Problems and Solutions

Reviewed work(s):
Source: The American Mathematical Monthly, Vol. 119, No. 4 (April 2012), pp. 344-351
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/10.4169/amer.math.monthly.119.04.344 .
Accessed: 21/09/2012 22:53

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
 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, Bog-
dan Petrenko, Richard Pfiefer, Cecil C. Rousseau, Leonard Smiley, Kenneth Stolarsky,
Richard Stong, Walter Stromquist, Daniel Ullman, Charles Vanden Eynden, Sam Van-
dervelde, 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 August 31, 2012. 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
11635. Proposed by Marian Tetiva, National College “Gheorghe Roşca Codreanu”,
Bârlad, Romania, and Nicuşor Minculete, “Dimitrie Cantemir” University, Braşov,
Romania.
(a) Let α and β be distinct nonzero real numbers. Let a, b, c, x, y, z be real, with
0 < a < b < c and a ≤ x < y < z ≤ c. Prove that if
x α + y α + z α = a α + bα + cα and x β + y β + z β = a β + bβ + cβ
then x = a, y = b, and z = c.
(b) Let α1 , α2 , α3 be distinct nonzero real numbers. Let a1 , a2 , a3 , a4 , x1 , x2 , x3 , x4 be
real, with 0 < a1 < a2 < a3 < a4 and a1 ≤ x1 < x2 < x3 < x4 ≤ a4 . If
4 4
X α X α
xk j = ak j
k=1 k=1

for 1 ≤ j ≤ 3, must ak then equal xk for 1 ≤ k ≤ 4?

11636. Proposed by Mowaffaq Hajja, Yarmouk University, Irbid, Jordan. Let ABCD
be a convex quadrilateral, and suppose there is a point M on the diagonal BD with the
property that the perimeters of ABM and CBM are equal and the perimeters of ADM
and CDM are equal. Prove that |AB| = |CB| and |AD| = |CD|.
11637. Proposed by Ovidiu Furdui, Technical University of Cluj-Napoca, Cluj, Roma-
nia. Let m ≥ 1 be a nonnegative integer. Let u = u − buc; the quantity u is called the
fractional part of u. Prove that
Z 1  m m
1 1 X
m
x dx = 1 − ζ (k + 1).
0 x m + 1 k=1
(Here ζ is denotes the Riemann zeta function.)
http://dx.doi.org/10.4169/amer.math.monthly.119.04.344

344 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 119

11638. Proposed by George Apostolopoulos, Messolonghi, Greece. Let a, b, c be pos-
itive real numbers. Prove that

1/3
a 3 + b3 + c3 + 3 ≥ 3 (a 2 b + 1)(b2 c + 1)(c2 a + 1) .

11639. Proposed by Omran Kouba,

R π/2Higher Institute for Applied Science and Technol-
ogy, Damascus, Syria. Evaluate 0 (log(2 sin x))2 d x.
11640. Proposed by Karl David, Milwaukee School of Engineering, Milwaukee, WI.
For x > 0 and x 6 = 1, let f (x) = x 1/(x−1) , and let f (1) = e. Show that f 00 (x) > 0 for
x > 0.
11641. Proposed by Nicolae Bourbăcuţ, Sarmizegetusa, Romania. Let f be a convex
function from R into R and suppose that f (x + y) + f (x − y) − 2 f (x) ≤ y 2 for all
real x and y.
(a) Show that f is differentiable.
(b) Show that for all real x and y,
| f 0 (x) − f 0 (y)| ≤ |x − y|.

SOLUTIONS

Adjacency Matrices of Acyclic Digraphs

11487 [2011, 183]. Proposed by Stephen Gagola Jr., Kent State University, Kent, OH.
Let A be a 0,1-matrix of order n with the property that tr(Ak ) = 0 for every positive
integer k. Prove or disprove: A is similar by way of a permutation matrix to a strictly
upper-triangular 0,1-matrix.
Solution by Victor S. Miller, CCR, Princeton, NJ. The statement is true. It is equivalent
to the standard proposition in graph theory that every acyclic directed graph has a
ranking, which is a vertex ordering such that every edge is directed from an earlier
vertex to a later vertex. (In computer science, this ranking is called a ‘topological sort’
of the directed graph.)
A 0,1-matrix A is the adjacency matrix of a directed graph. Position (i, j) of Ak
counts the directed walks from the ith vertex to the jth. If the traces of all powers
equal 0, then the digraph has no closed walk and hence no cycle. Ordering the vertices
by their distance from a source yields a ranking. Adjacency matrices with respect to
reorderings of the vertices are obtained by conjugating by a permutation matrix. Some
such matrix renumbers the vertices according to the ranking. The adjacency matrix
with respect to this ordering is strictly upper triangular.
Also solved by G. Apostolopoulos (Greece), R. Chapman (U. K.), P. P. Dályay (Hungary), C. M. da Fon-
seca (Portugal), A. Gewirtz (France), O. Kouba (Syria), O. P. Lossers (Netherlands), R. Martin (Germany),
M. Omarjee (France), J. Simons (U. K.), R. Stong, J. Stuart, M. Tetiva (Romania), GCHQ Problem Solving
Group (U.K.), Missouri State University Problem Solving Group, NSA Problems Group, and the proposer.

Positive Semidefinite Combinations of Hermitian Matrices

11488 [2010, 278]. Proposed by Dennis I. Merino, Southeastern Louisiana Univer-
sity, Hammond, LA, and Fuzhen Zhang, Nova Southeastern University, Fort Laud-
erdale, FL.

April 2012] PROBLEMS AND SOLUTIONS 345

(a) Show that if k is a positive odd integer, and A and B are Hermitian matrices of the
same size such that Ak + B k = 2I , then 2I − A − B is positive semidefinite.
(b) Find the largest positive integer p such that for all Hermitian matrices A and B of
the same size, 2 p−1 (A p + B p ) − (A + B) p is positive semidefinite.
Solution to Part (a) by Eugene A. Herman, Grinnell College, Grinnell, IA. More gen-
erally, let k be any positive integer such that Ak + B k = 2I .
First consider odd k. Let L be an eigenspace of A with eigenvalue λ. Since k is
odd and λ is real, L is also the eigenspace of Ak with eigenvalue λk . Hence L is the
eigenspace of B k with eigenvalue 2 − λk , and so L is also the eigenspace of B with
eigenvalue (2 − λk )1/k . Having common eigenspaces, A and B are simultaneously
unitarily diagonalizable. By applying this unitary similarity to both Ak + B k = 2I and
2I − A − B, we may assume that A and B are diagonal. With A = diag(α1 , . . . , αn )
and B = diag(β1 , . . . , βn ), we must show that αik + βik = 2 implies 2 − αi − βi ≥ 0.
For this, it suffices to show that x k + y k = 2 implies x + y ≤ 2.
Let z = (x + y)/2. If z > 1, then 2z k > 2 = x k + y k . Thus z k > (x k + y k )/2,
which contradicts the convexity of the kth-power function.
Next we show that if 2I − A2 j − B 2k is positive semidefinite (for j ∈ N), then
2I − A j − B j is positive semidefinite. Since k is expressible as an odd number times
a power of 2, this completes the proof for general k.
Since every eigenvalue of A2 j + B 2 j is real and at most 2, for x ∈ Cn we compute

kAk xk2 + kB k xk2 = hAk x, Ak xi + hB k x, B k xi

= h(A2k + B 2k )x, xi ≤ h2x, xi = 2kxk2 ,

and hence

kAk xk + kB k xk ≤ 2 kAk xk2 + kB k xk2 ≤ 4kxk2 .

Solution to Part (b) by Richard Stong, Center for Communications Research, San
Diego, CA. The largest such value of p is 2. For p = 2, we have 2(A2 + B 2 ) − (A +
B)2 = (A − B)2 ; hence the expression is always positive semidefinite. For larger odd
p, take A = −I and B = 0. Now 2 p−1 (A p + B p ) − (A + B) p = −(2 p−1 − 1)I , and
the expression is not positive semidefinite.
For larger even p, let x be a small positive number, and take A = 10 00 and B =


0 x
. Note that A p = A and B p = x p I . Let a p be the (2, 2)-entry of (A + B) p . Note


x 0
that a0 = 1, a1 = 0, and a p = a p−1 + x 2 a p−2 for p ≥ 2. Hence a p = x 2 + O(x 4 ).
Thus the (2, 2)-entry of 2 p−1 (A p + B p ) − (A + B) p is −x 2 + O(x 4 ), and the matrix
is not positive semidefinite.
Also solved by J. Grivaux (France), O. Kouba (Syria), J. H. Lindsey II, J. Simons (U. K.), and the proposers.

Twisted Semigroups
11490 [2010, 278]. Proposed by Gábor Mészáros, Kemence, Hungary. A semigroup S
agrees with an ordered pair (i, j) of positive integers if ab = b j a i whenever a and b
are distinct elements of S. Find all ordered pairs (i, j) of positive integers such that if
a semigroup S agrees with (i, j), then S has an idempotent element.
Solution by Nicolás Caro, Colombia. The required pairs are all pairs other than (1, 1).
Let (i, j) be a pair of positive integers, and let S be a semigroup agreeing with (i, j)
that has no idempotent elements. For k ≥ 0 and x ∈ S, we have x x 2 x k = (x 2 ) j x i x k

346 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 119

(since x 6= x 2 ). Hence x 2 j+i+k = x 3+k . Since x 3+k is not idempotent, x 2 j+i+k 6 = x 6+2k ,
and so 2 j + i + k 6 = 6 + 2k for k ≥ 0. This implies 3 ≤ 2 j + i ≤ 5. If 2 j + i = 4,
then x 4 = x 3 for all x implies x 3 = x 4 = x 5 = x 6 (contradicting x 3 6 = x 6 ). If 2 j + i =
5, then x 5 = x 3 for all x implies x 4 = x 6 = x 8 (contradicting x 4 6 = x 8 ). It follows that
2 j + i = 3, so (i, j) = (1, 1).
Finally, the positive integers under addition form a semigroup agreeing with (1, 1)
that has no idempotent elements. This concludes the proof.
Also solved by M. Angelelli (Italy), K. Benningfield, P. Budney, R. Chapman (U. K.), C. Curtis, P. P. Dályay
(Hungary), J. Guerreiro (Portugal), Y. J. Ionin, S. C. Locke, J. Lockhart, B. Mulansky (Germany), V. Pambuc-
cian, M. A. Prasad (India), K. Schilling, J. Simons (U. K.), R. Stong, B. Tomper, A. Wyn-Jones (U. K.), GCHQ
Problem Solving Group (U. K.), Missouri State University University Problem Solving Group, NSA Problems
Group, and the proposer.

A Matrix-Sum Inequality
11496 [2010, 370]. Proposed by Benjamin Bogoşel (student), West University of
Timisoara, Timisoara, Romania, and Cezar Lupu (student), University of Bucharest,
Bucharest, Romania. For a matrix X with real entries, let s(X ) be the sum of its
entries. Prove that if A and B are n × n real matrices, then
n s(A A T )s(B B T ) − s(AB T )s(B A T ) ≥

s(A A T )(s(B))2 + s(B B T )(s(A))2 − s(A)s(B) s(AB T ) + s(B A T ) .

Solution by Richard Stong, Center for Communications Research, San Diego, CA. The
required inequality has been corrected above; there was an extra “+” within the first
term, and A T B was printed instead of B A T .
Let 1 be the column vector with each entry 1, so s(X ) = 1T 1. For n × n matrices
A, B, and C, let
 T 
1 A
M = 1T B  ;
1T C
note that M is a 3 × n matrix. Also,
s(A A T ) s(AB T ) s(AC T )
 

M M T = s(B A T ) s(B B T ) s(BC T ) .

s(C A T ) s(C B T ) s(CC T )
Since det(M M T ) ≥ 0, expanding the determinant gives
s(CC T ) s(A A T )s(B B T ) − s(AB T )s(B A T ) ≥

s(A A T )s(BC T )s(C B T ) + s(B B T )s(AC T )s(C A T )

− s(AC T )s(C B T )s(B A T ) − s(BC T )s(C A T )s(AB T ).
Specializing to C = I yields the desired inequality.
Also solved by O. Kouba (Syria), O. P. Lossers (Netherlands), J. Simons (U. K.), GCHQ Problem Solving
Group (U. K.), and the proposer.

Areas in a Subdivided Quadrilateral

11498 [2010, 371]. Proposed by Y. N. Aliyev, Qafqaz University, Khyrdalan, Azerbai-
jan. Let ABCD be a convex quadrilateral. A line through the intersection O of the

April 2012] PROBLEMS AND SOLUTIONS 347

diagonals AC and BD intersects the interior of edge BC at L and the interior of AD at
N . Another line through O likewise meets AB at K and CD at M. This dissects ABCD
into eight triangles AKO, KBO, BLO, and so on. Prove that the arithmetic mean of the
reciprocals of the areas of these triangles is greater than or equal to the sum of the
arithmetic and quadratic means of the reciprocals of the areas of triangles ABO, BCO,
CDO, and DAO. (The quadratic mean is also known as the root mean square; it is the
square root of the mean of the squares of the given numbers.)
Solution by Richard Stong, Center for Communications Research, San Diego, CA . The
problem is invariant under affine transformations—which preserve ratios of areas—so
we may assume O = (0, 0), A = (0, a), B = (b, 0), C = (−c, 0), and D = (0, −d).
Let MK have slope m, so
   
ab mab cd mcd
K = , and M =− , .
b + ma b + ma d + mc d + mc
The areas of AOK, KOB, COM, and MOD are ma 2 b/(2(b + ma)), a 2 b/(2(b + ma)),
mc2 d/(2(d + mc)), and c2 d/(2(d + mc)), respectively. The sum S of their recip-
rocals is given by S = 4/ab + 4/cd + 2m(b−2 +pd −2 ) + 2m −1 (a −2 + c−2 ). Thus
by the AM-GM inequality, S ≥ 4/ab + 4/cd + 4 (b−2 + d −2 )(a −2 + c−2 ). Simi-
larly
p the sum of the reciprocals of the other four areas is at least 4/bc + 4/da +
4 (b−2 + d −2 )(a −2 + c−2 ). Adding these and dividing by 8 gives a lower bound of
v !

1 2 2 2 2
 u u 1  2 2  2 2  2 2  2 2
+ + + +t + + + .
4 ab bc cd da 4 ab bc cd da

Now ab/2, bc/2, cd/2, da/2 are the areas of AOB, BOC, COD, DOA, so this is the
desired result.
Also solved by P. P. Dályay (Hungary), O. Kouba (Syria), J. H. Lindsey II, GCHQ Problem Solving Group
(U. K.), and the proposer.

Everything in Its Own Place, Almost

11500 [2010, 371]. Proposed by Bhavana Deshpande, Poona College, Camp Pune,
Maharashtra, India, and M. N. Deshpande, Institute of Science, Nagpur, India. We
have n balls, labeled 1 through n, and n urns, also labeled 1 through n. Ball 1 is put
into a randomly chosen urn. Thereafter, as j increments from 2 to n, ball j is put into
urn j if that urn is empty, otherwise, it is put into a randomly chosen empty urn. Let
the random variable X be the number of balls that end up in the urn bearing their own
number. Show that the expected value of X is n − Hn−1 .
Solution I by Justin S. Dyer, Stanford University, Stanford, CA. Write X as X n , and let
µn = E(X n ). If the first ball is in the first urn, which happens with probability 1/n,
then every ball is in its own urn, contributing 1 to µn . If the first ball is in the kth urn,
where k ≥ 2, then the k − 2 balls 2, . . . , k − 1 are in their own urns, and n − k + 1
urns are left when the next random placement occurs. They are labeled 1, k + 1, k +
2, . . . , n, and this is the same situation as before, except that if the next ball lands in
urn 1, then it does not contribute. Hence the result of the remaining random placements
is X n−k+1 if the next ball is not in urn 1, and X n−k+1 − 1 if it is, so its expectation is
µn−k+1 − n−k+1 1
. Summing over all locations for the first ball, we have
n  
1X 1
µn = 1 + (k − 2) + µn−k+1 − .
n k=2 n−k+1

348 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 119

We now compute
1
nµn − (n − 1)µn−1 = 1 + (n − 2) + µn−1 − .
n−1
Rearranging yields µn = µn−1 + 1 − 1
n−1
, and hence µn = n − Hn−1 .
Solution II by Kenneth Schilling, University of Michigan-Flint, Flint, MI. Let S =
{2, 3, . . . , n}. We first prove that for T ⊆ S, there is exactly one placement of the
n balls such that the subset of S corresponding to balls in their Qown urns is precisely
S − T . Furthermore, the probability pT of this placement is n1 k∈T n−k+1 1
.
If T = ∅, then ball 1 is in urn 1. The rest fall into place, and p∅ = 1/n. Otherwise,
let T = {k1 , . . . , kt }, indexed in increasing order. Having ball i in urn i for 2 ≤ i < k1
but ball k1 not in urn k1 requires having ball 1 in urn k1 . Next, having ball i in urn i
for k1 < i < k2 and ball k2 not in urn k2 requires having ball k1 in urn k2 . Continuing,
ball ki must be in urn ki+1 for 1 ≤ i ≤ t − 1, and finally ball kt must be in urn 1. For
the probability claim, there are n urns in which ball 1 may be placed, for k ∈ / T ball k
is placed by rule, and the placement of ball ki is chosen from n − ki + 1 unfilled urns.
Now fix j ∈ S. We pair the subsets of S so that the sets T and T 0 in a pair differ
only in whether they contain j. By the claim above, if j ∈ T and j ∈ / T 0 , then (n −
j + 1) pT = pT . Summing over all pairs yields P(ball j is not in urn j) = n−1j+2 and
0
n− j+1
and P(ball j is in urn j) = n− j+2
= 1 − n−1j+2 .
Summing over j ∈ S and counting also 1/n for ball 1, we have E(X ) = n1 +
j=2 (1 − n− j+2 ) = n − Hn−1 .
Pn 1

Editorial comment. Justin S. Dyer also showed that X n /n tends to 1 almost surely as
n → ∞.
Also solved by D. Beckwith, D. F. Behan, D. Brown & J. Zerger, N. Caro (Columbia), R. Chapman (U. K.),
K. David & P. Fricano, P. J. Fitzsimmons, J. Freeman, D. Glass, N. Grivaux (France), S. J. Herschkorn,
B.-T. Iordache (Romania), J. H. Lindsey II, O. P. Lossers (Netherlands), J. & W. Matysiak (Poland), K. McIn-
turff, M. D. Meyerson, Á. Plaza & C. González-Alcón (Spain), R. Prasad (India), R. Pratt, B. Schmuland
(Canada), J. Simons (U. K.), N. C. Singer, T. Starbird, J. H. Steelman, R. Stong, S. Xiao, CMC 328, GCHQ
Problem Solving Group (U. K.), Szeged Problem Group “Fejéntaláltuka” (Hungary), and the proposer.

Runs of Heads and Covariance

11503 [2010, 458]. Proposed by K. S. Bhanu, Institute of Science, Nagpur, India, and
M. N. Deshpande, Nagpur, India. We toss an unbiased coin to obtain a sequence of
heads and tails, continuing until r heads have occurred. In this sequence, there will be
some number R of runs (runs of heads or runs of tails) and some number X of isolated
heads. (Thus, with r = 4, the sequence HHTHTTH yields R = 5 and X = 2.) Find the
covariance of R and X in terms of r .
Solution by Jim Simons, Cheltenham, U. K. If r = 1, then X = 1 with probability 1,
and the covariance is 0. For r > 1, we show that the answer is r/2.
For 1 ≤ i ≤ r , let ti be the random variable having value 1 if the ith head is imme-
diately preceded by a tail and value 0 otherwise. Clearly ti = 1 with probability 1/2,
and t1 , . . . , tr are independent.
The first head is isolated if and only if t2 = 1, and the last head is isolated if and
only if tr = 1. For 2 ≤ i ≤ r − 1, the ith head is isolated if and only if ti = ti+1 = 1.
Thus X = t2 + tr + ri=2
P −1
ti ti+1 , and
E(X ) = 1/2 + 1/2 + (r − 2)(1/4) = (r + 2)/4.

April 2012] PROBLEMS AND SOLUTIONS 349

 If all ti equal zero, then R = 1. Changing t1 from 0Pto 1 adds 1 to R. For 2 ≤ i ≤ r ,
switching ti to 1 adds 2 to R. Hence R = 1 + t1 + 2 rj=2 t j , and
E(R) = 1 + 1/2 + 2(r − 1)(1/2) = (2r + 1)/2.
Since ti2 = ti , expanding the product XR yields
r −1
! r

X X
XR = t2 + tr + ti ti+1 1 + t1 + 2 tj
i=2 j=2
   
r
X r −1
X
= X (1 + t1 ) + 2t2 1 + t j  + 2tr 1 + tj
j=3 j=2
  
r −1
X Xi−1 r
X
+2 ti ti+1  tj + 2 + t j  .
i=2 j=2 j=i+2

Since X is independent of t1 , we obtain

     
3 r +2 2 r −2 2 r −2
E(XR) = + 1+ + 1+
2 4 2 2 2 2
  
r −2 r −3
+2 2+ ,
4 2
which simplifies to (2r 2 + 9r + 2)/8. We can now compute the required covariance:
2r 2 + 9r + 2
  
r +2 2r + 1 r
E(XR) − E(X )E(R) = − = .
8 4 2 2

Also solved by D. Beckwith, N. Caro (Colombia), R. Chapman (U. K.), M. P. Cohen, P. J. Fitzsimmons,
J. Gaisser, S. J. Herschkorn, J. H. Lindsey II, K. McInturff, M. Nyenhuis (Canada), M. A. Prasad (India),
R. Pratt, K. Schilling, B. Schmuland (Canada), T. Starbird, R. Stong, and the proposers.

Computing Pi from Fibonacci Numbers

11505 [2010, 458]. Proposed by Bruce Burdick, Roger Williams University, Bristol, RI.
Define {an } to be the periodic sequence given by a1 = a3 = 1, a2 = 2, a4 = a6 = −1,
a5 = −2, and an = an−6 for n ≥ 7. Let {Fn } be the Fibonacci sequence with F1 =
F2 = 1. Show that
∞ ∞
X ak Fk F2k−1 X (−1)kn π
= .
k=1
2k − 1 n=0
Fkn+2k−1 Fkn+3k−1 4

Composite solution by Roberto Tauraso, Università di Roma “Tor Vergata”, Roma,

Italy; Rituraj Nandan, St. Peters, MO; and the proposer . By d’Ocagne’s identity,
(see http://mathworld.wolfram.com/dOcagnesIdentity.html), FN FM+1 − FN +1 FM =
(−1) M FN −M . With M = kn + 2k − 1 and N = k(n + 1) + 2k − 1, we have
(−1)kn −Fk(n+1)+2k−1 Fkn+2k + Fk(n+1)+2k Fkn+2k−1
=
Fkn+2k−1 Fkn+3k−1 Fk Fkn+2k−1 Fk(n+1)+2k−1
 
1 Fk(n+1)+2k Fkn+2k
= − .
Fk Fk(n+1)+2k−1 Fkn+2k−1

350 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 119

Hence we have the telescoping sum
∞ ∞ 
(−1)kn

X 1 X Fk(n+1)+2k Fkn+2k
= −
n=0
Fkn+2k−1 Fkn+3k−1 Fk n=0 Fk(n+1)+2k−1 Fkn+2k−1
 
1 F2k
= φ− ,
Fk F2k−1
√
where φ = 1+ 5
2
. Thus
∞ ∞ ∞
X ak Fk F2k−1 X (−1)kn X ak (φ F2k−1 − F2k )
=
k=1
2k − 1 n=0
Fkn+2k−1 Fkn+3k−1 k=1
2k − 1
∞  2k−1
X ak 1
= ,
k=1
2k − 1 φ

since Fn+1 − φ Fn = −φ . The sequence 1, 2, 1, −1, −2, −1, . . . is the sum of two
−n

sequences whose cycles are 1, −1, 1, −1, 1, −1 and 0, 3, 0, 0, −3, 0. Continuing the
computation,
∞ ∞ ∞
(−1)k+1 1 2k−1 X 3(−1)k+1 1 6k−3
 2k−1 X    
X ak 1
= +
k=1
2k − 1 φ k=1
2k − 1 φ k=1
6k − 3 φ
∞ ∞
(−1)k+1 1 2k−1 X (−1)k+1 1 2k−1
X    
= +
k=1
2k − 1 φ k=1
2k − 1 φ 3
   
1 1
= arctan + arctan .
φ φ3
Using the arctangent addition formula and φ 2 = φ + 1, this becomes
1
+ φ13
!
φ +φ
     3 
1 1 φ
arctan + arctan = arctan = arctan
φ φ3 1 − φ1 · φ13 φ4 − 1

φ3 + φ φ +φ π
   3 
= arctan = arctan = arctan(1) = .
φ3 + φ2 − 1 φ3 + φ 4

Editorial comment. Omran Kouba proved the general identity

∞
2x cos θ 2(−1)k−1 cos((2k − 1)θ ) 2k−1
  X
arctan = x
1 − x2 k=1
2k − 1

for θ ∈ R and |x| < 1, from which the second part of the proof above follows with
θ = 2π/3 and x = 1/φ.
Also solved by R. Chapman (U. K.), O. Kouba (Syria), K. D. Lathrop, M. A. Prasad (India), R. Stong,
S. Y. Xiao (Canada), and GCHQ Problem Solving Group (U. K.).

April 2012] PROBLEMS AND SOLUTIONS 351