortunate enough to pick for P the center P* of the equi-
lateral trianglewith vertices A, B, and C = A 7 B (see
Figure 2 ) , then the distances PA, PB, AR, and QS would
all equal d//3 « o.577<f, where d = AB. But we cannot
pick this point out exactly by eye; however, it is easy
to pick a random point P fairly close to P*. By a con-
tinuity argument, if P is close enough to P*, then
max {PA,PB,AR,QS} can be made arbitrarily close to
d/Jz.
In practice, it is very easy to pick a random
point P by eye so that max {PA,PB,AR,QS} <; 0.9<f.
Even if we were to pick for P the midpoint of AB, then
we would have
max {PA,PB,AR,QS} = di/3/2 « 0.866<f < o.9<3.

Thus, in practice, Theorem T is true for l+e = 1/0.9, or e = 1/9. We have there-
fore given a practical method of reproducing any distance. Unfortunately, we have
 - 279 -
not specified an effective method of picking a point P other than the vague
instruction: Pick a random point P by eye somewhere near the center of triangle
A B C \ where C* = A V B.
An effective method (i.e., one involving no randomness) would follow from
the following conjecture, which I have been unable to prove:
CONJECTURE. If P is a point such that PA = r and /PAB <> 30°, then

max {PA,PB,AR,QS} < d/(l+e)

for some fixed e > 0 independent of d - AB and r.

If the conjecture is true, the point

P can be used to prove Theorem T . We
could effectively construct such a
point P by drawing a circle
around A, picking any point Pi on Pi
this circle, and then swinging six
successive arcs with the rusty
compass to obtain the vertices of
a regular hexagon P1P2P3P4P5PG
(see Figure 3 ) . At least one of Pz
these six points P. must be such
that angle P.AB does not exceed
30°, Then we just apply the algorithm
of the lemma using P = P., l < i ^ 6, Figure 3
and abort the operation any time we
encounter a distance greater than 2r. By our conjecture, one of these six tries
would be guaranteed to succeed.
A comment on part (b) was also received from BIKASH K. GHOSH, Bombay, India.

REFERENCES

1« Leon Bankoff (proposer), Problem 438., Mathematics Magazine, 34 (1961) 174.

2. A.M. Gleason, R.E. Greenwood, and L.M. Kelly, The William Lowell Putnam
Mathematical Competition Problems and Solutions 1938-1964, Mathematical Association
of America, 1980, p. 498.
?'« & ;'c

526, C1980: 78; 1981: 87] Proposed by Bob Prielippy The University of
Wisconsin-Oshkosh.
The following are examples of chains of lengths 4 and 5, respectively:
 - 280 -
25, 225, 1225, 81225
25, 625, 5625, 75625, 275625.

In each chain, each link is a perfect square, and each link (after the first) is
obtained by prefixing a single digit to its predecessor.
Are there chains of length n for n = 6,7,8,...?

II. Solution by L. Csirmaz, Mathematical Institute, Hungarian Academy of

Sciences.
The answer to the question is NO. In fact, we show that the only chains of
length at least 3 are those listed in the earlier partial solution [1981: 88].
This list was exhaustive for squares less than loltf, so we now restrict our atten-
tion to squares greater than lo 1 M . For such squares, if there is a chain of
length at least 3, then it has a subchain of length 3, and we show that no such
chain of length 3 exists (with links containing no initial or final zeros).
Suppose, on the contrary, that (a 2 , 3 2 , y2) is such a chain, where lofa and

lO^" 1 < a 2 < 10*, k > 13; (1)

then there are nonzero digits a and h such that

10 a + a 2 = p 2 and 10 b + 32 = y 2 .
k
From 10 a - (3+a)($-a), we get

6 +a =a^V and 3 - a = a22k~"l5k~V\ (2)

where o < i3j < k and aisa2 are nonzero digits such that
= a
1 < a\ < a2 £ a and a\a2 * (3)
2 2 2
Using o.4*io < 4a < 4«io and 2a*lo = 26 - 2a , the first of which follows from
(l), we obtain
(O.H+2a)10k < (3+a)2 + (6~o02 < (i*+2a)10*\ (i+)
2 k 2 2 2
If we set (3+a) /lo = a 2^s^'/io* = As then (3-a) /10* = a /A, and (i+) is
equivalent to
a2
0.4+2a < A + ~ p < 4+2a. (5)

Observing that (e+a) 2 > 6 2 > 10 a9 so that A > a, we can solve (5) to get

0.2+a+/(0.2+a)z-az" < A < 2+aW(2-m)z~az,

from which follows
 - 281 -
Of the pairs (aua) which satisfy (3), the pair (3,9) minimizes the left member
of (6) and the pair (1,9) maximizes the right member; hence

1.234 < —~— < 17.325,

10*

which implies that

i.lll'10*/2 < 2 V < 4.163-10*/2. (7)

Since lo|a, we cannot simultaneously have 1 < j < k and 2 < i < k~l. How-
ever, j * 0 in (7) because 1.111-10 > 3 > 2 ; and j * k because (5//10V > 4.163
if k > 4 (and we have k > 13). Therefore 1 < j < k and, consequently, either i < 2
or I ^ k-l.
If i < 2, then 1 < 2^" ^ 2 and, from (7),

0.555-10^/2 < 5J' < 4.163-107</2;

hence (replacing j by fc-j in (2)) we have

3 + a = s^'J and e - a = e22*~V,

2 0 ana
where 0 < s^sa < *

-0.3658+0.7153& < fc-j < 0.8862+0.7154k.

Similarly, if i > fc-i, then J* 2* < 21 < 2K and

l . l l l ( 5 / 2 ) * / 2 < 5J' < 8 . 3 2 6 ( 5 / 2 ) 7 < / 2 9

so that
a = 8i5k~J and 8 + a = e22* V ,

with 0 < 8Us2 < 20 and

0.0654+0.2846^ < J < 1.3169+0.2847k.

Summing up each of these two cases, we get

26 = e^-* + s22J<-hJ
and
23 - \sx5k~j - s 2 2 7 < ~V|

with
-0.8862+0.2846fc < J < 1.3169+0.2847fc. (8)

The same result holds with 7<+I,8,Y instead of ^a,65 that is, there are integers
s{3&2i with 0 < s{ss2 < 20»andj' such that

23 = k{5/<+1*"J" - s ^ V ' l
 - 282 -
with
-0.8862+0.2846(fc+l) < j" < 1.3169+0.2847(fc+l), (9)

Now, changing the signs of s[ and s^ if necessary, we arrive at

(28 =) 9l5k'* + s 2 2 * - V = s{5k+±-r -ai2VJ

that is,
2k~1{81Sj + 2e45J") = s{5k+1~J' -^S^'. (10)

From (8) and (9) we know that fc+l-j" and k-j both exceed

I = -1.3169 + 0.7153&;

therefore the right member of (10) is divisible by 5 . The left member is not
divisible by 5 J + 1 if j < j \ nor by 5 J " + 1 if j" < j , nor by 5'7+3 if j = j". (In
the last case, the left member equals 2 5t/(s2+232)» and the claim follows from
|s2+ 2s2I < 20+2^20 = 60.) In any case, the left member is not divisible by
5 C r ] , where

£' = 3 + 1.3169 + 0.2847k,

and so I < V. But we also have I > V if k z 14, and this is the desired con-
tradiction. D
A modification of the above proof would show that there are infinitely many
chains of length 2.
* * *

556, C1980: 184; 1981: 189, 2411 Proposed by Paul Erdos, Mathematical Insti-
tute, Hungarian Academy of Sciences.
Every baby knows that

(n+l)(w+2)...(2n)
n(rc-l)...2.1

is an integer. Prove that for ey^ry k there is an integer n for which

(n+l)(n+2)...(2n-k) (1)
n(n-l).,.(tt-fc+1)

is an integer. Furthermore, show that if (1) is an integer, then k = <?(n), that is,
kin •+ 0.

II. Solution by L, Csirmaz, Mathematical Institute, Hungarian Academy of

Sciences.
Denote by n(k) the smallest integer n > k for which
 - 283 -
(n-H)(tt+2),..(2tt-fe) _ (2n-k)\{n-k)\
«(«-!)...(n-fe+1) " {nW (1 ! )

is an Integer.
First we prove that n(k) exists for every k. We easily find that n(i) = 6,
n(2) = 6, and n(3) = 9. For k > 4, we claim that (1') is an integer for n - k\ - 2 .
From this it will follow that n(k) exists for every k, and that n(k) < k\ -2 when
fc > M-.
Observe first that for the stated values of k and n we have Jn < n/2-fc,
k < n9 and there is no prime p such that n-fc/2 < p < n. Let

«•> • [ ^ ] * [=?] - »ia •

It is easy to verify that

> l, if 0 < x < j - k3

z 0, if | -fc< * <; n - ~,
fix)
= -1, if n - y < x £ ttj

> 0S if a? > n.

Now let p be any prime. The exponent of p in s! is E^ O / p ] ; therefore its

exponent in (l') is

p
-£=1

We prove that e > o, and from this our claim follows,, We have /(p ) > o except
possibly if n-k/7 < p1 <> n. In that case we must have i > 2 since there is no
prime In that interval. At the same time we have

p1"1 ^t-^|<„-| and pUl > 2pi > 2n-k > n;

so fip1) - -1 in (2) for at most one i. In this case, however,

i/2
2 < p < p = ^ <> Vn~ < | - k,

that Is, fip) > 1. Thus f{p) + fip1) > o, and £ > o since the remaining terms
are nonnegative.
Now let e > o and suppose that k/r?(k) > e for infinitely many k. This will
lead to a contradiction. It is well known that there exists a threshold number
M Q ( E ) such that for every n > WgCe) there is a prime p between n - (e/2)n and n.
Now choose k so large that k/n(k) > e and n{k) > « Q ( E ) . Then there is a prime
p such that n-k/2 < p < n. In the right member of ( i 1 ) , the numerator is divisible
 - 284 -
by p only, while the denominator is divisible by p 2 . Therefore (l 1 ) cannot be
an integer, and we have our contradiction. We conclude that Wmk/n(k) = o.
k->™

A partial solution was received from BIKASH K. GHOSH, Bombay, India.

* * sV

572, f^980: 251! Proposed by Paul Erdos, Technion - I.I.T., Haifa, Israel.
It was proved in Crux 458 T1980: 1571 that, if <f> is the Euler function
and the integer t > l, then each solution n of the equation
4»(n) ~ n - t (1)
satisfies t + ? < n < t2.
Let F i t ) be the number of solutions of (1). Estimate F i t ) as well as you can
from above and below.
Edi tor's comment.
The following solution was received, in slightly different form, through the
intermediary of the proposer. The editor has added, in square brackets and small
type, a few remarks drawing attention to points where some confirmation or clari-
fication may still be needed.
Solution by Imre Z. Bursa, Mathematical Institute, Hungarian Academy of
Sciences.
Our solution will require the following two lemmas,
LEW A 1. If n - pi1.. ,par, where the p. are distinct primes and the a. are
positive integers, then

where d = p\.. .p .
Proof. We have

• («) = iV\ ~V\ )---<Pr*,-Prr >

<zi -1 a •1 f , v , ,x
= PI* • ••*y (pi-i>---<pr-i)
ay a r
V l • * *P

= %<di.

LEMMA 2* The number w(n) of distinct prime factors of n satisfies

w(n) - O(loqn).

Proof, With the notation of Lemma 1, we have

 - 285 -
logn = ^ilogpi *-..,+ a^\ogpr
> log 2 + ... + log 2
= r log 2,
so r <: logn/log2, that is, w(n) = a(logn). 0
The answer to our problem is contained in the following
THEOREM. Let Fit) denote the number of solutions of the equation

n - «fr(n) = t, (1)
where <f» is Euler's function and t is a positive integer. Then we have

Fit) - Oit loglog£/log2t) (2)

but
Fit) * oit 1oglog£/log2t). (3)
Proof. Let F§it) denote the number of solutions of the form n = pq, where
p and q are distinct primes. Then (1) becomes
p + q = t +1.
From Theorem 3.11, page 117, of Halberstam and Richert's 5ievs Methods (Academic
Press, 197*0, we have
F0it) = oit loglogt/log2*), (m
and it is known [no reference was given for this and I have been unable to find
one] that
F0it) * oit loglog t/log2t). (5)

We will prove that

(6)
Fit) - F0(t) = Oit/Gogtf)

for every fixed a > o. Then (M-) and (5) together with (6) imply (2) and (3).
We consider in turn the possible forms of the solutions of (l) other than the
form n = pq, with p and q distinct primes. These are
(a) n = p a ;
(b) « = p a ^ a > 1;
/ \ a8
(c) n = p <?, a > 1, $ > 1;
(d) numbers n having at least three distinct prime factors.
For solutions of the form (a), (l)c t - becomes
1
p = t,
which has at most one solution.
 - 286 -
For solutions of the form (b), (l) becomes

Thus the only primes p which can give rise to solutions of (1) are those which
divide t. Hence the number of solutions of the form (b) is, by Lemma 2, less than
«(*) = 0(log£).
For solutions of the form (c), (l) becomes

p q
Here the only primes p and q which can give rise to solutions of (l) are those
which divide t. Hence the number of solutions of the form (c) is less than
Mt)}2 = 0(log 2 t).
We now turn our attention to solutions of (l) of the form (d). Let G{t) be
the number of solutions of (1) when n is restricted to square-free numbers having
at least three prime factors. We will show later that
G(t) = O(t/Oogtf). (7)

Assuming (7) for the moment, we obtain the number of not necessarily square-free
solutions of (1) of the form (d). Let m be the product of the distinct prime i
divisors of n , and set n=md. Then
n - <f>(n) - md - $(md)
= md - 4>(m)d (by Lemma l)
= (m - $(m))d,

so that (1) becomes

m - <Km) = j .

Hence the total number of solutions of type (d) is

< I G(t/d)
d\t
~ £ G(e) [here some care is needed to ensure that
e\t e > 1]

= o{ I ~ T T ) <for a f1xed
°)
e\t (log*)*
= 0{<ait) ~7T)
(logtf+1
 - 287 -

(logtf
completing the proof of (6).
We now turn to the proof of (7). Let p be a prime divisor of n and set
n = pk. (Recall that n is assumed square-free, so pXk.) We have
t =rc - $(rc) ~ pk - <f>(p&)
= pk - (p-m(k)
= k + (p-l)(k- *(&)),
so that
„ 1 - JLzJL .
P
" " fe-<Kfe) (8)
Since p-i > o, it follows from (8) that
k <t and k-$(k)\t-k. (9)
We will estimate the number of solutions k of (9); and n is then uniquely deter-
mined by means of (8) and n-pk. Let q be the largest prime divisor of k, and set
k = qli then I > l since « has at least three prime divisors. We have

k - <Kfc) = <?Z - <KqZ)

= ql - (q-m(l)
= <?U-<f>a)) + * a ) ,
and this divides t - ql by (9), Set
u = q(l-*(Z)) + <frU)
and
v = t(Z-<fr(n) + Z<f><7),
so that u\t-ql. Then
t s ql (modu)
and
«(Z-<KZ)) = -<KZ) (modu),
so
gy = q{t(l-$(l)) + 1(f)(1)}
= -qi$(i) + q^cn
= o (modu).
Suppose (q,u) > l. Then, as (7 is prime, we have q\us so <?|£ and q|^(Z). [we would
like this to be impossible or yield only a small number of k, but it is not clear
 - 288 -
why this should be so.3 Hence we must have (q,u) = 1 and so u\v. Now
v = t l - (t-7)(j>(l) < tl,
as I < k < t9 that is s v < t2. Hence for u the number of possibilities is d { v ) 9
the number of divisors of v. From Theorem 317 of Hardy and Wright's An i n t r o d u c t i o n
t o the Theory of Numbers (Fourth Edition, Oxford, I960) , we have
d(v) < 2° lo9tf/lo9logt? < ^logt/loglogt = ^/loglogi^

As I - $ ( l ) * o, thus u determines q ; therefore the total number of solutions for

which
t < t i-//loglog*
is
< ^/loglogt . ^1-f/loglogt = t l-(/-e)/loglog* = 0(t/c\0gtf).

If I > t l " ^ o g l o g t s thGn

„ - k < t * ^/loglogt
q
" T~ I v
As q was maximal, k is a product of primes, all less than
^/loglogt.
The number
of such numbers up to t is known to be o(t/C\oq t)°). [A reference for this
result would have been helpful.] This completes the proof.
Editor's comment.
The proposer had written: "It is an old unsolved problem of Sierpinski and
myself that infinitely many integers are not of the form n - <$>(n)."
A s'e s'e

582. C1980: 283; 1981: 254] Proposed by Allan Wm. Johnson Jr., Washington B.C.
In how many ways can five distinct digits A3 Bs cs D3 E be formed into
four decimal integers AB, CDE, EDC, BA for which the mirror-image multiplication
AB • CDE = EDC • BA

is true? (For example, the mirror-image multiplication AB • CD = DC • BA is true

for 13 • 62 = 26 • 31.)
III. Solution by Clayton W. Dodge, University of Maine at Orono.
The given equation can be written in the form
(1QA+B)(100C+10D+E) = (100E+10Z>+(7)(10B+4),
which is equivalent to
lll(A'C-B'E) = 10(A'E+B*D-A'D-B*C).
 - 289 -
Since IO\A'C-B-E, A*C*B-E implies

A'E+B'D-A*D-B'C = A(E-D) + B(D-C) > 111,

which is impossible. Hence A-C = B*E and c = B*E/A. Now

A-E+B-D = A'DiB'C = A-D+B7'E/A

which, since 4-B * o, reduces to

Clearly, B and E cannot be too large because D = E+C < 10, so we try various values
of As B, and E, keeping B'E/A an integer. The possibilities are soon exhausted,
and we find that there are exactly ll solutions (if solutions AB-CDE = EDC'BA and
BA'EDC - CDE'AB are considered to be identical).
[The 11 solutions are those listed on page 255, excepting the last two in the
first column, which have repeated digits. (Editor)]

IV. Comment by Stanley Rabinowitzs Digital Equipment Corp«5 Merrimack^ New

Hampshive.
If Clayton W. Dodge, Charles W. Trigg, Kenneth M. Wilke, and the proposer
all say that there are only ll solutions, then it is a fairly safe bet that there
are on!/ 11 solutions. In fact, the two "solutions" given at the bottom of the
first column on page 255 are invalid since they do not contain five distinct digits,
as required.
I employed a Rube Goldberq contraption of my own (also known as a VAX-jl/78o)
to confirm that, in fact, there are no other solutions.
While I was at it, I also decided to look at the related problem of finding
six distinct digits such that

ABC'DEF = FED'CBA.

I found that there are essentially only 3 solutions and that they can be written
down 1n the following interesting pattern:
143-682 = 286*341,
431-268 = 862-134,
314-826 = 628-413.

Additional comments have (so far) been received from RICHARD V. ANDREE,
University of Oklahoma; CHARLES W. TRIGG, San Diego, California? and the proposer.

EditorTs comment.
The editor had momentarily forgotten that the problem required distinct digits.
Then he licked his chops and made snide remarks about Rube Goldberg contraptions
[1981: 2551.
 - 290 -
Dodge, Trigg, Wilke, and the proposer had given the ll correct solutions, no
more and no less. In some cases, their processes for squeezing out solutions may
have been elaborate but, like all Rube Goldberg contraptions, they worked,
ft ft 5'*

5831 C1980: 283; 1981: 256] Inadvertently omitted from the list of solvers:
J.T. GROENMAN, Arnhem, The Netherlands*
ft ft ft

584 9 C1980: 283] Proposed by F.G.B. Maskell, Algonquin College, Ottawa.

If a triangle is isosceles, then its centroid, circumcentre, and the
centre of an escribed circle are col linear. Prove the converse.

Solution by Kesiraju Satyanarayana, Gagan Mahal Colonys Hyderabad, India.

Let ABC be a triangle with circumcentre 0, centroid G, orthocentre H, and
excentre I2 opposite angle B. We assume that 0, G, and I2 are col linear.
If 0 = G, then 0 = G = H (since GH = 20G) and the triangle is equilateral,
hence isosceles. If 0 * G, then 0,G,H are collinear on the Euler line of the
triangle, and hence 0,H,I2 are collinear. Let Mi »M3 be the midpoints of BC,BA;
F l s F 3 the feet of the altitudes on BC,BA; and let the escribed circle with centre
I 2 meet BC,BA produced in Tx ,T 3 . We know that BTx = BT 3 = s, the semiperimeter of
the triangle. Thus

0H:0I2 = MiFxiMiT! = (MxC-FiO^B^-BMj)

= (|-fccosC):(s-|) = {a2-2abcosC)ia(b+c)
= (c2-b2):a(b+c) = (e-b):a,
and similarly
0H:0I 2 = M 3 F 3 :M 3 T 3 = (a-b):c.

Thus {o-b):a = (a-b):c9 from which {c~a)(c+a-b) - 0. Since o+a-b > 0, we must have
c - a, and the triangle is isosceles.

Also solved by J.T. GROENMAN, Arnhem, The Netherlands; and the proposer.
ft ft ft

THE PUZZLE CORNER

Answer to Puzzle No. 3 [1981: ?37]: 19463
4 .
077852
Answer to Puzzle No. 4 T1981: 237]: Pieces of eight.
ft ft ft