172 THE MATHEMATICAL GAZETTE

Problem Corner
Solutions are invited to the following problems. They should be
addressed to Nick Lord at Tonbridge School, Tonbridge, Kent TN9 1JP
(e-mail: [email protected]) and should arrive not later than 10
August 2019.
Proposals for problems are equally welcome. They should also be sent
to Nick Lord at the above address and should be accompanied by solutions
and any relevant background information.

103.A (Michael Fox)

The point A is (6, 16, 12), B is (18, −9, 12) and P is a variable point
on the x-axis.
Find the ranges of possible values of (i) AP + BP, (ii) AP − BP, and
(iii) AP / BP.

103.B (Michel Bataille)

Let ABC be a triangle and D, E on the side BC be such that
BD = DE = EC. Let the circumcircles Γb and Γc of ABD and ACE,
respectively, intersect at A and A′ and let the tangents to Γb at D and to Γc at
E intersect in U .
Prove that the orthogonal projections P, Q, P′, Q′ of U onto AD, AE,
A′D, A′E, respectively, are collinear and that PP′ = QQ′.

103.C (K. S. Bhanu and M. N. Deshpande)

Consider the following method of generating a permutation of the
integers (1, 2, … , n). We have equal numbers of balls of three colours in
a bag. Balls are drawn randomly one by one n times with replacement. For
1 ≤ k ≤ n, the colour of the k th ball drawn is noted and we use this to
generate at most three cycles (a1, … , ak1), (b1, … , bk2) and (c1, … , ck3)
where the ai , bi and ci denote the draw numbers of balls of the same colour
and k 1 + k 2 + k 3 = n. From the product of these cycles, we generate a
random permutation (π1, … , πn).
Show that, for large n, E (π1) ≈ 4 and E (πn) ≈ 3.
[For example, if n = 7 and the draw number of balls of each colour are (1, 4),
(2, 3, 6) and (5, 7), then the generated permutation is (4, 3, 6, 1, 7, 2, 5).]

103.D (Ovidiu Furdui)

Let b, c, d ∈ r with bc > 0. Calculate

( )
n
b
1 sin
lim n .
n→∞ c d
sin cos
n n
Downloaded from https://www.cambridge.org/core. University of New England, on 14 Feb 2019 at 15:16:31, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/mag.2019.44
 PROBLEM CORNER 173

Solutions and comments on 102.E, 102.F, 102.G, 102.H (July 2018).

102.E (Martin Lukarevski)
Let a, b, c be the lengths of the sides of triangle ABC and let wa, wb, wc
be the lengths of the corresponding angle bisectors. Prove that
a2 b2 c2
+ + ≥ 4.
w2a w2b w2c
When does equality hold?

A wide variety of standard inequalities were used by solvers of this

intriguing geometrical inequality. All relied on the standard formulae for
2bc 2
the length of an angle bisector, wa = cos 12 A = bcs (s − a),
b + c b + c
where the first formula is quickly proved by expressing the area of triangle
ABC as the sum of the areas of the triangles created by wa and the second
s (s − a)
follows from the half-angle formula cos 12 A = .
bc
A number of solvers argued as follows:
a2 a2 (b + c)2 a2
= ≥ by the AM-GM inequality.
2
wa 4bcs (s − a) s (s − a)

( )
2
a2
But ∑s − a ∑ (s − a) ≥ ∑a (Cauchy-Schwarz) implies that
a2 a2 1
∑s − a ≥ 4s, whence ∑
w2a
≥ · 4s = 4, as required.
s
As Lienhard Wimmer observed, a method of proof also used by several
other solvers yields the generalisation

( )
n
a
∑ ≥ 31 − 2 .2n for n > 0.
n

wa
For

( ) ( )
n n/3
a abc
∑ wa
≥ 3
wawbwc
by the AM-GM inequality,

n
(b + c) (a + c) (a + b) ⎤ 3
= 3 ⎡⎢ ⎥ from the first formula for wa,
⎣ 8abc cos 12 A cos 12 B cos 12 C ⎦
3
≥ by the AM-GM inequality.
[ cos 1
2A cos 12 B cos 12 C] n/3
But Jensen's inequality applied to the concave function log cos x gives
A+ B+ C 3 3
cos 12 A cos 12 B cos 12 C ≤ cos3 =
6 8

Downloaded from https://www.cambridge.org/core. University of New England, on 14 Feb 2019 at 15:16:31, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/mag.2019.44
 174 THE MATHEMATICAL GAZETTE

( ) ( )
n
n
a 8 3

∑ = 31 − 2 .2n.
n
so that > 3
wa 3 3
(This also includes the result for n = 1 set by the proposer as Problem
11945 in the American Mathematical Monthly with solution by Dmitry
Fleischman using this argument in the October 2018 issue, pp. 758-759.)
a2
Finally, M. G. Elliott proved the stronger result ∑ 2 ≥ 4 involving
ma
the medians ma, mb, mc which implies the result in 102.E because wa ≤ ma,
wb ≤ mb, wc ≤ mc.
Correct solutions were received from: C. R. Aditya, M. Bataille, M. V. Channakeshava,
N. Curwen, S. Dolan, M. G. Elliott, GCHQ Problem Solving Group, G. Howlett, J. A. Mundie,
M. Nadapana, N. Neha, A. Ng, P. Nüetsch, V. Schindler, I. D. Sfikas, P. Shiu, G. B. Trustrum,
J. T. Varsha, L. Wimmer (2 solutions), S. Yip and the proposer Martin Lukarevski.

102.F (Micha³ Kremzer)

Find all pairs of positive real numbers (a, b) such that
ln (⎣a + b⎦) = ⎣a⎦ + ⎣b⎦, where ⎣x⎦ denotes the largest integer less than or
equal to x.

Answer: The solutions (a, b) comprise the triangular region of the plane
given by 0 < a, b < 1 and a + b ≥ 1.

The solvers of this attractive problem generally argued along similar

lines to Stan Dolan and Ioannis Sfikas whose succinct solution follows.

Writing a = ⎣a⎦ + δ, b = ⎣b⎦ + ε with 0 ≤ δ, ε < 1, we see that

⎣a + b⎦ = ⎣a⎦ + ⎣b⎦ + ⎣δ + ε⎦ ≤ ⎣a⎦ + ⎣b⎦ + 1. Now ex ≥ 1 + x
with equality if, and only if, x = 0. Thus
e⎣a⎦ + ⎣b⎦ ≥ (1 + ⎣a⎦) (1 + ⎣b⎦) ≥ 1 + ⎣a⎦ + ⎣b⎦ ≥ ⎣a + b⎦ .
The condition postulated in the problem then implies equality throughout
which occurs if, and only if, ⎣a⎦ = ⎣b⎦ = 0 and ⎣a + b⎦ = 1 so that
0 < a, b < 1 and a + b ≥ 1.
Peter Johnson investigated the effect of removing the restriction that
a, b are positive and showed that the solution set is then given by all the
translates of the triangular region above by translations of the form
n ∈ z. The problem is also worth investigating in other bases.
n
−n
, ( )
Correct solutions were received from: M. Bataille, M. V. Channakeshava, S. Dolan, M. G. Elliott,
GCHQ Problem Solving Group, P. F. Johnson, A. Ng, I. D. Sfikas, P. Shiu, G. B. Trustrum,
L. Wimmer and the proposer Micha³ Kremzer.

Downloaded from https://www.cambridge.org/core. University of New England, on 14 Feb 2019 at 15:16:31, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/mag.2019.44
 PROBLEM CORNER 175

102.G (John Melville)

1 ln (x + 1 − x2) π2
Show that ∫ dx = .
0 x 16
A wide variety of substitutions and other integration techniques were
employed to evaluate this integral which attracted many appreciative
comments from respondents. The solution below utilises ideas from several
solvers but most closely follows that given by M. G. Elliott.
1 ln (x + 1 − x2) π/2 ln (sin θ + cos θ )
Let I = ∫ dx = ∫ · cos θ dθ,
0 x 0 sin θ
on substituting x = sin θ .
The switch θ → π2 − θ shows also that
π/2 ln (sin θ + cos θ )
I = ∫ · sin θ dθ, so, on adding,
0 cos θ
π/2 ln (sin θ + cos θ ) π/2 ln (sin θ + cos θ )2
2I = ∫0 sin θ cos θ
dθ = ∫0 2 sin θ cos θ
dθ

π/2 ln (1 + sin 2θ )
= ∫0 sin 2θ
dθ

and
1 π/2 ln (1 + sin 2θ ) π/4 ln (1 + sin 2θ )
I =
2 ∫0 sin 2θ
dθ = ∫0 sin 2θ
dθ,

by symmetry.
The substitution t = tan θ then gives

I = ∫
1 ln 1 +( 2t
)
1 + t 2 · 1 dt = 1 ln (1 + t) − ln (1 + t ) dt.
2

0 2t 1 + t2 ∫0 t 2t
1 + t 2

1 ln (1 + t 2) 1 ln (1 + u)
But, setting u = t 2, shows that ∫ dt = ∫ du so that
0 2t 0 4u
3 1 ln (1 + t)
4 ∫0
I = dt
t

( )
∞
3 1 tn − 1
=
4 ∫0 n∑= 1 (−1)n − 1
n
dt

∞
3 (−1)n − 1
=
4 ∑ n2
, on switching the sum and integral
n=1

3⎡ ∞ 1 ∞
1 ⎤
4 ⎣n∑ ∑ (2n)2 ⎥⎦
= ⎢ 2
− 2
=1 n n=1

Downloaded from https://www.cambridge.org/core. University of New England, on 14 Feb 2019 at 15:16:31, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/mag.2019.44
 176 THE MATHEMATICAL GAZETTE
∞
3 1
=
8 ∑ n2
n=1
∞
π2 1 π2
=
16
, since ∑ n2 =
6
.
n=1

Another technique used quite frequently was differentiation under the

integral sign and we illustrate this with the integral used by the GCHQ
Problem Solving Group; similar arguments were given by Andrew Ng and
Ioannis Sfikas. We have
π/4 ln (1 + sin 2θ ) 1 π/2 ln (1 + sin θ )
I = ∫0 sin 2θ
dθ =
2 ∫0 sin θ
dθ

1 π ln (1 + sin θ )
4 ∫0
= dθ.
sin θ

(
1 + sin θ
)
π dθ
Consider f (λ) = ∫ ln for |λ| < π.
0 1 + cos λ sin θ sin θ
π sin λ
Then f ′ (λ) = ∫ dθ = 2λ, on evaluating the integral using
0 1 + cos λ sin θ
substitution t = tan 2 θ . It follows that, since f (0) = 0, f (λ) = λ2 so that
1

f ()
π
2
= ∫0
π ln (1 + sin θ)
sin θ
dθ =
π2
4
and I = 14 f
π
2
=
π2
16
. ()
Correct solutions were received from: M. Bataille, N. Curwen, S. Dolan, M. G. Elliott, GCHQ
Problem Solving Group, R. Gordon, G. Howlett, J. D. Mahony, A. Ng, V. Schindler, I. D. Sfikas,
P. Shiu, I. Sofair and the proposer John Melville.

102.H (Chris Starr)

Consider the ellipse given by the parametric equations x = a cos θ ,
y = b sin θ and let P (p, q) be a point not on the ellipse with p, q ≠ 0.

(a) Show the value of θ giving the minimum distance from P to the ellipse
is given by a quartic in t = tan 12 θ .

(b) Show that rational, non-zero values of a, b, p, q can be found such that
the quartic in part (a) factorises as the product of two quadratics with
rational coefficients.

(c) Find the exact value of the minimum distance from P (− 15

4 , 1) to the
y2
ellipse 4 + 9 = 1.
x2

Answer: (c) The minimum distance is 1

20 7965 − 1520 19.

Downloaded from https://www.cambridge.org/core. University of New England, on 14 Feb 2019 at 15:16:31, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/mag.2019.44
 PROBLEM CORNER 177

Parts (a) and (c) of this problem were tackled in much the same way by
all solvers.
(a) The distance D from P (p, q) to (a cos θ, b sin θ) is given by
D2 = (p − a cos θ )2 + (q − b sin θ ) .
2

2
dD
This is minimised when = 0 giving
dθ
(a2 − b2) sin θ cos θ − ap sin θ + bq cos θ = 0,
which is the condition for P to lie on the normal to the ellipse at
2t
(a sin θ, b cos θ ). Substituting t = tan 12 θ with sin θ = ,
1 + t2
1 − t 2
cos θ = and simplifying gives the quartic
1 + t2
bqt 4 + 2 (ap + a2 − b2) t 3 + 2 (ap − a2 + b2) t − bq = 0. (∗)

−15
(c) Here, a = 2, b = 3, p = , q = 1 and the quartic from part (a) is
4
3t 4 − 25t 3 − 5t − 3 = 0 or (3t 2 − t + 1) (t 2 − 8t − 3) = 0.
Geometrically, it is clear that the minimising point on the ellipse is in the
second quadrant, corresponding to a positive, real value of t , so the relevant
−1 − 2 19
root is t = 4 + 19 (from the second factor) giving cos θ = ,
10
−2 + 19
sin θ = and D = 20 1
7965 − 1520 19.
10
(b) Obviously, part (c) gives one example where the quartic in part (b)
factorises in the required manner, but respondents rose magnificently to the
challenge of finding other solutions.
Stan Dolan, Peter Shiu and the GCHQ Problem Solving Group observed
that when a = b (so the ellipse is a circle), (∗) becomes
aqt 4 + 2apt 3 + 2apt − aq = a (t 2 + 1) (qt 2 + 2pt − q)
which, for rational a, p, q, provides a solution.
Russell Gordon and Volkhard Schindler gave the following general
approach. The monic quartic corresponding to (∗) is
2
t4 + (ap + a2 − b2) t 3 + 2 (ap − a2 + b2) t − 1 = 0
bq bq
for which we seek a factorisation

(t 2 (
+ ct + d ) t 2 +
1 − d2
cd
t −
1
d
= 0 )
involving rational numbers c, d .

Downloaded from https://www.cambridge.org/core. University of New England, on 14 Feb 2019 at 15:16:31, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/mag.2019.44
 178 THE MATHEMATICAL GAZETTE

Equating coefficients shows that

1 − d2 2
A ≡ c + = (ap + a2 − b2)
cd bq
c 1 − d2 2
B ≡ −+ = (ap − a2 + b2) .
d c bq
Then Abq = 2ap + 2a2 − 2b2, Bbq = 2ap − 2a2 + 2b2
4 (b2 − a2) (B + A) (b2 − a2)
solves to give q = ,p = .
b (B − A) a (B − A)
Hence, given rational values for a, b, we can choose rational values for c, d
and then use the above equations to find P (p, q) with rational coordinates.
Other, specific solutions with a, b, p, q (where, by symmetry, we may
suppose that a, b, p, q > 0) included:
• a = m, b = n, p = 2n, q = 2m: a good way of generating
examples with integer parameters. (J. A. Mundie)
• The sub-case c = d = 1λ (λ ≠ 0, ±1, −2) with
a = 8 λ + 2λ + 6), b = 18 (2 + 6λ − λ2), ap = 14 λ (λ − 1) (λ + 2),
1( 2

bq = λ. (M. G. Elliott)
• Two families with a − b = ±1 and single rational parameter
Q > 1:
8Q (Q2 + 1) 4Q
a = 14 (Q2 + 3), b = 14 (Q2 − 1) , p = 2 ,q= 2
(Q − 1)(Q + 3)
2 Q −1
and
8Q (Q2 + 1) 4Q
a = 14 (Q2 − 1), b = 14 (Q2 + 3), p = 2 ,q =
( Q − 1)
2 Q2 + 3
(where Q = 3 corresponds to part (c) of 102.H). (G. Howlett)
• Integer solutions of the Pell-type equation b2 − 2a2 = q2 with
4bq (a2 − b2)
p = , where a = 2, b = 3, q = 1 gives part (c)
a (b2 − q2)
of 102.H. (Chris Starr, the proposer)
Russell Gordon commented that the evolute of an ellipse is an astroid. If P
is inside the astroid, the quartic (∗) has four real roots: outside (as in part (c))
it has just two real roots. And H. Azad and A. Laradji showed in the
November 2004 Gazette (pp. 548-551) that, in general, the construction of
the minimising point on the ellipse is not possible with ruler and compasses.
Correct solutions were received from: M. V. Channakeshava, N. Curwen, S. Dolan, M. G. Elliott,
GCHQ Problem Solving Group, R. Gordon, G. T. Q. Hoare, G. Howlett, J. D. Mahony,
J. A. Mundie, V. Schindler, I. D. Sfikas, P. Shiu, L. Wimmer and the proposer Chris Starr.
10.1017/mag.2019.44 N.J.L.

Downloaded from https://www.cambridge.org/core. University of New England, on 14 Feb 2019 at 15:16:31, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/mag.2019.44