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
ftftftftftftftftftftftftftftftftftftfti t ft ft ft ft ft ft ft i t ft ft ft i t ft ft ft & & sVftfti t ft

ISSN 0705-0348

CRUX MATHEMATICORU^

Vol. 5, No. 7
August - September 1979
Sponsored by
Carleton-Ottawa Mathematics Association Mathematique d'Ottawa-Carleton
Publie par le College Algonquin
The assistance of the publisher and the support of the Samuel Beatty Fund, the
Canadian Mathematical Olympiad Committee, the Carleton University Mathematics Depart-
ment, and the Ottawa Valley Education Liaison Council are gratefully acknowledged.
ftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftftfti t ft ft i tftftftftftftft
ftftftftftftftftftftftftftftftftftftftftftftftftftftftftfti t i t i t i t i t ft i tftftftftftft

CRUX MATHEMATICORUM is published monthly (except July and August). The yearly
subscription rate for ten issues is $10.00. Back issues: $1.00 each. Bound volumes
with index: Vol. 1-2 (combined), $10.00; Vol. 3-4, $10.00 each. Cheques or money
orders, payable in Canadian or U.S. funds to CRUX MATHEMATICORUM, 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, 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 A. Makila.
it it *

CONTENTS
Function Pathology Michael W. Ecker 182
Algorithms and Pocket Calculators: The Gamma Function . . . . Clayton W. Dodge 188
Addendum to Function Pathology Michael W, Ecker 190
Areas of Triangles Inscribed in a Given Triangle Dan Pedoe 191
Mama-thematics Charles W. Trigg 192
The Olympiad Corner: 7 Murray S. Klamkin 193
Problems -Probl&mes 199
Solutions 201
Some DON'T'S and DO'S: A Programmer's Blues Mans Havermann 212

- 181 -
 - 182 -

FUNCTION PATHOLOGY
MICHAEL W. ECKER

We shall be dealing here with real-valued functions of a real variable, with

some of the bizarre combinations of properties that such functions can have.
Both mirroring the real world and helping to define our conception of it, the
earliest, most frequently used, and most "elementary" functions were, for the most
part, well-behaved. That is, the functions encountered were generally continuous
and differentiate. But before long less smooth functions were introduced, and then
the unexpected began to happen. In some instances, it was the very failure of
intuition that led to the development of rigorous definitions for such things as the
limit concepts that we all more or less take for granted now. For example, in the
times of Weierstrass and Riemann, there was great concern with questions such as
whether a function could be continuous and still be nowhere differentiate (see [11,
[71, and [81). Even today, we have a legacy of unresolved questions of historical
and mathematical interest, because several specific functions have so far proved to
be intractable to Analysis (see [81).
What the mathematicians of the nineteenth century did not know was that most
such "anomalies" are in fact "normal". (This alludes to sets of first and second
categories. See [11.) The title of this paper, therefore, is a misnomer.
The temptation to make analogies is irresistible. In Measure Theory, one learns
or hears that most sets are nonmeasurable. Yet, where examples are provided, they
are often messy or abstruse, invoking such items as the axiom of choice (see, e.g.,
[91). The situation with real numbers is similar. While we, for whatever historical,
psychological, philosophical, practical, or mathematical reasons tend to choose
rational numbers in our dealings with the reals, the fact remains that, from the
point of view of either cardinality or measure, most reals are irrational. Likewise,
we have the inescapable fact that most functions are not differentiate.
The focus of our interest here will be with the continuity and differentiability
of functions. It is hoped that the examples we choose will reinforce our limited
intuition and enhance the importance and utility of good definitions. No claim is
made that there is anything exhaustive or even representative about the examples or
the references. However, the reader will find [61 a most enriching source of illumi-
nating examples.
Traditionally, undergraduates learn in precalcuius or calculus courses about
the so-called elementary functions, which include polynomials, trigonometric or
 - 183 -

circular functions, exponential and logarithmic functions, etc. From these, despite
our warnings to the contrary, students tend to expect a fair degree of "smoothness"
in all the functions they encounter.
The first kink introduced is usually the relatively innocuous absolute value
function, defined by y - \x\ , whose V-shaped graph makes manifest the possibility of
the combination of continuity and nondifferentiability at a point (here at a? = o).
When calculus students are taught the familiar theorem on differentiability implying
continuity, this is frequently the prime counterexample to the proposed converse
question.
Taking a quantum jump, we ask whether a function can be everywhere discontinuous.
An affirmative answer follows from the fact that the set of rationals, Q9 is dense
in the set of reals, R. (This simply means that every neighborhood of every xeR
contains infinitely many rationals.) The characteristic function of Q, Xn:R~*R>
H
defined by xn{x) = l or o according as x is rational or not, has the desired property.
A further quantum jump brings us to the question mentioned earlier, about con-
tinuous functions that are nowhere differentiate. But first let us back up a bit.
Can we, say, have a function which is differentiate (hence continous), but whose
derivative somewhere fails to be continuous? The function f : R+R defined by fix)
= x2 sin (l/x) if x*o and f(o) = o has this property, since Mmf'ix) does not exist
as x + o whereas continuity of /' at x = o requires this limit to exist. (Interestingly
enough, /'(o) exists and is o, but that's beside the point.) So continuous functions
that are differentiate need not have their derivatives continuous. In the light of
this somewhat surprising possibility, it is interesting to note that, by a theorem
of Darboux, while derivatives need not be continuous, they must satisfy the inter-
mediate value property (IVP). This, in fact, is a frequent starting point in trying
to characterize functions that are derivatives, as in [21. More generally, it is
true that a function which is continuous must satisfy IVP, while the converse fails.
It usually comes as a surprise that this is so, but experimentation soon leads one
to a function that does a lot of "wiggling", and one then comes up with something
like the function defined by g(x) = sin (l/x) if x*o and g(o)~c (arbitrary), for
which "Wmg(x) does not exist as x-+o9 so that g is not continuous at x = 0. However,
the function takes on all values in the closed interval [-1,1] infinitely often in
every neighborhood of o. It is frequently used in constructing counterexamples.
Of the two problems which follow, the first was suggested by [3] and [5], and
the second is a variation on the same theme. In both problems, we consider only real
numbers x in the interval (0,1) and write x = l(a /10 77 ), using the terminating
 - 184 -

expansion (i.e. with zeros) where such choice exists. This convention ensures that
the functions are well-defined. All sums are for n = 1,2,3
PROBLEM 1. Let p be a permutation of the digits 0 through 9. This induces, in
a natural way, a function p* defined by

, io n / io n
Discuss fully the continuity and differentiability properties of functions of this
type, the family of which we will designate Gx.
PROBLEM 2. Let i\ be a permutation of the natural numbers 1, 2, 3, ... . This
induces, in a natural way, a function TT* defined by

a , v

4%)
Discuss fully the continuity and differentiability
ion
properties of functions of this
type, the family of which we will denote Gz.
To illustrate, consider x = .12349 = .123490, where the 6 is added to emphasize
the necessity of treating all decimals as infinite series. Let p be defined by
p(7<) = k + 1 (mod lo), so that p(o) = 1, p(l) = 2,..., and p(9) = o. Then

p * ( x ) = p*(.123490) = .234501.
Note the infinite tail of ones. If ir is defined, say, by the rule "interchange each
pair of adjacent digits", that is, -n(k) = k-1 if k is even and ir(fe) =fc+ l if k is
odd, then
7T*(x) = TT*(. 123490) = .2143090.
Some preliminary observations.
(a) Functions in G1 change decimals by changing the digits into other prescribed
digits, while those in G2 permute the digits of the decimal in a prescribed way.
(b) Gx has 10! functions, while G2 is infinite.
(c) In each family, the appropriate identity permutation induces the identity
function on (o,l).
(d) The functions in both families are, or come tantalizingly close to being,
bijective. (To test yourself, find out which functions, other than the identity,
are bijective and why the others fail to be injective and/or surjective.)
(e) Recalling that a real number is rational if and only if its decimal
 - 185 -

representation is eventually periodic (this includes the terminating case), we get

a quick bonus by observing that all the functions of G1 have the interesting property:
p*(x)eQ if and only if xeQ. Many of the functions in G2 also have this property,
although by no means all. For example, the function ir* used above as an illustration
has it, although the periods of the decimals' repeating parts are doubled if originally
odd. Likewise, all functions TT* for which ir(k)=k for all k greater than some M have
this property, and without the period changing.
The following important results are noted. Some proofs will follow, and others
will appear in the solution of [5] when it is published.
(1) Every function p * in Gx is continuous except possibly on a countable dense
subset of the terminating decimals. The only (everywhere)continuous functions are
p*(x) = x and p*(x) = 1 -x9 induced respectively by p{k) - k and p(k) = 9 -k. The
others all have a one-sided discontinuity at a subset of the terminating decimals x.
Moreover, aside from these two continuous functions, which are everywhere differentiate,
the other lo! - 2 functions are all nowhere differentiate (see solution of [5]). The
one-sided discontinuity, it should be mentioned, is from below and cannot be removed
by changing p * at such a terminating decimal without introducing a new discontinuity
from above. Thus, if we were to use the expansion of x with nines, we would obtain
discontinuity from above. It is interesting to note that for every function we have
everywhere at least one-sided continuity.
(2) The functions IT* of G2 are also continuous except possibly at the terminating
decimals, the discontinuities there being one-sided as for G1. The functions, more-
over, are either differentiate nowhere, or else are differentiate except at finitely
many of the terminating decimals. In the latter case, the functions are characterized
by the property that they are induced by a permutation IT satisfying ij(,k) = k for all k
greater than some M9 and we then have in the order of 1 0 M points of nondifferentiability,
namely the finite decimals having M or fewer places before the terminal 6 in the
decimal representation. Finally, in the differentiate case (except at the afore-
mentioned points), we have dir*(x)/dx = 1 for all differentiate IT* in G2 and all
allowable x in (0,1).

Outlines of some proofs.

Consider a function p* of G1. It is almost immediate (although some small care
must be taken) that the function has at least the restricted continuity, since decimals
near a given x will tend to have images near that of x. This can only be extended
to terminating decimals x from above, since it is impossible to have two decimals,
one terminating, such that an arbitrary number of first digits agree and still have
 - 186 -

the other decimal less than the terminating one. Further experimentation leads
one to investigate the implications of continuity from below for finite decimals,
and one is eventually forced to conclude that the original permutation p must map
adjacent integers to adjacent integers, yielding either p(k)=k or p(k) = 9-fc,
inducing p*(x) = x or p*(x) - l -x, respectively.
The continuity of functions in G2 can be investigated in a similar manner. To
discover their differentiability properties, we set h=-lo~n where n = 1,2,3,...
and use the difference quotient
-1
i\*(x + h) - i\*(x) H^-W (n)
= 10
~h
For TT* to be different!*able at a given x, this quotient must approach either o or a
nonzero constant as n+°°, that is, n-is (n) must approach -°° or a constant o. The
-1
fact that TT is bijective rules out the first possibility altogether and, since n-ir (n)
_-i
is always an integer, the second leads to TT (n) = n-c for all sufficiently large
rc, or TT(n) = n + c. Considering the first M values of n which do not satisfy these
relationships, we conclude from the bijectivity of TT that t? = o. It can now be
verified that the functions TT* such that Tr(n)=n for n>M are indeed differentiate,
except at the finite decimals of M or fewer digits after the decimal point. Note
that TT* is differentiate at the other terminating decimals. While the above der-
ivation essentially (and tacitly) assumed that x was not a terminating decimal, no
generality is lost. We will leave it to the reader to verify that C?TT*(X)/dx = 1
subject to the stated limitations on TT and x.
Other "pathological" functions with similar continuity and differentiability
properties can be constructed. For example the author, in trying to find a function
/ that is discontinuous on a dense subset of its domain and differentiate on another
dense subset, only succeeded in producing an example of the earlier type (continuous
on a dense set, discontinuous on another dense set, differentiate nowhere). The
very method of construction is similar in appearance:
Define /: (o,l) -»• (0,1), with the same convention as before, by

fe)
1
n*
10

where b -a or 5 according as a * o or a = 0 . It is a straightforward exercise to

n n n n
demonstrate that f is continuous on a dense subset of its domain (e.g. at those x
with only finitely many a = o ) and discontinuous on another dense subset (at
those x with only finitely many a * o ) . The denseness of these subsets is clear,
 - 187 -
and by investigating the difference quotient for / one finds that no limit exists
as h + o9 due to the dependence of f(x + h) on h.
The functions discussed throughout this article are a very nice antidote to
primitive notions of continuity such as "drawing the graph without picking up the
pencil". They also have the virtue of all pathology: once studied, they are
available as examples or counterexamples. Lung Ock Chung appears to have made a
detailed study of such functions for, in addition to the Monthly problem earlier
cited [3], he has submitted another related problem [43: Find a function f : Co,l]
->[0,1] which is continuous everywhere except on two countable dense subsets DXy D2
of rationals such that on T)iy f is right continuous but not left continuous3 and on
V23 f is left continuous but not right continuous. It should be possible, by some
straightforward modification, to find functions from both families G1 and Gz which
have these properties.
It is customary to end with a few questions. What happens if we relax the
requirements that p and TT be permutations? What if these maps are abritrary? Note
that no reference has been made here to the group structure of the permutations p
and TT. Can this be utilized to derive any interesting consequences? The functions
p*(x) =x and p*(x) = 1 -x form a cyclic group of order 2 under composition, as do
the permutations which induce them, p(k)~k and p(k) = 9-fc. Is there any significance
to this?
REFERENCES
1. Ralph P. Boas, Jr., A Primer of Heal Functions, Second Edition, Carus
Mathematical Monograph No. 13, The Mathematical Association of America, 1972,
pp. 55, 61-62, 120.
2. Andrew M. Bruckner, "Derivatives: Why They Elude Classification", Mathematics
Magazines 49 (January 1976) 5-11.
3. Lung Ock Chung, Problem 6054, American Mathematical Monthly, 82 (November
1975) 941-942. Solution: 84 (June-July 1977) 494,
4. , Problem 6142, American Mathematical Monthly, 84 (March 1977) 222.
Solution: 85 (November 1978) 773.
5. Michael W. Ecker, Problem E 2738, American Mathematical Monthly, 85 (November
1978) 76M-765. Solution to appear.
6. Bernard R. Gelbaum and John M.H. Olmsted, Counterexamples in Analysis,
Hoi den-Day, San Francisco, 1964.
7. Brent Hail pern, "Continuous Non-differentiable Functions", Pi Mu Epsilon
Journal, 6 (Fall 1976) 249-260.
 - 188 -
8. Erwin Neuenschwander, "Riemann's Example of a Continuous, 'Nondifferentiable'
Function", The Mathematical Intelligencer* 1 (1978) M-0-44.
9. H.L. Royden, Real Analysis* Second Edition, Macmillan, New York, p. 63.
Department of Mathematics, Pennsylvania State University, Worthington Scranton
Campus, 120 Ridge View Drive, Dunmore, Pennsylvania 18512.
* * s'c

ALGORITHMS AND POCKET CALCULATORS:

THE GAMMA FUNCTION
CLAYTON W. DODGE

There are many functions for which accurate polynomial approximations are handy
to have. For any function defined on any given finite set of points, it is possible,
for any given maximum error e, relative or absolute, to find a polynomial approximation
with error less than e. Generally, the smaller e > o is, the larger is the degree of
the polynomial.
Since we are looking for approximating polynomials of relatively low degree, we
may have to sacrifice some accuracy. There is nothing magical about polynomials,
however, and the minimax method1 can be applied to rational functions as well. In
fact, some shrewd guessing with rational, algebraic and transcendental functions may
at times be appropriate.
Here we shall examine several approximations to the gamma function, defined by
•oo

/ 0

It is well known that

rGc + l) = xT(x) and TCI) = 1 (1)
so, when x is a nonnegative integer,
r(x + i) = x\
and the gamma function is a generalization of the factorial function.
We give the best first: Fike !~l, p. 1571 gives the minimax rational approxi-
mation

For the minimax method, see an earlier article in this series [1978: 217].
(Editor)
 - 189 -

re:r? + l ) .9999999851 + .6017813 806a: + .1861449035a? 2 + .0687440519a? 3

1 +1.17899457600? - .1223213114a?* - .2609958662a? 3 + .0609927504a?1* ' ( }

with absolute error on the interval Co,i] less than 1.49xio""8. We say this
approximation has deqree 7, the sum of the numerator and denominator polynomial
degrees. Since one can now calculate T(x) to seven-place accuracy for l<a?<2,
one can evaluate the gamma function to seven significant digits for any real a? for
which it is defined by using the reduction formula (1). Indeed, formula (2) was
the standard by which the author measured all other approximations. If you have
need to calculate Y(x) from time to time and if you have available a programmable
calculator with a card reader, you may wish to record equation (2) on a card for
your use.
Since equation (2) is rather cumbersome to use by hand, let us consider several
other minimax formulas for the gamma function. The formula

r(
* ) = r ^ H k ' 1-9507***3 (3)
is given by Tike [1, p. 1791 as having two-place accuracy (to the nearest .01).
Certainly it is easy to use with a hand calculator.
A search for greater accuracy produced

T(x) - ^ 7 - 3 . 4 4 4 a ? + 1.481a?2 1 < x < 2 (4)

I(a?) 1 . 7 5 6 8 - f - a ? ' 1 ^ a? < 2 (4)

with maximum absolute error .001, or almost three-place accuracy.

Observing that Stirling's formula states that

r(a? + l) » f^TCc f|J

for large a?, the author calculated values of T(x) • {e/x)x and applied the minimax
method to find
„, N 2.08635+ 1.09061a? - .078735a: 2 fxY , _,
(5)
r (a .) = A^T~X \7) '
which has maximum error .00004, hence four-place accuracy, for 1 < # < 2 .
One further refinement produced a slightly better degree 3 approximation. For
1 <x< 2, we have
r, , 2.56222 - .02783a? + ,00498a?2 fxY r (c.
T(X) = (6)
-.06582 fa? \e) ^

with maximum error .ooooil, or almost five-place accuracy.

We have presented five formulas for the gamma function on some unit interval
 - 190 -

giving accuracies of from two to seven places, so the choice of formula utilized
depends upon the accuracy desired and the degree of complexity one will accept.
These formulas are best, of course, for small x , becoming increasingly time-
consuming as x increases. Using the reduction formula (1) along with one of the
approximations in equations (2) to (6) grows less practical as x becomes large,
even with a programmable calculator. So we seek a formula to evaluate the gamma
function for large x .
The Stirling approximation, listed between equations (H) and (5), is not very
good, giving only two-place accuracy for 8 < # < 7 0 , the maximum rvalue usable with
a pocket calculator since r(7i) > lo 1 0 0 . The HP-35 Math Pac [2] lists
fx\X I 1 1 139 571 \
r(* + l) = / 2 ^ ^-J (1 + m < 2 8 8 ^ " 51840P" " 2488320*"j
as an approximation valid for large x . It gives seven-place accuracy for x > 13 and
five-place accuracy for x>2.l. The last term in the parentheses can be neglected
for #>14.7 or x>^.7, respectively. For five-place accuracy, the last two terms
can be omitted when X > 1 7 . 6 .
Perhaps the easiest way of all to calculate values of the gamma function is to
use the Commodore M55 Mathematician calculator which, according to John Free r3l,
has built-in programs for many functions of interest to statisticians and applied
mathematicians, including the gamma function.
REFERENCES
1. C.T. Fike, Computer Evaluation of Mathematical Functions, Prentice-Hall,
1968, pp. 75-76, 92, 111-112, 157, 169, 178, 207.
2. HP-35 Math Pac, Hewlett-Packard, 1974, p. 75.
3. J. Free, "Specialized Calculators Shortcut Tough Problems", Popular Science,
210 (June 1977) 171.

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

J'C J'J s'f

Addendum to Function Pathology,, pp. 182-188 in this issue:

Within the last two months I finally succeeded in producing an example of a
function f with the property that f is continuous on a dense subset of its domain,
discontinuous on another dense subset, and differentiate also on a dense subset S.
Moreover, the derivative of f can be manipulated (by modifying f) to be the re-
striction to S of any derivative we wish!
This is the major result of an article now being considered for publication
elsewhere.
MICHAEL W. ECKER
 - 191 -

AREAS OF TRIANGLES INSCRIBED IN A GIVEN TRIANGLE

DAN PEDOE

In a comment following the solution of Crux 320 [1978: 2*+o], the editor of this
journal refers to a problem proposed 1n 1977 M , which reads:
Let RST be a triangle such that M, N, and
L are the midpoints of its sides, (See figure.
It need not be the case that MF3 NB^ and LE
have the same length*) If triangles ABC and
DEF have vertices which lie on the sides of
triangle RST at equal distances from M, N, " D N B
and Lj prove that these triangles have the same area.
This is quite an old problem. It appears (with proof) as a Theorem in Johnson
[3]; it can be derived from Exercise 39 in Forder [23; and it is Exercise 11.7 in
my book [5, p. 55], where it is easily deduced from the preceding Exercise 11.6.
Forder uses the method of Grassmann, which in this case is nothing but a shorthand
for the determinantal expression for the area of a triangle, and in my book I explain
this method and give applications. All I wish to demonstrate in this note is the
ease of calculation which the method affords, compared with, say, the published
solution [1],
Exercise 11.6 in my book reads:
Let ABC be a given triangle, and let d3 E, F be points on BC3 CA,, and AB
respectively dividing the segments BC3 CA^ and AB in the ratio k1:li k2\ls and
k3:± respectively1. Show that
area of A DEF : area of A ABC = (1 + kxk2k3) : (1 H - S ^ K I +fc2)(l +k3).
If a, t9 and c are vectors from some origin to the vertices of triangle ABC,
the vectors to D, E, and F are given by the equations

(l + kl)d = b + kiS, (l+kz)e = c+k2a, (l+fc3)/ = a+kBb.

2
Hence

(1 -tk^il +kz)(l +k3)(d At *f) = (J +fcj?)A(J +fc22)A (2 + fc3ib)

The meaning here is that BD:DC = kitl, etc. ^ ^ ^

2
For the exterior (or wedge) product of three vectors, uAirAy, see my book
[5, p. 52].
 - 192 -

0 1 k1

K o ± (a A ib A<?)
l k3 o

= (1 +klkzk3)(a*b Ac?);
and since
d *t *f :a Ab A~C = area of A DEF : area of A ABC,
the required result follows.
To obtain the result we are discussing, we merely observe that the given ratio
is unchanged if we replace kx, k2, and k% by their respective reciprocals.
Similar methods produce a short proof of Routh's Theorem, which is given as
Exercise 11.8 on the same page in my book. With similar notation this reads, in
brief:
D., Ej and F are points on the sides BC3 CA., and AB of a triangle ABC such that

a = <?i£+Pi<2> e = q2c+p2a9 f = q^a+p^b,

where p1+q1 - p2 + q2 - p3 +q3 = 1 (so that BD : DC = pY : qx, etc). The lines AD.,
BEj and CF form a triangle PQR. Prove that the ratio of the areas of triangles PQR
and ABC is

Again, once the points P, Q, and R are found, the working is immediate.
REFERENCES
1. Mangho Ahuja, Solution to Problem 3707, School Science and Mathematics, 78
(December 1978) 715-716.
2. H.G. Forder, The Calculus of Extension, Chelsea, 1960, p. 21.
3. Roger A. Johnson, Modern Geometry, Houghton Mifflin Co., 1929 (reissued as
Advanced Euclidean Geometry, Dover, I960), p. 80.
4. Fred A. Miller, Problem 3707, School Science and Mathematics, 11 (December
1977) 714.
5. D. Pedoe, A Course of Geometry for College and Universities, Cambridge
University Press, 1970.
School of Mathematics, University of Minnesota, Minneapolis, Minnesota 55455.
ft * ft

MAMA-THEMATICS
Mrs. Dedekind to son Richard: "And I thought you wanted that knife to whittle with."
CHARLES W. TRIGG
 - 195 -

THE OLYMPIAD CORNER; 7

MURRAY S. KLAMKIN

The XXI International Mathematical Olympiad was held in London, England, on

July 2-3, 1979. The 22 participating nations were: Austria, Belgium, Brazil,
Bulgaria, Cuba, Czechoslovakia, Federal Republic of Germany, Finland, France, German
Democratic Republic, Great Britain, Greece, Hungary, Israel, The Netherlands, Poland,
Romania, Sweden, U.S.A., U.S.S.R., Vietnam, and Yugoslavia.
Canada, as usual, did not participate, no funds being available for the training
and travel expenses of a team.
The five top-scoring countries were:
1. U.S.S.R. Score 267 (maximum 320)
2. Romania 240
3. Federal Republic of Germany 235
H. Great Britain 218
5. U.S.A. 199
The members of the U.S. team were:
Richard G. Agin Age 18 Chicago, Illinois
Randy L. Ekl 17 Huntington, Pennsylvania
Michael V. Finn 16 Annandale, Virginia
Ronald F. Kaminsky 18 Albany, New York
Michael J. Larsen 17 Lexington, Massachusetts
Lawrence E. Penn 17 Great Neck, New York
Mark G. Pleszkoch 18 Manassas, Virginia
Bruce K. Smith 18 San Raphael, California
The problems posed at the International Olympiad are given below. Solutions
to these problems (together with those of the 1979 U.S.A. Mathematical Olympiad)
will appear later in a pamphlet compiled by S.L. Greitzer. The pamphlet will be
obtainable (50<£ per copy) from Dr. W.E. Mientka, 917 Oldfather Hall, University of
Nebraska, Lincoln, Nebraska 68588.

THE XXI INTERNATIONAL MATHEMATICAL OLYMPIAD

LONDON 1979
MONDAY JULY 2nd 1979
Time: M- hours

1, Let p and q be natural numbers such that

 - 194 -

P . i i i J
<7 2 3 4 '•" 1318 1319 *
Prove that p is divisible by 1979.
2, A prism with pentagons A 1 A 2 A 3 A 1+ A 5 and BjBjBgB^Bg as top and bottom faces
is given. Each side of the two pentagons and each of the line-segments A.B.,
for all i9j = l,..,, 5, is coloured either red or green. Every triangle whose
vertices are vertices of the prism and whose sides have all been coloured has two
sides of a different colour. Show that all 10 sides of the top and bottom faces are
the same colour.
3, Two circles in a plane intersect. Let A be one of the points of intersection.
Starting simultaneously from A two points move with constant speeds, each
point travelling along its own circle in the same sense. The two points return to
A simultaneously after one revolution. Prove that there is a fixed point P in the
plane such that, at any time, the distances from P to the moving points are equal.
TUESDAY JULY 3rd 1979
Time: M- hours

t\t Given a plane IT, a point P in this plane and a point Q not in TT, find all
points R in IT such that the ratio (QP + PR)/QR is a maximum.
5, Find all real numbers a for which there exist nonnegative real numbers
xl% x2, x3, x^, x5 satisfying the relations
5 5 5
I kxv - a, I k*xv - a2 , J fc5^, = a 3 .
K K K
fc=l fc=i fc=l
6, Let A and E be opposite vertices of a regular octagon. A frog starts
jumping at vertex A. From any vertex of the octagon except E, it may
jump to either of the two adjacent vertices. When it reaches vertex E, the frog
stops and stays there. Let a be the number of distinct paths of exactly n jumps
ending at E. Prove that
a a =
1(X , n-1 V n - L „n „
h
2n-l = °- 2n f! ~ " = 1-2'3'---'
where x - 2 + /2 and y - 2 - / 2 .
Note. A path of n jumps is a sequence of vertices (P0,..., P ) such that
(i) P 0 = A, P n = E;
(ii) for every t9 o < i ^ n - l , P. is distinct from E;
 - 195 -

(iii) for every i, o < i < n - l, P^ and p^ are adjacent.

As promised last month, I now give solutions to the problems posed at the Eleventh
Canadian Mathematics Olympiad. The problems and solutions were prepared by the
Canadian Mathematics Olympiad Committee, whose membership was listed in last month's
column.
ELEVENTH CANADIAN MATHEMATICS OLYMPIAD (1979)

1, Given: (i) a, b>o; (ii) a, A l 9 A 2 , b is an arithmetic progression; (iii)

a9 G19 G29 b is a geometric progression. Show that

A1A2 > GXG2.

Solution.
In the A.P., the common difference is d = (fc-a)/3; hence

AXA2 = (a+d)(b -d) = ab + (b -a)d-d2 = ab + 2d2 .

If r is the common ratio in the G.P., we have
GXG2 = (ar)(b/r) = ab.
Hence AXA2 -GXG2 = 2d1 > 0.
2, It is known in Euclidean geometry that the sum of the angles of a triangle
is constant. Prove, however, that the sum of the dihedral angles of a
tetrahedron is not constant.
NOTE: (i) A tetrahedron is a triangular pyramid,
(ii) A dihedral angle is the interior angle
between a pair of faces. &
Solution.
Consider a tetrahedron whose base is an equilateral triangle and whose vertex
P lies above the centroid C of this base. The dihedral angles at the horizontal
edges are all equal, say to a, and those at the other edges are also all equal, say
to 6. Thus the sum of all dihedral angles is 3(a + 6). As P-*C, a-*o and $ + TT, so
the sum can be made as close to 3TT as we want. As P recedes from C, a-^-rr/2 and
3->TT/3» so the sum can be made as close to 37r/2+7r = 5TT/2 as we want. This shows
that the sum of the dihedral angles is not constant.

3, Let a, b, o, d, e be integers such that ±<a<b<o<d<e. Prove that

1 1 1 1 < 15
tatbl +
£b9c! +
tc,dl +
£d,el ~ 16 9
where [>?,nl denotes the least common multiple of 777 and n (e.g. [4,6] = 12).
 - 196 -

Solution.
Let S denote the sum on the l e f t of the proposed i n e q u a l i t y . Since l<a< ... <e,
we have
Xa,bl > 2a, Xb,cl > 2b, Xo M ;> 2a9 Xd,el > 2d,

and b>29 c>3. We consider two c a s e s :

Case 1. o = 3. If ( f = 4 , then Xa,bl = 2 , Xb,cl = 6 , [<?,<?] = 1 2 , and W , e ] > 8;
hence
0 1 1 1 1 7 15
S + + +
^ 2 6 12 ¥ = 8 <
16-

I f d > 5 , then [<?,£?] > 6 and [<2,e] > 10; hence

e l l l l 14 15
S + + + =
* 2 6 6 10 15 < 1 6 '

Case 2. <?>4. I f 5 < < f < 7 , then Xa,dl = 20, 28, 30, 35, or 42, except when
e = 4 and d-6. In the exceptional case, we have

+ +
2 4 12 12 " 12 16 '
and otherwise

2 4 20 10 10 16

F i n a l l y , i f d>89 we have

5
„ ^ 1 + 1+ 1+ 1= 15
^ 2 4 8 16 16
Note. This is a special case of the following theorem originally conjectured
by P. Erdb's and proved by D. Borwein {Canadian Mathematical Bulletin, 21 (March 1978)
117-118):
Show that if 1< a0 <ax < ... < a,s then

1 1 1 < _ _1_
[<3A , a ] Ta .ao1 *" Xa. A, a. 1 0& '

with equality if and only if a. ~ 2 3 0 <i <k.

L\, A dog standing at the centre of a circular arena sees a rabbit at the wall.
The rabbit runs around the wall and the dog pursues it along a unique path
which is determined by running at the same speed and staying on the radial line
joining the centre of the arena to the rabbit. Show that the dog overtakes the
rabbit just as it reaches a point one-quarter of the way around the arena.
Solution.
 - 197 -
It is sufficient to show that the arcs
RR* and DD* are equal in length (see figure).
If the arena has radius r, then arc RR' =r6.
Also, /D'QD = /R'DR = e (corresponding sides
perpendicular), so arc DD' subtends an angle
of 26 at the centre of its circle (of radius
r/2). Thus arc DD' = (r/2)(20) = r6, as
required,

5, A walk consists of a sequence of

steps of length l taken in directions
north, south, east or west. A walk is self-
avoiding if it never passes through the same
point twice. Let fin) denote the number of
n-step self-avoiding walks which begin at
the origin. Compute f(l), /(2), /(3), / C O . and show that

2n < fin) < 4 • 3 n_1 .

Solution.
Since a one-step walk can be taken in any of the four directions, /(1) = 4. Two-
and three-step walks are self-avoiding provided they don't backtrack; therefore
/'(2) = 4 • 3 = 12 and f(3) = 4 • 3 • 3 = 36, To continue a three-step walk to a four-
step one,, we must avoid backtrackinn and closing a loop. The 8 walks of the form
*"9 | can only be continued in two ways, and the remaining 36 - 8 = 28 can be continued
in three ways. This shows that f{*\) = 8 • 2 + 28 • 3 = loo.
If v/e take n steps only to the north or west, we get 2n walks which must be
self-avoiding, so 2n < fin). If we only guard against backtracking, we get 4 . sn~
walks which may not be all self-avoiding but certainly include the self-avoiding ones,
SO fin) < M- • 371"1.

Finally, to give ardent problem solvers a full measure, we present the

SOLUTIONS TO PRACTICE SET 6

6-1, Determine all triplets of integers ix,y,z) satisfying the equation

oc3+y3+s3 = ix +y + s ) 3 .

Solution.
Expanding the right side, simplifying, and factoring, we get the equivalent
 - 198 -

equation
(y + z){z +x)(x+y) = 0.
Thus the solutions consist of all the triples
(x9y9z) = (a9b,-b)9 {b9a9-b)9 or (b9-b9a)9
where a and b are arbitrary integers.

6-21 Given are two points, one on each of two given skew lines (lines not
lying in a common plane). Prove that there exists a unique sphere
tangent to each of the given lines at each of the two given points.
Solution.
The center of the desired sphere must lie in the two planes TT I5 TT2, each passing
through one of the two given points Pl9 P 2 and perpendicular to the given line con-
taining the point. Since the two given lines are skew, the planes TT IS TT2 must
intersect in a line L which contains the center of the sphere. Since the center of
the sphere must be equidistant from Px , P 2 , it also lies on the perpendicular bi-
secting plane TT of the segment P ^ . Finally, -rr must intersect line L in the unique
center.
If
6-3• %>y>z * o, prove that
x3+y3+z3 £ y2z + z2x + x2y
and determine when there is equality.
Solution.
Without loss o f g e n e r a l i t y , we need only consider two cases: (1) x^yzz and
(2) x>z >y.
Case 1 . The desired i n e q u a l i t y follows from
2
x (x -y) + y2(y - z) > y2(x -y) +y2(y - z) = y 2 (x - z) ^ z2(x - z).
Case 2. Here i t follows from
2
y (z -y) +z2(x -z) < z2{z ~y) + z2(x - z) = z2 (x -y) < x2{x -y).
Alternatively, by Holder's inequality,

(y3 +z3 +x3)2/3(z3 +x* +y3)1/3 * y2z+z2x+x2y

and, more g e n e r a l l y ,

f n n nAn-k)ln, n n n.k/n ^ . n-k k n-k k n-k k.

(x. + x0 + . . . + x ) (as, + . . . + # + x.) £ (x, x„ +rc„ x0 + ...+x x.),
1 z l
n * n i z z A YI \
 - 199 -

with equality if and only if

Editor's note. All communications about this column should be sent to

Professor M.S. Klamkin, Department of Mathematics, University of Alberta, Edmonton,
Alberta, Canada T6G 2G1.
& ft ft

PROBLEMS--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 (it) 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 December 1, 1979, although solutions received after that date will also be
considered until the time when a solution is published.

461 f Proposed by the late R. Robinson Rowe3 Rowe Manses Nauhinway* Michigan.
Restore the digits in the decimal addition

C
DODGE
MAINE ,
0R0N0

with everyone agreeing that DODGE is the greatest.

4621 Propose" par Hippolyte Charless Waterloo, Qu&bec.

Soient A, B, C les angles d'un triangle. Montrer que

I A I
tan j cos A l
I p J
tan — cos B l = o.

C
tan — cos C 1

4631 Proposed by Jack Garfunkels Flushing* N.Y.

Construct an equilateral triangle so that one vertex is at a given
point, a second vertex is on a given line, and the third vertex is on a given
circle.
 - 200 -

464 • Proposed by J. Chris Fisher and E.L. Koh, University of Regina.

(a) If the two squares ABCD and AB'C'D' have vertex A in common and
are taken with the same orientation, then the centers of the squares together with
the midpoints of BD' and B'D are the vertices of a square.
(b) What is the analogous theorem for regular n-gons?
4651 Proposed by Peter A. Lindstrom, Genesee Community College, Batavia, N.Y.
For positive integer n9 let o(n) = the sum of the divisors of n and x(n)
= the number of divisors of n. Show that if o(n) is a prime then x(n) is a prime.

466 i Propose par Roger Fischler, University

Carleton, Ottawa.
Soient AB et BC deux arcs d'un cercle tels que arc
AB > arc BC, et soit D le point de milieu de 1'arc AC
(voir la figure). Si DEiAB, montrer que AE = EB + BC.
(Ce theoreme est attribue a Archim§de.)

4671 Proposed by Harold N. Shapiro, Courant

Institute of Mathematical Sciences, New York
University.
Let n ,..., n- be given positive integers and form the vectors (d19...9 <2-)
where, for each i = 1,..., k9 d. is a divisor of n.. Letting x(d) = the number of
divisors of d9 the number of these vectors is T(n1)T(n2)...Tin,). How many of these
have the property that their components are relatively prime in pairs?

468 i Proposed by Viktors Linis, University of Ottawa.

(a) Prove that the equation
k k
l 2
axx + a2x + .. •1 = 0,
where a1, , a are real and &.,..., k are natural numbers, has at most n positive
1
roots. n n

(b) Prove that the equation

ax (x + 1 r + bx (x + 1 y + ex (x + 1) -1 = 0,

where a9 b9 c are real and k9 l9 m9 p9 q9 r are natural numbers, has at most 14

positive roots.

•469 ( Proposed by Gali Salvatore, Perkins, Quebec.

Of the conies represented by the equations
 - 201 -

±x2 ±2xy±y2 ±2x±2y±l = 0,

how many are proper (nondegenerate)?

^701 Proposed by Allan Win. Johnson Jr., Washington, D.C.

Construct an integral square of eleven decimal digits such that, if
each digit is increased by unity, the resulting integer is a square. (A four-
digit example is 45 2 = 2025-*»3136 = 56 2 .)
ft ft ft

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.

387\ C1978: 2513 Proposed Harry D. Ruderman, Hunter College Campus School,
New York,
N persons lock arms to dance in a circle the traditional Israeli Hora. After
a break they lock arms to dance a second round. Let P(N) be the probability that
for the second round no dancer locks arms with a dancer previously locked to in the
first round. Find lim P(N).
N-*°°
Editor's comment.
Not one solution or comment was received for this problem, which therefore
remains wide open.
ft ft ft

3881 £1978: 251] Proposed by W. J. Blundon, Memorial University of Newfoundland.

Prove that the line containing the circumcenter and the incenter of a
triangle is parallel to a side of the triangle if and only if (in the usual notation)
2 (2Z?-i»)2(J?+iO
s
" R-r

I. Solution by Orlando Ramos, Instituto Politecnico Jose Antonio Echevarria,

Havana, Cuba.
The line containing the circumcenter and the incenter is parallel to side BC,
say, of AABC if and only if J"?cos A = r. Thus the required condition is equivalent
to
(i?cos A - r)(i?cos B - r)(/?cos C - r) = o,
that is,
i?3IT cos A - R2rl cos B cos C + Rr2Z cos A - r3 = 0.

The desired relation now follows upon substituting the well-known relations (see
 - 202 -

[1978: 591)
R +r
2: cos A =
R *

s2 -W2 tr2
£ cos B cos C

s2 -t|j? 2 -i±Rr -r2

ITcos A
m2

II. Comment by Hayo Ahlburg, Benidovm, Spain,

The relation also holds for the equilateral triangle, where "the" line con-
taining the circumcenter and the incenter does not exist. Perhaps the wording
could be changed slightly to include this "degenerate" case. How about "a" line?

Also solved by G.C. GIRI, Research Scholar, Indian Institute of Technology,

Kharagpur, India? ALLAN Wm. JOHNSON Jr., Washington, D.C.; HERMAN NYON, Paramaribo,
Surinam; KESIRAJU SATYANARAYANA, Gagan Mahal Colony, Hyderabad, India; and the
proposer.
* * *

389, C1978: 251] Proposed by Kenneth M, Wilke> Topeka, Kansas,

Prove that all the numbers in the sequence
100001, 10000100001, 1000010000100001,...
are composite.
I. Solution by A, Liuy University of Alberta,
For n> 2 , l e t
s n , 5c = looooiooooi...ooooi (n ones),
so that
= 10 5(n-l)+105(n-2) + < > > + 1 0 5 + 1 . l 0 ^
5
n,5 *•• 1 0 -i

. (ion-i)(io4%io3%io2n + io%i)^
(l)
io5-i

Using elementary divisibility tests, we find by inspection that ll\s t

2,5'
3Is , and ll\s g. If p is a prime divisor of S , we have p|lo 2 5 - 1 , so
lo 2 5 = l (mod p) and, by Fermat's Little Theorem, p = 25j+i. In fact, p being
odd, we must have j = 2k and p = 50& + 1. Testing primes of this form, we soon
find the prime factors 21401 and 25601, giving
S5 = 21401 '25601 '182521213001. (2)
 - 203 -

For n > 5 , each factor in the numerator of (l) is greater than the denominator,
so no amount of cancellation (to get rid of the denominator) will make (l) collapse
into a single prime. Thus S is composite for n > 5 , and hence for all rc>2.
II. Solution by Hayo Ahlburg* Benidorms Spain,
Let S be defined [as in solution I]. S _ is composite for n = 2, 3, 4, 5;
n, D n,5
in fact
Sn2,5c = 11 • 9091,
S 0 _ = 3 • 31 • 37 • 2906161,

St4,5 = 11 • 101 • 3541 • 9091 • 27961,

Sc _ = 21401 • 25601 • 182521213001.

We now assume n > 5 . If i? is the repunit with m ones (e.g. i?5 = m i l = Hi • 27l),
then we have R5c • sn, 5r = RSn
r . Since R \Rr and R contains at least one prime factor
n' 5rc n
p not occurring in any R <R , this factor p must divide S _ since it is not a
c
m n n,o
factor of R . So £ must be composite, unless p = 5 , But i? has only n digits,
0 ft y O rl> $ \J Tu

fewer than S r which has 5n - 4 digits; hence p < /? < S _ and so 5 _ is composite
r
w,5 * n n,5 n,5
also for all n > 5.
The factorization of sn c is not required here; but since I have it available,
b ,o
it may as well be recorded here:
S 6,5c = 3 • 7 • 11 • 13 • 31 • 37 • 211 • 241 • 2161 • 9091 • 2906161.
More information about repunits and their divisors can be found in Yates [13.
Also solved by ALLAN Wm. JOHNSON Jr., Washington, D.C.; MURRAY S. KLAMKIN,
Universiby of Alberta; LEROY F. MEYERS, The Ohio State University; and the proposer.
Partial solutions were received from HERMAN NYON, Paramaribo, Surinam; and KESIRAJU
SATYANARkYANA, Gagan Mahal Colony, Hyderabad, India.
Editor's comment.
Klamkin and the proposer located this problem in Gnedenko [2], where it is
stated that it was one of the problems presented in preparation for the eleventh
Moscow Olympiad. Klamkin also mentioned that the related problem, to prove that
S is composite for all rc£2, where
n
S = -io^
i i - — 1£ = loooioooi...oool (ft ones),
nA
io*-i
 - 2CW -

appeared in a 1975 Federal Republic of Germany Olympiad, This led him to a$(c for
which positive Integers r ts 1t true that $ is composite for all n*29 where

8n - ^ ^
>r iop-i
His question was parttally answered by Meyers, who proved the following:
If o>7) n^23 and r are positive integers, then
m _i
S
n,r - V f <3>
a -1
is composite except possibly when n is prime and r is a nonnegative integral power
of n (when a separate verification must be made). Furthermore, when n is composite,
say n - uv with u > 1 and v > 1, then
S =S •S (4)
n,r u9r v ,ur
and neither factor is 1.
Johnson, in fact, verified that this theorem is true when c-2 and r = 5 in
(3), that is, when the numbers in our proposal are interpreted as binary numbers.
Here again s is composite for all n>.2 and the special verification required
for sr r gives, in decimal notation,
5,5
S - 1082401 = 601 • 1801.
In our problem, <? = io, r = 5, and separate verification had to be made for rc=5,
yielding (2). When n - 6 and u = 2, v = 39 we get from (4)

6,5 2,5 3,10 2,5 i o 1 0 - l 2,5

and the factorizations given in solution II then yield

= 3 • 7 • 13 • 31 • 37 • 211 • 241 • 2161 • 2906161.

Similarly, if we put u = 3, v-2 instead, we get

5
6,5 = *3,5 * S 2 ( 1 5 = 5
3,5 ' i S f l = ^3,5 ' < " , S +*>.
whence
10 1 5 + 1 = 1000000000000001
= 7 • 11 • 33 • 211 • 241 • 2161 • 9091.
 - 205 -

A final remark. Several solvers gave the factorization (2), but only Klamkin
and Meyers seemed fully aware that the last factor there is a prime. Meyers
verified this fact himself, while Klamkin got the information from J. Brill hart
(University of Arizona) via M.V. Subbarao (University of Alberta).

REFERENCES
1, Samuel Yates, "Prime Divisors of Repunits", Journal of Recreational
Mathematics, 8 (1975) 33-38.
2. B.V. Gnedenko, "Mathematical Education in the U.S.S.R.", American
Mathematical Monthly„ 64 (July 1957) 389-408, esp. p. 397,
ft ft ft

3901 C1978: 251] Proposed by Gali Salvatore, Perkins, Quebec.

Show how to find the complete factorization of 2 3 8 + 1 using only pencil
and paper (no computers), having given that it consists of four distinct prime
factors, none repeated, one of which is 229.

I. Solution by Louis H. Cairolis student, Kansas State University, Manhattan,

Kansas,
We set rc=2 9 and use the given information to get

2 3 8 + 1 = 4a* + 1
= (2x2 - 2# + l)(2tf2 + 2a?+ 1)
= (2 1 9 - 2 1 0 + 1 ) ( 2 1 9 + 2 1 0 + 1)
= 523265 • 525313
= 5 • 104653 • 525313
= 5 • 229 • 457 • 525313.

II. Solution by Kenneth M, Wilke, Topeka, Kansas.

The given information about the number and nature of the factors is unnecessary:
the complete factorization can be found without it, using only pencil and paper.
We first find [as in solution I]
2 3 8 + 1 = 5 • 104653 • 525313.
We now use the following theorem of Legendre [l, p. 382]: Every prime divisor of
a + 1 is either of the form 2nk + 1 or divides a + 1 where w is the quotient of n
by an odd factor. Here this means that every prime divisor other than 2 2 + 1 = 5
 - 206 -

must be of the form 76& + 1. With or without a small table of primes (that's
allowable paper, in any case), it is easily determined that the only primes of the
form 76fc+1 that are less than /525313 « 725 are 229 and 457, and these are not
divisors of 525313. Thus 525313 is a prime and, since 229-457 = 104653, the
complete factorization is

Also solved by HAYO AHLBURG, Benidorm, Spain; G.P. HENDERSON, Campbellcroft,

Ontario; J.A.H. HUNTER, Toronto, Ontario; ALLAN Wm. JOHNSON Jr., Washington, D.C.
(two solutions); FRIEND H. KIERSTEAD, Jr., Cuyahoga Falls, Ohio; SIDNEY KRAVXTZ,
Dover, New Jersey; VIKTORS LINIS, University of Ottawa; LEROY F. MEYERS, The Ohio
State University; HERMAN NYON, Paramaribo, Surinam; BOB PRIELIPP, The University
of Wisconsin-Oshkosh; KENNETH S. WILLIAMS, Carleton University, Ottawa; and the
proposer.

Editor's comment.
Several solvers used the identity

n4n+2 „ ,„2n+l n+1 AS,„2n+l rtn+l

2 +1=(2 -2 + 1)(2 +2 + 1AS) ,

which Dickson Cl, p. 383] credits to H. LeLasseur and A. Aurifeuille, and earlier
!~1, p. 3821 to Sophie Germain (who, Ah!burg assures us, was born l April 1776 in
Paris and died 17 June 1831 in Paris). This is nothing but a straightforward
application of the identity

Hx* + 1 = (2x2 -2x + l)(2tf2 + 2# + l)

used in our solution I, an identity that used to be (but is no longer, alas!)
familiar to every high school student and which goes back at least to Euler (letter
to Goldbach, 1742) Cl, p. 381].
Here is an amusing anecdote about a related problem. Burton [2] quotes Landry
as writing in 1869: "No one of our numerous factorizations of the numbers 2n ± l
gave us as much trouble and labour as that of 2 5 8 +1." In fact, the number can be
broken down almost instantly, as in our solution I, to yield

2 5 8 + 1 = 5 • 107367629 • 536903681.
The last two factors are primes. This can be quickly verified with a table of
primes by the method of our solution II (there are only 52 primes of the form
116?: +1 that are less than /536903681, and none of them is a divisor of either
number).
Words worth uttering now: Landry! thou shouldst be living at this hour... .
 - 207 -

REFERENCES
1. Leonard Eugene Dickson, History of the Theory of Numbers, Vol. 1, Chelsea,
New York, 1952.
2. David M. Burton, Elementary Number Theory, Allyn and Bacon, Boston, 1976,
p. 238.
* * s%
3911 C1978: 2823 Proposed by Allan Wm. Johnson Jr., Washington, D.C.
Here is a word definition that is also an alphametic:
A
SUN
DRIED
GRAPE
RAISIN
Solve this decimal addition, bearing in mind that of course the digits of UP go that
way.
I. Solution by Charles W. Trigg, San Diego, California.
Immediately R = l, so 1 = 3 or 4 and
D + G = A + lo. W
Now from the columns, proceeding from the right,
A + D + E = lofc, (1)
U + E + P + f c = I +10m, (2)
I +A+w = ion. (3)

Since n>-l from (3), we must have 1 = 3. Also, m = l or 2 from (2), so A = 6 or 5.

Now, from (4),
1(7,9) or (9,7)
(D,G) I or
[(6,9), (7,8), (8,7), Or (9,6).
The only one of these pairs that does not produce duplicates in (l) is D = 9, G = 7,
with A = 6, E = 5,fc= 2, and w = i . From among the unidentified digits, o, 2, 4, 8,
equation (2) establishes that U = 2, P = M-. Finally N = o and S = 8.
The unique reconstruction is
6
820
91359
71645 .
163830
 - 208 -

II. Comment by the proposer.

This problem is the fruit of an effort that started when I happened on Crux
351 [1978: 259; 1979: 54] while eating raisins boxed by Sun-Maid. The effort
obviously went from frugivorous to fructiferous. Another fruit was the analogous
A
DRIED
PLUM ,
PRUNE
which has six solutions, but only two if the PLUM is picked in prime time. These
are

80538 80738
9721 and 9521 .
90263 90263

Also solved by HAYO AHLBURG, Benidorm, Spain; CHRIS DEARLING, Burlington,

Ontario; CLAYTON W. DODGE, University of Maine at Orono; ALLEN DRUZE, Brooklyn,
New York; J.A.H. HUNTER, Toronto, Ontario; FRIEND H. KIERSTEAD Jr., Cuyahoga Falls,
Ohio; EDGAR LACHANCE, Ottawa, Ontario; the following students of ARVON KYER and
JACK LeSAGE, Eastview Secondary School, Barrie, Ontario: JIM ENWRIGHT, DANIALL
GARBALLA, STEVE KOMAR, MIKE WALSH, IAN WILSON; PETER A. LINDSTROM, Genesee Community
College, Batavia, New York; HERMAN NYON, Paramaribo, Surinam; JEREMY D. PRIMER,
student, Columbia H.S., Maplewood, New Jersey? KENNETH M. WILKE, Topeka, Kansas;
and the proposer.
ft s'c ft

392. C1978: 282] Proposed by Steven R. Conrad, Benjamin N. Cardoso H.S.,

Bayside3 N.Y.
Find all natural numbers n for which nB -n2 is not divisible by 504.
I. Solution by Kenneth S. Williams, Carleton University, Ottawa.
We prove the following slight generalization:
Let p be an odd prime sueh that q - p2 -p + 1 is also a prime (necessarily odd).
Then
n*f ~n £ Q (mod 8p q) «=» n = 2 (mod 4) (1)

The proposed problem corresponds to the case p = 3, q~i.

Proof. By Format's L i t t l e Theorem, rfl"1 - 1 = o (mod q) when q\n\ hence, for
all n,
nin1'1 - D E O (mod q). (2)

If p \ n , then p 2 \ n 2 ; and if p \ n 9 then, by Euler's generalization of Fermat's Theorem,

rfl-^i = B P 2 - P - i -n^p2)-l = o (modp 2 )

 - 209 -

so t h a t , f o r a l l n ,
n2{nq"1 - l ) = o (mod p 2 ) . (3)

I f n i s odd, then rfl~ Is the square o f an odd i n t e g e r , and so

tfl'1 - 1 = o (mod 8 ) ; (if)

while i f n = o (mod 4 ) , then
n2 = o (mod 8 ) . (5)

Since p*q, it follow from (2)-(5) that

nq+1 -n2 = o (mod 8p2q),

except possibly when n = 2 (mod 4 ) . But in this last case, n*7 -1 is odd and
n = 4 (mod 8 ) , so that
n* + 1 ~ n 2 2 0 (mod 8p2<?).
This completes the proof of (l).

II. Comment by the proposer,

I have traced this problem back to the first round of the 13th Czechoslovakian
Mathematics Olympiad (1963-1964).
The simpler problem, to find the three smallest integers n>2 for which nB -nz
is not divisble by 504, was prepared by Harry Sitomer for use in the Nassau County,
New York Interscholastic Mathematics League Contest in 1972. In the allotted time
of 6 minutes, 2% of the students solved the problem.
The related problem, to prove that n 6 - l = o (mod 504) when n is not divisble
2, 3, or 7 occurs in Barnett Ci],

Also solved by HAYO AHLBURG, Benidorm, Spain; W.J. BLUNDON, Memorial University
of Newfoundland; CLAYTON W. DODGE, University of Maine at Orono; ALLEN DRUZE, Brooklyn,
New York; MICHAEL W. ECKER, Pennsylvania State University, Worthington Scranton Campus;
HERTA T. FREITAG, Roanoke, Virgina; T.J. GRIFFITHS, A.B. Lucas Secondary School,
London, Ontario; ALLAN Wm. JOHNSON Jr., Washington, D.C.; GILBERT W. KESSLER,^Canarsie
H.S., Brooklyn, New York; FRIEND H. KIERSTEAD Jr., Cuyahoga Falls, Ohio; ANDRE
LADOUCEUR, Ecole secondaire De La Salle, Ottawa; ANDY LIU, University of Alberta;
F.G.B. MASKELL, Algonquin College, Ottawa; L.F. MEYERS, The Ohio State University;
HERMAN NYON, Paramaribo, Surinam; BOB PRIELIPP, The University of Wisconsin-OshJcosh;
JEREMY D. PRIMER, student, Columbia H.S., Maplewood, New Jersey; DAVID STONE, Georgia
Southern College, Statesboro, Georgia; and KENNETH M. WILKE, Topeka, Kansas.

REFERENCE
1. I.A. Barnett, Elements of Number Theory, Prindle, Weber & Schmidt, Boston,
1969, p. 112, Exercise 3.
* * S'J
 - 210 -
393 • C1978: 283] Proposed by Sahib Ram Mandarin Indian Institute of
Technology9 Kharagpurj India.
If fn(a.) = (a.-a1)...(a.-a._1)(a.-a.+1)...(ai-an), prove that, for
k = 0, 1,.. ., n - 2,
k
n a.
^=l Jn ^
(This is given without proof in H.F. Baker's An Introduction to Plane Geometry,
Chelsea, Bronx, N.Y., 1971, p, 340.)

Solution by G.P. Henderson^ Campbellcrofts Ontario,

We assume the a. are all distinct and set

Fix) = (x -a. 1 )ix 2-a 0 )... ix -a n),

so that F'(a.) = / (a.). For any polynomial P of degree at most n - l , the Lagrange
interpolation formula gives the identity
n P(a.)F(x)
P(X) E
;L
t-=l
F'(a.)(x-a.)
^ t-
'

In particular, if o < k < n - 1 , we can set P(#) = x and get

fc _ n a^Fix)

and the required sum,

k
n a.
"t=l 1s

n-1
is then the coefficient of x on either side of (l). Thus we have

Sik) = 0 for 0 < k £ n - 2,

as required, and
Sik) = 1 forfc= n - 1.
For fc>n and k < o , £(fc) can be found from a recurrence formula involving the
elementary symmetric functions of the a.. Indeed, since
k k-1 k-2 , „ .n-1 k-n
a =
i °iai ~°2ai +
--'+^1> %ai >
where a. is the jth elementary symmetric function of the a., we have
J 1*
 - 211 -

S(k) = a±S(k-l) -a 2 $(fc-2) + ... + (-lf^o S(k - n ) , (2)

from which S(n), £(n + l)... can be obtained in succession. Furthermore, if all the
, that iis, if a * o , then (2) can be used to evaluate S(k) for k = ~1, -2,...
aa.. **o0,
Thus we find

5(n) = a i s S(tt + 1) = cTj - a 2 , S(rc + 2) = a' - 2a x a 2 + a3,...

and, if o * o ,
n

S(-l) = (-l)""1^. S(-2) = (-D""Vl/an' S(

" 3) = ( - " X ^ n - ^ - l ^ n

Also solved by W.J. BLUNDON, Memorial University of Newfoundland; G.C. GIRI,

Research Scholar, Indian Institute of Technology, Kharagpur, India; ALLAN Wm. JOHNSON
Jr., Washington, D.C.; MURRAY S. KLAMKIN, University of Alberta; KESIRAJU SATYANA-
RAYANA, Gagan Mahal Colony, Hyderabad, India; M. SELBY, University of Windsor;
KENNETH S. WILLIAMS, Carleton University, Ottawa (three solutions); and the proposer.

Editor's comment,
Klamkin found that our problem, as proposed, is equivalent to the following
theorem, which Mostowski and Stark CI] attribute to Euler:
If P is a polynomial of degree <> n - 2 and if al3 a2,.., 3 a are distinct
nvmbers and
Fix) = (x -a.)(x
1 2-a 9 )... (x -an ),
then
n P(a.)
VY i
L.
> F«(a.) = °"

No proof is given in Ci], but its truth follows from our own problem.
The relations derived in our featured solution are a fruitful source of algebraic
identities. Consider, for example,

a\ b" + ok + d\ = , .
a u
^ +
TB-a)(b-c)(b-d) (c-a)(c-b)(c-d) +
(d-a)(d-b)(d-c) ' ^ ^^^a
and
5 ^5 5
.^—_^ +_—_^—^ + _ _ — _ = (a+£+<?)3-2(a+2? + <?)(£><? +ca + ab) +abc.
(a-b)(a-c) (b-a)(b^c) {c~a){c-b)
Such identities occurred in profusion in some of the older treatises on algebra,
such as those of Hall and Knight and Chrystal in England, and Aubert et Papelier
in France. They were a frequent source of discouragement for an earlier generation
of students. But we see immediately that the first merely states that S(n) - ot
 - 212 -

when n = 4, and the second that 5(5) = a\ -2axa2+o% when n = 3. Any number of such
identities, each more monstrous than the last, can be manufactured at will for the
bafflement of the nonoognoscenti.

REFERENCE
1. A.M. Mostowski and M. Stark, Introduction to Higher Algebra, Polish
Scientific Publishers, Warsaw, 1964, p. 243.
it it it

SOME DON'T'S AND DO'S: A PROGRAMMER'S BLUES

As first-semesterer in 'Languages',
I've noticed that E-Fortran bandages
My programs with an eye to error:
'T should hardly cause me that much terror.
And yet, each time I run this through,
I get it stuck inside a 'DO'.
I get it stuck inside a 'DO'.
I get it stuck inside a 'DO'.
*** TEMPORARY PROGRAM INTERRUPT:
*** I'D HATE TO BE QUITE SO ABRUPT;
*** BUT REALLY NOW - YOU NINCOMPOOP,
*** YOU'VE STUCK YOURSELF INSIDE A LOOP.
Well, that's not all, let me continue;
Please hear me out if it's within you:
What really raises me to ire
Is merely my intense desire
That any run-amuck diagnostic
Need not be written quite so caustic.
Need not be written quite so caustic.
Need not be written quite so caustic.
*** QUITE SO CAUSTIC? WHAT THE $%$#*
*** THAT'S TWICE YOU'VE MANAGED TO GET STUCK,
*** ONCE MORE AND YOU'LL BE OUT OF LUCK.
Then, finally, I'd like to know
(If you can stand to stoop so low)
Just how you manage to malign
The 'READ' and 'WRITE' of my last line.
So there's the query; now get this:
My final line, and here it is...
My final line, and here it is...
My final line, and here it is...
*** THAT'S THRICE IT IS-YOU CARELESS SLOB,
*** MISSING BRAINS HAS STOPPED THIS JOB,

HANS HAVERMANN,
Weston, Ontario.