No.

1
2020
 CONTENTS
Articles

Some Inequalities Involving the Medians of a Triangle

by M. Bencze and M. Drăgan 2

The Last Three Decades of Lalescu Limit

by D. M. Bătineţu-Giurgiu, N. Stanciu,
and J. L. Díaz-Barrero 16

A New Bound for the Moduli of the Zeros

by José Luis Díaz-Barrero 27

Problems

Elementary Problems: E77–E82 33

Easy–Medium Problems: EM77–EM82 34

Medium–Hard Problems: MH77–MH82 36

Advanced Problems: A77–A82 38

Mathlessons

Seventeen Camels Problem

by O. Rivero Salgado and J. L. Díaz-Barrero 41

Contests

2nd Barcelona Spring Math-Olympiad

by O. Rivero Salgado and J. L. Díaz-Barrero 49
Solutions

Elementary Problems: E71–E76 54

Easy–Medium Problems: EM71–EM76 66

Medium–Hard Problems: MH71–MH76 78

Advanced Problems: A71–A76 91

Volume 7, No. 1, Spring 2020 1

Articles
Arhimede Mathematical Journal aims to publish interesting and
attractive papers with elegant mathematical exposition. Articles
should include examples, applications and illustrations, when-
ever possible. Manuscripts submitted should not be currently
submitted to or accepted for publication in another journal.

Please, send submittals to José Luis Díaz-Barrero, Enginyeria

Civil i Ambiental, UPC BARCELONATECH, Jordi Girona 1-3, C2,
08034 Barcelona, Spain, or by e-mail to
[email protected]
2 Arhimede Mathematical Journal

Some Inequalities Involving

the Medians of a Triangle

M. Bencze and M. Drăgan

Abstract

In this paper we present the best upper bounds for sums of

the medians of a triangle as homogeneous functions of the ra-
dius of the circumscribed circle R and the radius of the circle
inscribed to the triangle r . Using these results, we also give a
proof the Sun Wen Cai inequality (see [3]).

1 Introduction
Chu Xiao Guang and Yang Xue-Zhi [6] published the following
inequality:
(ma + mb + mc )2 ≤ 4s2 − 16Rr + 5r 2 ,
where ma , mb , mc are the medians, R the circumradius, r the
inradius, and s the semiperimeter of triangle ABC .
Sun Wen Cai posed a stronger conjecture in a personal commu-
nication which was solved by J. Liu [3, 4]. It states that, in any
triangle ABC ,
(ma + mb + mc )2 r2
≤2+ .
a2 + b2 + c2 R2
In this paper, we will find the best homogenous function f (R, r)
(ma + mb + mc )2
such that the inequality ≤ f (R, r) holds in
a2 + b2 + c2
every triangle ABC . To achieve this, we need the following well-
known result.
Volume 7, No. 1, Spring 2020 3

Theorem 1 (Blundon [1]). In any triangle ABC we have the in-

equality s1 ≤ s ≤ s2 , where
È p
s1 = 2R2 + 10Rr − r 2 − 2(R − 2r) R2 − 2Rr,
È p
s2 = 2R2 + 10Rr − r 2 + 2(R − 2r) R2 − 2Rr.

Equality occurs for the two isosceles triangles with sides

2r(R + r − d) (R + r − d)(R − d)
a1 = p , b1 = c1 = p ,
(R − d)2 − r 2 (R − d)2 − r 2
2r(R + r + d) (R + r + d)(R + d)
a2 = p , b 2 = c2 = p ,
(R + d)2 − r 2 (R + d)2 − r 2

where
p
d = kOIk = R2 − 2Rr. (1)

We will also need the following.

Lemma 1. In any triangle ABC , the following equalities hold:

X 9 2 
m2a m2b = (s + r 2 + 4Rr)2 − 16Rrs2 , (2)
16
(4mc mb mc )2 = s6 + (33r 2 − 12Rr)s4
− (33r 4 + 60R2 r 2 + 120Rr 3 ) − (r 2 + 4Rr)3 . (3)

Proof. In order to prove (2), we denote x = a2 + b2 + c2 . Then,

X 1 X
m2a m2b = (2x − 3a2 )(2x − 3b2 )
16
1 X 2  9 X 2 2
= 4x − 6x(x − c2 ) + 9a2 b2 = a b
16 16
9 2 
= (s + r 2 + 4Rr)2 − 16Rrs2 .
16
4 Arhimede Mathematical Journal

In order to prove (3), we denote r 2 + 4Rr = z . We have

Y Y
64m2a m2b m2c =[2(b2 + c2 ) − a2 ] = (2x − 3a2 )
X
= 8x3 − 12x3 + 9 a2 b2 2x − 27a2 b2 c2
X
= − 4x3 + 18x[( ab)2 − 4abc s] − 27 · 16R2 r 2 s2
= − 32(s2 − z)3 + 36(s2 − z)[(s2 + z)2 − 16Rrs2 ] − 432R2 r 2 s2
= − 32(s6 − 3s4 z + 3s2 z 2 − z 3 )
+ (36s2 − 36z)[s4 + (2z − 16Rr)s2 + z 2 ] − 432R2 r 2 s2
= − 32s6 + 96z s4 − 96z 2 s2 + 32z 3 + 36s6 + (72z − 576Rr)s4
+ 36z 2 s2 − 36zs4 − (72z 2 − 576Rrz)s2 − 36z 3 − 432R2 r 2 s2
= 4s6 + (96z + 72z − 576Rr − 36z)s4
− (96z 2 −36z 2 +72z 2 −576Rrz +432R2 r 2 )s2 + 32z 3 − 36z 3
= 4s6 + [132(r 2 + 4Rr) − 576Rr]s4
− [132(r 2 + 4Rr)2 − 576Rr(r 2 + 4Rr) + 432R2 r 2 ]s2 − 4z 3
= 4s6 + (132r 2 −48Rr)s4 − (132r 4 +480Rr 3 +240R2 r 2 )s2 − 4z 3 .

We then conclude that (4ma mb mc )2 = s6 + (33r 2 − 12Rr)s4 −

(33r 4 + 120Rr 3 + 60R2 r 2 )s2 − (r 2 + 4Rr)3 , as we wanted to see.

2 Main results

Below we present the main results obtained in this work. We start

with the following.

Theorem 2. In any triangle ABC , we have that

p p
(R − d)2 − r 2 + 8r 2 + (R − d)2 ma + mb + mc
p ≤ √
4r 2 + 2(R − d)2 a2 + b2 + c2
p p
(R + d)2 − r 2 + 8r 2 + (R + d)2
≤ p . (4)
4r 2 + 2(R + d)2
Volume 7, No. 1, Spring 2020 5

Proof. We denote x = a2 + b2 + c2 = 2(s2 − r 2 − 4Rr) and

Ê
ma + mb + mc X m2a
w= √ =
a2 + b2 + c2 x
È È È
= P (a, b, c) + P (b, c, a) + P (c, a, b).

After squaring, we obtain that

s Ê
Xm 2 X m2 m2b m2a m2b m2c
a a
w2 = +2 + 2 .
x x2 x3
If f, g, h : [s1 , s2 ] → R are the functions defined by
Ê
X m2 X m2 m2 m2a m2b m2c
a a b
f (s) = , g(s) = , h(s) = ,
x x2 x3
p
then we have (w2 − f (s))2 = 4(g(s) + 2h(s)w), with w > f (s).
”p Š
Let F : f (s), +∞ → R be defined by
hÈ 
F (u) = u4 − 2f (s)u2 − 8h(s)u + f 2 (s) − 4g(s), u ∈ f (s), +∞ .

We observe that F (w) = 0. Since

F 0 (u) = 4u3 − 4f (s)u − 8h(s) = 4u(u2 − f (s)) − 8h(s),

€ p Š €p Š
it follows that F 0 is increasing in 0, f (s) . But F 0 f (s) =
−8h(s) ≤ 0 and lim F 0 (u) = +∞. It follows that F 0 has only one
”p Š
u→∞
root in f (s), +∞ . Since
È  È
F f (s) = f 2 (s) − 2f 2 (s) − 8h(s) f (s) + f 2 (s) − 4g(s)
È
= −8h(s) f (s) − 4g(s) ≤ 0

and lim F (u) = +∞, it results that the equation F (u) = 0 has
u→∞ ”p Š
w as its only solution in f (s), +∞ .

In conclusion, we have that there exists a unique continuous

function u : [s1 , s2 ] → R such that F (u(s)) = 0, ∀ s ∈ [s1 , s2 ].
6 Arhimede Mathematical Journal

From the implicit function theorem it results that the function u

is differentiable in the interval (s1 , s2 ).

If we differentiate the equality

(u2 (s) − f (s))2 = 4(g(s) + 2h(s)u(s)), ∀ s ∈ (s1 , s2 ),

we obtain that

2(u2 (s) − f (s))(2u(s)u0 (s) − f 0 (s))

= 4g 0 (s) + 8h0 (s)u(s) + 8h(s)u0 (s), ∀ s ∈ (s1 , s2 )

or, equivalently,

(2u3 (s) − 2f (s)u(s) − 4h(s))u0 (s)

= f 0 (s)(u2 (s) − f (s)) + 2g 0 (s) + 4h0 (s)u(s), ∀ s ∈ (s1 , s2 ). (5)

From the AM-GM inequality, we have

XÈ È
u2 (s) = f (s) + 2 P (a, b, c)P (b, c, a) ≥ f (s) + 6 3 h2 (s) (6)

or È
u2 (s) − f (s) > 6 3
u(s). (7)
Furthermore,
È √ È
2 3
u (s) ≥ 6 h2 (s) or u(s) ≥ 6 3 u(s). (8)

From (7) and (8) we obtain that

2u3 (s) − 2f (s)u(s) − 4h(s) = 2u(s)(u2 (s) − f (s)) − 4h(s)

È
> 12 3 h2 (s)u(s) − 4h(s)
√
> 12 6h(s) − 4h(s)
€ √ Š
= 12 6 − 4 h(s) > 0,

so we obtain that

2u3 (s) − 2f (s)u(s) − 4h(s) > 0. (9)

Volume 7, No. 1, Spring 2020 7

P 2
We calculate that f (s) = ma /x = 3/4 and f 0 (s) = 0 for all
s ∈ [s1 , s2 ]. From (2), we have that
P – ™
m2a m2b 9 (s2 + r 2 + 4Rr)2 − 16Rrs2
g(s) = = .
x2 64 s2 − r 2 − 4Rr

To simplify the calculation, we denote t = s2 − r 2 − 4Rr. We have

9[(t + 2r 2 + 8Rr)2 − 16Rr(t + r 2 + 4Rr)]

g(s) =
64t2
9(t + 4r t + 4r + 16Rr 3 )
2 2 4
= . (10)
64t2

We consider the function H : [s1 , s2 ] → R, defined by H(s) = h2 (s).

From (3), we have that

1 s6 + xs4 − ys2 − z 3
H(s) = ,
32 (s2 − z)3

where we denote x = 33r 2 − 12Rr, y = 33r 4 + 60R2 r 2 + 120Rr 3

and z = r 2 + 4Rr, respectively. We obtain that

1 (t + z)3 + x(t + z)2 − y(t + z) − z 3

H(s) =
32 t3
t + (3z + x)t + (3z 2 + 2zx − y)t + z(xz − y)
3 2
=
32t3
•  ‹3  ‹2
1 1 2
1
= z(xz − y) + (3z + 2zx − y) 2
32 s2 − z s −z
 ‹ ˜  ‹
1 1 1
+ (3z + x) 2 +1 = Q 2 . (11)
s −z 32 s −z
√
We define Q : [1/ 4R2 + 2r 2 , +∞) → R by

Q(α) = a0 α3 + a1 α2 + a2 α + a3 ,

where we set β = R/r, a0 = z(xz − y) = −108β 2 (1 + 4β)r 6 < 0,

a1 = 3z 2 + 2zx − y = (−108β 2 + 144β + 36)r 4 ≤ 0 since β ≥ 2,
a2 = 3z + x = 36r 2 , and a3 = 1.
8 Arhimede Mathematical Journal

1 1
We have that ≥ . Since Q00 (α) = 6a0 α + 2a1 < 0,
s2 − z 4R2 + 2r 2
it follows that
 ‹
0 0
1
Q (α) ≤ Q . (12)
4R2 + 2r 2
We prove that
 ‹
0
1
Q ≤0 (13)
4R2 + 2r 2
or
3a0 + 2a1 (4R2 + 2r 2 ) + a2 (4R2 + 2r 2 )2 ≤ 0

or

−324(β 2+4β 3 )+(8β 2+4)(−108β 2+144β+36)+36(16β 4+16β 2 +4) ≤ 0

or

−324β 2 − 1296β 3 − 864β 4 + 1152β 3 + 288β 2

−432β 2 + 576β + 144 + 576β 4 + 576β 2 + 144 ≤ 0

or
288β 4 + 144β 3 − 108β 2 − 576β − 288 ≥ 0, ∀ β ≥ 2,

inequality which is true. From (12) and (13), we conclude that

Q0 (α) ≤ 0, which implies that H 0 (s) ≥ 0 for all s ∈ [s1 , s2 ].

From the AM-GM inequality we have that

m2a m2b m2c 27 m2a m2b m2c 1 1

H(s) = P 3 = 3 P 2
3 ≤ 3 , or ≥ 43 .
( a )2 4 ma 4 H(s)

From (6) we have that

Ê
3 È 3 È
u2 (s) ≥ +6 3
H(s), or u(s) ≥ +6 3
H(s).
4 4
Volume 7, No. 1, Spring 2020 9

Since H 0 (s) ≥ 0, we have that

1
g 0 (s) + 2h0 (s) · u(s) = g 0 (s) + p H 0 (s)u(s)
H(s)
Ê
1 3 È
≥ g 0 (s) + p +6 3
H(s)H 0 (s)
H(s) 4
Ì s ‹2
0
3 1 1
= g (s) + +6 3
H 0 (s)
4 H(s) H(s)
Ê
3 √
3
≥ g 0 (s) + · 43 + 6 46 H 0 (s)
4
p
= g (s) + 3 · 42 + 6 · 42 H 0 (s)
0

= g 0 (s) + 12H 0 (s) (14)

Next, we prove that the function G : [s1 , s2 ] → R defined by G(s) =

g(s) + 12H(s) is an increasing function. Indeed,
9 t2 +4r 2 t+4r 2 z 12 t3 +(3z +x)t2 +(3z 2 +2zx−y)t + z(xz −y)
G(s) = +
64 t2 32 t3
3 2 2 2 3 2 2
9t +36r t +36r zt+24t +24(3z+x)t +24(32 +2zx−y)t+24z(xz−y)
=
64t3
33t3 +[36r 2 +24(3z +x)t2 +(36r 2 z +24(3z 2 +2zx−y)]t+24z(xz −y)
=
‹3 – ‹2 ™
64 + 3
1
 1
 1
2 2
= 24z(xz −y) + 36r z +24(3z +2zx − y)
64 s2 −z s2 −z
•  1
‹ ˜
2
+ 36r +24(3z +x) +33
s2 −z
–  ‹3  ‹2  ‹ ™
1 1 1 1
= a0 + a1 + a2 + a3 .
64 s2 − z s2 − z s2 − z
h 1 
Now, we consider the function U : , +∞ → R defined
4R2 + 2r 2
3 2
by U (α) = a0 α + a1 α + a2 α + a3 , where
a0 = 24z(xz − y)
= 24r 6 (1 + 4β)(33 + 132β − 12β − 48β 2 − 33 − 60β 2 − 120β)
= −24 · 108r 6 β 2 (1 + 4β) < 0,
10 Arhimede Mathematical Journal

a1 = 36r 2 z + 24(3z 2 + 2zx − y)


= r4 36(1 + 4β)
 
+ 24 3(1+16β 2 +8β) + (2+8β)(33−12β) − 33 − 60β 2 − 120β

= r4 36 + 144β

+24(3+48β 2 +24β+66−24β+264β−96β 2 −33−60β 2 −120β)
 
= r 4 36 + 144β + 24(−108β 2 + 144β + 36)
= r 4 (−2592β 2 + 360β + 900) ≤ 0 for β ≥ 2,

and

a2 = 36r 2 + 24(3z + x) = r 2 [36 + 24(3 + 12β + 33 − 12β)]

= r 2 (36 + 24 · 36) = 900r 2 .

Thus,  ‹
1 1
G(s) = U . (15)
64 s2 − z
1 1
But α = ≥ , and we have that U 0 (α) = 3a0 α2 +
s2
−z + 4R2 2r 2
2a1 α + a2 . We will now prove that

U 0 (α) ≤ 0 (16)

1
for each α ≥ .
4R2 + 2r 2
 ‹
00 0 0
1
Since U (α) = 6a0 α + 2a1 ≤ 0, we have U (α) ≤ U .
4R2 + 2r 2
To prove (16), it will suffice to show that
 ‹
0
1
U ≤0
4R2 + 2r 2
or
3a0 + 2a1 (4R2 + 2r 2 ) + a2 (4R2 + 2r 2 )2 ≤ 0
or

−72·108β 2 (1+4β)+(8β 2+4)(−2592β 2+360β+900)+(4β 2+2)2 900 ≤ 0

Volume 7, No. 1, Spring 2020 11

or
−18 · 108(β 2 + 4β 3 ) + (2β 2 + 1)(−2592β 2 + 360β + 900)
+225(16β 4 + 16β 2 + 4) ≤ 0
or
−1994β 2 − 7776β 3 − 5184β 4 + 720β 3 + 1800β 2
−2592β 2 + 360β + 900 + 3600β 4 + 3600β 2 + 900 ≤ 0
or
1584β 4 + 7056β 3 − 814β 2 − 360β − 1800 ≥ 0,
which is true because β ≥ 2. Thus, U is a decreasing function.
From (15) it follows that G is an increasing function. Therefore,
G0 (s) = g 0 (s) + 12H 0 (s) ≥ 0. (17)
From (14) and (17), it results that
2g 0 (s) + 2h0 (s)u(s) ≥ 0, ∀ s ∈ [s1 , s2 ]. (18)
From (5), (9) and (18) it follows that u0 (s) ≥ 0 for all s ∈ [s1 , s2 ].
Thus, u is an increasing function, so u(s1 ) ≤ u(s) ≤ u(s2 ).
From (1) we now obtain that
"Ê Ê #
1 4b21 − a21 2a21 + b21
u(s1 ) = p 2 +2
a1 + 2b21 4 4
Ê p – ™
R−r−d 8r 2 +(R−d)2
= (R+r−d) 1+ p
(R+r−d)[4r 2 +2(R−d)2 ] (R−d)2 −r 2
p p
(R − d)2 − r 2 + 8r 2 + (R − d)2
= p
4r 2 + 2(R − d)2
and
"Ê Ê #
1 4b22 − a22 2a21 + b21
u(s2 ) = p 2 +2
a2 + 2b22 4 4
p p
(R + d)2 − r 2 + 8r 2 + (R + d)2
= p .
4r 2 + 2(R + d)2

The following inequality is proved in [6].

12 Arhimede Mathematical Journal

Theorem 3. In any triangle ABC we have

(ma + mb + mc )2 r2
≤2+ . (19)
a2 + b2 + c2 R2
p
Proof. We denote R/r = x, dx = x(x − 2). By (4), it turn out
that proving (19) is equivalent to proving that
ʁ ‹
È È 1
(x + dx )2 −1+ (x + dx )2 +8≤ 2+ [4 + 2(x + dx)2 ]
x2
and, after squaring,
È
2x2 (x + dx)2 + 7x2 + 2x2 [(x + dx )2 − 1][(x + dx )2 + 8]
≤ 8x2 + 4 + (4x2 + 2)(x + dx )2

or
È
2x2 [(x + dx )2 − 1][(x + dx )2 + 8] ≤ x2 + 4 + (2x2 + 2)(x + dx )2

or
È
2x2 [(x + dx )2 − 1][(x + dx )2 + 8]
≤ 4x4 − 4x3 + 5x2 − 4x + 4 + (4x3 + 4x)dx . (20)

After squaring, inequality (20) may be written as

4x4 (2x2−2x−1)(2x2 −2x+8)+16x6 (x2 −2x)+8x5 (4x2 −4x+7)dx

≤ (4x4 − 4x3 + 5x2 − 4x + 4)2 + (4x3 + 4x)2 (x2 − 2x)
+ (8x3 + 8x)(4x4 − 4x3 + 5x2 − 4x + 4)dx

or

16x8 − 16x7 + 64x6 − 16x7 + 16x6 − 64x5 − 8x6 + 8x5 − 32x4

+ 16x6 (x2 − 2x) + (32x7 − 32x6 + 56x5 )dx
≤ 16x8 + 16x6 + 25x4 + 16x2 + 16x2 − 32x7 + 40x6 − 32x5
+ 32x4 − 40x5 + 32x4 − 32x3 + (16x6 + 32x4 + 16x2 )(x2 − 2x)
+(32x7−32x6+40x5−32x4+32x3+32x5−32x4+40x3−32x2+32x)dx
− 40x3 + 40x2 − 32x
Volume 7, No. 1, Spring 2020 13

or

16x6 +16x5 −121x4 +72x3 −56x2 +32x−16 −(32x5 +16x3 )(x−2)

≤ (16x5 − 64x4 + 72x3 − 32x2 + 32x)dx

or

(x−2)(16x5 +48x4 −25x3 +22x2 −12x+8)−(x−2)(32x5 +16x3 )

≤ 8x(x − 2)2 (2x2 + 1)dx

or
 
(x − 2)2 8x(2x2 + 1)dx + 16x4 − 16x3 + 9x2 − 4x + 4 ≥ 0,

which is true since x ≥ 2.

In the following, we find the best constants α, β ∈ R and n ≥ 0

such that the inequality
(ma + mb + mc )2  r n
≤α+β (21)
a2 + b2 + c2 R
holds in every triangle ABC .
We prove that the inequality (19) is the best which we search.

Suppose that there exist α, β ∈ R and n ≥ 0 better than in

Theorem 3. From Theorem 2 we obtain that
€p p Š2
2 (R + d) 2 − r2 + 8r 2 + (R + d)2
(ma + mb + mc )
≤
a2 + b2 + c2 4r 2 + 2(R + d)2
 r n  r 2
≤α+β ≤2+ (22)
R R
holds in every triangle ABC .

In the case of the isosceles triangle with a = 0, b = c = 1, ma = 1,

mb = mc = 1/2, R = 1/2 and r = 0, from (22) we obtain α = 2.
β 1
In the case of the equilateral triangle, we obtain α + = 2+ .
 2n
‹n−2 4
n−2
x
Thus, β = 2 . Also, from (22) we have that ≥ 1 or
2
14 Arhimede Mathematical Journal

n ≥ 2. We consider n > 2, in which case

È È 2 ‚ 2
Œ
n−2  
(x + dx )2 −1 + (x + dx )2 +8 ≤ 2+ n 4 + 2(x+ dx )2
x

or, after performing some calculations,

È
2xn [(x + dx )2 − 1][(x + dx )2 + 8]
≤ 4xn+2 − 4xn+1 + xn + 2n x2 − 2n x + 2n + (4xn+1 + 2n x)dx

or

4x2n (2x2 − 2x − 1 + 2xdx )(2x2 − 2x + 8 + 2xdx )

≤ (4xn+2 − 4xn+1 +xn +2n x2 −2n x+2n )2 +(4xn+1 +2n x)2 (x2 −2x)
+ (8xn+1 + 2n+1 x)(4xn+2 − 4xn+1 + xn + 2n x2 − 2n x + 2n )dx

or

48x2n+2 + a1 x2n+1 + a2 x2n + a3 xn+4 + a4 xn+3 + a5 xn+2

+ a6 xn+1 + a7 xn + a8 x4 + a9 x3 + a10 x2 + a11 x + a12
≤ dx (−48x2n+1 +b1 xn+3 +b2 xn+2 +b3 xn+1 +b4 x3 +b5 x2 +b6 x.
(23)

Since n > 2, we have

2n + 2 6∈ {2n + 1, 2n, n + 4, n + 3, n + 2, n + 1, n, 4, 3, 2, 1}

and
2n + 1 6∈ {n + 3, n + 2, n + 1, 3, 2, 1}.
Dividing (23) by x2n+2 and taking x → ∞, we obtain that 48 ≤
−48, which represents a contradiction. Hence, we must have
n = 2. It results that inequality (19) is the best of type (21).
Volume 7, No. 1, Spring 2020 15

References
[1] Blundon, W. J. “Inequalities Associated with the Triangle”.
Canad. Math. Bull. 8.5 (1965), pp. 615–626. DOI: 10.4153/
CMB-1965-044-9.
[2] Blundon, W. J. “Problem 1935”. The Amer. Math. Monthly 73
(1966), p. 1122.
[3] Liu, J. “On a sharp inequality for the medians of a triangle”.
Transylv. J. Math. Mech. 2.2 (2010), pp. 141–148. ISSN: 2067-
239X.
[4] Liu, J. “On an inequality for the medians of a triangle”. J. Sci.
Arts 2(19) (2012), pp. 127–136. ISSN: 1844-9581.
[5] Mitrinović, D. S., Pečarić, J. E., and Volenec, V. Recent ad-
vances in geometric inequalities. Vol. 28. Mathematics and
its Applications (East European Series). Kluwer Academic
Publishers Group, Dordrecht, 1989. ISBN: 90-277-2565-9.
DOI : 10.1007/978-94-015-7842-4.
[6] Xi, G. C. and Xu, Z. Y. “On some inequalities involving me-
dians of a triangle”. Research in Inequalities, Tibet People’s
Press, Lha So (2000), pp. 234–247.

Mihály Bencze
Hărmanului 6, 505600 Săcele-Négyfalu,
Braşov, Romania.
[email protected]

Marius Drăgan
61311 Timişoara 35,
Bl. 0D6, Sc. E, et. 7, Ap. 176, Sect. 6,
Bucureşti, Romania
[email protected]
16 Arhimede Mathematical Journal

The Last Three Decades of

Lalescu Limit
D. M. Bătineţu-Giurgiu, N. Stanciu,
and J. L. Díaz-Barrero

Abstract

This article presents alternative solutions to some problems

involving Lalescu’s sequence that have been published in prob-
lem magazines around the world.

1 Introduction
In 2000, a paper entitled Hundred years of Lalescu Sequence [3]
was published. Since then, many problems have been published
and continue to be published today in almost all the journals with
Problem columns around the world. In this paper, continuing with
what has been published previously and from a problem-solving
point of view, some of the problems that have been published in
the last three decades are presented. We give below the statement
and solution of Lalescu’s Problem:
Problem (Lalescu, 1900-1901). Compute
√
n
È 
lim n! − n−1 (n − 1)! .
n→∞

Solution. Using the Cauchy-D’Alembert criterium, we have that

√
n
Ê
n! n n! (n + 1)! nn
lim = lim = lim ·
n→∞ n n→∞ nn n→∞ (n + 1)n+1 n!
 ‹n  ‹n
n 1 1
= lim = lim 1 − = .
n→∞ n + 1 n→∞ n+1 e
Volume 7, No. 1, Spring 2020 17

Let {un }n≥2 be the sequence defined by

p
n+1
(n + 1)!
un = √
n
.
n!
Then, we have
p p
n+1
(n + 1)! (n + 1)! n + 1 n
n+1

lim un = lim √
n
= lim · ·√
n
= 1,
n→∞ n→∞ n! n→∞ n+1 n n!
un − 1
lim = 1,
n→∞ ln un
and
‚ p Œn
n+1
(n + 1)! n+1
lim un
n = lim √n
= lim p = e.
n→∞ n→∞ n! n→∞ (n + 1)!
n+1

Finally, we have
√ È  ‚ p Œ
n n−1
√
n (n + 1)!
n+1

lim n! − (n − 1)! = lim n! √n

−1
n→∞ n→∞ n!
√
n
= lim n! (un − 1)
n→∞
√
n
n! un − 1 1
= lim · · ln un
n = .
n→∞ n ln un e

2 Lalescu type limits

Below we present alternative methods of solution or the authors’
solution to some problems related to the Lalescu limit published in
several magazines around the world. We will present the problems
in their chronological order of publication.

Problem 1 (1989). Compute

‚ Œ
(n + 1)2 n2
lim p − √
n
.
n→∞ n+1
(n + 1)! n!
18 Arhimede Mathematical Journal

n −n
√
Solution. From Stirling’s approximation
√ n! ∼ n e 2πn, we
n −1
may deduce the approximation p n! ∼ ne because, when n
n
√
tends to +∞, the sequence 2πn tends to 1. On account of
the preceding, we have that
‚ Œ ‚ Œ
(n + 1)2 n2 (n + 1)2 n2
lim p − √ = lim −
n→∞ n+1
(n + 1)! n
n! n→∞ (n + 1)e−1 ne−1
= lim ((n + 1)e − ne) = e.
n→∞

Next, we state a result published in 1998 [10] that we will use in

the following problem.

Lemma 1. If the positive sequence {pn } is such that

pn+1
lim = p > 0,
n→∞ npn
then p
√ √
lim ( n+1
pn+1 − n
pn ) = .
n→∞ e
Problem 2 (2012). Let b > 0 be a real number. If {an }n≥1 is a se-
an+1
quence of positive reals such that limn→∞ 2 = b, then compute
n an
 n+1
√ √ ‹
an+1 n
an
lim − .
n→∞ n+1 n

Solution 1. Putting pn = an /nn in Lemma 1, we have

pn+1 an+1 nn+1

lim = lim ·
n→∞ npn n→∞(n + 1)n+1 n2 an
 ‹n+1
an+1 n b
= lim 2 · lim = .
n→∞ n an n→∞ n + 1 e
Finally, on account of Lemma 1, we get
 n+1
√ √ ‹
an+1 n
an b
lim − = 2.
n→∞ n+1 n e
Volume 7, No. 1, Spring 2020 19

√ √
n+1 an+1 n
an
Solution 2. Let us denote xn = −
. Then, we have
n+1 n
√ √ √ ‚ √ Œ
n+1 an+1 n
an n
an n+1 a
n+1 n
xn = − = √ · −1 .
n+1 n n n
an n+1
Now, we apply the D’Alembert criterium to compute
√ É
n
an an an+1 n2n
lim = lim n
= lim ·
n→∞ n2 n→∞ n2n n→∞ (n + 1)2n+2 an
2n
an+1 n b
= lim · = .
n→∞ an (n + 1)2n+2 e2
√
n+1 a
n+1 n
Let us define the sequence un = √ · for all n ≥ 2.
n
an n+1
We have that
‚ √ Œ
n+1 a
n+1 n2 n
lim un = lim · √ · = 1,
n→∞ n→∞ (n + 1)2 n an n + 1
from which it follows that
un − 1
lim = 1.
n→∞ ln un
On the other hand,
‚  ‹n Œ
an+1 1 n
lim un = lim · √ ·
n→∞ n n→∞ an n+1 a
n+1 n+1
‚ 2  ‹n  ‹2 Œ
an+1(n + 1) n n
= lim 2
· √ · · = e.
n→∞ n an n+1 an+1 n+1 n+1
Finally, we obtain that
 n+1
√ √ ‹
an+1 n
an
lim − = lim xn
n→∞ n+1 n n→∞
‚√ ‚ √ ŒŒ
n
an n+1 a
n+1 n
= lim √ · −1
n→∞ n n
an n+1
√ ‹
n
an un − 1
= lim · · ln un
n→∞ n ln un
√ ‹
n
an un − 1 n
b
= lim 2
· · ln un = 2 .
n→∞ n ln un e
20 Arhimede Mathematical Journal

Problem 3 (2012). Compute limn→∞ In (t), where for every n ≥ 2,

the function In : R∗ → R is defined by

1 € √ √ Š
In (t) = ((n + 1) n+1
n + 1)t − (n n
n)t .
nt−1

Solution. For all n ≥ 2, we have that

√
n √ un − 1
In (t) = n nt · (un − 1) = ( n
n)t · · ln un
n,
ln un
 ‹t ‚ √
n+1
Œt
n+1 n+1
where un = · √
n
. Since limn→∞ un = 1, then
n n
limn→∞ ulnnu−1
n
= 1, as can be easily computed. On the other hand,

 ‹nt ‚ n+1
√ Œnt
n+1 n+1
lim un
n = · √
n
n→∞ n n
 ‹t
t
n+1 1
= e · lim · n+1
√ = et .
n→∞ n n+1

Finally, we have that

lim In (t) = 1 · 1 · ln( lim un t

n ) = ln e = t.
n→∞ n→∞

Problem 4 (2012). Compute

 √ √ 
lim n+1
u1 u2 . . . un un+1 − u1 u2 . . . un ,
n
n→∞

(n+2)n+1
where {un }n≥1 is the sequence defined by un = (n+1)n
.

Solution. For all n ≥ 2, we define the sequence {vn }n≥2 by

√
n+1u1 u2 . . . un un+1
vn = √ .
n
u1 u2 . . . u n
Volume 7, No. 1, Spring 2020 21

Then, we have that

√ √ √
n+1
u1 u2 . . . un un+1 − n
u1 u2 . . . un = u1 u2 . . . un (vn − 1)
n

√ vn − 1
= n u1 u2 . . . un · ln vn
ln vn
√
n
u1 u2 . . . un vn − 1
= · ln vnn .
n ln vn

Since
√ É
n
u1 u2 . . . u n u1 u2 . . . un
lim = lim n
n→∞ n n→∞
 nn ‹
u1 u2 . . . un un+1 nn
= lim ·
n→∞ (n + 1)n+1 u1 u2 . . . un
un+1
= e−1 · lim
n→∞ n + 1

−1
(n + 3)n+2
= e · lim = 1,
n→∞ (n + 2)n+1 (n + 1)

then limn→∞ vn = 1 and also limn→∞ vlnn −1

vn
= 1 on account of Stolz
and D’Alembert’s results. Furthermore,
‚ Œ
u1 u2 . . . un un+1 1
lim vnn = lim · √
n→∞ n→∞ u1 u2 . . . u n n+1 u1 u2 . . . un un+1
‚ Œ
un+1 n+1
= lim · √ = e · 1 = e.
n→∞ n+1 n+1 u1 u2 . . . un un+1

Finally, combining the preceding, we get that

√ √

lim n+1
u1 u2 . . . un un+1 − n
u1 u2 . . . un = 1 · 1 · ln e = 1.
n→∞

Problem 5 (2016). Compute

‚ p p Œ
n+1
(n + 1)!(2n + 1)!! n
n!(2n − 1)!!
lim − .
n→∞ n+1 n
22 Arhimede Mathematical Journal

n!(2n−1)!!
Solution 1. Putting pn = nn
in Lemma 1, we have
 ‹n
pn+1 2n + 1 n 2
lim = lim · lim = .
n→∞ npn n→∞ n n→∞ n+1 e
Finally, on account of Lemma 1, we get
‚ n+1
p p Œ
(n + 1)!(2n + 1)!! n
n!(2n − 1)!! 2
lim − = 2.
n→∞ n+1 n e

Solution 2. Let us define the sequence

p
n+1
(n + 1)!(2n + 1)!! n
un = p ·
n
n!(2n − 1)!! n+1

for all n ≥ 2. We have

p
n+1
(n + 1)!(2n + 1)!! n
lim un = lim p · =1
n→∞ n→∞ n
n!(2n − 1)!! n+1

because, on account of the D’Alembert criterium,

p Ê
n
n!(2n − 1)!! n n!(2n − 1)!!
lim = lim
n→∞ n n→∞ nn
 ‹
(2n + 1)!! nn 2
= lim · = ,
n→∞ (n + 1)n+1 (2n − 1)!! e
and also
p p √
n
n
n!(2n − 1)!! n
(2n − 1)!! n! 2
lim = lim · lim = .
n→∞ n2 n→∞ n n→∞ n e2
un − 1
Furthermore, we have lim = 1, and
n→∞ ln un
‚  ‹n Œ
(2n+1)!!(n+1)! n 1
lim un = lim p
n→∞ n n→∞ (2n − 1)!!n! n+1 n+1
(2n+1)!!(n+1)!
‚ Œ
2 n+1
= lim p = e.
e n→∞ n+1
(2n + 1)!!(n + 1)!
Volume 7, No. 1, Spring 2020 23

Finally, we get that

‚ n+1
p p Œ
(n + 1)!(2n + 1)!! n
n!(2n − 1)!!
lim −
n→∞ n+1 n
‚p Œ
n
n!(2n − 1)!!
= lim · n · (un − 1)
n→∞ n2
2 un − 1 2
= 2 · lim · ln un
n = 2.
e n→∞ ln un e

The next problem was published in the Romanian mathematical

journal Revista de Matematica din Timisoara, founded by Traian
Lalescu in 1921.
Problem 6 (2019). Calculate
1 X
n
k
lim p .
n→∞ n k=2
k
(2k − 1)!!

Solution. Let x denote the above limit. On account of Stolz and

D’Alembert’s results, we have that
n+1
X X
n
k k
p − p
k=2
k
(2k − 1)!! k=2
k
(2k − 1)!!
x = lim
n→∞ (n + 1) − n
Ê
n+1 n nn
= lim p = lim p = lim n
n→∞ n+1 (2n + 1)!! n→∞ n (2n − 1)!! n→∞ (2n − 1)!!
 ‹
(n + 1)n+1
(2n − 1)!! n+1 n+1 n e
= lim · = lim = .
n→∞ (2n + 1)!! nn n→∞ 2n + 1 n 2

Problem 7 (2020). Compute

 √
n
lim sin γn − sin γ n!,
n→∞

X
n
1
where {γn }n≥1 is the sequence defined by γn = − ln n+ with
k=1
k
limn→∞ γn = γ (γ is the Euler-Mascheroni constant).
24 Arhimede Mathematical Journal

Solution. Using well-known trigonometry formulae, we have

√
n
√n γn − γ γn + γ
(sin γn − sin γ) n! = 2 n! · sin cos
√ 2 2
n γn −γ
n! γn + γ sin 2
= · cos · γn −γ · n · (γn − γ).
n 2 2
√
n
Since we have limn→∞ n
n!
= e−1 , limn→∞ cos γn2+γ = cos γ , and
γn −γ
sin
limn→∞ 2
γn −γ = 1, it only remains to compute limn→∞ n (γn − γ).
2
We claim that limn→∞ n (γn − γ) = 1/2. Indeed, applying Stolz’s
result, we have
γn+1 − γn γn − γn+1
lim n (γn − γ) = lim 1 1 = lim 1 1
n→∞ n→∞
n+1
− n
n→∞
n
− n+1

and
 ‹  ‹
n+1 1 2
n+1 1
lim n(n + 1) ln − = lim n ln −
n→∞ n n+1 n→∞ n n+1
 ‹
n+1 1
+ lim n ln − .
n→∞ n n+1
Since
 ‹ 1

x
2
n+1 1 ln 1 + x
+ ln x − x+1 1
lim n ln − = lim =
n→∞ n n+1 x→0
x>0
x2 2

and
 ‹ 1

x
n+1 1 ln 1 + x
+ ln x − x+1
lim n ln − = lim =0
n→∞ n n+1 x→0
x>0
x

on account of L’Hospital’s rule, then

 √ 1 cos γ
n
lim sin γn − sin γ n! = e−1 · cos γ · = .
n→∞ 2 2e
Volume 7, No. 1, Spring 2020 25

References
[1] Bătineţu-Giurgiu, D. M. “Sequences”. Albatros Publishing
House (1979).
[2] Bătineţu-Giurgiu, D. M. “Problem C:890”. Gazeta Matemătica,
Seria B XCIV (1989), p. 139.
[3] Bătineţu-Giurgiu, D. M. and Semenescu, A. “Sirul Lalescu la
o suta de ani”. Creativitate Matematica 9 (1999–2000), pp. 1–
14.
[4] Bătineţu-Giurgiu, D. M. and Stanciu, N. “Problem 26615”.
Gazeta Matemătica, Seria B CXVII.5 (2012), p. 269.
[5] Bătineţu-Giurgiu, D. M. and Stanciu, N. “Problem 5208”.
School Science and Mathematics 112.4 (2012), p. 265.
[6] Bătineţu-Giurgiu, D. M. and Stanciu, N. “Problem XI.133”.
Recreatii Matematice Journal XIV.2 (2012), p. 165.
[7] Bătineţu-Giurgiu, D. M. and Stanciu, N. “Problem 5398”.
School Science and Mathematics 116.4 (2016), p. 236.
[8] Bătineţu-Giurgiu, D. M. and Stanciu, N. “Problem XI.499”.
Revista de Matemătica din Timişoara Anul XXIV (Seria a IV-
a).1 (2019), p. 52.
[9] Bătineţu-Giurgiu, D. M. and Stanciu, N. “Problem UP.328”.
Romanian Mathematicel Magazine Anul XXIV (Seria a IV-
a).Fall (2021), p. 91.
[10] Toader, G. “Lalescu sequences”. Publikacije-Elektrotehnickog
Fakulteta Univerzitet U Beogradu Serija Matematika 9 (1998),
pp. 25–28.

Dumitru Bătineţu-Giurgiu Neculai Stanciu

“Matei Basarab”, National College “George Emil Palade” School
Bucharest Buzău
Romania Romania
[email protected] [email protected]
26 Arhimede Mathematical Journal

José Luis Díaz-Barrero

Civil and Environmental Engineering
BarcelonaTech
Barcelona, Spain
[email protected]
Volume 7, No. 1, Spring 2020 27

A New Bound for the Moduli

of the Zeros
José Luis Díaz-Barrero

Abstract

In this note, we give an explicit bound for the modules of the

zeros of a polynomial that involves binomial coefficients and
recursive numbers.

1 Introduction
Because of its importance, the problem of finding limits to the
modules of the zeros of a polynomial plays a central role in the
theory of equations. From the beginning, a lot of papers dedicated
to giving new limits or to improving the known classics were written.
In some of them, binomial coefficients appear as part of explicit
bounds for zeros [1]. However, integer sequences such as Fibonacci
or Lucas numbers have rarely appeared as part of the bounds for
the modules of the zeros [2, 3].
In this work, we determined a circular domain on the complex
plane containing all zeros of a polynomial where binomial coef-
ficients and numbers defined recursively as the numbers of Fi-
bonacci, Lucas or Pell appear.

2 Preliminary results
Hereafter, we present two preliminary results, a numerical identity
relating the terms of a recurrence and a classical theorem on the
location of zeros. We begin with the following lemma.
28 Arhimede Mathematical Journal

Lemma 1. Let n be a non negative integer. Then,

‚ Œ
X
n
n
3k dk = d2n ,
k=0
k

where dn is the n th term of the sequence defined recursively by

d0 = 0, d1 = 1 and, for n ≥ 2, dn = 3dn−1 + dn−2 .
Proof. Since the characteristic equation of the recursion is
x2 − 3x − 1 = 0,
then √
13
dn = (un − v n ),
13
√ √
3 + 13 3 − 13
where u = and v = are the roots of x2 −3x−1 =
2 2
0. Therefore,
√ √ h i
13 n n
13 n n
d2n = (u − v ) = (3u + 1) − (3v + 1)
13‚ Œ ‚ 13Œ
Xn Xn ‚ Œ
n k uk − v k n k
= 3 √ = 3 dk
k=0
k 13 k=0
k
and the lemma is proved.
Next we present a well-known classical result [4] that will be used
later on.
n−1
X
Theorem 1 (Cauchy, 1829). Let A(z) = z n + ak z k be a poly-
k=1
nomial with complex coefficients. Then, all its zeros lie in the disk
C = {z ∈ C : |z| ≤ r}, where r is the unique positive root of the
equation
B(z) = z n − |an−1 |z n−1 − . . . − |a1 |z − |a0 | = 0.

Proof. If |z| > r, then we have

B(|z|) = |z|n − |an−1 ||z|n−1 − · · · − |a1 ||z| − |a0 |
> r n − |an−1 |r n−1 − · · · − |a1 |r − |a0 | = 0.
Volume 7, No. 1, Spring 2020 29

Since

|An (z)| ≥ |z|n − (|an−1 ||z|n−1 + . . . + |a0 |) = B(|z|),

then we conclude that |An (z)| > 0. That is, An (z) 6= 0, for |z| > r.
Hence, all zeros of An (z) must lie in the disk |z| ≤ r. Observe
that the limit is attained when

A(z) = z n − |an−1 |z n−1 − . . . − |a1 |z − |a0 |.

3 The main result

Now we state and prove our main result.
n−1
X
Theorem 2. Let A(z) = z + n
ak z k be a polynomial with com-
k=1
plex coefficients. Then, all its zeros lie in the disk C = {z ∈ C :
|z| ≤ r}, where
(s )
d2n
r = max n
k |an−k | .
k
1≤k≤n
k
3 d k

Proof. It is known [4] that all the zeros of A(z) have modulus less
or equal than ξ, the unique positive root of the equation

B(z) = z n − |an−1 |z n−1 − · · · − |a1 |z − |a0 | = 0.

Therefore, our statement will be proved if we show that r ≥ ξ or,

equivalently, if we prove that B(r) ≥ 0.

From the statement, we have that

n
k
3 dk k
|an−k | ≤ k r , k = 1, 2, · · · , n.
d2n
30 Arhimede Mathematical Journal

Then,
X
n
B(r) = r n − |an−k | r n−k
k=1
n ¨ n
k « n
k
!
X 3 dk X
n
3 dk
≥ rn − k
rk r n−k = r n 1 − k
= 0.
k=1
d2n k=1
d2n

Notice that in the preceding inequality we have used the fact that
Xn n
k
k
3 dk
= 1,
k=1
d2n

as follows immediately from Lemma 1.

To illustrate the preceding theoretical result, we give a numerical
example. For instance, if we consider the polynomial A(z) =
z 3 + 0.01z 2 + 0.3z + 0.8, then it has all its zeros in the disk C =
{z ∈ C : |z| ≤ r}, where r ' 1.15. This bound is sharper than the
explicit bound of Cauchy

|z| < 1 + max an−k = 1.18.

1≤k≤n

References
[1] Birkhoff, G. D. “An elementary double inequality for the roots
of an algebraic equation having greatest absolute value”. Bull.
Amer. Math. Soc. 21.10 (1915), pp. 494–495. DOI: 10.1090/
S0002-9904-1915-02684-4.
[2] Díaz-Barrero, J. L. “Advanced Problem H-580”. The Fibonacci
Quarterly 39.5 (2001), pp. 474–474.
[3] Díaz-Barrero, J. L. “An annulus for the zeros of polynomials”.
J. Math. Anal. Appl. 273.2 (2002), pp. 349–352. ISSN: 0022-
247X. DOI: 10.1016/S0022-247X(02)00229-9.
[4] Marden, M. Geometry of Polynomials. 4th ed. American Ma-
thematical Society, 1989, pp. xiii+243. ISBN: 0-8218-1503-2.
Volume 7, No. 1, Spring 2020 31

José Luis Díaz-Barrero

Civil and Environmental Engineering
BarcelonaTech
Barcelona, Spain
[email protected]
32 Arhimede Mathematical Journal

Problems
This section of the Journal offers readers an opportunity to ex-
change interesting and elegant mathematical problems. Proposals
are always welcome. Please observe the following guidelines when
submitting proposals or solutions:
1. Proposals and solutions must be legible and should appear on
separate sheets, each indicating the name and address of the
sender. Drawings must be suitable for reproduction.
2. Proposals should be accompanied by solutions. An asterisk (*)
indicates that neither the proposer nor the editor has supplied
a solution.
Please, send submittals to José Luis Díaz-Barrero, Enginyeria
Civil i Ambiental, UPC BARCELONATECH, Jordi Girona 1-3, C2,
08034 Barcelona, Spain, or by e-mail to
[email protected]
The section is divided into four subsections: Elementary Problems,
Easy–Medium High School Problems, Medium–Hard High School
Problems, and Advanced Problems mainly for undergraduates.
Proposals that appeared in Math Contests around the world and
most appropriate for Math Olympiads training are always welcome.
The source of these proposals will appear when the solutions are
published.

Solutions to the problems stated in this issue should be posted

before
October 31, 2020
Volume 7, No. 1, Spring 2020 33

Elementary Problems
E–77. Proposed by Gabriel T. Prǎjiturǎ, SUNY Brockport, Brock-
port, NY, USA. Prove that in every triangle we have
È √ √
P ≥2 3 3 A,

where P is the perimeter and A is the area, with equality for the
case of the equilateral triangle.

E–78. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Barce-

lona, Spain. Let p, q, r be three prime numbers. Determine the
number of ordered pairs of positive integers (a, b) such that the
least common multiple of a and b is pq q r r p .

E–79. Proposed by Mihály Bencze, Braşov, Romania. Let a, x be

real numbers and let f, g : R → R be two functions defined by
f (x) = eax sin x and g(x) = eax cos x, respectively. Show that

(f (n) (x))2 + (g (n) (x))2 = (a2 + 1)n e2ax .

Here, f (n) (x) represents the n th derivative of f (x).

E–80. Proposed by Miguel Amengual Covas, Cala Figuera, Ma-

llorca, Spain. The lengths of the sides of a triangle are in arithmetic
progression. If a denotes the side having a medium length or
“middle” side, then with the usual notations, prove that

a2 ≥ 6Rr.

E–81. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Barce-

lona, Spain. Let a, b, c, d be four distinct nonzero real numbers in
harmonic progression. Prove that (a + d)(b + c) > 4bc.

E–82. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Barce-

lona, Spain. Prove that, from a set of fourteen distinct three-digit
numbers in the decimal system, it is possible to select two disjoint
subsets whose elements have the same sum.
34 Arhimede Mathematical Journal

Easy–Medium Problems
EM–77. Proposed by Toyesh Prakash Sharma, Class XI, St. C.F.
Andrews School, Agra, India. Show that there are infinitely many
integers n ≥ 1 for which
π 2π 3π nπ
sin 4
+ sin 4
+ sin 4
+ . . . + sin 4
π 2π 3π nπ
cos 4
+ cos 4
+ cos 4
+ . . . + cos 4

is a positive integer.

EM–78. Proposed by Mihaela Berindeanu, Bucharest, Romania.

Find all pairs of real numbers (x, y) such that x ≥ 8, y ≥ 7 and

(x + y)2
√ p = 30.
x2 − 64 + y 2 − 49

EM–79. Proposed by Andrés Sáez Schwedt, Universidad de León,

Spain. Let ABC be a triangle with circumcenter O, circumcircle Γ
and incircle ω . Let E and F be the tangency points of ω with AC
and AB . A circle Ω is internally tangent to Γ at A and externally
tangent to ω . Show that the radius of Ω is equal to
AO · AE · AF
.
AB · AC

EM–80. Proposed by Nicolae Papacu, Slobozia, Romania. De-

termine all positive integers x, y, z such that x < y < z and
xyz = ax + by + cz , where a, b, c ∈ {1, 2, 3}.

EM–81. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Bar-

celona, Spain. Let n ≥ 1 be an integer number. Determine the
coefficients ak (1 ≤ k ≤ 2n) such that the points of the z -plane
corresponding to the roots of the equation
z 2n+1 + a2n z 2n + a2n−1 z 2n−1 + . . . + a2 z 2 + a1 z = 0
represent the vertices of a regular (2n + 1)-gon with center at the
point z = 1.
Volume 7, No. 1, Spring 2020 35

EM–82. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Bar-

celona, Spain. Prove that there exist infinitely many positive inte-
ger powers of 19 ending up in 0001.
36 Arhimede Mathematical Journal

Medium–Hard Problems
MH–77. Proposed by Mihály Bencze, Brasov, Romania and Gabriel
T. Prǎjiturǎ, SUNY Brockport, Brockport, NY, USA. Let z be a nonzero
complex number. Prove that
§ ª
1
max |z|,
|z|
¨ 2
p p «
|z +1| + |z 2 +1|2 + 4|z|2 |z 2 −1| + |z 2 −1|2 + 4|z|2
≤ min , .
2|z| 2|z|

MH–78. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Bar-

celona, Spain. From point P inside equilateral triangle ABC , per-
pendiculars are drawn to the sides meeting BC , CA, and AB at
points D, E , and F , respectively. Compute the value of
DE + EF + F D
.
PA + PB + PC

MH–79. Proposed by Oriol Baeza Guasch, CFIS, BarcelonaTech,

Terrassa, Spain. Let ABC be a triangle, and let M be the mid-
point of arc BC not containing A. Let the circumference with
center M that goes through B , C meet rays AB , AC again at
points J 6= B , K 6= C , respectively. Finally, let the line J K
intersect the circumcircle of 4ABC at points P , Q. Show that
4AP Q is isosceles.

MH–80. Proposed by Mihaela Berindeanu, Bucharest, Romania.

Find all functions f : N∗ → N∗ such that there exists k ∈ N∗ such
that, for all positive integers x, y ,
f (x + y) + f (x) kx + f (y)
= .
ky + f (x) f (x + y) + f (y)

MH–81. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Bar-

celona, Spain. A group of 2021 students sit in a circle, consecu-
tively counting numbers 1, 2 and 3 and repeating. Starting from
Volume 7, No. 1, Spring 2020 37

some student A with number 1, and counting clockwise round

the remaining students, students that count numbers 2 and 3
must leave the circle until only one remains. Determine, after this
process, who is the last student remaining in the circle?

MH–82. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Bar-

celona, Spain. Let a, b, c be three positive numbers such that
ab + bc + ca = 3abc. Prove that

bc arctan2 a ca arctan2 b ab arctan2 c 3π 2

+ + ≥ .
b+c c+a a+b 32
38 Arhimede Mathematical Journal

Advanced Problems
A–77. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Barce-
lona, Spain. In the ring Z19 [X], find a monic polynomial of degree
four knowing that its zeros x1 , x2 , x3 , x4 satisfy
X
4 X
4 X
4 X
4
xi = 13, x2i = 1, x3i = 13, and x4i = 18.
i=1 i=1 i=1 i=1

A–78. Proposed by Mihály Bencze, Braşov, Romania. Let A be

an n × n matrix with real entries such that A3 + 4A2 + 6A + 3I = 0
and let B = 2A2 + 7A + 8I . Prove that B is invertible and compute
its determinant knowing that det A = 1/3.

A–79. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Barce-

lona, Spain. Let n ≥ 2 be a positive integer. Compute
Z 2
dx
 q È p ‹.
1 3 4
√
x 1 + x x x x... x n

A–80. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Barce-

lona, Spain. Let a < b be two positive real numbers and let X
be a random variable with support the r ≥ 2 points a1 , a2 , . . . , ar ,
lying on the closed interval [a, b], with probabilities p1 , p2 , . . . , pr ,
respectively. Prove that
b−a
σX ≤
.
2
When does equality occur? (Here σX represents the standard
deviation of random variable X .)

A–81. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Bar-

celona, Spain. Let A be an n × n real matrix such that 4A4 =
A3 + A2 + A + I , where I is the identity matrix. Let
lim Aj = B.
j→∞

Show that B 2 = B .
Volume 7, No. 1, Spring 2020 39

A–82. Proposed by Vasile Mircea Popa, “Lucian Blaga” University

of Sibiu, Romania. Calculate the following integral:
Z ∞
ln(x + 2)
2
dx.
−1 x + 3x + 3
40 Arhimede Mathematical Journal

Mathlessons
This section of the Journal offers readers an opportunity to ex-
change interesting and elegant mathematical notes and lessons
with material useful to solve mathematical problems.

Please, send submittals to José Luis Díaz-Barrero, Enginyeria

Civil i Ambiental, UPC BARCELONATECH, Jordi Girona 1-3, C2,
08034 Barcelona, Spain, or by e-mail to
[email protected]
Volume 7, No. 1, Spring 2020 41

Seventeen Camels Problem

O. Rivero Salgado and J. L. Díaz-Barrero

1 Introduction
There is a very well-known and funny story that poses the problem
of dividing a property of 17 camels between the three sons of their
owner. This story originated many centuries ago and, although
it has been transmitted orally from generation to generation in
the various versions, the basic idea of the story remains the same.
Below, we present one version of this well-known story.

A long time ago, an old man lived with his three children in a de-
serted village, renting camels as a means of transportation in the
desert. He had 17 camels, and they were the main source of his
income. When he died, he left in his will the way to divide his prop-
erty among his three children. When the three children read the will,
they had knowledge of how their father had divided all the property
he had into three equal parts, while he had decided to distribute the
17 camels in a different way. The old man had ruled that the eldest
son would own half of the 17 camels, the middle son would get a
third of the 17 camels, and the youngest son would get his share
of the camels as a ninth! Everyone was astonished as they read
the will and wondered how to divide the 17 camels as mentioned
in the will. It is not possible to divide 17 camels and give half of
the 17 camels to the eldest and it is not possible to also divide the
camels for the other two children. They spent several days thinking
of ways to divide the camels as mentioned in the will, but none of
them could find the answer.
42 Arhimede Mathematical Journal

2 Camels problem
Below, we present the story stated in a problem form, give a
solution and generalize it.

Problem. An old man dies and leaves seventeen camels to be di-

vided among his three sons in the proportion 21 : 13 : 19 . May the
brothers carry out the father’s will?

Solution. To avoid calculating with fractions, a village wise man

conceived of an artificial method by which a camel is borrowed. He
then divided and obtained 18 : 2 = 9, 18 : 3 = 6, 18 : 9 = 2,
which meant that the sons received 9, 6 and 2 camels, for a total
of 17 camels. After this division, the borrowed camel was returned
to its owner and the problem was solved.

However, was the wise man’s trick an acceptable solution according

to the terms of the will? In fact, no, the division was not done in a
proper way and it is not possible. Indeed, since

1 1 1 17
+ + = < 1,
2 3 9 18

any division of 17 by 2, 3 or 9 gives a remainder and, due to the

borrowed camel, each of the brothers receives more than what the
will stated, since 9 > 17/2, 6 > 17/3 and 2 > 17/9.

In general, the trick of the wise man may be implemented when

n camels must be distributed into three brothers in proportions
1
a
: 1b : 1c . The preceding is equivalent to solving the diophantine
equation
1 1 1 n 1
+ + = =1− ,
a b c n+1 n+1
with the additional restriction that all a, b and c divide n + 1 (so in
particular n + 1 ≥ a, b, c). This excludes, for instance, a solution
like
1 1 1 14
+ + = .
2 3 10 15
Volume 7, No. 1, Spring 2020 43

We may write n + 1 = d, and we begin by finding all positive integer

solutions to the equation
1 1 1 1
+ + + = 1.
a b c d
We can assume, without loss of generality, that 2 ≤ a ≤ b ≤ c ≤ d.
It is clear that a ≤ 4 and, moreover, if a = 4, then a = b = c =
d = 4. To see this, note that
1 1 1 1 4
1= + + + ≤ ,
a b c d a
so a ≤ 4 and equality holds if and only if all four numbers are
equal to 4. Hence, there are two main cases to discuss, a = 2 and
a = 3.

Case 1: a = 2. We have now the equation

1 1 1 1
+ + = ,
b c d 2
with b ≥ 3. Again,
1 1 1 1 3
= + ,+ ≤
2 b c d b
so b ≤ 6 and equality holds if and only if b = c = d = 6. This
leads us to the study of three subcases: b = 3, b = 4, and b = 5.

1.1: b = 3. This is the more tedious option, since we still have

several options, since
1 1 1
+ = ,
c d 6
with c > 6. The same analysis as before shows that c ≤ 12.
Since
6c 36
d= =6+ ,
c−6 c−6
we have to consider those values of c ∈ (6, 12] such that c − 6
divides 36. The options for the pairs (c, d) are then (7, 42),
(8, 24), (9, 18), (10, 15), and (12, 12).
44 Arhimede Mathematical Journal

1.2: b = 4. We now have

1 1 1
+ = ,
c d 4
with c ∈ (4, 8]. Now,

4c 16
d= =4+ ,
c−4 c−4

whose solutions are (5, 20), (6, 12), and (8, 8).
1.3: b = 5. The equation

1 1 3
+ = ,
c d 10
with c, d ≥ 5, necessarily satisfies c ∈ {5, 6}. If c = 5, then
d = 10. When c = 6, the value of d is not an integer number.

Case 2: a = 3. We have the equation

1 1 1 2
+ + = ,
b c d 3
with b ≥ 3. Again,
2 1 1 1 3
= + + ≤ ,
3 b c d b
so b ∈ {3, 4} and we have two subcases: b = 3 and b = 4.

2.1: b = 3. Now we have

1 1 1
+ = ,
c d 3
so c ∈ {4, 5, 6}. Since

3c 9
d= =3+ ,
c−3 c−3

we have that the possibilities for the pair (c, d) are (4, 12) and
(6, 6).
Volume 7, No. 1, Spring 2020 45

2.2: b = 4. We now have

1 1 5
+ = ,
c d 12
with c ≥ 4, and c ≤ 24/5. We only have to analyze the case
c = 4, and in this case d = 6.

All in all, the possible solutions for the equation are the following
ones.

• (2, 3, 7, 42). Ok:

1 1 1 41
+
. + =
2 3 7 42
Observe that the cases corresponding to
1 1 1 1
+ + =
2 3 42 7
also work but with a different interpretation. Here, we begin
with 36 camels and the extra person gives 6. Then, the sons
take 21, 14, and 1, respectively, in such a way that the extra
person recovers the 6 camels. This serves to give a rough
explanation to the condition n + 1 ≥ a, b, c.
• (2, 3, 8, 24). Ok:
1 1 1 23
+ + = .
2 3 8 24
• (2, 3, 9, 18). Ok:
1 1 1 17
+ + = (Camels Problem).
2 3 9 18
• (2, 3, 10, 15). It does not work since 10 does not divide 15. In
this case, the situation would require to have 28 camels and
the extra person giving 2 camels, in such a way that the first
son takes 15, the second takes 10, and the third one takes
3 (so finally, the extra person recovers the 2 camels). This
explains the condition that n + 1 divides both a, b and c.
• (2, 3, 12, 12). Ok:
1 1 1 11
+ + = .
2 3 12 12
46 Arhimede Mathematical Journal

• (2, 4, 5, 20). Ok:

1 1 1 19
+ + = .
2 4 5 20
• (2, 4, 6, 12). Ok:
1 1 1 11
+ + = .
2 4 6 12
• (2, 4, 8, 8). Ok:
1 1 1 7
+ + = .
2 4 8 8
• (2, 5, 5, 10). Ok:
1 1 1 9
+ + = .
2 5 5 10
• (2, 6, 6, 6). Ok:
1 1 1 5
+ + = .
2 6 6 6
• (3, 3, 4, 12). Ok:
1 1 1 11
+ + = .
3 3 4 12
• (3, 3, 6, 6). Ok:
1 1 1 5
+ + = .
3 3 6 6
• (3, 4, 4, 6). Ok:
1 1 1 5
+ + = .
3 4 4 6
• (4, 4, 4, 4). Ok:
1 1 1 3
+ + = .
4 4 4 4
We conclude that there are thirteen solutions, of which only seven
split the number n into three parts with distinct number of ele-
ments.

References
[1] Dudeney, H. E. 536 Puzzles & Curious Problems. Charles
Scribner’s Sons, New York, 1967.
Volume 7, No. 1, Spring 2020 47

[2] Fontana ‘Tartaglia’, N. Questi et invenzioni diverse. Venetia,

1546.
[3] Fontana ‘Tartaglia’, N. Il general trattato de’ numeri e misure.
Curtio Troiano, Venetia, 1556-1560.

Óscar Rivero Salgado

Mathematics and Statistics Faculty
BarcelonaTech
Barcelona, Spain
[email protected]

José Luis Díaz-Barrero

Civil and Environmental Engineering
BarcelonaTech
Barcelona, Spain
[email protected]
48 Arhimede Mathematical Journal

Contests
In this section, the Journal offers sets of problems appeared in
different mathematical contests over the world, as well as their
solutions. This gives readers an opportunity to find interesting
problems and develop their own solutions.

No problem is permanently closed. We will be very pleased to

consider new solutions to problems posted in this section for pub-
lication. Please, send submittals to José Luis Díaz-Barrero, En-
ginyeria Civil i Ambiental, UPC BARCELONATECH, Jordi Girona
1-3, C2, 08034 Barcelona, Spain, or by e-mail to
[email protected]
Volume 7, No. 1, Spring 2020 49

2nd Barcelona Spring

Math-Olympiad

O. Rivero Salgado and J. L. Díaz-Barrero

1 Introduction
The 2nd edition of the Barcelona Spring Math-Olympiad took
place in April 2020 online. About 60 contestants from all the
autonomous communities of Spain participated in this edition of
the competition.

2 Problems and solutions

We now present the problems that appeared in the paper and their
solutions.

Problem 1. Find all integer numbers a1 , a2 , a3 such that

a21 < 3a1 − a2 , a22 < 3a2 − a3 , a23 < 3a3 − a1 .

Solution. First, we observe that for all x, y ∈ Z we have

x < y ⇐⇒ x + 1 ≤ y.

Then, solving in integers the given system is equivalent to solve

a21 + 1 ≤ 3a1 − a2 ,
a22 + 1 ≤ 3a2 − a3 ,
a23 + 1 ≤ 3a3 − a1 ,
50 Arhimede Mathematical Journal

or
a21 − 2a1 + 1 ≤ a1 − a2 ,
a22 − 2a2 + 1 ≤ a2 − a3 ,
a23 − 2a3 + 1 ≤ a3 − a1 .
Adding up the preceding yields
(a1 − 1)2 + (a2 − 1)2 + (a3 − 1)2 ≤ 0,
from which a1 = a2 = a3 = 1 follows. So, the only solution in
integers of the given system of inequalities is 1, 1, 1, and we are
done.

Problem 2. Boat 1 and boat 2 depart at the same time from

docks A and B , respectively, on the bank of a round lake. If
they made their way straight to docks C and D, respectively, they
would collide. Prove that if boat 1 goes straight to D and boat 2
goes straight to C , they will arrive at the same time.
Solution. Let M be the point where boat 1 and boat 2 collide.

M C

Figure 1: Scheme for solving problem 2.

By the statement of the problem, it would take the same time t

for boat 1 to cover the distance AM and for boat 2 to cover BM .
That is,
AM BM
t= = .
v1 v2
Volume 7, No. 1, Spring 2020 51

AD AD
So the time for boat 1 to travel from A to D is t1 = = t,
v1 AM
BC BC
and the time for boat 2 to move from B to C is t2 = = t
v2 BM
from which it follows that
AM BM
t1 = t2 .
AD BC
But triangles AM D and BM C are similar and we have
AM BM
= ,
AD BC
that jointly with the preceding implies that t1 = t2 , and we are
done.

Problem 3. The set S consists of 2021 positive integers, none

of which has a prime divisor larger than 24. Prove that S has four
elements, the product of which is equal to the fourth power of an
integer.
Solution. The elements of S can have nine different prime divi-
sors,
2, 3, 5, 7, 11, 13, 17, 19, 23.
Let us classify the elements of S just by the parity of the exponents
of the nine possible prime divisors in them. This classification
creates just 29 = 512 classes. Now pick two elements of S that
are in the same class, and remove them from S . Put their product
into a new set T . This procedure clearly decreased the size of S by
2. Then repeat this same procedure, that is, pick two elements of
S that are in the same class, remove them, and put their product
into T . Note that all elements of T will be squares as they will
contain all their prime divisors with even exponents. Do this until
you can, that is, until there are no two elements of S in the same
class. Stop when that happens. Then S has at most 511 elements
left, so we have removed at least 1500 elements from S . Therefore
T has at least 750 elements, all of which are squares of integers.
Now classify the elements of T according to the remainders of the
exponents of their prime divisors modulo four. As the elements
52 Arhimede Mathematical Journal

of T are all squares, all these exponents are even numbers, so

their remainders modulo four are either 0 or 2. So again, this
classification creates only 512 classes, and therefore, there will be
two elements of T in the same class, say u and v . Then uv is the
fourth power of an integer, and since both u and v are products of
two integers in S , our claim is proved.

Problem 4. Show that there exists a finite set A ⊂ R2 such that

for every P ∈ A there are points Y1 , Y2 , . . . , Y2020 in A such that
the distance between P and Yi is equal to one, for 1 ≤ i ≤ 2020.
Solution. Consider the set
¨‚ Œ «
m2 − 1 2m
X= ± 2 ,± 2 , 1 ≤ m ≤ 2020, m ∈ Z .
m +1 m +1
This set is contained in the unit circle, is finite and has 2020 points
in each quadrant of the unit circle. Indeed, if (a, b) is a point in
X , then
‚ Œ2  ‹2
2 2
m2 − 1 2m
a +b = ± + ± = 1,
m2 + 1 m2 + 1
|X| = 8080 so it is finite and it has 2020 points in each quadrant.

2
P
Yi

O 2
Vi

Figure 2: Scheme for solving problem 4.

Let
Y
2020
q= (m2 + 1),
m=1
Volume 7, No. 1, Spring 2020 53

and consider the set

§ ª
1 2 3 2q
B= 0, , , , . . . , .
q q q q

We claim that  A= ‹ B × B gives a solution for the problem. Indeed,

i j
given P = , ∈ B × B there is a quadrant of the unit circle
q q
centered at P that is contained in [0, 2] × [0, 2]. If we denote by
V1 , V2 , . . . , V2020 the 2020 distinct points of X lying in the same
quadrant of the unit circle, the points Y1 = P + V1 , Y2 = P +
V2 , . . . , Y2020 = P + V2020 are all in B × B , because
§ ‹ ª
r s
B × B = ([0, 2] × [0, 2]) ∩ , , r, s ∈ Z ,
q q

and the distance between P and Yi = P +Vi is 1, for 1 ≤ i ≤ 2020.

Observe that q is the product of all denominators that appear in

the points of X .

Óscar Rivero Salgado

Department of Mathematics
BarcelonaTech
Barcelona, Spain
[email protected]

José Luis Díaz-Barrero

Civil and Environmental Engineering
BarcelonaTech
Barcelona, Spain
[email protected]
54 Arhimede Mathematical Journal

Solutions
No problem is ever permanently closed. We will be very pleased to
consider new solutions or comments on past problems for publica-
tion.

Please, send submittals to José Luis Díaz-Barrero, Enginyeria

Civil i Ambiental, UPC BARCELONATECH, Jordi Girona 1-3, C2,
08034 Barcelona, Spain, or by e-mail to
[email protected]

Elementary Problems
E–71. Proposed by Marc Felipe i Alsina and José Luis Díaz-Ba-
rrero, BarcelonaTech, Barcelona, Spain. Let Fn be the n-th Fi-
bonacci number, defined by F0 = 0, F1 = 1 and Fn = Fn−1 + Fn−2
for all n ≥ 2. Is the number F370370367 odd or even?

Solution 1 by Henry Ricardo, Westchester Area Math Circle,

NY, USA. We have n = 370370367 = 9(41152263). Since it is
known that m | n implies Fm | Fn for positive integers m, n, we
see that F9 | Fn . But F9 = 34, which tells us that Fn is even.

Solution 2 by Michel Bataille, Rouen, France. First, we prove

by induction that, for any nonnegative integer n, F3n is even and
F3n+1 , F3n+2 are odd. Since F2 = 1, we see that F0 is even while
F1 and F2 are odd. Assume that F3k is even and F3k+1 , F3k+2
are odd for some nonnegative integer k. Then, we obtain that
F3(k+1) = F3k+3 = F3k+2 + F3k+1 is even as the sum of two odd
integers. Then, similarly, F3(k+1)+1 = F3k+4 = F3k+3 + F3k+2 is odd
Volume 7, No. 1, Spring 2020 55

and F3(k+1)+2 = F3k+5 = F3k+4 + F3k+3 is odd. This completes the

induction step and the proof.
In particular, since 370370367 = 3 × 123456789, we deduce that
F370370367 is even.
Solution 3 by Víctor Martín Chabrera, FME, BarcelonaTech,
Barcelona. We see that for all n ≥ 3 we have

Fn = Fn−1 + Fn−2 = Fn−2 + Fn−3 + Fn−2 ≡ Fn−3 (mod 2),

and therefore, it is clear that if a ≡ b (mod 3), then Fa ≡ Fb

(mod 2). As 370370367 ≡ 0 (mod 3), we have F370370367 ≡ F0 = 0
(mod 2), and therefore F370370367 is even.

Also solved by Miguel Amengual Covas, Cala Figuera, Mallorca,

Spain; Scott H. Brown, Auburn University Montgomery, Montgomery
AL; Alberto Espuny Díaz, University of Birmingham, Birmingham,
United Kingdom; Irene Fernández Fernández, I.E.S. El Carmen,
Murcia, Spain; Ángel Plaza, University of Las Palmas de Gran Ca-
naria, Spain, and the proposers.

E–72. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Barce-

lona, Spain. Find all real solutions of the equation

11x + 60x = 61x .

Solution 1 by Rovsen Pirkuliyev, Sumgayit City, Azerbaijan.

 ‹x  ‹x  ‹2  ‹2
11 60 11 60
We have + = 1. Since + = 1, then
61 61 61 61
11 60
there exists and angle α such that sin α = and cos α = .
61 61
Then, we have
(sin α)x + (cos α)x = 1.
Now we distinguish two cases:
• If x < 2, then (sin α)x > sin2 α, (cos α)x > cos2 α and

(sin α)x + (cos α)x > sin2 α + cos2 α = 1.

(Contradiction).
56 Arhimede Mathematical Journal

• If x > 2, then (sin α)x < sin2 α, (cos α)x < cos2 α and

(sin α)x + (cos α)x < sin2 α + cos2 α = 1.

(Contradiction).

Hence, x = 2 is the only root of the given equation.

Solution 2 by the proposer. Since (11, 60, 61) is a pythagorean

triple, then x = 2 is a solution. Now, we will see that it is unique.
Indeed, dividing both sides by 61x we get
 ‹x  ‹x
11 60
+ = 1.
61 61

Consider the functions f, g : R → R+ defined by

 ‹x  ‹x
11 60
f (x) = and g(x) = .
61 61

Both are strictly decreasing and, therefore, f + g is also strictly

decreasing. If x > 2, then we have (f + g)(x) < (f + g)(2) or
 ‹x  ‹x  ‹2  ‹2
11 60 11 60
+ < + = 1,
61 61 61 61

and the equation does not have solutions greater than 2.

On the other hand, if x < 2, then (f + g)(x) > (f + g)(2). That

is,
 ‹x  ‹x  ‹2  ‹2
11 60 11 60
+ > + = 1,
61 61 61 61
and the equation has no solutions less than 2. Therefore, the only
solution is x = 2, and we are done.

Also solved by Víctor Martín Chabrera, FME, BarcelonaTech, Bar-

celona, Spain, and Rovsen Pirkuliyev, Sumgayit City, Azerbaijan
(one more solution).
Volume 7, No. 1, Spring 2020 57

E–73. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Barce-

lona, Spain. Let x, y, z be positive real numbers. Prove that
É É Ê
xy yz zx
6
+ 6
+ 6
≥ 3.
z2 x2 y2
Solution 1 by Irene Fernández Fernández, I.E.S. El Carmen,
Murcia, Spain. Using the arithmetic-geometric mean inequality,
we have
É É Ê sÉ É Ê
xy yz zx xy yz zx
6
2
+ 6 2 + 6 2 ≥33 6 2 · 6 2 · 6 2
z x y z x y
Ê Ê
xy · yz · zx 18
z 2 · x2 · y 2
=3 18
=3 = 3.
z 2 · x2 · y 2 z 2 · x2 · y 2
Due to the properties of the arithmetic-geometric mean,
É É Ê
xy yz zx
6
+ 6
+ 6
=3
z2 x2 y2
È È È
when 6 xy
z2
= 6 yzx2
= 6 zx
y2
, which implies that x = y = z . Hence,
the inequality is true for all x, y, z > 0.
Solution 2 by the proposer. Taking into account the well-known
inequalities between arithmetic, geometric and harmonic means
(AM-GM-HM), we have
√ √ √ È
x yz + y zx + z xy 3
p p
≥ xyz x2 y 2 z 2 = 3 x2 y 2 z 2
3
3
≥ .
1 1 1
√ + √ + √
x yz y zx z xy
From the preceding, it immediately follows that
‚ Œ
1 1 1 3
√ + √ + √ ≥ p
x yz y zx z xy 3
x2 y 2 z 2
and ‚ Œ
1 1 1 p
3
√ + √ + √ x2 y 2 z 2 ≥ 3
x yz y zx z xy
58 Arhimede Mathematical Journal

or, equivalently,
√ √
6
√
6
yz zx 6
xy
√3
+ √ + √3
≥ 3.
x 3
y z

Notice that equality holds when x = y = z and the proof is

complete.

Also solved by Michel Bataille, Rouen, France; Víctor Martín Cha-

brera, FME, BarcelonaTech, Barcelona, Spain; Rovsen Pirkuliyev,
Sumgayit City, Azerbaijan; Ángel Plaza, University of Las Palmas
de Gran Canaria, Spain, and Henry Ricardo, Westchester Area
Math Circle, NY, USA.

E–74. Proposed by Marc Felipe i Alsina and José Luis Díaz-Ba-

rrero, BarcelonaTech, Barcelona, Spain. Let ABC be a triangle.
With the usual notations, show that

a2 + sin2 A b2 + sin2 B c2 + sin2 C

+ +
a + sin A b + sin B c + sin C
(a + b + c) + (sin A + sin B + sin C)2
2
= .
a + b + c + sin A + sin B + sin C

Solution 1 by Michel Bataille, Rouen, France. Let R be the

circumradius of the triangle ABC . The Law of Sines gives

a b c
= = = 2R.
sin A sin B sin C
We deduce that

x2 + sin2 X 4R2 sin2 X + sin2 X 4R2 + 1

= = · sin X,
x + sin X 2R sin X + sin X 2R + 1

where (x, X) can be either of (a, A), (b, B), (c, C), and it follows
that the left-hand side L of the equality is

4R2 + 1
L= (sin A + sin B + sin C).
2R + 1
Volume 7, No. 1, Spring 2020 59

In a similar way, the right-hand side M is

4R2 (sin A + sin B + sin C)2 + (sin A + sin B + sin C)2
M =
2R(sin A + sin B + sin C) + sin A + sin B + sin C
4R2 + 1
= (sin A + sin B + sin C),
2R + 1
and so L = M , as expected.
Solution 2 by Henry Ricardo, Westchester Area Math Circle,
NY, USA. We have
X a2 + sin2 A X a2 X sin2 A
= + = S1 + S2 .
cyclic
a + sin A cyclic
a + sin A cyclic
a + sin A

The Cauchy-Schwarz inequality gives us

€P Š2
cyclic a (a + b + c)2
S1 ≥ P = ,
cyclic (a + sin A) a + b + c + sin A + sin B + sin C

with equality if and only if a/(a + sin A) = b/(b + sin B) = c/(c +

sin C), or 1/(1 + sin A/a) = 1/(1 + sin B/b) = 1/(1 + sin C/c),
which is true by the Law of Sines.
Similarly,
(sin A + sin B + sin C)2
S2 ≥ ,
a + b + c + sin A + sin B + sin C
with equality if and only if sin A/(a + sin A) = sin B/(b + sin B) =
sin C/(c + sin C), or 1/(1 + a/ sin A) = 1/(1 + b/ sin B) = 1/(1 +
c/ sin C), which is again true by the Law of Sines.
Therefore,
X a2 + sin2 A (a + b + c)2
=
cyclic
a + sin A a + b + c + sin A + sin B + sin C
(sin A + sin B + sin C)2
+
a + b + c + sin A + sin B + sin C
(a + b + c)2 + (sin A + sin B + sin C)2
= .
a + b + c + sin A + sin B + sin C
60 Arhimede Mathematical Journal

Also solved by Miguel Amengual Covas, Cala Figuera, Mallorca,

Spain; Irene Fernández Fernández, I.E.S. El Carmen, Murcia, Spain;
Víctor Martín Chabrera, FME, BarcelonaTech, Barcelona, Spain;
Daniel Văcaru, Piteşti, Romania, and the proposers.

E–75. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Bar-

celona, Spain. Let ABC and ABD be two congruent triangles.
Vertices C and D lie on opposite sides of AB . If CD meets AB
in X , then prove that XC = XD.
Solution 1 by Miguel Amengual Covas, Cala Figuera, Mallorca,
Spain. We distinguish two cases, illustrated in figures A and B of
figure 1.
1. CA = AD (figure A). Then, BC = BD. Thus, AB is the
perpendicular bisector of segment CD, making right-angled
triangles AXC and AXD congruent with CX = XD.
2. CA = BD (figure B). Then, ∠CAB = ∠ABD. Thus, opposite
sides CA and BD of quadrilateral ADBC are equal and
parallel, making it a parallelogram. Because of the theorem
“the diagonals of a parallelogram bisect each other”, we have
CX = XD.
C C

A
X B A B
X

D Figure A Figure B D

Figure 1: Construction for both cases of Solution 1 of Problem

E–75.

Solution 2 by Irene Fernández Fernández, I.E.S. El Carmen,

Murcia, Spain. To solve the problem we consider two cases.
Case 1. As the opposite sides of the quadrilateral are equal and
parallel, and its opposite angles are equal, the congruent triangles
Volume 7, No. 1, Spring 2020 61

form a parallelogram. AB is one of the parallelogram’s diagonals,

and CD is the other diagonal, which intersects AB at X . Due to
a property of parallelograms, the diagonals cut themselves in two
halves, so CX = DX .

Figure 2: Two congruent triangles with common side AB , CB =

AD, AC = DB .

Case 2. Segment CD is composed of the heights of 4ABC and

4ABD over side AB , which form right angles with this segment
and intersect AB at X . 4ABC and 4ABD are congruent, so
their heights are equal. Therefore, as CX , DX are the triangles’
heights, segments CX and DX are equal.

Figure 3: Two congruent triangles with common side AB , CB =

BD, AC = AD.
62 Arhimede Mathematical Journal

Solution 3 by the proposer. Construct by A and B parallel

lines to BD and AD, respectively. Let us denote by E be its
intersection point.

C
E

A B
X Y

Figure 4: Construction for Solution 3 of Problem E–75.

Since AEBD is a parallelogram, then we have

4AEB = 4ADB = 4ACB

and AB k CE . Let X and Y be the intersection points of CD and

DE with the side AB . Since Y is the intersection of diagonals of
AEDB then it is the mid point of AB . Finally, we have in triangle
CDE that XY is a parallel to CE passing by the midpoint of DE
and, therefore, it meets CD in its midpoint X . So, XC = XD as
we wanted to prove.

Also solved by Rovsen Pirkuliyev, Sumgayit City, Azerbaijan.

E–76. Proposed by Marc Felipe i Alsina and José Luis Díaz-Ba-

rrero, BarcelonaTech, Barcelona, Spain. Let n be a positive integer.
Compute the following sum:
‚ Œ2 ‚ Œ2 ‚ Œ2 ‚ Œ2 ‚ Œ2
n n n n n
− + − + . . . + (−1)n .
0 1 2 3 n
Volume 7, No. 1, Spring 2020 63

Solution 1 by Michel Bataille, Rouen, France. Let Sn be the

sum to be computed. From the binomial theorem, we have
‚ Œ
X
n
n
(1 + x)n = xk
k=0
k

and ‚ Œ
X
n
n
n
(1 − x) = (−1)k xk .
k=0
k
Therefore, the coefficient of xn when expanding the product (1 +
x)n (1 − x)n is
‚ Œ‚ Œ ‚ Œ‚ Œ
X
n
n n Xn
n n
n−k
(−1) = (−1)j
k=0
k n−k j=0
j n−j
‚ Œ2
Xn
n
= (−1)j = Sn
j=0
j

n
n

(using n−j
= j
when 0 ≤ j ≤ n).

On the other hand, since (1 + x)n (1 − x)n = (1 − x2 )n , this

coefficient is also the coefficient of xn in
‚ Œ
X
n
n
(−1)k x2k .
k=0
k

By comparison, it clearly appears that

Sn = 0 (n odd)

and ‚ Œ
n
Sn = (−1)n/2 (n even).
n/2

Solution 2 by Henry Ricardo, Westchester Area‚MathŒ Circle,

n
NY, USA. We show that the sum equals (−1)n/2 if n is
n/2
even and 0 if n is odd.
64 Arhimede Mathematical Journal

We have ‚ Œ2 ‚ Œ‚ Œ
X
n
n X
n
n n
(−1)k = (−1)k ,
k=0
k k=0
k n−k
which is the coefficient of xn in
‚ Œ ‚ Œ
X
n
n X
n
n
2 n n n k k
(1 − x ) = (1 − x) (1 + x) = (−1) x · xk .
k=0
k k=0
k

If n is even, let n = 2m

and y = x2 .
Then, the coefficient of y m in
n n
(1 − y)2m is (−1)m m = (−1)n/2 n/2 . However, (1 − x2 )n has only
terms of even degree, so if n is odd, then clearly the coefficient of
xn is 0.
Solution 3 by Víctor Martín Chabrera, FME, BarcelonaTech,
P n
2
Barcelona, Spain. Let Sn = n k
k=0 k (−1) . Let us see that, if n
is odd, Sn = 0. Taking into account that (−1)n = −1,
‚ Œ2 ‚ Œ2
X
n
n X
n
n
−Sn = − (−1)k = (−1)n (−1)k .
k=0
k k=0
k
Taking j = n − k, we can rewrite this as
‚ Œ2 ‚ Œ2
X
n
n X
n
n
n k
(−1) (−1) = (−1)n (−1)n−j
k=0
k j=0
n−j
‚ Œ2
X
n
n
= (−1)j = Sn .
j=0
j

Since Sn = −Sn , Sn = 0.
Now, let us see what happens for n even. Let f (x) = (1 − x2 )n .
Let us rewrite f (x) in two different ways. On the one hand, we
n ‚ Œ
have
X n
2 n
f (x) = (1 − x ) = (−x2 )k .
k=0
k
On the other,
‚ Œ ! ‚ Œ !
X
n
n X
n
n
f (x) = (1−x2 )n = (1+x)n (1−x)n = xi (−x)j .
i=0
i j=0
j
Volume 7, No. 1, Spring 2020 65

We can therefore compute the term of degree n in two different

ways. On the one hand, from the first expression, taking into
account that n is even, we have
‚ Œ
n
[xn ]f (x) = (−1)n/2 .
n/2

On the other, from the second expression we have

‚ Œ‚ Œ ‚ Œ‚ Œ
X n n X
n
n n
n j
[x ]f (x) = (−1) = (−1)n−i
i+j=n
i j i=0
i n−i
‚ Œ‚ Œ ‚ Œ2
X
n
n n X
n
n
n i
= (−1) (−1) = (−1)i = Sn .
i=0
i i i=0
i

Hence, we conclude that

¨ n

n/2
(−1)n/2 if n ≡ 0 (mod 2),
Sn =
0 if n ≡ 1 (mod 2).

Also solved by Ángel Plaza, University of Las Palmas de Gran Ca-

naria, Spain; Daniel Văcaru, Pites̆ti, Romania, and the proposers.
66 Arhimede Mathematical Journal

Easy–Medium Problems
EM–71. Proposed by Oriol Baeza Guasch, CFIS, BarcelonaTech,
Terrassa, Spain. Let ABC be a triangle with orthocenter H . Let
M be the midpoint of arc BC , and let P be the reflection of H
over side BC . If M P is perpendicular to HB , find the value of
∠ABC
.
∠BCA
Solution by Miguel Amengual Covas, Cala Figuera, Mallorca,
Spain. If an altitude of a triangle is extended to the circumcircle,
the extension has the same length as the part between the ortho-
center and the foot of the altitude. Because of this, the points A,
P , M , and C lie on the circumcircle of 4ABC .

B C

Figure 5: Scheme for the solution of Problem EM–71.

Chords M P and CA are parallel since both are perpendicular

to BH (BH is an altitude), implying that AP M C is an isosceles
Volume 7, No. 1, Spring 2020 67

trapezium with
∠CAP = ∠M CA. (1)

Since
∠CAP = 90◦ − ∠BCA

and

∠M CA = ∠M CB + ∠BCA
= ∠M AB + ∠BCA (on chord M B)
1
= (∠CAB) + ∠BCA (AM bisects ∠CAB),
2

substitution in (1) yields

180◦ = ∠CAB + 4 · (∠BCA).

Combining this equation with ∠ABC + ∠BCA + ∠CAB = 180◦ ,

∠CAB may be eliminated. By elimination, we obtain

∠ABC = 3 · (∠BCA)

and
∠ABC
= 3.
∠BCA

Also solved by the proposer.

EM–72. Proposed by Mihaela Berindeanu, Bucharest, România.

Show that the inequality
2 2 2
(x2 + 2) (y 2 + 2) (z 2 + 2)
p + √
3
+ p ≥3
3 3 4y 4 z 4 + 4 3 4z 4 x4 + 4 3 3 4x4 y 4 + 4

holds for all non negative reals x, y, z .

68 Arhimede Mathematical Journal

Solution 1 by Víctor Martín Chabrera, FME, BarcelonaTech,

Barcelona. Using the AM-GM inequality, we have that
X (x2 + 2)2 X (x2 + 2)2
p = p
cyc 3 3 4y 4 z 4 + 4 cyc
3 3 (2y 2 )(2z 2 )(y 2 z 2 ) + 4
X (x2 + 2)2
≥
cyc
2y 2 + 2z 2 + y 2 z 2 + 4
X (x2 + 2)2
=
cyc
(y 2 + 2)(z 2 + 2)
Ê
3
(x2 + 2)2 (y 2 + 2)2 (z 2 + 2)2
≥3 = 3.
(x2 + 2)2 (y 2 + 2)2 (z 2 + 2)2
x2
Solution 2 by Michel Bataille, Rouen, France. Setting a = 2
,
2 2
b = y2 , c = z2 , the inequality becomes L ≥ 3 with
(a + 1)2 (b + 1)2 (c + 1)2
L= + + .
3(bc)2/3 + 1 3(ca)2/3 + 1 3(ab)2/3 + 1
From the Cauchy-Schwarz inequality, we have
L(3(bc)2/3 +1+3(ca)2/3 +1+3(ab)2/3 +1) ≥ (a+1+b+1+c+1)2 ,
hence to prove L ≥ 3, it is sufficient to prove that
(3 + (a + b + c))2 ≥ 9(1 + (ab)2/3 + (bc)2/3 + (ca)2/3 ),
which rewrites as
9(ab)2/3 + 9(bc)2/3 + 9(ca)2/3 ≤ (a + b + c)2 + 6(a + b + c).
Since a2 + b2 + c2 ≥ ab + bc + ca, we have
(a + b + c)2 ≥ 3(ab + bc + ca),
and the latter inequality will follow if we prove
3(ab)2/3 + 3(bc)2/3 + 3(ca)2/3 ≤ ab + bc + ca + 2(a + b + c). (1)
Now, by AM-GM, we have ab + a + b ≥ 3(ab · a · b)1/3 = 3(ab)2/3
and similarly bc + b + c ≥ 3(bc)2/3 and ca + c + a ≥ 3(ca)2/3 ;
adding these three inequalities, we obtain (1), so we are done.
Volume 7, No. 1, Spring 2020 69

Solution 3 by the proposer. First, we observe that

p È
4y z = yz(2y)2 (2z)2 .
3 4 4 3

According to the AM-GM inequality,

È y 2 z 2 + 2y 2 + 2z 2
3
y 2 z 2 (2y 2 )(2z 2 ) ≤ ,
3
so p
3· 3
4y 4 z 4 ≤ y 2 z 2 + 2y 2 + 2z 2
Therefore,
2 2
(x2 + 2) (x2 + 2)
p ≥
3 · 3 4y 4 z 4 + 4 y 2 z 2 + 2y 2 + 2z 2 + 4
or
2 2
(x2 + 2) (x2 + 2)
p ≥ .
3 · 3 4y 4 z 4 + 4 (y 2 + 2)(z 2 + 2)
Likewise, we get
2 2
(y 2 + 2) (y 2 + 2)
√3
≥ 2
3 · 4z 4 x4 + 4 (z + 2)(x2 + 2)
and
2 2
(z 2 + 2) (z 2 + 2)
p ≥ .
3 · 3 4x4 y 4 + 4 (x2 + 2)(y 2 + 2)
Applying again the AM-GM inequality,
2 2 2
(y 2 + 2) (z 2 + 2) (x2 + 2)
+ 2 + 2
(z 2 + 2)(x2 + 2) (x + 2)(y 2 + 2) (y + 2)(z 2 + 2)
s
2 2 2 2 2 2
3 (y + 2) (z + 2) (x + 2)
≥3 ≥ 3.
(y 2 + 2)2 (z 2 + 2)2 (x2 + 2)2
√
Equality occurs for x = y = z = 0 and x = y = z = 2, as can
be easily checked.

Also solved by José Luis Díaz-Barrero, BarcelonaTech, Barcelona,

Spain.
70 Arhimede Mathematical Journal

EM–73. Proposed by Andrés Sáez Schwedt, Universidad de León,

Spain. Let E be a point on side CD of a square ABCD, and
let H be the orthocenter of triangle ABE . The lines BE and
AE meet the circumcircles of triangles AEH and BEH again at
points P and Q, respectively. The lines AP and BQ intersect at
F . Prove that the circle of diameter AB is tangent to the incircle
of triangle ABF .

Solution 1 by Miguel Amengual Covas, Cala Figuera, Mallorca,

Spain. The angles HBP and HEP are equal since both subtend
the arc P H . But ∠HEP is also equal to ∠ABH ; this follows
from the fact that the arms of the two angles are respectively
perpendicular:

(altitude) EH ⊥ AB and (altitude) BH ⊥ EA.

Thus,
∠ABH = ∠HBF
and BH bisects ∠ABF . Similarly, AH bisects ∠F AB , and there-
fore H is the incenter of 4F AB .

D E C

Q
F

A MU B

Figure 6: Scheme for Solution 1 of Problem EM–73.

Volume 7, No. 1, Spring 2020 71

Let EH (extended) intersect AB at U . Since

∠U BH = ∠U EA

(∠U BH = ∠ABH = ∠HEP = ∠U EA, where the second equality

holds as shown above), right-angled triangles U BH and U EA are
similar. From the proportional sides we get

UB EU
= . (1)
HU AU

Let AB = a, AU = x, and let r denote the inradius of 4F AB .

Then U B = a − x, EU = a and HU = r. Substitution in (1)
yields
a−x a
= ,
r x
from which
x2 − ax = −ar.

a2
Adding 4
+ r 2 to each side gives
 a 2 a 2
x− + r2 = −r
2 2
and Ê
 a 2 a a
x− + r2 = −r = − r. (2)
2 2 2

Let M be the midpoint of side AB . Since

Ê
 a 2 È p
x− +r 2 = (AU −AM )2 +HU 2 = M U 2 +HU 2 = M H,
2

where the last equality holds by the Pythagorean theorem applied

to 4HM U , equality (2) tells us that the distance between the
center of the circle on AB as diameter and the incenter of 4F AB
equals the difference between their radii, making these circles
internally tangent, as we wish to prove.
72 Arhimede Mathematical Journal

Note. This problem provides an elementary and Euclidean solu-

tion to the following question.
Let P be any interior point on segment AB and let Γ denote a semi-
circle on AB as diameter. Draw a circle touching Γ and segment
AB at P .
Construction. In the plane of Γ, construct a line l through P per-
pendicular to AB , as in figure 7.
Q

l
R

A P B

Figure 7: Scheme for the extra construction in Solution 1 of Prob-

lem EM–73.

With AB as radius and P as center, draw an arc of circle intersect-

ing l at Q. Let R be the second point of intersection of QA and Γ,
and let RB intersects P Q at I .
The circle of centre I and radius IP touches AB at P and touches
Γ internally.
Solution 2 by the proposer. Let α = ∠BAE and β = ∠EBA.
Because of the cyclicity of the quadrilaterals AEP H and BHQE ,
one has
¨
∠HAP = ∠HEP = 900 − β = ∠HAB, and
∠HBQ = ∠HEQ = 900 − α = ∠HBA.
 Volume 7, No. 1, Spring 2020 73

Therefore, H is the incenter of triangle ABF because it is the

Proposal ASS-1of two internal bisectors.
intersection
Let E be a point on side CD of a square ABCD, and let H be the orthocenter of triangle ABE.
TheDenote by AE
lines BE and Ω the
meet circle of diameter
the circumcircles AB
of triangles , with
AEH center
and BEH againM (the Pmid-
at points and Q,
point ofThe
respectively. AB ) and
lines radius
AP and R, and
BQ intersect at Flet ω be
. Prove the
that theincircle of ABF
circle of diameter AB, iswith
tangent
to the incircleH
center of triangle ABF . r = HT , being T the foot of the perpendicu-
and radius
lar from E to AB .
Solution to Proposal ASS-1.
D E C

A M T B

Let α Figure
=  BAE, 8: β =Construction
 EBA. Because of the cyclicity of the quadrilaterals AEP H and BHQE,
for Solution 2 of Problem EM–73.
one has: 
 HAP =  HEP = 900 − β =  HAB, and
 HBQ =  HEQ = 900 − α =  HBA.

Therefore H is the incenter of triangle ABF because it is the intersection of two internal bisectors.
We shall
Denote by Ω use the offollowing
the circle criterion
diameter AB, with centerfor
M internal tangency:
(the midpoint of AB) and radius R; and
let ω be the incircle of ABF , with center H and radius r = HT , being T the foot of the perpendicular
from E AB.be tangent to Ω if we prove that M H = R − r.
ω towill
We shall use the following criterion for internal tangency:
ω will be tangent to Ω if we prove that M H = R − r.
Indeed, this will ensure the existence of a point G in the prolon-
Indeed, this will ensure the existence of a point G in the prolongation of M H beyond H such
thatgation
G belongs oftoM H Ωbeyond
both and ω, with such that
Ha common tangent Gatbelongs
G. to both Ω and ω ,
with a common tangent at G.
Consider the lengths a = T A and b = T B, assuming a ≥ b without loos of generality. Then AB
and T E are equal to a + b, in particular
Consider the lengths a =R = TAAB and a+b b = T B , assuming a ≥ b
= .
2 2
without loss of generality. Then, AB and T E are equal to a + b.
Also, the triangles T AH and T EB are similar, therefore
In particular,
TA TE TA · TB ab
= ⇒ r AB
= TH = a + b = .
TH TB TE a+b
√ R= = .
And of course, M H = T M 2 + T H 2 , so that 2 the desired2equality M H = R − r is equivalent to

   2
 a−b 2 ab a+b ab

+ = − ,
2 a+b 2 a+b
which is just a straightforward calculation.

5
74 Arhimede Mathematical Journal

Furthermore, the triangles T AH and T EB are similar, therefore

TA TE TA · TB ab
= =⇒ r = T H = = .
TH TB TE a+b
√
And of course, M H = T M 2 + T H 2 , so that the desired equality
M H = R − r is equivalent to
ʁ ‹2  ‹2
a−b ab a+b ab
+ = − ,
2 a+b 2 a+b
which is just a straightforward calculation.

EM–74. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Bar-

celona, Spain. Let A(x) = xn +an−1 xn−1 +an−2 xn−2 +. . .+a1 x+a0
be a polynomial of degree n ≥ 1 with nonnegative coefficients.
Prove that
X n
A(k) n+1
≤ .
k=1
A(k) + A(n − k) 2

Solution 1 by Michel Bataille, Rouen, France. Let

X
n
A(k)
Sn = .
k=1
A(k) + A(n − k)
Changing the indexation, we obtain
n−1
X n−1
X
A(n − j) A(n − k)
Sn = = .
j=0
A(n − j) + A(j) k=0
A(k) + A(n − k)
It follows that
n−1
!
A(n) X A(k)
2Sn = +
A(n) + A(0) k=1
A(k) + A(n − k)
n−1
!
A(n) X A(n − k)
+ +
A(0) + A(n) k=1
A(k) + A(n − k)
n−1
X
2A(n) A(k) + A(n − k)
= +
A(0) + A(n) k=1
A(k) + A(n − k)
2A(n)
= + (n − 1).
A(0) + A(n)
Volume 7, No. 1, Spring 2020 75

Now, A(0) = a0 ≥ 0, hence A(0) + A(n) ≥ A(n) and, therefore,

P
A(n)
A(0)+A(n)
≤ 1 (since A(0)+A(n) = a0 +nn + n−1 k n
k=0 ak n > n > 0).
We finally deduce that
2Sn ≤ 2 + (n − 1) = n + 1
and the desired inequality immediately follows.
A(k)
Solution 2 by the proposer. Let ak = . Then,
A(k) + A(n − k)
A(k) A(n − k)
ak + an−k = + =1
A(k) + A(n − k) A(n − k) + A(k)
and
n−1
X n−1
X
2 ak = (ak + an−k ) = n − 1,
k=1 k=1
which implies that
X
n
n−1 n−1 A(n)
ak = + an = + .
k=1
2 2 A(0) + A(n)
A(n)
Since ≤ 1 because A(0) = a0 ≥ 0, then
A(0) + A(n)
X
n
A(k) n+1
≤ .
k=1
A(k) + A(n − k) 2
Equality holds when a0 = 0, and we are done.

Also solved by Víctor Martín Chabrera, FME, BarcelonaTech, Bar-

celona, Spain, and Henry Ricardo, Westchester Area Math Circle,
NY, USA.

EM–75. Proposed by Oriol Baeza Guasch, CFIS BarcelonaTech,

Terrassa, Spain. Let ABC be a triangle with incenter I and cir-
cumcenter O. Denote by D, E and F the contact points of the
incircle with the sides BC , CA and AB , respectively. Let Q be
the second intersection of the circumcircles of ABC and BDF .
Finally, let N be the midpoint of arc BC containing A. Prove that
lines N Q, IO and AB are concurrent.
76 Arhimede Mathematical Journal

Solution by the proposer. First of all, let M be the midpoint of

arc BC not containing A. Notice M , O and N are collinear, as
well as A, I and M . Also, let P be the intersection of QN and
AB , and we will rather show that P , I and O are collinear.

Figure 9: Construction for the solution of Problem EM–75.

Next, it is clear that ∠BF D = 90◦ − ∠BAC

2
= ∠BID, and so I also
belongs to (BF D). Then, from ∠BQI = ∠BF I = 90◦ , we get that
QI ∩ (ABC) is the point diametrically opposed to B , say B 0 . That
means Q, I and B 0 are collinear, as well as B , O and B 0 .

Now, applying Pascal to hexagon QN M ABB 0 we get that QN ∩

AB = P , N M ∩ BB 0 = O and M A ∩ B 0 Q = I are collinear,
which is exactly what we wanted to show.

EM–76. Proposed by Mihaela Berindeanu, Bucharest, Romania.

Let P be an interior point in 4ABC so that AP ∩ BC = E ,
BP ∩ AC = F, CP ∩ AB = G and ∠BGC = ∠AEC = ∠BF A.
−→ − −→ −→ −
→
Show that, if AP + BP + CP = 0 , then ABC is an equilateral
triangle.

Solution by the proposer. Let us denote ∠AF B = ∠BGC =

∠AEC = α, ∠ABC = β and let O be the circumcenter of 4ABC .
Volume 7, No. 1, Spring 2020 77

Figure 10: Scheme for the solution of Problem EM-76.

We write the solution in the following steps:

Claim 1. Point P is the orthocenter of 4ABC .
Indeed, ∠AEC = α, so ∠BEP = 180◦ − α. Using this and
the fact that ∠BGC = α, we conclude that BEP G is cyclic.
Now, we see that AGF P is cyclic because ∠AF P = ∠BGC = α
and ∠AGP + ∠AF P = 180◦ . We also observe that quadrilateral
GBCF is cyclic because ∠AP G = ∠GF A = β (both see the
same chord AG) and then we have that ∠GBC + ∠GF C = 180◦ .
Moreover, ∠BGC = ∠BF C = α both see the same chord BC .
Then, α = 90◦ and we have BF ⊥ AC , CG ⊥ AB and AE ⊥ BC .
This means that P is the orthocenter of 4ABC as claimed.
Claim 2. Triangle ABC is equilateral.
Indeed, according to Sylvester’s Theorem, with P the orthocenter
and O the circumcenter of 4ABC , we have
−→ −→ −→ −→ −→ −→ −→ −→ − −→ −→ −→
OP = OA+ OB + OC ⇒ OP = OP + P A+ OP + P B + OP + P C.
−→ −
−→ −→ −→ −→ −→ −
→
Since AP + BP + CP = 0, then OP = 3OP , OP = O and
P ≡ O. So 4ABC is equilateral, as we wanted to prove.

Also solved by José Luis Díaz-Barrero, BarcelonaTech, Barcelona,

Spain.
78 Arhimede Mathematical Journal

Medium–Hard Problems
MH–71. Proposed by Andrés Sáez Schwedt, Universidad de León,
Spain. ABCD is a convex quadrilateral satisfying ∠CAD =
∠DCA and ∠DCB = ∠CBA. Let r be the external bisector
of the angle ∠DAB and let s be the reflection of the line AC with
respect to BC . Prove that the lines r, s and BD are parallel or
Proposal ASS-2
concurrent.
ABCD is a convex quadrilateral satisfying  CAD =  DCA and  DCB =  CBA. Let r be the
external bisector
Solution by theof the angle  DAB and let s be the reflection of the line AC with respect to BC.
proposer. Denote by α, β and γ the angles of
Prove that the lines r, s and BD are parallel or concurrent.
ABC . It is clear that ∠CAD = ∠DCA = β − γ . Moreover, the
angles between
Solution (r, AB),
to Proposal (r, AD) and (s, BC) are γ , and the angle
ASS-2.
between (s, CD) is α.
Denote by α, β, γ the angles of ABC. It is clear that  CADC =  DCA = β − γ. Moreover, the
angles between (r, AB), (r, AD) and (s, BC) are γ, and the angle between (s, CD) is α.
Case β < 2γ 
B

Case r||s (β = 2γ)

r s

γ γ
A α Q Q γ
   
C
B
β−γ β−γ
 γ
γ
γ α
α γ r
A C
 
s
γ β − γ β−γ 


D
P

If r||s (see the left picture), it is easy to check that this condition coincides with β = 2γ, and in
Figure
this case ABCD11: isConstruction
cyclic. Since DA = DC, for itthe solution
follows that BD is ofthe Problem MH–71.
bisector of  ABC, and the
slope of BD reveals that it is parallel to r and to s.
Assume β < 2γ (see the right picture). Consider the points P = r ∩ s, Q = BP ∩ AC and
If r||s
Q = DP (see∩ AC,the
and left picture),
note that it D-halfplane
P lies on the is easy with to check
respect to that
the line this
AC. condition
QA Q A
We claim that = . To prove this, brackets
coincides withQC β Q=C 2γ , and in this case ABCD is cyclic.
 [ ] will be used to denote areas of triangles:
Since
1
DA = DCQA ,[PitBA]follows· AB that
· AP · sin(
BD  P AB)
is =the
AB ·bisector of=∠ABC
AP · sin(α + β) AB · AP , and the
= = 12 , (1)
QC [P BC] · CB · CP · sin( P CB) CB · CP · sin(α + β) CB · CP
slope of BD reveals

2 that 1
it is parallel to r and to s.
QA [P DA] 2
 · AD · AP · sin( P AD) AP · sin(γ) AP · AB
= = 1 = = , (2)
Q C [P DC] · CD · CP · sin( P CD) CP · sin(α) CP · CB
Assume β < 2γ (see the right picture). Consider the points P =
2
where we have used the sines law in triangle
0 ABC, and we have cancelled AD = CD.
r ∩ s,Now, = claim
Q the ∩ AC
BPimplies Q =and
Q andQ =the
hence DP ∩ AC
problem , and
is solved: sincenote
in thisthat lies lies
case D P inside on
triangle ACP and ABCP is a convex quadrilateral, both Q and Q lie between A, C and so there is
QA Q A
no danger that the ratios QC and Q  C have different signs.

Finally, the case β > 2γ is similar, with P lying on the B-halfplane with respect to the line AC.
Remark: An alternative solution can be obtained by equating the cross-ratios
−→ −→ −→ −−→ −→ −−→ −−→ −→ −−→ −→ −→ −−→
[AB, AP , AC, AD] = [CP , CB, CD, CA] = [CB, CP , CA, CD].
This implies that the lines AD and CD intersect BP at the same point, which can only be D.

6
Volume 7, No. 1, Spring 2020 79

the D-halfplane with respect to the line AC .

QA QA0
We claim that QC = Q 0 C . To prove this, brackets [ ] will be used to

denote areas of triangles:

1
QA [P BA] 2
· AB · AP · sin(∠P AB)
= = 1
QC [P BC] · CB · CP · sin(∠P CB)
2
AB · AP · sin(α + β) AB · AP
= = ,
CB · CP · sin(α + β) CB · CP
1
Q0 A [P DA] 2
· AD · AP · sin(∠P AD)
= = 1
Q0 C [P DC] 2
· CD · CP · sin(∠P CD)
AP · sin(γ) AP · AB
= = ,
CP · sin(α) CP · CB

where we have used the sines law in triangle ABC , and we have
cancelled AD = CD.

Now, the claim implies Q = Q0 and hence the problem is solved:

since in this case D lies inside triangle ACP and ABCP is a
convex quadrilateral, both Q and Q0 lie between A, C and so there
QA Q0 A
is no danger that the ratios QC and Q 0 C have different signs.

Finally, the case β > 2γ is similar, with P lying on the B -halfplane

with respect to the line AC .

Remark: An alternative solution can be obtained by equating the

cross-ratios
−→ −→ −→ − −→ −→ − −→ −−→ −→ −
−→ −→ −→ − −→
[AB, AP , AC, AD] = [CP , CB, CD, CA] = [CB, CP , CA, CD].

This implies that the lines AD and CD intersect BP at the same

point, which can only be D.

MH–72. Proposed by Pedro H. O. Pantoja, Natal/RN,

√ Brazil. Find
the solution of the equation sin8 x − cos8 x = 7 3/16.
80 Arhimede Mathematical Journal

Solution by Michel Bataille, Rouen, France. The left-hand side

of the equation is

sin8 x − cos8 x = (sin2 x + cos2 x)(sin2 x − cos2 x)(sin4 x + cos4 x)

‚ Œ
1 − cos2 (2x)
= − cos(2x) 1 −
2

(indeed, this follows since sin2 x + cos2 x = 1, cos2 x − sin2 x =

2
cos 2x and 2 sin2 x cos2 x = sin2 2x ). It readily follows that the given
equation is equivalent to
√
3
7 3
cos (2x) + cos(2x) + = 0,
8
that is,
‚ √ Œ‚ √ Œ
3 3 7
cos(2x) + cos2 (2x) − cos(2x) + = 0.
2 2 4
√
3
The quadratic equation X 2 − 2
X + 74 = 0 having no real solution,
√
the initial equation finally is equivalent to cos 2x = − 23 and
therefore the required solutions are the numbers 5π12
+ kπ (k ∈ Z)
5π
and − 12 + `π (` ∈ Z).

Also solved by Miguel Amengual Covas, Cala Figuera, Mallorca,

Spain; Rovsen Pirkuliyev, Sumgayit City, Azerbaijan, and the pro-
poser.

MH–73. Proposed by Oriol Baeza Guasch, CFIS BarcelonaTech,

Terrassa, Spain. Let ABC be a triangle (AB ≤ AC ) with incenter
I and circumcenter O. Let E be the contact point of the incircle
with side CA. Denote by N the midpoint of arc BC containing
A, and denote by M the midpoint of arc AC not containing B .
Finally, suppose line IO intersects side AB at point P . Show that
lines N P and M E meet at the circumcircle of ABC .

Solution by the proposer. First of all, let D and F be the other

contact points of the incircle with BC and AB , respectively. Let
Volume 7, No. 1, Spring 2020 81

Q be the second intersection of the circumcircles of ABC and of

BDF . We claim that Q is, in fact, the intersection of lines N P
and M E .
Claim 1. Q belongs to M E .
Proof. By construction, Q is the spiral similarity center that sends
DC to F A. Hence, 4QDC ∼ 4QF A and
QC CD CE
= =
QA FA AE
and, by the angle bisector theorem, that means QE is the angle
bisector of ∠AQC . Thus, Q, E , M are collinear.

Figure 12: Construction for the solution of Problem MH–73.

Claim 2. Q belongs to N P .
Proof. Denote by N 0 the midpoint of arc BC not containing A.
Notice A, I , N 0 are collinear as well as N , O, N 0 .
Let B 0 be the point diametrically opposite to B in (ABC). Notice
B , O, B 0 are collinear, but also Q, I , B 0 since I belongs to BDF
and ∠BQI = ∠BDI = 90◦ = ∠BQB 0 .
Now, applying Pascal’s theorem to the hexagon N QB 0 BAN 0 we
have that point N Q ∩ BA, point QB 0 ∩ AN 0 = I and point
82 Arhimede Mathematical Journal

B 0 B ∩ N 0 N = O are collinear. Thus, point N Q ∩ BA ≡ P , which

means that Q, P , N are collinear.
And this completes the proof.

MH–74. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Bar-

celona, Spain. Solve in real numbers the following system of equa-
tions: √ √ √ √ 
a b + b c + c a = 3 abc, 
a2 c b2 a c2 b 3
3
+ 3
+ 3
= . 
b + abc c + abc a + abc 2
Solution 1 by Miguel Valdivieso, Colegio Alemán, Madrid, Spain.
In order both the square roots and the fractions to be defined, a, b,
c must be positive real numbers. As it turns out, the first equation
is rather an inequality: in fact, in virtue of the AM-GM inequality,
√ È √ √
√ √ 3
a b + b c + c a > 3 · abc · abc = 3 abc.
√ √ √
Equality is achieved if and only if a b = b c = c a, that is, if
and only if a = b = c. Thus, the first equation is equivalent to
a = b = c, which fulfils the second equation:
a2 c b2 a c2 b a3 3
+ + =3· = .
b3 + abc c3 + abc a3 + abc 2a3 2
Hence, the system of equations is satisfied if and only if a = b =
c > 0.
Remark. Just as the first one, the second equation is but an in-
equality, whereas proving it is far more complicated:
For positive reals a, b, c, prove the inequality
a2 c b2 a c2 b 3
+ + > .
b3 + abc c3 + abc a3 + abc 2

Due to homogeneity, one may assume abc = 1. Hence,

a2 c b2 a c2 b a b c
+ + = + + .
b3 + abc c3 + abc a3 + abc b4 + b c4 + c a4 + a
Volume 7, No. 1, Spring 2020 83

1 1 1
After applying the substitution a = , b = v
, c = w
, and thus
u
uvw = 1, we have
‚ Œ ‚ Œ ‚ Œ
a b c 1 v4 1 w4 1 u4
+ + = · + · + · .
b4 + b c4 + c a4 + a u 1 + v3 v 1 + w3 w 1 + u3
Titu’s Lemma (also known as Engel’s form of Cauchy-Schwarz’s
inequality) yields
‚ Œ ‚ Œ ‚ Œ
1 v4 1 w4 1 u4
· + · + ·
u 1 + v3 v 1 + w3 w 1 + u3
u4 + v 4 + w4 + 2u2 v 2 + 2v 2 w2 + 2w2 u2
> .
(u + v + w) · uvw + (u3 w + v 3 u + w3 v)
Combining the AM-GM inequality and Cauchy-Schwarz, we have
(u4 + v 4 + w4 ) + (u2 v 2 + v 2 w2 + w2 u2 )
È
>2 · (v 4 + w4 + u4 )(u2 v 2 + v 2 w2 + w2 u2 )
> 2 · (v 3 u + w3 v + u3 w). (1)
Moreover, in virtue of Muirhead’s inequality, we obtain that
u2 v 2 + v 2 w2 + w2 u2 > u2 vw + uv 2 w + uvw2
since (2, 2, 0)  (2, 1, 1). Together with the last result, we get
u4 + v 4 + w4 + 2u2 v 2 + 2v 2 w2 + 2w2 u2
(u + v + w) · uvw + (u3 w + v 3 u + w3 v)
2 · (v 3 u + w3 v + u3 w) + u2 vw + uv 2 w + uvw2
>
(u + v + w) · uvw + (u3 w + v 3 u + w3 v)
v 3 u + w3 v + u3 w
=1 + . (2)
(u + v + w) · uvw + (u3 w + v 3 u + w3 v)
Finally, due to Cauchy-Schwarz’s inequality (recall the constraint
uvw = 1), we have that
‚ Œ
3 3 3
u2 v2 w2
(u w + v u + w v)(u + v + w) = + + (v + w + u)
v w u
> (u + v + w)2 ,
84 Arhimede Mathematical Journal

which is equivalent to
u3 w + v 3 u + w3 v > u + v + w.
Substitution in (2) yields
v 3 u + w 3 v + u3 w v 3 u + w 3 v + u3 w
1+ >1+
(u + v + w) + (u3 w + v 3 u + w3 v) 2(v 3 u + w3 v + u3 w)
3
= .
2

In conclusion, we have proven, as desired, that

a b c 3
+ + > ,
b4 +b c4 +c a4 +a 2
with equality if and only if every inequality becomes an equality.
In particular, regarding the AM-GM inequality (1), we must have
u4 + v 4 + w 4 = u2 v 2 + v 2 w 2 + w 2 u2
⇐⇒ (u2 − v 2 )2 + (v 2 − w2 )2 + (w2 − u2 )2 = 0
⇐⇒ u = v = w.
This suffices for the other inequalities to be equalities. Thus,
equality holds if and only if u = v = w ⇐⇒ a = b = c.
Solution 2 by the proposer. By inspection, it immediately follows
that (α, α, α), where α is a positive real, is a solution. Now, we
claim that these
√ are the only solutions of the given system. Indeed,
dividing by abc both terms of the first equation, we get
É Ê É Ê Ê É
a b c a b b c c a
+ + = 3 ⇐⇒ · + · + · = 3.
c a b b c c a a b
On the other hand, we have that
a2 c b2 a c2 b 3
+ + ≥
b3 + abc c3 + abc a3 + abc 2
can be written as
a2 b2 c2 3
 ‹+  ‹+ c a  ≥ .
a b b c a2 + 2
b2 + c2 +
b c c a a b
Volume 7, No. 1, Spring 2020 85

a
The preceding suggest the change of variables x = b
, y = bc , z = c
a
.
Then, we have to prove that

x2 y2 z2 3
+ + ≥
x+y y+z
z+x 2
√ √ √
when x, y , z are positive reals such that xy + yz + zx = 3.
To prove it, apply the HM-GM inequality to the numbers x, y to
obtain that
√ √
xy xy xy xy
≤ and − ≥− ,
x+y 2 x+y 2
from which it follows that
√
x2 x2 + xy − xy x(x + y) − xy xy xy
= = = x− ≥ x− .
x+y x+y x+y x+y 2
Likewise, we have
√ √
y2 yz z2 zx
≥z− and ≥y− .
y+z 2 z+x 2
Adding up the preceding inequalities and taking into account the
first equation yields
√ √ √
x2 y2 z2 xy + yz + zx
+ + ≥x+y+z−
x+y y+z z+x 2
3
≥x+y+z− .
2
√ 2 √ 2 √ 2 √ √ √
Since x + y + z = ( x) + ( y) + ( z) ≥ xy + yz + zx = 3
on account of the well-known inequality a2 + b2 + c2 ≥ ab + bc + ca,
then
x2 y2 z2 3 3 3
+ + ≥x+y+z− ≥3− = .
x+y y+z z+x 2 2 2
Equality holds when a = b = c, and we are done.

Also solved by Michel Bataille, Rouen, France, and Víctor Martín

Chabrera, FME, BarcelonaTech, Barcelona, Spain.
86 Arhimede Mathematical Journal

MH–75. Proposed by Mihály Bencze, Braşov, Romania. Let n ≥

1 be an integer number. Prove that
$ % $ %
X
2n p X
n p
16n2 + k = 7n2 + n2 + k ,
k=1 k=1

where bxc represents the integer part of x.

Solution by the proposer. Let 1 ≤ k ≤ n be an integer. Then,

for all k we have
2k − 1 p k
n+ < n2 + k < n + .
4n 2n
Adding up the preceding expressions yields

Xn  ‹ X n p Xn  ‹
2k − 1 2
k
n+ < n +k < n+
k=1
4n k=1 k=1
2n

or, equivalently,

X
n
n p
n+1
2
n + < n2 + k < n2 + ,
4 k=1
4

from which it follows that

$ %
X
n p jnk
n2 + k = n2 + .
k=1
4

Likewise, for all k we have

16k − 1 p k
2n + < 4n2 + k < 2n + .
64n 4n
Adding up the preceding expressions yields

Xn  ‹ X n p Xn  ‹
16k − 1 2
k
2n + < 4n + k < 2n +
k=1
64n k=1 k=1
4n
Volume 7, No. 1, Spring 2020 87

or, equivalently,

n Xn p
n+1
2n2 + < 4n2 + k < 2n2 + ,
8 k=1
8

from which it follows that

$ %
X
n p jnk
4n2 + k = 2n2 + .
k=1
8

Putting in the last identity 2n instead of n, we get

$ % $ %
X
2n p jnk X
n p
16n2 + k = 8n2 + = 7n2 + n2 + k ,
k=1
4 k=1

and we are done.

Also solved by José Luis Díaz-Barrero, BarcelonaTech, Barcelona,

Spain, and Ángel Plaza, University of Las Palmas de Gran Canaria,
Spain.

MH–76. Proposed by Mihaela Berindeanu, Bucharest, Romania.

Calculate
Z1 n+1
(4x3 + 1)
lim dx.
n→∞ (5x4 + 1)n
0

Solution 1 by Michel Bataille, Rouen, France. For each non-

negative integer n, let
Z 1
(4x3 + 1)n+1
In = dx.
0 (5x4 + 1)n

We show that lim In = ∞.

n→∞

Let f (x) = 4x3 + 1 and g(x) = 5x4 + 1. For the sake of simplifica-
R R1 R n+1
tion, we use the notation h for 0 h(x) dx. Note that In = f gn
f n+1
is positive (since gn
is a continuous, positive function on [0, 1]).
88 Arhimede Mathematical Journal

R R
The key remark is that f = [x4 + x]10 = 2 = [x5 + x]10 = g . It
follows that I1 ≥ I0 since, by the Cauchy-Schwarz inequality,
Z ‹ ‚Z Œ Z ‹ ‚Z Œ
f2 f2
I1 · I0 = f · = g ·
g g
‚Z Œ2 Z ‹2
√ f
≥ g·√ = f = I02 . (1)
g

Using the Cauchy-Schwarz inequality again, we obtain

Z Z ‚Z Œ2
f n+1 f n+3 f n+2 2
In · In+2 = · ≥ = In+1
gn g n+2 g n+1

In+2 In+1
€I Š
n+1
and so In+1
≥ In
for all n ≥ 0. The sequence In
is
n≥0
In+1
increasing, and in particular ≥ II10 ≥ 1 for all n ≥ 0. Thus,
In
In+1 ≥ In for all n ≥ 0 and the sequence (In ) is increasing.

For the purpose of a contradiction, assume that (In ) is convergent

€ I Šthat ` > 0 (since ` ≥ I0 ). Since
and let ` be its limit. Note
In+1 `
lim In = ` = 1 and n+1 In
is increasing, we have II10 ≤ 1
n→∞ n≥0
and so I1 = I0 . This means that equality holds in (1) so that
f √
√ = λ g for some (positive) constant λ. But then f = λg and
g
from f (0) = λg(0), we deduce λ = 1 and so f = g , a clear
contradiction. We conclude that the sequence (In ) is divergent,
more precisely limn→∞ In = ∞ since (In ) is increasing.

Solution 2 by Víctor Martín Chabrera, FME, BarcelonaTech,

Barcelona, Spain. Let f (x) = 4x3 + 1 and g(x) = 5x4 + 1. It is
clear that, for x ≥ 0, both f (x) and g(x) are bigger or equal than
n+1 (x)
1 and increasing, and so f gn (x) is positive as well for any integer
n.

We can observe that f (1/32) > g(1/16), since 4 3213 + 1 ≥ 5 1615 +

2 18
1 ⇐⇒ 2215 > 2516 ⇐⇒ 2215 > 5 ⇐⇒ 8 > 5, which is true.
f (1/32)
Let a = g(1/16)
. By the previous observation, a > 1. We will have,
Volume 7, No. 1, Spring 2020 89

f n+1 (x)
since g n (x)
> 0 for all n ≥ 0, that
Z Z Z
1
f n+1 (x) 1/16
f n+1 (x) 1/16
f n+1 (1/32)
dx ≥ dx ≥ dx
0 g n (x) 1/32 g n (x) 1/32 g n (1/16)
Z 1/16 Z 1/16
f n (1/32) 1
≥ dx = an dx = an .
1/32 g n (1/16) 1/32 32

1 n
R1 f n+1 (x)
Since a > 1, limn→∞ 32
a = ∞, and so, limn→∞ 0 g n (x)
= ∞.

Comment by Henry Ricardo, Westchester Area Math Circle,

NY, USA. As the problem is stated, the limit is infinite. Instead we
consider sZ
1
(4x3 + 1)n+1
lim n dx.
n→∞ 0 (5x4 + 1)n

n
Since the function ((4x3 + 1)/(5x4 + 1)) is continuous on the
closed interval [0, 1], it must attain its maximum at some point
x = ξ, 0 ≤ ξ ≤ 1. Let ε > 0, δ > 0 be such that
‚ Œn ‚ Œn ‚ Œn
4ξ 3 + 1 4x3 + 1 4ξ 3 + 1
−ε≤ ≤
5ξ 4 + 1 5x4 + 1 5ξ 4 + 1

whenever |x − ξ| < δ , 0 ≤ x ≤ 1. Then,

‚ Œn Z Z ‚ Œn
4ξ 3 + 1 ξ+δ 1
4x3 + 1
−ε (4x3 + 1) dx ≤ (4x3 + 1) dx
5ξ 4 + 1 ξ−δ 0 5x4 + 1
‚ Œn Z
4ξ 3 + 1 1
≤ (4x3 + 1) dx.
5ξ 4 + 1 0

If ξ − δ < 0, the lower limit ξ − δ of the leftmost integral is replaced

by 0. If ξ + δ > 1, the upper limit ξ + δ is replaced by 1. Now take
the n-th root, let n increase to infinity, and let ε tend to 0. The
squeeze principle implies that
sZ
n
1
(4x3 + 1)n+1 4ξ 3 + 1 4x3 + 1
lim dx = = max .
n→∞ 0 (5x4 + 1)n 5ξ 4 + 1 0≤x≤1 5x4 + 1
90 Arhimede Mathematical Journal

Using elementary calculus techniques, we find that

f 0 (x) = −4x2 (5x4 + 5x − 3)/(5x4 + 1)2 ,

and the critical point x ≈ 0.5244 in the interval [0, 1] yields a

maximum value of approximately 1.1442. Thus
sZ
1
n
(4x3 + 1)n+1
lim dx ≈ 1.1442.
n→∞ 0 (5x4 + 1)n

Also solved by the proposer.

Volume 7, No. 1, Spring 2020 91

Advanced Problems
A–71. Proposed by José Luis Díaz-Barrero, BarcelonaTech, Barce-
lona, Spain. Evaluate
Z π/2
cos4 x + 2 sin2 x
dx.
0 2 + sin4 x + cos4 x
Solution by Henry Ricardo, Westchester Area Math Circle, NY,
USA. Let I denote the given integral. Using the known identity
Z b Z b
f (x) dx = f (a + b − x) dx (easily proved using the substi-
a a
tution x 7→ a + b − x), we see that
Z π/2
cos4 ( π2 − x) + 2 sin2 ( π2 − x)
I= dx
0 2 + sin4 ( π2 − x) + cos4 ( π2 − x)
Z π/2
sin4 x + 2 cos2 x
= dx.
0 2 + cos4 x + sin4 x
Adding, we get
Z Z
π/2
cos4 x + 2 sin2 x π/2
sin4 x + 2 cos2 x
2I = dx + dx
0 2 + sin4 x + cos4 x 0 2 + cos4 x + sin4 x
Z π/2
cos4 x + sin4 x + 2 π
= 4 dx = ,
0 2+ cos4 x + sin x 2
π
so that I = .
4

Also solved by Michel Bataille, Rouen, France; Moti Levy, Rehovot,

Israel; Víctor Martín Chabrera, FME, BarcelonaTech, Barcelona, Spain;
Rovsen Pirkuliyev, Sumgayit City, Azerbaijan; Ángel Plaza, Univer-
sity of Las Palmas de Gran Canaria, Spain, and the proposer.

A–72. Proposed by Ander Lamaison Vidarte, Berlin, Germany.

For any matrix A ∈ M5×2 (R) prove that A · At has at most 16
strictly negative entries.
92 Arhimede Mathematical Journal

Solution 1 by Michel Bataille, Rouen, France. (In what follows,

negative = strictly negative.)

We consider the row vectors of the matrix A as vectors

−
→
u1 , −
→
u2 , −
→
u3 , −
→
u4 , −
→
u5

of the usual plane so that the (i, j)-entry of the symmetric matrix
A · At is the dot product −
→
ui · −
→
uj .

Since the diagonal entries − →ui · −

→
ui (i = 1, . . . , 5) are nonnegative,
it is sufficient to show that it cannot occur that nine of the ten
numbers − →ui · −
→
uj (i 6= j ) are negative.

For the purpose of a contradiction, we assume that at least 9 of

these numbers are negative. First, we show how to select four of
the five vectors −→
u1 , −
→
u2 , −
→
u3 , −
→
u4 , −
→
u5 whose mutual dot products are
all negative. If the ten dot products − →
ui · −→
uj are negative, pick any
four of the five vectors. Otherwise, say if − →·−
u →
k u` is nonnegative for
−
→
some (k, `) with k 6= `, select the vectors ui with i 6= k. Without
loss of generality, suppose that − →
u1 , −
→
u2 , −
→
u3 , −
→
u4 are such that the dot
products u1 · u2 , u1 · u3 , u1 · u4 , u2 · u3 , −
−
→ −
→ −
→ −
→ −
→ −
→ −
→ −
→ →
u2 · −
→
u4 , −
→
u3 · −
→
u4 are all
−
→ ⊥
negative. Then, u4 is a line (subspace of dimension 1) of the
plane; let −→, −
w → − →
1 w2 , w3 be the orthogonal projections of u1 , u2 , u3
−
→ − → −→
onto this line. We have
−
→
ui = −
→+λ−
w → (i = 1, 2, 3)
i i u4

for some real numbers λi .

Now, expressing that − →

ui · −
→
u4 < 0 we see that λi < 0 (i = 1, 2, 3)
(note that wi · u4 = 0). We readily deduce that −
−
→ −
→ →·−
w → − → − →
1 w2 , w1 · w3 ,
−→·−→ −
→ − → −→ − → −
→2
w 2 w3 are negative: for example since u1 · u2 = w1 · w2 + λ1 λ2 u4 ,
we have w1 · w2 = u1 · u2 − λ1 λ2 u4 · u4 < 0. However, w2 = µ−
−→ −
→ −
→ −
→ −
→ −
→ −
→ w→,
1
−
w→ = ν− → , where the real numbers µ, ν are negative (since, for
w
3 1
example, µ−w→·− → −
→ − → −
→ − → −
→ − →
1 w1 = w2 · w1 < 0); but then w2 · w3 = µν w1 · w1 > 0,
a contradiction.

We conclude that at most eight of the numbers − →

ui · −
→
uj (i 6= j ) are
t
negative and so at most 16 of the entries of A · A are negative.
Volume 7, No. 1, Spring 2020 93

Note. Using induction, one can easily prove (by mimicking the proof
above) that in an n-dimensional Hilbert space, at most n + 1 vec-
tors can form an obtusangle set, that is, can have all their mutual
dot products negative.

Solution 2 by Víctor Martín Chabrera, FME, BarcelonaTech,

Barcelona, Spain. First of all, note that the statement is equiv-
alent to saying that A · At has at least 9 non-negative elements.
Let vi be the i-th row of A and let B = A · At . It is clear that
Bij = (A · At )ij = vi · vj , that is, B is the matrix of scalar products
of the vectors v1 , . . . , v5 ∈ R2 . Thus, for all i ∈ {1, . . . , 5}, we have
Bii = vi · vi = kvi k2 ≥ 0. This way we have found our first 5
non-negative elements.

Let us say {i, j}, with i 6= j is a good pair if vi · vj ≥ 0 (and let

us say it is bad otherwise). Since B is symmetric, we just need to
show there are at least two different good pairs, which will give us
the remaining 4 non-negative elements.

Assume there is at most one good pair. This implies that there are
three vectors vi1 , vi2 , vi3 (the ones not belonging to the pair if there
is such a pair, or any three vectors otherwise) such that, for each
of them, the scalar product with any of the remaining four vectors
is negative.

Since rotations of the configuration of the 5 vectors will not change

the values of the scalar products, we can assume WLOG vi1 =
(0, r), for some r ≥ 0. Let us cover the plane with the four common
quadrants (consider them as closed sets). Let us notice that vi1
belongs to the two upper quadrants.

It is trivial to see that two vectors on the same quadrant have non-
negative scalar product, since the angle they form cannot exceed
π
2
. Therefore, vi2 and vi3 will belong one to the lower left quadrant,
and the other, to the lower right one.

So, any new vector we add to the configuration will belong to a

quadrant already covered by at least one of the previous three
vectors, meaning that any new vector added will form at least a
new good pair. Since we will be adding two more vectors, we will
94 Arhimede Mathematical Journal

obtain two good pairs, getting a contradiction and completing the

proof.
Remark. We can achieve this bound. If A, B , and C are the
vertices of an equilateral triangle centered in the origin O, and we
−→ −→ −→
let v1 = v2 = OA, v3 = v4 = OB and v5 = OC , we would obtain
a configuration that attains the bound.
Solution 3 by the proposer. We will show that A · At has at least
9 non-negative entries. For 1 ≤ i ≤ 5, let ~ vi ∈ R2 be the vector
corresponding to the i-th row of A. Then, the (i, j) entry of A · At
vi · ~
is ~ vj . This immediately produces five non-negative entries in
the diagonal, as ~ vi · ~
vi ≥ 0.
We need to find four non-negative off-diagonal entries, or equiv-
alently, two distinct pairs {i, j}, {i0 , j 0 } of indices such that
vi · ~
~ vj ≥ 0, ~
v i0 · ~
vj 0 ≥ 0. If one of the five vectors ~
vi is the zero
vector, then the pairs involving it will suffice.
We claim that, in any set of four nonzero vectors, there is a pair
which has non-negative scalar product. Let w
~ 1, w
~ 2, w ~ 4 be those
~ 3, w
four vectors, with indices taken so that the four vectors lie in
clockwise order. Then,
2π = ∠w ~ 2 + ∠w
~ 1w ~ 3 + ∠w
~ 2w ~ 4 + ∠w
~ 3w ~ 1,
~ 4w
where the angles are measured clockwise. There is a pair with
∠w ~ j ≤ π2 , and therefore w
~ iw ~i · w
~ j ≥ 0.
The set S = {~v1 , ~
v2 , ~
v3 , ~ v5 } has five elements, therefore it con-
v4 , ~
tains a pair {~ vj } with ~
vi , ~ vi · ~
vj ≥ 0. The set S \ {~vi } has four
elements, and thus it contains a different pair {~ vj 0 }, with
v i0 , ~
vi · ~
~ 0 vj ≥ 0. This concludes the proof of the statement.
0

Solution 4 by Moti Levy, Rehovot, Israel. Let us express the

matrix A as an array of row vectors,
 t 
v1
 vt 
 2t 
A :=  
 v3 ,
 v 
t
4
v5t
Volume 7, No. 1, Spring 2020 95

where • ˜ • ˜ • ˜
a1,1 a2,1 a5,1
v1 = , v2 = , . . . , v5 = .
a1,2 a2,2 a5,2
Then, the matrix A · At is
 
hv1 , v1 i hv1 , v2 i hv1 , v3 i hv1 , v4 i hv1 , v5 i
 hv2 , v1 i hv2 , v2 i hv2 , v3 i hv2 , v4 i hv2 , v5 i 
 
A · At = 
 hv3 , v1 i hv3 , v2 i hv3 , v3 i hv3 , v4 i hv3 , v5 i ,

 hv4 , v1 i hv4 , v2 i hv4 , v3 i hv4 , v4 i hv4 , v5 i 
hv5 , v1 i hv5 , v2 i hv5 , v3 i hv5 , v4 i hv5 , v5 i

where hv, ui is the inner product of the vectors v and u.

We observe that A · At is a symmetric matrix with non-negative

numbers on the diagonal.

Lemma. In R2 , there are at most 3 vectors with pairwise negative

inner product.

Proof. Suppose that there are 4 vectors v1 , v2 , v3 , v4 with pairwise

negative inner product. Any three vectors in R2 are linearly de-
pendent, hence there exist numbers α1 , α2 , α3 , not all zero, such
that
α1 v1 + α2 v2 + α3 v3 = 0.
Suppose α1 , α2 , α3 ≥ 0. Then, on one hand

hα1 v1 + α2 v2 + α3 v3 , v4 i = h0, v4 i = 0,

and on the other hand

X
3
hα1 v1 + α2 v2 + α3 v3 , v4 i = hαk vk , v4 i < 0.
k=1

A contradiction.

Now suppose that α1 < 0 and α2 , α3 ≥ 0. Then, let w := −α1 v1 =

α2 v2 + α3 v3 . On one hand,

hw, wi = h−α1 v1 , α2 v2 + α3 v3 i ≥ 0,
96 Arhimede Mathematical Journal

and on the other hand

h−α1 v1 , α2 v2 + α3 v3 i = (−α1 α2 )hv1 , v2 i + (−α1 α2 )hv1 , v3 i < 0.
Again, a contradiction.
The case α1 ≥ 0 and α2 , α3 < 0 is treated similarly.
We conclude that there are at most 3 vectors in R2 with pairwise
negative inner product.
To show that this upper limit is achieved, we give an example:
 ‹  ‹
1 1√ 1 1√
v1 = (1, 0), v2 = − , 3 , v3 = − , − 3 .
2 2 2 2

Now let us draw a graph with 5 vertices where a vertex represents

a vector vi . We connect two vertices vi and vj whenever their inner
product hvi , vj i is negative. Now we ask, what is the maximum
number of edges for any choice of the five vectors. According to the
lemma, the graph cannot have a clique with more than 3 vertices.
To find the maximum number of edges such that there are no
4-vertex cliques, we begin with the complete graph K5 (with 10
edges). Clearly, this graph has 4-vertex cliques. Now remove one
edge and clearly this graph still has 4-vertex cliques. Only if two
edges are removed from K5 , then we can have a graph without
4-vertex cliques. For example, see the following graph:
Thus, the maximum number of edges is 8. Because of the symme-
try of A · At , we conclude that the maximum number of negative
entries is 16.
Remark 1. This result can be generalized to A ∈ Mm×n (R) using
the following two theorems:
Theorem. In Rn , there are at most n + 1 vectors with pairwise
negative inner product.
Theorem (Turán). Let G be a graph with n vertices, such that G
is free of (r + 1)-vertex cliques. Then, the number of edges in G is

2
at most 1 − 1r · n2 .
   '&&, ,/' .*,"+  &  /!&.* ,!"* "&&* (*'-,    "+
& ,". '/ / +# /!, "+ ,! %0"%-% &-%* '  + '* &1 !'" '
,! 3. .,'*+ '*"& ,' ,! $%% ,! *(! &&', !.  $")- /",!
%'* ,!& .*,"+
' 3& ,! %0"%-% &-%* '  + +-! ,!, ,!* * &' .*,"+
$")- /  "& /",! ,! '%($, *(!  /",!   + $*$1 ,!"+ 
*(! Volume
!+ .*,"+ $")-
7, No. 1, '/
Spring *%'. '&   & $*$1 ,!"+ *(!
2020 97 +,"$$
!+ .*,"+ $")-+ &$1 " ,/'  + * *%'. *'%   ,!& / & !.
 *(! /",!'-, .*,"+ $")-+ '* 0%($ + ,! '$$'/"& *(!

!-+ ,! %0"%-% &-%*

Figure 13: Scheme ' for+Solution
"+  -+ ' ,! +1%%,*1
4 to Problem A–72. '    
/ '&$- ,!, ,! %0"%-% &-%* ' & ,". &,*"+ "+  

Alsosolved by José Luis Díaz-Barrero, BarcelonaTech, Barcelona,
!"+ *+-$, &  &*$"2 ,'     -+"& ,! '$$'/"& ,/'
Spain.
,!'*%+
   by 
A–73. Proposed  
José LuisDíaz-Barrero,
   BarcelonaTech,
  Bar-
 
  celona,
  Spain. Let a, b be real numbers such that 0 < a < b.

Prove that-*4&
for anyx1 ,
x2 , . . 
. , xn ∈   exists
[a, b] there   a    ξ  
number
  
(a ≤ ξ ≤ b) such that
                         
arctan5 (x1 ) + arctan5 (x2 ) + . . . + arctan5 (xn ) = n arctan5 (ξ).
Solution by Michel Bataille, Rouen, France. Let f and g be the
functions defined by f (x) = arctan5 (x) (x ∈ [a, b]) and g(x) = x5
(x ∈ R). Since g and arctan are strictly increasing and continuous,
the same is true for f = g ◦ arctan. It follows that f is a bijection
from [a, b] onto [f (a), f (b)].
Now, assuming without loss of generality that x1 ≤ x2 ≤ · · · ≤ xn ,
the number
f (x1 ) + f (x2 ) + · · · + f (xn )
m=
n
satisfies f (x1 ) ≤ m ≤ f (xn ), hence f (a) ≤ m ≤ f (b). As a result,
m is the image under f of a unique number ξ of [a, b], so that
f (x1 ) + f (x2 ) + · · · + f (xn ) = nf (ξ),
the desired result.
98 Arhimede Mathematical Journal

Also solved by Moti Levy, Rehovot, Israel; Víctor Martín Chabrera,

FME, BarcelonaTech, Barcelona, Spain; Henry Ricardo, Westchester
Area Math Circle, NY, USA, and the proposer.

A–74. Proposed by Marc Felipe i Alsina, BarcelonaTech, Barce-

lona, Spain. Let f : [0, 1] → R be a twice differentiable function
such that f 00 (x) < 0 for all x ∈ [0, 1]. For all n ≥ 2, prove that
Z 1  ‹2 Z 1
n−1 n−1
f (x) dx ≥ f (1) + f (x) dx.
1/n n2 n 0

Solution 1 by Henry Ricardo, Westchester Area Math Circle,

NY, USA. The function f is concave, so we have f (ta + (1 − t)b) ≥
tf (a) + (1 − t)f (b) for 0 ≤ t ≤ 1 and a, b ∈ [0, 1]. Then, using the
substitution x 7→ (n − 1)x/n + 1/n, we see that
Z 1 Z 1  ‹
n−1 n−1 1
f (x) dx = f · 1 dx ·x+
1/n n 0 n n
Z 1 ‹
n−1 n−1 1
≥ f (x) + f (1) dx
n n n
‚ 0
Z 1 Œ
n−1 n−1 1
= f (x) dx + f (1)
n n 0 n
 ‹2 Z 1
n−1 n−1
= f (1) + f (x) dx.
n2 n 0

Solution 2 by Michel Bataille, Rouen, France. It is readily

checked that the required inequality rewrites as
Z 1  ‹2 Z 1
1
[f (x) − f (1)] dx ≥ 1 − [f (x) − f (1)] dx.
1/n n 0

Let g(x) = f (x) − f (1). Then, g(1) = 0, g is concave on the

interval [0, 1] (since g 00 (x) = f 00 (x) < 0 for all x ∈ [0, 1]) and it is
sufficient to prove that for x ∈ [0, 1) we have
Z 1 Z 1
2
g(t) dt ≥ (1 − x) g(t) dt. (1)
x 0
Volume 7, No. 1, Spring 2020 99

To this end, we consider the function φ defined on [0, 1) by

Z 1 Z 1
2
φ(x) = g(t) dt − (1 − x) g(t) dt.
x 0

Clearly, φ is differentiable on [0, 1) with

Z 1
0
φ (x) = −g(x) + 2(1 − x) g(t) dt.
0

Now, let x be any number in [0, 1). The graph of the concave
function g is above the chord whose extremities are the points
(x, g(x)) and (1, g(1)) = (1, 0). This means that for any t ∈ [x, 1]
the following inequality holds:

g(x)
g(t) ≥ · (t − 1).
x−1
Integrating, we deduce that
Z – ™1
1
g(x) (t − 1)2 g(x)
g(t) dt ≥ =
0 x−1 2 0
2(1 − x)

and therefore
Z 1
0
φ (x) = 2(1 − x) g(t) dt − g(x) ≥ 0
0

(since 1 − x > 0).

As a result, φ is a nondecreasing function. In particular, φ(x) ≥

φ(0) = 0 and (1) holds. This completes the proof.

Solution 3 by the proposer. Consider the function graphed on

the plane and the area beneath it. Figure 14 illustrates the proof.
We have taken n = 3 to do the drawings.

Perform an homothety centered at the point S = (1, f (1)) with

ratio n−1
n
, thus getting a new function g , defined between n1 and 1.
The area between f and the x axis becomes, after the homothety,
the area between the new function and the line y = n1 f (1), which
100 Arhimede Mathematical Journal

Figure 14: Scheme for Solution 3 of Problem A–74.

is the image of the x axis by the transformation. So, the latter has
a value of  ‹ Z
n−1 2 1
f (x) dx.
n 0
If we wanted to calculate the area beneath g , we would have to
add the area between the line y = n1 f (1) and the x axis, which is



1 − n1 n1 f (1) = n−1
n2
f (1):
Z 1  ‹ Z
n−1 n−1 2 1
g(x) dx = f (1) + f (x) dx.
1
n
n2 n 0

Finally, since f 00 (x) < 0 in all the interval, we have that f is

concave, so each point P = (a, g(a)), which is the image under
homothety of another point Q = (b, f (b)) and so lies on the line
joining Q to S = (1, f (1)), is located under the corresponding
point
R = (a, f (a)), so we have that g(x) ≤ f (x) for all x in
1
n
, 1 . Therefore,
Z 1 Z 1  ‹ Z
n−1 n−1 2 1
f (x) dx ≥ g(x) dx = f (1) + f (x) dx.
1
n
1
n
n2 n 0
Volume 7, No. 1, Spring 2020 101

Also solved by José Luis Díaz-Barrero, BarcelonaTech, Barcelona,

Spain, and Víctor Martín Chabrera, FME, BarcelonaTech, Barcelona,
Spain.

A–75. Proposed by Nicolae Papacu, Slobozia, Romania. Let A,

B ∈ GL(n, R) be two matrices distinct from the identity and let
p ≥ 1 be an integer number. If B 3p = I and AB = B 2p A, then
prove that:

1. B 3 = I ,
2. AB = BA if and only if 3 | (p − 2).

Solution 1 by Michel Bataille, Rouen, France. 1. From AB =

B 2p A, we deduce that B 2p = ABA−1 and so

B 6p = (B 2p )3 = ABA−1 · ABA−1 · ABA−1 = AB 3 A−1 .

Since B 3p = I , we also have B 6p = (B 3p )2 = I 2 = I . As a result,

AB 3 A−1 = I and so B 3 = A−1 IA = I .

2. The relation AB = BA is successively equivalent to B =

ABA−1 , B = B 2p , I = B 2p−1 . Now, from B 3 = I and
B 6= I , we deduce that the prime 3 is the order of B in the group
GL(n, R) and, therefore, the equality I = B 2p−1 is equivalent to
3 | (2p − 1). To complete the proof, there just remains to observe
that 2p − 1 = 3(p − 1) − (p − 2) so that 3 | (2p − 1) if and only if
3 | (p − 2).

Solution 2 by Henry Ricardo, Westchester Area Math Circle,

NY, USA. In the proof that follows, we will use the result that
B p AB = A. Indeed, B p AB = B p (B 2p A) = B 3p A = A.

Now we have, noting that A−1 exists,

AB 3 = (B 2p A)B 2 = B p (B p AB)B = B p AB = A,

so B 3 = I .

To prove property 2, first suppose that AB = BA. It follows that

AB = B(B p AB) = B p+1 AB, or B p+1 = I

102 Arhimede Mathematical Journal

since A−1 and B −1 exist.

If we assume that p − 2 ≡ 1 mod 3, then B p−2 = B 6= I and

I = B p+1 = B 3 · B = B , a contradiction of an hypothesis. Also, if
p − 2 ≡ 2 mod 3, we have B p−2 = B 2 , so B p−1 = B 3 = I . But

AB = BA ⇐⇒ B 2p A = BA ⇐⇒ B 2p = B ⇐⇒ B 2p−1 = I
⇐⇒ B p−2 B p+1 = I ⇐⇒ I = B 2 B p+1 = B p+3 = B p .

Since we have already shown that B p+1 = I , we have B p = B p+1 ,

or B = I , a contradiction. Thus p − 2 must be a multiple of 3.

On the other hand, if 3 | (p − 2), then for some positive integer

k, B p−2 = B 3k = (B 3 )k = I , so that B p+1 = B 3 = I . Then
AB = B p−2 B 3 AB = B(B p AB) = BA by the result established
earlier.

Also solved by José Luis Díaz-Barrero, BarcelonaTech, Barcelona,

Spain; Javier F. González and Fabio A. Velandia, Universidad Na-
cional de Colombia, Bogotá, Colombia; Moti Levy, Rehovot, Israel;
Víctor Martín Chabrera, FME, BarcelonaTech, Barcelona, Spain, and
the proposer.

A–76. Proposed by Mihály Bencze, Braşov, Romania, and José

Luis Díaz-Barrero, BarcelonaTech, Barcelona, Spain. Let a, b, c
be three positive reals numbers such that abc = 1. Prove that

Γ(a) Γ(b) Γ(c)

+ + ≥ 1,
1 + a + ab 1 + b + bc 1 + c + ca

where Γ is the gamma function.

Solution by Víctor Martín Chabrera, FME, BarcelonaTech, Bar-

Volume 7, No. 1, Spring 2020 103

celona, Spain. We can rewrite the formula as

Γ(a) Γ(b) Γ(c)

+ +
1 + a + ab 1 + b + bc
1 + c + ca
Γ(a) a
ab Γ(b)
Γ(c)
= + +
1 + a + ab a 1 + b + bc ab 1 + c + ca
Γ(a) aΓ(b) abΓ(c)
= + +
1 + a + ab a + ab + abc ab + abc + abca
Γ(a) aΓ(b) abΓ(c)
= + +
1 + a + ab a + ab + 1 ab + 1 + a
Γ(a) + aΓ(b) + abΓ(c)
= .
1 + a + ab

R∞
Let us show that Γ(x) is convex. Since Γ(x) = 0
tx−1 e−x dt,
Z ∞ Z ∞
d2 Γ(x) ∂2 x−1 −x
= t e dt = (log t)2 tx−1 e−x dt ≥ 0.
dx2 0 ∂ 2x 0

So, we can apply Jensen’s inequality to see that

 ‹  ‹
Γ(a) + aΓ(b) + abΓ(c) a + ab + abc a + ab + 1
≥Γ =Γ
1 + a + ab 1 + a + ab 1 + a + ab
Z ∞
= Γ(1) = e−x dt = 1.
0

Also solved by Michel Bataille, Rouen, France; Rovsen Pirkuliyev,

Sumgayit City, Azerbaijan, and the proposers.
Arhimede Mathematical Journal

Volume 7, No. 1 Spring 2020

Editor-in-Chief

José Luis Díaz-Barrero

BarcelonaTech, Barcelona, Spain.

Editors
Alberto Espuny Díaz Birmingham, United Kingdom.
Ander Lamaison Vidarte Berlin, Germany.
Óscar Rivero Salgado Barcelona, Spain.

Editorial Board
Mihály Bencze Braşov, Romania.
Marc Felipe i Alsina Barcelona, Spain.
José Gibergans-Báguena Barcelona, Spain.
Nicolae Papacu Slobozia, Romania.

Managing and Subscription Editors

Petrus Alexandrescu Bucharest, Romania.
José Luis Díaz-Barrero Barcelona, Spain.
 Aim and Scope

The goal of Arhimede Mathematical Journal is to provide a means of pub-

lication of useful materials to train students for Mathematical Contests
at all levels. Potential contributions include any work involving fresh
ideas and techniques, problems and lessons helpful to train contestants,
all written in a clear and elegant mathematical style. All areas of mathe-
matics, including algebra, combinatorics, geometry, number theory and
real and complex analysis, are considered appropriate for the journal.

Information for Authors

A detailed statement of author guidelines is available at the
journal’s website:
http://www.amj-math.com

Edited in Barcelona by the Arhimede Association

ISSN 2462-537X © 2020. All rights reserved.