Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
PROBLEMS
SOLUTIONS
72
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
Can One Recover a (Spherical) Triangle from Its Medial Triangle?
11201 [2006, 179]. Proposed by Robert Russell, New York, NY. Given the midpoints
of the sides of a spherical triangle, provide a construction for the original triangle.
Composite solution by Christoph Soland (Switzerland), William Dickenson, and the
editors. There is one configuration of midpoints, namely, mutually perpendicular, for
which many choices of original triangle work. Otherwise, the underlying triangle may
be recovered by a construction.
We take it to be part of the definition of a spherical triangle that the three vertices
do not lie on a common great circle. We also assume (by definition) that the sides of
a spherical triangle are less than π radians. Thus, to construct a spherical triangle, it
suffices to construct the three vertices. Let O be the center of our sphere S . Without
loss of generality, we take O to be the origin of R3 and the radius of S to be 1. When
A and B are not antipodal, let m(A, B) be the midpoint of the spherical line segment
AB, defined as the shorter of the two portions of the great circle through A and B
delimited by A and B.
We first consider the exceptional case. If any side of the medial triangle M1 M2 M3
is π/2, then, we claim, all of them are, and there are infinitely many triangles with the
same three midpoints, so we cannot uniquely construct the “original triangle”. For the
proof of this claim, suppose that M1 ⊥ M2 . Without loss of generality, M1 = (1, 0, 0)
and M2 = (0, 1, 0). By assumption, there exist distinct S1 , S2 , and S3 , no two antipodal,
such that for any permutation (i, j, k) of (1, 2, 3), m(Si , S j ) = Mk . Note that the ends
of a line segment in spherical geometry are nearer the midpoint than its antipode.
Now if S2 = (x2 , y2 , z 2 ), then x2 > 0, S3 = (x2 , −y2 , −z 2 ) so that y2 < 0, and S1 =
(−x2 , −y2 , z 2 ). Now z 2 = 0 since S1 = −S2 , and it follows that M3 = (0, 0, sign(z 2 )).
Thus all three axes are perpendicular. Glancing over the foregoing algebra, we see that
any choice of S2 = (x2 , y2 , z 2 ) with x2 , z 2 > 0 > y2 and x22 + y22 + z 22 = 1 gives rise to
S2 , S3 , and S1 for which (1, 0, 0), (0, 1, 0), and (0, 0, 1) are the vertices of the medial
triangle.
We now consider the composition R = H3 ◦ H2 ◦ H1 , where H j is the bijection of
S obtained by reflecting it through the axis of ±M j . Either R is the identity map on S ,
or it is a nontrivial rotation that fixes two antipodal points, call them ±S. We claim that
the M j are mutually perpendicular if and only if R = I . In one direction this is trivial
by coordinate algebra after setting M1 = (1, 0, 0) and so on. In the other direction,
suppose R = I . Then for any permutation (i, j, k) of (1, 2, 3), Hi ◦ H j = Hk . Thus
the M j are distinct, and no two are antipodal. Now let P = H2 M1 and Q = H3 M1 . If
M1 ⊥ M2 , then M1 and M2 determine a line L on S , P, M1 , and M2 are distinct points
on L, and P is not antipodal to M1 .
Furthermore, H1 Q = P since H1 H3 H2 = I . Thus Q = H1 P. Since P lies on L
and H1 carries any line through M1 onto itself, Q is on L. From Q = H3 M1 , we
have also that Q is on the line L through M1 and M3 . Thus Q = M1 , Q = −M1 , or
L = L . But Q = M1 because M3 = ±M1 , and Q = −M1 because that would make
P = H1 Q = −M1 , a contradiction. Finally, if L = L , each H j transposes the poles
of S if we take L as the equator, and so R transposes those poles, contrary to the the
assumption that R = I . This proves that M1 ⊥ M2 , and by the same logic, all three
M j are mutually perpendicular.
If, on the other hand, no two of the M j are perpendicular, then the underlying S j
may be recovered from the M j by a construction.
Henceforth, we assume that no two of the M j are orthogonal. Thus, R = I . We
first show that the axis ±S of rotation for R is not perpendicular to M1 . To see this,
suppose to the contrary that H3 H2 H1 S = S, and S ⊥ M1 . Take S = H1 S and S =
74
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
converges uniformly, so g is differentiable and g (x) is the sum of this derived series.
We claim that g is discontinuous at each rational q j . Write (1) as
−2k k
g (x) = 2−2 j f 2 j (x − q j ) + 2 f 2 (x − qk ) .
k = j
The last series defines a function continuous at q j since each term is continuous there
and the series converges uniformly. The discontinuity of f at 0 then shows that g is
discontinuous at q j .
For n a positive integer, let E n = ∞ k=n (qk − 2
1−k
, qk + 21−k ) and E = ∞ n=1 E n .
Thus E n has Lebesgue measure at most 23−n and E has measure zero, so the set of
irrationals in the complement of E is dense. We claim g is differentiable at every
irrational x0 ∈ E. Since x0 is neither qk nor qk + 2−k for any k, each term 2−3k f (2k (x −
qk )) is twice differentiable at x0 . Let n be such that x0 ∈ E n . Then the derived series
(1) reduces to a finite series, and its derivative at x0 is
n−1
2−k f 2k (x0 − qk ) . (2)
k=1
α(x) − α(x0 ) ≤ C 2−k |x − x0 | ≤ 2C|x − x0 |2 ,
2−k ≤|x−x 0 |
where C is a bound for | f |. Hence α (x0 ) = 0, so g (x0 ) is indeed given by the finite
series (2).
Also solved by J. H. Lindsey II, M. D. Meyerson, A. Stadler (Switzerland), R. Stong, Szeged Problem Solving
Group “Fejéntaláltuka” (Hungary), GCHQ Problem Solving Group (U. K.), and the proposer.
A Unique Solution
11226 [2006, 460]. Proposed by Franck Beaucoup, Ottawa, Canada, and Tamás
Erdélyi, Texas A& M University, College Station, TX. Let a1 , . . . , an be real numbers,
each greater than 1. For n ≥ 2, show that there is exactly one solution in the interval
(0, 1) to
n
(1 − x a j ) = 1 − x.
j =1
n a j −2 a j a j −2
j =1 a j x (x − 1 − a j x + a j )(1 − x ) and is thus positive on (0, 1).
Also solved by K. F. Andersen (Canada), R. Bagby, P. P. Dályay (Hungary), J. H. Lindsey II, O. P. Lossers
(Netherlands), T. L. McCoy (Taiwan), R. Mortini (France), L. Pebody, J. Rooin and A. Mahmoodi (Iran), A.
Stadler (Switzerland), R. Stong, BSI Problems Group (Germany), GCHQ Problem Solving Group (U. K.), and
the proposers.
76
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
A Triangle Inequality
11228 [2006, 460]. Proposed by Marian Tetiva, Bı̂rlad, Romania. Prove that in an
acute triangle with angles A, B, and C radians,
(1 − cos A)(1 − cos B)(1 − cos C)
cos A cos B cos C
8(tan A + tan B + tan C)3
≥ .
27(tan A + tan B)(tan A + tan C)(tan B + tan C)
A Function Inequality
11232 [2006, 567]. Proposed by Michael W. Botsko, Saint Vincent College, Latrobe,
PA. Let f be a continuous mapping from R into R that is bounded below. Show that
there exists a real number x0 such that f (x0 ) − f (x) < |x − x0 | holds for all x other
than x0 .
Solution by Albert Stadler, Dübendorf, Switzerland. Put g(x) := f (x) + |x|/2. Then
g is continuous, bounded below, and g(x) → +∞ as x → ±∞. Therefore there is a
real number x0 such that g(x0 ) = minx∈R g(x). Thus f (x) + |x|/2 ≥ f (x0 ) + |x0 |/2
or f (x0 ) − f (x) + |x0 |/2 − |x|/2 ≤ 0. We conclude that for x = x0 ,
f (x0 ) − f (x) − |x − x0 | < f (x0 ) − f (x) − |x − x0 |/2
≤ f (x0 ) − f (x) + |x0 |/2 − |x|/2 ≤ 0.
A Derivative Formula
11233 [2006, 568]. Proposed by Robert Downes, Mountain Lakes High School, Moun-
tain Lakes, NJ. Show that for positive integer n, and for x = 0,
dn 1 (−1)n 1 nπ
x n−1
sin = n+1 sin + .
dxn x x x 2
Solution by Jean-Pierre Grivaux, Paris, France. We will show more generally that if
f is an n-times differentiable real-valued function, then
dn 1 (−1)n (n) 1
x n−1
f = n+1 f .
dxn x x x
The proof is by induction on n. The case n = 1 is obvious. Suppose it is true for
n − 1. Applying the inductive hypothesis, and taking g(x) = x n−2 f (1/x) in Leibniz’s
formula
dn dn g d n−1 g
n
(xg(x)) = x n + nx n−1 ,
dx dx dx
we compute
dn 1 dn 1 d n−1 1
n
x n−1
f =x n x n−2
f + nx n−1 x n−2
f
dx x dx x dx x
d (−1) n−1
1 (−1) n−1
1
=x n
f (n−1) + nx n
f (n−1)
dx x x x x
(−1) n (n−1) 1
n
(−1) (n) 1
n
(−1) n (n−1) 1
n−1
=x f + f + f
x n+1 x x n+2 x x n−1 x
(−1) n
1
= n+1 f (n) .
x x
78
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
PROBLEMS AND SOLUTIONS
Edited by Gerald A. Edgar, Doug Hensley, Douglas B. West
with the collaboration of Paul T. Bateman, Mario Benedicty, Itshak Borosh, Paul
Bracken, Ezra A. Brown, Randall Dougherty, Tamás Erdélyi, Zachary Franco, Chris-
tian Friesen, Ira M. Gessel, Jerrold Grossman, Frederick W. Luttmann, Vania Mas-
cioni, Frank B. Miles, Richard Pfiefer, Cecil C. Rousseau, Leonard Smiley, John
Henry Steelman, Kenneth Stolarsky, Richard Stong, Walter Stromquist, Daniel Ull-
man, Charles Vanden Eynden, and Fuzhen Zhang.
PROBLEMS
11341. Proposed by Cezar Lupu, University of Bucharest, Bucharest, Romania (stu-
dent), and Tudorel Lupu, Decebal High School, Constanza, Romania. Consider an
acute triangle with side-lengths a, b, and c, with inradius r and semiperimeter p. Show
that
√
3 3r
(1 − cos A)(1 − cos B)(1 − cos C) ≥ cos A cos B cos C 2 − .
p
11344. Proposed by Albert Stadler, Meilen, Switzerland. Let μ be the Möbius μ func-
tion of number theory. Show that if n is a positive integer and n > 1, then
n
(n−1)/2
n/(2 j +1)
μ( j) = − j μ(k).
j =1 j =1 k=(n+1)/(2 j +3)
11345. Proposed by Roger Cuculière, France. Find all nondecreasing functions f from
R to R such that f (x + f (y)) = f ( f (x)) + f (y) for all real x and y.
11346. Proposed by Christopher Hillar, Texas A&M University, College Station, TX,
and Lionel Levine, University of California, Berkeley, CA. Let n be an integer greater
166
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
than 1 and let S = {2, . . . , n}. For each nonempty subset A of S, let π(A) = j ∈A j.
Prove that when k is a positive integer and k < n,
n
lcm({1, . . . , n/i}) = gcd ({π(A) : |A| = n − k}) .
i=k
n
(In particular, setting k = 1 yields i=1 lcm({1, . . . , n/i}) = n!.)
√
11347. Proposed by Mihály Bencze, Brasov, Romania. Let A = (x + y)/2, G = x y,
and
1 x x 1/(x−y)
I = .
e yy
Determine all ordered 4-tuples (α, β, γ , δ) of positive numbers with α > β and γ > δ
such that for all distinct positive x and y,
α A + βG
1/(γ +δ) √
I > > Aγ G δ > AG.
α+β
SOLUTIONS
1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9
4 5 6 7 8 9 1 2 3 5 6 4 8 9 7 2 3 1
7 8 9 1 2 3 4 5 6 9 7 8 3 1 2 6 4 5
3 1 2 6 4 5 9 7 8 6 4 5 9 7 8 3 1 2
6 4 5 9 7 8 3 1 2 7 8 9 1 2 3 4 5 6
9 7 8 3 1 2 6 4 5 2 3 1 5 6 4 8 9 7
2 3 1 5 6 4 8 9 7 8 9 7 2 3 1 5 6 4
5 6 4 8 9 7 2 3 1 3 1 2 6 4 5 9 7 8
8 9 7 2 3 1 5 6 4 4 5 6 7 8 9 1 2 3
1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9
8 9 7 2 3 1 5 6 4 9 7 8 3 1 2 6 4 5
6 4 5 9 7 8 3 1 2 5 6 4 8 9 7 2 3 1
5 6 4 8 9 7 2 3 1 8 9 7 2 3 1 5 6 4
3 1 2 6 4 5 9 7 8 4 5 6 7 8 9 1 2 3
7 8 9 1 2 3 4 5 6 3 1 2 6 4 5 9 7 8
9 7 8 3 1 2 6 4 5 6 4 5 9 7 8 3 1 2
4 5 6 7 8 9 1 2 3 2 3 1 5 6 4 8 9 7
2 3 1 5 6 4 8 9 7 7 8 9 1 2 3 4 5 6
Editorial comment. Lyle Ramshaw provided a proof that 6 is the maximum number of
pairwise orthogonal Latin squares of order 9. These results and others are dealt with
in “Sudoku, gerechte designs, resolutions, affine space, spreads, reguli, and Hamming
codes” by R. A. Bailey, Peter J. Cameron, and Robert Connelly, to appear shortly in
this M ONTHLY.
Also solved by R. Bagby, J. Brawner, R. Chapman (U. K.), K. Dale (Norway), D. Degiorgi (Switzerland),
G. T. Fala, R. T. Guy, D. E. Knuth, O. P. Lossers (Netherlands), M. D. Meyerson & T. S. Michael, R. M. Ped-
ersen, L. Ramshaw, T. Q. Sibley, J. H. Steelman, R. Stong, H. Stubbs, R. Tauraso (Italy), L. Wenstrom, the
ABC Student Problem Solving Group, the BSI Problems Group (Germany), Szeged Problem Solving Group
“Fejéntaláltuka”(Hungary), the GCHQ Problem Solving Group (U. K.), the Northwestern University Math
Problem Solving Group, the VMI Problem Solving Group, and the proposer.
168
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
Observe that
1
L 1 2 L 1
f (t) dt = 1− +L −1 + 1− = 1.
0 2 L L 2 L
We have v(t) ≤ f (t) for t ∈ [0, 1]. Equality cannot hold everywhere, since f is not
differentiable at 1/L. Hence v(t1 ) < f (t1 ) at some t1 ∈ [0, 1] and, by continuity,
v(t) < f (t) in some neighborhood of t1 . Our assumption that |v
| ≤ M on [0, 1] thus
leads to this contradiction:
1 1
1 = x(1) − x(0) = v(t) dt < f (t) dt = 1.
0 0
Also solved by T. Achenbach, M. W. Botsko, D. Chakerian, P. Corn, J.-P. Grivaux (France), E. A. Herman, J. H.
Lindsey II, O. P. Lossers (Netherlands), T. L. McCoy (Taiwan), K. McInturff, M. D. Meyerson, R. E. Prather,
K. Schilling, J. Silver, J. G. Simmonds, R. Stong, J. B. Zacharias, L. Zhou, Szeged Problem Solving Group “Fe-
jéntaláltuka” (Hungary), GCHQ Problem Solving Group (U. K.), Houghton College Problem Solving Group,
Microsoft Research Problems Group, NSA Problems Group, and the proposer.
170
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
different colors. Similarly B, D, E, F have different colors. Thus A and B have the
same color. Since B was an arbitrary point on the circle with center A and radius 2,
that circle is monochromatic. This gives a contradiction, because there are two points
on that circle with distance √1. √
Next we prove P((1 + 3)/ 2, 4); since this is not listed in (a), it will be an
example for (b). It is equivalent to prove: In every 4-coloring of the
√ Euclidean plane
√
there are two points having the same color that are distance 1 + 3 or distance 2
apart. Let A and B be any two points with distance√ 2. Let AB be the hypotenuse of
isosceles right
√ triangle ABC so that AC = BC = 2. On the circle centered at A
with radius 2, take D and E to be the two points such that D AB =√ E AB = 105
A B
E
√
√ BC = AC = AD = AE = C D = 2, while B D = D E =
In this configuration
E B = EC = 1 + 3. Assuming a 4-coloring of the plane contrary to the claim, the
four points A, C, D, and E have different colors, as do the four points B, C, D, and
E. Therefore A and B have the same color. Since A and B were arbitrary points with
distance 2, the entire circle centered at A with radius 2 has just one color. This gives
√ a
contradiction, because√there are two points on that circle separated by distance 2.
Next
√ consider P( 2, 4). Assume a 4-coloring of the plane inconsistent√ with
P( 2, 4), that is, one √ in which
√ any two points with distance 1 or 2 have differ-
√ colors. By P((1 + 3)/ 2, 4), just proved, since any two points with distance
ent √
2 have different colors, there must exist two points X and T with distance 1 + 3
having the same color; call it color 1. Let X Y Z T U V be a convex hexagon such that
points Y and V are symmetric with respect to line X T , points Z and U are symmetric
with respect to line X T , and X Y V and T U Z are equilateral triangles with side 1.
Now Y , Z, U , and V all have colors different from color 1, so some two of them have
the same color.
√ This gives a contradiction, because Y ZU V forms a square with side 1
and diagonal 2.
Solution to (c) by √Matthew Huddleston, Washington State University, Pullman, WA.
We prove P((1 + 5)/2, 5). Suppose √ there is a 5-coloring of the plane so that no two
points with distance 1 or (1 + 5)/2 have the same color. Let S be a set of five points
forming the vertices of a regular pentagon with side 1. Let Q be the set of ordered
quintuples chosen from S, so that Q has 55 elements. Color Q by assigning to each of
its 5-tuples the color, in the original coloring, of the sum of its five entries.
For any 4-tuple of points in S, note that adding the sum of its entries to each of the
five points of S gives a regular pentagon with side 1, so these five points have different
colors. Therefore, each of the five colors is assigned to 54 of the 55 elements of Q. On
the other hand, permuting an ordered 5-tuple in Q cannot change its color. The number
Ex-To-In-Radius Ratio
11240 [2006, 655]. Proposed by Pál Péter Dályay, Deák Ferenc High School, Szeged,
Hungary. Let a, b, and c be the lengths of the sides of a triangle, and let R and r be
the circumradius and inradius of that triangle, respectively. Show that
R (a − b)2 (b − c)2 (c − a)2
≥ exp + + .
2r 2c2 2a 2 2b2
Solution by A. K. Shafie, IASBS, Zanjan, Iran. Write A, B, C for the angles opposite
sides a, b, c, repectively, and s for the semiperimeter. We have
A A B C A B C
r = (s − a) tan = a sec sin sin = 4R sin sin sin ,
2 2 2 2 2 2 2
so sin(A/2) sin(B/2) sin(C/2) = r/(4R). Now e ≤ 1/(1 − x) for all x ∈ [0, 1).
x
172
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
Iterational Rate of Convergence
11244 [2006, 759]. Proposed by Jolene Harris and Bogdan Suceavă, California State
University, Fullerton, CA. Let f be a differentiable function from the positive reals to
the positive reals with the property that f (x) < x for all x. Suppose that x1 > 0, and
for n > 1 let xn = f (xn−1 ). Suppose further that limn→∞ xn = 0 and that there exist
positive numbers a and k such that
x a − ( f (x))a 1
lim = a.
x→0 x ( f (x))
a a k
(a) Prove that limn→∞ n 1/a xn = k.
(b) Suppose that 0 < x1 < 1, and specialize to the case where f is given
√ by f (x) =
√
sin x if x < π/2 and f (x) = 1 if x ≥ π/2. Show that limn→∞ xn n = 3.
(c) Finally, suppose instead that 0 < x1 < 1 and f (x) = 1 − e−x . Show that, in this
case, limn→∞ nxn = 2.
Solution by Knut Dale, Telemark University College, Bø, Norway. (a) We are given
f (x)−a − x −a = k −a + (x), where (x) → 0 as x → 0. So
n−1
1 1 1 1 n−1
− a = a − a = + (x1 ) + · · · + (xn−1 ) and
xna x1 i=1
x i+1 x i ka
1 1 n − 1 (x1 ) + · · · + (xn−1 ) 1
a
= a + a
+ → a
nxn nx1 nk n k
as n → ∞. Therefore n 1/a xn → k.
(b) Let a = 2. Compute
x 2 − sin2 x x 2 − sin2 x 1
lim 2
= lim 4
=
2
x→0 x sin x x→0 x 3
√
by L’Hopital’s rule (or Maclaurin series). Hence k = 3.
(c) Let a = 1. Compute
x − (1 − e−x ) 1
lim =
x→0 x(1 − e−x ) 2
by writing e−x = 1 − x + x 2 /2 − x 2 ω(x) where ω(x) → 0 as x → 0. Hence k = 2.
Editorial comment. Note that the differentiability of f is not needed.
Also solved by S. Amghibech (Canada), M. R. Avidon, O. Bagdasar (Romania), P. Bracken, D. R. Bridges, R.
Chapman (U. K.), P. P. Dályay (Hungary), J.-P. Grivaux (France), J. H. Lindsey II, O. López & N. Caro (Brazil),
M. McMullen, M. D. Meyerson, J. Rooin & A. Morassaei (Iran), K. Schillling, H.-J. Seiffert (Germany),
N. C. Singer, A. Stadler (Switzerland), A. Stenger, R. Stong, M. Tetiva (Romania), D. Văcaru (Romania),
J. Vinuesa (Spain), Z. Vörös (Hungary), BSI Problems Group (Germany), GCHQ Problem Solving Group
(U. K.), Microsoft Research Problems Group, NSA Problems Group, Northwestern University Math Problem
Solving Group, and the proposers.
PROBLEMS
11348. Proposed by Richard P. Stanley, Massachusetts Institute of Technology, Cam-
bridge, MA. A polynomial f over a field K is powerful if every irreducible factor of f
has multiplicity at least 2. When q is a prime or a power of a prime, let Pq (n) denote
the number of monic powerful polynomials of degree n over the finite field Fq . Show
that for n ≥ 2,
Pq (n) = q n/2 + q n/2−1 − q (n−1)/3 .
11350. Proposed by Bhavana Deshpande, Poona College of Arts, Science & Com-
merce Camp, Pune, India, and M. N. Deshpande, Institute of Science, Nagpur, In-
dia. Given a positive integer n and an integer k with 0 ≤ k ≤ n, form a permutation
a = (a1 , . . . , an ) of (1, . . . , n) by choosing the first k positions at random and filling
the remaining n − k positions in ascending order. Let E n,k be the expected number of
left-to-right maxima. (For example, E 3,1 = 2, E 3,2 = 11/6, and E 4,2 = 13/6.) Show
that E n+1,k − E n,k = 1/(k + 1). (A left-to-right maximum occurs at k when a j < ak
for all j < k.)
11351. Proposed by Marian Tetiva, National College “Gheorghe Roşca Codreanu”,
Bârlad, Romania. Given positive integers p and q, find the least positive integer m
such that among any m distinct integers in [− p, q] there are three that sum to zero.
262
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
11352. Proposed by Daniel Reem, The Technion-Israel Institute of Technology, Haifa,
Israel. Let I be an open interval containing the origin, and let f be a twice-
differentiable function from I into R with continuous second derivative. Let T2 be
the Taylor polynomial of order 2 for f at 0, and let R2 be the corresponding remain-
der. Show that
R2 (u) − R2 (v)
lim √ = 0.
(u,v)→(0,0) (u − v) u 2 + v 2
u =v
11353. Proposed by Ernst Schulte-Geers, √ BSI, Bonn, Germany. For s > 0, let f (s) =
∞ s −x
0
(1 + x/s) e d x and g(s) = f (s) − sπ/2. Show that g maps R+ onto (2/3, 1)
and is strictly decreasing on its domain.
11354. Proposed by Matthias Beck, San Francisco State University, San Francisco,
CA, and Alexander Berkovich, University of Florida, Gainesville, FL. Find a polyno-
mial f in two variables such that for all pairs (s, t) of relatively prime positive integers,
s−1
t−1
|mt − ns| = f (s, t).
m=1 n=1
SOLUTIONS
Unsolved in 1990
6576 [1986, 1036]. Proposed by Hans V. Gerber, University of Lausanne, Switzerland.
Suppose X 1 , X 2 , . . . are independent identically distributed real random variables with
E(X k ) = μ. Put Sk = X 1 + X 2 + · · · + X k for k = 1, 2, . . . .
(a) If ρ < μ < 1, where ρ = −0.278465 . . . is the real root of xe1−x = −1, show
that the series
∞
Skk e−Sk /k!
k=1
is μ/(1 − μ).
∗
(c) In (b) is it possible to relax the condition that the random variables are positive?
For example, would it suffice to assume E(|X k |) < ∞ and ρ < μ < 1?
Solution to (c) by Daniel Neuenschwander, Université de Lausanne, Lausanne, and
Universität Bern, Bern, Switzerland. We prove the validity of (c) under the additional
assumption that supp (X 1 ) [the support of X 1 , i.e., the intersection of all closed subsets
A of the real line for which P(X 1 ∈ A) = 1] is contained in the open interval (ρ, −ρ).
Let λ = sup{|x| : x ∈ supp (X 1 )}. Note that λ < −ρ. By Stirling’s formula, one sees
that
(kλ)k (nλ)n
< ∞. (1)
k,n≥0
k! n!
A Determinant Identity
11242 [2006, 656 & 848]. Proposed by Gerd Herzog and Roland Lemmert, Univer-
sität Karlsruhe, Karlsruhe, Germany. (corrected) Let f and g be entire holomorphic
functions of one complex variable, and let A and B be complex n × n matrices. If
the application of such a function to a matrix means applying the power series of this
function to the matrix, prove that
det ( f (A) f (B) + g(A)g(B)) = det ( f (B) f (A) + g(B)g(A)) .
Solution by Roger A. Horn, University of Utah. The crucial property of these matrix
functions is that f (Z) and g(Z) commute whenever both are defined. The following
result is key. (See D. Carlson et al., Linear Algebra Gems, MAA, 2002, p. 13.)
Lemma. Let C, D, X, Y be n × n complex matrices. If C commutes with X , then
C Y
det = det(C D − X Y ).
X D
264
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
Let
f (A) g(B) f (B) −g(A) 0 1
M= , N= , and P= .
−g(A) f (B) g(B) f (A) 1 0
Since M = P N P, det M = (det P)2 det N = det N . Because f (A) commutes with
−g(A) and f (B) commutes with g(B), the lemma ensures that
det( f (A) f (B) + g(A)g(B)) = det M = det N = det( f (B) f (A) + g(B)g(A)).
Also solved by S. Amghibech (Canada), R. Chapman (U. K.), K. Dale (Norway), G. Dospinescu (France), H.
Flanders, J. Grivaux (France), E. A. Herman, J. H. Lindsey II, O. P. Lossers (Netherlands), K. Schilling, R.
Stong, BSI Problems Group (Germany), GCHQ Problem Solving Group (U. K.), Microsoft Research Problems
Group, and the proposers.
A Strip Problem
11247 [2006, 760]. Proposed by Jürgen Eckhoff, University College London, London,
U. K. Let A, B, C, and D be distinct points in the plane with the property that any
three of
√ them can be covered by some strip of width 1. Show that there is a strip of
width
√ 2 covering all four points, and demonstrate that if no strip of √
width less than
2 covers all four, then the points are the corners of a square of side 2. (A strip of
width w is the closed set of points bounded by two parallel lines separated by distance
w.)
Solution by Li Zhou, Polk Community College, Winter Haven, FL. Let S be the set
{A, B, C, D}. If the convex hull of S is a triangle, then that triangle is covered by
a strip of width 1, and so is S. It thus suffices to assume that ABC D is a convex
quadrilateral. Given an arbitrary triangle X Y Z, let h X denote the altitude of X Y Z at
vertex X . For X, Y, Z ∈ S, by the assumed property of ABC D, min{h X , h Y , h Z } ≤ 1.
√
Lemma. Consider a triangle X Y Z√with X, Y, Z ∈ S. If h X ≥ 2, then√ X ≤ 45◦ .
Equality holds if and only if h X = 2, h Y = 1, and Y = 90◦ , or h X = 2, h Z = 1,
and Z = 90◦ .
Now assume that ABC D is labeled clockwise. Without loss of generality, assume that
−→ −→ −→ −→
rays B A and C D do not intersect, and similarly C B and D A do not intersect. The
minimum width of all strips that cover ABC D is the minimum length of the perpen-
diculars √
from A to BC, A to C D, B to AD, and D to AB. Suppose this minimum
exceeds 2. Applying the lemma repeatedly, taking X Y Z to be ABC, AC D, B AD,
and D AB, we conclude that all of B AC, C AD, AB D, and AD B are less than
45◦ . This contradicts the fact that these angles sum to 180◦ as internal
√ angles of tri-
angle AB√ D. Thus ABC D may be covered by a strip of width 2. If the minimum
equals 2, then each of the four angles must equal 45◦ . By the conditions for equality
in the lemma, it follows that ABC D is a square.
Also solved by E. A. Herman, J. H. Lindsey II, B. Schmuland (Canada), R. Stong, M. Tetiva (Romania), GCHQ
Problem Solving Group (U. K.), and the proposer.
Solution
√ by Vitaly Stakhovsky,
√ Redwood City, CA. Let n > 1, 0 ≤ x ≤ 1, x0 = 1/n,
β = 1 + 2x0 , f (x) = nx, and
√ √
n+ n+2 1+β
g(x) = √ √ = √ .
n + n + 2nx 1 + 1 + 2x
266
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
We want to prove that
1 n
1 n
f (xi ) g(xi ) ≤ 1.
n i=1 n i=1
ψ(x)/2, where
√
( 1 + 2x − β)2 2 1 √ 2
ψ(x) = √ + ≤ 1 + 2x − β
1+β 1 + 1 + 2x β
√ √ 2
√ 2 1 + 2x + 1 + 2x0 2 √ 2
≤ 1 + 2x − β √ √ = nx − 1 ≤ φ(x).
2x + 2x0 n
n
Using i=1 (xi − x0 ) = 0, we obtain:
1 n
1 n
1 n
1 n
f (xi ) g(xi ) = 1 − φ(xi ) 1+ ψ(xi )
n i=1 n i=1 2n i=1 2n i=1
1 n
1 n
≤ 1− φ(xi ) 1+ φ(xi ) ≤ 1.
2n i=1 2n i=1
Also solved by D. R. Bridges, G. Crandall, P. P. Dályay (Hungary), K.-W. Lau (China), J. H. Lindsey II, O. P.
Lossers (Netherlands), B. Schmuland (Canada), A. Stenger, R. Stong, J. Sun, L. Zhou, GCHQ Problem Solving
Group (U. K.), Microsoft Research Problems Group, and the proposer.
Solution by Vitaly Stakhovsky, Redwood City, CA. The inequality follows from
1 1 1 3 (a − 1)(b − 1)(c − 1) 3
+ + ≥ + (1)
a+b b+c c+a 2 (a + b)(b + c)(c + a) 2
and x 2 + 1/4 ≥ x with x = 1/(a + b), 1/(b + c), and 1/(c + a). Multiplying by the
positive factor (a + b)(b + c)(c + a), we see that (1) is equivalent to
(b + c)(c + a) + (a + b)(c + a) + (a + b)(b + c)
3 3
≥ (a − 1)(b − 1)(c − 1) + (a + b)(b + c)(c + a). (2)
2 2
Using S1 , S2 , and S3 for the symmetric expressions a + b + c, ab + bc + ca, and abc,
inequality (2) becomes
3 3 3
S12 + S2 ≥ (S3 − S2 + S1 − 1) + (S1 S2 − S3 ) = (S1 − 1)(S2 + 1). (3)
2 2 2
A Productive Inequality
11252 [2006, 847]. Proposed by Ovidiu Bagdasar, Babeş Bolyai University, Cluj-
Napoca, Romania.Let
n n be an integer greater than 2 and let a1 , . . . , an be positive
n
numbers. Let S = i=1 ai . Let bi = S − ai , and let S = i=1 bi . Show that
n n
ai bi
n i=1 ≤ n i=1 .
i=1 (S − ai ) i=1 (S − bi )
268
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
Also solved by P. P. Dályay (Hungary), O. P. Lossers (Netherlands), B. Schmuland (Canada), R. Stong, and
the proposer.
PROBLEMS
11355. Proposed by Jeffrey C. Lagarias, University of Michigan, Ann Arbor, MI. De-
termine for which integers a the Diophantine equation
1 1 1 a
+ + =
x y z x yz
has infinitely many integer solutions (x, y, z) such that gcd(a, x yz) = 1.
11356. Proposed by Michael Poghosyan, Yerevan State University, Yerevan, Armenia.
Prove that for any positive integer n,
n 2
n
24n (n!)4
k
2n = .
k=0 (2k + 1) 2k
(2n)!(2n + 1)!
11361. Proposed by Finbarr Holland, University College Cork, Cork, Ireland. The
Lemoine point of a triangle is the unique point L inside the triangle such that the
distances from L to the sides are proportional to the corresponding side lengths.
Given a circle G and distinct fixed points B, C on G, let K be the locus of the
Lemoine point of ABC as A traverses the circle. Show that K is an ellipse.
SOLUTIONS
k+n−1 k
Solution I by Harris Kwong, SUNY, Fredonia, NY. From (1 − t)−n = ∞ t
n k=0 n−1
and (1 − t ) = r =0 (−1) r t , we obtain
2 n r n 2r
∞
(1 − t 2 )n n
r n k + n − 1 k+2r
(1 + t)n = = (−1) t .
(1 − t)n k=0 r =0
r n−1
Since k < n, no k-person committees can be formed from n couples using at least one
member of each couple. To count these committees using inclusion-exclusion, let Ai
be the set of k-person committees using neither member of couple i. Now i∈S Ai =
2n−2|S|
k
. Hence the sum on the left is the inclusion-exclusion sum to count the empty
set.
Solution III by Nicholas Singer, Annandale, VA. Indeed, rn=0 (−1)r nr P(r ) = 0 for
any polynomial P of degree less than
n. The nth forward
difference of the sequence
(P(0), P(1), . . . ) has first term (−1)n rn=0 (−1)r nr P(r ). Since P is a polynomial of
degree less than n, its nth forward difference is identically zero.
Also solved by 62 other readers and the proposer.
Sometimes an Integer
11213 [2006, 268]. Proposed by Stanley Rabinowitz, Chelmsford, MA. For positive
integers n and m with n odd and greater than 1, let S(n, m) = (n−1)/2
k=1 sec2m ( n+1
kπ
).
(a) Show that if n is one less than a power of 2, then S(n, m) is a positive integer.
(b∗ ) Show that if n does not have the form of part (a), then S(n, m) is not an integer.
366
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
Solution to (a) by NSA Problems Group, Fort Meade, MD. Let n + 1 = 2s with s ≥ 2.
We use induction on s. The base case s = 2 is easy: sec2m (π/4) = 2m . Note that
2
s −1
2s−1
kπ (2 j − 1)π
S(2 s+1
− 1, m) = 2m
sec = S(2 − 1, m) +
s
sec2m
.
k=1
2s+1 j =1
2s+1
By the induction hypothesis, it suffices to show that the last sum is integral.
The Chebyshev polynomial Tr is a polynomial of degree r defined by Tr (x) =
cos(r arccos(x)). In particular, T2s is a polynomial of degree 2s with T2s (x) = 0 when
x = cos((2 j − 1)π/2s+1 ) for 1 ≤ j ≤ 2s . Furthermore, since T2 (x) = 2x 2 − 1 and
T2s (x) = 2 cos2 (2s−1 arccos(x)) − 1 = 2T2s−1 (x)2 − 1,
it follows that T2s (x) ∈ Z[x] and, apart from the constant term, all of its coeffi-
cients are even. Since T2s (0) = cos(2s π/2) = 1 for s ≥ 2, the reciprocal polynomial
s
x 2 T2s (1/x) is a monic integral polynomial that has a zero at each of the 2 S numbers
xk := sec((2 j − 1)π/2s+1 ), 1 ≤ j ≤ 2s . Furthermore, every elementary symmetric
function in 2 S variables evaluates to an even integer at (x1 , . . . , x2S ). Now
2s−1
1 1
s S
(2 j − 1)π 2
(2 j − 1)π 2
sec 2m
= sec2m = x 2m ,
j =1
2s+1 2 j =1 2s+1 2 j =1 j
Solution by Bruce S. Burdick, Roger Williams University, Bristol, RI. In describing the
first round of the process, “less than n” should be “at most n”.
then f (n) = 2i+1 , and f (n)/n > 2/3. Similarly, if 3 · 2i ≤ n ≤ 2i+2 − 1, then f (n) ≥
3 · 2i , and f (n)/n > 3/4. Hence limn→∞ f (n)/n = 2/3.
Solved also by M. R. Avidon, D. Beckwith, P. Corn, D. Cranston, D. Fleischman, J. W. Frommeyer, J.-P. Gri-
vaux (France), C. C. Heckman, E. A. Herman, M. Hildebrand, M. Huibregtse, G. Keselman, J. H. Lindsey
II, O. P. Lossers (The Netherlands), D. Lovit, F. B. Miles, D. Opitz, W. Y. Pong, M. A. Prasad (India),
A. Stadler (Switzerland), R. Staum, J. H. Steelman, R. Stong, M. Tetiva (Romania), L. Zhou, BSI Problem
Solving Group (Germany), Szeged Problem Solving Group “Fejéntaláltuka” (Hungary), GCHQ Problem Solv-
ing Group (U. K.), Manchester Problem Solving Group (U. K.), Microsoft Research Problems Group, and
Northwestern University Problem Solving Group.
11248 [2006, 760]. Proposed by Pál Péter Dályay, Deák Ferenc High School, Szeged,
Hungary. Let n be a positive integer, and let f be a continuous real-valued func-
1
tion on [0, 1] with the property that 0 x k f (x) d x = 1 for 0 ≤ k ≤ n − 1. Prove that
1
0
( f (x))2 d x ≥ n 2 . (The case n = 2, due to Ioan Raşa, appeared on the 55th Roma-
nian Mathematical Olympiad, and the solution was published in Gazeta Matematică
(CIX) 5-6 (2004), page 227.)
368
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
(d k /d x k )[(x 2 − 1)k ]. These polynomials satisfy Pk (1) = 1 and
⎧
1 ⎪
⎨0 if j
= k,
P j (x)Pk (x) d x = 2
−1 ⎪
⎩ if j = k.
2k + 1
√
Let Q k (x) = 2k + 1 Pk (2x − 1). 1 Now {Q 0 , Q 1 , . . . , Q n−1 } is√an orthonormal set
in L 2 [0, 1]. Moreover, f, Q k
= 0 f (x)Q k (x) d x = Q k (1) = 2k + 1, so Bessel’s
inequality yields
1
n−1
n−1
( f (x))2 d x = f 2 ≥ | f, Q k
|2 = (2k + 1) = n 2 .
0 k=0 k=0
Editorial comment. As some contributors noted, the inequality is sharp, and it holds
for all f ∈ L 2 [0, 1]. Some solvers observed that it suffices to construct a polynomial p
1
of degree at most n − 1 satisfying 0 ( p(x))2 d x = 1 and p(1) = n. The desired result
then follows from the Schwarz inequality using p, f
= p(1). Various devices—e.g.,
orthogonal polynomials, Lagrange interpolation, Hilbert matrices—were used in con-
structing such a polynomial.
Also solved by U. Abel (Germany), O. J. L. Alfonso (Colombia), S. Amghibech (Canada), K. F. Andersen
(Canada), A. Bandeira & E. Dias (Portugal), M. Benito & Ó. Benito & E. Fernández (Spain), M. W. Botsko,
P. Bracken, D. R. Bridges, M. A. Carlton, R. Chapman (U. K.), A. Chaudhuri & R. Selukar, P. R. Chernoff,
K. Dale (Norway), Y. Dumont (France), B. Dunn III, J. Fabrykowski & T. Smotzer, S. Foucart, J.-P. Grivaux
(France), J. Groah, E. A. Herman, F. Holland (Ireland), E. J. Ionascu (Romania), M. E. H. Ismail, D. Jespersen,
J. Kalis, G. Keselman, O. Kouba (Syria), A. V. Kumchev, K.-W. Lau (China), J. H. Lindsey II, O. P. Lossers
(The Netherlands), R. Martin (U. K.), W. Matysiak (Poland), T. Nowak (Austria), M. A. Prasad (India), H.
Ricardo, J. Rooin & M. Hassani (Iran), H.-J. Seiffert (Germany), J. G. Simmonds, N. C. Singer, A. Stadler
(Switzerland), V. Stakhovsky, A. Stenger, R. Stong, J. Sun, M. Tetiva (Romania), S. Vagi, Z. Vörös (Hun-
gary), R. Whitley, L. Zhou, BSI Problems Group (Germany), Szeged Problem Solving Group “Fejéntaláltuka”
(Hungary), GCHQ Problem Solving Group (U. K.), Microsoft Research Problems Group, and the proposer.
A Symmetric Inequality
11255 [2006, 848]. Proposed by Slavko Simic, Mathematical Institute SANU, Bel-
grade, Serbia. Let n be a positive integer, x1 , . . . , xn be real numbers,
n and k let
p1 , . . . , pn be real numbers summing to 1. For k ≥ 1, let Sk = i=1 pi xi −
n k
i=1 pi x i . Show that for m ≥ 1,
1
S2m S2m+2 ≥ 1 − S2 .
m(2m + 1) 2m+1
or, equivalently,
t 2 S2m+2 2t S2m+1 S2m
+ + ≥ 0.
(2m + 2)(2m + 1) 2m(2m + 1) 2m(2m − 1)
The expression on the left, considered as a polynomial in t, must therefore have a
nonpositive discriminant:
2
2S2m+1 S2m+2 S2m
−4 · ≤ 0,
2m(2m + 1) (2m + 2)(2m + 1) 2m(2m − 1)
and this is equivalent to the desired inequality.
Editorial comment. The requirement that pi ≥ 0 was given by the proposer but lost
by the editors. A more subtle oversight is the unstated requirement that m should be
an integer. Byron Schmuland provided a counterexample with m = 3/2, x1 = x2 = 1,
x3 = −1, and p1 = p2 = p3 = 1/3, in which case S3 S5 < (5/6)S42 .
Also solved by B. Schmuland (Canada), Microsoft Research Problems Group, and the proposer.
Perimeter by Integral
11256 [2006, 848]. Proposed by Finbarr Holland, University College Cork, Cork, Ire-
land. For complex a, b, and c, let f (x) = max{Re(aei x ), Re(bei x ), Re(cei x )}. Find
2π
0
f (x) d x.
2π
Solution by Michel Bataille, Rouen, France. Let I = 0 f (x) d x. We show that I =
|a − b| + |b − c| + |c − a|; this is the perimeter of the triangle formed by the three
complex numbers.
Let g(x) = min{Re(aei x ), Re(bei x ), Re(cei x )}. Now
2π
2π
g(x) d x = − max{−Re(aei x ), −Re(bei x ), −Re(cei x )} d x
0 0
2π
=− max{Re(aei(x+π) ), Re(bei(x+π) ), Re(cei(x+π) )} d x
0
3π
=− max{Re(aeit ), Re(beit ), Re(ceit )} dt = −I,
π
370
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
Now let a = α + iα and b = β + iβ where α, α , β, and β are real numbers. Then
Re(aei x ) − Re(bei x ) = (α − β) cos x − (α − β ) sin x = A sin(x + φ),
where A = ((α − β)2 + (α − β )2 )1/2 = |a − b|, sin φ = (α − β)/A, and cos φ =
−(α − β )/A. Thus,
2π
2π
[Re(aei x ) − Re(bei x )| d x = |a − b| | sin(x + φ)| d x
0 0
2π
= |a − b| | sin x| d x = 4|a − b|,
0
2π
and Ja,b = 2|a − b|. Similarly, with K a,b = 0 min{Re(aei x ), Re(bei x )} d x, we have
K a,b = −2|a − b|. Finally, for real numbers u, v, and w, we have
1
max{u, v, w} − min{u, v, w} = (max{u, v} + max{v, w} + max{w, u}
2
− min{u, v} − min{v, w} − min{w, u}),
so
2π
1
2I = (Ja,b + Jb,c + Jc,a − K a,b − K b,c − K c,a )
( f (x) − g(x)) d x =
0 2
1
= 2|a − b| + 2|b − c| + 2|c − a| + 2|a − b| + 2|b − c| + 2|c − a|
2
= 2|a − b| + 2|b − c| + 2|c − a|.
This is the stated result.
Also solved by S. Amghibech (Canada), M. R. Avidon, D. R. Bridges, B. S. Burdick, R. Chapman (U. K.),
A. Demis (Greece), J.-P. Grivaux (France), E. A. Herman, F. Hjouj, G. Keselman, J. H. Lindsey II, M. D.
Meyerson, J. G. Simmonds, V. Stakhovsky, R. Stong, R. Tauraso (Italy), E. I. Verriest, J. Vinuesa (Spain),
H. Widmer (Switzerland), J. H. Zacharias, L. Zhou, Szeged Problem Solving Group “Fejéntaláltuka” (Hun-
gary), GCHQ Problem Solving Group (U. K.), Microsoft Research Problems Group, and the proposer.
A Choice Sum
11263 [2006, 940]. Proposed by Gregory Keselman, Oak Park, MI, and formerly of
Lvov Polytechnic Institute, Ukraine. Show that when n is a positive integer and a is
real,
⎧ √ √
⎪ (1+ 1−4a)n+1√ −(1− 1−4a)n+1
n/2 ⎨ , if a < 1/4;
n−k 2 n+1 1−4a
(−1)k a k = (n + 1)/2n , if a = 1/4;
k ⎪
⎩ n/2 √
k=0
a cos(nβ) + sin(nβ)/ 4a − 1 , if a > 1/4.
√
Here, β denotes arcsin 1 − 1/(4a).
Solution by E. Leonard, University of Alberta, Edmonton, AL, Canada. For each non-
negative integer n, define
n/2
n−k n
k k n−k
bn = (−1) a
k k
= (−1) a .
k=0
k k=0
k
1 1√ 1
λ= ± 1 − 4a if a < ,
2 2 4
1 1
λ= if a = ,
2 4
1 1√ 1
λ= ±i 4a − 1 if a > .
2 2 4
(i) For a < 1/4 there are two distinct roots, so the general solution of the recurrence
is
√ n √ n
1 + 1 − 4a 1 − 1 − 4a
bn = A +B .
2 2
We now compute A and B from the initial conditions b0 = b1 = 1 to arrive at
√ n+1 √ n+1
1 + 1 − 4a − 1 − 1 − 4a
bn = √ .
2n+1 1 − 4a
(ii) For a = 1/4 there is a double root, so the general solution of the recurrence is
A Bn
n
+ n. bn =
2 2
Again using the initial conditions to determine A and B, we obtain
n+1
. bn =
2n
(iii) For a > 1/4, there are two distinct complex roots, which may be written in
polar form as
√ √ √ √
a cos β + i sin β = aeiβ , and a cos β − i sin β = ae−iβ ,
√
where β = arcsin 1 − 1/(4a). The general solution of the recurrence is
bn = Aa n/2 einβ + Ba n/2 e−inβ .
Applying the initial conditions to evaluate A and B, and using
√
4a − 1 1
sin β = √ , cos β = √ ,
2 a 2 a
we get
a n/2
bn = a n/2 cos nβ + √ sin nβ.
4a − 1
Also solved by U. Abel (Germany), S. Amghibech (Canada), M. R. Avidon, M. Bataille (France), D. Beckwith,
J. Borwein (Canada), P. Bracken, R. Chapman (U.K.), P. P. Dályay (Hungary), P. De (Ireland), G.C. Greubel,
J.-P. Grivaux (France), H. Kwong, G. Lamb, O. P. Lossers (The Netherlands), J. Minkus, C. R. Pranesachar
(India), T. L. & V. Râdulescu (Romania), H. Roelants (Belgium), J. N. Senadheera, A. Stadler (Switzerland),
R. Stong, M. Tetiva (Romania), H. Widmer (Switzerland), C. Zhou (Canada), BSI Problems Group (Germany),
GCHQ Problem Solving Group (U.K.), Microsoft Research Problems Group, and the proposer.
372
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
PROBLEMS AND SOLUTIONS
Edited by Gerald A. Edgar, Doug Hensley, Douglas B. West
with the collaboration of Paul T. Bateman, Mario Benedicty, Itshak Borosh, Paul
Bracken, Ezra A. Brown, Randall Dougherty, Tamás Erdélyi, Zachary Franco, Chris-
tian Friesen, Ira M. Gessel, Jerrold Grossman, Frederick W. Luttmann, Vania Mas-
cioni, Frank B. Miles, Richard Pfiefer, Cecil C. Rousseau, Leonard Smiley, John
Henry Steelman, Kenneth Stolarsky, Richard Stong, Walter Stromquist, Daniel Ull-
man, Charles Vanden Eynden, and Fuzhen Zhang.
PROBLEMS
11362. Proposed by David Callan, University of Wisconsin, Madison, WI. A bit string
arc diagram is an undirected graph in which the vertices are the positions in a single
string of bits and the edges are called arcs due to the visual representation in which
they are drawn joining positions in the string. To be a good diagram, arcs must occur
only between unequal bits, and each bit may be the left endpoint of at most one arc.
Thus, the first diagram is good but, for two reasons, the second is not.
1 1 0 0 1 1 0 1 1 0 0 1 1 0
There are six good diagrams on two bits, four with no arc and two with a single arc.
How many good diagrams are there on n bits?
11363. Proposed by Oleh Faynshteyn, Leipzig, Germany. Let m a , m b , and m c be the
lengths of the medians of a triangle T . Similarly, let Ia , Ib , Ic , h a , h b , and h c be the
lengths of the angle bisectors and altitudes of T , and let R, r , and S be the circumra-
dius, inradius, and area of T . Show that
Ia Ib Ib Ic Ic Ia
+ + ≥ 3(2R − r ),
Ic Ia Ib
and
m a Ib m b Ic m c Ia √
+ + ≥ 35/4 S.
hc ha hb
11364. Proposed by Pál Péter Dályay, Szeged, Hungary. Let p be a prime greater than
3, and t the integer nearest p/6.
(a) Show that if p = 6t + 1, then
2t−1
1 1
( p − 1)! (−1) j
+ ≡ 0 (mod p).
j =0
3j + 1 3j + 2
11366. Proposed by Nicolae Anghel, University of North Texas, Denton, TX. Let φ :
R → R be a continuously differentiable function such that φ(0) = 0 and φ is strictly
increasing. For a > 0, let Ca denote the space of all continuous functions from [0, a]
a
into R, and for f ∈ Ca , let I ( f ) = x=0 (φ(x) f (x) − xφ( f (x))) d x. Show that I has
a finite supremum on Ca and that there exists an f ∈ Ca at which that supremum is
attained.
√
11367.
Proposed by Andrew Cusumano, Great Neck, NY. Let x = 1 + 2, x2 =
√ 1
1 + 2 1 + 3, and in general, let xn+1 be the number obtained by replacing the
innermost expression
√ (1 + (n + 1)) in the nested square root formula for xn with
1 + (n + 1) 1 + (n + 2). Show that
xn − xn−1
lim = 2.
n→∞ xn+1 − xn
SOLUTIONS
f (i 1 ) f (i 2 ) f (i 3 ) · · · f (i n ) = (−1)m−k mk k n .
i 1 =0 i 2 =0 i 3 =0 i n =0 m=1 k=1
Composite solution by the proposer and the editors. We begin by forming a generating
function from each side. Multiply by (−1)n q n , sum over n ≥ 1, and use 1 for the
constant term. It suffices to show that the two resulting generating functions are the
same. We start with the one on the left.
462
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
We first convert the power series to a continued fraction. With a = (a0 , a1 , . . . ), let
pn (a) be the coefficient of q n in the expression
1 1 a0 q a1 q ∞
= ··· = (−1)n pn (a)q n . (1)
a0 q 1+ 1+ 1+
1+ n=0
1 + a1 q
..
1+ .
We prove by induction on n that pn (a) is the sum of nj =1 ai j over all n-tuples
i 1 , . . . , i n such that i 1 = 0 and 0 ≤ i j ≤ i j −1 + 1 for 2 ≤ j ≤ n. This is precisely our
generating function when ak = f (k) for k ≥ 0.
For n = 0, we have the empty product and coefficient 1. For n ≥ 1, let a =
(a1 , a2 , . . . ). By the induction hypothesis,
∞
∞
(−1)n pn (a)q n 1 + a0 q (−1)k pk (a )q k = 1,
n=0 k=0
n
By examining the coefficient of q on both sides, it follows that
n−1
pn (a) = a0 p (a) pn−1− (a ). (2)
=0
Group the terms in the claimed formula pn (a) according to the maximum index
j such that i j = 0. By the induction hypothesis, the sum of the terms for which
the last position with subscript 0 in the n-tuple is position l + 1 can be written as
p (a)a0 pn−1− (a ). Hence (2) completes the proof of the formula for pn (a).
The sequence from which we form the continued fraction can be expressed as
a2r −2 = r and a2r −1 = 2r for r ≥ 1. Let s = 1/q. Our next aim is
k
1 a0 /s a1 /s 1 ∞
s 1
··· = . (3)
1+ 1+ 1+ 2 k=0 s + k 2
In step 1 we expand (1 + kq)−1 by the geometric series and interchange the order
of summation. In step 2 we invoke k n = S(n, m)k(m) , where S(n, m) is the Stirling
number of the second kind (the number of partitions of n positions into m unordered
blocks) and k(m) is the falling factorial. Both sides count the k-ary n-tuples, on the right
∞ k
−k
grouped by the number of distinct entries. In step 3 we use the identity k=0 m 2 =
k
k
2; this is the instance of ∞ k=0 m x = x /(1 − x)
m m+1
at x = 1/2, where the series
converges. Finally, in step 4 we use the inclusion-exclusion formula for the Stirling
numbers.
No solutions were received.
464
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
let γG be an ordered Abelian group. The support Supp(a) of a formal sum a =
and
γ aγ t with coefficients aγ in K and exponents γ in G is the set {γ ∈ G : aγ
= 0}.
The generalized power series ring K ((G ≤0 )) is the set of formal sums a for which
Supp(a) is a well-ordered subset of the nonpositive elements of G. Addition and mul-
tiplication in K ((G ≤0 )) are defined in the same way they are for ordinary power se-
ries. Show that K ((G ≤0 )) is Noetherian if and only if either G = {0} or G is order-
isomorphic to Z with the usual ordering. (An ordered Abelian group is an Abelian
group G with a total order ≤ such that a ≤ b implies a + c ≤ b + c for all a, b, c
in G.)
Solution by Reid Huntsinger, Lansdale, PA. Let R = K ((G ≤0 )). Assume G
= {0}.
Given a positive element γ0 in G, let I (γ0 ) = {a ∈ R : aγ = 0 if γ > −γ0 }. Always
I (γ0 ) is an additive
subgroup. In fact, I (γ0 ) is an ideal, since a ∈ R and b ∈ I (γ0 )
imply (ab)γ = δ≤0 aδ bγ −δ . The sum is zero unless γ − δ ≤ −γ0 for some δ ≤ 0,
which requires γ ≤ −γ0 . Thus ab ∈ I (γ0 ), making I (γ0 ) an ideal.
On the other hand, if a ∈ R with min Supp(A) = 0 then a is invertible in R. (The
inversion algorithm is essentially the customary algorithm for inverting power series
with real coefficients with nonzero constant coefficient.) As a result, all proper ideals
of R have the form I (γ ) for some positive γ .
If γ0 < γ1 , then I (γ1 ) ⊂ I (γ0 ), and vice-versa. Thus the Noetherian property (no in-
finite ascending chains of ideals) is equivalent to having no infinite descending chains
in G >0 . In other words, R is Noetherian if and only if the positive elements of G are
well-ordered. This latter condition means that the positive elements are all powers of
the smallest positive element. Thus G is order-isomorphic to Z exactly when R is
Noetherian.
Also solved by N. Caro (Brazil), R. Chapman (U. K.), D. Fleischman, Szeged Problem Solving Group “Fején-
találtuka” (Hungary), and the proposers.
Solution for odd B by the GCHQ Problem Solving Group Cheltenham, UK. For odd
B we have n ≡ s(n) mod 2. Therefore we can write (−1)n rather than (−1)s(n) in the
infinite product. Thus the corresponding infinite series
∞ nB + k ∞ 1
(−1)n log = (−1)n+1 log 1 +
n=0 k odd
nB + k + 1 n=0 k odd
nB + k
0<k<B 0<k<B
converges by the classical alternating series test. Hence the product converges to a
finite positive limit. By induction (for example), it follows for each positive integer M
that
2M−1 n B + k (−1)n (2M B)!22M (M!)2
= 2M B
n=0 k odd
nB + k + 1 2 (M B)!(M B)!(2M)!
0<k<B
so
2M
n B + k (−1)n 2M−1
n B + k (−1)n
= 1 + O(1/M) .
n=0 k odd
nB + k + 1 n=0 k odd
nB + k + 1
0<k<B 0<k<B
466
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
Letting A = (2k 3 )−1/2 (M 2 − k M), we have
A2 − M = (2k 3 )−1 (M 4 − 2k M 3 + k 2 M 2 − 2k 3 M)
= (2k 3 )−1 (M − 2k I )(M 3 + k 2 M) = 0,
as desired (here I is the identity matrix). Writing out A in terms of a and b, we obtain
1 t
Solution by Ronald A. Kopas, Clarion University, Clarion, PA. Side a is the sum of
the two segments on either side of the point of tangency with the incircle, so
B C cos(B/2) sin(C/2) + cos(C/2) sin(B/2)
a = r cot + cot =r
2 2 sin(B/2) sin(C/2)
r sin(B/2 + C/2) r sin(π/2 − A/2) r cos(A/2)
= = = ,
sin(B/2) sin(C/2) sin(B/2) sin(C/2) sin(B/2) sin(C/2)
and r = a sin(B/2) sin(C/2)/ cos(A/2).
From the law
of sines, a = 2R sin A =
4R sin(A/2) cos(A/2), so R = a/ 4 sin(A/2) cos(A/2) . Therefore
r A B C
= 4 sin sin sin . (1)
R 2 2 2
Now we compute a lower bound for E = 2a 2 − (b − c)2 . We have
E = 2 a 2 − (b − c)2 + (b − c)2
which has all positive terms and diverges. So assume β > α ≥ 0. Now
n
α + k log k β n
α + k log k n+1
α + k log k
= − ,
k=1
β + (k + 1) log(k + 1) β − α k=1 β + k log k k=1 β + k log k
so we have
N −1
n
α + k log k α β N
α + k log k
= − .
n=1 k=1
β + (k + 1) log(k + 1) β − α β − α k=1 β + k log k
∞
Since k=1 (β − α)/(β + k log k) has all positive terms and diverges, we have
∞ ∞
β −α α + k log k
1− = = 0.
k=1
β + k log k k=1
β + k log k
Also solved by O. Furdui, J.-P. Grivaux (France), O. Kouba (Syria), J. H. Lindsey II, O. P. Lossers (The
Netherlands), A. Stadler (Switzerland), Szeged Problem Solving Group “Fejéntaláltuka” (Hungary), GCHQ
Problem Solving Group (U. K.), Microsoft Research Problems Group, and the proposer.
468
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
PROBLEMS AND SOLUTIONS
Edited by Gerald A. Edgar, Doug Hensley, Douglas B. West
with the collaboration of Paul T. Bateman, Mario Benedicty, Itshak Borosh, Paul
Bracken, Ezra A. Brown, Randall Dougherty, Tamás Erdélyi, Zachary Franco, Chris-
tian Friesen, Ira M. Gessel, Jerrold Grossman, Frederick W. Luttmann, Vania Mas-
cioni, Frank B. Miles, Richard Pfiefer, Cecil C. Rousseau, Leonard Smiley, John
Henry Steelman, Kenneth Stolarsky, Richard Stong, Walter Stromquist, Daniel Ull-
man, Charles Vanden Eynden, and Fuzhen Zhang.
PROBLEMS
11369. Proposed by Donald Knuth, Stanford University, Stanford, CA. Prove that for
all real t, and all α ≥ 2,
α
eαt + e−αt − 2 ≤ et + e−t − 2α .
11370. Proposed by Michael Goldenberg and Mark Kaplan, Baltimore Polytechnic In-
stitute, Baltimore, MD. Let A0 , A1 , and A2 be the vertices of a non-equilateral triangle
T . Let G and H be the centroid and orthocenter of T , respectively. Treating all indices
modulo 3, let Bk be the midpoint of Ak−1 Ak+1 , let Ck be the foot of the altitude from
Ak , and let Dk be the midpoint of Ak H .
The nine-point circle of T is the circle through all Bk , Ck , and Dk . We now introduce
nine more points, each obtained by intersecting a pair of lines. (The intersection is not
claimed to occur between the two points specifying a line.) Let Pk be the intersection
of Bk−1 Ck+1 and Bk+1 Ck−1 , Q k the intersection of Ck−1 Dk+1 and Ck+1 Dk−1 , and Rk
the intersection of Ck−1 Ck+1 and Dk−1 Dk+1 .
Let e be the line through {P0 , P1 , P2 }, and f be the line through {Q 0 , Q 1 , Q 2 }.
(By Pascal’s theorem, these triples of points are collinear.) Let g be the line through
{R0 , R1 , R2 }; by Desargues’ theorem, these points are also collinear.
(a) Show that the line e is the Euler line of T , that is, the line through the circumcenter
and centroid of T .
(b) Show that g coincides with f .
(c) Show that f is perpendicular to e.
(d) Show that the intersection S of e and f is the inverse of H with respect to the
nine-point circle.
11371. Proposed by Ovidiu Furdui, University of Toledo, Toledo, OH. Let A denote
the Glaisher-Kinkelin constant, given by
2 /2−n/2−1/12 n 2 /4
n
A = lim n −n e k k = 1.2824 · · · .
n→∞
k=1
SOLUTIONS
568
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
N 2
Using the standard formula i=1 i = N (N + 1)(2N + 1)/6 and algebraically simpli-
fying, we obtain m 2 = k 2 + (n − 1)k + (n − 1)(2n − 1)/6. Thus (n − 1)(2n − 1)/6
must be an integer, which forces n ≡ ±1 (mod 6). Completing the square on the right
side yields m 2 = (k + (n − 1)/2)2 + (n 2 − 1)/12, so
(b) Given a blocky integer n, taking all possible factorizations of the form (n 2 −
1)/12 = (2α)(2β) with α, β > 0 and setting k = α − β − (n − 1)/2 yields all pos-
sible starting points k for the summation. (Notice that if α = −a and β = −b with
a, b > 0, then k also arises from the positive factorization (n 2 − 1)/12 = (2a)(2b),
because k = α − β − (n − 1)/2 = b − a − (n − 1)/2.)
(c) From the construction above, we conclude that the number of such k is the
number of positive divisors of (n 2 − 1)/48, which is easily computed from its prime
factorization.
Also solved by K. Alpar-Vajk (Italy), M. R. Avidon, D. Beckwith, R. Chapman (U. K.), P. Corn, P. P. Dályay
(Hungary), J. Ferdinands, J. W. Frommeyer, S. M. Gagola, Jr., E. A. Herman, O. P. Lossers (Netherlands),
P. Magli (Italy), L. Pebody, F. Sami, R. Stong, Z. Vörös (Hungary), Con Amore Problem Group (Denmark),
GCHQ Problem Solving Group (U. K.), Microsoft Research Problems Group, NSA Problems Group, and the
proposer.
570
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
where is the sum over the even permutations. Writing
x = sin a1 sin α1 sin(α2 − a2 ) sin(α3 − a3 ),
y = sin a2 sin α2 sin(α3 − a3 ) sin(α1 − a1 ),
z = sin a3 sin α3 sin(α1 − a1 ) sin(α2 − a2 ),
we have t1 = z/y, t2 = x/z, t3 = y/x. So
x yz = T1 T2 T3 sin(α1 − a1 ) sin(α2 − a2 ) sin(α3 − a3 ),
y + z = sin(α1 − a1 ) [sin a2 sin α2 sin(α3 − a3 ) + sin a3 sin α3 sin(α2 − a2 )] .
Expanding and rearranging gives
y + z = sin(α1 − a2 ) [sin a2 sin a3 sin α1 + sin a1 sin α2 sin α3 ] = sin(α1 − a1 ) S(123) .
Using this and similar formulas for z + x and x + y, we get
2T1 T2 T3
[B1 B2 B3 ] = [A1 A2 A3 ],
S
as desired.
Also solved by V. Schindler (Germany) and the proposer.
A Combinatorial Inequality
11269 [2007, 78]. Proposed by Hervé Moulin, Rice University, Houston, TX. Let n be a
positive integer and let a1 , . . . , an benonnegative p real numbers. For integers 1 ≤ k ≤
n, and for p ≥ 1, let A(k, p) = S j ∈S a j , where the outer summation extends
p
n
over all subsets of {1, . . . , n} having exactly k elements. Thus, A(n, p) = j =1 a j
while A(1, p) = nj =1 a jp .
(a) Show that if p ≥ n − 1 or p is a positive integer, then
A(n, p) n−1
(−1)k A(n − k, p)
≤ .
n k=0
k+1
n−1
(−1)k A(n − k, p) 1 n+1 p
≤ A(n, p).
k=0
k+1 n+1 n
(c)∗ Does the conclusion of part (a) hold also for nonintegral p with 1 < p < n − 1?
Solution by the editors. (a) More generally, let f be a function differentiable n times,
such that f (x), . . . , f (n) (x) ≥ 0 on (0, ∞) and f (0) = 0. For any subset S of [n] let
aS = as , bS = (−1)|S|−|T | f (aT ).
s∈S T ⊆S
|
n−|T
(−1)k n − |T | bT
= bT =
T ⊆[n] k=0
k+1 k T ⊆[n]
n + 1 − |T |
1 1 n
≥ bT = f aj .
n T ⊆[n] n j =1
Here in the third equality we let k = n − |S|. For the inequality we have used bT ≥ 0
for nonempty T , and b∅ = 0. For part (a) we take f (x) = x p , which has the desired
nonnegativity property exactly when p ≥ n − 1 or p is a positive integer. For use in
part (b), note that similarly
(−1)n−|S| | (−1)k n − |T |
n−|T
f (a S ) = bT
S⊆[n]
n + 2 − |S| T ⊆[n] k=0
k+2 k
bT
= ≥0
T ⊆[n]
(n + 1 − |T |)(n + 2 − |T |)
The desired result is then the special case f (x) = x p . Let F(a1 , . . . , an ) denote the
left side of this desired inequality and suppose that an > an−1 . Since
∂F (−1)n−|S|
= f (a S ),
∂ak S:k∈S
n + 1 − |S|
we have
∂F ∂F (−1)n−2−|S|
− =− f (an + a S ) − f (an−1 + a S ) ≤ 0.
∂an ∂an−1 S⊆[n−2]
(n − 2) + 2 − |S|
Here for the inequality we have used the result of the previous paragraph for the func-
tion g given by g(x) = f (an + x) − f (an−1 + x). Since the symmetric result holds
for the other variables, it follows that for fixed nj =1 a j , F attains its maximum when
a1 = · · · = an . Thus it suffices to prove the inequality when a1 = · · · = an = a. If we
take a1 = · · · = an = an+1 = a then
n+1−k n + 1 (−1)n−k (n + 1)
n+1 n
b[n+1] = (−1) f (ka) = f ((n + 1)a) − f (ka).
k=0
k k=0
n+1−k
Since b[n+1] ≥ 0,
n
(−1)n−k 1
f (ka) ≤ f ((n + 1)a).
k=0
n+1−k n+1
572
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
This is exactly the desired inequality in the case where a1 = · · · = an = a. Note that if
f (x) = x p with n − 1 < p < n, then f (n+1) (x) < 0 on (0, ∞). Hence b[n+1] < 0 and
the reverse inequality holds. Thus (b) does not hold for n − 1 < p < n.
Editorial comment. No solution of (c)∗ is available.
Parts (a) and (b) were also solved by the proposer.
a + b2 + c2
2
− 8 a 2 b2 c2 = (a 2 + b2 + c2 ) − (a 2 + b2 + c2 )2
R
+ 4(a 2 + b2 + c2 )(a 2 b2 + b2 c2 + c2 a 2 ) − 8a 2 b2 c2
= (a 2 + b2 + c2 − 2a 2 )(a 2 + b2 + c2 − 2b2 )(a 2 + b2 + c2 − 2c2 ),
which is positive since ABC is acute.
−→ −→
Now let O V = O H − λ(a + b + c), where λ is a parameter. This is a parametric
description of the Euler line H O, since λ = 0 corresponds to V = H and λ = 1 cor-
−→ −→ −→
responds to V = O. Since AV = O V − O A = −λa + (1 − λ)b + (1 − λ)c, we see
by Lemma 1 that
|AV |2 = R 2 (−λ + 1 − λ + 1 − λ)2 − a 2 (1 − λ)2 + b2 λ(1 − λ) + c2 λ(1 − λ)
= R 2 (2 − 3λ)2 + λ(1 − λ)(a 2 + b2 + c2 ) + (λ − 1)a 2 .
Therefore, if we can take λ satisfying the equation
R 2 (2 − 3λ)2 + λ(1 − λ)(a 2 + b2 + c2 ) = 0, (2)
then we have |AV |2 = (λ − 1)a 2 . Similarly, for the same λ we have |BV |2 =
(λ − 1)b2 and |C V |2 = (λ − 1)c2 , so we obtain the relation (1).
Let p(λ) be the quadratic expression on the left in (2). Then
p(λ) = 9R 2 − (a 2 + b2 + c2 ) λ2 − 12R 2 − (a 2 + b2 + c2 ) λ + 4R 2 .
574
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
PROBLEMS AND SOLUTIONS
Edited by Gerald A. Edgar, Doug Hensley, Douglas B. West
with the collaboration of Paul T. Bateman, Mario Benedicty, Itshak Borosh, Paul
Bracken, Ezra A. Brown, Randall Dougherty, Tamás Erdélyi, Zachary Franco, Chris-
tian Friesen, Ira M. Gessel, Jerrold Grossman, Frederick W. Luttmann, Vania Mas-
cioni, Frank B. Miles, Richard Pfiefer, Cecil C. Rousseau, Leonard Smiley, John
Henry Steelman, Kenneth Stolarsky, Richard Stong, Walter Stromquist, Daniel Ull-
man, Charles Vanden Eynden, and Fuzhen Zhang.
PROBLEMS
11376. Proposed by Bogdan M. Baishanski, The Ohio State University, Columbus OH.
Given a real number a and a positive integer n, let
1
Sn (a) = √ .
an<k≤(a+1)n kn − an 2
664
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
11380. Proposed by Hugh Montgomery, University of Michigan, Ann Arbor, MI, and
Harold
x S.Shapiro, Royal Institute of Technology, Stockholm, Sweden. For x ∈ R, let
k−1
= 1
j =0 (x − j). For k ≥ 1, let ak be the numerator and qk the denominator of
k k!
the rational number −1/3 k
expressed as a reduced fraction with qk > 0.
(a) Show that qk is a power of 3.
(b) Show that ak is odd if and only if k is a sum of distinct powers of 4.
11381. Proposed by Jesús Guillera, Zaragoza, Spain, and Jonathan Sondow, New York,
NY. Show that if x is a positive real number, then
n 1/n
∞ n
(−1)k+1
e =
x
(kx + 1) k .
n=1 k=0
(Two rationals are congruent modulo d if their difference can be expressed as a reduced
fraction of the form da/b with b relatively prime to a and d.)
SOLUTIONS
A Special Prime
11235 [2006, 568]. Proposed by Lenny Jones and Rachel Keller, Shippensburg Uni-
versity, Shippensburg, PA. Find all primes p such that 128 p + 1 is composite and a
divisor of 264 p − 1.
Solution by Toni Ernvall and Anne-Maria Ernvall-Hytönen, University of Turku, Fin-
land. For a fixed prime p, let d be the smallest positive integer such that 2d ≡ 1
(mod 128 p + 1). Suppose that 128 p + 1 is composite and is a divisor of 264 p − 1.
Note that d | 64 p and d | φ(128 p + 1), where φ is Euler’s function. We divide the
problem into cases according to whether p divides d.
If p | d, then p | φ(128 p + 1). Since p does not divide 128 p + 1, there exists a
prime q such that q ≡ 1 (mod p) and q | 128 p + 1. Write q = r p + 1, so that
128 p + 1 = (r p + 1)(sp + 1) = r sp 2 + r p + sp + 1
for some positive integer s, and r sp + r + s = 128. If p = 2, then there are no so-
lutions, since 257 is prime. Hence p is odd, and r ≡ s ≡ 0 (mod 2). Let r = 2k
and s = 2. Now 2kp + k + = 64, and kp ≤ 31. We may assume that k ≤ . By
straightforward calculations we notice that the only positive integer solution is p = 3,
k = 1, and = 9, yielding 128 · 3 + 1 = 385 = 5 · 7 · 11. However, 11 does not divide
264·3 − 1, so this solution does not satisfy the original conditions.
Assume now that p does not divide d. Now d | 64, and 128 p + 1 | 264 − 1. We
compute
264 − 1 = (232 + 1)(216 + 1)(28 + 1)(24 + 1)(22 + 1)(2 + 1)
= 641 · 6700417 · 65537 · 257 · 17 · 5 · 3.
Checking through the possibilities gives 128 p + 1 = 65537 · 6700417. Thus p =
3430665851 is the only solution.
Also solved by M. Avidon, R. Chapman (U. K.), P. Corn, S. M. Gagola Jr., O. P. Lossers ((Netherlands)),
A. Nakhash, A. Stadler (Switzerland), R. Stong, M. Tetiva (Romania), D. B. Tyler, BSI Problems Group
(Germany), Microsoft Research Problems Group, NSA Problems Group, and the proposers.
666
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
of singletons occurring in all partitions of n − 1, where a singleton in a partition is a
part occurring once. (For example, partitions of 5 yield four 2s: one from (3, 2), two
from (2, 2, 1) and one from (2, 1, 1, 1); partitions of 4 yield four singletons: one from
(4), two from (3, 1) and one from (2, 1, 1).)
Solution I by John H. Smith, Needham, MA. Let p(n) denote the number of partitions
of n, with p(0) = 1 and p(n) = 0 for n < 0. The number of partitions of n with at
least i occurrences of j is p(n − i j). Hence the total number of 2s in all partitions of
∞
n is i=1 p(n − 2i), since a partition with exactly k occurrences of 2 is counted once
in each of the first k terms.
On the other hand, a partition of n with a singleton j is formed by appending j to
a partition of n − j that has no j. Since p(n − 2 j) partitions of n − j contain at least
one j, the number of partitions of n with a singleton j is p(n − j) − p(n − 2 j). Thus
the total number of singletons in partitions of n − 1 is ∞ j =1 ( p(n − 1 − j) − p(n −
1 − 2 j)). This sum equals
( p(n − 2) − p(n − 3)) + ( p(n − 3) − p(n − 5)) + ( p(n − 4) − p(n − 7)) + · · · .
Terms p(n − r ) occur
∞positively when r is even and with both signs when r is odd.
Cancellation yields i=1 p(n − 2i), which is the formula obtained for the total number
of 2s.
Solution II by Albert Stadler, Dübendorf, Switzerland. Euler obtained the generating
function for partitions:
∞
∞
1
P(x) = p(n)x n = .
n=1 n=1
(1 − x n )
On the other hand, the number of partitions of n − 1 containing the singleton k is the
coefficient of x n−1 in x k (1 − x k )P(x). Hence the total number of singletons over all
partitions of n − 1 is the coefficient of x n in x ∞ k=1 x (1 − x )P(x). Note that
k k
∞
x x2 x2
x x k (1 − x k )P(x) = x − P(x) = P(x).
k=1
1−x 1−x 2 (1 − x 2 )
Solution by Reiner Martin, Teddington, U.K. Since n − 1 < (nφ − 1)/φ ≤ nφ /φ <
n, we have nφ /φ = n − 1. With φ = 1 + 1/φ, this implies
Thus n = (n + 1)φ − nφ + 1, and so (a) follows since 1 < φ < 2.
Note that nφ
φ = nφ + nφ
/φ
≤ nφ
+ n = n(φ
+ 1)
=
nφ2 . Thus
nφ ≤ nφ 2
2
2
/φ ≤ nφ, and so nφ 2
/φ = nφ. Thus nφ 2
φ = nφ 2
+
nφ /φ = nφ + nφ, which
is claim (c).
Moreover, it follows that nφ 2 φ = 2 nφ + n. Applying this to n + 1 and to n
and invoking n = (n + 1)φ − nφ + 1 yields (b).
Finally, using nφ φ = nφ + n − 1, (d) follows from
2
nφ φ = 2 nφ + n = nφ + nφ φ + 1
= nφ (1 + φ) + 1 = nφ φ 2 + 1.
Also solved by M. R. Avidon, D. Beckwith, N. Caro & O. López (Brazil), R. Chapman (U. K.), J. Christopher,
P. P. Dályay (Hungary), M. Goldenberg & M. Kaplan, G. Keselman, R. Lampe, J. H. Lindsey II, O. P. Lossers
((Netherlands)), A. Nijenhuis, S. Northshield, J. Rebholz, K. Schilling, A. Stadler (Switzerland), A. Stenger,
R. Stong, BSI Problems Group (Germany), Con Amore Problem Group (Denmark), GCHQ Problem Solving
Group (U. K.), Microsoft Research Problems Group, and the proposer.
668
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
1 to the formula, we require at least one row. We prove that
n + 1
c(n) = μ 2d−1 ,
d|(n+1)
d
Solution I by David Beckwith, Sag Harbor, NY. Let I denote the required integral.
Introduce the change of variables x = (z + t)/2, y = (z − t)/2 which gives
∞ z 2
1 t log(z/t)
I =2 dt dz.
z=0 sinh z t=0 z2 − t 2
In the inner integral, set t = zw to obtain the product of two known integrals,
∞ 1
z dz 1 π2 π2 π 2 (π 2 − 8)
I =2 1− log w dw = 2 − 1 = .
0 sinh z 0 1 − w2 4 8 16
670
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
Let Iu denote the inner integral in the previous formula for I . Substituting t = e−(u+1)y
gives
1 1 ∞
−2 ln t −2
Iu = dt = t 2k ln t dt
(1 + u)2 0 t 2 − 1 (1 + u)2 t=0 k=0
2 ∞
1 π2
= = .
(1 + u)2 k=0 (2k + 1)2 4(1 + u)2
So I takes the form
∞
π2 (u − 1)2 log((u + 1)/(u − 1))
I = du.
4 1 u(1 + u)2
Finally, substitute s = (u − 1)/(u + 1):
π 2 1 s 2 ln s π2 1 ln s π 2 (π 2 − 8)
I =− ds = − − ln s ds = .
2 0 1 − s2 2 0 1 − s2 16
A Matrix Inequality
11296 [2007, 252]. Proposed by Said Amghibech, Quebec, Canada. Let A be a real
symmetric positive definite matrix, with entries ai, j , 1 ≤ i, j ≤ n. Let Ak be the k × k
matrix in the upper left corner of A. Show that
n
1 n
(det Ak )1/k ≤ 1 + Tr(A)
k=1
n
where Tr(A) denotes the trace of A.
Solution by Joanna Chachulska and Wojciech Matysiak, Polytechnica Warszawska,
Warsaw, Poland. Hadamard’s inequality says det Ak ≤ a1,1 a2,2 · · · ak,k . Carleman’s
inequality says
n
n
1/k 1 n
x1 x2 · · · xk ≤ 1+ xk
k=1
n k=1
PROBLEMS
11383. Proposed by Michael Nyblom, RMIT University, Melbourne, Australia. Show
that
√ √
∞
1 + n 2 + 2n n 2 + 4n + 3 π
−1
cos = .
n=1
(n + 1)(n + 2) 3
n=1
pn
converges.
11385. Proposed by José Luis Dı́az-Barrero, Universidad Politécnica de Cataluña,
Barcelona, Spain. Let α0 , α1 , and α2 be the radian measures of the angles of an acute
triangle, and for i ≥ 3 let αi = αi−3 . Show that
2
αi2 1/4 √
3 + 2 tan2 αi ≥ 3 3.
i=0
αi+1 αi+2
11386. Proposed by Greg Markowsky, Somerville, MA. Consider a triangle ABC. Let
O be the circumcircle of ABC, r the radius of the incircle, and s the semiperimeter.
Let arc(BC) be the arc of O opposite A, and define arc(C A) and arc(AB) similarly.
Let O A be the circle tangent to AB and AC and internally tangent to O along arc(BC),
and let R A be its radius. Define O B , OC , R B , and RC similarly. Show that
1 1 1 s2
+ + = .
a RA b RB c RC rabc
SOLUTIONS
Solution by E. Omey, EHSAL, Brussels, Belgium. The answer is (π 2 /16) log 2. The
integral to be evaluated is
π/2 π/2
I = f (sin θ cos φ) sin θ dθ dφ,
φ=0 θ=0
1
log(1 + x) π log 2
dx = ,
0 1+x 2 8
758
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
appears in various tables and its evaluation was given as Problem A5 on the Sixty-Sixth
William Lowell Putnam Competition (see this M ONTHLY 113 (2006) 733-743).
Also solved by R. Bagby, D. H. Bailey & J. M. Borwein (U. S. A., Canada), D. Beckwith, R. Chapman (U. K.),
H. Chen, K. Dale (Norway), B. E. Davis, G. de Marco (Italy), A. Fok (Hong Kong), O. Furdui, M. L. Glasser,
J. Grivaux (France), E. A. Herman, O. Kouba (Syria), G. Lamb, K. McInturff, P. Perfetti (Italy), O. G,
Ruehr, H.-J. Seiffert (Germany), J. G. Simmonds, A. Stadler (Switzerland), V. Stakhovsky, R. Stong, J. Sun,
R. Tauraso (Italy), GCHQ Problem Solving Group (U. K.), Microsoft Research Problems Group, University
of Sharjah Problem Solving Group (United Arab Emirates), and the proposer.
Entire Limit
11278 [2007, 259]. Proposed by Slavko Simic, Mathematical Institute SANU, Bel-
grade, Serbia. Let f be a nonconstant entire function with nonnegative Taylor series
coefficients. Prove that limr →∞ f (r )/r f (r ) exists and is rational.
Solution by John H. Lindsey II, Cambridge, MA. First consider the case that f =
a0 + · · · + an x n , a polynomial with an > 0. We have
f (r ) a0 + · · · + an r n a0 /r n + · · · + an 1
lim = lim = lim = .
r →∞ r f (r )
r →∞ a1 r + · · · + an nr n r →∞ a1 /r + · · · + an n
n n
∞
Now consider the case in which f = k=0 ak x k with all coefficients nonnegative, and
not eventually zero. For n ∈ N with an > 0,
n−1 ∞
f (r ) k=0 ak r
k
k=n ak r
k
lim sup ≤ lim sup + lim sup ∞
r →∞ r f (r ) an nr n k
r →∞ r →∞ k=n ak kr
∞
ak r k 1
≤ 0 + lim sup ∞k=n k
= .
r →∞ k=n ak nr n
Sliding Beads
11279 [2007, 259]. Proposed by Vitaly Stakhovsky, Redwood City, CA. Two test-mass
beads are sliding along a vertical circular track under (Newtonian) constant gravity,
without friction. The first bead M1 starts from the highest point on the circle, at some
nonzero velocity. Some time later the second bead M2 starts from the same position at
the top of the circle and with the same initial velocity as M1 had. Prove that there is a
circle to which the line through the current positions of M1 and M2 is always tangent,
and find the center and radius of that circle in terms of the original circle.
Solution by the proposer. Assume that the original circle has unit radius and center
at the origin O. Let OY point in the upward direction. Describe the position of bead
i by its current location Pi and the angle ϕi =
Pi OY . Assume both particles are
moving with ϕi increasing. Let vi = ϕ̇i be the ith bead’s velocity. The equation of
760
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
If n = 1 the answer is f (0) whatever properties f has. For general n, we show that
if f is a positive nondecreasing function on the real line (not necessarily concave or
differentiable, or even continuous), then
n
xk /S
sup f (S − xk ) = lim f (y). (1)
x y→+∞
k=1
We claim first that the limit exists. Indeed, if f is bounded above, its least upper bound
is its limit; if
f is not bounded above, the limit is +∞. Let this limit be ϕ. Next, if ϕ
is finite, then nk=1 ( f (S − xk ))xk /S ≤ nk=1 ϕ xk /S = ϕ. Finally, if x1 = · · · = xn = z,
then
n
xk /S
1/n
f (S − xk ) = f ((n − 1)z) = f ((n − 1)z) → ϕ
k=1
as z → ∞. This completes the proof of (1). The maximum is attained if and only if
the limit is.
There are of course functions on the nonnegative half of the real line with the full
range of properties given in the problem statement. The result given here applies to
them.
Editorial comment. It seems that in the intended problem f is defined on the positive
real line, and n and S are given values.
Then maximize over all lists x satisfying the
conditions. Answer: f (1 − 1/n)S .
Also solved by R. A. Agnew, K. F. Andersen (Canada), J.-P. Grivaux (France), E. A. Herman, J. H. Lindsey II,
V. Stakhovsky, R. Stong, M. Tetiva (Romania), and Microsoft Research Problems Group.
A Triangular Result
11285 [2007, 358]. Proposed by Yakub Aliyev, Qafqaz University and Baku State Uni-
versity, Baku, Azerbaijan. Let six points be chosen in cyclic order on the sides of
triangle ABC: A1 and A2 on BC, B1 and B2 on C A, and C1 and C2 on AB. Let K
denote the intersection of A1 B2 and C1 A2 , L the intersection of B1 C2 and A1 B2 , and
M the intersection of C1 A2 and B1 C2 . Let T , U , and V be the intersections of A1 B2
and B1 A2 , B1 C2 and B2 C1 , and C1 A2 and C2 A1 , respectively. Prove that lines AK ,
B L, and C M are concurrent if and only if points T , U , and V are collinear.
Solution I (one direction) by Apostolis Denis, Varvakeio High School, Athens, Greece.
We prove the first direction of the implication. Suppose AK , B L, C M are concur-
rent. Write u.v for the intersection of lines u and v (extended if necessary). Let X =
A1 C2 .B1 A2 , Y = B1 A2 .C1 B2 , and Z = C1 B2 .A1 C2 . Given triangles ABC and abc,
Desargues’ Theorem with its dual states that Aa, Bb, Cc are concurrent if and only
if AB.ab, BC.bc, C A.ca are collinear. We consider parallel lines to intersect at infinity
and allow for collinearity and concurrency at infinity (see Editorial comment below).
Application of Desargues’ Theorem to the following three pairs of triangles gives the
claimed implication:
(i) [ABC and K L M]: AK , B L, C M are concurrent if and only if AB.K L,
BC.L M, C A.M K are collinear. Since C1 lies on AB, and B2 lies on K L, we have
AB.K L = C1 B.B2 L. Similarly, BC.L M = B A2 .B1 L, and C A.M K = B1 B2 .C1 A2 .
(ii) [C1 B A2 and B2 L B1 ]: C1 B.B2 L, B A2 .L B1 , and A2 C1 .B1 B2 are collinear if and
only if C1 B2 , B L, A2 B1 are concurrent. Note: Since C1 B2 .A2 B1 = Y , this concurrency
is equivalent to B L = LY . Similarly, AK = K X and C M = M Z so K X , LY , M Z
are concurrent.
Continuous Blackjack
11287 [2007, 359]. Proposed by Stephen J. Herschkorn, Highland Park, NJ. Players
1 through n play “continuous blackjack.” At his turn, Player k considers a random
number X k drawn from the uniform distribution on [0, 1]. He may either accept X k as
his score or draw a second number Yk from the same distribution, in which case his
score is X k + Yk if X k + Yk < 1 and 0 otherwise. The highest score wins. Give a rule
for when player k should draw a second number, in terms of k, n, the result of X k , and
the highest score attained so far.
Solution by BSI Problems Group, Bonn, Germany, and the editors. We renumber the
players P0 , P1 , . . . , Pn−1 in reverse order, so that Pk has k players yet to draw after
his turn. It turns out that the optimal strategy for Pk is the same whether the remaining
players play as a team or selfishly.
Let Z k be Pk ’s score; that is, Z k = X k if Pk accepts his first number, Z k = X k + Yk
if he draws a second number Yk and X k + Yk < 1, and Z k = 0 if he draws a second
number and X k + Yk ≥ 1. For k = 0, . . . , n − 1, let Mk = max j ≥k Z j , and let Mn = 0.
We assume selfish play. That is, Pk should take a second number if and only if
doing so improves his probability of winning, assuming Pk−1 through P0 do the same.
We ignore ties, which have zero probability. Note that Pk ’s situation depends only on
the number k of players yet to draw and on Mk+1 . How Mk+1 was produced, and how
many players came earlier to produce it, are irrelevant.
Let f k (z) denote the probability that if players P j for j ≥ k have already played and
the best score so far is Mk = z, then for all j < k, Z j < z—that is, all later players
achieve scores lower than z, so z is the winning score. We will prove that for all k:
1. f k is a continuous, nondecreasing function on [0, 1] and f k (z) > 0 for z > 0.
1
2. There is a unique number gk ∈ [0, 1) such that f k (gk ) = gk f k (t) dt, and Pk
should take a second number if and only if X k < max(Mk+1 , gk ).
We first observe that statement 2 follows from statement 1. To see why, suppose
1
statement 1 holds. Clearly z f k (t) dt is a strictly decreasing continuous function of z,
762
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
1 1
f k (t) dt ≥ f k (0), and 1 f k (t) dt = 0 < f k (1), so by the intermediate value theo-
0 1
rem there is a unique number gk ∈ [0, 1) such that f k (gk ) = gk f k (t) dt. Furthermore,
1 1
if z < gk then z f k (t) dt > f k (z), and if z > gk then z f k (t) dt < f k (z). To deter-
mine how player Pk should play, suppose that all players P j for j > k have already
played. Clearly if X k < Mk+1 then player Pk must take a second number to have any
hope of winning. If X k > Mk+1 , then player Pk ’s probability of winning is f k (X k ) if
1
he doesn’t take a second number, and X k f k (t) dt if he does. The latter probability is
larger than the former if and only if X k < gk , so Pk should take a second number if
and only if X k < max(Mk+1 , gk ).
To prove statement 1, we proceed by induction on k. In the case k = 0, if all of the
players have played and M0 = z, then z is definitely the winning score, so f 0 (z) = 1.
Now assume statement 1 (and therefore also statement 2) holds for k. Suppose that
players P j for j ≥ k + 1 have already played and Mk+1 = z. There are two ways for
player Pk to end up with a score z Z k < z: either X k + Yk < z, or X k + Yk > 1. The
probability of the first event is 0 (z − x) d x = z 2 /2, and the probability of the second
max(z,gk )
is 0 x d x = (max(z, gk ))2 /2. Thus, the probability that Z k < z is:
1
(g 2 + z 2 ), if z ≤ gk ;
h k (z) = 22 k
z , if z > gk .
If Z k < z then Mk = z, and therefore the probability that Z j < z for all j < k is
f k (z). Thus, we have f k+1 (z) = h k (z) f k (z), and combining our formula for h k (z) with
statement 1 for k, we see that statement 1 holds for k + 1 as well.
Finally, we wish to determine the numbers gk . We first observe that for all k, gk+1 >
gk . To see why, suppose that gk+1 ≤ gk . Then, on putting l = k + 1, we have
1
fl (gl ) = h k (gl ) f k (gl ) ≤ h k (gl ) f k (gk ) = h k (gl ) f k (t) dt
gk
1 1 1
≤ h k (gl ) f k (t) dt < h k (t) f k (t) dt = fl (t) dt = fl (gl ),
gl gl gl
In summary, the best strategy is for Pk to take a second number if and only if
X k < max(Mk+1 , gk ), where gk is the unique nonnegative solution to z 2k = (1 −
z 2k+1 )/(2k + 1).
Editorial comment. If the remaining players are not selfish, but cooperate to try to stop
Pk , this is still Pk ’s best strategy. For j < k, player P j now accepts any score X j that
is better than M j . The result is that h j (z) = z 2 and f j (z) = z 2 j for j < k. Thus Pk
should still take a second number if and only if X k < max(Mk+1 , gk ).
Also solved by Z. Ahmed (India), S. Amghibech (Canada), M. Bataille (France), R. Chapman (U.K.), P. P.
Dályay (Hungary), P. De (Ireland), A. Demis (Greece), A. Fok (China), W. Fosheng (China), V. V. Garcı́a
(Spain), M. Goldenberg & M. Kaplan, O. Kouba (Syria), H. Kwong, K.-W. Lau (China), O. P. Lossers (Nether-
lands), Y. Mikata, E. Mouroukos (Greece), R. D. Nelson (U.K.), J. H. Nieto (Venezuela), P. E. Nuesch (Switzer-
land), C. R. Pranesachar (India), J. Rooin & A. Alikhani (Iran), V. Schindler (Germany), L. Sega, H.-J. Seiffert
(Germany), M. Shattuck, R. A. Simon (Chile), A. Stadler (Switzerland), D. Vacaru (Romania), M. Vowe
(Switzerland), L. Zhou, GCHQ Problem Solving Group (U.K.), Microsoft Research Problems Group, North-
western University Math Problem Solving Group, and the proposer.
764
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
PROBLEMS AND SOLUTIONS
Edited by Gerald A. Edgar, Doug Hensley, Douglas B. West
with the collaboration of Paul T. Bateman, Mario Benedicty, Itshak Borosh, Paul
Bracken, Ezra A. Brown, Randall Dougherty, Tamás Erdélyi, Zachary Franco, Chris-
tian Friesen, Ira M. Gessel, Jerrold Grossman, Frederick W. Luttmann, Vania Mas-
cioni, Frank B. Miles, Richard Pfiefer, Cecil C. Rousseau, Leonard Smiley, Kenneth
Stolarsky, Richard Stong, Walter Stromquist, Daniel Ullman, Charles Vanden Eynden,
and Fuzhen Zhang.
PROBLEMS
11390. Proposed by Jeffrey C. Lagarias, University of Michigan, Ann Arbor, MI. Let
G be the undirected graph on the vertex set V of all pairs (a, b) of relatively prime
integers, with edges linking (a, b) to (a + kab, b) and (a, b + kab) for all integers k.
(a) Show that for all (a, b) in V , there is a path joining (a, b) and (1, 1).
(b)∗ Call an edge linking (a, b) to (a + kab, b) or (a, b + kab) positive if k > 0, and
negative if k < 0. Let the reversal number of a path from (1, 1) to (a, b) be one more
than the number of sign changes along the path, and let the reversal value of (a, b)
be the minimal reversal number over all paths from (1, 1) to (a, b). Are there pairs of
arbitrarily high reversal value?
11391. Proposed by Marian Tetiva, National College “Gheorghe Roşca Codreanu”,
Bı̂rlad, Romania. Let p be a positive prime and s a positive integer. Let n and k be
integers such that n ≥ k ≥ p s − p s−1 , and let x1 , . . . , xn be integers. For 1 ≤ j ≤ n,
let m j be the number of expressions of the form xi1 + · · · + xi j with 1 ≤ i 1 < · · · <
i j ≤ n that evaluate to 0 modulo p, and let n j denote the number of such expressions
that do not. (Set m 0 = 1 and n 0 = 0). Apart from the cases (s, k) = (1, p − 1) and
s = p = k = 2, show that
j n−k + j
k
(−1) m k− j ≡ 0 (mod p s ),
j =0
j
SOLUTIONS
Generalizing (1 − 1)n = 0
11230 [2006, 567]. Proposed by Gregory Keselman, Oak Park, MI, formerly of Lvov
Polytechnic Institute, Ukraine. Let n be a positive integer, let [n] = {0, 1, . . . , n − 1},
and for a subset P of [n] let s(z, P) = j ∈P z j . With the usual understanding that
00 = 1, show that
(−1)|P| s k (z, P) = 0 (k < n),
P⊆[n]
(−1)|P| s n (z, P) = (−1)n (n!)z n(n−1)/2 ,
P⊆[n]
(n + 1)!(z n − 1)z n(n−1)/2
(−1)|P| s n+1 (z, P) = (−1)n .
P⊆[n]
2(z − 1)
856
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
After canceling equal terms and interchanging sums,
k−1
k−1
k n(k−i) k n(k−i)
S[n + 1, k] = − z (−1)|P| s i (z, P) = − z S[n, i].
i=0
i P⊆[n] i=0
i
n−1
m ∞
yk k
1 − e yz = (−1)|P| e ys(z,P) = (−1)|P| s (z, P)
m=0 P⊆[n] P⊆[n] k=0
k!
∞
yk
= (−1)|P| s k (z, P)
k=0
k! P⊆[n]
and also
n−1
n−1
m 1
1 − e yz = (−1)n yz m + y 2 z 2m + O(y 3 )
m=0 m=0
2
1
n−1
= (−1)n y n z n(n−1)/2 + (−1)n y n+1 z n(n−1)/2 z m + O(y n+2 ),
2 m=0
k+1
2n−k F1+2xi ,
k=0 x∈S[k,n] i=1
858
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
We show first that the given expression equals an . Call each 4-by-2 rectangle a tooth. If
the top part of a tooth is not tiled with two horizontal dominos or two vertical dominos,
then the tooth is tiled like the second tooth shown above. Let k be the number of teeth
tiled in this way. For each of the n − k remaining teeth, the top part can be tiled in
two ways. This leaves regions that are 2-by-2xi rectangles, where x0 , . . . , xk is an
element of S[k, n]. It is well known (by induction using the Fibonacci recurrence) that
the number of domino tilings of a 2-by-m rectangle is F1+m . Hence we have proved
that the number of tilings equals the given sum.
Next we obtain a recurrence for the sequence an by considering how the last tooth
can be tiled. Consider whether the domino covering the lower right corner is horizontal
or vertical. If it is horizontal, then no domino crosses from the last tooth to the one
before it, and there are three ways to tile the rest of the last tooth. Hence there are
3an−1 tilings of this type.
If the lower right domino is vertical, then there are two ways to tile the top part of the
tooth, and what remains is a 4-by-2(n − 1) region with the two lower rows extended
by one unit. Let bn−1 be the number of ways to tile this region. With two ways to tile
the upper half of the last tooth, we have an = 3an−1 + 2bn−1 .
For bn−1 , we also consider whether the domino covering the lower right corner is
horizontal or vertical. If vertical, then there are an−1 ways to complete the tiling. If
horizontal, then there is another horizontal tile above it, two ways to tile the top half
of tooth n − 1, and bn−2 ways to complete the rest. Hence bn−1 = an−1 + 2bn−2 .
Using the first recurrence to substitute into the second and eliminate the auxiliary
sequence yields an = 7an−1 − 6an−2 . With a0 = 1 and a1 = 5, the solution is an =
(1 + 4 · 6n )/5.
Also solved by S. Amghibech (Canada), D. Beckwith, R. Chapman (U. K.), P. P. Dályay (Hungary), J. W. From-
meyer, C. C. Heckman, G. Keselman, K. McInturff, N. C. Singer, A. Stadler (Switzerland), A. Stenger,
R. Stong, BSI Problems Group (Germany), GCHQ Problem Solving Group (U. K.), Microsoft Research Prob-
lems Group, and the proposer.
k+1
, summing P
i= j −1 i (n − 1) yields
n
2n − 2 − i n
2n − 2 − i 2n − j 2n − j
P j (n) = − = − .
i= j −1
n−2 i= j −1
n n−1 n+1
n+1
Now evaluate j =1 P j (n) to find P(n).
Solution II by Li Zhou, Polk Community College, Winter Haven, FL. The answer is
1 2n+2
the Catalan number Cn+1 , which equals n+2 n+1
. Identify each member of genera-
tion n with the list (a0 , . . . , an ) of labels on the path to it from the root. For n ≥ 0,
860
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
Editorial comment. Equality holds if and only if all aσ are equal, which requires that
ai j = αi β j for all i and j, where all αi and β j are positive numbers.
Also solved by S. Amghibech (Canada), J. C. Binz (Switzerland), P. Budney, R. Chapman (U. K.), K. Dale
(Norway), P. P. Dályay (Hungary), S. M. Gagola Jr., J.-P. Grivaux (France), E. A. Herman, S. Hitotumatu
(Japan), D. Karagulyan (Armenia), P. T. Krasopoulos (Greece), O. López & N. Caro (Brazil), O. P. Lossers
(Netherlands), R. Martin (U. K.), J. Rooin & M. A. Maleki (Iran), K. Schilling, V. Schindler (Germany),
A. Stadler (Switzerland), A. Stenger, R. Stong, R. Tauraso (Italy), M. Tetiva (Romania), J. Vinuesa (Spain), BSI
Problems Group (Germany), GCHQ Problem Solving Group (U. K.), Microsoft Research Problems Group,
Missouri State University Problem Solving Group, NSA Problems Group, and the proposer.
f (n) = 2j k,
j =1 S k∈S
where the sum is over all j-element subsets S of the set {1, . . . , n − 1}. Show that
4(2n − 1)! + ( f (n))2 is never the square of an integer.
Solution by Marian Tetiva, Bı̂rlad, Romania. By expanding n−1k=1 (1 + ck x) and setting
n−1
ck = k and x = 2, we obtain f (n) = k=1 (2k + 1) − (n − 1)!2n−1 − 1. Letting a =
n−1
n−1
k=1 (2k + 1) and b = k=1 2k, we have f (n) = a − b − 1, and our task is to show
that 4ab + (a − b − 1)2 is not a square. This follows from
(a + b − 1)2 < 4ab + (a − b − 1)2 < (a + b)2 ,
which holds whenever a > b > 0. That condition holds when n ≥ 2.
Also solved by S. Amghibech (Canada), A. Bandeira & J. Moreira (Portugal), D. Beckwith, R. Chapman
(U. K.), P. P. Dályay (Hungary), O. Kouba (Syria), O. P. Lossers (Netherlands), C. R. Pranesachar (India),
A. Stadler (Switzerland), R. Stong, R. Tauraso (Italy), BSI Problems Group (Germany), FAU Problem Solving
Group, and the proposer.
862
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
PROBLEMS AND SOLUTIONS
Edited by Gerald A. Edgar, Doug Hensley, Douglas B. West
with the collaboration of Mario Benedicty, 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,
Richard Pfiefer, Cecil C. Rousseau, Leonard Smiley, Kenneth Stolarsky, Richard
Stong, Walter Stromquist, Daniel Ullman, Charles Vanden Eynden, Sam Vandervelde,
and Fuzhen Zhang.
PROBLEMS
11397. Proposed by Grahame Bennett, Indiana University, Bloomington, IN. Let
a, b, c, x, y, z be positive numbers such that a + b + c = x + y + z and abc = x yz.
Show that if max{x, y, z} ≥ max{a, b, c}, then min{x, y, z} ≥ min{a, b, c}.
11398. Proposed by Stanley Huang, Jiangzhen Middle School, Huaining, China. Sup-
pose that acute triangle ABC has its middle-sized angle at A. Suppose further that the
incenter I is equidistant from the circumcenter O and the orthocenter H . Show that
angle A has measure 60 degrees and that the circumradius of I BC is the same as that
of ABC.
11399. Proposed by Biaggi Ricceri, University of Catania, Catania, Italy. Let
(, F , μ) be a measure space with finite nonzero measure M, and let p > 0. Let
f be a lower semicontinuous function on R with the property that f has no global
minimum, but for each λ > 0, the function t → f (t) + λ|t| p does have a unique
global minimum. Show that exactly one of the two following assertions holds:
(a) For every u ∈ L p () that is not essentially constant,
1/ p
1
Mf |u(x)| dμ
p
< f (u(x)) dμ,
M
948
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
11401. Proposed by Marius Cavachi, “Ovidius” University of Constanţa, Constanţa,
Romania. Let A be a nonsingular square matrix with integer entries. Suppose that for
every positive integer k, there is a matrix X with integer entries such that X k = A.
Show that A must be the identity matrix.
11402. Proposed by Catalin Barboianu, Infarom Publishing, Craiova, Romania. Let
f : [0, 1] → [0, ∞) be a continuous function such that f (0) = f (1) = 0 and f (x) >
0 for 0 < x < 1. Show that there exists a square with two vertices in the interval (0,1)
on the x-axis and the other two vertices on the graph of f .
11403. Proposed by Yaming Yu, University of California Irvine, Irvine, CA. Let n be
an integer greater than 1, and let f n be the polynomial given by
n
i−1
n
f n (x) = (−x)n−i (x + j).
i=0
i j =0
SOLUTIONS
Solution by√Richard Stong, Rice University, Houston, TX. Let φ denote the golden
ratio (1 + 5)/2, and recall the Binet formula for the Fibonacci numbers:
√
Fn = φ n + (−1)n−1 φ −n / 5.
For odd m,
√
5(Fm + i F2m ) = iφ 2m + φ m + φ −m − iφ −2m = −iφ −2m (iφ m + 1)(iφ 3m + 1)
and
1 − 2φ m 1 + 2φ −1 −m
Fm − 2F1+m = √ φ + √ φ = −φ m + φ −m .
5 5
Hence
√ √
Fm − 2F1+m √ −iφ 3m + iφ m 5 5
= 5 m = − .
Fm + i F2m (iφ + 1)(iφ 3m + 1) iφ 3m + 1 iφ m + 1
A Sinh of a Series
11286 [2007, 78]. Proposed by M. L. Glasser, Clarkson University, Potsdam, NY. Show
that
∞
π/4
e − (−1)n sinh(n + 1/2)π e−n(n+1)π = 0.
n=0
950
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
is even and periodic in x with period 1. Therefore it is the sum of a cosine series
1 ∞
F(x, y) = A0 + An cos(2nπ x),
2 n=1
∞ 2
with An = 2 −∞
e−x y cos(2πnx) d x. It follows that
π
∞
−π 2 n 2 /y
F(x, y) = 1+2 e cos(2πnx) . (2)
y n=1
Substituting x = 1/2 and y = π into the expression for F(x, y) given in (1) leads to
∞
2π
−1
2π
∞
2π
F(1/2, π) = e−(k+1/2) = e−(k+1/2) + e−π/4 + e−(k+1/2)
k=−∞ k=−∞ k=1
∞
2π
∞
2π
∞
2
= e−(k−1/2) + 2e−π/4 + e−(k+1/2) =2 e−(k+1/2) π .
k=2 k=1 k=0
∞
2
∞
2
= (−1)n e−πn + (−1)n+1 e−π(n+1) .
n=0 n=0
Applying the definition of the hyperbolic sine, we obtain the desired equation.
Also solved by R. Chapman (U.K.), J. Grivaux (France), O. Kouba (Syria), G. Lamb, M. A. Prasad (India),
O. G. Ruehr, A. Stadler (Switzerland), R. Stong, J. Sun, FAU Problem Solving Group, GCHQ Problem Solving
Group, and the proposer.
Editorial comment. (i) The functions f and g need not be continuous—it is sufficient
that they be integrable. This was observed by Botsko, Pinelis, Schilling, and Schmu-
1 1
land. (ii) Keselman, Martin, and Pinelis noted that 0 x f (x) d x and 0 xg(x) d x can
1 1
be replaced with 0 φ(x) f (x) d x and 0 φ(x)g(x) d x, where φ(x) satisfies suitable
conditions—roughly speaking, that φ is differentiable and strictly monotonic, although
the specific conditions vary from one of these solvers to another.
Also solved by U. Abel (Germany), S. Amghibech (Canada), M. W. Botsko & L. Mismas, R. Chapman (U. K.),
J. G. Conlon & W. C. Troy, P. P. Dályay (Hungary), J. W. Hagood, E. A. Herman, S. J. Herschkorn, E. J.
Ionascu, G. L. Isaacs, G. Keselman, O. Kouba (Syria), J. H. Lindsey II, O. P. Lossers (Netherlands), G. Martin
(Canada), J. Metzger & T. Richards, M. D. Meyerson, A. B. Mingarelli & J. M. Pacheco & A. Plaza (Spain), E.
Mouroukos (Greece), P. Perfetti (Italy), I. Pinelis, M. A. Prasad (India), K. Schilling, B. Schmuland (Canada),
H.-J. Seiffert (Germany), J. Sun, R. Tauraso (Italy), M. Tetiva (Romania), L. Zhou, GCHQ Problem Solving
Group (U. K.), Microsoft Research Problems Group, NSA Problems Group, and the proposer.
Double Integral
11295 [2007, 452]. Proposed by Stefano Siboni, University of Trento, Trento, Italy.
For positive real numbers and ω, let M be the mapping of [0, 1) × [0, 1) into itself
defined by M(x, y) = ({2x}, {y + ω + x}), where {u} denotes u −
u, the fractional
part of u. For integers a and b, let ea,b (x, y) = e2πi(ax+by). Let
1 1
Cn (a, b; p, q) = ea,b (M n (x, y))e p,q (x, y) d x dy.
y=0 x=0
952
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
Since a and b are integers, the outer braces do not matter for Cn (a, b; p, q) =
1 1
e (2n a − p)x + (b − q)y + bnω + b({x} + {2x} + · · · + {2n−1 x}) d x dy.
0 0
For n ≥ 1 define
1
Fn (α, β) = exp iαx + 2iβx + iβ({2x} + · · · + {2n−1 x}) d x,
0
For n > 1, the last term in the exponent has period 1/2, so we split the integral into
two parts to obtain
i(α/2+β)
1/2
Fn (α, β) = e +1 exp i(α + 4β)x + iβ({4x} + · · · + {2n−1 x}) d x,
0
Finally, Cn (a, b; p, b) = e2πibωn Fn (2π(2n a − p − b), 2πb) yields the required re-
sult after some simplifications based on the assumption that a and b are integers.
Also solved by R. Chapman (U. K.), D. Fleischman, GCHQ Problem Solving Group, and the proposer.
A Tricky Minimum
11297 [2007, 452]. Proposed by Marian Tetiva, Bı̂rlad, Romania. For positive a, b,
and c, let
a 2 b2 c2 − 64
E(a, b, c) = .
(a + 1)(b + 1)(c + 1) − 27
Find the minimum value of E(a, b, c) on the set D consisting of all positive triples
(a, b, c) such that abc = a + b + c + 2, other than (2, 2, 2).
Solution by John H. Lindsey II, Cambridge, MA. Let m be the geometric mean, de-
fined by m = (abc)1/3 . By the arithmetic-geometric mean inequality, a + b + c ≥ 3m,
n B + k (−1)s(n) 1
=√ .
n=0 k odd
nB + k + 1 B
0<k<B
Solution by the proposer. Set (n) = (−1)s(n) . If B is odd, then s(n) ≡ n mod 2, since
all powers of B are odd. If B is even, then the constant term in the base-B expansion of
2m cannot be B − 1, and hence s(2m + 1) = s(2m) + 1. Hence in both cases (2m +
1) = −(2m). Let δk = 1 if k = 0, and otherwise δk = 0, and let
∞
n B + k (n)
Pk,B = Pk = .
n=δk
nB + k + 1
954
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 115
Then Pk converges because it is a product over m ≥ 0 of factors of the form
(2m)B + k + 1 (2m + 1)B + k ±1
· ,
(2m)B + k (2m + 1)B + k + 1
B−1
n B + k (n) B−1
k (0) ∞ B−1
n B + k (n)
=
k=0 n=δk
nB + k + 1 k=1
k+1 n=1 k=0
nB + k + 1
(1)
∞
1 m (m)
= .
B m=1 m + 1
m (m) B−1 ∞
n B + k (−1)k (n)
= .
m=1
m+1 k=0 n=δk
nB + k + 1
Substitute this into (1). Since the infinite products are all nonzero (being convergent
and having no zero factors), the factors for even k are the same on both sides and
cancel out. This yields
∞
n B + k (n) 1 ∞
n B + k −(n)
= .
k odd n=0
nB + k + 1 B k odd n=0 n B + k + 1
0<k<B 0<k<B
All the products are positive, and the desired formula follows.
Editorial comment. The solution of Problem 11222 for odd values of B given in the
May, 2008 issue of the Monthly was selected at a time when the proposer’s solution had
become separated from the file of solutions. The previously published solution treated
odd B only by reference to the literature. The proposer’s elegant solution covers all
cases simultaneously and efficiently. Fortunately, it was recovered, and we are pleased
to present it.
The names of solver Apostolis Demis, of Athens, Greece (11285, [2007,358]), and
William Dickinson, (11201, [2008,73]) were misspelled. Our apologies.
Paolo Perfetti gives a counterexample to the
if-direction of part (a) in 11257,[2008,
269]. If z k = (−1)k−1 / ln(k + 1) (and sk = 1j z k ), then sn converges, but z k /sk
diverges, tending to −∞.