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

CRUX ^ A T H E W A T I C O R U M

Vol. 4 S No. 8
October 1978
Sponsored by
Carleton-Ottawa Mathematics Association Mathematique d"Ottawa-Carleton
A Chapter of the Ontario Association for Mathematics Education
Publie par le College Algonquin
(32 issues of this journal, from Vol. 1, No. 1 (March 1975) to Vol. 4, No. 2 (February
1978) were published under the name EUREKA.)

.'. A A A A A A A A A .»- A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A

CRUX MATHEMATICORUM is published monthly (except July and August). The yearly
subscription rate for ten issues is $8.00 in Canadian or U.S. dollars ($1.50 extra
for delivery by first-class mail). Back issues: $1.00 each. Bound volumes: Vol.
1-2 (combined), $10.00; Vol. 3, $10.00. Cheques or money orders, payable to CRUX
MATHEMATICORUM, should be sent to the managing editor.
All communications about the content of the magazine (articles, problems,
solutions, permission to reprint, 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, Mathematics Department, Algonquin College, 281 Echo Drive,
Ottawa, Ontario, K1S 1N3.
Managing Editor: F.G.B. Maskell, Mathematics Department, Algonquin College,
200 Lees Ave., Ottawa, Ontario, K1S 0C5.
Typist-compositor: Nancy Makila.

CONTENTS
Computer Oriented Solutions Alan G. Henney and Dagmar R. Henney 212
A Riddle Andrejs Dunkels 216
Algorithms and Pocket Calculators: Square Roots: III . . . . Clayton W. Dodge 217
A Formula for Polytopes? Dr. A.H. Noel 223
Our Own Mother Goose Clayton W. Dodge 223
Problems - Problernes 224
Publishing Note 226
Solutions . , 227
The Decapus Supaced and Edith Orr 2^0

For Vol. 4 (1978), the support of Algonquin College, the Samuel Beatty Fund,
and Carleton University is gratefully acknowledged.

- 211 -
 - 212 -

COMPUTER ORIENTED SOLUTIONS

ALAN P. HENNEY and DAGMAR R. HENNEY

A solution of the celebrated Four Color Problem has recently been advanced by
Kenneth ApDel and Wolfaang Haken, of the University of Illinois, through extensive
use of the computer. (See [2] for a discussion of the problem and its solution.)
The conjecture had baffled scientists for over 120 years.
One hopes that their approach marks a breakthrough in scientific thought.
There are many conjectures which, at the present time, can be proved or disproved
only by lengthy computations. Their solution may be hastened by the success of
Haken and Appel, which may lead to new methods for humans to interact with computers.
Solutions of the famous "15-14 Puzzle" have appeared recently [l] and the
following "Canadian I.Q. Problem" has also been partly analyzed with the assistance
of the computer.

A Monte Carlo Approach to Sequential Machine Puzzles or the Canadian I.Q. Problem.
Many popular puzzles that involve a rearrangement of pegs or blocks can be
considered models of what are called "sequential machines". Each movement of a
piece is an input and each arrangement or state of the pieces is an output.
Reference [l 1 discusses a systematic approach to finding solutions to one type of
sequential machine. However, a simpler aid often effective approach is the Monte
Carlo method (method of statistical trials). Figure 1 shows a popular puzzle
which can be solved by the Monte Carlo technique.
Fourteen pegs are arranged as shown on a triangular board X X X X
with fifteen holes. The objective of the puzzle is to X X X X
remove one peg at a time by jumping in a straight line. The X X X
jumped peg is removed. The process is continued, if possible, X X
until only one peg remains. One popular version uses the X
starting configuration shown. Fourteen other starting config-
urations are possible by simply relocating the position of the Figure 1
hole.
Such puzzles are usually solved manually by trial and error. During the process
a person hopes to gain experience indicating the type of spatial arrangement of the
pegs that will lead to a successful solution. The game ends when one or more pegs
remain on the board in such an arrangement that no more jumps are possible.
After a score of attempts one begins to realize how difficult it is to find a
 - 213 -

sequence of inputs that will permit the game to continue until thirteen pegs have
been removed. Insight gained through experience does not lead one quickly to a
solution. Although experience does help one to reduce the number of pegs remaining
at the end of the game, one can readily see the value of using an electronic
computer to either solve the problem or at least to gather statistical data which
can be used in reaching a solution.
The Monte Carlo approach is one way to attack the problem. Monte Carlo is
simply a fast automatic method for employing the trial and error process. With an
electronic computer the game can be played thousands of times in a few minutes.
Although the computer cannot employ insight to eliminate obviously poor moves 9
statistical information can be easily tabulated and printed to help future program-
ming efforts. Examples of such information will be given later.
The computer program to solve the puzzle of Figure 1 was written in FORTRAN.
The technique used is interesting because it can be applied to many types of
similar puzzles. It consists of three major parts:

1. input and output routines;

2. a routine that indicates the possible jumps for a peg positioned in
each of the fifteen different holes;
3. a method for selecting all possible jumps at each stage of the game and
a random number generator for selecting one move randomly from these possible jumps.
The input-output routines enter into the computer such information as the
location of the hole prior to the first move, the number of games to be played, and
a seed number for the random number generator. An electronic computer generates a
set of random numbers relative to a given initial number. Unless the initial seed
number is changed the random number generator will produce exactly the same set of
random numbers each time it is used. This feature is useful while a computer
program is being debugged9 but during use each series of games requires a different
set of random numbers and hence a different seed number.
The output information can include many parameters. Some obviously important
ones are the number of trials, if any, leading to a solution and the sequence of
moves resulting in a solution. Also the percentage of games ending with exactly
2 5 3 9 ... pegs remaining is interesting. Table l is an example of this type of data.
The second part of the computer program is the most tedious to write. To
illustrate how it is constructed, identify each position in Figure l by a number as
shown in Figure 2. For each number identify the possible paths that a peg positioned
in that hole could take. For example, arrows indicate that a peg in position 2 can
 - 214 -

follow only one of 5 different paths. For each 1-2-3-4-5

possible path (there are 37 in all) instructions
are written indicating how the arrangement of the
6/18 9
board is changed when that oath is followed. It 10 11 12
is assumed in each case that the peg-hole arrangement 13 1[1
/ | ^
is such that a move can be completed.
15
The third part of the computer program examines
each position in turn and lists the jumps that are Figure 2
possible at this stage of the game. For example, consider Figure 1 at the start of
a game with the hole at position 4. The computer program examines each of the 15
positions in turn. Only two of the possible 37 paths can be taken: a peg can jump
from position 2 to 4 or from position ll to 4. Now a random number is generated (a
fraction between o and l ) . If the fraction is greater than \ the second path is
foil oweds otherwise the first path is chosen. The arrangement is then changed
accordingly and the process is repeated, starting with the selection of all possible
jumps for the second move. The game ends when no more jumps are possible. If only
one peg remains, the game has ended successfully and the sequence of moves is
printed. Otherwise the computer program restores the initial arrangement on the
peg board and starts another game.
The arrangement shown in Figure l was run on a computer three times with the
number of games per run increased each time. Table 1 summarizes the results.

Table l

No. Of Games ending w i t h i n d i c a t e d no. of pegs remaining (% of t o t a l )

games
J^ 2 3 4 5 6 >7
20 0 0 20.0 65.0 5.0 5.0 5.0
200 0 2.5 28.5 46.0 10.5 8.5 4.0
2000 0.2 > 3.U 25.6 44.3 13.7 8.5 4.3

The first computer run limited the number of games to 20 mainly to check the
computer program to make certain it was running properly. The number of trials was
then increased by a factor of 10. No solutions were found but several games did
end with only two pegs remaining. The number of trials was increased again by a
factor of 10. The results are shown on the third line of the table. This time
four different solutions were obtained. The sequence of jumps for one of these
solutions, using the notation of Figure 2, is 2-4s 12-3, 4-2, io-3, 13-8, 2-4s
 - 215 -

5-12, 14-9, 1-10, 4-11, 10-12, 9-14, 15-12.

Table 1 shows that for each computer run a game was most likely to end with
u pegs remaining. Many games ended with 3 pegs on the board. It was more difficult
to play a game leaving only two pegs* and it was yery difficult to find a sequence
of moves solving the puzzle.
By changing the initial location of the hole one can, using the same computer
program, attempt to solve fourteen additional puzzles. The symmetric nature of
the peg board allows a reduction in the number of puzzles that must be solved to
determine a solution for each of the fourteen remaining puzzles. For example, a
puzzle with the initial hole in position 4 is symmetric to that with the initial
hole in position 2. Looking at Figure 2 and the solution given above, one can
quickly write down a solution to the puzzle with initial hole in position 2. It
is 4-2, 10-3, 2--49 12-3, 14-7, 4-2, 1-10, 13-6, 5-12, 2-11, 12-10, 6-13, 15-10.
Note by observing Figure 2 that each move in the new sequence is symmetrical
to that given in the previous sequence. Therefore, to show that each of the four-
teen additional puzzles has at least one solution it is only necessary to repeat
the computer run for configurations with the initial hole in positions l, 3, 6 9 7,
10, 11, 13, and 15 of Figure 2. Table 2 shows the result of those runs playing
2000 games for each initial configuration.

Table 2

Initial hole Gaines ending with indicated no. of pegs remaining (% of total)
position
1 2 3 4 5 6 >7

1 0.60 6.25 24.95 39.75 12.00 11.15 5.30

3 0.35 7.80 31.15 34.65 12.55 5.85 7.65
6 0.35 3.65 29.40 43.55 15.55 2.65 4.85
7 0.05 4.25 14.40 46.05 20.75 10.95 3.55
10 0.80 6.90 28.05 43.05 10.90 4.80 5.50
11 0.15 4.10 19.75 43.55 15.65 4.60 12.20
13 0.25 4.10 25.25 46.70 13.15 7.40 3.15
15 0.30 6.30 31.30 36.20 1^.25 2.50 9.15

Table 2 shows that each of the remaining fourteen puzzles has at least one
solution. The percent of games ending with l, 2, 3,... pegs left on the board has
the same general distribution as that in Table l, For each configuration less than
 - 216 -

l0' of the 2000 games played ended in solutions. Solutions using the Monte Carlo
techniaue were found to all fifteen variations of the puzzle.
Althouqh the Monte Carlo technique was used to find solutions to each variation
of the puzzle9 one must proceed differently to find all possible solutions to any
aiven variation of the puzzle. A systematic approach similar to that discussed in
fil could be used. The computer program would be more difficult to construct than
that involvina the Monte Carlo method. All combinations of moves that could lead
to a solution must be considered. Techniques must be devised to quickly eliminate
any combination that cannot result in a solution. Such techniques are necessary
to reduce the required computer time,

REFERENCES
1. Alan 6. Henney and Dagmar R. Henneys Systematic Solutions of the Famous
15-14 Puzzles9 Pi Mu Epsilon Journal* Vol. 6 9 No. 4 (Spring 1976) 197-201.
2. Lynn Arthur Steens Solution of the Four Color Problem9 Mathematics Magazine,
49 (September 1976) 219-222.

Department of Mathematics, The George Washington University, Washington, D.C. 20052.

ft & *

A riddle
What goes and goes and neyer reaches the door?

Answer: Footies on a Mobius strip.

ANDREJS DUNKELS,
University of Lulea, Sweden.
 - 217 -

ALGORITHMS AND POCKET CALCULATORS:

SQUARE ROOTS: 111
AND THE MNIMAX METHOD
CLAYTON W. DODGE

Many functions are accurately approximated by polynomials; generally, the higher

the degree of the polynomial the more accurate is the approximation. As an example,
suppose we wish to write a linear approximation to the function / of Figure l defined
on the interval [0,1]. A line through the endpoints of the graph9 such as y - gix)
in Figure 2, would be one such approximation. Here gio) = fio) and g(±) = f i i ) 9 and
there is a point x =a in Co,1] such that the error eix) = fix) -g(x) has maximum
magnitude. We search for that line y - hix) such that \eix)\ has as small a maximum
value as possible; we wish to minimize the maximum error on the stated interval.
Hence the process is called the minimax method.

y = fix)
/

Figure l

In this case, by lowering the line y = gix) of Figure 2, we decrease the error
\ z i o ) \ and increase |e(o)| and |e(l)|. When these three quantities are equal, as
shown by the line y = hix) of Figure 3, we have attained our goal; the error six) -
fix) -hix) has the smallest possible maximum value, and this maximum error is attained
at least three times: at both endpoints and at one intermediate point. We note that
between these points are two values p and q such that the error is zero: f(p) = hip)
and fiq) = hiq).
Rather than working with the pair of simultaneous equations
fip) = hip) and fiq) = hiq)
and trying to determine what linear polynomial hix) minimizes the maximum error, we
 - 218 -

generally encounter less work to let e = e(o) be that maximum error (or its negative)
and solve the trio of equations
/(0) = ft(o)+e, f(o) = Hc)-e9 f(l) = h(l)+e.

This is the technique of the minimax method. Observe that the error term changes
sign from one rvalue to the next, as one generally expects would occur.
More generally, an approximating polynomial h(x) of degree n would attain its
maximum error n + 2 times. The big question with the minimax method is how to deter-
mine the intermediate c-values at which the maximum error is realized; we need n such
values for a degree n approximation. Although algebraic methods may be availables
the usual techniques except when n is quite small, is trial. Guess at the locations
of flp c29..., cn3 and solve the resulting n + 2 (linear) equations for the n + 2
constants* which are e and the rc+l coefficients of the polynomial h ( x ) . Then calcu-
late the actual error at sufficiently many ar-values to locate all points of relative
maximum error (both positive and negative). Now recalculate the coefficients using
these new rvalues for the rvalues* Usually only two or three repetitions are required
to obtain a very good (almost) minimax approximation.
As a specific example, according to Fike Ci], in the IBM Fortran library appears
a square root algorithm based on Newton's method. To find 7x where x = m • 16 is a
base 16 numeral with 1/16 < m < i, assume a first approximation
k/2
Pj = 16 p(m) if k is even
or
r = a
i ^ I6 (fe+1)/2 p(m) if k is odd,

where p(m) = (2 + 8m)/9 is the minimax linear approximation to Jm on the interval

[1/16,1] whose maximum relative error has been minimized. This algorithm has the
advantage that, to obtain double precision square roots, one need only perform the
divide»and»average routine (see [1978? 154]) just one extra time.
The reader may enjoy verifying that p(m) = (2 + 8w)/9 is indeed the linear mini-
max approximation claimed. The maximum relative error

occurs at the endpoints 1/16 and 1 and also at the interior point 1/4.
Pocket calculators do not utilize base 16, but rather binary-coded decimal (BCD)9
in which each decimal digit is separately written in binary notation. Thus, in BCD9
 - 219 -

we have
239 = 0 0 1 0 S 0 0 1 1 9 1 0 0 1 B C D S
whereas in base 2,
239 = 11101111 .

It is much easier to program a calculator to work in BCD 9 even though more bits are
required, because conversion to decimal for display is very easy and absolutely
accurate. Hence, if we are to consider the calculation of square roots on a cal-
culator,, we need work only in base ten.
Consider applying Newton's method to calculator square roots. For Jx9 take
x - m71••1010 with k even and l < m < ioo. Then our first approximation would have
the form
rx = 10 p O ) ,
where p(m) is an approximation to Jm on the interval [1,100]. Suppose we take p(m)
linear and minimize the maximum relative error for best operation of Newton's method;
say
p (777) = am + b .
If e denotes the maximum relative error (or perhaps its negative) 9 then we have
/ T = l = a + £> + £ 9 (1)
Jo = ao + b - JQZ , (2)

/Too" = 10 = lOOa + b "- lOe (3 s)

to solve simultaneously for a , b , and e. The general equation for the relative
error ER(m) is

Jm Jm
In order that the intermediate value m - c be a relative maximum of \EE(m)\, we must
have the derivative of ER(m) equal to zero. Thus

2mJm

so ERl (m) = o at m - b / a ; that i s 9 c - b i a . From ( l ) and (3) we next get

lOe = 10 - 10a - 102? = 10 - 100a - b,
so b - 10a and e = I0a/a = 10. Now (2) yields
 - 220 -

/To" = 10a + lGa - vToCl - a - 1 0 a ) ,

so
2/To 2
0.115443,
20 + 11/10 2v / 10~+li

20
10a 1.15443,
2i/To + l l
and
22
e = 1 - « -0.2699.
2vT0 + l l
Taking p(w) = 0.115443O + 10), we arrive at the following values and graph for the
r e l a t i v e error (see Figure 4 ) :

m p(m) V^7 ER(rr)

1 1.2699 1.0000 + .270

2 1.3853 1.4142 -.020
5 1.7316 2.2361 -.226
10 2.3089 3.1623 -.270
20 3.4633 4.4721 -.226
50 6.9266 7.0711 -.020
90 11.5443 9.4868 + .217
100 12.6987 10.0000 + .270

Figure 4

When the error is negative (when r < /x) , the convergence is not quite as rapid for
the divide-and-average method as when rx > v^Fs as the algebraic proof showed [1978:
157]. So applying the technique to the extreme case m = lo § we get p(io) = 2.3089 and

rx = 2.3089,
r2 = 3.319983371,

T = 3.1660233259
3
r = 3.16227987588,
r, - 3.16227766016,
the last of which is a satisfactory lo-digit approximations even for those lo-digit
 - 221 -

calculators that work internally with 12 or 13 digits.

Curiously, even if we skirrp somewhat on the initial accuracy^ we can still
obtain just as qood a final result. If we round our initial estimate to
Pl(m^ = 0.12(777 + 10),
then ER(l) = EE(loo) = 0.32 and ER(lo) = -o.24 s increasing the maximum relative
error when it is positive but decreasing its absolute value when the relative error
is negative. Here again four iterations always suffice for full lo-digit calculator
accuracy. For the extreme cases /F and vfo, we have
rx = p 1 ( l ) = 1.32 and r 1 = p.^10) = 2.4
v - 1.038787879 i» = 3.283333333
2 2
r, = 1.000724161 r = 3.164509306
3 3
r. = 1.00000026201 r, = 3.16227844705

rc = 1.00000000000 v = 3.16227766016.
5 5

It is reasonable to ask whether a significant improvement would result if a

higher degree polynomial first approximation to Jm on [i,ioo] were used. Let us
try a quadratic polynomial
q(m) = am2 -tbm + c*

We may pick four points on !!i,ioo] at which the maximum error is to occur: the two
endpoints l and loo and two intermediate values, say 9 and 36, Assuming the maximum
relative error is e 9 we must have
1 = q(l) - e = a + b + o - e,

3 = q(9)+3£ = 81a + 9& + e + 3e,

6 = q(36) - 6e = 1296a + 36b +e - 6e,

10 = <7(10)+10e = 10000a + 1002? ^o + lOe.

With the aid of a calculator and careful algebra, we find that

a = -0.0012626, b = 0,20580819 o = 0.9090909,

and the maximum relative error e is

e = 0,1136,

By constructing a table of values for the relative error

 - 222 -

an2 +hn + Q - \/m

lull \m ' - 9

vri
we locate the extreme values
ER(1) = 0.1136, EP(H.85) = - 0 . 1 4 7 4 9

Eff(49.5) = 0 . 1 3 7 5 , and Eff(lOO) = - 0 . 1 1 3 6 .

So our original choice of 9 and 36 for intermediate values was incorrect9 as

expected9 and we should use new choices, say 5 and 50.
There seems to be little benefit from further work in this directi on 9 however,
since e = 0.1136 is not sufficiently better than the maximum relative error obtained
from the linear approximation; four iterations of Newton's method are still required
and the details of the quadratic polynomial would consume much more memory space in
a calculator. Hence we drop the search for a higher degree approximation to i/m.
As an aside comments if we do find the new approximation q1(m) using 5 and 50
for intermediate values, we get
ql(m) = -0,0012981m2 +0.206932m + 0,9301780

with e = 0.1358* This approximation is quite close to the desired minimax quadratic;
extreme error occurs at l, 4.95§ 48.2s and 100, the relative error at 48.2 being
0.13629 sufficiently close to the 0.1358 so that further refinement seems unncessary.
Thus we have found a simple, fast technique for finding square roots by Newton's
divide-and-average method starting from a modified minimax linear first approximation.
Whether this technique or one similar to it is actually used in a calculator is
unknown to me. Such information is a closely-guarded industrial secret, as I have
been informed by one well-known calculator company. Surely the method seems reasonable.
As we shall see next, however, there is a better method that seems to have gained
wide acceptance.

REFERENCE
1, C.T. Fike? Computer Evaluation of Mathematical Functions, Prentice-Hall ,
19689 pp. 75-76s 929 111-112, 1579 1699 178, 207.

Mathematics Department, University of Maine, Orono, Maine 04469.

ft ft ft

Every theorem in category theory is either a pushover or its dual—-a put-on.

From Category Theory, by Herrlich

and Strecker, Allyn and Bacon, 1973.
 - 223 -

A FORMULA FOR POLYTOPES ?

DR. A.H. NOEL1

For o<i<ns let F^ denote the number of -^-dimensional faces of an n»simplex.

We have the binomial expansion

(1 +x) = 1 + ) F.x ^ + 1
i=o %
and setting x = -l gives

o = i + y (-I)^+1F.

or
n
I (-l)V. = 1. (1)
^
t=0

Now let F. be the number of t-dimensional faces of an n-dimensional hyper-cube.

Here we have the binomial expansion
(2+xf = I F.x
%
i=0

and setting x = -l yields again (1).

Now (l) is known to hold for all simply-connected polytopes for n - o, l, 2 S 3.
If n = 3 s for example9 (l) reduces to (note that F3 =l)
F0~F1+F2 = 2, (2)

a well-known formula due to an obscure French mathematician named Relue who never
received proper credit for it. If it holds as well for n > 3 9 then (l) would provide
a nice generalization of (2). But what in the (n-dimensional) world would (3+x)
n-2
(4 +x) ,... represent?

Our Own Mother Goose

Jack be nimble,
Jack be quick,
If you are going to check
The addition at McDonald's.
CLAYTON W, DODGE

This note was found among the unpublished papers of the late Dr, Noel by A, Liu,
of University of Alberta, a distant relative of the good doctor.
 - 224 -

PROBLEMS — PROBLEl^ES
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 submitted
without a solution.
Original problems are particularly sought. But other interesting problems may
also be acceptable provided they are not too well known and references are given as
to their provenance. Ordinarily, if the originator of a problem can be located, it
should not be submitted by somebody else without his permission.
To facilitate their consideration, your solutions, typewritten or neatly
handwritten on signed, separate sheets, should preferably be mailed to the editor
before January 1, 1979, although solutions received after that date will also be
considered until the time when a solution is published.
371B Proposed by Charles W. Triggs San Diegos California.
In the following skeleton multiplications each letter uniquely replaces
a digit in the decimal system. Reconstruct the multiplication.

EASY
M A T H E M A T I C S
M ± * * s

A * * *C
H ***i
S ± * * j
Y * * *A
I -k -k * M
J * * *£
C * * *H
s * * * j

Y * * *A
j * * * ^

372 i Proposed jointly by Steven R. Conrad3 Benjamin N„ Cardozo H,S.S Bay side,
N.Y.; and Gilbert W. Kesslers Canarsie H.S.S Brooklyn, N.Y,
A triangle ABC has area 1. Point P is on side a 9 a units from B\ point Q is on
2?9 B units from C; and point R is on c, y units from A. Prove that, if a/a9 B/b, and
y/e are the zeros of a cubic polynomial / whose leading coefficient is unity, then the
area of A PQR is given by f(l) - /(o).
 - 225 -

37i» • Proposed by Leroy F. Meyers3 The Ohio State University.

Suppose that the human population of the Earth is increasing exponentially
at a constant relative rate k, that the average volume of a person stays at VQ$ and
that the present population is NQ. If people are assumed packed solidly into a sphere,
how long will it be until the radius of that sphere is increasing at the speed of
light, c5 and what will the radius of the sphere be then?
The following approximate data may be used: NQ = 4 x i o 9 , k - 1%/yr, 1 yr = 365,25
da, l da = 24-60-60 sec; and, in English units, VQ = 4 ft 3 and c = 186300 mi/sec,
whereas in metric units V0 = o.l m 3 and c = 3 * 1 0 8 m/sec.
(I heard of this problem several years ago. It must be a well-known bit of
mathematical folklore.)

3747 Proposed by Sidney Penners Bronx Community Colleges New York,

Prove or disprove the following

THEOREM. Let R be the set of real numbers and let the function f: R~*R be such
that f"(x) exists^ is continuous and is positive for every x in R. Let Pj and P2
be two distinct points on the graph of fj let Lx be the line tangent to f at ?1 and
define Lz analogously. Let 0 be the intersection of L and L and let S be the
intersection of the graph of f with the vertical line through Q. Finallys let Rx be
the region bounded by segment P 0, segment SQ and arc PjS, and define R2 analogously.
Ifj for each choice of P and P2J the areas of Rx and Pz are equals then the graph
of f is a parabola with vertical axis.

3751 Proposed by M.S, Klamkins University of Alberta,

A convex n-gon P of cardboard is such that if lines are drawn parallel
to all the sides at distances x from them so as to form within P another polygon
P\ then Pl is similar to P. Now let the corresponding consecutive vertices of P
and p' be A 2 , A 2 ,..., A n and AJ, A^,..., A^, respectively. From A', perpendiculars
A B ? A B
2 i 2 2 a r e d r a w n t 0 A A 9 A 2 A 3 S respectively, and the quadrilateral A^B J A 2 B 2 is
cut away. Then quadrilaterals formed in a similar way are cut away from all the
other corners. The remainder is folded along A{A 2 , A^A•,..., A^AJ so as to form an
open polygonal box of base AJA£...A^ and of height x. Determine the maximum volume
of the box and the corresponding value of x.

376, Proposed by V.G. Hobbes3 Westmount3 Quebec.

Isosceles triangles can be divided into two types: those with equal
sides longer than the base and those with equal sides shorter than the base. Of
all possible isosceles triangles what proportion are long-legged?
 - 226 -

377 8 Proposed by Michael W. Eckers Pennsylvania Stave University,

For n = i, 2, 3 s . a a , let f(n^ be the number of zeros in the decimal
representation of n, and let

rc=l n^

Find the domain of F 9 that is, the real values of p for which the series F(p)
converges.
(This problem was suggested by Problem E 26759 American Mathematical Monthly,
84 (October 1977) 652.)

3781 Proposed by Allan Wm, Johnson Jr.s Washingtons B.C.

(a) Find four positive decimal integers in arithmetic progression,
each having the property that if any digit is changed to any other digit, the
resulting number is always composite.
(b)* Can the four integers be consecutive?
(This problem was suggested by Problem 1029- in Mathematics Magazine5 51
(January 1978) 69.)

5791 Proposed by Peter Arendss Algonquin College3 Ottawa,

Construct a triangle ABC 9 given angle A and the lengths of side a and
t (the internal bisector of angle A ) .

380. Proposed by G.P. Henderson^ Campbellcrofts Ontario.

Let P be a point on the graph of y - f(x)$ where f is a third-degree
polynomial, let the tangent at P intersect the curve again at Q, and let A be the
area of the region bounded by the curve and the segment PQ. Let B be the area of
the region defined in the same way by starting with Q instead of P. What is the
relationship between A and B?
A »»J J.

PUBLISHING NOTE
We would like to remind our readers of the very fine series of booklets (« 90
pages each) entitled 1001 Problems in High School Mathematics published from time
to time by the Canadian Mathematical Society. They are edited by the well-known
Canadian problemists E. Barbeau, M, Klamkin, and WB Moser. Each booklet contains
150 problems and the solutions to 100 of them. Book 1, Book 2, and Book 3 are now
available, the last of which was published quite recently and contains several
problems from this journala For more information see [1977: 90] or write to
Canadian Mathematical Society,
3421 Drummond St., Suite 15,
Montreal, Quebec, Canada H3G 1X7.
 - 227 -

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.

305 a [1978: 11, 180] Late solutions CARL W* NORRIS, Memorial University
of Newfoundland.

3151 C1978: 35] Proposed by Orlando Ramoss Havana, Cuba.

Prove that9 if two points are conjugate with respect to a circle* the
sum of their powers is equal to the square of the distance between them.
Solution by Kesiraju Satyanarayanas Gagan Mahal Colony3 Hyderabad3 India.
Let y be a circle whose equation we assume without loss of generality to be

x2 +yz - r1 - 09
and let Pl(x1,yl) and P2(x29y2) be two points distinct from the origin. Since the
polar of a point ?(x\y%) * (0,0) with respect to y is the line

it follows that P2 and P2 are conjugate points with respect to y if and only if

Now the powers of P and P2 with respect to y are respectively

x
Pi =
i +
^i2 " p 2 and
p2=x2+y2-rz;
,
hence the following statements ^2)-(5 i are all equivalent the last being a consequence
of (l):
P^P2 = iP^al 2 . (2)
2 2 2 2 2 2
(xf+yf-r ) + (x +y -r ) = (xx -x2) + (y x - y 2 ) * (3)

x x =
i 2 ^y^i *2* ( 4 )

Px and P 2 are conjugate points with respect to y. (5)

The equivalence of (2) and (5) proves the stated theorem as well as its converse.
An analogous result clearly holds for a sphere.

Also solved by LEON BANROFF, Los Angeles, California; ROLAND H. EDDY, Memorial
University of Newfoundland? G.C. GIRI, Research Scholar, Indian Institute of
Technology, Kharagpur, India; JIM MATTICE, Parkside H.S., Dundasf Ontario? DAN PEDOE,
University of Minnesota? SAHIB RAM MANDAN, Indian Institute of Technology, Kharagpur,
India (two solutions)? HARRY D. RUDERMAN, Hunter College, New York; and the proposer.
 - 228 -

Editorfs comment.
Pedoe mentioned that this problem is very, very old, and Ruderman located it
in Johnson [l].
One solution was less than satisfactory because the solver confused the notions
of conjugate and inverse points; so it may be useful to recall that every point on
the polar of a point P 9 not only its inverse point, is conjugate to P.
A synthetic proof is not hard to find, and several of the solutions submitted
were of this type. But9 as frequently happens with synthetic proofs, in order to
have a complete proof one must tiresomely consider several cases: whether, for
example9 any of the points Px and P 2 lie on the circle, or whether one or both of
them lie outside the circle. The analytic solution we have featured is valid in
eyery case* and for this we are grateful to Monsieur Descartes.
REFERENCE
1. Roger A. Johnsons Modern Geometry, Houghton Mifflin Co. 9 Cambridge9 Mass.s
19299 p. 105 9 Corollary d.
ft ft ft

316 8 [1978: 36] Propose par Hippolyte Charles3 Waterloo3 Quebec.

DSmontrer 1 ' i m p l i c a t i o n

a -x
• ••.,l.. l ,.~.-y„. ™
a~d c-y
• y ^ u ^ * ^ S mmmwM-
^a-y
=SS=£» • , , . . .im~&L. ~
a-d G-X
J. 8 j „

x-b b-Q y -d y -h b -c x -d
I. Solution by the late R. Robinson Rowe5 Sacramento, California.
We assume that both equations are meaningful, that is 9 that no denominator
vanishes in either one.
The first equation is then equivalent to

f(x,y) = 09 (1)
where
f(x9y) = (a-x)(b-c)(y -d) - (x-b)(a - d)(c-y)
= (a - b + c - d)xy + (bd -ac)(x + y) + bc{a - d) - ad{b - c) 9
and the second is equivalent to
f(y,x) = o. (2)
Since / is symmetric in x and y% it follows immediately that C D <£==» (2).
II. Solution by Eerta T, Freitag, Roanoke, Virginia.
If we assume that no denominator vanishes in either equation, then the first
 - 229 -

eauation is equivalent to

(b-c)(a-x)(y -d) = (a - d){x - b)(c - y),

and subtracting this from the identity

(b -c)(d-aHx-y) E (a - d)(c - b){x - y)

yields
(b-o)(x-d)(y -a) = (a-d)(c-x)(b-y)t
which is eauivalent to the second equation.

III. Solution by David C» Kays University of Oklahoma.

Note that the given equation is equivalent to

x - a m c -b _ y - c @ a - d
x -b a-a y -d a-a
and also to
x ~ a d -b _ y - o b - d
x -b d - a y - a b - a
Letting a* & 9 c9 d9 x 9 and z/ be the coordinates of the points A 9 B 9 C 9 D 9 X 9 and Y
on some line* each of the above equations yields an equation in cross ratios:
(XC9AB) = (YA9CD) = X
(XD9AB) = (YB9CD) = p.
Since the same permutations of points will yield the same function of A (and of y ) ,
we obtain
(XB9CA) = (YD9AC) = Pia^ Cn
(XB 9 AD) = (YD 9 CB) = p2(\i). (2)
Multiplying in ( l ) and (2\
(XB9CA)(XB9AD) = (YDSAC)(YDSCB)S
Therefore
x-c b-a x -a b -d y-amd - c y- c d - b
x-a b-c x-a b-a y~o a - a y-b a-c

x-c b-d y- a d-b

a: - a 0 - e? z/ - 2? a - a

x - o a - d y -a

which is equivalent to the desired eauation.

 - 230 -

Also solved by LEON BANKOFF, Los Angeles, California; W.J. BLUNDON, Memorial
University of Newfoundland; RICHARD BURNS and KRISTIN DIETSCHE, East Longmeadow
H.S., East Longmeadow, Massachusetts (jointly); CLAYTON W. DODGE, University of
Maine at Orono; G.C„ GIR1, Research Scholar, Indian Institute of Technology,
Kharagpur, India; ALLAN Wm„ JOHNSON JR., Washington, D.C.j F.G.B. MASKELL, College
Algonquin, Ottawa; HERMAN NYON, Paramaribo, Surinam; HARRY D. RUDERMAN, Hunter
College, New York; KESIRAJU SATYANARAYANA, Gagan Mahal Colony, Hyderabad, India;
and KENNETH Sa WILLIAMS, Carleton University, Ottawa.

Editorrs comment.
Solution III, while not the simplest, is the most illuminating in that it
shows one way in which implications such as the proposed one can be generated.
* * *
r
317s 1978: 36] Proposed by James Gary Propps Great Necks N,I.
In triangle ABC 9 let D and E be the trisection points of side BC with D
between B and E 9 let F be the midpoint of side AC 9 and let G be the midpoint of side
AB. Let H be the intersection of segments EG and DF. Find the ratio EH : HG by means
of mass points (see [1976: 55]) or otherwise.
I. Solution by Richard Burns and Kristin Dietsche3 East Longmeadow R>S.3 East
Longmeadow3 Massachusetts (jointly).
Since GFljBC (see Figure l ) , triangles FGH
and DEH are similar; hence

EH DE |BC
HG GF 4BC

II. Solution by Harry D. Ruderman3 Runte:

College3 New York.
More generally, suppose
BD : DE : EC = u : 1 : v9
AG : GB = 1 : r9
AF: FC = l : s 9
(see Figure 2 9 where the numbers 1, r$ s, u, v
indicate r a t i o s , not lengths). It will be
convenient to assign weights r s , s+x9 y+r to
A 9 B 9 C 9 respectively. These weights can be
replaced either by
D
(s+x) (x+ii+s^ {x+h^r^
Figure 2
 - 231 -

rs+s at G and x+y+r at E or by rs+r at F and x+y+s at D; and with either of these
weight assignments the center of mass is at H. Now the following relations hold:

x(u + 1) = v(y+r), u(s+x^ = y(v + l)9

and subtracting these equations yields x + y = rv + su. Hence

EH _ rs + s _ sir + 1)
HG ~ x+y+r ~ r(v + 1) + su
and
DH _ r s t r _ r(s + 1)
HF ~ x +y + s ~ s(u + 1) +rv
In particular, for the problem at hand we have r=s=u=v=ls and EH : HG = 2 : 3 .
Also solved by LEON BANKOFF, Los Angeles, California; STEVEN R. CONRAD,
Benjamin N. Cardozo H.S., Bayside, N.Y.? SAHIB RAM MANDAN, Indian Institute of
Technology, Kharagpur, India? W.A. McWORTER JR., The Ohio State University?
HERMAN NYON, Paramaribo, Surinam? FREDERICK ROTHSTEIN, New Jersey Department
of Transportation, Trenton, N*J.; the late R. ROBINSON ROWE, Sacramento, California?
KESIRAJU SATYANARAYANA, Gagan Mahal Colony, Hyderabad, India? HARRY SITOMER,
Huntington, N.Y.; CHARLES W. TRIGG, San Diego, California? and the proposera

Editor fs comment.
The problem as proposed could be solved easily by traditional methods, as
shown in solution I. But the generalization of solution II could hardly be
solved more efficiently than by the method of mass points, as shown above.
For those who need a refresher course in the method of mass points, see
[1976: 55],

318, C1978: 36] Proposed by C.A. Davis in James Cook Mathematical Notes
No, 12 (October, 1977J, p. 6,
Given any triangle ABC, thinking of it as in the complex planes two points L
and N may be defined as the stationary values of a cubic that vanishes at the vertices
A 9 B s and C. Prove that L and N are the foci of the ellipse that touches the sides
of the triangle at their midpoints* which is the inscribed ellipse of maximal area.
I. Solution by Kesiraju Satyanarayana3 Gagan Mahal Colony3 Hyderabad3 India
(revised by the editor).
We will make use of the following lemmas, which describe known properties of
the ellipse:
LEMMA 1. Let ATX and AT2 be tangents to an ellipse with foci F and F2 (see
 - 232 -

Figure 1). Then

(a) Z F / T 1 = /F 2 AT 2 ;

(b) /AF 1 T 1 = ZAF a F 2 and / A F 2 T 2 = ZAF 2 F X .

This is a theorem of Ponce!et. For a proof9 see T2"1.

LEMMA 2. If Fx and F2 are £ne /bci of an ellipse

inscribed in A ABC^ then Fx and F2 ape isogonal
conjugate points with respect to A ABC. Conversely3
if Fj and F2 ape isogonal conjugate points with
respect to A ABC., tfoere ts a unique inscribed ellipse
Figure 1
with foci Fx and F 2 . (See Figure 2.)

Outline of proof\ The first part follows from Lemma 1

(a) and the fact that the isogonal conjugates of AF l S BF l S
CF, are concurrent in F2 (see fi"i). For the converse9
given Fx and F2 we can construct T 1 9 T 2 9 T 3 by Lemma
t(b)\ and the unique ellipse through T with foci
Fj and F2 will be inscribed in AABC 9 the points
of tangency being T 1 9 T 2 9 T 3 .

LEMMA 3. For a given triar^gle^ there

is a unique inscribed ellipse whose centre
lies at the centroid of the triangle. It
is the inscribed Steiner ellipse3 which
touches the sides of the triangle at their
midpoints and is the inscribed ellipse of Figure 2
maximal area.
This is a well-known result. For a proof. see [3].
Consider now the triangle of the proposal. Let the affixes of A 9 B, C 9 L 9 N
be a9 b9 c, z. z . The cubic that vanishes at the vertices is then defined by

f(z) = (z -a)(z -b)(z -e),

and we have
f'(z) = (z -b){z - c) + (z - c)(z -a) + (z -a)(z -b)

= 3s" 2(a+b + c)z + be + ca + ah (1)

= 3 ( 3 - 3 j }<Z z2\
 - 233 -

Now
f{a) = (a-b)(a-a) = 3(a - 24 )(a - s 2 );
hence

AL AN " a -Sj ° a - s 2
Thus the sum of the amplitudes of AB/AL and AC/AN is zero9 so AL and AN are isogonal
conjugates. Similarly BL and BN9 and CL and CN S are isogonal conjugates, and so L
and N are isogonal conjugate points with respect to A ABC. It now follows from
Lemma 2 that L and N are the foci of a uniquely determined inscribed ellipse. From
(l), the centre of this ellipse has affix
z + z
i z _ a+b +a .
=
2 3

hence the centre lies at the centroid of the triangle and9 from Lemma 3, the ellipse
is the inscribed Steiner ellipse, as required.
II. Solution by 0. Bottemas Delfts The Netherlands.
This problem is well-known in Dutch mathematical literature. As far as I know
it was first proved by van den Berg [5] and other proofs were later given by Scheffer
[4] and van Veen [61. Here is still another proof.
Let AJA^Ag be a triangle in the complex plane; BJS B 2 the zeros of the derivative
of the cubic whose zeros are the vertices of the triangle; and El the inscribed
Steiner ellipse* which touches the sides of the triangle at their midpoints. It is
well-known that E% is the inscribed ellipse of maximal area and that its center
coincides with the centroid of the triangle.
All displacements in the complex plane are transformations of the form
z ' = as + 3, (2)
where a and 3 are complex numbers with |a| = 1 , and there is always a displacement
(2) which transforms E* into an ellipse E with standard equation

Under (2), A{, A£, A^ 9 BJ9 B^ are transformed into A x , A 2 , A 3 , B x > B 2 , where B l 9 B 2
are the zeros of the derivative of the cubic with zeros at A 2 , A 2 , A 3 ; and E is the
inscribed Steiner ellipse of triangle AjA2A .
The transformation x ' - a~lx? y l - b~ly transforms E into the unit circle c\ it
is an affine transformation with the property that tangents remain tangents and
 - 234 -

midpoints remain midpoints. The triangles with o as their inscribed Steiner

ellipse are the equilateral triangles inscribed in the circle C with center 0
and radius 2. Hence the vertices of the triangles with E as their Steiner ellipse
are
A- = (2a cos 4),9 2fc sin <(>*), k = l, 2 9 3

w i t h $!=$ ( a r b i t r a r y ) , 4>2 = <f)+2iT/3 9 <J>3 = 4> - 2 T T / 3 . Thus

£ cos <j>, = Z s i n <|), = o 9

which confirms that the centroid coincides with the center 0 of E. Furthermore,

E cos 2 4>fc = l sin 2 $k = | ± i I cos 2 ^ = |

Consider now the cubic equation whose roots s- are the points A-, so that

z« = 2a COS 0. + 2ib$it)$.9 k = 1, 2 9 3.
The equation reads
s 3 + dxzz + <i23 + <23 = 0 9 (3)
with
f
! = -^fc = °
and
d2 = s 2 s 3 +S30J + s 1 s 2 r - j(s 2 + s 2 + s 2 ) = -2 £ (a cos <J>* + ££ sin <J>, )
Since
2 E cos <j>. sin $, = Z sin 20, = o,

we obtain
3 3
d, = -2|^a^22 -4Z>
T , 2Z |
I.
and (3) becomes
zz -3(a 2 -2>2 )*+<*, = 0, (4)
2
The roots of the derivative of O ) are s = ±/a - M ; they coincide therefore
with the foci of E«
Editor's comment.
A solution of this problem* which included a proof of Lemma 3 in our solution I,
was published in James Cook Mathematical Notes No. 14 (January 1978). It was by
Basil C. Rennie9 the editor of JCMN,
 - 235 -

REFERENCES
1. Nathan Altshiller Court, College Geometry, Barnes and Noble, New Yorks
19529 pp. 269-270.
2. F.G.~^!.S Cours de Geometries Librairie Generale9 Paris, 1922s pp. 439-440,
3= F.G.-M., Exercioes de Geometries Mame & Filss Tours, 1907, p. 1059.
4. M. Scheffer9 Mathematical hi (1938-39) 285-288.
5. F.J. van den Berg9 Over het verband tusschen de wortels eener verge!ijking
en die van haar afgeleide (On the relationship between the roots of an equation and
those of its derivative), N. Arch. v. wisk* , IX (1882) 10.
6. S.C. van Veen9 Mathematical A8 (1939-40) 14-16.
i'z «k «

3191 [1978: 36] Proposed by Leigh Janess Rocky Hills Connecticut.

If a solution of the following type has not yet occurred in a student
paper, it soon will. "Cancelling the exponents" yields

37 3 + 13 3 _ 37 + 13 _ _5£
37 3 + 24 3 37 + 24 61
which is correct.
Find necessary and sufficient conditions for the positive integer triple
(A9B9C) to satisfy
A 3 + B3 _ A + B
A3+C3 A+C
I. Solution by David Y.arrs student s Bronx High School of Science9 Bronxs A 7 .7.
For positive A , B, and C (integers or not), we have
3
iLtll = d±£ ^ (i4 + £ 7 ) M 3 + B 3 ) = U+B)(A +C3)
A3+C3 A+C
<=> A2-AB + B2 = A2-AC + C2

«==» (B -C)(A - B -C) = 0;

so the required necessary and sufficient condition is:

B =c or A - B + c.

II. Comment by Allan Wm. Johnson Jr.s Washingtons B.C.

There is ample equipment in the high school algebra toolbox to construct more
examples of "cancelling the exponents".
(a) To find positive integers A9 5, c9 D such that
 - 236 -

As +g 3 = A+B
C* +D3 C+D *

we first clear fractions to obtain the equivalent condition

A2 -AB + B2 = C2 -CD + D2 . (1)

Next we call upon the following identity Ci]: if either

2» = mp -nq9 s - np + (m - n)q or r = mq -np, s = nq + (m - w)p9

then
O 2 -mn -k-n2)(p2 - pq + q-2 ) = r 2 - vs + s 2 .

This furnishes the following parametric representation of solutions of (l):

A - mp -nq, B ~ np + (m - n)q 9 C - mq -np, D = nq + (m - n)p.

For example, for m - 4 9 n = l, p = 2 8 <? = 3 9 we get

5^ + ll 3 _ 5 + 11
10 3 + 9 3 10 + 9 *

(b) For
A2 +ff2 _ A +B
A2 +C2 " A+C '
we assume B * C to eliminate trivial solutions and clear fractions to obtain the
equivalent condition
A2 - (B + C)A -BC = 0 9 (2)
from which
A = ~ LB + C ± /(B + C)2 + 45C ] .

To make the radicand a perfect square, we seek solutions of

s2 = (B + C)2 + ^BC = (5 + ZC)2 - SC2 ,

an equation of the form s 2 + 2 t 2 = u2 to which we apply the following identity [2]:

(p2 -2q2)Z+2(2pq)2 = (p2+2q2)\

This yields the following two parametric representations for solutions of (2):
A = pip -q) A = q(2q -p)
B - (p -q)(p -2q) and B = (2q -p)(q -p)
C ~ pq C - pq

For example9 if we use p = 3, q-l in the first representation and p = l 9 q - 3

in the second, we get
 - 237 -

6 Z + 2^ 6+2 a . 152+102 15 + 10
a 2 = — and = — —
62+32 6+3 152+3Z 15 + 3
(c) To find solutions of
A2 + g 2 _ A+B
C2+D2 C+D
(3)

we f i r s t set k = U 2 + £2)/G4 + B), which is equivalent to

^ 2 -kA+B(B -k) = 0g
giving

Proceeding as in (b) to make kz + ^kB - ^B2 = (k + 2B)2-8B2 a square* we find

A = p(p - q), B = pq9 k = p 2 - 2pq + 2^ 2 .

Solving the last equation for p gives p = <y + Jk -q%, which requires fc = <?2 +rz.
We note that if k can be expressed as a sum of two squares in two ways 9 then two
different pairs (p9q) result: a pair (pl»q1) to solve (A2 + B2)/(A + B) = k and a
2 2
pair (p2,q2) to solve (C + D )/(C + D) = k. Using
fc = O 2 + n 2 )(sz + t 2 ) = O s + n t ) 2 + (mt ™ns) 2 = (mt +ns)z + (ms - nt)2 9

we obtain the following parametric representation of solutions of (3):

A = p1(p1 - qt) p = mis + t) - n(s - t)

B = p1ql q = ms +nt
where
C = p2(p2 -q2) pz = mis + t) +n(s -£)

Z? = p2<72 g 2 = m£+ns.
As an example, for m = 2 8 n = 1, s = 3 9 t = 2, we obtain
9 2 + 7 2 2 _ 9 + 72
I442 + 77 2 " 44 + 77 '

The above discussions, of course, give no assurance that all solutions of (a),
(b), and (c) have been found.

Also solved by LEON BANKOFF, Los Angeles, California; RICHARD BURNS AND KRISTIN
DIETSCHE, East Longmeadow H.S 0# East Longmeadow, Massachusetts; LOUIS H8 CAIROLI,
Kansas State University, Manhattan, Kansas? STEVEN R. CONRAD, Benjamin N, Cardozo
H.S., Bayside, N.Ya; HERTA T. FREITAG, Roanoke, Virginia? G«C, GIRI, Research Scholar,
Indian Institute of Technology, Kharagpur, India; J.A.H. HUNTER, Toronto, Ontario;
ALLAN Wm„ JOHNSON JR., Washington, D.C.; ROBERT S. JOHNSON, Montreal, Quebec; LAI
 - 238 -

LANE LUEY, Willcwdale, Ontario; SAHIB RAM MANDAN, Indian Institute of Technology,
Kharagpur, India,- F.GoB. MASKELL, College Algonquin, Ottawa; HERMAN NYON, Paramaribo,
Surinam? SIDNEY PENNER, Bronx Community College, New York; BOB PRIELIPP, The
University of Wisconsin-Oshkosh; HYMAN ROSEN, Yeshiva University H.S., Brooklyn,
N.Y.; FREDERICK N. ROTHSTEIN, New Jersey Department of Transportation, Trenton, N.J*;
the late R. ROBINSON ROWE, Sacramento, California; HARRY D. RUDERMAN, Hunter College,
New York? HARRY SITOMER, Huntington, N.Y.; CHARLES W„ TRIGG, San Diego, California;
KENNETH M. WILKE, Topeka, Kansas; KENNETH S. WILLIAMS, Carleton University, Ottawa;
JOHN A. WINTERINK, Albuquerque Vocational Institute, Albuquerque, New Mexico? and
the proposer.

REFERENCES
1. R.D. Carmichael, Diophantine Analysis, Dovers New York9 1959s p. 25.
2. J.V. Uspensky and M.A. Heas1et9 Elementary Number Theory, McGraw-Hill,
New Yorks 19399 p. 393.

520 C1978: 36] Proposed by Dan Sokolowskys Antioch College.

s

The sides of A ABC are trisected by the points P , P 2 , Q 1 ? Q 2 , x

a*
as shown in the figure below. Show that:
(a) APJQJRJ s AP2Q2R2;
(b) JPJQJRJ| = |!ABC|9 where the bars denote area;
(c) the sides of As P 1 Q 1 R 1 and P 2 0 2 R 2 trisect one another;
(d) If Mj is the midpoint of AB S then C 9 S s T, Mx are collinear.

I. Solution by Clayton Wa Dodge, University of Maine at Orono*

Because QjRx is parallel to
and 2/3 the length of CM X , as also
is P 2 R 2 , it follows that P 2 Q 1 R 1 R 2
is a parallelogram; and similarly
Q
2R1P1P2 a i l d R
2 P 1 C ?1 C !2 a f e
Pafal"

lelograms. Their diagonals

QjR 2S RjP 29 and P X Q 2 all concur
at a point 6 that bisects each
diagonal. Also9 G is the
center of a half-turn that
carries &?lQ1R1 into
A Q 2 R 2 P 2 9 so these two
triangles are congruent
(in that order rather
than the order given in
 - 239 -

part (a) of the proposal). Hence part (a).

Because ARX = AB/3 and AQ X = 2AC/39 we have |AQ1Rl | = | « | • jABCj. From this
and two similar results, we get

jAQ^J = |BR1P11 = ICP^J =||ABC|9

and part (b) follows by subtraction.
PJQJ and P;2Q2 are medians of A C Q 2 P l 9 so S is its centroid, and P J Q l and P 2 Q 2
trisect each other. Similarly Q 1 R 1 and Q2%z* and RjP and R2P,,9 trisect each other.
The half-turn above takes S into T 9 so T trisects R 2 Q 2 and R X P 1 S and similarly other
trisection points are taken into trisection points. This proves (c).
The similarity that takes ACQ 2 Pj into A CAB shows that C 9 S 9 and Mx are collinear;
and the half-turn then shows that C 9 T 9 and Mx are collinear. Hence (d).
It would not be hard to shows in addition9 that G is the common centroid of
As P ^ R ^ P 2 Q 2 R 2 , and ABC.
11. Extracted from a comment by Sahib Ram Mandan3 Indian Institute of
Technologys Kharagpur3 India.
If the sides of A ABC are internally n-sected by P 9 P 2 , Q , Q 2 S R , R2 (so
that BPj : BC = CP 2 : CB = l/n, etc.), the corresponding results are:

(a) A P J Q J R J is not congruent to A Q 2 R 2 P 2 unless n = 3 or A ABC is equilateral.

w 2 3
(b) I P ^ R J = |P 2 Q 2 R 2 | = - y » |ABC|.

(c) The sides of As P 1 Q 1 R 1 and P 2 Q 2 R 2 divide each other in the same ratios.
Specifically (see figure),
P S : SCL = Q„U : UR0 = R X : XP = 1 : (n - 1)
2 ^2 ^2 , 2 2 2
and
P2V : VR2 = R2T : TQ 2 = Q 2 Y : YP 2 = (n-2) : ( n - 1 ) ,
(d) C 9 S, T 9 and Mx are collinear for all n.
P R a n d A B C h a v e a comon
For all n, also, the As P 1 Q 1 R 1 » ZQ2 2> centroid G.
Also solved by LEON BANKOFF, Los Angeles, California; STEVEN R. CONRAD,
Benjamin N. Cardozo H.S., Bayside, N.Y.; RICHARD BURNS and KRISTIN DIETSCHE, East
Longmeadow H.S., East Longmeadow, Massachusetts; G.C. GIRI, Research Scholar,
Indian Institute of Technology, Kharagpur, India; SAHIB RAM MANDAN, Indian Institute
of Technology, Kharagpur, India; W.A. McWORTER JR., The Ohio State University; HERMAN
NYON, Paramaribo, Surinam; the late R. ROBINSON ROWE, Sacramento, California; HARRY
D. RUDERMAN, Hunter College, New York; KESIRAJU SATYANARAYANA, Gagan Mahal Colony,
Hyderabad, India; HARRY SITOMER, Huntington, N.Y.; CHARLES W. TRIGG, San Diego,
California; KENNETH M. WILKE, Topeka, Kansas; JOHN A. WINTERINK, Albuquerque Technical
Vocational Institute, Albuquerque, New Mexico; and the proposer.
 - 2*JQ -

Editorfs comment.
Trigg mentioned Problem 3707 in ll1, which generalizes part (b) of our comment
II. In our notation, it states that if P j S P 2 S Q l S Q 2 , R x , R 2 are equidistant from
the midpoints of the sides upon which they lie (the distances not being necessarily
the same on the three sides), then fP 1 Q 1 R 1 ! = l p 2 Q 2 R 2 ! 0 A solution to this problem
has not yet been published.
REFERENCE
1. School Science and Mathematics , 77 (December 1977) 7149 Problem 3707,
proposed by Fred A, Miller.

THE DECAPUS

Behold the Decapus9 my friends.

This son of Hecto has ten ends 9
One of which he used9 I think,
To draw himself with his own ink.
A Metric Monster, he is urging
You to become acquainted with
His family, both sires and offspring,
In facts with his whole kin and kith.

EDITH ORR

W O METRIC

As this is written, the Canadian postal workers are on strike, so the Great
Postmaster»in-the-Sky only knows when readers will receive this issue.