The American Mathematical Monthly

ISSN: 0002-9890 (Print) 1930-0972 (Online) Journal homepage: http://maa.tandfonline.com/loi/uamm20

Problems and Solutions

Gerald A. Edgar, Daniel Ullman & Douglas B. West

To cite this article: Gerald A. Edgar, Daniel Ullman & Douglas B. West (2018)
Problems and Solutions, The American Mathematical Monthly, 125:7, 660-668, DOI:
10.1080/00029890.2018.1483682

To link to this article: https://doi.org/10.1080/00029890.2018.1483682

Published online: 03 Aug 2018.

Submit your article to this journal

Article views: 341

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://maa.tandfonline.com/action/journalInformation?journalCode=uamm20
 PROBLEMS AND SOLUTIONS
Edited by Gerald A. Edgar, Daniel H. Ullman, Douglas B. West
with the collaboration of Paul Bracken, Ezra A. Brown, Zachary Franco, Christian Friesen,
László Lipták, Rick Luttmann, Frank B. Miles, Lenhard Ng, Kenneth Stolarsky, Richard
Stong, Daniel Velleman, Stan Wagon, Elizabeth Wilmer, and Fuzhen Zhang.

Proposed problems should be submitted online at

americanmathematicalmonthly.submittable.com/submit.
Proposed solutions to the problems below should be submitted by December 31,
2018 via the same link. More detailed instructions are available online. Proposed
problems must not be under consideration concurrently at any other journal nor be
posted to the internet before the deadline date for solutions. An asterisk (*) after the
number of a problem or a part of a problem indicates that no solution is currently
available.

PROBLEMS
12055. Proposed by Donald E. Knuth, Stanford University, Stanford, CA. Let a1 , a2 , . . .
be a sequence of nonnegative integers with a1 ≥ a2 ≥ · · · and with finite sum. For a
positive integer j, let b j be the number of indices i such that ai ≥ j. (The sequence
b1 , b2 , . . . is the conjugate of a1 , a2 , . . . .) Prove that the multisets {a1 + 1, a2 + 2, . . . } and
{b1 + 1, b2 + 2, . . . } are equal. For example, if ai  = 5, 3, 2, 2, 0, 0, 0, . . . , then b j  =
4, 4, 2, 1, 1, 0, 0, . . . , and the corresponding multisets are {6, 5, 5, 6, 5, 6, 7, 8, . . . } and
{5, 6, 5, 5, 6, 6, 7, 8, . . . }.
12056. Proposed by Leonard Giugiuc, Drobeta Turnu Severin, Romania, Kadir Altintas,
Emirdağ, Turkey, and Florin Stanescu, Gaesti, Romania. Let ABCD be a rectangle inscribed
in a circle S of radius R, and let P be a point inside S. The lines AP, BP, CP, and DP inter-
sect S a second time at K, L, M, and N, respectively. Prove AK 2 + BL2 + CM 2 + DN 2 ≥
16R4 /(R2 + OP2 ).
12057. Proposed by Peter Kórus, University of Szeged, Szeged, Hungary.
(a) Calculate the limit of the sequence defined by a1 = 1, a2 = 2, and
a2k−1 + a2k √
a2k+1 = and a2k+2 = a2k a2k+1
2
for positive integers k.
(b) Calculate the limit of the sequence defined by b1 = 1, b2 = 2, and
b2k−1 + b2k 2 b2k b2k+1
b2k+1 = and b2k+2 =
2 b2k + b2k+1
for positive integers k.
12058. Proposed by Max A. Alekseyev, George Washington University, Washington, DC.
Let b be an integer greater than 1. For a positive integer n, let ub (n) be the number of
nonzero digits in the base b representation of n. Prove that for any positive integers n and
k, there exists a positive integer m such that ub (mn) = ub (n) + k.

doi.org/10.1080/00029890.2018.1483682

660 
C THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 125
12059. Proposed by George Stoica, Saint John, NB, Canada. Let n be an integer greater
than 1, and let R be the ring of polynomials in the variables x1 , . . . , xn with real coefficients.
Let S be the ideal in R generated by the elementary symmetric polynomials e1 , . . . , en ,
where

ek (x1 , . . . , xn ) = xi1 · · · xik
1≤i1 <···<ik ≤n

for 1 ≤ k ≤ n. The degree of a monomial x1m1 · · · xnmn is m1 + · · · + mn . Prove that

n(n − 1)/2 is the largest degree among all monomials that do not belong to S.

12060. Proposed by Ovidiu Furdui and Alina Sîntămărian, Technical

 University of Cluj-
Napoca, Cluj-Napoca, Romania. Let ζ (3) be Apéry’s constant ∞ n=1 1/n3
, and let Hn be
the nth harmonic number 1 + 1/2 + · · · + 1/n. Prove
∞
 Hn Hn+1 5 π2
= − − ζ (3).
n=2
n3 − n 2 24

12061. Proposed by Dao Thanh Oai, Thai Binh, Viet Nam, and Le Viet An, Hue, Viet Nam.
Two triangles ABC and A B C in the plane are perspective from a point if the lines AA ,
BB , and CC are concurrent (the common point is the perspector) and are perspective from
a line if the points of intersection of AB and A B , of AC and A C , and of BC and B C are
collinear (the common line is the perspectrix). Desargues’s theorem states that two triangles
are perspective from a point if and only if they are perspective from a line. Consider three
triangles, each pair of which are perspective from a point, hence per Desargues’s theorem
perspective from a line. Show that the three perspectrices are identical if and only if the
three perspectors are collinear.

SOLUTIONS
A Triangle out of Pieces
11934 [2016, 832]. Proposed by Leonard Giugiuc, Drobotu Turnu Severin, Romania. Let
ABC be an isosceles triangle, with |AB| = |AC|. Let D and E be two points on side BC
such that D ∈ BE, E ∈ DC, and m(∠DAE ) = 12 m(∠A). Describe how to construct a triangle
XY Z such that |XY | = |BD|, |Y Z| = |DE|, and |ZX| = |EC|. Also, compute m(∠BAC) +
m(∠Y XZ) (in radians).
Solution by Pál Péter Dályay, Szeged, Hungary. Write α, β, γ for the radian measures of
the angles at A, B, C, respectively. Construct three circles C1 , C2 , C3 with center A and radii
r1 , r2 , r3 , respectively, with r1 = |AB|, r2 = |AD|, r3 = |AE|. Let X be the intersection of
the ray from A to the midpoint of BC with C1 , let Y be the intersection of ray AE with C2 ,
and let Z be the intersection of the ray AD with C3 . We claim that XY Z meets the required
conditions.
Let ABD be rotated around A by α/2 to bring B to X and D to Y . Since ABD is
congruent to AXY , we have |XY | = |BD| and m(∠AXY ) = m(∠ABD) = β.
Similarly, let ACE be rotated around A by α/2 to bring C to X and E to Z. As before
we conclude |ZX| = |EC| and m(∠AXZ) = m(∠ACE ) = γ .
Triangles ADE and AY Z are congruent, since they share an angle A, |AY | = |AD|, and
|AZ| = |AE|. Thus |Y Z| = |DE|, and triangle XY Z satisfies the required conditions.

August–September 2018] PROBLEMS AND SOLUTIONS 661

 Since m(∠Y XZ) = m(∠AXY ) + m(∠AXZ) = β + γ , we have m(∠BAC) + m(∠Y XZ)
= α + (β + γ ) = π .
Editorial comment. The problem as originally published had ∠XY Z for the last angle where
∠Y XZ was intended.

Also solved by R. B. Campos (Spain), R. Chapman (U. K.), I. Dimitric, J. Han (Korea), E. Ionascu,
B. Karaivanov (U. S. A) & T. S. Vassilev (Canada), O. Kouba (Syria), O. P. Lossers (Netherlands), M. Meyer-
son, R. Stong, Armstrong State University Problem Solvers, GCHQ Problem Solving Group (U. K.), GWstat
Problem Solving Group, and the proposer.

Hidden Mersenne
11936 [2016, 941]. Proposed by William Weakley, Indiana University–Purdue University
at Fort Wayne, Fort Wayne, IN. Let S be the set of integers n such that there exist integers i,
j, k, m, p with i, j ≥ 0, m, k ≥ 2, and p prime, such that n = mk = pi + p j .
(a) Characterize S.
(b) For which elements of S are there two choices of (p, i, j)?
Solution by Anthony J. Bevelacqua, University of North Dakota, Grand Forks, ND.
(a) The set S is the union of three sets: (1) {2d : d ≥ 2}, (2) {(2t 3)2 : t ≥ 0}, and
(3) {(2pt )k : t ≥ 0 and p = 2k − 1 (a Mersenne prime)}.
First, we prove that if 1 + pd = v k , where p is a prime, d ≥ 1, and v, k ≥ 2, then either
p = 2 and d = 3 (that is, 1 + 23 = 32 ), or p = 2k − 1 (a Mersenne prime) and d = 1 (so
1 + (2k − 1) = 2k ).
We prove this claim in two cases. Suppose first that p = 2 and 1 + 2d = v k . In this case,
2 = v k − 1 = (v − 1)(v k−1 + · · · + v + 1). Since all factors of 2d are even, it follows that
d

v is odd and v k−1 + · · · + v + 1 is even, so k must be even, with k = 2t for some t ∈ N.

This yields 2d = v k − 1 = (v t − 1)(v t + 1), so v t − 1 and v t + 1 are powers of 2 differing
by 2. Thus they must be 2 and 4, so v t − 1 = 2, and this implies v = 3, t = 1, and d = 3.
In the remaining case, p is an odd prime. Factor k as pt m, where t ≥ 0 and p  m, and let
t
w = v p . We have pd = v k − 1 = w m − 1 = (w − 1)(wm−1 + · · · + w + 1). If w − 1 > 1,
then p divides both w − 1 and wm−1 + · · · + w + 1, but then w ≡ 1 mod p and wm−1 +
· · · + w + 1 ≡ m mod p. Hence p divides m, a contradiction. Therefore w = 2, and so
v = 2 and 1 + pd = 2k . Since we are given k ≥ 2, it follows that 1 + pd ≡ 0 mod 4, so
d is odd. If d > 1, then d = qs, where q is an odd prime and s is odd. We must have s = 1,
since otherwise
 
2k = pd + 1 = (pq + 1) (pq )s−1 + (pq )s−2 − · · · − pq + 1 ,
and the second factor is an odd number larger than 1. Thus d is an odd prime, and p has
order 2d modulo 2k , because p2 ≡ 1 mod 2k (since 1 < p2 < pd < 2k ) and pd ≡ −1 mod
2k . Thus 2d divides φ(2k ) = 2k−1 , a contradiction. We conclude d = 1, and p = 2k − 1
must be a Mersenne prime, finishing the proof of the claim.
Now consider the general case n = mk = pi + p j . If i = j, then mk = 2pi , so p = 2
and n = 2i+1 . Thus n can be 2d with d ≥ 2 (d = 1 is excluded by m, k ≥ 2). If i < j,
then mk = pi (1 + p j−i ). Since pi and 1 + p j−i are relatively prime, we have i = kt for
some t ≥ 0, and 1 + p j−i = v k for some v ≥ 2. By our claim we have either p = 2 with
j − i = 3 (so v = 3 and k = 2), or p = 2k − 1 is a Mersenne prime with j − i = 1 (so
v = 2). Thus n = (2t 3)2 = 22t + 22t+3 for some t ≥ 0, or n = (2pt )k = pkt + pkt+1 for
some t ≥ 0, where p = 2k − 1 is a Mersenne prime. Hence the set S is as claimed above.
(b) We further assume i ≤ j to exclude two such trivial representations obtained by switch-
ing i and j, so each member of (1), (2), or (3) has only one representation in that family.

662 
C THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 125
Clearly, values of n in (1) and (2) cannot be the same. If n is in both (1) and (3), then
t = 0 and d = k (so n = 2k , where 2k − 1 is a Mersenne prime), while if n has a represen-
tation in (2) and (3), then p = 3 (which is a Mersenne prime), t = 1 (in both representa-
tions), and k = 2 (so n = 36). Hence the only numbers in S with two different representa-
tions are 36 (represented as 22 + 25 and 32 + 33 ) and 2k (represented as 2k−1 + 2k−1 and
(2k − 1)0 + (2k − 1)1 ) whenever 2k − 1 is a Mersenne prime.

Editorial comment. To simplify the proof, several solvers referred to Catalan’s conjecture
(proved by Mihăilescu in 2004) that the only consecutive integers that are powers of integers
with exponents at least 2 are 23 and 32 .

Also solved by B. Karaivanov (U. S. A) & T. S. Vassilev (Canada), GCHQ Problem Solving Group (U. K.), and
NSA Problems Group. Part (a) also solved by Y. J. Ionin, M. Josephy (Costa Rica), O. P. Lossers (Netherlands),
R. Stong, and the proposer.

A Double Integral for the Digamma Function

11937 [2016, 941]. Proposed by Juan Carlos Sampedro, University of the Basque Country,
Leioa, Spain. Let s be a complex number that is not a zero of the gamma function (s).
Prove
 1 1
(xy)s−1 − y  (s)
dx dy = .
0 0 (1 − xy) log(xy) (s)

Composite solution by O. P. Lossers, Eindhoven University of Technology, Eindhoven,

Netherlands, and Omran Kouba, Higher Institute for Applied Sciences and Technology,
Damascus, Syria. No finite complex number is a zero of (s), but we must assume Re s > 0
for the integral to converge. Write the integral as
 1 1  1 1
1 − (xy)s−1 dx dy 1 − y dx dy
I(s) = − + .
0 0 1 − xy log(xy) 0 0 1 − xy log(xy)
Notice that (1 − (xy)s−1 )/(1 − xy) has finite limit as xy → 1, the functions xs−1 and ys−1 are
11
integrable at 0, and 0 0 dx dy/| log(xy)| < +∞. Therefore, the first integral converges ab-
solutely. Since 0 ≤ (1 − y)/(1 − xy) ≤ 1 whenever 0 < x, y < 1, the second integral con-
verges absolutely as well.
Now I(s) is an analytic function of s in the right half-plane, so it suffices to prove the
result for 0 < s < 1. In this case, the integrand is real and has constant sign, so we may
interchange the order of integration. Thus,
 1  1  1  y
(xy)s−1 − y dx t s−1 − y
I(s) = dy = dt dy
0 0 1 − xy log(xy) 0 0 y(1 − t ) log t
 1  1 s−1
1 t −y
= dy dt
0 (1 − t ) log t t y
 1 s−1  1  s−1
−t log t − (1 − t ) −t 1
= dt = − dt.
0 (1 − t ) log t 0 1−t log t
This is a well-known integral representation of the digamma function ψ (s) =  (s)/(s)
due to Gauss.

Also solved by M. Arnold, A. Berkane (Algeria), P. Bracken, R. Chapman (U. K.), H. Chen, B. Davis,
C. Georghiou (Greece), G. Greubel, J.-P. Grivaux (France), J. A. Grzesik, E. Herman, R. Nandan, M. O’Brien,
M. Omarjee (France), F. Perdomo & Á. Plaza (Spain), P. Perfetti (Italy), S. Sharma, A. Stadler (Switzerland),

August–September 2018] PROBLEMS AND SOLUTIONS 663

R. Stong, R. Tauraso (Italy), M. Vowe (Switzerland), M. Wildon (U. K.), Y. Zhang, GCHQ Problem Solving
Group (U. K.), and the proposer.

An Inequality for Triangles

11938 [2016, 941]. Proposed by Martin Lukarevski, University “Goce Delcev,” Stip, Mace-
donia. Let a, b, c be the lengths of the sides of a triangle, and let A be its area. Let R and r
be the circumradius and inradius of the triangle, respectively. Prove

R − 2r
a + b + c ≥ (a − b) + (b − c) + (c − a) + 4A 3 +
2 2 2 2 2 2
.
R
Solution by John G. Heuver, Grande Prairie, AB, Canada. Let ∠A = α, ∠B = β, and
∠C = γ . By the law of cosines
α
a2 = b2 + c2 − 2bc cos α = (b − c)2 + 2bc(1 − cos α) = (b − c)2 + 4A tan ,
2
where we have used 2A = bc sin α and (1 − cos α)/ sin α = tan(α/2). It follows that

α β γ
a + b + c = (a − b) + (b − c) + (c − a) + 4A tan + tan + tan
2 2 2 2 2 2
.
2 2 2
We have
α β γ 4R + r
+ tan + tan =
tan ,
2 2 2 s
where s is the semiperimeter of the triangle. (This is equation 83 on page 59 of D. S. Mitri-
novic (1989), Recent Advances in Geometric Inequalities, Dordrecht: Kluwer.) Kooi’s
inequality
R(4R + r)2
s2 ≤
2(2R − r)
(see, for example, item 5.7 in O. Bottema, et. al. (1969), Geometric Inequalities, Groningen:
Wolters-Noordhoff) then gives

α β γ R − 2r
tan + tan + tan ≥ 3 + .
2 2 2 R
This completes the proof. Equality holds if and only if the triangle is equilateral.

Also solved by A. Ali (India), R. Boukharfane (France), P. P. Dályay (Hungary), L. Giugiuc (Romania),
B. Karaivanov (U. S. A.) and T. S. Vassilev (Canada), O. Kouba (Syria), K.-W. Lau (China), J. H. Lindsey II,
D. Moore, R. Nandan, P. Nüesch (Switzerland), P. Perfetti (Italy), V. Schindler (Germany), M. Stănean (Ro-
mania), R. Stong, M. Vowe (Switzerland), T. Wiandt, J. Zacharias, L. Zhou, GCHQ Problem Solving Group
(U. K.), and the proposer.

Summing Errors in Approximations to Euler’s Constant

11939 [2016, 941]. Proposed by Moubinool Omarjee, Lycée Henri IV, Paris, France. Find
∞ 
1 1 1 1
1 + + · · · + − log(k) − γ − + .
k=1
2 k 2k 12k2

Here γ is Euler’s constant.

664 
C THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 125
Solution by Roberto Tauraso, Università di Roma “Tor Vergata,” Rome, Italy. Let
Hk = 1 + 1/2 + · · · + 1/k. We have

n 
n 
k
1 
n
1
n n
n+1− j
Hk = = 1= = (n + 1)Hn − n.
k=1 k=1 j=1
j j=1
j k= j j=1
j

Hence,
n 

1 Hn
Hk − log(k) − γ − = (n + 1)Hn − n − log(n!) − nγ −
k=1
2k 2
 
1 1
= n+ log(n) + γ + + O(1/n ) − n − nγ
2
2 2n

log(2π ) log(n)
− n log(n) − n + + + O(1/n)
2 2
1 + γ − log(2π )
= + O(1/n),
2
where we have used the approximations Hn = log(n) + γ + 1
2n
+ O(1/n2 ) and log(n!) =
n log(n) − n + log(2π
2
)
+ log(n)
2
+ O(1/n). Also,
∞
1 1 π2 π2
= · = .
k=1
12k2 12 6 72

Combining these results, we obtain

∞ 
1 1 1 + γ − log(2π ) π 2
Hk − log(k) − γ − + = + .
k=1
2k 12k2 2 72

Editorial comment. Several solvers noted that the requested sum, without the final term
1/(12k2 ), appears as Problem 3.42 on page 195 of O. Furdui (2013), Limits, Series, and
Fractional Part Integrals: Problems in Mathematical Analysis, New York: Springer. The
more general formula

∞  
π 
p−1
1 log(p) + γ − log(2π ) 1 kπ
Hpk − log(pk) − γ − = + + 2 k cot ,
k=1
2pk 2 2p 2p k=1
p

where p is a positive integer, appears in O. Kouba (2016), Inequalities for finite trigono-
metric sums. An interplay: with some series related to harmonic numbers, J. Inequal. Appl.,
Paper No. 173, 15 pp.
Also solved by A. Balfaqih (Yemen), A. Berkane (Algeria), R Boukharfane (France), P. Bracken, R. Chapman
(U. K.), H. Chen, R. Guculiére (France), R. Dutta (India), O. Furdui (Romania), N. Ghosh, M. L. Glasser,
J. A. Grzesik, L. Han, E. A. Herman, E. J. Ionaşcu, B. Karaivanov (U. S. A.) & T. S. Vassilev (Canada),
O. Kouba (Syria), C. W. Lienhard, O. P. Lossers (Netherlands), G. N. Macris, P. Magli (Italy), C. R. McCarthy,
R. Nandan, P. Perfetti (Italy), F. A. Rakhimjanovich (Uzbekistan), E. Schmeichel, A. Stadler (Switzerland),
A. Stenger, R. Stong, M. Vowe (Switzerland), S. Wagon, H. Widmer (Switzerland), J. Zacharias, Y. Zhang,
GCHQ Problem Solving Group (U. K.), and the proposer.

A Hypergeometric Identity
11940 [2016, 942]. Proposed
 by Hideyuki Ohtsuka, Saitama, Japan. Let Tn = n(n + 1)/2
and C(n, k) = (n − 2k) nk . For n ≥ 1, prove

August–September 2018] PROBLEMS AND SOLUTIONS 665

 
n−1  
n3 − 2n2 + 4n Tn Tn+1
C(Tn , k)C(Tn+1 , k) = .
k=0
n+2 n n

Solution I by Pierre Lalonde, Kingsey Falls, QC, Canada. Let m be a positive integer. We
prove by induction on m the more general formula

m−1  
m2 (n2 + 2n − 4m + 4) Tn Tn+1
C(Tn , k)C(Tn+1 , k) = .
k=0
n(n + 2) m m

For m = 1 both sides give Tn Tn+1 . Given the formula for m, we compute

m 
m−1
C(Tn , k)C(Tn+1 , k) = C(Tn , k)C(Tn+1 , k) + C(Tn , m)C(Tn+1 , m)
k=0 k=0
  
m2 (n2 + 2n − 4m + 4) Tn Tn+1
= + (Tn − 2m)(Tn+1 − 2m)
n(n + 2) m m
 
(n2 + 2n − 4m) (n2 + n − 2m)(n2 + 3n − 2m + 2) Tn Tn+1
=
n(n + 2) 4 m m
 
(m + 1)2 (n2 + 2n − 4m) (Tn − m)(Tn+1 − m) Tn Tn+1
=
n(n + 2) (m + 1)2 m m
 
(m + 1) (n + 2n − 4m)
2 2
Tn Tn+1
= ,
n(n + 2) m+1 m+1
where the step from the second to the third line is easy (though tedious) to check. The
special case m = n gives the desired result.
Solution II by Akalu Tefera, Grand Valley State University,
 Allendale, MI. Dividing both
sides of the desired equality by its right side yields n−1
k=0 F (n, k) = 1, where

n+2 C(Tn , k)C(Tn+1 , k)

F (n, k) = Tn Tn+1  .
n3 − 2n + 4n
2
n n

Applying Gosper’s algorithm to F (n, k) produces a rational function

4k2 (n2 + 2n − 4k + 4)
R(n, k) =
n(n + 2)(n2 + n − 4k)(n2 + 3n − 4k + 2)
such that setting G(n, k) = F (n, k)R(n, k) yields F (n, k) = G(n, k + 1) − G(n, k), which
can be confirmed easily. Summing both sides of this equality with respect to k then gives
the telescoping sum

n−1 
n−1
 
F (n, k) = G(n, k + 1) − G(n, k) = G(n, n) − G(n, 0) = 1.
k=0 k=0

Also solved by R. Chapman (U. K.), R. Stong, R. Tauraso (Italy), and the proposer.

Rate of Convergence for an Integral

11941 [2016, 492]. Proposed by Ovidiu Furdui, Technical University of Cluj-Napoca, Cluj-
Napoca, Romania. Let
 1
L = lim x + (1 − x)n dx.
n n

n→∞ 0

666 
C THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 125
(a) Find L.
(b) Find
 1
lim n 2 n
xn + (1 − x)n dx − L .
n→∞ 0

Solution by Hongwei Chen, Christopher Newport University, Newport News, VA. (a) We
1 √
prove L = 3/4. To see this, let In = 0 n xn + (1 − x)n dx. We have
 1/2   1 
In = x + (1 − x) dx +
n n n x + (1 − x)n dx
n n

0 1/2
 1/2  1
3
≥ (1 − x) dx + x dx = .
0 1/2 4
On the other hand, since x ≤ 1 − x for x ∈ [0, 1/2] and 1 − x ≤ x for x ∈ [1/2, 1],
 1/2   1 
In = x + (1 − x)n dx +
n n
x + (1 − x)n dx
n n

0 1/2
 √  √
1/2 1
3√
≤ 2 (1 − x) dx + 2 x dx =
n n n
2.
0 1/2 4
The squeeze theorem implies that L = limn→∞ In = 3/4.
 1/2 1
(b) The limit is π 2 /48. Notice that 0 (1 − x) = 1/2 x dx = 3/8. We claim
 1/2
 3 π2
lim n 2
x + (1 − x) dx −
n n n = (1)
n→∞ 0 8 96
and

1  3 π2
lim n 2 n
xn + (1 − x)n dx − = , (2)
n→∞ 1/2 8 96
from which the required limit follows. To prove (1), we compute
 1/2 

lim n 2
x + (1 − x) − (1 − x) dx
n n n
n→∞ 0
   n 
1/2
x
= lim n 2
(1 − x) n
1+ −1 dx
n→∞ 0 1−x
 1 √
1
= lim n2 + n−1 (letting t = x/(1 − x))
n
1 t dt
0 (1 + t )
n→∞ 3
 1 √
1
= lim n + − 1/n−1
(letting u = t n )
n
1 u 1 u du
n→∞ 0 (1 + u 1/n )3
 1 √
1
= + − u1/n−1 du
n
lim n 1 u 1
0 n→∞ (1 + u
1/n )3
 ∞
1 1 ln(1 + u) 1  (−1)n+1 π2
= du = = .
8 0 u 8 n=1 n2 96

Equation (2) follows from (1) upon substituting 1 − x for x.

August–September 2018] PROBLEMS AND SOLUTIONS 667

Editorial comment. Chen noted that the results can be generalized as follows. For part (a):
If f and g are nonnegative and integrable on [a, b], then

 b   b
lim n
f (x)n + g(x)n dx = max{ f (x), g(x)} dx.
n→∞ a a

For part (b): If f is a positive continuous function on [0, 1] with f (0) = 1 and g(x) is con-
tinuous on [0, 1], then
 1  1  1
 ln f (x)
lim n2 n
f (xn )g(x) dx − g(x) dx = g(1) dx.
n→∞ 0 0 0 x
Letting f (x) = 1 + x and g(x) = 1/(1 + x)3 yields the result in part (b).
Also solved by R. Agnew, K. F. Andersen (Canada), A. Berkane (Algeria), R. Boukharfane (France), P. Bracken,
R. Chapman (U. K.), P. P. Dályay (Hungary), B. E. Davis, R. Dutta (India), D. Fleischman, N. Ghosh,
J.-P. Grivaux (France), L. Han, F. Holland (Ireland), E. J. Ionaşcu, O. Kouba (Syria), J. H. Lindsey II,
O. P. Lossers (Netherlands), S. de Luxán (Germany) & Á. Plaza (Spain), M. Omarjee (France), N. Osipov
(Russia), P. Perfetti (Italy), A. Stadler (Switzerland), A. Stenger, R. Stong, R. Tauraso (Italy), GCHQ Problem
Solving Group (U. K.), NSA Problems Group, and the proposer.

On Perpendicularity
11942 [2016, 492]. Proposed by Florin Parvanescu, Slat, Romania. In acute triangle ABC,
let D be the foot of the altitude from A, let E be the foot of the perpendicular from D to AC,
and let F be a point on segment DE. Prove that AF is perpendicular to BE if and only if
|FE|/|FD| = |BD|/|CD|.
Solution by Wei-Kai Lai and John Risher (student), University of South Carolina Salke-
−
→ − →
hatchie, Walterboro, SC. Note that since AD · BD = 0,
−
→ − → −→ −→ − → −→ −→ −→ −→ − → −→ −→
AF · BE = (AD + DF ) · (BD + DE ) = AD · DE + DF · BD + DF · DE
−→ −→ −→
= (AE − DE ) · DE + |DF| |BD| cos(∠EDC) + |DF| |DE|
|DE|
= −|DE|2 + |DF| |BD| + |DF| |DE|. (1)
|DC|
Consider first the necessity of the condition. When AF ⊥ BE, (1) yields |DF| |BD| +
|DF| |DC| = |DE| |DC|. Since |DE| = |DF| + |FE|, we get
|DF| |BD| + |DF| |DC| = |DF| |DC| + |FE| |DC|,
which implies |DF| |BD| = |FE| |DC| as required.
Now consider the sufficiency of the condition. Since |DE| = |DF| + |FE|, and
|FE|/|FD| = |BD|/|CD| is assumed, we can write (1) in the equivalent form
−
→ − → |FE|
AF · BE = −(|DF| + |FE|)2 + |DF| |DE| · + |DF|(|DF| + |FE|)
|FD|
= −|DF|2 − 2|DF| · |FE| − |FE|2 + (|DF| + |FE|)|FE| + |DF|2 + |DF| · |FE|.
This equals zero, and hence AF is perpendicular to BE, as desired.
Also solved by A. Ali (India), H. Bailey, R. Chapman (U. K.), P. P. Dályay (Hungary), P. De (India), I. Dim-
itrić, A. Fanchini, D. Fleischman, O. Geupel, L. Giugiuc (Romania), N. Grivaux (France), J. Han (South Korea),
E. A. Herman, S. Hitotumatu (Japan), E. J. Ionaşcu, Y. Ionin, S.-H. Jeong (Korea), O. Kouba (Syria), J. H. Lind-
sey II, O. P. Lossers (Netherlands), M. D. Meyerson, I. Mihăilă, J. Minkus, R. Nandan, A. Stadler (Switzerland),
R. Stong, R. Tauraso (Italy), M. Vowe (Switzerland), T. Wiandt, L. Zhou, T. Zvonaru & N. Stanciu (Romania),
Armstrong Problem Solvers, GCHQ Problem Solving Group (U. K.), and the proposer.

668 
C THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 125