Problems and Solutions

Source: The American Mathematical Monthly, Vol. 119, No. 10 (December 2012), pp. 880-887
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/10.4169/amer.math.monthly.119.10.880 .
Accessed: 27/03/2013 20:37

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 137.140.1.131 on Wed, 27 Mar 2013 20:37:53 PM

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 Mike Bennett, Itshak Borosh, Paul Bracken, Ezra A. Brown,
Randall Dougherty, Tamás Erdélyi, Zachary Franco, Christian Friesen, Ira M. Ges-
sel, László Lipták, Frederick W. Luttmann, Vania Mascioni, Frank B. Miles, Richard
Pfiefer, Dave Renfro, 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. Proposed problems should never
be under submission concurrently to more than one journal. Submitted solutions
should arrive before April 30, 2013. Additional information, such as general-
izations and references, is welcome. The problem number and the solver’s name
and address should appear on each solution. An asterisk (*) after the number of
a problem or a part of a problem indicates that no solution is currently available.

PROBLEMS
11677. Proposed by Albert Stadler, Herrliberg, Switzerland. Evaluate
∞ 
Y √ √ 
1 + 2e−m 3 cosh(mn/ 3) .
n=1

11678. Proposed by Farrukh Ataev Rakhimjanovich, Westminster International Uni-

versity in Tashkent, Tashkent, Uzbekistan. Let Fk be the kth Fibonacci number, where
F0 = 0 and F1 = 1. For n ≥ 1 let An be an (n + 1) × (n + 1) matrix with entries a j,k
given by a0,k = ak,0 = Fk for 0 ≤ k ≤ n and by a j,k = a j−1,k + a j,k−1 for j, k ≥ 1.
Compute the determinant of An .
11679. Proposed by Tim Keller, Orangeville, CT. Let n be an integer greater than 2,
and let a2 , . . . , an be positive real numbers with product 1. Prove that
n
Y 2  n 2n−1
(1 + ak )k > .
k=2
e 2

11680. Proposed by Benjamin Bogoşel, University of Savoie, Savoie, France, and

Cezar Lupu, University of Pittsburgh, Pittsburgh, PA. Let x1 , . . . , xn be nonnegative
real numbers. Show that
n
!4 n n
X xi X x i x j X x k xl
≤ 2π 2 .
i=1
i i, j=1
i + j k,l=1
(k + l) 3

11681. Proposed by Des MacHale, University College Cork, Cork, Ireland. For any
group G, let AutG denote the group of automorphisms of G.
http://dx.doi.org/10.4169/amer.math.monthly.119.10.880

880 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 119

This content downloaded from 137.140.1.131 on Wed, 27 Mar 2013 20:37:53 PM

All use subject to JSTOR Terms and Conditions
(a) Show that there is no finite group G with |AutG| = |G| + 1.
(b) Show that there are infinitely many finite groups G with |AutG| = |G|.
(c) Find all finite groups G with |AutG| = |G| − 1.
11682. Proposed by Ovidiu Furdui, Technical University of Cluj-Napoca, Cluj-
Napoca, Romania. Compute
∞ ∞
!2
X X (−1)k−1
(−1)n .
n=0 k=1
n+k

11683. Proposed by Raimond Struble, Santa Monica, CA. Given a triangle ABC, let
FC be the foot of the altitude from the incenter to AB. Define FB and FC similarly.
Let C A be the circle with center A that passes through FB and FC , and define C B and
CC similarly. The Gergonne point of a triangle is the point at which segments AFA ,
B FB , and C FC meet. Determine, up to similarity, all isosceles triangles such that the
Gergonne point of the triangle lies on one of the circles C A , C B , or CC .

SOLUTIONS

Matrices Whose Powers Have Bounded Entries

11530 [2010, 834]. Proposed by Pál Peter Dályay, Szeged, Hungary Let A be an
m × m matrix with nonnegative entries aQ i, j and with the property that there exists
a permutation σ of {1, . . . , m} for which i=1 m
ai,σ (i) ≥ 1. Show that the union over
n ≥ 1 of the set of entries of An is bounded if and only if some positive power of A is
the identity matrix.
Solution by John H. Smith, Needham, MA. If some positive power of A is the identity,
then the powers An run over a finite set and hence have bounded entries. We prove the
converse.
Define A0 to agree with A in the positions (i, σ (i)) and be 0 elsewhere. Let A00 =
A − A0 ; note that A00 has nonnegative entries. Let k be the least common multiple
of the lengths of the cycles in σ , so k is the order of σ . Let B = Ak − A0k ; since
A = A0 + A00 , the entries of B are nonnegative.
The matrix A0k is diagonal. Let b j be the product of the entries (i, σ (i)) in A corre-
sponding to the jth cycle in σ , and let l j be its length. The entries on the diagonal of
k/l
A0k corresponding to this cycle are b j j .
If b j > 1, then the entries of A0kn are unbounded as n grows. Since Akn − A0kn has
nonnegative entries, the entries of Akn would also be unbounded. Hence b j ≤ 1 for
each j; since the product over all cycles is at least 1, we have b j = 1 for each j.
Hence Ak = I + C, where C has nonnegative entries. If C 6 = 0, then I + nC has
unbounded entries (over all n). The same also holds for Akn , which equals I + nC +
Dn , where Dn has nonnegative entries. Hence boundedness of the entries of An over
all n implies Ak = I .
Also solved by P. Budney, R. Chapman (U. K.), D. Constales (Belgium), D. Fleischman, E. A. Herman,
R. A. Horn & J. L. Stuart, J. H. Lindsey II, O. P. Lossers (Netherlands), R. Martin (Germany), S. Pierce,
É. Pité (France), R. E. Prather, A. R. Schep, J. Simons (U. K.), R. Stong, M. Tetiva (Romania), Z. Vörös
(Hungary), Barclay Capital Problems Solving Group, Con Amore Problem Group (Denmark), NSA Problems
Group, and the proposer.

December 2012] PROBLEMS AND SOLUTIONS 881

This content downloaded from 137.140.1.131 on Wed, 27 Mar 2013 20:37:53 PM

All use subject to JSTOR Terms and Conditions
 A Sufficient Condition for a Boolean Ring
11533 [2010, 835]. Proposed by Erwin Just (emeritus), Bronx Community College of
the City College of New York, Bronx, NY. Let t be a positive integer and let R be a ring,
not necessarily having an identity element, such that x + x 2t+1 = x 2t + x 10t+1 for each
x in R. Prove that R is a Boolean ring, that is, x = x 2 for all x ∈ R.
Solution by Richard Stong, CCR, San Diego CA. Summing the given identity for x and
−x yields 2x 2t = 0. The identity for 2x yields

2x = 2x 2t 22t−1 − 22t x + 210t x 8t+1 = 0.

 

Thus, R has characteristic 2. Now fix x and define a linear map Z2 [X ] → R by

p(X ) 7→ x p(x). The kernel of this map is an ideal I in Z2 [X ]. Since Z2 is a field, I is
a principal ideal; let q(X ) be a generator of I . Note that X 10t − X 2t + X 2t−1 − 1 ∈ I ,
so X - q(X ).
Suppose that h(X ) is an irreducible factor of q(X ) having degree m, and let α ∈ F2m
be a root of h. For each nonzero β ∈ F2m , there is a polynomial g(X ) with αg(α) = β.
Since the given identity applies to xg(x),

X 10t+1 g(X )10t+1 − X 2t+1 g(X )2t+1 + X 2t g(X )2t − Xg(X )

is in I and hence is a multiple of the irreducible polynomial h(X ). Setting X = α,

we have β 10t+1 − β 2t+1 + β 2t − β = 0. Hence every nonzero element of F2m is a root
m
of X 10t+1 − X 2t+1 + X 2t − X . This means that X 2 −1 − 1 divides X 10t+1 − X 2t+1 +
2t 10t+1 2t+1 m
X − X . When we reduce X −X + X − X modulo X 2 −1 − 1, we replace
2t
m
every exponent by its residue mod 2 − 1. Hence, since R has characteristic 2, the four
exponents 10t + 1, 2t + 1, 2t and 1 must pair up when reduced mod 2m − 1. If they
pair up as (10t + 1, 2t + 1) and (2t, 1), then 2m − 1 divides both 10t + 1 − (2t + 1)
and 2t − 1, which implies m = 1 since gcd(8t, 2t − 1) = 1. Similarly, the other two
pairings also lead to m = 1. Hence, h(X ) = X − 1.
Now let k(X ) = X 10t+1 − X 2t+1 + X 2t − X . Since k 0 (1) = 1, we cannot have 1 as
a double root of k(X ). Thus q(X )|(X − 1), and X − 1 ∈ I . In terms of x this yields
x 2 = x = 0 or x = x 2 , as desired.
Also solved by A. J. Bevelacqua, D. Constales (Belgium), P. P. Dályay (Hungary), O. P. Lossers (Netherlands),
J. M. Sanders, R. Tauraso (Italy), and the proposer.

An Application of Fermat’s Little Theorem

11537 [2010, 929]. Proposed by Lang Withers, Jr., MITRE, McClean, VA. Let p be a
prime and a be a positive integer. Let X be a random variable having a Poisson distri-
bution with mean a, and let M be the pth moment of X . Prove that M ≡ 2a mod p.
Solution by Alin Bostan and Bruno Salvy, INRIA, France. For n ≥ 0, the nth moment
Mn (a) of a random variable having a Poisson distribution with mean a is given by
X kn ak
Mn (a) = e−a .
k≥0
k!

Mn (a)t n /n! as follows:

P
We compute its exponential generating function n≥0

X Mn (a) X a k X (tk)n X ak t
t n = e−a = e−a etk = ea(e −1) .
n≥0
n! k≥0
k! n≥0
n! k≥0
k!

882 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 119

This content downloaded from 137.140.1.131 on Wed, 27 Mar 2013 20:37:53 PM

All use subject to JSTOR Terms and Conditions
 By coefficient extraction, it follows that Mn (a) is a monic polynomial of degree n in
n (a) on a , denoted
k
 k
a (it is known as the Touchard polynomial). Its coefficient a M
n

k
, is the Stirling number of the second kind, given by
k k
t (e − 1)k (−1)k− j k t j
   n t  nX    
n t 1 X k n
= = e = (−1)k− j j .
k n! k! n! j=0 k! j k! j=0 j

By Fermat’s Little Theorem, if p is a prime number and 2 ≤ k ≤ p − 1, then mod p,

  X k   k  
p k− j k k− j k − 1
X
k! ≡ (−1) j =k (−1) = k(1 − 1)k−1 = 0.
k j=1
j j=1
j − 1
 p
Therefore, p divides k
when 2 ≤ k ≤ p − 1, yielding
M p ≡ a p + a ≡ 2a (mod p).

Also solved by O. J. L. Alfonso (Colombia), T. Becker (Germany), L. Bogdan (Canada), J. R. Buchanan,

M. Caragiu, M. A. Carlton, N. Caro (Brazil), R. Chapman (U. K.), D. Constales (Belgium), W. J. Cowieson,
N. Grivaux (France), K. Kim, O. Kouba (Syria), J. Lobo (Costa Rica), O. P. Lossers (Netherlands), R. D. Nel-
son (U.K.), M. A. Prasad (India), D. Promislaw (Canada), C. M. Russell, J. Simons (U. K.), N. C. Singer,
A. Stenger, R. Stong, R. Tauraso (Italy), M. Tetiva (Romania), Z. Vörös (Romania), M. Vowe (Switzerland),
D. M. Warme, GCHQ Problem Solving Group (U. K.), Texas State University Problem Solvers Group, and the
proposer.

A Condition for Multiplicative Identities in Commutative Rings

11538 [2010, 929]. Proposed by Marian Tetiva, National College “Gheorghe Roşca
Codreanu”, Bı̂rlad, Romania. Prove that a finite commutative ring in which every
element can be written as a product of two (not necessarily distinct) elements has a
multiplicative identity.
Solution by John H. Smith, Needham, MA. Let R be such a ring. For x ∈ R, let x R =
{x z : z ∈ R}. We call x R maximal if it is not properly contained in y R for any y. By
the hypothesis, every element of R is in some such set. Since R is finite, each y R is
contained in a maximal such set.
When x R is maximal, we show that (i) x ∈ x R, (ii) x R contains an element ex that
acts as a multiplicative identity on x R, (iii) if y R is also maximal, then x R = y R, and
(iv) x R = R. Together, (ii) and (iv) yield the desired multiplicative identity on R.
(i): We are given x = yz, which yields x R ⊆ y R. If x ∈ / x R, then the containment
is proper, contradicting the maximality of x R.
(ii): Since x ∈ x R, we have x = xex = ex x (by commutativity) for some ex ∈ R.
Hence x R ⊆ ex R, and these sets are equal by the maximality of x R. Thus also ex R is
maximal, and by (i) we have ex ∈ ex R = x R. Since ex x = x, it follows for x y ∈ x R
that ex x y = x y, and hence ex is a multipicative identity on x R.
(iii): Suppose that x R and y R are both maximal; let ex and e y be the corresponding
identity elements, and let f = ex + e y − ex e y . We compute
f x = ex x + e y x − (ex )e y x = x + e y x − e y x = x.
Similarly, f y = y, so f R contains both x R and y R. By maximality, f R equals both
x R and y R, and hence they equal each other.
(iv): Every element of R lies in some maximal set y R and hence in x R. Thus R =
x R.

December 2012] PROBLEMS AND SOLUTIONS 883

This content downloaded from 137.140.1.131 on Wed, 27 Mar 2013 20:37:53 PM

All use subject to JSTOR Terms and Conditions
Also solved by G. Apostolopoulos (Greece), H. E. Bell (Canada), J. Bergen, A. J. Bevelacqua, W. D. Burgess,
J. Cade, N. Caro, R. A. Caruthers, R. Chapman (U. K.), J. E. Cooper III, A. Habil (Syria), A. Hays, C. Lanski,
O. P. Lossers (Netherlands), A. Nakhash, D. Opitz, D. Promislow (Canada), J. Simons (U. K.), T. Smith,
R. Stong, R. Tauraso (Italy), X. Wang, J. W. Ward, the Microsoft Research Problems Group, the NSA Problems
Group, the Texas State University Problem Solvers Group, and the proposer.

Quadratic form plus a cube

11539 [2010, 929]. Proposed by William C. Jagy, MSRI, Berkeley, CA. Let E be
the set of all positive integers not divisible by 2 or 3 or by any prime q repre-
sented by the quadratic form 4u 2 + 2uv + 7v 2 . (Thus, the first few members of E
are 1, 5, 11, 17, 23, and 25.) Show that 4x 2 + 2x y + 7y 2 + z 3 is not an element of
{2n 3 , −2n 3 , 32n 3 , −32n 3 } for n ∈ E and x, y, z ∈ Z.
Solution by Robin Chapman, Exeter, UK. Let Q 1 (x, y) = x 2 + 27y 2 and Q 2 (x, y) =
4x 2 + 2x y + 7y 2 . Both Q 1 and Q 2 are primitive positive definite integral quadratic
forms with discriminant −108. By special cases of quadratic and cubic reciprocity, we
have the following (Theorems 2.13 and 4.15 of [1]):
(i) a prime p is represented by one of these forms if and only if p ≡ 1 (mod 3);
(ii) p is represented by the quadratic form Q 1 if and only if p ≡ 1 (mod 3) and 2
is a cubic residue modulo p.
Therefore, if p is represented by Q 2 , then 2 is not a cubic residue modulo p.
Now suppose that 4x 2 + 2x y + 7y 2 + z 3 = kn 3 with k ∈ {±2, ±32}, for x, y, z ∈ Z
and n ∈ E. We have Q 2 (x, y) = kn 3 − z 3 , but we cannot have x = y = 0, since k is
not the cube of a rational. Now Q 2 (x, y) = 41 ((4x + y)2 + 27y 2 ) ≥ 27 4
> 6 if y 6 = 0,
and 4x 2 + 2x y + 7y 2 = 4x 2 ≥ 4 if y = 0.
Consider the possible values of Q 2 (x, y). If Q 2 (x, y) is even, then y is even; let
y = 2y 0 . Now Q 2 (x, y) = 4(x 2 + x y 0 + 7(y 0 )2 ). For x 2 + x y 0 + 7(y 0 )2 to be even,
both x and y 0 must be even. Repeating this argument, we see that the 2-adic val-
uation v2 (Q 2 (x, y)) must be even. (Here v2 (m) is defined as the highest power of
2 that divides m.) Since n ∈ E, n is odd, and therefore v2 (kn 3 ) is 1 or 5. Since
z 3 = kn 3 − Q 2 (x, y), we have v2 (z 3 ) = min(v2 (kn 3 ), v2 (Q 2 (x, y))) ∈ {0, 1, 2, 4, 5}.
However, v2 (z 3 ) is a multiple of 3, so v2 (z 3 ) = 0, and hence Q 2 (x, y) is odd.
Now Q 2 (x, y) ≡ x 2 + 2x y + y 2 = (x + y)2 (mod 3). If 3 | Q 2 (x, y), then x ≡
−y (mod 3), so we can write x = 3t − y, where t ∈ Z. Now Q 2 (x, y) = 36t 2 −
18t + 9y 2 so it is a multiple of 9, so z 3 ≡ kn 3 (mod 9). However, n ∈ E yields
3 - n, so n ≡ ±1 (mod 3). Therefore, n 3 ≡ ±1 (mod 9), leading to z 3 ≡ ±2 or ±5
(mod 9), which is impossible. We conclude that Q 2 (x, y) is coprime to 6.
With m = gcd(x, y) we get Q 2 (x, y) = m 2 Q 2 (x 0 , y 0 ), where x = mx 0 , y = my 0 ,
and gcd(x 0 , y 0 ) = 1. Now Q 2 (x 0 , y 0 ) ≥ 4, and we claim that Q 2 (x 0 , y 0 ) has a prime
factor p represented by Q 2 . Let p be any prime factor of Q 2 (x 0 , y 0 ). We have
4Q 2 (x 0 , y 0 ) ≡ (4x 0 + y 0 )2 + 27(y 0 )2 (mod p), and hence p ≡ 1 (mod 3) by (i). Now
p is represented by Q 1 or Q 2 ; let us suppose by Q 1 so that p = u 2 + 27v 2 .
Define a = ux 0 − vx 0 − 7vy 0 and b = uy 0 + 4vx 0 + vy 0 to get
4a 2 + 2ab + 7b2 = (u 2 + 27v 2 ) 4(x 0 )2 + 2x 0 y 0 + 7(y 0 )2 = p Q 2 (x 0 , y 0 ).

Since u 2 ≡ −27v 2 (mod p), we can write u ≡ ξ v (mod p), where ξ 2 ≡ −27
(mod p) (ξ exists, since −3 is a quadratic residue modulo p). Since (4x 0 + y 0 )2 ≡
27(y 0 )2 (mod p), we also have 4x 0 + y 0 ≡ ±ξ y 0 (mod p). Replacing u by −u and
ξ by −ξ , if necessary, we may assume that u ≡ ξ v and 4x 0 ≡ −(1 + ξ )y 0 (mod p).
Now
4a ≡ −ξ(1 + ξ ) + (1 + ξ ) − 28 vy 0 = −(ξ 2 + 27)vy 0 ≡ 0 (mod p)

884 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 119

This content downloaded from 137.140.1.131 on Wed, 27 Mar 2013 20:37:53 PM

All use subject to JSTOR Terms and Conditions
and b ≡ ξ − (1 + ξ ) + 1 vy 0 ≡ 0 (mod p). Thus p | a and p | b, so

Q 2 (x 0 , y 0 ) Q 2 (a, b)
 
a b
= = Q2 ,
p p2 p p
and hence Q 2 (x 0 , y 0 )/ p is represented by Q 2 . Iterating this argument must eventually
find a prime factor of Q 2 (x 0 , y 0 ) represented by Q 2 .
Let the prime p divide Q 2 (x 0 , y 0 ) and be represented by Q 2 , so 2 is not a cubic
residue modulo p. We have kn 3 ≡ z 3 (mod p). Since also k = ±2 or k = ±25 , we
conclude that k is not a cubic residue modulo p. Hence p | n, contradicting n ∈ E.
This shows that Q 2 (x, y) + z 3 6 ∈ {±2n 3 , ±32n 3 : n ∈ E}.
References
[1] David A. Cox, Primes of the form x 2 + ny 2 , John Wiley & Sons, 1989.
Also solved by the proposer.

A Stirling sum
11545 [2011, 84]. Proposed by Manuel Kauers, Research Institute for Symbolic Com-
putation, Linz, Austria, and Sheng-Lan Ko, National Taiwan University, Taipei, Tai-
wan. Find a closed-form expression for
n  
X 2n
(−1) k
s(n + k, k),
k=0
n+k
where s refers to the (signed) Stirling numbers of the first kind.
(2i − 1). Let c(n, k)
Qn
Solution I by Jim Simons, Cheltenham, U. K. The answer is i=1
denote the unsigned Stirling number of the first kind, the number of permutations of
[n] with k cycles. By definition, s(n, k) = (−1)n−k c(n, k). Subsituting this definition
into the sum and then setting k = n − i transforms the sum to
n  
i 2n
X
(−1) c(2n − i, n − i).
i=0
i

Now 2ni c(2n − i, n − i) is the number of ways to construct a permutation of [2n]

with n cycles by choosing i fixed points and constructing a permutation with n −
i cycles on the remaining 2n − i elements. By inclusion-exclusion, the sum is the
number of permutations of [2n] having n cycles and no fixed points. Each cycle in
such a permutation
Qn must be a 2-cycle. It is well known that the number of pairings of
2n elements is i=1 (2i − 1).
Solution II by Kim McInturff, Santa Barbara, CA. We obtain the answer in the form
(2n)!/2n n!. It is well known that
X tn
s(n, k) u k = (1 + t)u ,
n≥k
n!

Now
∞
X (ut)i X t j+k
e−ut (1 + t)u = (−1)i s( j + k, k) uk
i=0
i! j,k
( j + k)!

t i+ j+k
 
X i + j +k
= (−1)i s( j + k, k) u i+k .
i, j,k
j + k (i + j + k)!

December 2012] PROBLEMS AND SOLUTIONS 885

This content downloaded from 137.140.1.131 on Wed, 27 Mar 2013 20:37:53 PM

All use subject to JSTOR Terms and Conditions
The coefficient of t 2n u n /n! in this sum is obtained by taking the terms in which j = n
and i = n − k, yielding
n  
X 2n
(−1)n−k s(n + k, k)
k=0
n+k

which is (−1)n times the desired sum. Computing the coefficient another way yields
e−ut (1 + t)u = e−ut+u log(1+t) = exp u(−t 2 /2 + t 3 /3 − · · · )

so the coefficient of u n t 2n /(2n)! is (−1)n (2n)!/2n n!.

Also solved by D. Beckwith, C. Burnette, D. Callan, R. Chapman (U. K.), D. Constales, K. David, C. Fuerst
(Austria), J.-P. Grivaux (France), O. Kouba (Syria), O. P. Lossers (Netherlands), Á. Plaza (Spain), J. Quain-
tance, N. C. Singer, R. Stong, R. Tauraso (Italy), M. Wildon, Barclays Capital Problems Solving Group (U.K.),
CMC 328, GCHQ Problem Solving Group (U. K.), and the proposer.

2-adic Valuation of Bernoulli-style Sums

11546 [2011, 84]. Proposed by Kieren MacMillan, Toronto, Canada, and Jonathan
Sondow, New York, NY. Let d, k, and q be positive integers, with k odd. Find the
P dk q
highest power of 2 that divides 2n=1 n .
Solution by Barclays Capital Problems Solving Group, London, (U.K.). If q is even or
equals 1, then the answer is 2d−1 . Otherwise, the answer is 22(d−1) .
In the case q = 1, the sum equals 2d−1 (k(2d k − 1)), which clearly is divisible by
2 d−1
and not by 2d . Now restrict to q > 1.
If d = 1, then the sum contains exactly k odd terms and hence is odd, so the answer
is 20 . We proceed by induction on d; consider d > 1. We pair high and low terms from
the sum, using
d
2 k d−1 k
2X
X
n = (2 k) − (2
q d q d−1 q
n q + (2d k − n)q .


k) +
n=1 n=1

For even q, we take residues modulo 2d . Since q(d − 1) ≥ d, both (2d k)q and
(2 k)q are divisible by 2d . Also, (2d k − n)q ≡ n q mod 2d (since q is even). Thus
d−1

2 k d d−1 k
2X
X
q
n ≡2 n q (mod 2d ).
n=1 n=1

P2d−1 k
By the induction hypothesis, n=1 n q is divisible by 2d−2 but not 2d−1 . Hence the
sum on the left is divisible by 2d−1 but not 2d .
For odd q, we instead work modulo 22d−1 . Since q(d − 1) ≥ 2d − 1 (using q ≥ 3)
both (2d k)q and (2d−1 k)q are divisible by 22d−1 . Since q is odd, expanding the binomial
yields (2d k − n)q = −n q + 2d kqn q−1 − 22d k 2 q2 n q−2 + · · · ; all terms after the second

are divisible by 22d−1 . Thus n q + (2d k − n)q ≡ 2d qkn q−1 mod 22d−1 , so
d
2 k d−1 k
2X
X
q
n ≡ 2 qk d
n q−1 (mod 22d−1 ).
n=1 n=1

P2d−1 k
By the induction hypothesis, n=1 n q−1 is divisible by 2d−2 but not 2d−1 . Hence the
sum on the left is divisible by 22d−2 but not 22d−1 .

886 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 119

This content downloaded from 137.140.1.131 on Wed, 27 Mar 2013 20:37:53 PM

All use subject to JSTOR Terms and Conditions
Also solved by R. Chapman (U. K.), J. Christopher, P. P. Dályay (Hungary), Y. J. Ionin, O. P. Lossers (Nether-
lands), Á. Plaza (Spain), R. Stong, R. Tauraso (Italy), J. Vondra (Australia), Z. Vörös (Hungary), CMC 328,
GCHQ Problem Solving Group (U. K.), NSA Problems Group, and the proposers.

Largest odd divisors

11553 [2011, 178]. Proposed by Mihály Bencze, Brasov, Romania. For a positive in-
teger k, let α(k) be the largest odd divisor of k. Prove that for each positive integer
n,
n
n(n + 1) X n − k + 1 n(n + 3)
≤ α(k) ≤ .
3 k=1
k 3

α( j)
Solution by O. P. Lossers, Eindhoven, The Netherlands. Let T (k) =
Pk
j=1 j
. Since
α(2 j) = α( j) and α(2 j − 1) = 2 j − 1,
X α(2 j − 1) bk/2c
dk/2e
X α(2 j)  k  T (bk/2c)
T (k) = + = + , (1)
j=1
2j − 1 j=1
2j 2 2

We prove by induction on k (with trivial basis k = 1) that

2k 2(k + 1)
< T (k) < . (2)
3 3
For k > 1, using (1) and the induction hypothesis yields
   
2k k bk/2c k bk/2c + 1 2(k + 1)
< + < T (k) < + ≤ .
3 2 3 2 3 3
Since nk=1 T (k) counts each instance of α( j)/j exactly n − j + 1 times, summing
P
(2) over 1 ≤ k ≤ n yields
n
n(n + 1) X n − j + 1 n(n + 3)
< α( j) < .
3 j=1
j 3

Editorial comment. R. A. MacLeod (On the Largest Odd Divisor of n, Amer. Math.
Monthly 75 (1968) 647–648) proved that 2k3 + 3k1 ≤ T (k) ≤ 2(k+1) 3
2
− 3(k+1) , with
m m
equality for k = 2 in the first and for 2 − 1 in the second inequality. Many
Pnk =n−k+1
solvers also showed that n(n+7/4)
3
≤ k=1 k
α(k) ≤ n(n+2)
3
and that equality is
achieved in the second inequality whenever n = 2m − 1. O. Kouba (On Certain
Sums Related to the Largest Odd Divisor,
P arXiv:1103.2295v1 [math.NT], Mar 2011,
arxiv.org) gave a sharp lower bound for nk=1 n−k+1
k
α(k) and described when equality
holds.
Also solved by M. Bataille (France), D. Beckwith, C. Burnette, P. P. Dályay (Hungary), D. Fleischman, O. Ge-
upel (Germany), N. Grivaux (France), A. Habil (Syria), M. E. Kidwell & M. D. Meyerson, O. Kouba (Syria),
J. H. Lindsey II, D. Nacin, Á. Plaza (Spain), C. R. Pranesachar (India), R. E. Prather, J. Simons (U. K.),
N. C. Singer, A. Stenger, R. Stong, R. Tauraso (Italy), D. B. Tyler, J. Vinuesa (Spain), Z. Xintao (China),
C. Y. Yıldırım (Turkey), GCHQ Problem Solving Group (U. K.), and the proposer.

Errata and End Notes for 2012. In the credits for problem 10912 [2003,745],
P. T. Krasopoulos’ name was misspelled.

December 2012] PROBLEMS AND SOLUTIONS 887

This content downloaded from 137.140.1.131 on Wed, 27 Mar 2013 20:37:53 PM

All use subject to JSTOR Terms and Conditions