Problems and Solutions

Source: The American Mathematical Monthly, Vol. 123, No. 1 (January 2016), pp. 97-105
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/10.4169/amer.math.monthly.123.1.97
Accessed: 27-01-2016 02:25 UTC

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 130.49.147.152 on Wed, 27 Jan 2016 02:25:44 UTC
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, Steven J. Miller,
Mohamed Omar, Richard Pfiefer, Dave Renfro, Cecil C. Rousseau, Leonard Smiley,
Kenneth Stolarsky, Richard Stong, Walter Stromquist, Daniel Ullman, Charles Vanden
Eynden, 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, nor posted to the internet
before the due date for solutions. Submitted solutions should arrive before May 31,
2016. Additional information, such as generalizations and references, is welcome.
The problem number and the solver’s name and address should appear on each solu-
tion. An asterisk (*) after the number of a problem or a part of a problem indicates
that no solution is currently available.

PROBLEMS
11880. Proposed by Dorin Andrica, Babeş-Bolyai University, Cluj-Napoca, Romania.
Let ABCD be any plane quadrilateral (not necessarily convex or even simple). Let a
parallelogram be created by constructing through the ends of each diagonal of ABCD
lines parallel to the other diagonal. Show that each diagonal of this parallelogram
passes through the intersection point of a pair of opposite sides of ABCD.
11881. Proposed by Richard Bagby, New Mexico State University, Las Cruces, NM.
Given a triangle T , let T  be the triangle whose vertices are the points where the three
excircles of T are tangent to the sides of T .
(a) Prove that the ratio of the area of T  to the area of T is r/2R, where r and R are
the inradius and circumradius of T .
(b) Show also that the triangle formed by the points of tangency of the incircle likewise
has area r/2R times that of the original triangle.
11882. Proposed by David Callan, University of Wisconsin, Madison, WI. In a list of
distinct positive integers, say that an entry a is left-full if the entries to the left of a
include 1, . . . , a − 1. For example, the left-full entries in 241739 are 1 and 3. Show
that the number of arrangements of n elements from {1, 2, . . . , 2n + 1} that contain
1 but no other left-full entry is equal to (2n − 1)!/n! times the sum of the entries of
the n × n Hilbert matrix M with Mi, j = 1/(i + j − 1). (The seven arrangements for
n = 2 are 13, 14, 15, 21, 31, 41, and 51.)
11883. Proposed by Hideyuki Ohtsuka, Saitama, Japan. For |q| > 1, prove that

∞
1 1 
∞
1
= .
k=0
(q 20 + q)(q 21 + q) · · · (q 2 + q)
k
q − 1 i=0 q 1−2i + 1

http://dx.doi.org/10.4169/amer.math.monthly.123.1.97

January 2016] PROBLEMS AND SOLUTIONS 97

This content downloaded from 130.49.147.152 on Wed, 27 Jan 2016 02:25:44 UTC
All use subject to JSTOR Terms and Conditions
11884. Proposed by Cezar Lupu, University of Pittsburgh, Pittsburgh, PA and Tudorel
Lupu, Decebal High School, Constanţa, Romania. Let f be a real-valued function on
[0, 1] such that f and its first two derivatives are continuous. Prove that if f (1/2) = 0,
then
 1  1 2

( f (x)) d x ≥ 320
2
f (x) d x .
0 0

11885. Proposed by Cornel Ioan Vălean, Teremia Mare, Romania. Prove that

∞ 
∞ 
∞
1 3 5
= ζ (3) − ζ (4).
p=1 n=1 m=1
(m + n)4 + ((m + n)(m + p)) 2 2 4

Here ζ denotes the Riemann zeta function.

11886. Proposed by Finbarr Holland, University College Cork, Cork, Ireland. Sup-
pose n ≥ 3, and let y1 , . . . , yn be a list of
 real numbers such that 2yk+1 ≤ yk + yk+2
for 1 ≤ k ≤ n − 2. Suppose further that nk=1 yk = 0.
Prove that

n 
n
k 2 yk ≥ (n + 1) kyk ,
k=1 k=1

and determine when equality holds.

SOLUTIONS

A Twisted Enumeration of the Positive Integers

11733 [2013, 854]. Proposed by Donald Knuth, Stanford University, Stanford, CA.
Let V = {0, 1, 2, 3, 4}2 . Say that nonnegative integers
a and b are adjacent when their base-5 expansions
· · · a2 a1 a0 and · · · b2 b1 b0 satisfy the condition that if i >
j ≥ 0 and (ai , a j ) = (bi , b j ), then (ai , a j ) and (bi , b j ) are
consecutive in the path through V shown at right (horizon-
tal coordinate listed first). Thus, for example, 0 is adja-
cent to 1. Similarly 48 (expansion 1435 ) is adjacent to 47
(expansion 1425 ) and 73 (expansion 2435 ).
(a) Prove that every positive integer is adjacent to exactly two nonnegative integers.
(b) Prove that with this definition of adjacency, the nonnegative integers form a path
x0 , x1 , x2 , . . .  starting with x0 = 0.
(c) Explain how to compute efficiently from n the number xn that comes n steps after
0, and determine x1,000,000 .
Solution by Richard Stong, Center for Communications Research, San Diego, CA. We
first define a raster path Pd from 1d to 3d in {1, 2, 3}d . For d = 0 the path is the single
vertex given by a 0-tuple. Denote by a P the path given by prepending a to every vertex

of a path P. For d ≥ 1, the path Pd is the concatenation of 1Pd−1 , 2Pd−1 , and 3Pd−1 ,

where P denotes the reverse of P. By induction on d, every step along the path moves
one digit up or down by 1. Call a step an i| j step if it changes i to j or j to i in one
position.

98
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 123

This content downloaded from 130.49.147.152 on Wed, 27 Jan 2016 02:25:44 UTC
All use subject to JSTOR Terms and Conditions
 To find the nth vertex of Pd (starting from 0), write n in base 3, replace the kth digit
by its difference from 2 if the number of 1s to the left is odd, and then increment each
digit by 1. When two vertices differ only in position k, having 1 and 2 there, they are
adjacent via a 1|2 step in Pd if and only if all entries to the right are 3. When in position
k they have 2 and 3, they are adjacent via a 3|2 step in Pd if and only if all entries to
the right are 1. Because Pd has 1d at the beginning and 3d at the end, it follows by
induction on d that 1|2 steps alternate with 3|2 steps.
We next define a double raster path Dd from 0d to 4d in {0, 1, 2, 3, 4}d . Begin
with the path in {0, m, 4}d obtained from Pd by replacing every 1 with 0, every
2 with m, and every 3 with 4. Steps become 0|m steps or 4|m steps, alternating.
Turn Dd into a path through {0, 1, 2, 3, 4}d as follows: When entering an element v
of {0, m, 4}d with j entries equal to m, expand v into a copy of P j in the coordi-
nates equal to m, leaving the other coordinates fixed. If we enter v by a 0|m step,
then we follow P j from 1 j to 3 j . If we enter v by a 4|m step, then we follow
P j from 3 j to 1 j . For example, the state 0m04m follows 0m044 and expands to
03043, 03042, 03041, 02041, 02042, 02043, 01043, 01042, 01041.
We summarize important properties of double raster paths, all of which can be ver-
ified by induction on d. Exactly two of properties (3)–(6) apply at each vertex other
than 0d and 4d (where exactly one applies), governing the steps entering and leaving
that vertex. Left and right are relative to the coordinate that changes in a step.
(1) All steps change a single coordinate to an adjacent value.
(2) The raster path is followed forward in the blocks arising from elements whose
number of copies of m is odd and backwards in those where it is even.
(3) A vertex is involved in a 0|1 step when each entry to the right is 4 and each entry
to the left is 0, 1, or 4.
(4) A vertex is involved in a 1|2 step when each entry to the right is 0, 3, or 4.
(5) A vertex is involved in a 2|3 step when each entry to the right is 0, 1, or 4.
(6) A vertex is involved in a 3|4 step when each entry to the right is 0 and each entry
to the left is 0, 3, or 4.
For example, when d = 4 the vertex 2233 with leftmost entry 2 cannot satisfy (3)
or (6), so its transitions must involve 2; its neighbors are 2133 and 2232.
A position cannot first change away from 0 until each entry to the right is 4. Thus
the first 5d−1 steps of the double raster path Dd are a copy of Dd−1 through vertices
with leading coordinate 0. Hence the double raster paths of all dimensions combine to
give a single path D∞ through {0, 1, 2, 3, 4}∞ .
We now return to the given problem. Each edge in the graph shown corresponds to
changing the entry in one position to a neighboring value. An edge of V that changes
a vertical coordinate from 0 to 1 has horizontal coordinate 0, 1, or 4. An edge of V
that changes a horizontal coordinate from 0 to 1 has vertical coordinate 4. Thus two
integers that differ in a single digit where one has a 0 and the other a 1 are adjacent if
and only if all digits to the left are 0, 1, or 4 and all digits to the right are 4. Comparing
this to (3), we see that they are adjacent if and only if they are consecutive in D∞ .
Analyzing the other three cases similarly and comparing them to (4), (5), and (6), we
see that nonnegative integers are adjacent if and only if they are consecutive in D∞ .
This establishes parts (a) and (b) of the problem.
For part (c), let yk be the string of symbols from {0, m, 4} corresponding to the
vertex xk . If k has d digits in base 5, then yk will have d characters (since D∞ starts
with Dd ). If the leading digit of k is 4, say k = 4 · 5d−1 + k  , then yk will be given
by prepending a 4 to yk  (with leading zeroes added if necessary to make yk  of length
d − 1), since the last segment of Dd is 4Dd−1 . If the leading digit is 1, 2, or 3, then let

January 2016] PROBLEMS AND SOLUTIONS 99

This content downloaded from 130.49.147.152 on Wed, 27 Jan 2016 02:25:44 UTC
All use subject to JSTOR Terms and Conditions
s =
(k − 5d )/3 < 5d−1 . Now yk will be given by prepending m to the complement

(swapping symbols 0 and 4) of ys , since the middle section of Dd is m Dd−1 and every
block has three times as many vertices as the corresponding block of Dd−1 due to the
extra m. Running this calculation backwards, we can find the least j such that y j = yk .
Hence we know how far into the raster path k is. Specifically, take the string yk and
swap the symbols 0 and 4 whenever there is an odd number of symbols m to the left.
Call the resulting sequence z d−1 · · · z 0 . Define k0 = 0 and
⎧
⎪
⎨k j if z j = 0;
k j+1 = 3k j + 5 j if z j = m;
⎪
⎩k + 4 · 5 j if z = 4.
j j

Now kd is the least term in the block. From (2) we know whether the raster path is
followed forwards or backwards, so we can compute the values for these coordinates
as described in the opening paragraph and use them to find xk .
For the specific case k = 1000000, we compute as follows. Letting an overbar
denote complementation, and m k to denote k consecutive copies of m, we obtain
y1000000 = m y203125 = m 2 y41666 = m 3 y8680 = m 4 y1851 = m 5 y408
= m 6 y94 = m 7 y23 = m 7 0y3 = m 7 0m.
We then calculate k1 = 1, k2 = 21, k3 = 88, k4 = 389, k5 = 1792, k6 = 8501, k7 =
41128, k8 = 201509, and k9 = 995152. Thus, the first step in this block occurs at
index 995152, and x1000000 is the 4848th term in this block. We follow the raster path
backwards in this block, because it has eight symbols m. Writing 4848 = 201221203
in base 3 and complementing every digit with an odd number of 1s to its left gives
20100120. Hence the 4848th step of the raster path P8 is 31211231 and the 4848th
step counting backwards from the end is the complement 13233213. Hence x1000000 =
1323321035 = 667778.
Also solved by Y. J. Ionin, O. P. Lossers (Netherlands), and the proposer. Parts (a) and (b) also solved by
TCDmath Problem Group (Ireland).

A Powerful Equation
11734 [2013, 854]. Proposed by Vahagn Aslanyan, Yerevan State University Yerevan,
Armenia. Find all lists (a, k, m, n) of positive integers such that
a m+n + a n − a m − 1 = 15k .

Solution by GCHQ Problem Solving Group, Cheltenham, U. K. There are two such
lists, (2, 1, 2, 2) and (4, 1, 1, 1).
We require 15k = (a m + 1)(a n − 1). We show first that 15 cannot divide both
factors. If 3 | a m + 1, then a ≡ 2 (mod 3) and m is odd. If 3 | a n − 1, then a ≡ 2
(mod 3) requires n even.
If 5 | a m + 1, then m odd requires a ≡ 4 (mod 5).
Since n is even, we have a n − 1 = (a n/2 − 1)(a n/2 + 1). These factors differ by 2,
so neither 3 nor 5 can divide both. Since 15 | a n − 1, n ≥ 4, so neither factor can be 1.
Hence we have two cases.
Case 1: a n/2 − 1 = 3r and a n/2 + 1 = 5s , with r, s ≥ 1. If s = 1, then a n/2 = 4, and
a ≡ 4 (mod 5) requires a = 4 and n = 2. This yields 15 | 4m + 1, which fails since
4m + 1 is 2 or 5 modulo 15. If s > 1, then by Mihăilescu’s theorem (Catalan’s conjec-
ture) a n/2 is not a proper power, so n = 2. This yields a = 1 + 3r , which contradicts
a ≡ 2 (mod 3).

100
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 123

This content downloaded from 130.49.147.152 on Wed, 27 Jan 2016 02:25:44 UTC
All use subject to JSTOR Terms and Conditions
Case 2: a n/2 + 1 = 3r and a n/2 − 1 = 5s , with r, s ≥ 1. If s = 1, then 3r = 7, which
is nonsense. If s > 1, then again we have n = 2. This yields a = 1 + 5s , which con-
tradicts a ≡ 4 (mod 3).
Hence 15 does not divide both factors. There remain four cases.
Case 3a: a n − 1 = 3r and a m + 1 = 3s 5r +s .
Case 3b: a n − 1 = 3s 5r +s and a m + 1 = 3r .
Case 3c: a n − 1 = 5r and a m + 1 = 3r +s 5s .
Case 3d: a n − 1 = 3r +s 5s and a m + 1 = 5r .
These cases allow r = 0 or s = 0; we eliminate this first. In Cases 3b and 3d, a ≥ 1
implies r > 0. In Cases 3a and 3c, r = 0 requires a = 2 and then 2m + 1 = 15s , which
fails modulo 15. Thus in all cases r > 0.
In Cases 3b and 3c, s = 0 yields a n − 1 = 5r and a m + 1 = 3r . If n = 1, then
a = 5r + 1, so 3r = a m + 1 ≥ a + 1 = 5r + 2. If n ≥ 2, then Mihăilescu’s theorem
yields r = 1, but then a n = 6.
In Cases 3a and 3d, s = 0 yields a n − 1 = 3r and a m + 1 = 5r . If m = 1, then
a = 5r − 1, so 3r = a n − 1 ≥ a − 1 = 5r − 2, which requires r = 1. Now a n = 4,
yielding the solution (a, k, m, n) = (4, 1, 1, 1). If m ≥ 2, then Mihăilescu’s theorem
yields r = 1, but then a m = a n = 4. Now (a, k, m, n) is (4, 1, 1, 1) or (2, 1, 2, 2).
Henceforth we may assume r, s > 0.
Case 3a: Since 3 divides a n − 1 and a m + 1, we have a ≡ 2 (mod 3) with m odd
and n even. Since m is odd and 5 | a m + 1, we have a ≡ 4 (mod 5). Since n is even,
5 | a n − 1, which contradicts a n − 1 = 3r .
Case 3c: Since 3 divides a m + 1 but not a n − 1, we have a ≡ 2 (mod 3) with m odd
and n odd. Since m is odd and 5 | a m + 1, we have a ≡ 4 (mod 5). Since n is odd,
5  a n − 1, which contradicts a n − 1 = 5r .
Case 3b: If r = 1, then (a, m) = (2, 1), so 2n − 1 = 3s 51+s . Thus 25 | 2n − 1, which
requires 20 | n and hence 5 | n, yielding the contradiction 31 | 2n − 1.
If r = 2, then (a, m) is (2, 3) or (8, 1). With either, we need the form 2t − 1 =
s r +s
3 5 , which fails as when r = 1. If r ≥ 3, then by Mihăilescu’s theorem m = 1, so
a = 3r − 1. Thus a ≡ 2 (mod 3), but a n ≡ 1 (mod 3), so n is even. Now a n − 1 =
3s 5r +s fails, reasoning as in Cases 1 and 2.
Case 3d: If r = 1, then a = m = 2 or m = 1. This requires 5 | 2n − 1 or 5 | 4n − 1,
each of which makes n even. Now a n − 1 = 3r +s 5s fails, reasoning as in Cases 1 and
2. If r ≥ 2, then by Mihăilescu’s theorem m = 1, so a = 5r − 1. Thus a ≡ 4 (mod 5),
so n is even.
Also solved by D. Henderson, Y. J. Ionin, O. P. Lossers (Netherlands), R. S. and S. S. Bhatt (India),
R. Stong, NSA Problems Group, and the proposer.

Perpendicular Lines Constructed in a Triangle

11737 [2013, 855]. Proposed by Nguyen Thanh Binh, Hanoi, Vietnam. Given an acute
triangle ABC, let O be its circumcenter, let M be the intersection of lines AO and BC,
and let D be the other intersection of AO with the circumcircle of ABC. Let E be
that point on AD such that M is the midpoint of ED. Let F be the point at which the
perpendicular to AD at M meets AC. Prove that EF is perpendicular to AB.
−→
Solution I by Dao Tanh Oai, Vietnam. Extend FM to intersect AB at G and consider
triangle AFG. The feet of the perpendiculars to its sides from D are B, M, and C,
which are collinear. Thus D has a Simson line. By the Wallace–Simson theorem, D
lies on the circumcircle of AFG. Point E has the property of the orthocenter of AFG;
to wit, its reflection D across side FG lies on the circumcircle. Thus EF ⊥ AG.

January 2016] PROBLEMS AND SOLUTIONS 101

This content downloaded from 130.49.147.152 on Wed, 27 Jan 2016 02:25:44 UTC
All use subject to JSTOR Terms and Conditions
Solution II by O.P. Lossers, Department of Mathematics and Computing Science,
Eindhoven University of Technology, Eindhoven, The Netherlands. There are three
cases, depending on the location of F. If F and C coincide, then ∠FDM = ∠BDM =
∠ADB = γ , where γ = ∠ACB. If F lies outside the circle, then ∠FMD = ∠FCD =
π/2, so MCFD is cyclic and ∠FDM = π − ∠FCM = γ . If F lies between A and
C, then ∠FMD + ∠FCD = π, so MFCD is cyclic and FDM = γ . Since EM = MD,
we have ∠FEM = ∠FDM = γ . Since ∠DAB = π/2 − γ , vertical angles at E imply
EF ⊥ AB.
Solution III by Jerry Minkus, San Francisco, CA. For any point X , let X  be its reflec-
tion across AD. By Pascal’s theorem applied to hexagon AC B  DBC, if we take M,
Q, and Q  to be the intersections BC ∩ B  C  , DB ∩ AC, and DB ∩ AC , then M, Q,
and Q  are collinear. Thus F = Q and lies on DB and F  = Q  and lies on DB. From
rhombus EFDF  , we have EF  DB, so EFAB.
Also solved by M. Abate, H. Bailey, M. Bataille (France), E. Bojaxhiu & E. Hysnelaj (Albania & Australia),
R. Chapman (U. K.), P. P. Dályay (Hungary), P. De (India), C. Delorme (France), A. Ercan (Turkey), D. Fleis-
chman, O. Geupel (Germany), L. Giugiuc (Romania), M. Goldenberg & M. Kaplan, J.-P. Grivaux (France),
A. Habil (Syria), R. Hall, D. He, E. A. Herman, I. M. Isaacs, A .G. Karagulyan (Armenia), B. Karaivanov,
J. H. Lindsey II, J. McHugh, V. Mikayelyan (Armenia), P. Nüesch (Switzerland), J. Schaer (Canada), A. Sine-
fakopoulos (Greece), N. Stanciu & T. Zvonaru (Ramonia), R. Stong, R. S. Tiberio, E. I. Verriest, T. Viteam
(South Africa), T. Wiandt, L. Wimmer (Germany), L. Zhou, GCHQ Problem solving Group (U. K.), and the
proposer.

Estimate an Integral by Moments of the Square

11746 [2013, 942]. Proposed by Pál Péter Dályay, Szeged, Hungary. Let f be
a continuous function from [0, ∞) to R such that the following integrals con-
∞ ∞ ∞
verge: S = 0 f 2 (x) d x, T = 0 x 2 f 2 (x) d x, and U = 0 x 4 f 2 (x) d x. Let V =
√
T + T 2 + 3SU . Given that f is not identically 0, show that
 ∞ 4
π 2 S(T + V )2
| f (x)| d x ≤ .
0 9V

Solution by Solution by Finbarr Holland, School of Mathematical Sciences, University

College Cork,
 ∞ Cork, Ireland. The continuity of f implies that S, T , and U are positive.
Let J = 0 | f (x)|d x and suppose t > 0. By the Cauchy–Schwarz inequality,
 ∞  ∞
1
J2 ≤ (1 + t 2 x 2 )2 f 2 (x)d x · d x.
0 0 (1 + t 2 x 2 )2
∞
The first integral is S + 2t 2 T + t 4 U , and the second is 1t 0 (1+u1 2 )2 du = 4tπ . Choose t
to minimize their product. A calculus argument shows that this occurs when −S/t 2 +
2T + 3t 2 U = 0, or
√
−T + T 2 + 3SU V − 2T S(V − 2T )
t =
2
= = .
3U 3U S(3U )
Since 3SU = V 2 − 2V T , we have t 2 = S/V . Hence
  √
π S S2 π S/V (V 2 + 2V T + SU )
J2 ≤ √ S + 2 T + 2U = .
4 S/V V V 4V

√ + 2V T + SU ) =4 3V 2+ 6V T +2V − 2V T = 4V (V + T ), we have
2 2 2
Since 3(V
π
J ≤ 3 S/V (V + T ) and J ≤ π S(V + T ) /(9V ).
2

102
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 123

This content downloaded from 130.49.147.152 on Wed, 27 Jan 2016 02:25:44 UTC
All use subject to JSTOR Terms and Conditions
Editorial comment. Borislav Karaivanov pointed out that Hardy used the multi-
√  ∞ 4
plier a + bx 2 to prove Carlson’s inequality 0 | f (x)|d x ≤ π 2 ST , and using

p + q x 2 + r x 4 will produce even better bounds. Richard Stong also improved the
bound to π 2 ST − π 2 T 3 /U. We note
√ that π 2 S(T + V )2 /(9V ) is an improvement over
π 2 ST only if SU/T 2 < (11 + 5 5)/2. Taking f (x) = 1/x on [1, n] shows this is
not always the case.
Also solved by R. Chapman (U. K.), E. A. Herman, B. Karaivanov, O. Kouba (Syria), R. Stong, and the
proposer.

A Series That Cannot Converge

11748 [2014, 83]. Proposed by Cezar Lupu, University of Pittsburgh, Pittsburgh, PA,
and Tudorel Lupu, Decebal High School, Constanţa, Romania. Is there a sequence
a1 , a2 , . . . of positive real numbers such that ∞ 1
k=1 ak converges and n
k=1 ak < n for
n

all n?
Solution
 I√ by Traian Viteam,  Cape Town, South Africa. No. We use Carleman’s inequal-
ity, nk=1 k u 1 · · · u k < e nk=1 u k . From this we get

n
1  n 1
n
1  n
1
< ··· <e .
k=1
k k=1
a1 ak k=1 k
a
 n
Letting n → ∞ shows that ∞ 1
k=1 ak diverges given the condition k=1 ak < n (which
n
n ∞ −1/n
may be replaced by k=1 ak < bn provided k=1 bn = ∞).
Composite Solution II by Nicolás Caro, Universidade Federale de Pernambuco,
Recife, Brazil, and Borislav Karaivanov, Lexington, SC. Suppose a1 , √a2 , . . . is such
 n a ···a
a sequence. Let Sn = nk=1 a1k with Sn → S < ∞. We are given Tn = 1n n < 1. By
the AM–GM inequality, Sn ≥ 1/Tn . Also,
2n
1 n n 2 Tn Tn Tn
S2n − Sn = ≥ √ = = > .
a
k=n+1 k
n an+1 an+2 · · · a2n (2nT2n ) 2 2
4T2n 4

By the Cauchy criterion, S2n − Sn → 0, so Tn → 0, whence Sn → ∞. This contradicts

the existence of the sequence.
√ 
Editorial comment. Carleman’s inequality, √ k u 1 · · · u k < e u k follows from the
AM–GM n inequality on u 1 , 2u 2 , 3u 3 , . . . and k k! > (k + 1)/e. Finbarr Holland noted
that n k=1 ak = Sn − S1 +···+S
1 k n
, so by Cesaro (but not Cezar) summation, it tends to
n
√
0. The result follows via AM–GM and the Stirling limit  n
n! → n/e. Allen Stenger
pointed out the problem can be viewed as the minimizing ∞ k=1 x k subject to the con-
n k−1 −1
straints k=1 (k−1)k−1 xk = n for n ≥ 1. The optimal solution is xk = (k−1)
kk
kk 
≈ ek .
Also, some misuse of the limit comparison test was noted. We may have 1/ak <
∞ and 1/bk = ∞, but lim ak /bk = 0. For example, let b2k = 2k and b2k+1 =
a2k+1 /k.
Also solved by A. Alt, R. Bagby, P. Bracken, H. Chen, D. Fleischman, J. W. Hagood, F. Holland (Ire-
land), J. C. Kieffer, O. Kouba (Syria), J. H. Lindsey II, O. P. Lossers (Netherlands)R. Martin (Germany),
V. Mikayelyan (Armenia), R. Molinari, M. Omarjee (France), P. Perfetti (Italy), N. C. Singer, A. Stenger,
R. Stong, R. Tauraso (Italy), M. Wildon (U. K.), GCHQ Problem Solving Group (U. K.), Iowa State Uni-
versity Student Problem Solving Group, NSA Problems Group, TCDmath Problem Group (Ireland), and the
proposers.

January 2016] PROBLEMS AND SOLUTIONS 103

This content downloaded from 130.49.147.152 on Wed, 27 Jan 2016 02:25:44 UTC
All use subject to JSTOR Terms and Conditions
 An Unexpected Hyperbola
11758 [2014, 171]. Proposed by Bryan Brzycki, Whittier, CA. Acute triangle ABC has
several ancillary points and properties shown in the figure. Segment AX A is perpen-
dicular to BC, and segments marked with the same symbol have the same length. The
angle at C is less than the angle at B, and lines PN and RM are parallel and perpen-
dicular, respectively, to BC.

XB

R
P O N
Q YB
M

B C
XA YA

(a) Prove that |RO|/|OM| = 2.

(b) Show that PQ is not parallel to BC.
(c) Letting D be the intersection of PQ and BC, show that if AD is perpendicular to
BX B , then P, R, N , and M lie on a common circle.
(d) For fixed B and C, describe the set of A such that AD is perpendicular to BX B .
Solution by Ahmad Habil, Damascus University, Damascus, Syria. Let coordinates be
introduced in the plane of ABC so that B is the origin and C is on the positive x-axis.
Write B = (0, 0), A = (α, β), and C = (γ , 0). Now γ > 0 by choice of coordinates,
α > 0 since ∠B is acute, and β = 0 since this is a triangle.
We have
   
γ − α 2β 2(γ − α) β
X A = (α, 0), Y A = (2α, 0), X B = α + , , YB = α + , .
3 3 3 3
The equation of the line AX A is x = α. The equation of the line AY A is y = 2β −
βx/α. It will be convenient to put μ = 2α + γ and ν = 4α + γ . With this notation,
the equation of the line BX B is y = 2βx/μ. The equation of the line BY B is y =
βx/(α + 2γ ).
Solving pairs of equations, we get
   
2αβ 2α(μ) 4αβ
P = α, , R= , ,
μ 4α + γ ν
   
2α(μ) 2αβ α(α + 2γ ) αβ
O= , ,Q = , .
ν μ α+γ α+γ
The equations of lines PQ and RO are y = 3αβ/μ − βx/μ and y = 2βx/μ, respec-
tively. The intersection of these two lines is the point
 
2αμ αβ(8α + γ )
M= , .
ν μν

104
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 123

This content downloaded from 130.49.147.152 on Wed, 27 Jan 2016 02:25:44 UTC
All use subject to JSTOR Terms and Conditions
 2αβγ αβγ
(a) |OR| = y B − y O = and |OM| = y O − y M = , so |OR| = 2|OM|.
μν μν
2αβ αβ
(b) If PQ  BC, then y P = y Q , so = , so at least one of α, β, γ is zero,
μ α+γ
which is false.
(c) Point D is the intersection of line P Q with line BC, so D = (3α, 0). The equa-
tions of lines AD and BX B are
3β βx 2βx
y= − and y= ,
2 2α μ
respectively. If AD ⊥ BX B , then
  
β 2β
− = −1,
2α μ
2
so α = βμ . To show that P, R, N , M lie on a circle, we must prove |OM| · |OR| =
|OP| · |ON|. First,
2α 2 β 2 γ 2 2α 3 γ 2
|OM| · |OR| = = .
μ2 ν 2 ) μν 2
Lines PO and AY A have equations
2αβ βx
y= and y = 2β − ,
μ α
respectively. Point N is the intersection of these two lines, so
 
2α(α + γ ) 2αβ
N= , .
μ μ
Now
2α 2 γ αγ
|ON| = x N − x O = and |OP| = x O − x P = .
μν ν
Thus |ON| · |OP| = |OM| · |OR|.
(d) Points B and C are fixed, so γ is constant. Using α = β 2 /μ, we see that A
describes the curve
 γ 2 γ2
y 2 = 2x 2 + γ x, or 2 x + − y2 = ,
2 2
which is a hyperbola.
Editorial comment. Omran Kouba (Syria) noted that the assumption that ABC is
acute is not necessary. Only ∠B must be acute. The assumption ∠C < ∠B is also
unnecessary.
Also solved by M. Bataille (France), E. Bojaxhiu & E. Hysnelaj (Albania & Australia), R. Boukharfane
(Canada), R. Chapman (U. K.), P. P. Dályay (Hungary), P. De (India), O. Geupel (Germany), J.-P. Grivaux
(France), B. Karaivanov & T. S. Vassilev (Canada), O. Kouba (Syria), P. McPolin (Ireland), C. R. Pranesachar
(India), R. Stong, T. Wiandt, GCHQ Problem Solving Group (U. K.), and the proposer.

January 2016] PROBLEMS AND SOLUTIONS 105

This content downloaded from 130.49.147.152 on Wed, 27 Jan 2016 02:25:44 UTC
All use subject to JSTOR Terms and Conditions