GAZETA MATEMATICĂ

SERIA A

ANUL XLI (CXX) Nr. 1 – 2/ 2023

ARTICLES
A class of inequalities involving length elements of a triangle
Jian Liu1)

Abstract. In this paper, we establish a class of inequalities involving

length elements of a triangle. With the help of a computer, we also propose
some similar conjectured inequalities as open problems.
Keywords: Triangle, altitude, angle-bisector, median, symmedian, radius
of excircle.
MSC: 51M16.

1. Introduction

Given a triangle ABC with side lengths a, b, c, denote by ha , hb , hc

the altitudes, wa , wb , wc the angle-bisectors, ra , rb , rc the radii of excircles,
ma , mb , mc the medians, and ka , kb , kc the symmedians.
In a Chinese paper [2], the author established the following two inequal-
ities involving the altitudes and medians of a triangle ABC:
4mb mc ≥ (hb + hc )2 , (1)
(mb + mc )2 ≥ 2(h2b + h2c ). (2)
In another Chinese paper [4], the author considered improvements of inequal-
ity (1) and proved the following double inequality
wb2 + wc2 ≥ 2ma (mb + mc − ma ) ≥ h2b + h2c , (3)
in which the second inequality actually improves (1) by a simple algebraic
inequality. We also gave some applications of the double inequality (3) in [4].
Note that the following known relation (cf.[7])
2bc
ka = 2 ma (4)
b + c2
1)
East China Jiaotong University, Nanchang, China, [email protected]
2 J. Liu, Triangle inequalities

is equivalent to
2hb hc
ka = ma . (5)
h2b + h2c
We conclude that the second inequality of (3) is equivalent to
ka (mb + mc − ma ) ≥ hb hc . (6)
Motivated by inequality (6), the author has recently studied similar inequal-
ities in a triangle and proved some new results, which can be stated in the
following four theorems.
Theorem 1. In any triangle ABC the following double inequality holds
ha (mb + mc − ha ) ≥ hb hc ≥ ha (hb + hc − wa ). (7)
Equality holds in either (7) if and only if triangle ABC is equilateral.
Theorem 2. In any triangle ABC the following two inequalities hold:
ha (ma + ra − wa ) ≥ hb hc , (8)
wa (ma + ra − wa ) ≥ wb wc . (9)
Equality holds in either (8) and (9) if and only if triangle ABC is equilateral.
Theorem 3. In any triangle ABC the following two inequalities hold:
mb mc ≥ ha (mb + mc − wa ), (10)
mb mc ≥ ha (ma + ra − wa ). (11)
Equality holds in either (10) and (11) if and only if triangle ABC is equilat-
eral.
Theorem 4. In any triangle ABC the following two inequalities hold:
ma ra ≥ ha (mb + mc − wa ), (12)
ma ra ≥ ha (ma + ra − wa ). (13)
Equality holds in (12) if and only if triangle ABC is equilateral and equality
holds in (13) if and only if b = c.
From the inequalities given in the above theorems, we can obtain some
inequality chains. For example, by (10) and the first inequality of (7) we can
obtain the double inequality
mb mc hb hc
+ wa ≥ mb + mc ≥ + ha . (14)
ha ha
Also, it follows from inequalities (13), (8) and the second inequality of (7)
that
ma ra hb hc
+ wa ≥ ma + ra ≥ + wa ≥ hb + hc . (15)
ha ha
The purpose of this paper is to prove these theorems. We shall also
propose similar conjectural inequalities as open problems in the last section.
 Articles 3

2. Proof of Theorem 1.1

We first give the following lemma.

Lemma 5. In any triangle ABC it holds that
9
(mb + mc )2 ≥ a2 + h2a , (16)
4
with equality if and only if b = c.
Proof. We recall that for any triangle ABC the following known inequality
(cf. [3] or [5]) holds:
√
b + c ≥ a2 + 4h2a . (17)
Applying this inequality to the triangle GBC (G is the centroid of the triangle
ABC) and noting that the altitude from G is equal to ha /3, we get
√ ( )2
1
GB + GC ≥ a + 4 2 ha .
3
2 2
Also, we have BG = mb , CG = mc . Hence, inequality (16) can be ob-
3 3
tained immediately from the above inequality. Note that equality in (17)
holds if and only if b = c. Then we easily determine that the equality condi-
tion of (16) is the same as that of (17). This completes the proof. 2
Next, we prove Theorem 1.
Proof. We first prove the left inequality of (7), i.e.,
ha (mb + mc − ha ) ≥ hb hc , (18)
which is equivalent to
h2a (mb + mc )2 ≥ (h2a + hb hc )2 .
By Lemma 5, it is enough to prove that
h2a (9a2 + 4h2a ) − 4(h2a + hb hc )2 ≥ 0.
We denote by S the area of triangle ABC in the sequel. Using the formula
ha = 2S/a, one sees that the above inequality is equivalent to
A0 ≡ (bc)2 (9a4 + 16S 2 ) − 16S 2 (a2 + bc)2 ≥ 0. (19)
Applying Heron’s formula
√
S= s(s − a)(s − b)(s − c), (20)
where s = (a + b + c)/2, we easily obtain
A0 = a 2 A1 , (21)
4 J. Liu, Triangle inequalities

where
A1 =a6 − 2(b2 − bc + c2 )a4 + (b4 − 4b3 c + 7b2 c2 − 4bc3 + c4 )a2
+ 2(b − c)2 (b + c)2 bc.
Thus, we have to prove A1 ≥ 0. But A1 can be rewritten as
A1 = 2bc(a + b + c)(b + c − a)(b − c)2 + a2 (a2 − b2 − c2 + bc)2 , (22)
which shows A1 ≥ 0. Hence inequality (18) is proved. Furthermore, it is
easily seen that equality in (18) holds if and only if a = b = c.
We now prove the right inequality of (7), i.e.,
hb hc + wa ha ≥ ha (hb + hc ). (23)
Let I be the incenter of the triangle ABC. It is clear that wa ≥ AI + r, and
then
r
wa ≥ r + . (24)
A
sin
2
Therefore, to prove inequality (23) it is sufficient to prove that
rha
hb hc + ha r + ≥ ha (hb + hc ).
A
sin
2
Applying r = S/s and ha = 2S/a etc., one sees that the above inequality is
equivalent to ( )
2 1 1 2 1 1
+ + ≥ + .
bc sa A a b c
sa sin
2
Using s = (a + b + c)/2 and simplifying gives the equivalent inequality
A bc
sin ≤ 2 . (25)
2 b + c2 + bc − a2
In view of the well known formula
√
A (s − b)(s − c)
sin = , (26)
2 bc
for proving (25) we need to show
b3 c3 − (b2 + c2 + bc − a2 )2 (s − b)(s − c) ≥ 0.
Using once more s = (a + b + c)/2 and factoring gives
1
(a + b + c)(b + c − a)(b2 + c2 − bc − a2 )2 ≥ 0,
4
which clearly is true. Thus, inequalities (25) and (23) are proved. Also, it is
easy to deduce that the equality in (23) occurs only when a = b = c. This
completes the proof of Theorem 1. 2
 Articles 5

3. Proof of Theorem 2

We first give the following snappy double inequality.

Lemma 6. In any triangle ABC, we have
ma wa ≥ rb rc ≥ wa2 . (27)
Equality holds in either inequality of (27) if and only if b = c.
Proof. We recall (see [8] or [7] p. 223) that for any triangle ABC it holds
ma (b + c)2
≥ , (28)
wa 4bc
with equality if and only if b = c. Also, it is well known that the angle-bisector
wa is given by
2 √
wa = s(s − a)bc. (29)
b+c
Multiplying both sides of (28) by wa2 and using (29) and the following known
identity
rb rc = s(s − a), (30)
we obtain
ma wa ≥ rb rc . (31)
which is just the first inequality of (27). The second inequality of (27) can
be rapidly obtained from another known inequality (see [1]), viz.,
wa2 ≤ s(s − a), (32)
and identity (30). Moreover, we easily determine the equality condition of
(27). This completes the proof of Lemma 6. 2

Remark 1. Inequality (28) is equivalent to (31) and the following two in-
equalities:
A
4ma sin ≥ hb + hc , (33)
2
1 A
ma ≥ (b + c) cos . (34)
2 2
The later can be obtained from the identity
√
1 A A
ma = (b + c) cos2 + (b − c)2 sin2 . (35)
2 2 2
We are now in the position to prove Theorem 2.
Proof. We first prove inequality (8), which has the following two analogues:
hb (mb + rb − wb ) ≥ hc ha , (36)
hc (mc + rc − wc ) ≥ ha hb . (37)
6 J. Liu, Triangle inequalities

Without loss of generality we may suppose that a ≥ b ≥ c. Clearly, in

order to prove inequality (8) is valid for any triangle we must prove that all
inequalities (8), (36) and (37) hold under the hypothesis that a ≥ b ≥ c.
By ha = 2S/a and ra = S/(s − a), we get
4(a − b)(a − c)S 2
ha ra − hb hc = . (38)
abc(b + c − a)
Hence, we have ha ra ≥ hb hc , hb rb ≤ hc ha and hc rc ≥ ha hb under the hy-
pothesis a ≥ b ≥ c. Noting that ma ≥ wa and mc ≥ wc , we deduce that
inequalities (8) and (37) hold.
It remains to prove inequality (36) under the hypothesis a ≥ b ≥ c.
According to Lemma 6, it is sufficient to prove
( )
rc ra
hb + rb − wb ≥ hc ha ,
wb
i.e.,
hb (rc ra − wb2 ) ≥ wb (hc ha − hb rb ). (39)
By Lemma 6, we see that the left hand side of (39) is non-negative. Also,
the above inequality hb rb ≤ hc ha shows that the right hand side of (39) is
non-negative, too. Thus, for proving inequality (39) we need to prove that
B0 ≡ h2b (rc ra − wb2 )2 − wb2 (hc ha − hb rb )2 ≥ 0. (40)
With the help of software Maple, using the following known formulas
√
2 s(s − a)(s − b)(s − c)
ha = , (41)
a
√
s(s − b)(s − c)
ra = , (42)
s−a
and Heron’s formula, we easily obtain the identity
(a + b + c)2 S 2
B0 = B1 , (43)
4cab2 (c + a)4
where
B1 = − 4(c + a)2 b6 + 8(c + a)3 b5 − 24(c + a)2 cab4 − 8(c2 − 3ca
+ a2 )(c + a)3 b3 + (4c6 + 17c5 a − 12c4 a2 − 34c3 a3
− 12c2 a4 + 17ca5 + 4a6 )b2 − 2ca(c + a)(5c2 + 6ca
+ 5a2 )(c − a)2 cab + (c − a)2 (c + a)4 .
Consequently, we have to prove that B1 ≥ 0 holds under the hypothesis
a ≥ b ≥ c.
 Articles 7

Putting s−a = x, s−b = y and s−c = z, then we get a = y+z, b = z +x

and c = x + y (x, y, z > 0). Substituting a = y + z etc. into B1 and
reorganizing gives
B1 = − 64zxy 6 + 64(z + x)zxy 5 + (4z 4 + 32z 3 x − 8z 2 x2 + 32zx3
+ 4x4 )y 4 + 4(z + x)(z 4 − 12z 3 x − 10z 2 x2 − 12zx3 + x4 )y 3
− 4(−z + 3x)(−3z + x)(z + x)2 zxy 2 + 32(z + x)(z 2 + x2 )z 2 x2 y
− 16(z + x)2 z 3 x3 . (44)
Let us denote by B1 (x, y, z) the polynomial B1 , i.e., B1 = B1 (x, y, z). From
the hypothesis a ≥ b ≥ c we have x ≤ y ≤ z, thus we may set y = x + m, z =
x + m + n (m ≥ 0, n ≥ 0). Substituting them into B1 and reorganizing gives
B1 (x, x + m, x + m + n)
= 16(m2 + 6mn + n2 )(m − n)2 x4 + 8(m − n)(7m4 + 36m3 n + 2m2 n2
− 12mn3 − n4 )x3 + 4(18m5 + 77m4 n − 4m3 n2 − 50m2 n3
+ 2mn4 + 5n5 )mx2 + 8(m + n)(5m3 + 17m2 n + 3mn2 − n3 )m3 x
+ 4(2m + n)(m + n)4 m3 . (45)
Therefore, we only need to prove that B1 (x, x + m, x + m + n) ≥ 0 holds for
the positive number x, the non-negative numbers m and n. If m > n, the
inequality clearly holds. If n ≥ m, we may set n = m+t (t ≥ 0). Substituting
this into B1 (x, x + m, x + m + n), we can obtain the identity
B1 (x, x + m, x + 2m + t) = B2 t3 + 16m2 B3 , (46)
where
B2 = (8x3 + 20x2 m + 4m3 )t2 + (16x4 + 128x3 m + 108x2 m2 − 8xm3
+ 44m4 )t + 16(8x4 + 20x3 m + 2x2 m2 − xm3 + 12m4 )m,
B3 = (8x4 − 23x2 m2 + 10xm3 + 26m4 )t2 − 4m2 (x + m)(4x2 − xm
− 7m2 )t + 12m4 (x + m)2 .
As m ≥ 0, n ≥ 0, it is easy to show that
108x2 m2 − 8xm3 + 44m4 = 4m2 (27x2 − 2xm + 11m2 ) > 0,
2x2 m2 − xm3 + 12m4 = m2 (2x2 − xm + 12m2 ) > 0.
So we have B2 ≥ 0. Note that B3 is a quadratic function in t and 8x4 −
23x2 m2 + 26m4 > 0. Also, we easily compute its discriminant Ft as
Ft = −16m4 (8x4 + 8mx3 − 14m2 x2 + 16m3 x + 29m4 )(x + m)2 .
Note that 8x4 −14m2 x2 +29m4 > 0, hence Ft ≤ 0, and then inequality B3 ≥ 0
holds. Since B2 ≥ 0 and B3 ≥ 0, inequality B1 (x, x + m, x + 2m + t) ≥ 0
follows from (46). Hence, we have proved that for x > 0, m ≥ 0, n ≥ 0
8 J. Liu, Triangle inequalities

inequality B1 (x, x + m, x + m + n) ≥ 0 holds. We also proved that B1 ≥ 0

holds under the hypothesis a ≥ b ≥ c. This completes the proof of inequality
(8). Moreover, it is easy to determine that the equality in (8) holds if and
only if a = b = c.
Next, we prove inequality (9).
By the inequality ma wa ≥ rb rc given in Lemma 6, to prove inequality
(9) it suffices to show that

rb rc − wa2 ≥ wb wc − wa ra , (47)

which has the following two analogues:

rc ra − wb2 ≥ wc wa − wb rb , (48)

ra rb − wc2 ≥ wa wb − wc rc . (49)
Suppose that a ≥ b ≥ c. In this setting, we need to prove that all three
inequalities (47)–(49) hold. Using the previous formulas (29) and (42), we
easily obtain

(wa ra )2 − (wb wc )2
bc(a − b)(a − c)(c + a − b)(a + b − c)(a2 + 3ab + 3ac + b)(a + b + c)2
= .
4(b + c)2 (c + a)2 (a + b)2
Thus, one can see that if a ≥ b ≥ c then wa ra ≥ wb wc , wb rb ≤ wc wa and
wc rc ≥ wa wb hold. Then, we conclude by Lemma 6 that both inequalities
(47) and (49) are true. Thus, it remains to prove inequality (48). Noting that
the values of both sides of (48) are non-negative, because we have rc ra ≥ wb2
(which follows from Lemma 6) and wc wa ≥ wb rb we only need to prove

(rc ra − wb2 )2 − (wc wa − wb rb )2 ≥ 0,

i.e.,

C0 ≡ (rc ra )2 + wb4 + 2wa wb wc rb − 2rc ra wb2 − (wc wa )2 − (wb rb )2 ≥ 0. (50)

Applying (29) and Heron’s formula, we get

8abcsS
wa wb wc = . (51)
(b + c)(c + a)(a + b)
Substituting this into C0 and using (29) and (42), with the help of software
Maple we obtain the identity

(a + b + c)2
C0 = C1 , (52)
16(a + b)2 (b + c)2 (c + a)4
 Articles 9

where
C1 = (c4 − 8c3 a − 2c2 a2 − 8ca3 + a4 )b6 + 8(c + a)3 cab5
− (2c6 − 6c5 a + 30c4 a2 + 44c3 a3 + 30c2 a4 − 6ca5 + 2a6 )b4
− 2(c + a)(5c4 − 8c3 a − 10c2 a2 − 8ca3 + 5a4 )cab3
+ (c8 + 2c7 a + 17c6 a2 − 10c5 a3 − 36c4 a4 − 10c3 a5
+ 17c2 a6 + 2ca7 + a8 )b2 + 2(c + a)(c4 − 5c3 a − 8c2 a2
− 5ca3 + a4 )(c − a)2 cab + (c − a)2 (c + a)4 c2 a2 .
Consequently, we have to prove C1 ≥ 0 under the hypothesis a ≥ b ≥ c. If
we set s − a = x, s − b = y and s − c = z, then a = y + z, b = z + x, and
c = x + y (x, y, z > 0). Substituting a = y + z etc. into C1 and reorganizing
gives
C1 = − 64zxy 8 + (4z 4 + 96z 3 x + 56z 2 x2 + 96zx3 + 4x4 )y 6
+ 8(z + x)(3z 2 − 4zx + 3x2 )(z − x)2 y 5 + (52z 6 − 144z 5 x
− 132z 4 x2 + 64z 3 x3 − 132z 2 x4 − 144zx5 + 52x6 )y 4
+ 8(z + x)(6z 4 + z 3 x − 18z 2 x2 + zx3 + 6x4 )(z − x)2 y 3
+ 4(4z 6 − 4z 5 x − 15z 4 x2 + 46z 3 x3 − 15z 2 x4 − 4zx5
+ 4x6 )(z + x)2 y 2 + 16(z + x)(z − x)2 z 3 x3 y
− 16(z + x)2 z 4 x4 . (53)
From the hypothesis a ≥ b ≥ c we have x ≤ y ≤ z. Thus, it remains to prove
that C1 ≥ 0 holds for 0 < x ≤ y ≤ z. Let C1 = C1 (x, y, z). We may let
y = x + m, z = x + m + n (m ≥ 0, n ≥ 0). Then, we need to prove
C1 (x, x + m, x + m + n) ≥ 0. (54)
Making use of software Maple and expanding C1 (x, x + m, x + m + n), we
find that all the terms are non-negative. So, inequality (54) holds. This
completes the proof of inequality (9). Moreover, the above proof clearly
shows that equality in (9) holds if and only if a = b = c. Theorem 2 is
proved. 2

4. Proof of Theorem 3

In the proofs of both Theorem 3 and Theorem 4 we will use the following
lemma.
Lemma 7. In any triangle ABC the following inequality holds
mb mc ≥ wa ra , (55)
with equality if and only if triangle ABC is equilateral.
10 J. Liu, Triangle inequalities

Proof. By the first inequality of (27) and the fact that ma ≥ wa , we obtain

m2a ≥ rb rc , (56)
which was used in my recent paper [6]. Using two inequalities similar to (56)
and the second inequality of (27), we have
√ √ √
mb mc ≥ rc ra · ra rb = ra rb rc ≥ ra wa ,
which proves (55). It is easily seen that equality in (55) holds if and only if
a = b = c. This completes the proof of Lemma 7. 2

Remark 2. Inequality (55) can be extended to

mb mc ≥ wa ra ≥ hb hc ≥ ha (hb + hc − wa ), (57)
in which the second inequality could be proved by using the previous inequal-
ity (24), the third inequality is just the second inequality of (7).
Next, we prove Theorem 3.
Proof. We first prove inequality (10), that is
mb mc + wa ha ≥ ha (mb + mc ). (58)
Squaring both sides gives
(mb mc )2 + (wa ha )2 + 2mb mc wa ha ≥ h2a (mb + mc )2 .
By Lemma 7, it is sufficient to prove that
(mb mc )2 + (wa ha )2 + 2ra ha wa2 ≥ h2a (mb + mc )2 . (59)
Using Cauchy’s inequality, we have
( )
1 1
(mb + mc ) ≤
2
(bm2b + cm2c ) + . (60)
b c
Thus to prove (59) we need to show that
[ ]
D0 ≡ bc (mb mc )2 + ha (ha + 2ra )wa2 − (b + c)(bm2b + cm2c )h2a ≥ 0. (61)
With the help of software Maple, using the previous formulas (29), (41), (42)
and the well-known median formula
4m2a = 2b2 + 2c2 − a2 , (62)
we easily obtain
D1
D0 = , (63)
16a2 (b+ c)2
 Articles 11

where
D1 = (2b4 + 12b3 c + 16b2 c2 + 12bc3 + 2c4 )a6 − 8(b + c)b2 c2 a5
− (5b6 + 15b5 c + 15b4 c2 + 26b3 c3 + 15b2 c4 + 15bc5 + 5c6 )a4
+ 16(b + c)(b2 + c2 )b2 c2 a3 + (4b6 − 4b4 c2 + b3 c3
− 4b2 c4 + 4c6 )(b + c)2 a2 − 8(b − c)2 (b + c)3 b2 c2 a
− (b2 + bc + c2 )(b − c)4 (b + c)4 .
Consequently, we have to prove D1 ≥ 0. Putting s−a = x, s−b = y, s−c = z,
then we have a = y + z, b = z + x, c = x + y. Substituting the later three
relations into D1 and reorganizing gives
D1 = 4(y + z)2 x8 + 16(y + z)(y 2 + 18yz + z 2 )x7 + (9y 4 + 816y 3 z
+ 846y 2 z 2 + 816yz 3 + 9z 4 )x6 − (y + z)(29y 4 − 688y 3 z + 358y 2 z 2
− 688yz 3 + 29z 4 )x5 − (29y 6 − 9y 5 z + 971y 4 z 2 + 1762y 3 z 3
+ 971y 2 z 4 − 9yz 5 + 29z 6 )x4 + (y + z)(9y 6 − 152y 5 z − 921y 4 z 2
− 752y 3 z 3 − 921y 2 z 4 − 152yz 5 + 9z 6 )x3 + (16y 6 − 13y 5 z
− 213y 4 z 2 + 404y 3 z 3 − 213y 2 z 4 − 13yz 5 + 16z 6 )(y + z)2 x2
+ (4y 6 + 24y 5 z + 31y 4 z 2 + 282y 3 z 3 + 31y 2 z 4 + 24yz 5
+ 4z 6 )(y + z)3 x + (y + 2z)2 (2y + z)2 (y + z)4 yz. (64)
Consequently, we need to prove D1 ≥ 0 for the positive real numbers x, y, z.
We set D1 = D1 (x, y, z). Since D1 (x, y, z) is symmetric with respect to y
and z we may suppose that y ≥ z and let y = z + m (m ≥ 0). With the help
of Maple, we easily obtain
D1 (x, z + m, z)
= 4(z + x)m9 + 8(3z + x)(3z + 2x)m8 + (569z 3 + 547z 2 x
+ 147zx2 + 9x3 )m7 + (2591z 4 + 2600z 3 x + 358z 2 x2
− 80zx3 − 29x4 )m6 + (7491z 5 + 8635z 4 x − 78z 3 x2
− 1742z 2 x3 − 165zx4 − 29x5 )m5 + (14257z 6 + 19778z 5 x
− 1353z 4 x2 − 8868z 3 x3 − 1361z 2 x4 + 514zx5 + 9x6 )m4
+ 4(z + x)(4464z 6 + 2977z 5 x − 3416z 4 x2 − 1994z 3 x3
+ 460z 2 x4 + 209zx5 + 4x6 )m3 + 4(3546z 6 − 166z 5 x
− 3435z 4 x2 − 168z 3 x3 + 664z 2 x4 + 86zx5 + x6 )(z + x)2 m2
+ 16(z − x)(405z 4 + 90z 3 x − 196z 2 x2 − 58zx3
− x4 )(z + x)3 zm + 16(81z 2 + 38zx + x2 )(z − x)2 (z + x)4 z 2 . (65)
12 J. Liu, Triangle inequalities

Thus, we have to prove D1 (x, z +m, z) ≥ 0 for positive numbers x, z and non-
negative number m. If z > x, then it is easy to see that D1 (x, z + m, z) ≥ 0
holds. If z ≤ x, we may let x = z + n (n ≥ 0) and substitute it into
D1 (x, z + m, z). Then, we can obtain the identity
D1 (z + n, z + m, z)
= m8 z 8 + m7 z 7 + m6 z 6 + m5 z 5 + m4 z 4
+ m3 z 3 + m2 z 2 + m1 z + m0 , (66)
where
m8 = 8448m2 − 30720mn + 30720n2 ,
m7 = 21632m3 − 62720m2 n + 15360mn2 + 71680n3 ,
m6 = 22976m4 − 44608m3 n − 54336m2 n2 + 117248mn3 + 66816n4 ,
m5 = 14112m5 − 12352m4 n − 63616m3 n2 + 50816m2 n3
+ 122368mn4 + 31232n5 ,
m4 = 5440m6 + 2448m5 n − 30848m4 n2 − 1824m3 n3 + 75312m2 n4
+ 55808mn5 + 7424n6 ,
m3 = 1272m7 + 2960m6 n − 6584m5 n2 − 8992m4 n3 + 20584m3 n4
+ 32688m2 n5 + 12160mn6 + 768n7 ,
m2 = 4(m + n)(40m7 + 177m6 n − 191m5 n2 − 482m4 n3 + 818m3 n4
+ 1213m2 n5 + 268mn6 + 4n7 ),
m1 = 2m(4m8 + 52m7 n + 87m6 n2 − 98m5 n3 − 155m4 n4 + 284m3 n5
+ 482m2 n6 + 192mn7 + 8n8 ),
m0 = (m + n)(m − n)2 (2m + n)2 (m + 2n)2 m2 n.
Clearly, in order to prove D1 (z + n, z + m, z) ≥ 0 we only need to show that
the coefficients m8 , m7 , . . . , m0 are all non-negative. As m ≥ 0 and n ≥ 0, it
is clear that m0 ≥ 0. Also, it is easy to show that m8 ≥ 0 and m7 ≥ 0. In
addition, we can rewrite m6 , m5 , m4 and m3 as follows:
m6 = m(22976m + 47296n)(m − 2n)2 + 64(671m2 − 1124mn
+ 1044n2 )n2 , (67)
m5 = m(m − 2n)2 (14112m2 + 44096mn + 56320n2 )
+ 128(779m2 − 804mn + 244n2 )n3 , (68)
m4 =m(5440m + 24208n)(m − n)4 + 16(464n + 1975m)n5
+ 16(2084m2 − 7832mn + 10419n2 )m2 n2 , (69)
 Articles 13

m3 = m(636m + 3706n)(2m + n)(m − n)4 + 2(7135m2 − 18684mn

+ 14634n2 )m3 n2 + 2(19732m2 + 4227mn + 384n2 )n5 . (70)
Note that 671m2 − 1124mn + 1044n2 ≥ 0 etc. One sees that m6 , m5 , m4
and m3 are non-negative. Furthermore, it is easy to verify the following two
identities:
2(40m7 + 177m6 n − 191m5 n2 − 482m4 n3 + 818m3 n4
+ 1213m2 n5 + 268mn6 + 4n7 )
= m(2m + n)(40m + 317n)(m − n)4
+ (3020m2 + 219mn + 8n2 )n5
+ (1517m2 − 3420mn + 2350n2 )m3 n2 , (71)

4m8 + 52m7 n + 87m6 n2 − 98m5 n3 − 155m4 n4 + 284m3 n5

+ 482m2 n6 + 192mn7 + 8n8
= m(2m2 + 33mn + 151n2 )(m − n)4 (2m + n)
+ (751m2 + 41mn + 8n2 )n6
+ (699m2 − 1293mn + 650n2 )m3 n3 . (72)
We can easily show that both right hand sides of the above two identities are
non-negative. So, we have m2 ≥ 0 and m1 ≥ 0. And D1 (z + n, z + m, z) ≥ 0
follows from identity (66). We thus proved that inequality D1 (x, z +m, z) ≥ 0
is valid for non-negative numbers m, n and positive number x. This completes
the proof of inequality (58).
Next, we prove inequality (11), that is
mb mc + wa ha ≥ ha (ma + ra ). (73)
Squaring both sides gives
(mb mc )2 + (wa ha )2 + 2mb mc wa ha ≥ h2a (ma + ra )2 .
By Lemma 7, to prove the above inequality we need to show
(mb mc )2 + (wa ha )2 + 2ra ha wa2 ≥ h2a (ma + ra )2 ,
or equivalently
(mb mc )2 + (wa ha )2 + 2ra ha wa2 − h2a (m2a + ra2 ) ≥ 2ma ra h2a . (74)
By the simplest arithmetic mean–geometric mean inequality we have
m2a
ha + ≥ 2ma . (75)
ha
Thus we only need to prove
E0 ≡ (mb mc )2 +(wa ha )2 +2ra ha wa2 −h2a (m2a +ra2 )−ra h3a −ra ha m2a ≥ 0. (76)
14 J. Liu, Triangle inequalities

Making use of formulas (29), (41), (42), and (62), with the help of software
we obtain
E1
E0 = , (77)
16a (b + c)2
3

where
E1 = 4(b2 + bc + c2 )a7 − 8(b + c)bca6 + (b2 + 6bc + c2 )(b − c)2 a5
− 2(b + c)(b − c)4 a4 − (b2 + bc − c2 )(b2 − bc − c2 )(b + c)2 a3
+ (3b2 − 2bc + 3c2 )(b − c)2 (b + c)3 a2 − (b − c)4 (b + c)5 .
To prove inequality (76) we need to show E1 ≥ 0. Putting s − a = x, s − b =
y, s − c = z, we easily show that E1 ≥ 0 is equivalent to
E1 ≡ 4(y + z)3 x6 + (12y 4 + 176y 3 z − 184y 2 z 2 + 176yz 3 + 12z 4 )x5
− (y + z)(3y 4 − 380y 3 z + 514y 2 z 2 − 380yz 3 + 3z 4 )x4
− 2(13y 2 − 38yz + 13z 2 )(y 2 − 6yz + z 2 )(y + z)2 x3
− 3(y 4 − 10y 3 z + 62y 2 z 2 − 10yz 3 + z 4 )(y + z)3 x2
+ 2(6y 4 + 5y 3 z − 16y 2 z 2 + 5yz 3 + 6z 4 )(y + z)4 x
+ (4y 4 + 12y 3 z − 15y 2 z 2 + 12yz 3 + 4z 4 )(y + z)5 ≥ 0. (78)
Through analysis, we find the following identity which can be verified by
expanding:
4E1 = t1 + t2 x(x − y)2 (x − z)2 + t3 xyz(y + z − 2x)2
+ yz(a0 x2 + b0 x + c0 ), (79)
where
t1 = 4y 5 (33x2 + 20xy + 4y 2 )(x − y)2 + 4z 5 (33x2 + 20zx + 4z 2 )(x − z)2 ,
t2 = 16(y + z)3 x + 80(y 4 + z 4 ) + 32yz(26y 2 − 17yz + 26z 2 ),
t3 = (y + z)(805y 2 − 859yz + 805z 2 )x + 3(y 2 + yz + z 2 )(y − z)2
+ 2(351y 4 − 242y 2 z 2 + 351z 4 ),
a0 = (y + z)(2259y 4 + 381y 3 z − 4208y 2 z 2 + 381yz 3 + 2259z 4 ),
b0 = − 473y 6 − 1167y 5 z − 1087y 4 z 2 + 1166y 3 z 3 − 1087y 2 z 4
− 1167yz 5 − 473z 6 ,
c0 = 4(y + z)(32y 6 + 53y 5 z + 44y 4 z 2 + 10y 3 z 3 + 44y 2 z 4 + 53yz 5 + 32z 6 ).
As x, y, z > 0, we have t1 > 0. Also, it is easy to show that t2 > 0 and
t3 > 0. Clearly, to prove E1 ≥ 0 it remains to prove
a0 x2 + b0 x + c0 ≥ 0. (80)
 Articles 15

The left hand side is a quadratic function in x. Note that a0 > 0 and the
discriminant Fx is given by
Fx =b20 − 4a0 c0
= − (932879y 10 + 5185696y 9 z + 12185266y 8 z 2 + 16635726y 7 z 3
+ 16037951y 6 z 4 + 15621924y 5 z 5 + 16037951z 6 y 4 + 16635726z 7 y 3
+ 12185266z 8 y 2 + 5185696yz 9 + 932879z 10 )(y − z)2 .
It is clear Fx ≤ 0, thus inequality (80) holds. This completes the proof of
inequality (11). Moreover, it is easy to determine that the equality in (11)
holds if and only if a = b = c. Theorem 3 is proved. 2

5. Proof of Theorem 4

Proof. We first prove inequality (12), that is

ma ra + wa ha ≥ ha (mb + mc ). (81)
Squaring both sides gives
(ma ra )2 + (wa ha )2 + 2ma wa ra ha ≥ h2a (mb + mc )2 .
By Lemma 7, we only need to prove that
(ma ra )2 + (wa ha )2 + 2rb rc ra ha ≥ h2a (mb + mc )2 ,
which, by the previous inequality (60) follows from
[ ]
F0 ≡ bc (ma ra )2 + (wa ha )2 + 2rb rc ra ha − (b + c)(bm2b + cm2c )h2a ≥ 0. (82)
Applying the previous formulas (29), (41), (42), and (62), we can obtain the
identity
(c + a − b)(a + b − c)(a + b + c)
F0 = F1 , (83)
16(b + c − a)a2 (b + c)2
where
F1 = − (2b4 + 7b3 c + 14b2 c2 + 7bc3 + 2c4 )a4 + 2(b + c)(2b4 + 7b3 c
+ 14b2 c2 + 7bc3 + 2c4 )a3 − (b2 + 6bc + c2 )(b2 + 3bc
+ c2 )(b + c)2 a2 − 2(b4 − 2b3 c − 2b2 c2 − 2bc3
+ c4 )(b + c)3 a + (b2 + bc + c2 )(b − c)2 (b + c)4 .
Thus, we need to show F1 ≥ 0. After analyzing, we obtain the nice identity
F1 = x(x2 − yz)2 F2 + (y − z)2 F3 , (84)
16 J. Liu, Triangle inequalities

where x = s − a, y = s − b, z = s − c and
F2 =64(y + z)x2 + (112y 2 + 32yz + 112z 2 )x + 64y 3 + 64z 3 ,
F3 = 20(y 2 + 6yz + z 2 )x4 + 12(y + z)(y 2 + 10yz + z 2 )x3
+ (5y 4 + 40y 3 z + 22y 2 z 2 + 40yz 3 + 5z 4 )x2
+ (y + z)(y 2 + 14yz + z 2 )(y − z)2 x + (y + z)4 yz.
As x, y, z > 0, we have F1 ≥ 0. This completes the proof of inequality (81).
Also, the above proof immediately implies that equality in (81) holds if and
only if a = b = c.
Finally, we prove inequality (13), i.e.,
ma ra + wa ha ≥ ha (ma + ra ), (85)
which is equivalent to
(ma ra )2 + (wa ha )2 + 2ma wa ra ha ≥ h2a (m2a + ra2 + 2ra ma ).
According to the first inequality ma wa ≥ rb rc of (27) and the inequality
1 m2a
ma ≤ (rb + rc ) + , (86)
4 rb + rc
which follows from the simplest arithmetic mean–geometric mean inequality,
we only need to prove
(ma ra )2 + (wa ha )2 + 2rb rc ra ha
[ ]
1 2ra
≥ h2a m2a + ra2 + ra (rb + rc ) + m2a ,
2 rb + rc
i.e.,
[ ]
G0 ≡ 2(rb + rc ) (ma ra )2 + (wa ha )2 + 2rb rc ra ha
[ ]
− h2a 2(rb + rc )(m2a + ra2 ) + ra (rb + rc )2 + 4ra m2a . (87)
Making use of the previous formulas (29), (41), (42), and (62), we obtain the
factorization
(a + b + c)(b − c)2 S
G0 = G1 , (88)
2(b + c − a)a2 (b + c)2
where
G1 = a5 − 2(b + c)a4 + (b + c)2 a3 + 2(b + c)3 a2 − (3b2 + 2bc
+ 3c2 )(b + c)2 a + 2(b2 + c2 )(b + c)3 .
Thus, to prove inequality (85) we need to prove the strict inequality G1 > 0.
But it is easy to get the following identity:
G1 = 32x5 + 48(y + z)x4 + 8(5y 2 + 6yz + 5z 2 )x3 + 4(y + z)(7y 2 + 6yz
+ 7z 2 )x2 + 8(y 2 + yz + z 2 )(y + z)2 x + (y + z)5 , (89)
 Articles 17

where x = s − a, y = s − b, z = s − c. So we have G1 > 0 and inequality

(13) is proved. From Lemma 7 and identity (88), we immediately deduce
that equality in (13) holds if and only if b = c. This completes the proof of
Theorem 4. 2

6. Several conjectures

In this section, we shall propose several conjectures as open problems.

Conjecture 1. In any triangle ABC the following double inequality holds:
ha (ma + ra − ha ) ≥ wa ra ≥ ha (mb + mc − ma ). (90)
Conjecture 2. In any triangle ABC the following two inequalities hold:
wa (mb + mc − ma ) ≥ hb hc , (91)
ha (mb + mc − wa ) ≥ hb hc . (92)
Remark 3. In the acute triangle ABC we have wa ≥ ka etc., thus by
inequality (6) we know that inequality (91) holds for the acute triangle.
Conjecture 3. In any triangle ABC the following two inequalities hold:
wb wc ≥ ha (wb + wc − wa ), (93)
wb wc ≥ ha (mb + mc − ma ). (94)
Conjecture 4. In any triangle ABC the following two inequalities hold:
mb mc ≥ wa (mb + mc − ma ), (95)
ma ra ≥ wa (mb + mc − ma ). (96)

References
[1] O. Bottema, R.Ž. Djordjević, R.R. Janić, D.S. Mitrinović, P.M. Vasić, Geometric
inequalities, Wolters–Noordhoff Press, Groningen, 1969.
[2] J. Liu, Two half-symmetry inequalities involving medians and altitudes of a triangle,
(in Chinese) Middle School Mathematics Teaching 1 (2004), 18–19.
[3] J. Liu, A new proof of the Erdős–Mordell inequality, Int. Electron. J. Geom. 4 (2011),
114–119.
[4] J. Liu, A refinement of Panaitopol’s inequality and some half-symmetry inequality
chains, (in Chinese) Elementary Mathematics Research in China 2 (2019), 98–112.
[5] J. Liu, Two new refinements of a linear geometric inequality, Int. J. Geom. 9 (2020),
81–92.
[6] J. Liu, A beautiful inequality chain for a triangle, Int. J. Open Problems Compt. Math.
15 (2022), 54–69.
[7] D.S. Mitrinović, J.E. Pečarić, V.Volenec, Recent Advances in Geometric Inequalities,
Kluwer Academic Publishers, Dordrecht–Boston–London, 1989.
[8] G. Tsintsifas, L. Bankoff, L. Goldstone, Problem E 2471, Amer. Math. Monthly, 81
(1974) and 82 (1975), 523–524.
18 D. Popa, Refined asymptotic expansions for some recurrent sequences

Refined asymptotic expansions for some recurrent sequences

Dumitru Popa1)

Abstract. Let a ∈ R and f : (a, ∞) → R be a continuous function such

that f (x) > x for all x > a, and (xn )n≥1 the sequence defined by x1 > a
and xn+1 = f (xn ) for all n ≥ ( 1. We prove that )if there exist b0 , b1 , b2 ∈ R,
b0 ̸= 0, such that lim x2 f (x) − x − b0 − bx1 = b2 , then there exists
( x→∞ )
C := lim xn − b0 n − bb10 · ln n ∈ R. Moreover,
n→∞ ( ) ( )
b2 b2
xn = b0 n + bb01 · ln n + C + b31 · lnnn − 2b
b1
0
+ b2 −b
b 2
1C
− b31 · n1 + o n1 .
0 0 0
Many and various concrete examples are given.
Keywords: Recurrent sequences, asymptotic expansion of a function, as-
ymptotic expansion of a sequence, Cesàro lemma.
MSC: 35C20, 11B37, 40A05, 40A25.

1. Introduction and a preliminary result

In the recent paper [5] we have proved that there is a connection between
the asymptotic expansions of functions and the asymptotic expansions of
some recurrent sequences; several similar results can be found in [1, 6]. We
recall Theorem 2 from the paper [5].
Theorem 1. Let a ∈ R and f : (a, ∞) → R be a continuous function such
that f (x) > x for all x > a. Let (xn )n≥1 be the sequence defined by x1 > a
and xn+1 = (f (xn ) for all n ≥ 1. ) If there exist b0 , b1 , b2 ∈ R, b0 ̸= 0, such
that lim x f (x) − x − b0 − x = b2 , then there exists C ∈ R such that
2 b1
x→∞
b2 ( )
xn = b0 n + bb01 · ln n + C + b13 · lnnn + o lnnn .
0

A look at the statement of Theorem 1 reveals the absence of b2 in the

conclusion. The main purpose of this paper is to complete Theorem 1, see
Theorem 3. All our notation and notions are standard, see [3]. We need the
following well-known evaluation. We include one of its possible proofs.
Proposition 2. The following evaluation holds
( )
ln (n + 1) ln n ln n 1 1
− =− 2 + 2 +o .
n+1 n n n n2
( ) ( )
Proof. From ln (n + 1) = ln n + ln 1 + n1 = ln n + n1 + o n1 we deduce
ln(n+1) (1) (1)
n+1 = n+1 + n2 + o n2 . From n+1 = n − n2 + n3 + o n3 we deduce
ln n 1 1 1 1 1
( ) ( )
ln n ln n ln n ln n ln n ln n ln n 1
= − 2 + 3 +o = − 2 +o
n+1 n n n n3 n n n2
1) Department of Mathematics, Ovidius University, Constanţa, Romania,
[email protected]
 Articles 19

and the proof is finished. 2

2. The main result

Theorem 3. Let a ∈ R and f : (a, ∞) → R be a continuous function such
that f (x) > x for all x > a. Let (xn )n≥1 be the sequence defined by x1 > a
and xn+1 = (f (xn ) for all n ≥ 1. ) If there exist b0 , b1 , b2 ∈ R, b0 ̸= 0, such
that lim x f (x) − x − b0 − x = b2 , then there exists
2 b1
x→∞ ( )
C := lim xn − b0 n − bb10 · ln n ∈ R. Moreover,
n→∞
( ) ( )
b1 b2 ln n b1 b2 − b1 C b21 1 1
xn = b0 n + · ln n + C + 13 · − + − · +o .
b0 b0 n 2b0 b20 b30 n n
b2
Proof. For all n ≥ 1 let us put zn = xn − b0 n − bb01 · ln n − C − b13 · lnnn . Using
( ) ( ) 0
xn+1 = f (xn ), ln 1 + n1 = n1 − 2n1 2 + o n12 and Proposition 2, we get
( ) ( )
b1 1 b21 ln (n + 1) ln n
zn+1 − zn = xn+1 − xn − b0 − ln 1 + − 3 −
b0 n b0 n+1 n
( ) ( ) ( )
b1 1 1 b21 ln n 1 1
= f (xn ) − xn − b0 − − − 3 − 2 + 2 +o
b0 n 2n2 b0 n n n2
( ) ( )
b1 b1 b1 1 1
= f (xn ) − xn − b0 − + − −
xn xn b0 n 2n2
( ) ( )
b21 ln n 1 1
− 3 − 2 + 2 +o . (1)
b0 n n n2
( )
Since lim xn = ∞, the hypothesis lim x2 f (x) − x − b0 − bx1 = b2 im-
n→∞ ( x→∞)
plies that lim x2n f (xn ) − xn − b0 − xb1n = b2 . Passing to the limit in the
n→∞
equality
( ) ( ) 2
b1 b1 n
n f (xn ) − xn − b0 −
2
= xn f (xn ) − xn − b0 −
2
· 2
x x xn
and remembering that lim xnn = b0 , we get
n→∞
( )
b1 b2
lim n f (xn ) − xn − b0 −
2
= 2
n→∞ xn b0
or equivalently
( )
b1 b2 1 1
f (xn ) − xn − b0 − = 2 · 2 +o . (2)
xn b0 n n2
20 D. Popa, Refined asymptotic expansions for some recurrent sequences

From (1) and (2) we deduce that

( ) ( )
b1 b1 1 b21 ln n b2 b1 b21 1 1
zn+1 − zn = − · + · + 2+ − · 2 +o . (3)
xn b0 n b30 n2 b0 2b0 b30 n n2
( )
From C := lim xn − b0 n − bb10 · ln n , that is xn = b0 n + bb10 · ln n + C + o (1),
n→∞
we get 1
xn = 1
b0 n · b1 ln n
1
. But
1+ · n + bC · n
1 1
( )
+o n
b2
0 0

( )
1 b1 ln n C 1 1
(1) = 1 − 2 · − · +o
1+ b1
b20
· ln n
n + bC0 · 1
n +o n b0 n b0 n n

and hence
( )
b1 b1 b2 ln n b1 C 1 1
= − 31 · 2 − 2 · 2 + o . (4)
xn b0 n b0 n b0 n n2
Relations (3) and (4) give us that
( ) ( )
b1 b2 − b1 C b21 1 1
zn+1 − zn = + 2 − 3 · 2 +o ,
2b0 b0 b0 n n2
that is
zn+1 − zn b1 b2 − b1 C b21
lim 1 = + − . (5)
n→∞
n2
2b0 b20 b30
[ ]
Since by Theorem 1, lim zn = 0 from the Cesàro lemma in the case 00 ,
n→∞ ( )
b2 −b1 C b21
see [2], relation (5) gives us that lim z1n = − 2b b1
+ b20
− b30
, that is,
( ) n→∞ n
( )
0
b2
zn = − 2b b1
0
+ b2 −b
b2
1C
− b13 · n1 + o n1 and the proof is finished. 2
0 0

3. Some applications
We apply in the sequel Theorem 3 to refine asymptotic evaluations of
some recurrent sequences from the paper [5].
Theorem 4. Let φ : [0, ∞) → R be a twice differentiable function such that
φ (x) > 0 for all x ≥ 0. Let (xn )n≥1 be the sequence defined by x1 > 0 and
( )
xn+1 = xn + φ x1n for all n ≥ 1. Then there exists C ∈ R such that

φ′ (0) [φ′ (0)]2 ln n

xn = φ (0) n + · ln n + C + ·
φ (0) [φ (0)]3 n
( ) ( )
φ′ (0) φ′′ (0) − 2Cφ′ (0) [φ′ (0)]2 1 1
− + 2 − 3 · +o .
2φ (0) 2 [φ (0)] [φ (0)] n n
 Articles 21
(1)
Proof. Let f : (0, ∞) → R, f (x) = x + φ x . In this case φ (0) = b0 > 0,
φ′′ (0) b21 φ′ (0) ′′ (0)−2Cφ′ (0) ′ 2
φ′ (0) = b1 , 2 = b2 , and b1
2b0 + b2 −b
b2
1C
− b30
= 2φ(0) +φ 2[φ(0)]2
− [φ (0)]
[φ(0)]3
.
0
We apply now Theorem 3. 2

Theorem 5. Let φ : [0, ∞) → R be a three times differentiable function

such that φ (x) > 1 for all x > 0, φ (0) = 1, φ′ (0) ̸= 0. Let (xn )n≥1 be the
( )
sequence defined by x1 > 0 and xn+1 = xn φ x1n for all n ≥ 1. Then there
exists C ∈ R such that
φ′′ (0) [φ′′ (0)]2 ln n
xn = φ′ (0) n + ′ · ln n + C + ·
2φ (0) 4 [φ′ (0)]3 n
( ) ( )
φ′′ (0) φ′′′ (0) − 3Cφ′′ (0) [φ′′ (0)]2 1 1
− ′
+ − · +o .
2φ (0) ′
6 [φ (0)]2 ′
4 [φ (0)] 3 n n
(1) ′ φ′′ (0)
Proof. Let f : (0, ∞) → R, f (x) = xφ x . We have φ (0) = b0 , 2 = b1 ,
φ′′′ (0) b21 φ′′ (0) φ′′′ (0)−3Cφ′′ (0) [φ′′ (0)]2
6
b1
= b2 , and 2b 0
+ b2 −b
b20
1C
− = b30
−
2φ′ (0) + 6[φ′ (0)]2 4[φ′ (0)]3
. From
Theorem 3 we deduce the evaluation from the statement. 2
( )
Proposition 6. (i) If αn3 = o n12 , then for all r ∈ R \ {0}, (1 + αn )r =
( )
1 + rαn + r(r−1)
2 αn2 + o n12 .
(ii) Let a, b, c, d ∈ R. Then for all r ∈ R \ {0}
( ( ))r
ln n 1 ln n 1 1 ln n 1
1+a· +b· +c· 2 +d· 2 +o 2
= 1 + ra · + rb · +
n n n n n n n
[ ] ( )
r (r − 1) a2 ln2 n ln n r (r − 1) b2 1 1
· 2 +[abr (r − 1) + rc]· 2 + + rd · 2 +o .
2 n n 2 n n2

Proof. (i) We will use the well-known evaluation ( (1 +) x)r = 1+rx+ r(r−1) 2
2 x +
r(r−1)(r−2) 3 ( )
6 x + o x3 as x → 0. Since αn = o n2/3 1
, lim αn = 0, and hence
r r(r−1) r(r−1)(r−2) ( ) n→∞
( )
(1 + αn ) = 1+rαn + 2 αn + 2 αn +o αn . Now from αn3 = o n12
3 3
6 (1)
we get (1 + αn )r = 1 + rαn + r(r−1) 2
2 αn + o n2 .
(ii) From (i) it follows that
( ( ))r
ln n 1 ln n 1 1
1+a· +b· +c· 2 +d· 2 +o
n n n n n2
( )
ln n 1 ln n 1
=1+r a· +b· +c· 2 +d· 2
n n n n
( )2 ( )
r (r − 1) ln n 1 ln n 1 1
+ a· +b· +c· 2 +d· 2 +o . (6)
2 n n n n n2
22 D. Popa, Refined asymptotic expansions for some recurrent sequences

Also we have
( ) ( ) ( )
ln n 1 ln n 1 2 ln n 1 2 1
a· +b· +c· 2 +d· 2 = a· +b· +o
n n n n n n n2
2 ( )
2 ln n ln n 1 1
= a · 2 + 2ab · 2 + b · 2 + o
2
. (7)
n n n n2
Replacing (7) in (6), we get the stated evaluation. 2

Proposition 7. Let α, β, γ, δ, C (∈ R, ) α ̸= 0, and r ∈ R \ {0}. If an =

αn + β ln n + C + γ · lnnn + δ · n1 + o n1 , then
rβ ln n rC 1 r (r − 1) β 2 ln2 n
arn = αr nr + · + · + · 2−r
α1−r n1−r α1−r n1−r 2α2−r n
( )
r (r − 1) βC + rαγ ln n r (r − 1) C 2 + 2rαδ 1 1
+ · 2−r + · 2−r + o .
α2−r n 2α2−r n n2−r
Proof. From Proposition 6 we have
( ( ))r
β ln n C 1 γ ln n δ 1 1
an = α n 1 + ·
r r r
+ · + · 2 + · 2 +o
α n α n α n α n n2
(
rβ ln n rC 1 r (r − 1) β 2 ln2 n
= α r nr 1 + · + · + · 2
α n α n 2α2 n
[ ] [ ]
r (r − 1) βC rγ ln n r (r − 1) C 2 rδ 1
+ 2
+ · 2 + 2
+ · 2
α α n 2α α n
( ))
1
+o
n2
(
rβ ln n rC 1 r (r − 1) β 2 ln2 n
= α r nr 1 + · + · + · 2
α n α n 2α2 n
( ))
r (r − 1) βC + rαγ ln n r (r − 1) C + 2rαδ 1
2 1
+ 2
· 2 + 2
· 2 +o .
α n 2α n n2
By calculations we get the stated evaluation. 2
In the next result we refine the evaluation from the exercise 2.8 in [4].
Corollary 8. Let (xn )n≥1 be the sequence defined by x1 > 0 and xn+1 =
xn + x1n for all n ≥ 1. Then there exists K ∈ R such that
√ √ 1 ln n 1 1 ln2 n
xn = 2 n+ √ · √ +K · √ − √ · √
4 2 n n 64 2 n n
√ ( √ )2 ( )
2K 2 − 1 ln n 2K 2 − 1 1 1
− √ · √ − √ · √ +o √ .
16 2 n n 16 2 n n n n
 Articles 23

Proof. Squaring the recurrence relation, we obtain x2n+1 = x2n +2+ x12 , ∀n ≥ 1.
n ( )
For all n ≥ 1 we denote x2n = an and thus an+1 = an + 2 + a1n = an + φ a1n ,
where φ (x) = 2 + x. From Theorem 4 there exists C ∈ R such that
φ′ (0) [φ′ (0)]2 ln n
an = φ (0) n + · ln n + C + ·
φ (0) [φ (0)]3 n
( ) ( )
φ′ (0) φ′′ (0) − 2Cφ′ (0) [φ′ (0)]2 1 1
− + 2 − 3 · +o .
2φ (0) 2 [φ (0)] [φ (0)] n n
( )
or, x2n = 2n + 21 · ln n + C + 18 · lnnn + 2C−1 8 · n1 + o n1 . By Proposition 7
( ( ))1/2
1 1 ln n 2C − 1 1 1
xn = 2n + · ln n + C + · + · +o
2 8 n 8 n n
√ √ β ln n C 1 β2 ln2 n
= α n+ √ · √ + √ · √ − √ · √
2 α n 2 α n 8α α n n
( )
2αγ − βC ln n 4αδ − C 2 1 1
+ √ · √ + √ · √ +o √ ,
4α α n n 8α α n n n n
where α = 2, β = 12 , γ = 18 , δ = 2C−1
8 . Hence
√ √ 1 ln n C 1 1 ln2 n
xn = 2 n+ √ · √ + √ · √ − √ · √
4 2 n 2 2 n 64 2 n n
2 ( )
C − 1 ln n (C − 1) 1 1
− √ · √ − √ · √ +o √ .
16 2 n n 16 2 n n n n
If we denote C
√
2 2
= K, we get the stated evaluation. 2

Proposition 9. Let A > 0 and g : [0, A) → R be a function with the property ( )

that there exist a1 , a2 , a3 ∈ R such that g (x) = x+a1 x2 +a2 x3 +a3 x4 +o x4
as x → 0, x > 0. Then
1 1 ( ) ( ) ( )
= − a1 + a21 − a2 x − a31 − 2a1 a2 + a3 x2 + o x2 as x → 0, x > 0.
g (x) x
In particular,
( )
1 a2 − a2 a31 − 2a1 a2 + a3 1
( 1 ) = x − a1 + 1 − +o as x → ∞.
g x x x2 x2
1
(a21 −a2 )x
− x1 +a1 − ( )
Proof. By direct calculus lim g(x)
x2
= − a31 − 2a1 a2 + a3 .
x→0,x>0
A different proof can be found in [5, Proposition 16]. 2

Theorem 10. Let A > 0, g : [0, A) → [0, ∞) be a continuous function such

that 0 < g(x) < x for all 0 < x < A. Let (xn )n≥1 be the sequence defined by
24 D. Popa, Refined asymptotic expansions for some recurrent sequences

x1 ∈ (0, A) and xn+1 = g(xn ) for all n ≥ 1. If there exist the( real ) numbers
a1 , a2 , a3 , a1 ̸= 0 such that g (x) = x + a1 x + a2 x + a3 x + o x as x → 0,
2 3 4 4

x > 0, then there exists K ∈ R such that

( 2 )2
1 1 a21 − a2 ln n 1 a1 − a2 ln2 n
xn = − · + · + K · − ·
a1 n a3 n2 n2 a51 n3
( 2 ) ( 21 )
a − a2 a1 − a2 − 2a31 K ln n
+ 1 · 3
a51 n
( ) ( )
2a4 K 2 − 2a31 a21 − a2 K + a41 − a21 a2 + 2a1 a3 − 2a22 1 1
− 1 · + o .
2a51 n3 n3
( )
Proof. Let f : A1 , ∞ → R, f (x) = g 11 . Then f is continuous and from
(x)
0 < g(x) < x, ∀0 < x < A, it follows that f (x) > x, ∀x > A1 . For
all n ≥ 1 we define vn = x1n . Then v1 > A1 and the recurrence relation
becomes vn+1 = f (vn ) for all ( 1 n) ≥ 1. From Proposition 9 it follows that
f (x) = x + b0 + x + x2 + o x2 as x → ∞, where b0 = −a1 , b1 = a1 − a2 ,
b1 b2 2

b2 = −a31 + 2a1 a2 − a3 . From Theorem 3 we have

( )
ln n 1 1
vn = αn + β ln n + C + γ +δ +o ,
n n n
( )
b2 b2 −b1 C b21
where α = b0 , β = bb10 , γ = b13 , δ = − 2b b1
0
+ b2
− b3
. By Proposition 7
0 0 0

1 β
ln n C 1 ln n 2βC − γα ln n
β2 2
xn = 1
vn = − 2· 2
− 2· 2+ 3· 3 + · 3
αn α n α( n) α n α3 n
C 2 − αδ
1 1
+ · 3 +o .
α3 n n3
( 2 )2
a21 − a2 a1 − a2
Since α = −a1 , β = − ,γ=− , and
a1 a31
( 3 ) ( ) ( 2 )2
a21 − a2 a1 − 2a1 a2 + a3 + a21 − a2 C a1 − a2
δ= + − ,
2a1 a21 a31
we obtain
( )( )
2βC − γα a21 − a2 a21 − a2 + 2a1 C
= ,
α3 a51
2
a21 −a2 (a31 −2a1 a2 +a3 )+(a21 −a2 )C (a21 −a2 )
C2 − αδ C2 + 2 + a1 − a21
= − 3
α3 a1
( )
2a21 C 2 + 2a1 a21 − a2 C + a41 − a21 a2 + 2a1 a3 − 2a22
= − .
2a51
 Articles 25

Hence
( 2 )2
1 1 a21 − a2 ln n C 1 a1 − a2 ln2 n
xn = − · + 3 · 2 − 2· 2− 5 · 3
a1 n a n a1 n a1 n
( 2 ) ( 21 )
a − a2 a1 − a2 + 2a1 C ln n
+ 1 · 3
a51 n
( )
2a2 C 2 + 2a1 a21 − a2 C + a41 − a21 a2 + 2a1 a3 − 2a22 1
− 1 · 3
2a51 n
( )
1
+o .
n3
C
If we denote K = − , we get the stated evaluation. 2
a21
In the next result we refine the evaluation from Exercise 2.3 in [4].
Corollary 11. Let (xn )n≥1 be the sequence defined by x1 > 0 and xn+1 =
ln(1 + xn ) for all n ≥ 1. Then there exists K ∈ R such that
1 2 ln n 1 2 ln2 n 2 (3K − 1) ln n
xn = 2 · + · 2 +K · 2 + · 3 + · 3
n 3 n n 9 n 9 n
( )
18K 2 − 3K + 1 1 1
− · 3 +o .
144 n n3
2 3 4 ( )
Proof. As is well known, ln (1 + x) = x − x2 + x3 − x4 + o x4 as x → 0, i.e.,
a1 = − 12 , a2 = 31 , a3 = − 41 . By Theorem 10 we get
( 2 )2
1 1 a21 − a2 ln n 1 a1 − a2 ln2 n
xn = − · + · + K · − ·
a1 n a3 n2 n2 a51 n3
( 2 ) ( 21 )
a − a2 a1 − a2 − 2a31 K ln n
+ 1 · 3
a51 n
( ) ( )
2a41 K 2 − 2a31 a21 − a2 K + a41 − a21 a2 + 2a1 a3 − 2a22 1 1
− 5 · 3 +o .
2a1 n n3
a21 − a2 2 (a1 − a2 )
2 2
We have a21 − a2 = − 12
1
, = 3 , = − 92 , a41 − a21 a2 + 2a1 a3 −
a31 a51
( )( ) 3K − 1 ( )
1
2a22 = 144 , a21 − a2 a21 − a2 − 2a31 K = − , 2a41 K 2 − 2a31 a21 − a2 K
144
18K 2 − 3K + 1
+ a1 − a1 a2 + 2a1 a3 − 2a2 =
4 2 2 , and therefore we get the stated
144
evaluation. 2
In the end we refine the asymptotic evaluation for the logistic map.
26 A. Negrescu, V. Pop, R. Strugariu, SEEMOUS 2023

Corollary 12. Let (xn )n≥1 be the sequence defined by x1 ∈ (0, 1) and xn+1 =
xn − x2n for all n ≥ 1. Then there exists K ∈ R such that
( )
1 ln n 1 ln2 n ln n 2K 2 + 2K + 1 1 1
xn = − 2 +K · 2 + 3 −(2K + 1)· 3 + · 3 +o .
n n n n n 2 n n3
Proof. We apply Theorem 10 for the function g : [0, 1) → R, g (x) = x − x2
with a1 = −1, a2 = a3 = 0. 2

References
[1] N.G. de Bruijn, Asymptotic methods in analysis, Dower Publications, Inc., New York,
1981.
[2] E. Cesàro, Sur une question de limites, Mathesis X, 25-28 (1890).
[3] I. Colojoară, Analiză matematică, Editura Didactică şi Pedagogică, Bucureşti, 1983.
[4] D. Popa, Exerciţii de analiză matematică, Colecţia Biblioteca S.S.M.R., Editura Mira,
Bucureşti 2007.
[5] D. Popa, Recurrent sequences and the asymptotic expansion of a function, Gazeta
Matematică, Seria A, Nr. 3-4, 37 (2019), 1–16.
[6] S. Stević, Asymptotic behaviour of a sequence defined by iteration, Mat. Vesn. No. 3-4,
48 (1996), 99–105.

17th South Eastern European Mathematical Olympiad for

University Students, SEEMOUS 2023
Alexandru Negrescu1) , Vasile Pop2) , Radu Strugariu3)

Abstract. The 17th South Eastern European Mathematical Olympiad for

University Students (SEEMOUS 2023) took place on March 7–12, 2023,
in Struga, North Macedonia. We present the competition problems and
their solutions, as given by the authors, or in the line of the official ones.
Alternative solutions provided by members of the jury or by contestants
are also included.
Keywords: Trace, commutator, Jacobson’s Lemma, normal matrix,
Schur’s Triangularization Theorem, Riemann integral, integral conver-
gence, differentiable function, sequences of real numbers, series of real
numbers.
MSC: Primary 15A21; Secondary 40A05, 26A24, 26A42.

The Mathematical Society of South-Eastern Europe (MASSEE) orga-

nized in 2023 the 17th South Eastern European Mathematical Competition
for University Students with International Participation (SEEMOUS 2023),
which is addressed to students in the first or second year of undergraduate
studies, from universities in countries that are members of MASSEE or from
invited countries that are not affiliated to MASSEE.
1) University Politehnica of Bucharest, Romania, [email protected]
2) Technical University of Cluj-Napoca, Romania, [email protected]
3) Gheorghe Asachi Technical University of Iaşi, Romania, [email protected]
 Articles 27

This year’s competition was hosted between March 7–12, 2023, in Struga
by the Union of Mathematicians of Macedonia and the Faculty of Natural
Sciences and Mathematics, Ss. Cyril and Methodius University in Skopje.
The number of students that participated in the contest was 85, representing
24 universities from Albania, Bulgaria, Greece, North Macedonia, Romania,
and Turkmenistan. The jury awarded 10 gold medals, 21 silver medals and
30 bronze medals. The student Mario–Cosmin Drăguţ from University of
Bucharest, Romania, was the absolute winner of the competition, being the
only contestant that obtained full marks to all four problems.
We present the problems from the contest and their solutions as given
by the corresponding authors, together with alternative solutions provided
by members of the jury or by the contestants.
Problem 1. Prove that if A and B are n × n square matrices with complex
entries satisfying
A = AB − BA + A2 B − 2ABA + BA2 + A2 BA − ABA2 ,
then det (A) = 0.

***
Authors’ solution. Let k be an arbitrary positive integer. By multiplying
the previous equality with Ak−1 , we obtain
Ak = Ak B − Ak−1 BA + Ak+1 B − 2Ak BA + Ak−1 BA2 + Ak+1 BA − Ak BA2 .
Taking the trace and considering that Tr (M N ) = Tr (N M ), for all M, N ∈
Mn (C), we deduce
( ) ( ) (( ) ) ( ) (( ) )
Tr Ak = Tr Ak B − Tr Ak−1 B A + Tr Ak+1 B − 2 Tr Ak B A
(( ) ) (( ) ) (( ) )
+ Tr Ak−1 B A2 + Tr Ak+1 B A − Tr Ak B A2
= 0.
( )
Therefore, we proved that Tr Ak = 0 for any positive integer k, which
implies (see [3], p. 144) that the matrix A is nilpotent, so det (A) = 0.
Alternative solution. Suppose on the contrary that det (A) ̸= 0. By
multiplying on the left the equality from the statement with A−1 , we obtain
In = B − A−1 BA + AB − 2BA + A−1 BA2 + ABA − BA2 .
As in the previous solution, taking the trace and considering that Tr (M N ) =
Tr (N M ), for all M, N ∈ Mn (C), we deduce
(( ) )
Tr (In ) = Tr (B) − Tr A−1 B A + Tr (AB) − 2 Tr (BA)
( ( )) ( )
+ Tr A−1 BA2 + Tr (A (BA)) − Tr BA2
= 0,
28 A. Negrescu, V. Pop, R. Strugariu, SEEMOUS 2023

which is a contradiction. Hence det (A) = 0.

Second alternative solution. The following solution was given by Carol–
Luca Gasan and Horia Mercan, from University Politehnica of Bucharest
(contestants).
Let us consider the linear operator ∆A : Mn (C) → Mn (C), with
∆A (X) = AX − XA. With this notation, the equality in the statement can
be rewritten
A = ∆A (B) + ∆A (AB) − ∆A (BA) + ∆A (ABA) ,
so A = ∆A (B + AB − BA + ABA). As an obvious consequence, the ma-
trix A commute with the matrix ∆A (B + AB − BA + ABA). Therefore,
considering Jacobson’s Lemma (see [1], p. 126), we obtain that the matrix
∆A (B + AB − BA + ABA) is nilpotent, that is, the matrix A is nilpotent,
and the conclusion follows immediately.

Problem 2. For the sequence

1 1 1
Sn = √ +√ + ··· + √ ,
2
n +1 2 2
n +2 2 n + n2
2

find ( )
( ( √ ) ) 1
lim n n ln 1 + 2 − Sn − √ ( √ ) .
n→∞ 2 2 1+ 2

***
Author’s solution. Observe that, if f is a C2 function on [0, 1], by Taylor
formula, we can write
∫ 1 ∑n ∫ k ∑n ( ( ) ( ))
n k k−1
f (x) dx = f (x) dx = F −F
0 k−1 n n
k=1 n k=1
∑n ( ( ) ( ) )
1 k−1 1 ′ k−1 1 ′′
= f + 2f + 3 f (ck,n ) ,
n n 2n n 6n
k=1
( )
where F is an antiderivative of f and ck,n ∈ k−1 k
n , n . Moreover,
( ) ( )
1∑ 1∑
n n
k k−1 f (0) − f (1)
f = f − ,
n n n n n
k=1 k=1
hence
(∫ ( ))
1∑
1 n
k
n f (x) dx − f
0 n n
k=1
∑
n ( ( ) )
1 ′ k−1 1 ′′
= f + 2 f (ck,n ) + (f (0) − f (1)) . (1)
2n n 6n
k=1
 Articles 29

Furthermore, also by Taylor formula,

∑n ( ( ) ( ))
k k−1
f (1) − f (0) = f −f
n n
k=1
∑n ( ( ) )
1 ′ k−1 1 ′′
= f + 2 f (dk,n ) , (2)
n n 2n
k=1
( )
with dk,n ∈ k−1 k
n , n . By combining relations (1) and (2), we obtain
( (∫ ( )) )
1∑
n
1
k f (0) − f (1)
n n f (x) dx − f −
0 n n 2
k=1
( n ( )
∑ 1 (k − 1) 1
)
f (1) − f (0)
=n f′ + 3 f ′′ (ck,n ) −
2n n 6n 2
k=1
( n ( ) )
∑ 1 ′′ 1
=n 2
f (ck,n ) − 2 f ′′ (dk,n )
6n 4n
k=1

1 1∑n
1 1 ∑ ′′
n
= · f ′′ (ck,n ) − · f (dk,n ) ,
6 n 4 n
k=1 k=1
which converges to
∫ ∫
1 1 ′′ 1 1 ′′ 1 ( ′ )
f (x) dx − f (x) dx = − f (1) − f ′ (0) .
6 0 4 0 12
1
In the particular case of f (x) = √1+x 2
, we obtain that
∫ 1 ( √ )
f (x) dx = ln 1 + 2 ,
0
( )
1∑
n
k
Sn = f ,
n n
k=1
f (0) − f (1) 1
= √ (√ ),
2 2 2 2+1
i.e., the requested limit can be obtained by the procedure described before.
Therefore, the result is
1 ( ′ ) 1
− f (1) − f ′ (0) = √ .
12 24 2
Remark. This problem is closely related to a well-known result (see [2], p.
7), stating that if f is a C 1 function on [0, 1] , then
( n ( ) ∫ )
1∑ k 1
f (1) − f (0)
lim n f − f (x) dx = ,
n→∞ n n 0 2
k=1
30 A. Negrescu, V. Pop, R. Strugariu, SEEMOUS 2023

[ ]
whose proof uses the Taylor expansion on each interval k−1 k
n , n for an anti-
derivative of f. So, the idea and the path of the proof described before should
naturally have been visible, and for this reason the problem was considered
to be of easy to medium difficulty. The presented proof follows the author’s
official solution, but tries to better emphasize the natural way to solve the
problem.
Problem 3. Prove that: if A is n × n square matrix with complex entries
such that A + A∗ = A2 A∗ , then A = A∗ . (For any matrix M , denote by
t
M ∗ = M , the conjugate transpose of M .)
Mihai Opincariu, Avram Iancu National College, Brad, Romania
Vasile Pop, Technical University of Cluj-Napoca, Romania
Authors’ solution. We will show first that the matrix A is normal, i.e.,
AA∗ = A∗ A. The equality A + A∗ = A2 A∗ leads to
( )
A = A2 − I n A∗ , (3)
wherefrom A ± In = (A − In ) (A + In ) A∗ ± In , so
(A − In ) [(A + In ) A∗ − In ] = In and (A + In ) [In − (A − In ) A∗ ] = In ,
relations that imply that the matrices A − In and A + In are invertible. Then
( )−1
A2 − In is also invertible and by (3) it follows that A∗ = A2 − In A. On
( 2 )−1
the other hand, using the Cayley–Hamilton theorem, we get that A − In
is a polynomial of A2 − In , hence a polynomial of A, so AA∗ = A∗ A.
Since the matrix A is normal, it is unitarily diagonalizable, that is,
there exists an unitary matrix U ∈ Mn (C) and a diagonal matrix D =
diag (λ1 , . . . , λn ) such that A = U DU ∗ . Then A∗ = U DU ∗ , which, due to
the equality in the statement, leads to D + D = D2 D, from where we find
that λi + λi = λ2i λi , for all i ∈ 1, n. Then 2 Re (λi ) = λi |λi |2 , so λi ∈ R, for
all i ∈ 1, n, hence D = D and the conclusion follows immediately.
Alternative solution. The following solution was given by George–Vlad
Manolache and Horia Mercan, from University Politehnica of Bucharest (con-
testants).
By multiplying on the right with A∗ , the equality in the statement
implies that AA∗ + (A∗ )2 = A2 (A∗ )2 , so the conjugate transposes of the
matrices in the two members of the previous( )∗ equality are also equal, that
is, AA∗ + A2 = A2 (A∗ )2 , hence A2 = A2 . By multiplying on the left
with A∗ , the equality in the (statement
) implies that A∗ A + (A∗ )2 = A∗ A2 A∗
∗
and, considering that A2 = A2 , we obtain that A∗ A + (A∗ )2 = (A∗ )4 =
A2 (A∗ )2 = AA∗ + (A∗ )2 , so A∗ A = AA∗ , which means that the matrix A is
normal.
As in the previous solution, there exists an unitary matrix U ∈ Mn (C)
and a diagonal matrix D = diag (λ1 , . . . , λn ) such that A = U DU ∗ . From
 Articles 31
( )∗ ( )∗
A2 = A2 we obtain that D2 = D2 , that is λ2i = λ2i , so λ2i ∈ R, for all
i ∈ 1, n. ( )∗
Considering that A2 = A2 , the equality in the statement also implies
that A + A∗ = (A∗ )3 , so (A + A∗ )∗ = A3 , that is A + A∗ = A3 , which means
( )∗ ( )∗
that A3 = A3 . Thus D3 = D3 , that is λ3i = λ3i , so λ3i ∈ R, for all
i ∈ 1, n.
Taking into account those established in the last two paragraphs, we
obtain that λi ∈ R, for all i ∈ 1, n, and, as in the previous solution, the
conclusion follows immediately.

Second alternative solution. The following solution of the fact that the
matrix A is normal was given by Carol–Luca Gasan, from University Po-
litehnica of Bucharest (contestant).
It can be shown (for example, as in the previous solution) that A2 =
∗ 2
(A ) . According to Schur’s Triangularization Theorem (see [1], p. 101),
there exists an unitary matrix U ∈ Mn (C) and an upper triangular matrix
T ∈ Mn (C) such that A = U T U ∗ . It immediately follows that T 2 = (T ∗ )2 .
On the other hand, the equality in the statement implies that T +T ∗ = T 2 T ∗ ,
so T = (T ∗ )3 − T ∗ , which means that the matrix T is normal. Since T is
an upper triangular matrix, it follows (see [1], p. 132) that T is a diagonal
matrix and the conclusion is obvious.

Third alternative solution. The following solution of the fact that the ma-
trix A is normal was given by Alexandru–Constantin Ariton, from University
Politehnica of Bucharest (contestant). ( )∗
The equality in the statement implies that (A + A∗ )∗ = A2 A∗ , so
A + A∗ = A (A∗ )2 . It follows that A2 A∗ = A (A∗ )2 .
By multiplying on the right with A, the equality in the statement implies
that A2 + A∗ A = A2 A∗ A. By multiplying on the left with A, the equality in
the statement implies that A2 + AA∗ = A3 A∗ . Subtracting the last equality
from the previous one, we obtain that A∗ A − AA∗ = A2 (A∗ A − AA∗ ).
Suppose on the contrary that there exists λ ̸= 0, eigenvalue of the
matrix A∗ A − AA∗ , and x ̸= 0, a corresponding eigenvector. That is,
(A∗ A − AA∗ ) x = λx, wherefrom we obtain that A2 (A∗ A − AA∗ ) x = λx,
so λA2 x = λx. Hence A2 x = x, which means that ω = 1 is an eigenvalue
of the matrix A2 , so ω = 1 is also an eigenvalue of the matrix (A∗ )2 . Hence
there exists y ̸= 0 such that (A∗ )2 y = y.
On the other hand, the equality A + A∗ = A (A∗ )2 implies that Ay +
A∗ y = A (A∗ )2 y. Then Ay + A∗ y = Ay, so A∗ y = 0. As an immediate
consequence, we obtain that y = (A∗ )2 y = 0, which is a contradiction.
Hence, all the eigenvalues of the self-adjoint matrix A∗ A − AA∗ are equal to
zero, which means that A∗ A − AA∗ = On , that is, the matrix A is normal.
32 A. Negrescu, V. Pop, R. Strugariu, SEEMOUS 2023

Problem 4. Let f : R → R be a continuous, strictly decreasing function

such that f ([0, 1]) ⊂ [0, 1] .
(a) For all n ∈ N∗ , prove that there exists a unique an ∈ (0, 1) , solution
of the equation
f (x) = xn . (4)
Moreover, if (an ) is the sequence defined as above, prove that lim an = 1.
n→∞
(b) Suppose f is C 1 with f (1) = 0 and f ′ (1) < 0. For any x ∈ R, we
define
∫ 1
F (x) = f (t) dt.
x
∞
∑
Study the convergence of the series (F (an ))α , with α ∈ R.
n=1

Radu Strugariu, Gheorghe Asachi Technical University of Iaşi, Romania

Author’s solution. (a) Consider the continuous function g : [0, 1] → R
given by g (x) = f (x) − xn , and observe that g (0) = f (0) > 0, and g (1) =
f (1) − 1 < 0. It follows the existence of an ∈ (0, 1) such that g (an ) = 0. For
uniqueness, observe that if would exist two solutions of the equation (4), say
an < bn , we would obtain
f (an ) > f (bn ) ⇔ ann > bnn ⇔ an > bn ,
a contradiction.
We prove that the sequence (an ) is strictly increasing. If it would exist
n ∈ N∗ such that an ≥ an+1 , we would obtain that
f (an ) ≤ f (an+1 ) ⇔ ann ≤ an+1 n
n+1 < an+1 ,

since f is strictly decreasing and an+1 ∈ (0, 1) . It follows that an < an+1 ,
a contradiction. Hence, (an ) is strictly increasing and bounded above by 1,
so it converges to ℓ ∈ (0, 1] . Suppose, by contradiction, that ℓ < 1. Since
f (an ) = ann for any n, using the continuity of f it follows that f (ℓ) = 0
for ℓ < 1, contradicting the fact that f is strictly decreasing with f (1) ≥ 0.
Hence, lim an = 1.
n→∞

(b) Observe that F is well-defined, of class C 2 , with F (1) = 0, F ′ (x) =

−f (x) ⇒ F ′ (1) = 0, F ′′ (x) = −f ′ (x) ⇒ F ′′ (1) > 0. Moreover, notice that
F (x) > 0 on [0, 1) . Using the Taylor formula on the interval [an , 1] , it follows
that for any n, there exist cn , dn ∈ (an , 1) such that
F ′′ (cn ) F ′′ (cn )
F (an ) = F (1) + F ′ (1) (an − 1) + (an − 1)2 = (an − 1)2 ,
2 2
f (an ) = f (1) + f ′ (dn ) (an − 1) = f ′ (dn ) (an − 1) . (5)
 Articles 33

Hence, since cn → 1 and F is C 2 , we obtain

(1 − an )2 2
lim = ′′ ∈ (0, ∞) ,
n→∞ F (an ) F (1)
so, due to the comparison test,
∞
∑ ∞
∑
(F (an ))α ∼ (1 − an )2α .
n=1 n=1
But
(an − 1)
lim n (1 − an ) = − lim n · · ln an
n→∞ n→∞ ln (1 + (an − 1))
( )
= − lim ln ann = − lim ln f (an ) = − ln lim f (an ) = ∞.
n→∞ n→∞ n→∞
∞
∑ ∞
∑
It follows that (1 − an ) diverges and, furthermore, (1 − an )2α diverges
n=1 n=1
for any 2α ≤ 1.
Remark that
n (1 − an ) (an − 1) n ln an (an − 1)
=− · = → 1.
− ln f (an ) ln (1 + (an − 1)) − ln f (an ) ln (1 + (an − 1))
Next, consider arbitrary γ ∈ (0, 1) . Using (5) and the fact that dn → 1,
we obtain
lim nγ (1 − an ) = lim [n (1 − an )]γ · (1 − an )1−γ
n→∞ n→∞
[ ]
γ f (an ) 1−γ
= lim [n (1 − an )] ·
n→∞ −f ′ (dn )
1 [ ]1−γ
γ
= · lim [− ln f (a n )] · eln f (an )
.
(−f ′ (1))1−γ n→∞
Observe that
xγ
− ln f (an ) → ∞ and lim
= 0,
e(1−γ)x
x→∞

hence lim nγ (1 − an ) = 0. So, if α > 12 , we obtain that there exists ε > 0

n→∞
1+ε
such that 2α > 1 + ε, hence for γ := 2α < 1, we get
lim n2αγ (1 − an )2α = lim n(1+ε) (1 − an )2α = 0.
n→∞ n→∞
∞
∑
Using the comparison test, it follows that the series (1 − an )2α converges.
n=1
∞
∑
In conclusion, the series (F (an ))α converges iff α > 12 .
n=1
34 A. Negrescu, V. Pop, R. Strugariu, SEEMOUS 2023

References
[1] R.A. Horn, C.R. Johnson, Matrix Analysis, Second Edition, Cambridge University
Press, New York, 2013.
[2] W.J. Kaczor, M.T. Nowak, Problems in Mathematical Analysis III. Integration, Amer-
ican Mathematical Society, Rhode Island, 2003.
[3] F. Zhang, Matrix Theory. Basic Results and Techniques, Second Edition, Springer, New
York, 2011.
 Mathematical Notes 35

MATHEMATICAL NOTES
A series involving a product of three consecutive harmonic
numbers
Ovidiu Furdui1) , Alina Sı̂ntămărian2)

Abstract. In this paper we calculate the following remarkable series

∑∞ Hn Hn+1 Hn+2
n=1 n(n+1)(n+2) , where Hn denotes the nth harmonic number. We
prove that this series equals a rational linear combination of ζ(2), ζ(3),
and ζ(4).
Keywords: Harmonic series, product of harmonic numbers, Euler’s series,
Goldbach’s series, Riemann zeta function values.
MSC: 40A05, 40C99

1. Introduction and the main result

∑
∞
Hn Hn+1 Hn+2
In this paper we calculate the remarkable series n(n+1)(n+2) , where
n=1
Hn = 1 + 12 + · · · + n1 denotes the nth harmonic number. This paper is
motivated by the following series formula (see [1]) involving the product of
two consecutive harmonic numbers
∞
∑ Hn Hn+1
= ζ(2) + 2ζ(3). (1)
n(n + 1)
n=1
We extend this result by calculating the corresponding series involving the
product of consecutive harmonic numbers Hn , Hn+1 and Hn+2 .
The main result of this paper, which answers to an open problem posed
in [3, pag. 119], is the following theorem.
Theorem 1. The following identity holds
∞
∑ Hn Hn+1 Hn+2 1 5 5
= − ζ(2) + ζ(3) + ζ(4).
n(n + 1)(n + 2) 2 4 8
n=1

Before we prove the theorem we compile the following lemma.

Lemma 2. An Euler and a Goldbach series
The following identities hold:
∑∞
Hn
(a) = 2ζ(3);
n2
n=1

1) Department of Mathematics, Technical University of Cluj-Napoca, Romania,

[email protected]
2) Department of Mathematics, Technical University of Cluj-Napoca, Romania,
[email protected]
36 O. Furdui, A. Sı̂ntămărian, A series involving a product

∞
∑ Hn 5
(b) = ζ(4).
n3 4
n=1

Proof. A proof of the lemma can be found in [3, pp. 238–240]. The two
parts of the lemma also appear as problems 3.55 and 3.58 in [2]. We mention
that the series in part (a) is due to Euler and the sum in part (b) is due to
Goldbach, and thus we correct the statement of problem 3.58 in [2], where
the problem is wrongly attributed to Klamkin. 2
Proof. (Proof of Theorem 1.) Let xn = Hn(n+1)(n+2)
n Hn+1 Hn+2
. We calculate the series
∑∞
n=1 xn by showing that it telescopes. We have
( )
Hn Hn+1 Hn+2 Hn Hn+1 Hn+2 1 1
= −
n(n + 1)(n + 2) 2 n(n + 1) (n + 1)(n + 2)
[ ( )]
1 Hn Hn+1 Hn+2 Hn+1 Hn+2 1 1 1
= − Hn+3 − − −
2 n(n + 1) (n + 1)(n + 2) n+1 n+2 n+3
( )
1 Hn Hn+1 Hn+2 Hn+1 Hn+2 Hn+3
= −
2 n(n + 1) (n + 1)(n + 2)
( )
1 Hn+1 Hn+2 Hn+1 Hn+2 Hn+1 Hn+2
+ + + .
2 (n + 1)2 (n + 2) (n + 1)(n + 2)2 (n + 1)(n + 2)(n + 3)

We calculate

Hn+1 Hn+2 Hn+1 Hn+2

+
(n + 1) (n + 2) (n + 1)(n + 2)2
2
( )
1 1
= Hn+1 Hn+2 −
(n + 1)2 (n + 1)(n + 2)
( )
1 1
+ Hn+1 Hn+2 −
(n + 1)(n + 2) (n + 2)2
Hn+1 Hn+2 Hn+1 Hn+2
= −
(n + 1)2 (n + 2)2
( ) ( )
Hn+1 Hn+1 + n+2 1
Hn+2 − n+21
Hn+2
= −
(n + 1)2 (n + 2)2
2
Hn+1 2
Hn+2 Hn+1 Hn+2
= 2
− 2
+ +
(n + 1) (n + 2) (n + 1) (n + 2) (n + 2)3
2
2 2 ( )
Hn+1 Hn+2 Hn+2 1 1
= − + + Hn+1 −
(n + 1)2 (n + 2)2 (n + 2)3 (n + 1)2 (n + 1)(n + 2)
2 2 ( )
Hn+1 Hn+2 Hn+2 Hn+1 1 1
= − + + − Hn+1 −
(n + 1)2 (n + 2)2 (n + 2)3 (n + 1)2 n+1 n+2
 Mathematical Notes 37
( )
Hn+1 Hn+2 − n+2
2 2 1
Hn+1 Hn+2 Hn+2 Hn+1
= 2
− 2
+ 3
+ −−
(n + 1) (n + 2) (n + 2) (n + 1)2
n+1 n+2
2 2 ( )
Hn+1 Hn+2 Hn+2 Hn+1 Hn+1 Hn+2
= − + + − −
(n + 1)2 (n + 2)2 (n + 2)3 (n + 1)2 n+1 n+2
1
− .
(n + 2)2

Now we calculate
( )
Hn+1 Hn+2 Hn+1 Hn+2 1 1
= −
(n + 1)(n + 2)(n + 3) 2 (n + 1)(n + 2) (n + 2)(n + 3)
 ( )
1 Hn+1 Hn+2 Hn+2 H n+3 − 1
n+2 − 1
n+3
=  − 
2 (n + 1)(n + 2) (n + 2)(n + 3)
(
1 Hn+1 Hn+2 Hn+2 Hn+3
= −
2 (n + 1)(n + 2) (n + 2)(n + 3)
)
Hn+2 Hn+2
+ + .
(n + 2)2 (n + 3) (n + 2)(n + 3)2
(2)
We have
( )
Hn+2 Hn+2 1 1
+ = Hn+2 −
(n + 2)2 (n + 3) (n + 2)(n + 3)2 (n + 2)2 (n + 2)(n + 3)
( )
1 1
+ Hn+2 −
(n + 2)(n + 3) (n + 3)2
Hn+2 Hn+2
= −
(n + 2)2 (n + 3)2
Hn+2 Hn+3 1
= 2
− 2
+ .
(n + 2) (n + 3) (n + 3)3
(3)
Combining (2) and (3) we have that

Hn+1 Hn+2
(n + 1)(n + 2)(n + 3)
[ ]
1 Hn+1 Hn+2 Hn+2 Hn+3 Hn+2 Hn+3 1
= − + − + .
2 (n + 1)(n + 2) (n + 2)(n + 3) (n + 2)2 (n + 3)2 (n + 3)3

It follows that
38 O. Furdui, A. Sı̂ntămărian, A series involving a product

( )
1 Hn Hn+1 Hn+2 Hn+1 Hn+2 Hn+3
xn = −
2 n(n + 1) (n + 1)(n + 2)
( )
1 Hn+1 Hn+2 Hn+2 Hn+3
+ −
4 (n + 1)(n + 2) (n + 2)(n + 3)
( )
1 Hn+2 Hn+3 1
+ 2
− 2
+
4 (n + 2) (n + 3) 4(n + 3)3
( 2 2 )
1 Hn+1 Hn+2 Hn+1 Hn+2
+ 2
− 2
+ 2
+
2 (n + 1) (n + 2) 2(n + 1) 2(n + 2)3
( )
1 Hn+1 Hn+2 1
− − − .
2 n+1 n+2 2(n + 2)2
This implies, based on Lemma 2, that
∑∞
Hn Hn+1 Hn+2 1 H1 H2 H3 1 H2 H3 1 H3
= · + · + ·
n(n + 1)(n + 2) 2 2 4 6 4 9
n=1
( )
1 1 1 1 H2
+ ζ(3) − 1 − 3 − 3 + · 2
4 2 3 2 4
( ) ( )
1 H1 1 5 H1 H2
+ 2ζ(3) − 2 + ζ(4) − 3 − 3
2 1 2 4 1 2
( )
1 H2 1 1
− · − ζ(2) − 1 − 2
2 2 2 2
1 5 5
= − ζ(2) + ζ(3) + ζ(4),
2 4 8
and the theorem is proved. 2
Motivated by formula (1) and by Theorem 1 we conjecture that, if k ≥ 1
is an integer, then
∞
∑ Hn Hn+1 · · · Hn+k
= ak,2 ζ(2) + ak,3 ζ(3) + · · · + ak,k+2 ζ(k + 2),
n(n + 1) · · · (n + k)
n=1
where ak,i , i = 2, . . . , k + 2, are rational numbers.

References
[1] O. Furdui, Series involving products of two harmonic numbers, Math. Mag. 84 (5),
2011, 371–377.
[2] O. Furdui, Limits, Series and Fractional Part Integrals, Problems in Mathematical
Analysis, Springer, New York, 2013.
[3] A. Sı̂ntămărian, O. Furdui, Sharpening Mathematical Analysis Skills, Springer, Cham,
2021.
 Problems 39

PROBLEMS

Authors should submit proposed problems to [email protected].

Files should be in PDF or DVI format. Once a problem is accepted and considered
for publication, the author will be asked to submit the TeX file also. The referee
process will usually take between several weeks and two months. Solutions may also
be submitted to the same e-mail address. For this issue, solutions should arrive
before 15th of November 2023.

PROPOSED PROBLEMS
536. Let p be a prime number, Fp the field with p elements, and n ≥ 1 an
integer. If f ∈ Fp [X] is the polynomial X p − X ∈ Fp [X] composed with itself
n times, determine the splitting field of f over Fp .
Proposed by Tudor Păis, anu, École Polytechnique, Paris, France.

537. Let A, B ∈ Mn (C) be such that A2 = A and B 2 = B.

Prove that
Im(AB − BA) = Im(A + B − In ) ∩ Im(A − B),
where Im M = {M X | X ∈ Mn,1 (C)} for every M ∈ Mn (C).
Proposed by Vasile Pop, Technical University of Cluj-Napoca,
Romania.

538. Let n ≥ 1 be an integer and let a1 , . . . , a2n ∈ Z be pairwise distinct.

Prove that ( 2n )2
∑
2n ∑ n(n + 1)(2n + 1)
a2i + ai ≥ .
3
i=1 i=1
When do we have equality?
Proposed by Leonard Giugiuc, Traian National College, Drobeta-
Turnu Severin, Romania.

539. Let n ≥ 1 be an integer and let X = {1, . . . , n}. We denote by FX

the set of all functions f : X → X and by SX the symmetric group on X,
i.e. the set of all permutations on X. If f, g ∈ FX , we say that f and g are
conjugate and we write f ∼ g if there is σ ∈ SX such that g = σf σ −1 .
Let MX be the set of all f ∈ FX such that for every ∅ ̸= Y ⊆ X with
f (Y ) ⊆ Y we have f (Y ) = f (X).
(i) Prove that if f ∈ MX and g ∼ f then g ∈ MX .
1∑
(ii) Prove that |MX / ∼ | = ϕ(d) 2n/d − 1.
n d|n
Proposed by Constantin-Nicolae Beli, IMAR, Bucharest,
Romania.
40 Proposed problems

t
540. For any matrix M , denote M ∗ = M the transpose conjugate of M .
Let A, B ∈ Mn (C) be such that A∗ B = On . Prove that
rank (A∗ A + B ∗ B) ≤ rank (AA∗ + BB ∗ ) .
Proposed by Mihai Opincariu, Brad, Romania, and Vasile Pop,
Technical University of Cluj-Napoca, Romania.

541. Calculate
∞
∑ Hn Hn+1
,
(2n + 1)(2n + 3)
n=1
where Hn = 1 + 1
2 + ··· + 1
n denotes the nth harmonic number.
Proposed by Ovidiu Furdui and Alina S^
ıntămărian, Technical
University of Cluj-Napoca, Romania.

542. Let n ≥ 2 be an integer and let f : Rn → R given by f (x1 , . . . , xn ) = 0

if (x1 , . . . , xn ) = (0, . . . , 0) and
√
5
x41 · · · x4n
f (x1 , . . . , xn ) = √
3
x21 · · · x2n + (x2 − x1 )2 + (x3 − x1 )2 + · · · + (xn − x1 )2
otherwise.
Prove that:
(i) f is continuous at (0, . . . , 0).
(ii) f is Fréchet differentiable at (0, . . . , 0) if and only if n ≥ 8.
Proposed by Dumitru Popa, Department of Mathematics, Ovidius
University of Constant, a, Romania.
 Solutions 41

SOLUTIONS

521. Prove that

∑p ( )( ) ( )
(−4a2 )i+j p + i p + j 1 1+a
lim = ln ∀a ∈ (0, 1).
p→∞ 2i + 2j + 1 p − i p − j 4a 1−a
i,j=0

Proposed by Florin Stănescu, Şerban Cioculescu School, Găeşti,

D^
amboviţa, Romania.

Solution by the author. We denote our sum by Sp (a). Then

∫ ( )( )
1 a ∑
p
2 i+j p + i p+j
Sp (a) = (−4x ) dx
a 0 p−i p−j
i,j=0
∫ ( p ( ))2
1 a ∑ p + i
= (−4x2 )i dx.
a 0 p−i
i=0
We make the change of index k = p − i and we get
∑p ( ) ∑ p ( )
2 i p+i p−k 2p − k
(−4x ) = (−1) (2x)2p−2k = (−1)p U2p ,
p−i k
i=0 i=0
where Un is the Chebyshev polynomial of the second kind, which is defined
by the relation Un (cos θ) sin θ = sin(n+1)θ for all θ ∈ R, and writes explicitly
as
∑
[n/2] ( )
n−k
Un (x) = (−1)k (2x)n−2k .
k
k=0
∫
1 a 2
∫a
Hence Sp (a) = a 0 U2p dx. We put In = 0 Un2 (x)dx and make the
substitution x = cos u. Since sin(n + 1)u = sin uUn (cos u), we get
∫ θ ∫ π/2
2 sin2 (n + 1)u
In = Un (cos u)(− sin u)du = du,
π/2 θ sin u
∫ π/2
where θ = arccos u. We have I0 = θ sin udu = cos u and for n ≥ 1 it holds
1 1
sin2 (n + 1)u − sin2 nu = (1 − cos 2(n + 1)u) − (1 − cos 2nu)
2 2
= sin u sin(2n + 1)u,
so we obtain
∫ ∫
sin2 (n + 1)u − sin2 nu
π/2 π/2
In − In−1 = du = sin(2n + 1)udu
θ sin u θ
cos(2n + 1)θ
= .
2n + 1
42 Problems

It follows that
cos 3θ cos(2n + 1)θ
In = I0 + (I1 − I0 ) + · · · + (In − In−1 ) = cos θ + + ··· +
3 2n + 1
cos 2θ cos(2n + 1)θ cos 2θ cos 4θ cos 2nθ
= cos θ + + ··· + − − − ··· − .
2 2n + 1 2 4 2n
∑ cos kα ( α)
By using the formula = − ln 2 sin , we get
k 2
k≥1
( ) ( )
θ 1 1 2 θ 1
lim In = − ln 2 sin + ln(2 sin θ) = − ln 4 sin + ln(4 sin2 θ)
n→∞ 2 2 2 2 4
( )
1 1 1 4(1 − cos2 θ)
= − ln(2(1 − cos θ)) + ln(4(1 − cos θ)) = ln
2
2 4 4 4(1 − cos θ)2
( ) ( )
1 1 + cos θ 1 1+a
= ln = ln .
4 1 − cos θ 4 1−a
It follows that
( )
1 1 1+a
lim Sp (a) = lim I2p = ln .
p→∞ p→∞ a 4a 1−a

Editor’s note. One may write Un2 in terms of the first Chebyshev polyno-
1 − T2n+2 (x)
mials of the first kind as Un2 (x) = . This follows from the fact
2(1 − x2 )
that
Un2 (cos θ) = sin2 ((n + 1)θ)/ sin2 θ = (1 − cos(2n + 2)θ)/2 sin2 θ
1 − T2n+2 (cos θ)
= .
2(1 − cos2 θ)
We get
∫ a ∫ ∫
1 a 1 1 a T4p+2 (x)
Sp (a) = 2
U2p (x)dx
= dx − dx
0 a 0 2(1 − x2 ) a 0 2(1 − x2 )
( ) ∫
1 1+a 1 a T4p+2 (x)
= ln − dx.
4a 1−a a 0 2(1 − x2 )
∫ a n (x)
To complete the proof, we show that limn→∞ 0 T1−x 2 dx = 0. We have

∫ a ∫ a
Tn (x) cos(n arccos x)
dx = dx.
0 1 − x 2
0 1 − x2
We use integration by parts. Let f (x) = sin(n arccos x). Then f ′ (x) =
− √1−x
n
2
cos(n arccos x), so cos(n1−x
arccos x)
2 = f ′ (x)g(x), with g(x) = − n√1−x
1
2
.
 Solutions 43
√
Since g ′ (x) = n4 x 1 − x2 , we get
∫ a
cos(n arccos x) sin(n arccos x) a
dx = − √
0 1 − x2 n 1 − x2
∫ a 0
4 √
− x 1 − x2 sin(n arccos x)dx.
0 n
To conclude the proof, we show that if an and bn are the two terms in the
right side of the above formula, then limn→∞ an = limn→∞ bn = 0. We have
sin(n arccos x) sin(nπ/2) 1 1
|an | ≤ √ + ≤ √ + ,
n 1−a 2 n n 1−a 2 n
which proves that limn→∞ an = 0, and
∫ a ∫ a
4 √ 4 √
|bn | ≤ x 1 − x sin(n arccos x) dx ≤
2 x 1 − x2 dx
0 n 0 n
∫ a√
4
= x 1 − x2 dx,
n 0
whence the conclusion that limn→∞ an = 0.
We encounter essentially this approach in a solution we received from
G. C. Greubel, from Newport News, VA, USA. Instead of Chebyshev poly-
nomials, he uses the Fibonacci and Lucas polynomials Fn and Ln , which
are related to the Chebyshev polynomials of the first and second kind by
Fn (x) = in−1 Un−1 (−ix/2) and Ln (x) = 2in Tn (−ix/2).
He starts by writing
 2
∫ 2a ∑ p ( ) ∫ 2a
1  p+j 1
Sp = (ix)2j  dx = 2
F2p+1 (ix)dx.
2a 0 p−j 2a 0
j=0

L2n (x) − 2(−1)n

Then he uses the formula Fn2 (x) = 2
to get F2p+1 (ix) =
x2 + 4
L4p+2 − 2
. (Note that L4p+2 is an even function.) This gives
x2 − 4
∫ 2a ∫
1 L4p+2 (it) 1 2a dx
Sp = dx −
2a 0 x2 − 4 a 0 x2 − 4
∫ 2a ( )
1 L4p+2 (it) 1 1+a
= dx + log .
2a 0 x2 − 4 4a 1−a
From here he noticed that L4p+2 (ix) = −T2p+2 (x/2), but he failed to provide
a correct proof of the fact that
∫ 2a ∫ 2a
1 L4p+2 (ix) 1 T4p+2 (x/2)
lim dx = lim − dx = 0.
p→∞ 2a 0 x −4
2 p→∞ 2a 0 x2 − 4
Hence his proof is incomplete.
44 Problems

522. Evaluate the integral

∫ ∞
( )
log 1 + x10
dx.
0 1 + x2
Proposed by Seán M. Stewart, King Abdullah University of Science
and Technology (KAUST), Saudi Arabia.

Solution by√the author. Denote the integral to be found by I. We claim

I = π log(10 + 4 5).
Consider the parametric integral
∫ ∞
log(a + x10 )
J(a) = dx, a > 0.
0 1 + x2
Observe that ∫ ∞
log(x)
J(0) = 10 dx = 0,
0 1 + x2
a result that can be readily seen on enforcing a substitution of x 7→ 1/x,
which proves that J(0) = −J(0). The value for the desired integral we seek
is J(1). Differentiating with respect to the parameter a we have
∫ ∞
′ dx
J (a) = ,
0 (1 + x )(a + x10 )
2

or
∫ ∞ ∫ ∞ 8
′ 1 dx 1 x − x6 + x4 − x2 + 1
J (a) = − dx,
a − 1 0 1 + x2 a − 1 0 a + x10
after a partial fraction decomposition has been applied to the integrand. The
first of the integrals appearing to the right of the equality is elementary. It
has a value equal
∫ ∞ xto π/2. The second integral can be written as a linear com-
n
bination of 0 a+x 10 dx for n = 8, 6, 4, 2, 0. These integrals can be written
in terms of the beta function,
∫ ∞
tx−1
B(x, y) = dt, x, y > 0,
0 (1 + t)x+y
which satisfies B(x, y) = Γ(x)Γ(y)/Γ(x + y), where Γ(·) denotes the gamma
function.
We make the substitution ay = x10 , i.e. x = (ay)1/10 , and we get
∫ ∞ ∫ ∞ n/10 n/10 ∫ ∞ − n−9
xn a y 1/10 −9/10 n−9 y 10
10
dx = a y dy = a 10 dy
0 a+x 0 a + ay 0 1+y
( ) n−9
n−9 n+1 9−n πa 10
= a 10 B , = .
10 10 sin (n+1)π
10
Here we used the fact that Γ(1) = 1 and the Euler’s reflection formula to get
B (z, 1 − z) = Γ(z)Γ(1 − z)/Γ(1) = Γ(z)Γ(1 − z) = sinππz .
 Solutions 45

By plugging n = 8, 4, 2, 0 in the formula above, we get

(
′ π/2 π a−1/10 a−3/10 a−1/2 a−7/10
J (a) = + − + −
a − 1 10(a − 1) sin 9π/10 sin 7π/10 sin π/2 sin 3π/10
)
a−9/10
+ .
sin π/10
But
1 1 √ 1 1 √
= = 5 + 1 =: α, = = 5 − 1 =: β,
sin π/10 sin 9π/10 sin 3π/10 sin 7π/10
so
π
J ′ (a) = (5 − αa−1/10 + βa−3/10 − a−1/2 + βa−7/10 − αa−9/10 ).
10(a − 1)
∫1
We make the substitution a = u10 in the integral J(1) = 0 J ′ (a) da
and we get
∫ 1 9
5u − αu8 + βu6 − u4 + βu2 − α
J(1) = π dx. (4)
0 u10 − 1
Over R the denominator u10 − 1 decomposes into the factors u − 1, u + 1
and u2 − 2 cos θu + 1,√with θ = π/5, 2π/5, 3π/5 and 4π/5. But the cosines of
these numbers are ± 45±1 , i.e. ±α/4 and ±β/4. Hence we have the factors
u2 ± α2 u + 1 = 12 (2u2 ± αu + 2) and u2 ± β2 u + 1 = 12 (2u2 ± βu + 2). It turns out
that in the fraction above the factors u−1, 21 (2u2 −αu+2) and 12 (2u2 −βu+2)
simplify and we get the following partial fraction decomposition
5u9 − αu8 + βu6 − u4 + βu2 − α 1 4u + α 4u + β
= + + .
u10 − 1 u + 1 2u2 + αu + 2 2x2 + βu + 2
Then our integral becomes
∫ 1 ∫ 1 ∫ 1
du 4u + α 4u + β
J(1) = π +π 2 + αu + 2
du + π 2 + βu + 2
du
0 u + 1 0 2u 0 2u
[ ( ) ( )]1
= π log(u + 1) + log 2u2 + αu + 2 + log 2u2 + βu + 2 0
( √ √ )
(5 + 5)(3 + 5) ( √ )
= π log = π log 10 + 4 5 ,
2
as announced. 

Solution by Constantin-Nicolae Beli, IMAR, Bucharest. We denote by

10 )
I our integral. Since the real map x 7→ log(1+x
1+x2
is even, we have
∫ ∞
log(1 + x10 )
dx = 2I.
−∞ 1 + x2
To evaluate this integral we will use complex contour integration.
46 Problems

∏
We have 1 + x10 = λ∈A (x − λ), where A ⊆ C is the set of roots of the
polynomial X 10 + 1, A = {ekπi/10 | k odd}. We write A = A+ ∪ A− , where
A+ and A− are the roots of X 10 + 1 in the upper and lower half-plane in C,
respectively. We have
A± = {eiθ | θ = ±π/10, ±3π/10, ±5π/10, ±7π/10, ±9π/10}.
(Here the ± sign from A± is the same as the ± signs for the values of θ.)
Note that λ 7→ λ̄ is a bijection from A− to A+ so for every x ∈ R we
have
∏ ∏ ∏
1 + x10 = |1 + x10 | = (x − λ) = (x − λ)(x − λ̄) = |x − λ|2 .
λ∈A λ∈A− λ∈A−
∑
(We have |x − λ| = |x − λ̄|.) Hence log(1 + x10 ) = 2 λ∈A− log |x − λ|.
We now denote by log : C \ (−∞, 0] → C the principal branch of the
logarithm. If z ∈ C \ (−∞, 0] then z writes uniquely as z = reiθ , with
r > 0 and θ ∈ (−π/2, π/2) and log(z) is defined as log(z) = log(r) + θi.
Moreover θ will be called the argument of z and denoted arg(z) = θ. Note
that ℜ log(z) = log(r) = log |z|.
We now define the function
2 ∑
f (z) = 2
log(z − λ).
1+z
λ∈A−

Note that each function log(z−λ) is defined everywhere except on (−∞, 0]+λ,
which is the horizontal half-line to the left of λ. In particular, log(z − λ) is
defined in the half-plane {z ∈ C | Im z > Im λ}. Since the imaginary parts
of the elements λ ∈ A− are − sin π/10, − sin 3π/10, . . . , − sin 9π/10 and the
largest of these imaginary parts is − sin π/10, the function f is defined on the
half-plane {z ∈ C | Im z > − sin π/10}, which contains the upper half-plane
{z ∈ C | Im z ≥ 0}.
Now for z ∈ R we have
2 ∑
ℜf (z) = ℜ log(z − λ)
1 + z2
λ∈A−
2 ∑
= log |z − λ|
1 + z2
λ∈A−
1
= log(1 + z 10 ).
1 + z2
It follows that ∫ ∞ ∫ ∞
2I = ℜf (z)dz = ℜ f (z)dz.
−∞ −∞
Let R ≫ 0. We consider the closed contour made of the interval [−R, R]
and the semicircle CR of diameter [−R, R] in the upper half-plane. The only
 Solutions 47

singularity of f in this domain is at i and this singularity is simple. We have

( )
2 ∑ 1 ∑
Resz=i (f ) = lim(z−i)f (z) = log(z−λ) = log(i−λ).
z→i z+i z=i i
λ∈A− λ∈A−

It follows that
∫ R ∫ ∑
f (z)dz + f (z)dz = 2πi Resz=i (f ) = 2π log(i − λ).
−R CR λ∈A−

When we take the real parts in both sides, since ℜ log(z) = log |z|, we get
∫ R ∫ ∑
ℜ f (z)dz + ℜ f (z)dz = 2π log |i − λ|.
−R CR λ∈A−

We now take limits when R → ∞. Let R ≫ 0. If z ∈ CR , then for every

λ ∈ A− we have |z| = R and |λ| = 1, so 1 ≤ R − 1 ≤ |z − λ| ≤ R + 1. Hence
0 ≤ log |z − λ| ≤ log(R + 1) and so |ℜ log(z − λ)| = | log |z − λ|| ≤ log(R + 1).
We also have | Im log(z − λ)| ≤ π, so | log(z − λ)| ≤ log(R + 1) + π. Since
|A− | = 5, we get
∑ ∑
log(z − λ) ≤ | log(z − λ)| ≤ 5(log(R + 1) + π).
λ∈A− λ∈A−

We also have |1 + z 2 | ≥ |z 2 | − 1 = R2 − 1, so 2
1+z 2
≤ R22−1 . It follows that
|f (z)| ≤ R22−1 · 5(log(R + 1) + π) = 10(log(R+1)+π)
R2 −1
∀z ∈ CR . Since the length
∫ 10πR(log(R+1)+π)
of CR is πR, this implies that | CR f (z)dz| ≤ . It follows
∫ R2 −1 ∫
that limR→∞ f (z)dz = 0, which implies that limR→∞ ℜ CR f (z)dz = 0.
CR
∫R ∫∞
We also have limR→∞ ℜ −R f (z)dz = ℜ −∞ f (z)dz = 2I. Thus
∑ ∑
2I + 0 = 2π log |i − λ|, whence I = π log |i − λ|.
λ∈A− λ∈A−

To evaluate |i − λ|, we note that for every α, β ∈ R we have

|e αi
− eβi | = |e(α+β)i/2 | · |e(α−β)i/2 − e−(α−β)i/2 | = 1 · |2i sin(α − β)/2|
= 2| sin(α − β)/2|.
In our case |i − λ| = |eπi/2 − eθ | = |2 sin(π/4 − θ/2)|, for λ = −π/10, . . . ,
−9π/10. We get
∑ ∏
log |i − λ| = log |i − λ|
λ∈A− λ∈A−

= log |25 sin(3π/10) sin(2π/5) sin(π/2) sin(3π/5) sin(7π/10)|.

But sin(π/2) = 1, sin(3π/10) sin(7π/10) = sin2 (3π/10) = 12 (1 − cos(3π/5)) =
√
1
8 (3 + 5) and sin(2π/5) sin(3π/5) = sin2 (2π/5) = 12 (1 − cos(4π/5)) = 18 (5 +
48 Problems

√ √
5). (Here we used the formulas cos(3π/5) = − cos(2π/5) = − −1+4 5
and
√
−1− 5
cos(4π/5) = 4 .) We get
∑ ( 1 √ 1 √ ) (1 √ )
log |i − λ| = log 25 · (3 + 5) · (5 + 5) = log (20 + 8 5)
8 8 2
λ∈A−
√
= log(10 + 4 5),
√
which implies I = π log(10 + 4 5). 

523. Prove that

rank(A − ABA) − rank(B − BAB) = rank(A) − rank(B)
for all A, B ∈ Mn (C).
Proposed by Vasile Pop, Technical University of Cluj-Napoca,
Romania and Mihai Opincariu, Avram Iancu National College, Brad,
Romania.

Solutions by the authors. First solution. This solution is based on the

observation that
Ker(A) ⊆ Im(In − BA)
for all A, B ∈ Mn (C). (If x ∈ Ker(A), then x = x − BAx = (In − BA)x, so
x ∈ Im(In − BA).)
It follows that in the Sylvester inequality for A and In − BA we have
equality, i.e.
n − rank(A) = rank(In − BA) − rank(A(In − BA))
= rank(In − BA) − rank(A − ABA) .
By symmetry,
n − rank(B) = rank(In − AB) − rank(B − BAB) .
Since rank(In − BA) = rank(In − AB) for all A, B ∈ Mn (C), the required
identity follows by subtracting termwise the previous two equalities. 

Second solution. Consider the block matrices

[ ] [ ] [ ]
A − ABA On A On A AB
M= , N= , and K = .
On B On B − BAB BA B
Since
rank(M ) = rank(A − ABA) + rank(B)
and
rank(N ) = rank(B − BAB) + rank(A) ,
 Solutions 49

the required identity is equivalent to rank(M ) = rank(N ), which follows from

the representation L1 M R1 = L2 N R2 = K, where
[ ] [ ] [ ] [ ]
In A In On In On In B
L1 = , R1 = , L2 = , R2 =
On In A In B In On In
are all non-singular matrices. 

524. Determine all bijective and differentiable functions f : R → R, with

f ′ (x) ̸= 0 ∀x ∈ R, satisfying
( )
1
f (x) + f − x = 1, for every x ∈ R.
f ′ (x)
Proposed by Mircea Rus, Technical University of Cluj-Napoca,
Romania.
1
Solution by the author. Let g : R → R, g(x) = f (x) − . Then g is
2
bijective and differentiable, with g ′ = f ′ and
( )
1
g − x = −g(x), for every x ∈ R. (1)
g ′ (x)
Let h : R → R be the inverse function of g. For every y ∈ R if we take
x = h(y) in (1) we get
( )
g h′ (y) − h(y) = −y, for every y ∈ R, (2)
1 1
as h is differentiable, with h′ (y) = = ′ . We apply h to (2) and
g ′ (h(y)) g (x)
we get h′ (y) − h(y) = h(−y), so
h′ (y) = h(y) + h(−y), for every y ∈ R, (3)
from which we also get
h′ (−y) = h(−y) + h(y) = h′ (y), for every y ∈ R.
Hence,
(h(y) + h(−y))′ = h′ (y) − h′ (−y) = 0, for every y ∈ R,
and so h(y) + h(−y) is a constant.
Then, by (3), we get that h′ is a constant, i.e. h(y) = ay + b, with
a, b ∈ R and a ̸= 0 (h is assumed bijective). After plugging h(y) = ay + b
a 1 1
into (3) we get a = 2b, so that h(y) = ay + ∀y ∈ R and so g(x) = x − ,
2 a 2
x
∀x ∈ R. Consequently, f (x) = , i.e. the functions f satisfying the required
a
properties are of the form
fc : R → R, f (x) = cx, for every x ∈ R,
50 Problems

where c ∈ R∗ . One verifies that each such function fc satisfies the required
properties. 
Editor’s note. We received a similar solution from Leonard Giugiuc,
except that he doesn’t consider the map g(x) = f (x) − 1/2 and its inverse
h = g −1 , but he takes h = f −1 . Then the relation from the hypothesis
applied to x = h(y) yields f (h′ (y) − h(y)) = 1 − y which, after applying
h, gives h′ (y) = h(y) + h(1 − y). After substituting y 7→ 1 − y one gets
(h(y) + h(1 − y))′ = h′ (y) − h′ (1 − y) = 0, so h′ (y) = h(y) + h(1 − y) is a
constant. It follows that h is affine and so is its inverse f . If f (x) = ax + b,
then, after plugging this into the original relation, one gets f (x) = ax, with
a ∈ R∗ .

525. Let n ≥ 4 be an integer and a1 , . . . , an be positive real numbers.

1 1
(i) Prove that if + ··· + ≤ 1, then
a1 an
∑
2 ai aj − n3 (n − 1) ≥ 2(n − 1)2 (a1 + · · · + an − n2 ).
1≤i<j≤n

1 1 n2 − n + 1
(ii) Prove that if + ··· + ≥ 1 and k = , then
a1 an (n − 1)3
 
∑
a21 + · · · + a2n − n3 ≥ k 2 ai aj − n3 (n − 1) .
1≤i<j≤n

Proposed by Vasile C^ırtoaje, Petroleum-Gas University of

Ploieşti, Romania and Leonard Giugiuc, Traian National College,
Drobeta-Turnu Severin, Romania.
∑n
∑n by the authors.∑(i) Let a1 , . . . , an > 0, with i=1 1/ai ≤ 1.
Solution
Put p = i=1 ai ∑
and q = i<j ai aj . These relations are equivalent to
∑n n
i=1 ai = p − 2q. By The Equal Variables Theorem
2 2
i=1 ai = p and
applied
∑n to the map∑ θ 7→ 1/θ, i.e. Corollary ∑n 1.4 in [1], of ∑all c1 , . . . , cn with
n
i=1 ci = p − 2q,
c = p and c c = q, i.e. with c = p and 2 2
i=1 i ∑ni<j i j i=1 i
the minimum of i=1 1/ci is reached for (c1 , . . . ∑ , cn ) = (x, . . . , x, y) for some
x, y with 0 < x ≤ y. In this case we have p = ni=1 ci = (n − 1)x + y and
∑ (n − 1)(n − 2) 2
q = i<j ci cj = x + (n − 1)xy. Hence the relation we want
2
to prove writes as
( )
(n − 1)(n − 2)x2 + 2(n − 1)xy − n3 (n − 1) ≥ 2(n − 1)2 (n − 1)x + y − n2 ,
that is,
( )
(n − 2)x2 + 2xy − n3 ≥ 2(n − 1) (n − 1)x + y − n2 ,
 Solutions 51

or equivalently
2(x − n + 1)y + (n − 2)x2 − 2(n − 1)2 x + n2 (n − 2) ≥ 0.
∑n
∑n The minimality property implies that (n − 1)/x + 1/y = i=1 1/ci ≤
i=1 1/an = 1. Then (n − 1)/x < (n − 1)/x + 1/y ≤ 1, so x > n − 1, i.e.
x
x − n + 1 > 0. Also (n − 1)/x + 1/y ≤ 1 implies that y ≥ . Since
x−n+1
also y ≥ x, we have
{ } { x
x if n − 1 < x ≤ n,
y ≥ max , x = x−n+1
x−n+1 x if x > n.
When n − 1 < x ≤ n, the inequality we want to prove follows from
y≥ x
x−n+1 and
x
2(x − n + 1) + (n − 2)x2 − 2(n − 1)2 x + n2 (n − 2) ≥ 0,
x−n+1
which is equivalent to (n − 2)(x − n)2 ≥ 0. The equality holds when x = n
x
and y = x−n+1 = n.
When x > n, the inequality we want to prove follows from y ≥ x and
2(x − n + 1)x + (n − 2)x2 − 2(n − 1)2 x + n2 (n − 2) ≥ 0,
which is equivalent to n(x − n + 2)(x − n) ≥ 0. Since x > n, we have strict
inequality, so the only case of equality is the one from the first case, i.e. for
(a1 , . . . , an ) = (n, . . . , n).
(ii) We keep the notations p and q from (i). We first prove the following
statement
Lemma 1. One of the following holds.
(a) There are unique 0 < y ≤ x such that (n − 1)x + y = p and
(n − 1)(n − 2)x2 + 2(n − 1)xy = 2q. ∑n
∑ (b) There are b1 ≥ · · · ≥ bn−1 ≥ 0 such that i=1 bi = p and
i<j bi bj = q.

∑n Proof. Our ∑ lemma states that there are 0 ≤ c1 ≤ · · · ≤ cn with

i=1 i c = p and i<j ci cj = q such that (c1 , . . . , cn ) is either of the form
(x, . . . , x, y), for some unique 0 < y ≤ x, or the form (b1 , . . . , bn−1 , 0), for
some b1 ≥ · · · ≥ bn−1 ≥ 0. ∑ ∑
Again the relations ni=1 ci = p and i<j ci cj = q are equivalent to
∑n ∑n 2
i=1 ci = p − 2q.
2
i=1 ci = p and
For (a) let s = p/n, so that p = ns. Then the condition that (n − 1)x +
y = p = ns with x ≥ y means that x = s + t and y = s − (n − 1)t for some
t ≥ 0. The extra condition y > 0 means s > (n − 1)t. And the condition
(n − 1)x2 + y 2 = p2 − 2q write as (n − 1)(s + t)2 + (s − (n − 1)t)2 = p2 − 2q, i.e.
ns2 +n(n−1)t2 = p2 −2q. But ns = p, so after multiplying by n, the equation
becomes p2 + n2 (n − 1)t2 = np2 − 2nq, i.e. n2 (n − 1)t2 = (n − 1)p2 − 2nq.
52 Problems

∑ ∑
Note that the inequality ( ni=1 ai )2 ≤ n ni=1 a2i writes as p2 ≤ n(p2 − 2q),
i.e. (n − 1)p2 ≥ 2nq. Thus there is a unique t ≥ 0 with this property, namely
√
(n − 1)p2 − 2nq
t= .
n2 (n − 1)
We still have to check the condition p/n = s > (n − 1)t, i.e. p > n(n − 1)t,
which is equivalent to p2 ≥ n2 (n − 1)2 t2 > (n − 1)2 p2 − 2n(n − 1)q, i.e.
2(n − 1)q > (n − 2)p2 .
For (b) we look for (b1 , . . . , bn−1 ) of the type (y, x, . . . , x), with 0 < x ≤
y. Our conditions write as (n − 2)x + y = p and (n − 2)x2 + y 2 = p2 − 2q. We
denote u = p/(n − 1), that is, p = (n − 1)u. Then (n − 2)x + y = p with x ≤ y
is equivalent to x = u − v and y = u + (n − 2)v for some v ≥ 0. The extra
condition x > 0 writes as u > v. By the reasoning from (a) applied with n, s, t
replaced by n − 1, u, −v, respectively, the second equation, (n − 2)x2 + y 2 =
p2 − 2q, is equivalent to (n − 1)2 (n − 2)v 2 = (n − 2)p2 − 2(n − 1)q. It has a
solution iff (n − 2)p2 − 2(n − 1)q ≥ 0, i.e. 2(n − 1)q ≤ (n − 2)p2 . Assuming
this happens, we have (n − 1)2 (n − 2)v 2 = (n − 2)p2 − 2(n − 1)q < (n − 2)p2 ,
which is equivalent to v < p/(n − 1) = u, so all conditions are fulfilled.
Hence if 2(n−1)q > (n−2)p2 then (a) holds and if 2(n−1)q ≤ (n−2)p2
then (b) holds. This concludes the proof of lemma. 
( )
The relation we want to prove writes as (p2 −2q)−n3 ≥ k 2q−n3 (n−1)
i.e. as
( ) ( )
p2 − 2q − n3 − k 2q − n3 (n − 1) = p2 − n4 − (k + 1) 2q − n3 (n − 1) ≥ 0.
Recall that in the case (a) of Lemma 1 we have p2 − 2q = (n − 1)x2 + y 2 and
∑
in the case (b) we have p2 − 2q = n−1 2
i=1 bi . We now consider the two cases
of Lemma 1. ( )
(a) We have p2 − n4 − (k + 1) 2q − n3 (n − 1) = f (y), where f is a
quadratic polynomial in the variable Y with coefficients depending on x, viz.
f (Y ) = (n − 1)x2 + Y 2 − n3 − k((n − 1)(n − 2)x2 + 2(n − 1)xY
− n3 (n − 1))
( ) ( )
= Y 2 − 2k(n − 1)xY + (n − 1) 1 − k(n − 2) x2 + k(n − 1) − 1 n3 .
Note that f is monic and the coefficient of Y is −2k(n − 1)x. It follows
2 −n+1
that f is decreasing on (−∞, k(n − 1)x]. But k(n − 1) = n(n−1) 2 > 1, so that

k(n − 1)x > x. Thus f is decreasing on (−∞, x]. Since y ≤ x, we have

f (y) ≥ f (x) = (n − kn(n − 1))(x2 − n2 ).
But k(n − 1) > 1, so n − kn(n − 1) < 0. Hence, if x < n, then f (y) < 0 and
we are done.
 Solutions 53

Assume now that x ≥ n. By The Equal Variables Theorem∑applied to

∑ θ 7→ 1/θ, i.e. Corollary 1.4 in∑[1],
the map of all c1 , . . . , cn with ni=1 ci = p
and i<j ci cj = q, the maximum of ni=1 1/ci is reached for (c1 , . . . , ci ) =
(x,
∑n. . . , x, y). This maximality property implies that (n − 1)/x + 1/y ≥
i=1 1/ai ≥ 1, so y ≤ x−n+1 < x. (We have x ≥ n.) Since f is decreasing
x
( )
on (−∞, x], we have that f (y) ≥ f x−n+1 , so it is enough to prove that
x

f (z) ≥ 0, where z = x−n+1

x
.
Note that f can be rewritten as
((n − 1)x + Y )2 − n4 − (k + 1)((n − 1)(n − 2)x2 + 2(n − 1)xY − n3 (n − 1)).
Thus, in order to get f (z), we first compute
((n − 1)x + z)2 − n4 = ((n − 1)x + z − n2 )((n − 1)x + z + n2 )
(n − 1)(x − n)2 (n − 1)x2 − 2nx − n2 (n − 1)
= · .
x−n+1 x−n+1
With
x2 (x − n)(x − n(n − 1))
xz − n2 = − n2 = ,
x−n+1 x−n+1
we get
(n − 2)x2 + 2xz − n3 = (n − 2)(x2 − n2 ) + 2(xz − n2 )
( )
x − n(n − 1)
= (x − n) (n − 2)(x + n) + 2 ·
x−n+1
(n − 2)x + nx − n (n − 1)
2 2
= (x − n) ·
x−n+1
(x − n)2 ((n − 2)x + n(n − 1))
= .
x−n+1
Hence, since
( 2 )
n −n+1 n(n2 − 2n + 2)
(k + 1)(n − 1) = (n − 1) + 1 = ,
(n − 1)3 (n − 1)2
we have
( )2 ( )
f (z) = (n − 1)x + z − n2 − (k + 1)(n − 1) (n − 2)x2 + 2xz − n3
( )
(n − 1)(x − n)2 (n − 1)x2 − 2nx − n2 (n − 1)
=
(x − n + 1)2
( )
n(n2 − 2n + 2) (x − n)2 (n − 2)x + n(n − 1)
− ·
(n − 1)2 x−n+1
(x − n) g(x)
2
= ,
(n − 1)2 (x − n + 1)2
54 Problems

where
( )2
g(x) = x2 − 2n(n − 1)x + n2 (n − 1)2 = x − n(n − 1) .
2 2
Thus f (y) ≥ f (z) = (x−n) (x−n(n−1))
(n−1)2 (x−n+1)2
≥ 0. The equality holds iff x = n
or n(n − 1) and y = z = x−n+1 , i.e. y = n or n−1
x n
, respectively.
∑ ∑n−1 2
(b) We have (n − 1) n−1 i=1 bi ≥ ( i=1 bi ) , i.e. (n − 1)(p − 2q) ≥ p ,
2 2 2

which implies that (n − 2)p ≥ 2(n − 1)q, so (n − 2)(p − 2q) ≥ 2q, i.e.
2 2

p2 − 2q ≥ n−2
2
q. But (n − 2)(n2 − n + 1) = n3 − 3n2 + 3n − 2 < (n − 1)3 , so
2−n+1 n −n+1
2
3 = k. Hence p − 2q > 2kq. We also have (n − 1)k = (n−1)2 >
1
n−2 > n(n−1) 2

1, so −n3 > −kn3 (n − 1). It follows that p2 − 2q − n3 > k(2q − n3 (n − 1))

and we are done.
The equality holds for (a1 , . . . , an ) = (n, . . . , n) or (n(n − 1), . . . , n(n −
n
1), ) and permutations. 
n−1

References
[1] Vasile Cı̂rtoaje, The equal variable method, J. Inequal. Pure Appl. Math. 8 (2007), no.
1, art. 15, 21 pp. https://www.emis.de/journals/JIPAM/article828.html?sid=828

526. If a, b ∈ C and k ≥ 2, then we define

Fa,b
k
= {f : C → C | f (k) (z) = az + b},
where f (k) = f ◦ · · · ◦ f , with k copies of f .
For given a, b ∈ C and k ≥ 2 find necessary and sufficient condition
such that a set A ⊆ C can be written as A = Fix(f ) for some f ∈ Fa,b
k .

Proposed by Vasile Pop, Technical University of Cluj-Napoca,

Romania and Constantin-Nicolae Beli, IMAR, Bucharest, Romania.

Solution by the authors. We prove the following.

1. If a ̸= 1, then the only possible A is A = { 1−a b
}.
2. If a = 1, b ̸= 0, then the only possible A is A = ∅.
3. If a = 1, b = 0, then A can be any subset of C except a set of the
form A = C \ X, where X ⊂ C is a finite set with |X| ∈ / p1 N + · · · + ps N,
where p1 , . . . , ps are all prime divisors of k and N := Z≥0 .
In particular, if k is a power of a prime p, then the only sets A that do
not qualify are A = C \ X, where X is finite and p - |X|.
And if k is not a power of a prime, i.e. if s ≥ 2, then every n ≥
p1 p2 − p1 − p2 + 1 writes as n = p1 n1 + p2 n2 for some integers n1 , n2 ≥ 0, so
n ∈ p1 N + p2 N ⊆ p1 N + · · · + ps N. Hence S = N \ (p1 N + · · · + ps N) is a finite
set contained in {1, 2, . . . , p1 p2 − p1 − p2 }. Then the only sets A that don’t
qualify are A = C \ X, where X is finite and |X| ∈ S.
 Solutions 55

To prove these statements, we use a couple of observations. Let a, b ∈ C

be fixed. We denote by g : C → C, the function given by g(z) = az + b. So
f ∈ Fa,b
k means f (k) = g.

First note that Fix(f ) ⊆ Fix(g). Indeed, if f (z) = z, then, by induction

on i, we have f (i) (z) = z ∀i ≥ 1. In particular, g(z) = f (k) (z) = z.
Next, we note that if z ∈ Fix(g), then g(z) = z, so f (g(z)) = f (z).
From f ◦ g = f ◦ f (k) = f (k+1) = f (k) ◦ f = g ◦ f it follows that g(f (z)) =
f (g(z)) = f (z) and so f (z) ∈ Fix(g). Hence f (Fix(g)) ⊆ Fix(g).

We now prove our claims.

1. If a ̸= 1, then Fix(g) = {z0 }, where z0 is the only solution of
az + b = z, i.e. z0 = 1−a b
. So Fix(f ) ⊆ Fix(g) = {z0 }. But we also have
f (z0 ) ∈ f (Fix(g)) ⊆ Fix(g) = {z0 } and so f (z0 ) = z0 , i.e. z0 ∈ Fix(f ). It
follows that Fix(f ) = {z0 }, as claimed.
To complete the proof, we must show that Fa,b k ̸= 0, i.e. that there

are functions f with f (k) = g. We look for affine functions f (z) = αz + β.

By induction on i, f (i) has the form f (i) (z) = αi z + βi for some βi ∈ C. In
particular, f (k) (z) = αk z + βk . In order that f (k) (z) = az + b we need that
αk = a and, in order that f (z0 ) = z0 , we must have f (z) = α(z − z0 ) + z0 .
Conversely, if f has this form, then f (k) (z) = az + βk for some βk ∈ C
and z0 is a fixed point of f , so it will be a fixed point of f (k) , as well.
Since f (k) (z) = az + βk , g(z) = az + b, and f (k) (z0 ) = g(z0 ) = z0 , we
must have f (k) = g. Hence Fa,b has precisely k affine functions, namely
f (z) = α(z − z0 ) + z0 for each α ∈ C with α ∈ C with αk = a.
2. If a = 1, b ̸= 0, i.e. g(z) = z + b, then Fix(g) = ∅, so, since
Fix(f ) ⊆ Fix(g), we must have Fix(f ) = ∅, as claimed.
To complete the proof, we must show that F1,b k ̸= ∅. An obvious element

of F1,b
k is f (z) = z + b . In fact, this is the only affine function from F k .
k 1,b
Indeed, if f (z) = αz + β then, same as in the case 1, we have f (k) (z) =
αk z + βk . Therefore, in order that f (k) (z) = z + b, we must have αk = 1.
Assuming that αk = 1, we have f (k) (z) = αk z + βk = z + βk . If α ̸= 1, then
β
z0 = 1−α is fixed point of f , so it will be a fixed point of f (k) , as well. It
follows that z0 = f (k) (z0 ) = z0 + βk and so f (k) (z) = z ̸= g. If α = 1, then
f (z) = z + β and f (k) (z) = z + kβ. Thus f (k) (z) = g(z) = z + b if and only
if β = kb .
3. If a = 1 and b = 0, then g(z) = z. Let A ⊆ C. We denote X = C \ A.
We have two cases.
(a) X is infinite. Then |X| = k|X| = |Zk × X|. (See, e.g., [1].) Hence
there is a bijection ϕ : Zk ×X → X. Then every element of X writes uniquely
as ϕ(c̄, x) for some c̄ ∈ Zk and x ∈ X. (Here c̄ denotes the class in Zk of
56 Problems

c ∈ Z.) Then we define the function f : C → C by

{
z, z ∈ C \ X,
f (z) =
ϕ(c̄ + 1̄, x), z = ϕ(c̄, x) with c̄ ∈ Zk , x ∈ X.

Obviously, Fix(f ) = C \ X = A. If z ∈ A, then f (z) = z, so f (k) (z) = z. If

z ∈ X, then z = ϕ(c̄, x) for some c̄ ∈ Zk and x ∈ X. By induction on l, we
have f (l) (z) = f (l) (ϕ(c̄, x)) = ϕ(c̄+ ¯l, x). In particular, f (k) (z) = ϕ(c̄+ k̄, x) =
ϕ(c̄, x) = z. Hence f (k) is the identity map, i.e. f ∈ F1,0 k .

(b) X is finite, |X| = n. If f ∈ F1,0 k , then f (k) is the identity map g,

so f is invertible, with f −1 = f (k−1) . Then f generates a group of invertible

complex functions (G, ◦) . Since f (k) is the identity map, we necessarily have
|G| = ord f | k. We consider the action of G on C given by f (i) · z = f (i) (z).
For every z ∈ C we denote by Oz its orbit with respect to the action of G.
We have |Oz | = |{f (i) (z) | i ∈ Z}| = 1 iff z ∈ Fix(f ) = A. Hence X = C \ A
is the union of all orbits Oz with |Oz | ≥ 2. Since |X| < ∞, there ∑ are a finite
number of such orbits, say, Oz1 , . . . , Ozm . Then n = |X| = m j=1 dj , with
dj = |Ozj |. But we have dj | |G| | k . Hence n writes as a sum of divisors of
k which are > 1.
Conversely, assume that |X| = n = d1 + · · · + dm , where dj | k, dj > 1.
Then X writes as a disjoint union X = X1 ∪ · · · ∪ Xm with |Xk | = dk . We
fix an index j, 1 ≤ j ≤ m, and we index the elements of Xj by the set Zdj ,
i.e. we put Xj = {zc̄ | c̄ ∈ Zdj }. We define fj : Xj → Xj by fj (zc̄ ) = zc̄+1̄ .
Obviously, fj (z) ̸= z ∀z ∈ Xj . By induction on l we get f l (zc̄ ) = zc̄+l̄ . In
particular, f (dj ) (zc̄ ) = zc̄+d¯j = zc̄ . Hence f (dj ) is the identity map on Xj .
(k) (k)
Since dj | k, this implies that fj is the identity map, i.e. fj (z) = z
∀z ∈ Xj .
We now define f : C → C by
{
z, z ∈ C \ X,
f (z) =
fj (z), z ∈ Xj , 1 ≤ j ≤ m.

Since f (z) ̸= z for every 1 ≤ j ≤ m and every z ∈ Xj , we have Fix(f ) =

C \ X = A. If z ∈ C \ X, then f (z) = z and so f (k) (z) = z. If z ∈ Xj for
(k)
some j, then f (k) (z) = fj (z) = z. Hence f (k) is the identity map and so
f ∈ F1,0k .

Let now M be the monoid M = p1 N + · · · + ps N, where p1 , . . . , ps are

all primes divisors of k. If n = d1 + · · · + dm , with dj | k, dj > 1, then for
every j we have pt | dj for some 1 ≤ t ≤ s. It follows that dj ∈ pt N ⊆ M and
so n = d1 + · · · + dm ∈ M . Conversely, if n ∈ M , then n is a sum of copies
of pt , with 1 ≤ t ≤ s, and for each t we have pt | k and pt > 1.
 Solutions 57

Hence an equivalent necessary and sufficient condition is that n ∈ p1 N+

· · · + ps N. 

References
[1] S.M. Srivastava, A Course on Borel Sets, Graduate Texts in Mathematics, 180,
Springer-Verlag, New York, 1998.

527. For every non-negative integer n, evaluate the integral

∫ ∞
logn (x) sin(x)
In = dx.
0 x
The answer may be given either in a closed form or recursively. Also it
may include values of the zeta function at integers ≥ 2.
Proposed by Seán M. Stewart, King Abdullah University of Science
and Technology (KAUST), Saudi Arabia.

Solution by the author. Denote the integral to be found by In . We claim

for n ∈ N
∑n { }
n! |Bk+1 |π k+1
In+1 = −γIn − (−1)k ζ(k + 1) + In−k ,
(n − k)! (k + 1)!
k=1

subject to the initial conditions of I0 = π2 and I1 = − γπ

2 . Here ζ(s) de-
notes the Riemann zeta function, Bn denotes the Bernoulli numbers, while
γ denotes the Euler–Mascheroni constant.
Denote the integral to be evaluated by In . Thus
∫ ∞
logn (x) sin(x)
In = dx.
0 x
The value for the integral will be found using a technique that exploits an ex-
ponential
∑ generating function. Consider the exponential generating function
G(t) = ∞ In tn
n=0 n! . We have
∫ ∞ ∞ ∫ ∞ ( )
sin x ∑ (t log x)n t−1 πt
G(t) = dx = x sin x dx = Γ(t) sin ,
0 x n! 0 2
n=0
(5)
where the value for the integral corresponds to the well-known value for the
Mellin transform of the sine function. Here Γ(t) denotes the gamma function.
Taking the logarithmic derivative of (5) with respect to t yields
( ) ( )
G′ (t) π πt 1 π πt
= ψ(t) + cot = ψ(t + 1) − + cot . (6)
G(t) 2 2 t 2 2
58 Problems

where ψ(t) denotes the digamma function and we have made use of the
functional relation for the digamma function of
1
ψ(t + 1) = ψ(t) + .
t
From the series expansion for the cotangent function
∞
1 ∑ 2n+1 |Bn+1 |tn
cot(t) = − , |t| < π,
t (n + 1)!
n=1

and the Maclaurin series expansion for the digamma function

∞
∑
ψ(t + 1) = −γ − (−1)n ζ(n + 1)tn , |t| < 1,
n=1

we may rewrite (6) as

∑ ∞ ∑ |Bn+1 |π n+1 tn ∞
G′ (t)
= −γ − (−1)n ζ(n + 1)tn − ,
G(t) (n + 1)!
n=1 n=1
or
∞
∑ ∞
∑ ∞
In tn In tn ∑
G′ (t) = −γ − · (−1)n ζ(n + 1)tn
n! n!
n=0 n=0 n=1
∞
∑ ∞
∑
In tn |Bn+1 |π n+1 tn
− · .
n! (n + 1)!
n=0 n=1

Noting that
∞
∑ ∞
∑
In tn−1 In+1 tn
G′ (t) = = ,
(n − 1)! n!
n=1 n=0
adjusting all sums so they start at n = 1 we have
∞
∑ ∞
∑ ∞ ∞
In+1 tn In tn 1 ∑ In−1 tn ∑
I1 + = −γI0 − γ − · (−1)n ζ(n + 1)tn
n! n! t (n − 1)!
n=1 n=1 n=1 n=1
∞ ∞
1 ∑ In−1 tn ∑ |Bn+1 |π n+1 tn
− · . (7)
t (n − 1)! (n + 1)!
n=1 n=1

For the initial conditions, we have

∫ ∞ ∫ ∞
sin x π log(x) sin(x) γπ
I0 = dx = and I1 = dx = − .
0 x 2 0 x 2
One way to find the value for I1 is given in the Appendix. With these values,
on employing the Cauchy product of power series allows one to rewrite (7)
 Solutions 59

as
∞
∑ ∑∞ ∞ n { } n
tn tn ∑ ∑ n! t
In+1 = − γIn − k
(−1) ζ(k + 1)In−k
n! n! (n − k)! n!
n=1 n=1 n=1 k=1
∑∞ ∑ n { }
n! |Bk+1 |π k+1 In−k tn
− .
(n − k)! (k + 1)! n!
n=1 k=1

Comparing coefficients of tn /n! on either side yields the recurrence relation

of
∑n { }
n! |Bk+1 |π k+1
In+1 = −γIn − k
(−1) ζ(k + 1) + In−k ,
(n − k)! (k + 1)!
k=1

subject to the initial conditions of I0 = π

2 and I1 = − γπ
2 , as announced.

Appendix. The value for the initial condition I1 will be found using the
following property for the Laplace transform
∫ ∞ ∫ ∞
f (x)g(x) dx = L{f (x)}(s) · L−1 {g(x)}(s)ds, (8)
0 0
provided both integrals exist.
We start by considering the integral
∫ ∞
log(x)
J= dx.
0 1 + x2
To evaluate this integral, enforcing a substitution of x 7→ 1
x yields
∫ ∞
log(x)
J =− dx = −J,
0 1 + x2
from which J = 0. Now the Laplace transform for the logarithmic function
is known (see, for example, [1, Entry 6.1, p. 48]). Here
log(s) + γ
L{log(x)}(s) = − ,
s
so on applying (8) to the integral J we find
∫ ∞ ∫ ∞ { }
log(x) −1 1
dx = L{log(x)}(s) · L (s)ds
0 1 + x2 0 1 + x2
∫ ∞
log(s) + γ
=− · sin(s)ds,
0 s
whence ∫ ∫
∞ ∞
log(s) sin(s) sin(s)
0=− ds − γ ds.
0 s 0 s
60 Problems

The integral appearing on the right is just the Dirichlet integral. Its value is
π
2 . Substituting for this value, after replacing the dummy variable s with x
and rearranging, we find
∫ ∞
log(x) sin x γπ
dx = − ,
0 x 2
as announced. 

References
[1] F. Oberhettinger and L. Badii, Tables of Laplace transforms, Springer-Verlag, New
York–Heidelberg, 1973.

Editor’s note. We don’t need two initial terms of the sequence (In )n≥0
in order to use the recurrence relation. It suffices to determine I0 . The
recurrence relation at n = 0 gives I1 = −γI0 = − γπ
2 .

Solution by Brian Bradie, Christopher Newport University, Newport

News, VA, USA. Note,
∫ ∞ ∫ ∞ (∫ ∞ )
logn x sin x −sx
In = dx = fn (s)e ds sin x dx
0 x 0 0
∫ ∞ (∫ ∞ ) ∫ ∞
−sx fn (s)
= e sin x dx fn (s)ds = ds,
0 0 0 1 + s2
where
logn x
L {fn (s)} = .
x
Here L{f } denotes the Laplace transform of f . For n = 0 we obviously have
1
L{f0 (s)} = , so f0 (s) = 1. Hence
x
∫ ∞
1 π
I0 = 2
ds = .
0 1+s 2
Γ(a + 1)
For n > 0 we start with L{sa } = . By differentiating with respect
xa+1
to a, we get
( )
dn Γ(a + 1)
L{sa logn s} =
dan xa+1
( )
1 ∑
n
j n j dn−j
= (−1) log x Γ(a + 1),
xa+1
j=0
j dan−j
 Solutions 61

so
∑n ( ) j
j n log x (n−j)
L{log s} = lim L{s log x} =
n a
(−1) n
Γ (1)
a→0 j x
j=0
∑n ( )
n
= (−1)j L{fj (s)}Γ(n−j) (1)
j
j=0

∑
n−1 ( )
n
= (−1) j
L{fj (s)}Γ(n−j) (1) + (−1)n L{fn (s)}.
j
j=0

It then follows that

( ) ∑
n−1
n (n−j)
fn (s) = (−1) log s − n
(−1) n
Γ (1)fj (s) j−n
j
j=0
∑n ( )
n n j−1 n
= (−1) log s + (−1) Γ(j) (1)fn−j (s),
j
j=1

and
∫ ∞ ∑
n ( )
n logn s j−1 n
In = (−1) ds + (−1) Γ(j) (1)In−j .
0 1 + s2 j
j=1
Now, for n odd,
∫ ∞ ∫ ∫ ∞
logn s 1
logn s logn s
ds = ds + ds
0 1 + s2 0 1 + s2 1 1 + s2
∫ ∫ 1
1
logn s logn s
= ds − ds = 0,
0 1 + s2 0 1+s
2

while for n even,

∫ ∞ ∫ 1 ∞ ∫ 1
∑
logn s logn s
ds = 2 ds = 2 (−1)k s2k logn sds
0 1 + s2 0 1+s
2
0 k=0
∞
∑ 1
= 2n! (−1)k = 2n!β(n + 1)
(2k + 1)n+1
k=0
(−1)n/2 E nπ
n+1
= n+1
,
2
where β(n) is the Dirichlet beta function and En is the nth Euler number.
Finally, for the values of the derivatives of the Gamma function evaluated at
1, there is the recursion
∑
n
(−1)k+1
Γ(n+1) (1) = −γΓ(n) (1) + n! ζ(k + 1)Γ(n−k) (1),
(n − k)!
k=1
62 Problems

see equation (2.2) in J. Choi and H.M. Srivastava, Evaluation of higher-order

derivatives of the Gamma function, Univ. Beograd. Publ. Elektrotehn. Fak.
Ser. Mat. 11 (2000), 9–18. Thus,
π
I0 =
2
and for n ≥ 1,
∑
2n−1 ( )
2n − 1 (j)
I2n−1 = (−1)j−1 Γ (1)I2n−1−j
j
j=1

and
( )
(−1)n E2n π 2n+1 ∑
2n
j−1 2n
I2n = + (−1) Γ(j) (1)I2n−j .
22n+1 j
j=1
We have two recurrences, one for the derivatives of Γ at 1 and one for
the values of In . We start with Γ′ (1) = −γΓ(1) = −γ. Next we get
Γ′ (1) = −γΓ(1) = −γ,
Γ′′ (1) = −γΓ′ (1) + ζ(2)Γ(1) = γ 2 + ζ(2),
( )
Γ′′′ (1) = −γΓ′′ (1) + 2 ζ(2)Γ′ (1) − ζ(3)Γ(1) = −γ 3 − 3γζ(2) − 2ζ(3),
( )
′′′ 1 ′′ ′
Γ (1) = −γΓ (1) + 6
(4)
ζ(2)Γ (1) − ζ(3)Γ (1) + ζ(4)Γ(1)
2
= γ 4 + 6γ 2 ζ(2) + 3ζ 2 (2) + 8γζ(3) + 6ζ(4).
π
For the first values of In we start with I0 = and we get
2
γπ
I1 = Γ′ (1)I0 = − ,
2
E2 π 3 π3 π π3 γ 2π
I2 = − 3 + 2Γ′ (1)I1 − Γ′′ (1)I0 = + γ 2 π − (γ 2 + ζ(2)) = + ,
2 8 2 24 2
I3 = 3Γ′ (1)I2 − 3Γ′′ (1)I1 + Γ′′′ (1)I0
( 3 )
π γ2π 3γπ 2 π
= −3γ + + (γ + ζ(2)) − (γ 3 + 3γζ(2) − 2ζ(3))
24 2 2 2
γπ 3 3
γ π
=− − − πζ(3),
8 2
E4 π 5
I4 = + 4Γ′ (1)I3 − 6Γ′′ (1)I2 + 4Γ′′′ (1)I1 − Γ(4) (1)I0
25
19π 5 γ 2 π 3 γ 4 π
= + + + 4γπζ(3).
480 4 2

528. If a1 , a2 , a3 , a4 , a5 , a6 are non-negative real numbers such that

a1 a2 + a2 a3 + a3 a4 + a4 a5 + a5 a6 + a6 a1 = 6,
 Solutions 63

a1 ≥ a2 ≥ a3 ≥ a4 ≥ a5 ≥ a6 ,
then
1 1 1 1 1 1 3
+ + + + + ≥ .
a1 + 3 a2 + 3 a3 + 3 a4 + 3 a5 + 3 a6 + 3 2
Proposed by Vasile C^
ırtoaje, Petroleum-Gas University of
Ploieşti, Romania.

Solution by the author. We apply Jensen’s inequality to the function

1
f (x) = , which is strictly convex on (−3, ∞). We have
x+3
1 1 1 1 4
+ + + ≥ ,
a2 + 3 a3 + 3 a4 + 3 a5 + 3 b+3
where
a2 + a3 + a4 + a5
b= , a 2 ≥ b ≥ a5 .
4
Thus, it suffices to show that
1 1 4 3
+ + ≥ .
a1 + 3 a6 + 3 b + 3 2
First we will show that
3b2 + b(a1 + a6 ) + a1 a6 ≤ 6.
Using the constraint from the hypothesis, the inequality writes as
3b2 + b(a1 + a6 ) + a1 a6 ≤ a1 a2 + a2 a3 + a3 a4 + a4 a5 + a5 a6 + a6 a1
⇐⇒ 3b2 + a1 (b − a2 ) + a6 (b − a5 ) ≤ a2 a3 + a3 a4 + a4 a5 .
Since b − a2 ≤ 0, b − a5 ≥ 0, a1 ≥ a2 , and a5 ≥ a6 , it suffices to show that
3b2 + a2 (b − a2 ) + a5 (b − a5 ) ≤ a2 a3 + a3 a4 + a4 a5 ,
which can be rewritten as
a2 a3 + a3 a4 + a4 a5 ≥ 3b2 + b(a2 + a5 ) − a22 − a25 .
Since the sequences a2 , a3 , a4 and a3 , a4 , a5 are decreasing, as a consequence
of the rearrangement inequality we have
3(a2 a3 + a3 a4 + a4 a5 ) ≥ (a2 + a3 + a4 )(a3 + a4 + a5 ) = (4b − a5 )(4b − a2 ).
Thus, it suffices to show that
(4b − a5 )(4b − a2 ) ≥ 9b2 + 3b(a2 + a5 ) − 3(a22 + a25 ),
which is equivalent to 7b2 − 7b(a2 + a5 ) + 3(a22 + a25 ) + a2 a5 ≥ 0 and
7(2b − a2 − a5 )2 + 5(a2 − a5 )2 ≥ 0.
Denote now a1 and a6 by a and c, respectively. So, we need to show
that
1 1 4 3
+ + ≥
a+3 c+3 b+3 2
64 Problems

for a ≥ b ≥ c ≥ 0 such that 3b2 + b(a + c) + ac ≤ 6. From 6 ≥ 3b2 + b(a + c) +

√ a+c
ac > 3b2 , it follows that b < 2. Denoting S = , we need to prove that
2
2S + 6 4 3
+ ≥ .
ac + 6S + 9 b + 3 2
Since ac ≤ 6 − 2bS − 3b , it suffices to show that
2

2S + 6 1 + 3b
− ≥ 0.
15 − 3b + 2(3 − b)S 2(b + 3)
2

After multiplying by 3 − b > 0, we get the equivalent inequalities

18 − 6b + 2(3 − b)S 3b2 − 8b − 3
+ ≥ 0,
15 − 3b2 + 2(3 − b)S 2(b + 3)
3(b − 1)2 3b2 − 8b − 3
1+ + ≥ 0,
15 − 3b2 + 2(3 − b)S 2(b + 3)
3(b − 1)2 3(b − 1)2
+ ≥ 0.
15 − 3b2 + 2(3 − b)S 2(b + 3)
The proof is completed. The equality occurs for a2 = a3 = a4 = a5 = 1
and a1 + a6 + a1 a6 = 3 (a1 ≥ 1 ≥ a6 ).