PROBLEMS AND SOLUTIONS

Source: The American Mathematical Monthly, Vol. 117, No. 8 (October 2010), pp. 741-748
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/10.4169/000298910X515820 .
Accessed: 24/03/2013 04:38

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to
The American Mathematical Monthly.

http://www.jstor.org

This content downloaded from 150.108.161.71 on Sun, 24 Mar 2013 04:38:49 AM

All use subject to JSTOR Terms and Conditions
 PROBLEMS AND SOLUTIONS
Edited by Gerald A. Edgar, Doug Hensley, Douglas B. West
with the collaboration of Itshak Borosh, Paul Bracken, Ezra A. Brown, Randall
Dougherty, Tamás Erdélyi, Zachary Franco, Christian Friesen, Ira M. Gessel, László
Lipták, Frederick W. Luttmann, Vania Mascioni, Frank B. Miles, Bogdan Petrenko,
Richard Pfiefer, Cecil C. Rousseau, Leonard Smiley, Kenneth Stolarsky, Richard
Stong, Walter Stromquist, Daniel Ullman, Charles Vanden Eynden, Sam Vandervelde,
and Fuzhen Zhang.

Proposed problems and solutions should be sent in duplicate to the MONTHLY

problems address on the back of the title page. Submitted solutions should arrive
at that address before February 28, 2011. Additional information, such as gen-
eralizations and references, is welcome. The problem number and the solver’s
name and address should appear on each solution. An asterisk (*) after the num-
ber of a problem or a part of a problem indicates that no solution is currently
available.

PROBLEMS
11523. Proposed by Timothy Chow, Princeton, NJ. Given boxes 1 through n, put balls
in k randomly chosen boxes. The score of a permutation π of {1, . . . , n} is the least
i such that box π(i) has a ball. Thus, if π = (3, 4, 1, 5, 2) with (n, k) = (5, 2), and
boxes 1 and 4 have balls, then π has score 2.
(a) A permutation π is fair if, regardless of the value of k, the probability that π scores
lower than the identity permutation equals the probability that it scores higher. Show
that π is fair if and only if for each i in [1, n], either π(i) > i and π −1 (i) > i, or
π(i) ≤ i and π −1 (i) ≤ i.
(b) Let f (n) be the number  of fair permutations of {1, . . . , n}, with the convention
that f (0) = 1. Show that ∞ n=0 f (n)x n
/n! = e x
sec(x).
(c) Assume now that n = m 3 with m ≥ 2, and the boxes are arranged in m rows of
length m 2 . Alice scans the top row left to right, then the row below it, and so on, until
she finds a box with a ball in it. Bob scans the leftmost column top to bottom, then the
next column, and so on. They start simultaneously and both check one box per second.
For which k are Alice and Bob equally likely to be the first to discover a ball?
11524. Proposed by H. A. ShahAli, Tehran, Iran. A vector v in Rn is short if v ≤ 1.
(a) Given six short vectors in R2 that sum to zero, show that some three of them have
a short sum.
(b)∗ Let f (n) be the least M such that, for any finite set T of short vectors in Rn that
sum to 0, and any integer k with 1 ≤ k ≤ |T |, there is a k-element subset S of T such
that  v∈S v ≤ M. The result of part (a) suggests f (2) = 1. Find f (n) for n ≥ 2.
11525. Proposed by Grigory Galperin, Eastern Illinois University, Charleston, IL, and
Yury Ionin, Central Michigan University, Mount Pleasant, MI.
(a) Prove that for each n ≥ 3 there is a set of regular n-gons in the plane such that
every line contains a side of exactly one polygon from this set.
doi:10.4169/000298910X515820

October 2010] PROBLEMS AND SOLUTIONS 741

This content downloaded from 150.108.161.71 on Sun, 24 Mar 2013 04:38:49 AM

All use subject to JSTOR Terms and Conditions
(b) Is there a set of circles in the plane such that every line in the plane is tangent to
exactly one circle from the set?
(c) Is there a set of circles in the plane such that every line in the plane is tangent to
exactly two circles from the set?
(d) Is there a set of circles in the plane such that every line in the plane is tangent to
exactly three circles from the set?
11526. Proposed by Marius Cavachi, “Ovidius” University of Constanta, Constanta,
Romania. Prove that there is no function f from R3 to R2 with the property that
 f (x) − f (y) ≥ x − y for all x, y ∈ R3 .
11527. Proposed by Cezar Lupu, student, University of Bucharest, Bucharest, Roma-
nia. Prove that in an acute triangle with sides of length a, b, c, inradius r , and circum-
radius R,
a2 b2 c2 3 R
+ + ≥ · .
b +c −a
2 2 2 c +a −b
2 2 2 a +b −c
2 2 2 2 r

11528. Proposed by Alina Sı̂ntămărian, Technical University of Cluj-Napoca, Cluj-

Napoca, Romania. Let p, a, and b be positive integers with a < b. Consider a sequence
xn defined by the recurrence nxn+1 = (n + 1/ p)xn and an initial condition x1
= 0.
Evaluate
xan + xan+1 + · · · + xbn
lim .
n→∞ nxan

11529. Proposed
 by Walter Blumberg, Coral Springs, FL. For n ≥ 1, let An =
n  k 2 
3 k=1 n − n 2 . Let p and q be distinct primes with p ≡ q (mod 4). Show
that A pq = A p + Aq − 2.

SOLUTIONS

Splitting Elements of Set Systems

11372 [2008, 568]. Proposed by Jennifer Vandenbussche and Douglas B. West, Uni-
versity of Illinois at Urbana-Champaign, Urbana, IL. In a family of finite sets, let a
splitting element be an element that belongs to at least two of the sets and is omit-
ted by at least two of the sets. Determine the maximum size of a family of subsets of
{1, . . . , n} for which there is no splitting element.
Solution by David Gove, California State University, Bakersfield, CA. The maximum
size is n + 1. Consider a largest such family. Removing x from the sets it lies in and
adding it to the others yields another such family. Hence we may assume that each
element appears in at most one of the sets. If any of the sets has more than one element,
then we can obtain a bigger family by replacing that set by its singleton subsets. Thus
the family consisting of the empty set and the singleton sets is a largest such family.
Editorial comment. By the argument above, there are 2n extremal families. Marian
Tetiva sent a thorough discussion of a more general problem. Let gs (n) be the maxi-
mum size of a family of subsets of {1, . . . , n} such that every element appears in at
most s sets or avoids at most s sets; the stated problem is g1 (n) = n + 1, and clearly
g0 (n) = 1. By the complementation argument above, we may equivalently seek the
largest family such that every element appears in at most s sets. Tetiva proved a bound

742
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 117

This content downloaded from 150.108.161.71 on Sun, 24 Mar 2013 04:38:49 AM

All use subject to JSTOR Terms and Conditions
and conjectured equality. Intuitively, the idea is that one should take all the small sets
until the bound on the number of appearances of each element is reached. For example,
k n−1
if s = i=1 i−1
, then one should take all the sets of size at most k. When s is not of
this form, the exact solution is more difficult.
Also solved by D. Beckwith, P. Corn, D. L. Craft, C. Curtis, P. P. Dályay (Hungary), K. David & P. Fricano,
D. Degiorgi (Switzerland), J. Gately, J. Guerreiro (Portugal), H. S. Hwang (Korea), K. Kneile, J. H. Lindsey II,
O. P. Lossers (Netherlands), R. Martin (Germany), M. D. Meyerson, J. H. Nieto (Venezuela), R. E. Prather,
T. Rucker, V. Rutherfoord, K. Schilling, E. Schmeichel, B. Schmuland (Canada), R. Stong, J. Swenson,
M. Tetiva (Romania), B. Tomper, Fisher Problem Group, Szeged Problem Group “Fejéntaláltuka” (Hungary),
GCHQ Problem Solving Group (U. K.), Houghton College Problem Solving Group, Microsoft Research Prob-
lems Group, NSA Problems Group, and the proposers.

A Determinant Generated by a Polynomial

11377 [2008, 664]. Proposed by Christopher Hillar, Texas A&M University, College
Station, TX and Lionel Levine, Massachusetts Institute of Technology, Cambridge, MA.
Given a monic polynomial p of degree n with complex coefficients, let A p be the
(n + 1) × (n + 1) matrix with p(−i + j) in position (i, j), and let D p be the determi-
nant of A p . Show that D p depends only on n, and find its value in terms of n.
Solution by John H. Lindsey II, Cambridge, MA. The value of D p is (n!)n+1 , which
we prove by induction on n. The result is trivial when n = 0. For n > 0, use indices
0, . . . , n for the rows and columns of A p . In A p , let C j be column j and Ri be row i.
Given a function f , define  f by  f (k) = f (k + 1) − f (k). By induction on n, if f
is a monic polynomial of degree
n n, then n f (x) = n! for all x.
n
Replacing Cn with j =0 j (−1) C j does not change the determinant, but it turns
n− j

the ith entry of column n into n p(−i), which equals n!. Now for 0 ≤ i ≤ n − 1 in
order, subtract the next row from Ri , replacing Ri with Ri − Ri+1 . This puts 0 in the last
column, except for the last row. For j < n, the new entry ai, j is p(−i + j) − p(−i −
1 + j), which equals p(−i − 1 + j). Since p has leading coefficient n, the upper
left (n − 1)-by-(n − 1) block has the form n A f , where f (x) = (1/n)p(x − 1).
Since f is a monic polynomial with degree n − 1, by the induction hypothe-
sis D f = (n − 1)!n . Expanding the altered D p down the last column yields D p =
n! n n (n − 1)!n = n!n+1 .
Editorial comment. Solvers used a variety of methods, including Vandermonde deter-
minants. Roger Horn proved a substantial generalization. Given a matrix A, let p(A)
denote the entrywise application of the polynomial p to A; that is, the (i, j)-entry of
p(A) is p(ai, j ). For x ∈ Cn+1 , let A(x) be the matrix given by ai, j = xi + j − 1. If p
is a monic polynomial of degree n, then


(−1)(n+1)/2(n!)
n

det p(A(x)) = (xi − x j ) n−1 , (1)

i> j i=1 i!

which depends only on x and n, not p. The originally stated problem is the case x =
(0, −1, . . . , −n)T .
Also solved by D. Beckwith, R. Chapman (U. K.), P. Corn, P. P. Dályay (Hungary), J. Grivaux (France), J. Hart-
man, C. C. Heckman, R. A. Horn, R. Howard, G. Keselman, O. Kouba (Syria), S. C. Locke, O. P. Lossers
(Netherlands), K. McInturff, J. H. Nieto (Venezuela), É. Pité (France), C. R. Pranesachar (India), M. A. Prasad
(India), N. C. Singer, J. H. Smith, A. Stadler (Switzerland), V. Stakhovsky, R. Stong, T. Tam, M. Tetiva (Ro-
mania), B. Tomper, M. Vowe (Switzerland), L. Zhou, BSI Problems Group (Germany), FAU Problem Solving
Group, GCHQ Problem Solving Group (U. K.), NSA Problems Group, and the proposers.

October 2010] PROBLEMS AND SOLUTIONS 743

This content downloaded from 150.108.161.71 on Sun, 24 Mar 2013 04:38:49 AM

All use subject to JSTOR Terms and Conditions
 The Column Space of a Very Nilpotent Matrix
11379 [2008, 664]. Proposed by Oskar Maria Baksalary, Adam Mickiewicz Univer-
sity, Poznań, Poland, and Götz Trenkler, Technische Universität Dortmund, Dortmund,
Germany. Let A be a complex matrix of order n whose square is the zero matrix. Show
that R(A + A∗ ) = R(A) + R(A∗ ), where R(·) denotes the column space of a matrix
argument.
Solution by M. Andreoli, Miami Dade College, Miami, FL. Note first that A2 = 0
implies R(A) ⊆ N (A), where N (A) is the nullspace of A. This holds because y =
Ax implies Ay = A2 x = 0.
For y ∈ R(A + A∗ ), there exists x such that y = (A + A∗ )x = Ax + A∗ x. Hence
y ∈ R(A) + R(A∗ ), and we conclude that R(A + A∗ ) ⊆ R(A) + R(A∗ ).
Conversely, for y ∈ R(A) + R(A∗ ), there exist vectors x1 and x2 such that y =
Ax1 + A∗ x2 . Since N (A∗ ) and R(A) are orthogonal complements in Cn , there exist
vectors u ∈ R(A) and v ∈ N (A∗ ) such that x1 − x2 = u + v. Since R(A) ⊆ N (A),
we have u ∈ N (A). Letting x = x1 − u = x2 + v, we have
(A + A∗ )x = Ax + A∗ x = A(x1 − u) + A∗ (x2 + v)
= Ax1 − Au + A∗ x2 + A∗ v = Ax1 + A∗ x2 = y.
Thus y ∈ R(A + A∗ ), and hence R(A) + R(A∗ ) ⊆ R(A + A∗ ).
Also solved by M. Bataille (France), P. Budney, R. Chapman (U. K.), P. Corn, C.-K. Fok, J. Freeman, J.-
P. Grivaux (France), J. Hartman, E. A. Herman, R. A. Horn, O. Kouba (Syria), C. Lanski, J. H. Lindsey II,
O. P. Lossers (Netherlands), R. Martin (Germany), I. Pinelis, É. Pité (France), N. C. Singer, J. H. Smith,
R. Stong, J. Stuart, F. Vrabec (Austria), BSI Problems Group (Germany), GCHQ Problem Solving Group
(U. K.), and the proposers.

A Generalized Binomial Coefficient

11380 [2008, 665]. Proposed by Hugh Montgomery, University of Michigan, Ann
Arbor, MI, and Harold
 S. Shapiro, Royal Institute of Technology, Stockholm, Sweden.

For x ∈ R, let xk = k!1 k−1j =0 (x − j). For k ≥ 1, let ak be the numerator and qk the de-
 
nominator of the rational number −1/3 k
expressed as a reduced fraction with qk > 0.
(a) Show that qk is a power of 3.
(b) Show that ak is odd if and only if k is a sum of distinct powers of 4.
Solution by Stephen M. Gagola Jr., Kent State University, Kent, OH. We prove more
−1/m  that if m > 1 and m + 1 is a power of a prime p, and the rational number
generally
k
has numerator ak and denominator qk in lowest terms with qk > 0, then

(a ) all prime factors of qk divide m, and
(b ) p  ak if and only if k is a sum of distinct powers of m + 1.
The stated problem isthe case m = 3, where p = 2.
(a ) For clarity, let ck = −1/m
k
= ak /qk . In the formal power series ring Q[[x]],

∞
(1 + x)−1/m = ck x k . (1)
k=0

Therefore,
m

∞
∞
∞
(−1)k x k = (1 + x)−1 = ck x k = ci1 · · · cim x k , (2)
k=0 k=0 k=0

744
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 117

This content downloaded from 150.108.161.71 on Sun, 24 Mar 2013 04:38:49 AM

All use subject to JSTOR Terms and Conditions
where the inner sum extends over all m-tuples (i 1 , . . . , i m ) of nonnegative integers
summing to k. Exactly m such m-tuples have k as an entry. Equating coefficients of x k
in (2) then yields

mck + ci1 · · · cim = (−1)k , (3)

where the sum extends over m-tuples (i 1 , . . . , i m ) with sum k and entries less than k.
Let Rm = i≥0 (1/m i )Z. Note that Rm is the subring of Q consisting of all rational
numbers whose denominators factor into primes dividing m. Also, c0 = 1, so c0 ∈ Rm .
Since m is a unit of Rm , (3) yields ck ∈ Rm for all k, inductively. Thus (a ) follows.
(b ) View (1) and (2) above in the formal power series ring Rm [[x]]. We write
f (x) ≡ g(x) when f (x) − g(x) = ph(x) for some power series h(x) ∈ Rm [[x]].
Since f (x) p ≡ f (x p ) for all f (x) ∈ Rm [[x]], also f (x)m+1 ≡ f (x m+1 ). Therefore,
m+1

∞
∞
−1/m −1/m m+1
ck x = (1 + x)
k
= (1 + x)((1 + x) ) = (1 + x) ck x k

k=0 k=0

∞
∞
≡ (1 + x) ck x (m+1)k = (ck x (m+1)k + ck x (m+1)k+1 ). (4)
k=0 k=0

We conclude that ck ≡ 0 mod p if k is not congruent to 0 or 1 modulo m + 1, and the

same holds for ak .
Note that c0 = 1 and c1 = −1/m ≡ 1 mod p. Hence p divides neither a0 nor a1 . For
k > 1, if m + 1 divides k or k − 1, then write k = (m + 1)k  + , where  ∈ {0, 1}.
Note that k is a sum of distinct powers of m + 1 if and only if k  is. The congruence in
(4) implies that ck ≡ ck  mod p, and (b ) follows by induction.
Also solved by R. Chapman (U. K.), H. Chen, P. Corn, P. P. Dályay (Hungary), Y. Dumont (France), E. Errthum,
S. M. Gagola Jr., J. H. Lindsey II, O. P. Lossers (Netherlands), J. Minkus, M. A. Prasad (India), B. Schmuland
(Canada), N. C. Singer, J. H. Smith, A. Stadler (Switzerland), R. Stong, M. Tetiva (Romania), Z. Vörös (Hun-
gary), BSI Problems Group (Germany), GCHQ Problem Solving Group (U. K.), NSA Problems Group, and
the proposers.

Convergence of a Prime-denominated Series

11384 [2008, 757]. Proposed by Moubinool Omarjee, Lycée Jean-Lurçat, Paris,
France. Let pn denote the nth prime. Show that
√

∞
(−1) n

n=1
pn

converges.
Solution by Greg Martin, University of British Columbia, Vancouver, CA. Let S N =
 N (−1)√n
n=1 pn
. It suffices to show that the subsequence {S M 2 −1 : M ≥ 1} converges,
since S N is between S M 2−1 and S(M+1)2−1 for N between M 2 − 1 and (M + 1)2 − 1.
M
However, S M 2 −1 = m=2 Tm , where
√

m 2 −1
(−1) n
m 2 −1
1
Tm = = (−1)m−1 .
n=(m−1)2
pn n=(m−1)2
pn

October 2010] PROBLEMS AND SOLUTIONS 745

This content downloaded from 150.108.161.71 on Sun, 24 Mar 2013 04:38:49 AM

All use subject to JSTOR Terms and Conditions
Since {Tm : m ≥ 1} alternates in sign, it suffices to show that lim Tm = 0, by the alter-
nating series test. Using the crude inequality pn > n, we obtain

m 2 −1
1 1
m 2 −1
2m − 1
|Tm | < < 1= ,
n=(m−1)2
n (m − 1)2 n=(m−1)2
(m − 1)2

and thus lim Tm = 0.

A similar proof
√ works if the sequence of primes is replaced by an arbitrary sequence
q satisfying qn / n → ∞.
Editorial comment. Many solvers used detailed information about the distribution of
the prime numbers, but the proof above shows that this is unnecessary.
Also solved by R. Bagby, H. Chen, P. P. Dályay (Hungary), Y. Dumont (France), V. V. Garcia (Spain), S. James
(Canada), O. Kouba (Syria), K. Y. Li (China), J. Oelschlager, P. Perfetti (Italy), É. Pité (France), Á. Plaza
(Spain), C. R. Pranesachar (India), M. T. Rassias (Greece), V. Schindler (Germany), B. Schmuland (Canada),
N. C. Singer, A. Stadler (Switzerland), R. Stong, T. Tam, R. Tauraso (Italy) & M. Lerma, D. B. Tyler, J. Vinuesa
(Spain), Z. Vörös (Hungary), GCHQ Problem Solving Group (U. K.), and the proposer.

Capturing Eigenvalues in an Interval

11387 [2008, 758]. Proposed by Oskar Maria Baksalary, Adam Mickiewicz Univer-
sity, Poznań, Poland, and Götz Trenkler, Technische Universität Dortmund, Dortmund,
Germany. Let Cn,n denote the set of n × n complex matrices. Determine the shortest
interval [a, b] such that if P and Q in Cn,n are nonzero orthogonal projectors, that is,
Hermitian idempotent matrices, then all eigenvalues of P Q + Q P belong to [a, b].
Solution I by O. P. Lossers, Eindhoven University of Technology, Einhoven, The
Netherlands. The eigenvalues of P Q + Q P lie in [− 14 , 2]. The matrix P + Q is Her-
mitian, and hence there is an orthonormal basis of its eigenvectors. The eigenvalues
of P + Q are real and in [0, 2], since |(P + Q)x| ≤ |P x| + |Qx| ≤ 2|x|. The matrix
P Q + Q P equals (P + Q)2 − (P + Q) and thus has the same eigenvectors as P + Q,
with eigenvalues of the form λ2 − λ with 0 ≤ λ ≤ 2. It follows that the eigenvalues of
P Q + Q P lie in [−1/4, 2].
The maximum is attained when P and Q both equal the identity matrix, while the
minimum is attained for the projections on two lines intersecting at an angle of π/3.
Solution II by Fuzhen Zhang, Nova Southeastern University, Fort Lauderdale, FL.
Since |(P Q + Q P)x| ≤ |P Qx| + |Q P x| ≤ 2|x| for all x, the eigenvalues are at most
2. For the lower bound, write X ≥ Y if X and Y are Hermitian and X − Y is positive
semidefinite. Note that
 
1 2 1 1
0 ≤ P + Q − I = P 2 + Q 2 + I + P Q + Q P − P − Q = I + P Q + Q P.
2 4 4

It follows that P Q + Q P ≥ − 14 I , and therefore each eigenvalue of P Q + Q P is at

least − 14 .
For the extreme cases,
 √ taking
 P = Q = I gives the largest eigenvalue 2. Setting
1 0
P = 0 0 and Q = 4 3 3 yields − 14 as an eigenvalue.
1 √1 3

Editorial comment. The part P Q + Q P ≥ − 14 I of this problem appeared in F. Zhang,

Linear Algebra: Challenging Problems for Students (2nd ed.), Johns Hopkins Univer-
sity Press, Baltimore, 2009, p. 81.

746
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 117

This content downloaded from 150.108.161.71 on Sun, 24 Mar 2013 04:38:49 AM

All use subject to JSTOR Terms and Conditions
Also solved by R. Chapman (U. K.), J. Freeman, J.-P. Grivaux (France), J. Hartman, E. A. Herman, O. Kouba
(Syria), T. Laffey & H. Šmigoc (Ireland), J. H. Lindsey II, M. Omarjee (France), R. Stong, S. E. Thiel,
N. Thornber, Szeged Problem Solving Group “Fejéntaláltuka” (Hungary), GCHQ Problem Solving Group
(U. K.), and the proposers.

Distinct Multisets with the Same Pairwise Sums

11389 [2008, 758]. Proposed by Elizabeth R. Chen and Jeffrey C. Lagarias, Univer-
sity of Michigan, Ann Arbor, MI. Given a multiset A = {a1 , . . . , an } of n real num-
bers (not necessarily distinct), define the sumset S(A) of A to be {ai + a j : 1 ≤ i <
j ≤ n}, a multiset with n(n − 1)/2 not necessarily distinct elements. For instance, if
A = {1, 1, 2, 3}, then S(A) = {2, 3, 3, 4, 4, 5}.
(a) When n is a power of 2 with n ≥ 2, show that there are two distinct multisets A1
and A2 such that S(A1 ) = S(A2 ).
(b) When n is a power of 2 with n ≥ 4, show that if r distinct multisets A1 , . . . , Ar all
have the same sumset, then r ≤ n − 2.
(c*) When n is a power of 2 with n ≥ 4, can there be as many as 3 distinct multisets
with the same sumset?
(Distinct multisets are known to have distinct sumsets when n is not a power of 2.)
Solution by BSI Problems Group, Bonn, Germany.
(a) We recursively construct multisets Am and Bm of size 2m for m ≥ 0. For m ≥ 0,
choose arbitrary positive cm . Let A0 = {0} and B0 = {c0 }. For m > 0, let Am =
Am−1 ∪ {b + cm : b ∈ Bm−1 } and Bm = Bm−1 ∪ {a + cm : a ∈ Am−1 }. Inductively,
|Am | = |Bm | = 2m and S(Am ) = S(Bm ). Also min Am = 0 < min Bm , which yields
Am
= Bm .
(b) First we prove three claims. Let A = {a1 , . . . , an } with a1 ≤ · · · ≤ an , and let
S(A) = {s1 , . . . , sn(n−1)/2 } with s1 ≤ · · · ≤ sn(n−1)/2 .
Claim 1: a2 + a3 ∈ {s3 , . . . , sn }. Since a1 + a2 ≤ a1 + a3 ≤ a2 + a3 , we have a2 +
a3 ≥ s3 . Also, the only sums that can be strictly smaller than a2 + a3 are {a1 + ai : 2 ≤
i ≤ n}. Thus a2 + a3 ≤ sn .
Claim 2: Let B = {b1 , . . . , bn } with b1 ≤ · · · ≤ bn . If a1 = b1 and S(A) = S(B),
then A = B. We prove ai = bi by induction on i. Let A(i) = {a1 , . . . , ai } and B(i) =
{b1 , . . . , bi }. If A(i − 1) = B(i − 1), then a1 + ai and b1 + bi are both minimal among
S(A) − S(A(i − 1)). Thus ai+1 = bi+1 .
Claim 3: Let B = {b1 , . . . , bn } with b1 ≤ · · · ≤ bn . If a2 + a3 = b2 + b3 and
S(A) = S(B), then A = B. Since the two smallest sums from the two sets are
equal, a1 + a2 = s1 = b1 + b2 and a1 + a3 = s2 = b1 + b3 . With the hypothesis
a2 + a3 = b2 + b3 , we have a1 = b1 . Claim 2 now applies.
Given these claims, let A1 , . . . , An−1 be multisets of size n having the same sum-
set. Write Ak = {a1(k) , . . . , an(k) } with a1(k) ≤ · · · ≤ an(k) . By Claim 1, there are at most
n − 2 values for the sum of the second and third smallest elements. By the pigeonhole
principle, there exist distinct k and l such that a2(k) + a3(k) = a2(l) + a3(l) . By Claim 3,
Ak = Al . Thus at most n − 2 multisets can have the same sumset.
(c) The answer is yes. Let A = {0, 4, 4, 4, 6, 6, 6, 10}, B = {2, 2, 2, 4, 6, 8, 8, 8}, and
C = {1, 3, 3, 3, 7, 7, 7, 9}. With exponents denoting multiplicity, S(A), S(B), and
S(C) all equal {4(3) , 6(3) , 8(3) , 10(10) , 12(3) , 14(3) , 16(3) }.
Editorial comment. The GCHQ Problem Solving Group solved part (a) by letting A1
be the set of nonnegative integers less than 2n whose binary expansion has an even
number of ones and setting A2 = {0, 1, . . . , 2n − 1} − A1 . This results from the con-
struction given above by setting cm = 2m .

October 2010] PROBLEMS AND SOLUTIONS 747

This content downloaded from 150.108.161.71 on Sun, 24 Mar 2013 04:38:49 AM

All use subject to JSTOR Terms and Conditions
 For part (c), Daniele Degiorgi gave the example A = {0, 6, 7, 9, 11, 13, 14, 20},
B = {1, 5, 6, 8, 12, 14, 15, 19}, and C = {2, 4, 5, 9, 11, 15, 16, 18}, showing that it
can be solved with sets (i.e., multisets with no repeated elements).
It remains open whether there are quadruples of multisets of size greater than 2
with the same sumset, or whether there are triples of multisets of any size greater than
2 other than 8 with the same sumset. Richard Stong showed that the search for such
triples can be restricted to multisets whose size is an odd power of 2.
Also solved by D. Degiorgi (Switzerland), R. Stong, and the GCHQ Problem Solving Group (U. K.). Parts (a)
and (b) solved also by O. P. Lossers (Netherlands), M. A. Prasad (India), Microsoft Research Problems Group,
and the proposers.

Tetrahedral Cevians
11405 [2009, 82]. Proposed by Ovidiu Furdui, Campia Turzii, Cluj, Romania. Let P be
an interior point of a tetrahedron ABCD. When X is a vertex, let X  be the intersection
of the opposite face with the line through X and P. Let X P denote the length of the
line segment from X to P.
(a) Show that P A · P B · PC · P D ≥ 81P A · P B  · PC  · P D  , with equality if and
only if P is the centroid of ABC D.
(b) When X is a vertex, let X  be the foot of the perpendicular from P to the plane of
the face opposite X . Show that P A · P B · PC · P D = 81P A · P B  · PC  · P D  if
and only if the tetrahedron is regular and P is its centroid.
Solution by Kit Hanes, Bellingham, WA. We will consider the more general case of an
let Ai be the
n-simplex with vertices A0 , . . . , An . Let P be a point in the interior, and
pointwhere the line Ai P meets the face opposite Ai . We will show that nk=0 P Ak ≥
n n+1 nk=0 P Ak , with equality if and only if P is the centroid of the simplex. Let P =
a0 A0 + · · · + an An where a0 + · · · + an = 1 and each ai is positive. For each j, Aj is
a convex combination of the Ai with A j omitted and P is a convex linear combina-
tion of A j and Aj . Hence P = a j A j + (1 − a j )Aj . Hence P A j /P Aj = (1 − a j )/a j .
 
The inequality of (a) is equivalent to nj =0 (1 − a j ) ≥ n n+1 nj =0 a j . This inequality
follows by applying the arithmetic-geometric mean inequality
1 − aj a0 + · · · + aj + · · · + an 
= ≥ n a0 . . . aj . . . an
n n
to each term separately and taking the product. (Here, the hats indicate that the hatted
term is to be skipped.) Equality holds if and only if all the ai are equal, and hence
ai = 1/(n + 1) for all i and P is the centroid of the simplex. For part (b), note that
P Ai ≥ P Ai with equality if and only if Ai = Ai , i.e., if and only if the line P Ai is
an altitude of the simplex. Hence the stated equality holds exactly when P is both the
centroid and the orthocenter of the simplex. That this is equivalent to the simplex being
regular is half of Problem 11087 from this M ONTHLY, December, 2005.
Editorial comment. Part (a) of this problem is the generalization from triangles to
tetrahedra of Problem 11325, this M ONTHLY, November, 2007.
Also solved by S. Amghibech (Canada), M. Bataille (France), M. Can, R. Chapman (U. K.), P. P. Dályay
(Hungary), O. Geupel (Germany), M. Goldenberg & M. Kaplan, J. Grivaux (France), K. Hanes, J. G. Heuver
(Canada), B.-T. Iordache (Romania), O. Kouba (Syria), J. H. Lindsey II, O. P. Lossers (Netherlands), J. Schaer
(Canada), R. Stong, M. Tetiva (Romania), Z. Vörös (Hungary), M. Vowe (Switzerland), GCHQ Problem Solv-
ing Group (U. K.), and the proposer.

748
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 117

This content downloaded from 150.108.161.71 on Sun, 24 Mar 2013 04:38:49 AM

All use subject to JSTOR Terms and Conditions