Mathproblems

ISSN: 2217-446X, url: http://www.mathproblems-ks.com

Volume 2, Issue 4 (2012), Pages 118–143

Editors: Valmir Krasniqi, José Luis Dı́az-Barrero, Valmir Bucaj, Mihály

Bencze, Ovidiu Furdui, Enkel Hysnelaj, Paolo Perfetti, József Sándor, Armend Sh.
Shabani, David R. Stone, Roberto Tauraso, Cristinel Mortici.

PROBLEMS AND SOLUTIONS

Proposals and solutions must be legible and should appear on separate sheets, each
indicating the name of the sender. Drawings must be suitable for reproduction.
Proposals should be accompanied by solutions. An asterisk (*) indicates that nei-
ther the proposer nor the editors have supplied a solution. The editors encourage
undergraduate and pre-college students to submit solutions. Teachers can help by
assisting their students in submitting solutions. Student solutions should include
the class and school name. Solutions will be evaluated for publication by a com-
mittee of professors according to a combination of criteria. Questions concerning
proposals and/or solutions can be sent by e-mail to: [email protected]

Solutions to the problems stated in this issue should arrive before

February 15, 2013

Problems
50. Proposed by Anastasios Kotronis, Athens, Greece. Show that
+∞
X (−1)k 1
= + O(n−1 ln−2 n) n → +∞.
ln(n + k) 2 ln n
k=0

51. Proposed by D.M. Bătineţu-Giurgiu, Matei Basarab National Colege, Bucharest,

Romania and Neculai Stanciu, George Emil Palade Secondary School, Buzǎu, Ro-
mania (Jointly). Let a ∈ (1, ∞) and b ∈ (0, ∞). Calculate
n
!!
X (n + k)a−1
lim n 2 − exp
n→∞ (n + k)a + b
k=1

52. Proposed by Yuanzhe Zhou, The School of Physics and Technology (SPT) at
Wuhan University. Let a, b, c > 0, prove that,
 a 2  b 2  c 2 3
cos + cos + cos >
b+c a+c a+b 2
c 2010 Mathproblems, Universiteti i Prishtinës, Prishtinë, Kosovë.
118
 119

53. Proposed by Ovidiu Furdui, Technical University of Cluj-Napoca, Cluj-Napoca,

Romania. The Stirling numbers of the first kind denoted by s(n, k), are the spe-
cial
Pn numbers defined by the generating function z(z − 1)(z − 2) · · · (z − n + 1) =
k
k=0 s(n, k)z . Let n and m be nonnegative integers with n > m − 1. Prove that
(
1
lnn x (−1)n · n! · ζ(n + 1), m = 1,
Z
dx =
(1 − x)m (−1)n+m−1 · (m−1)!
n!
· m−1 i
P
0 i=1 (−1) · s(m − 1, i) · ζ(n + 1 − i), m ≥ 2,
where ζ denotes the Riemann zeta function.
54. Proposed by Moubinool Omarjee, Paris, France. Let f : [0, +∞) → R be
a measurable function such that g(t) = et f (t) ∈ L1 (R+ ); the space of Lebesgue
integrable functions. Find
Z      n1
nt + 1 nt − 1
lim f (t) 4 sinh sinh dt,
n→+∞ R
+
2 2
ex −e−x
where sinh(x) = 2

55. Proposed by Enkel Hysnelaj, University of Technology, Sydney, Australia. Let

x, y, z, α be real positive numbers. Show that if
X nx3 + (n + 1)x
=α
x2 + 1
cycl

then
X1 2α 27n3
> + 2
x 3 9n α + α3
cycl
where n is a natural number.
56. Proposed byJosé Luis Dı́az–Barrero, BARCELONA TECH, Barcelona, Spain.
Let n ≥ 2 be a positive integer. Find all possible values of the number k such that
2
Fn2 (1 + Fn+1 2
Fn+2 2
) Fn+2 2
(1 + Fn2 Fn+1 ) F 2 (1 + Fn+2
2
Fn2 )
+ = k + n+1
Fn−1 Fn+1 Fn Fn+1 Fn−1 Fn

57.∗ Proposed by Naim L. Braha, Department of Mathematics and Computer Sci-

ences, University of Prishtina, Republic of Kosova. If f0 and f1 are constants, then
prove for which values of λ, the following series converges:
 
b(n−1)/2c
X X j 
Y 
2n−1−2j 1
fn = f1 −1 
λ(im + 1)2
j=0 1i1 ···ij ≤n−2 m=1
 
b(n−2)/2c
X X j 
Y 
2n−2−2j 1
+ f0 − 1 ,
λ(im + 1)2
j=0 2i1 ···ij ≤n−1 m=1

where a  b if and only if a + 1 < b.

120

Solutions
No problem is ever permanently closed. We will be very pleased considering for
publication new solutions or comments on the past problems.

43. Proposed by D.M. Bătineţu-Giurgiu, Matei Basarab National Colege, Bucharest,

Romania and Neculai Stanciu, George Emil Palade Secondary School, Buzǎu, Ro-
mania (Jointly). Let a be a positive real number and Γ(x) be the Gamma function
(or Euler’s second integral). Calculate
    
1 1 1 1
lim (x + a) (Γ(x + 2)) x+1
sin − x (Γ(x + 1)) sin
x
x→∞ x+a x

Solution by Anastasios Kotronis, Athens, Greece. From Stirling’s formula,

we have
√
2πxx+1/2

Γ(x + 1) = 1 + O(x−1 ) x → +∞,
ex
so
ln x ln 2π
ln (Γ(x + 1)) = x ln x − x + + + O(x−1 )
2 2
and
 
1 ln x ln 2π −2
(Γ(x + 1)) = exp ln x − 1 +
x
+ + O(x )
2x 2x
x ln x ln 2π

= + + + O x−1 ln2 x .
e 2e 2e
Putting x + 1 in the preceding expression instead of x, we get
1 x ln x ln 2π 1

(Γ(x + 2)) x+1 = + + + + O x−1 ln2 x .
e 2e 2e e
Furthermore,
 
1
x sin = 1 + O(x−2 ),
x
 
1
(x + a) sin = 1 + O(x−2 )
x+a
and therefore
 
1 1 x ln x ln 2π

x (Γ(x + 1)) x sin = + + + O x−1 ln2 x and
x e 2e 2e
 
1 1 x ln x ln 2π 1

(x + a) (Γ(x + 2)) x+1 sin = + + + + O x−1 ln2 x .
x+a e 2e 2e e
On account of the above
   
1 1 1 1 1
1
(x+a) (Γ(x + 2)) x+1
sin −x (Γ(x + 1)) sin
x
= +O x−1 ln2 x → .
x+a x e e
 121

Also solved by Omran Kouba, Higher Institute for Applied Sciences

and Technology, Damascus, Syria; Paolo Perfetti, Department of Mathe-
matics, Tor Vergata University, Rome, Italy; Konstantinos Tsouvalas,
University of Athens, Athens, Greece; and the proposer

44. Proposed by Ovidiu Furdui, Technical University of Cluj-Napoca, Cluj-Napoca,

Romania. Let A denote the Glaisher–Kinkelin constant defined by
n
Y
−n2 /2−n/2−1/12 n2 /4
A = lim n e k k = 1.282427130 . . . .
n→∞
k=1

Prove that
∞
X ζ(2p + 1) − 1 γ 7
=−
− 6 ln A + ln 2 + ,
p=1
p+2 2 6
P∞
where ζ is the Riemann zeta function defined by ζ(s) = k=1 1/k s for <(s) > 1.

Solution 1 by Paolo Perfetti, Department of Mathematics, Tor Vergata

University, Rome, Italy. First, we have
∞
X ∞ X
X ∞
ζ(2p + 1) − 1 1 1
= 2p+1 p + 2
p=1
p+2 p=1
k
k=2

Since
∞ ∞
!  
X z 2p+1 1 X z 2r z4 1 z4
2 2 2
= 3 −z − =− 3 ln(1 − z ) + z +
p=1
p+2 z r=1
r 2 z 2

then
∞
X ∞ X
X ∞ ∞
X    
ζ(2p + 1) − 1 1 1 3 1 1 1
= = − k ln 1 − + +
p=1
p+2 p=1
k 2p+2 p + 2 k2 k2 2k 4
k=2 k=2

Now, we have
n
X   X
n n
X n
X
3 1 3 3
k ln 1 − 2 = k ln(k + 1) + k ln(k − 1) − 2 k 3 ln k
k
k=2 k=2 k=2 k=2
Xn n
X n
X
= (k + 1)3 ln(k + 1) + (k − 1)3 ln(k − 1) − 2 k 3 ln k
k=2 k=2 k=2
n
X
+ (3(k − 1) ln(k − 1) + 3(k + 1) ln(k + 1))
k=2
Xn


+ 3(k − 1)2 ln(k − 1) − 3(k + 1)2 ln(k + 1)
k=2
Xn
+ (ln(k − 1) − ln(k + 1))
k=2
122

and
n
X    
1
k 3 ln 1 − 2 = (n + 1)3 ln(n + 1) − n3 ln n − 8 ln 2
k
k=2
n
X
+6 k ln k − 3n ln n + 3(n + 1) ln(n + 1) − 6 ln 2
k=1
 
+ −3n2 ln n − 3(n + 1)2 ln(n + 1) + 12 ln 2 − ln n − ln(n + 1) + ln 2
n
X n 1
=6 k ln k − 3n2 ln n + n2 − 3n ln n − − ln n +
2 3
k=1

Moreover
n
X n
X
n(n + 1) 1 1 γ 1
k= − 1, = ln n + − + o(1)
2 2k 2 2 2
k=2 k=2

Adding up the three preceding contributions, we get

n
X ln n 3 2 γ 7
6 k ln k − 3n2 ln n − 3n ln n − + n − ln 2 + −
2 2 2 6
k=1
γ
whose limit when n → ∞ is 6 ln A − ln 2 + 2 − 76 , and the statement follows.

Solution 2 by Anastasios Kotronis, Athens, Greece.

First we observe that for |x| < 1 is

p  
X x2p+1 1 X x2 1 x4
2 2
= = − ln(1 − x ) + x +
p+2 x3 p x3 2
p≥1 p≥3

x 1 ln(1 − x2 )
= − − −
2 x x3
Now, we have
 

+∞
X ζ(2p + 1) − 1 X 1 X 1 X X n−1 2p+1
= − 1 =
p=1
p+2 p+2 n2p+1 p+2
p≥1 n≥1 p≥1 n≥2

X X n−1 2p+1 X  1 
1

3
= − + n + n ln 1 − 2
p+2 2n n
n≥2 p≥1 n≥2

XN    XN   
1 1 1 1
= − lim + n + n3 ln 1 − 2 = − lim + n + n3 ln 1 − 2
N →+∞
n=2
2n n N →+∞
n=2
2n n

N  2 n 3 !!
HN − 1 (N + 2)(N − 1) Y n −1
= − lim + + ln
N →+∞ 2 2 n=2
n2
 
HN N2 N 3
= − lim + + − + ln AN
N →+∞ 2 2 2 2
 123

But
YN N
(n − 1)n Y (n + 1)n
3 3

AN =
n=2
n n3 n=2
n n3
N −1 N
1 Y (k+1)3 −k3 (N + 1)N Y (k−1)3 −k3
3

= k · k
N N3 223
k=2 k=3
N3 N
Y
1 (N + 1)
= · k 6k
2 N (N +1)3
k=1
N
!6
Y 3 2
(N + 1)N N 3N +3N +1/2
−N 2 /2−N/2−1/12 N 2 /4 k
= N e k · ,
k=1
2N (N +1)3 e3N 2 /2

Since HN = ln N + γ + o(1) and

3 2
(N + 1)N N 3N +3N +1/2 N2 N ln N 1
ln (N +1) 3 3N 2 /2 =− − − + − ln 2 + o(1),
2N e 2 2 2 3
then on account of the preceding, we get
+∞
X  
ζ(2p + 1) − 1 γ 7
= lim − − 6 ln A + ln 2 + + o(1)
p=1
p+2 N →+∞ 2 6

as desired.

Also solved by Omran Kouba, Higher Institute for Applied Sciences

and Technology, Damascus, Syria; Konstantinos Tsouvalas, University
of Athens, Athens, Greece; and the proposer.
 
a b
45. Proposed by Moubinool Omarjee, Paris, France. Let M = ∈ M2 (Z).
c d
Prove that there exist (A, B, C) ∈ M2 (Z) × M2 (Z) × M2 (Z) such that M = A2 +
B2 + C 2.

Solution by AN-anduud Problem Solving Group, Ulaanbaatar, Mongo-

lia. (a) Let a − d = 2k + 1, k ∈ Z or a − d = 4k, k ∈ Z. Since 2k + 1 = (k + 1)2 − k 2
and 4k = (k + 1)2 − (k − 1)2 , there exists x, t ∈ Z such that a − d = x2 − t2 . Then we
have a − 1 − x2 = d − 1 − t2 and let us denote a − 1 − x2 = d − 1 − t2 = yz, y, z ∈ Z.
Then, we have
   2 
a b x + yz + 1 b
=
c d c t2 + yz + 1
 2   
x + yz xy + yt 1 b − xy − yt
= +
xz + zt t2 + yz c − xy − yz 1
 2    
x y 1 0 0 b − xy − yt
= + +
z t c − xy − yz 0 0 1
 2  2  2
x y 1 0 0 b − xy − yt
= + + .
z t c − xy − yz 0 0 1
124

(b) Let a − d = 4k + 2, k ∈ Z. Also we have

a − d − 2 = 4k = (k + 1)2 − (k − 1)2 ⇔ a − 2 − (k + 1)2 = d − (k − 1)2 .
Let a − 2 − (k + 1)2 = d − (k − 1)2 = yz, x, y ∈ Z. Then we also can show
   
a b 2 + (k + 1)2 + yz b
=
c d c (k − 1)2 + yz
 
(k + 1)2 + yz 2ky
=
2kz (k − 1)2 + yz
 
2 b − 2ky
+
c − 2kz 0
 2  2
k+1 y 1 0
= +
z k−1 c − 2kz 0
 2
1 b − 2ky
+ .
0 0

Also solved by Omran Kouba, Higher Institute for Applied Sciences and
Technology, Damascus, Syria; and the proposer.

46. Proposed by Anastasios Kotronis, Athens, Greece, and Serafeim Tsipelis, Ion-
nina, Greece (Jointly). Evaluate
+∞
X ζ(2n)
(−1)n−1 ,
n=1
n
where ζ is the Riemann zeta function.

Solution 1 by Ovidiu Furdui, Technical University of Cluj-Napoca, Cluj-

Napoca, Romania. More generally it is known that (cf., [1, Formula 11, p. 160])
∞
X t2k
ζ(2k) = ln Γ(1 + t) + ln Γ(1 − t), |t| < 1.
k
k=1

Letting t → i, in the preceding formula, we get

∞
X ζ(2k)
(−1)k−1 = − ln (Γ(1 + i) · Γ(1 − i))
k
k=1
= − ln (i · Γ(i) · Γ(1 − i))
 
π·i
= − ln
sin(π · i)
 π 
e − e−π
= ln ,
2π
eiz − e−iz
where the last equality follows on account of Euler’s formula, sin z = .
2i
References
[1] H. M. Srivastava and J. Choi, Series Associated with the Zeta and Related Functions, Kluwer
Academic Publishers, 2001.
 125

Solution 2 by Robinson Higuita and Joel Restrepo(Jointly), University

of Antioquia, Colombia. We know that
∞
X 1
ζ(2n) = ,
(k 2 )n
k=1
therefore
∞
X ∞ ∞ ∞ X∞
ζ(2n) X (−1)n−1 X 1 X (−1)n−1 1
(−1)n−1 = = .
n=1
n n=1
n (k 2 )n n=1
n (k 2 )n
k=1 k=1
But   X∞
1 (−1)n−1 1
ln 1 + 2 = .
k n=1
n (k 2 )n
Thus   !
∞
X ∞
X ∞ 
Y
n−1 ζ(2n) 1 1 + k2
(−1) = ln 1 + 2 = ln .
n=1
n k k2
k=1 k=1
We see that
Y∞  
1 sinh π
1+ 2 =
n=1
n π
All zeros of the entire function sinh z are first order and they are the points
0, −πi, πi, · · · , −nπi, nπi, · · · , the convergence exponent of this sequence obviously
is τ = 1. Moreover,
X∞ X ∞ X∞
1 1 1
= = +∞, while , < +∞
n=1
|nπi| n=1 nπ n=1
(nπ)2
i.e. κ = 1. Hence, by Theorem (Hadamard’s Factorization)
∞  ∞  
Y z  −z/nπi  z  z/nπi Y z2
sinh z = eg(z) z 1+ e 1− e = eg(z) z 1+ .
n=1
nπi nπi n=1
(nπ)2
Further, the order of the entire function sinh z obviously is ρ = 1, and hence
g(z) = c0 + c1 z. The easiest way of determining the coefficients c0 and c1 is to
observe that
sinh z
eg(z) = Q  
∞ z2
z n=1 1 + (nπ) 2

is an even function, and hence ec0 +c1 z = ec0 −c1 z . Hence c1 = 0. Putting z → 0 in
the above expression we conclude that ec0 = 1, and hence g(z) = 0. Thus,
Y∞  
z2
sinh z = z 1+ .
n=1
(nπ)2
Finally,
Y∞  
sinh π 1
= 1+ 2 .
π n=1
n

Also solved by Omran Kouba, Higher Institute for Applied Sciences and
Technology, Damascus, Syria; Paolo Perfetti, Department of Mathemat-
ics, Tor Vergata University, Rome, Italy; AN-anduud Problem Solving
Group, Ulaanbaatar, Mongolia; Konstantinos Tsouvalas, University of
126

Athens, Athens, Greece; Joaquin Rivero Rodriguez, Spain; and the pro-
poser.

47. Proposed by David R. Stone, Georgia Southern University, Statesboro, GA,

USA. Identify the graph of the equation

y 4 − 10x2 y 3 + 25x4 y 2 − 50x6 y + 24x8 = 0

Solution by Omran Kouba, Higher Institute for Applied Sciences and

Technology, Damascus, Syria. Let

G = (x, y) ∈ R2 | y 4 − 10x2 y 3 + 25x4 y 2 − 50x6 y + 24x8 = 0 .

Clearly,
[n y  o
G = {(0, 0)} (x, y) ∈ R∗ × R : f = 0 ,
x2
where f (t) = t4 − 10t3 + 25t2 − 50t + 24. Now, a simple study of the variations
of f shows that the equation f (t) = 0 has exactly two real zeros λ1 ∈ (0, 1) and
λ2 ∈ (7, 8). A numeric evaluation shows that λ1 ≈ 0.63270759 and λ2 ≈ 7.49826796.
Cosequently, G is the union of the two parabolas P1 and P2 of equations y = λ1 x2
and y = λ2 x2 respectively.

Also solved by the proposer.

48. Proposed by Florin Stanescu, School Cioculescu Serban, Gaesti, jud. Dambovita,
Romania. Let f : (0, ∞) → (0, ∞) be a differentiable function for which there exist
real numbers a and b, 0 < a < b such that
R b f (x) f (b) f (a)
(i): a 2 dx = −
0
x b a
f (a) 2
(ii): =
f (a) a
(a) Find an example of a function that satisfies the preceding conditions
(b) Show that there exists a c ∈ (a, b) such that

f (c)  c 2  0
f (c)

(c−a) f (c) − 2c
= e
f (a) a

Solution by Adrian Naco, Polytechnic University, Tirana, Albania.

(a) The function f (x) = kx2 , where k is a constant number, satisfy the condition
(ii), ∀x ∈ (0, ∞).

(b) The function F is defined by

    

 f (x) f (a)
 ln
 −ln
 x2 a2
, x ∈ (a, ∞)
F (x) = x−a




 0 x=a
 127

which is everywhere continuous and differentiable in (0, ∞) and even continuous in

the point a. Indeed from the definition
  0 0
f (x) f (a) 2
lim+ F (x) = ln = − = 0 = F (0)
x→a x2 x=a f (a) a
On account of (i) and applying the integration by part method we have
Z b  0  b Z b
f (x) f (x) f (x)
x 2
dx = x 2 − dx
a x x a a x2

  Z b
f (b) f (a) f (x)
= − − dx = 0
b a a x2
 0  0
f (x) f (x)
Case 1. If ≥ 0 or ≤ 0 for all x ∈ (0, ∞). Then, considering the
x2 x2
 0
f (x)
above result, we get = 0 for all x ∈ (0, ∞) ⇒ f (x) = kx2 , with k ∈ R in
x2
which case, the condition (b) holds for every c ∈ (0, ∞).

Case 2. There are at least two points α, β ∈ (a, ∞) such that

 0  0
f (x) f (x)
>0 and <0
x2 x=α x2 x=β
f (x)
Since the function is continuous and differentiable in [α, β], based on Darboux
x2
 0
f (x)
Theorem it implies that there exists a point γ ∈ (α, β) such that = 0.
x2 x=γ
Thus
  0 0
f (x) f (γ) 2
lim F (x) = ln = − =0
x→γ x2 x=γ f (γ) γ
The function F is continuous in [α, γ] and differentiable in (α, γ), thus, considering
the fact that F (α) = F (γ) = 0 and based on Rolle’s Theorem, we have that there
 ln f (x) f (a) 
x2 − ln a2
0
exists c ∈ (α, γ) such that F 0 (c) = 0 or = 0. That is,
x−a x=c
 0   
f (c) 2 f (c) f (a)
(c − a) f (c) − c − ln c2 − ln a2  0   
f (c) 2 f (c) f (a)
= 0 ⇒ (c−a) − − ln 2 −ln 2 = 0
c−a f (c) c c a
or
   
0 0
f (c) f (c)
(c−a) − 2c (c−a) − 2c
f (c) f (a) f (c) f (c) f (a) f (c)
ln 2 = ln 2 + ln e ⇔ ln 2 = ln 2 + ln e
c a c a
and    
0 0
f (a)
− 2c
 2 (c−a) f (a)
− 2c
f (c) f (a) (c−a) f (c) f (c) c f (c)
= e ⇔ = e
c2 a2 f (a) a
128

Also solved by AN-anduud Problem Solving Group, Ulaanbaatar, Mon-

golia; and the proposer.

49. Proposed by Mihály Bencze, Braşov, Romania. Let a, b, c be positive real

numbers. Prove that
p
a b c 3
(a − b)2 (b − c)2 (c − a)2
+ + ≥3+
b c a ab + bc + ca

Solution 1 by Adrian Naco, Polytechnic University, Tirana, Albania.

The generalization of the problem 49. Let x1 , x2 , . . . xn , (n ≥ 3), be positive real
numbers. Prove that
v
uY
u n
t
n
(xi − xi+1 )2
Xn
xi n i=1
≥3+ n ,
x 2 X
i=1 i+1
xi xi+1
i=1

where xn+1 = x1 . Based on Chauchy-Schwarz inequality we have

X n 2 X n 2 Xn  2 Xn  2
xi √ xi √
xi = √ xi xi+1 ≤ √ xi xi+1
i=1 i=1
xi xi+1 i=1
xi xi+1 i=1

X
n 2
X
n X
n  n Xn  xi
xi 2 X xi xi 2
= xi xi+1 ⇒ = ≥ ni=1
xi xi+1 x x x X
i=1 i=1 i=1 i+1 i=1 i i+1
xi xi+1
i=1

Furthermore
X
n 2
n Xn  xi
X xi xi 2 i=1
−3= −3≥ n −3
x xx X
i=1 i+1 i=1 i i+1
xi xi+1
i=1

n
X n
X n
X
xi 2 + 2 xi xj 3 xi xi+1
i=1 1≤i<j≤n i=1
= n − n
X X
xi xi+1 xi xi+1
i=1 i=1

n
X n
X n
X n
X n
X X
n 
2 2 1 2
xi + 2 xi xi+1 3 xi xi+1 xi − xi xi+1 2 (xi − xi+1 )
i=1 i=1 i=1 i=1 i=1 i=1
≥ n − n ≥ n = n
X X X X
xi xi+1 xi xi+1 xi xi+1 xi xi+1
i=1 i=1 i=1 i=1
 129

v v
uY uY
u n u n
nt
n
(xi − xi+1 )2 t
n
(xi − xi+1 )2
1 i=1 n i=1
≥ n = n
2 X 2 X
xi xi+1 xi xi+1
i=1 i=1
The last inequality is based on the well-known AM-GM inequality
√
a1 + a2 + . . . an ≥ n n a1 a2 . . . an

where a1 , a2 , . . . , an ∈ R+ , and ai = (xi − xi+1 )2 , ∀i ∈ {1, 2, . . . , n}.

The equality holds only for n = 3 and xi = x2 = . . . = xn . For n ≥ 4 the equality
is not true because
n
X n
X
xi xj > xi xi+1
1≤i<j≤n i=1

Solution 2 by Eric Milesi Vidal, BARCELONA TECH, Barcelona, Spain.

First we subtract 3 from each side of the inequality
p
a−b b−c c−a 3
(a − b)2 (b − c)2 (c − a)2
+ + ≥
b c a ab + bc + ca
Now we prove that the following inequalities hold
X 1 1 1 (a − b)2 + (b − c)2 + (c − a)2 p
(a−b)ca( + + ) ≥ ≥ 3 (a − b)2 (b − c)2 (c − a)2
a b c 3
cyclic

RHS is holds applying AM −GM inequality. Equality holds if and only if a = b = c.

To prove LHS inequality, we have see that
X ca2 (a − b)2 + (b − c)2 + (c − a)2 X 3ca2 X
a2 −ba−bc+ ≥ ⇔ a2 + ≥4 ab,
b 3 b
cyclic cyclic cyclic

or equivalently,
X X X ca2 X
a2 ≥ ab and 3× ≥3× ab
b
cyclic cyclic cyclic cyclic

The first inequality is well-known. For the second, we may assume WLOG that
a ≥ b ≥ c so cb ca ba
a ≤ b ≤ c Since both sequences are sorted in the opposite way by
applying the rearrangement inequality, we get
ca2 ab2 bc2
ab + bc + ca ≤+ +
b c a
Equality holds if and only if a = b = c and we are done.

Also solved by AN-anduud Problem Solving Group, Ulaanbaatar, Mon-

golia; Paolo Perfetti, Department of Mathematics, Tor Vergata Univer-
sity, Rome, Italy; Ioan Viorel Codreanu, Satulung, Maramure, Romania
and the proposer.
130

MATHCONTEST SECTION
This section of the Journal offers readers an opportunity to solve interesting and ele-
gant mathematical problems mainly appeared in Math Contest around the world
and most appropriate for training Math Olympiads. Proposals are always wel-
comed. The source of the proposals will appear when the solutions be published.

Proposals
36. Find all positive integers n such that 17n−1 + 19n−1 divides 17n + 19n .

37. Let α, β and γ be three distinct complex numbers. Show that they are collinear
if, and only if, Im(αβ + βγ + γα) = 0.

38. Through the midpoint M of a chord P Q of a circle, any other chords AB

and CD are drawn; chords AD and BC meet P Q at points X and Y respectively.
Prove that M is the midpoint of XY.

39. The students of a University Course in Mathematics take their exams in Calcu-
lus, Algebra, Physics and Geometry. It is known that 73% passed Calculus exam,
82% passed Algebra, 77% passed Physics and 89% passed Geometry. At least, how
many students have passed the exam of all four subjects?

40. Let α, β, γ be the angles of an acute triangle ABC. Prove that

 2/3
1 X tan2 α 1
+3 ≥2
3 tan β tan γ tan α + tan β + tan γ
cyclic
 131

Solutions
32. Two circles C1 and C2 have in common two points X and Y . We draw a
line that cuts C1 at points A and B. Next we draw the lines AX, AY which cut
C2 at points AX and BY and lines BX, BY which cut C2 at points BX and BY
respectively. Show that the three lines AB, AX BY and AY BX are parallel.

(Short List of International Mathematical Arhimede Contest 2012)

Solution by Guillem Alsina Oriol, BARCELONA TECH, Barcelona,

\ = BXY
Spain. In the following figure we observe that BAY \ by spanning arc
\ \ \ \
(or capacius arc), and BXY = BXP + P XY . Moreover, BXP is equal to QXB \X
(opposite by the vertex), and they are equal to XA\ Y BX on account of the half
inscribed angle. Also by using again the half inscribed angle property, we have
\
P XY = Y\ AY X.

Q BX
AX

C1 X

BY
B Y

P
C2
AY

So, we have
\ = XA
BAY \ \ \
Y BX + Y A Y X = Y A Y BX

\ and Y \
Since line AAY cuts lines AB and AY BX and the angles BAY AY BX are
equal, then lines AB and AY BX are parallel.

Now, we only need to see that AX BY is also parallel to these two lines. It will be
suffice to see that the angles AX\ BY BX and AY\ BX BY are equal. We know that
AX\ BY BX is equal to AX \ XBX by spanning arc. Furthermore, the last one is equal
\ because they are angles opposite by the vertex. Likewise, AY\
to AXB BX BY is
equal to A\Y Y B Y , which is equal to \
AY B. Finally, since \
AXB = \
AY B by spanning
\ \
arc, then AX BY BX = AY BX BY , and the three lines AX BY , AY BX and AB are
parallel. 2

Also solved by José Luis Dı́az-Barrero, BARCELONA TECH, Barcelona,

Spain, Adrian Naco, Polytechnic University, Tirana, Albania, Omran
132

Kouba, Higher Institute for Applied Sciences and Technology, Damas-

cus, Syria and José Gibergans-Báguena, BARCELONA TECH, Barcelona,
Spain.

33. Four dice are thrown at the same time on a table. Find the probability that the
sum of the points appeared in the upper faces lies between 14 and 18 points.

(Training Sessions of Spanish Team for IMO 2012)

Solution by José Luis Dı́az-Barrero, BARCELONA TECH, Barcelona,

Spain. We have to count how many possibilities there are to obtain sums of 15, 16
and 17 points, respectively. To do it we consider the polynomial
z + z2 + z3 + z4 + z5 + z6
Here, the powers of z keep track of the different faces of the dice and the coefficients
of the powers of z show the number of occurrences of each face. The second, third
and fourth dies represented by the same polynomial and the outcome of throwing
are represented quite naturally by the polynomial
(z+z 2 +z 3 +z 4 +z 5 +z 6 )(z+z 2 +z 3 +z 4 +z 5 +z 6 )(z+z 2 +z 3 +z 4 +z 5 +z 6 )(z+z 2 +z 3 +z 4 +z 5 +z 6 )
By expanding this, we get
z 4 + 4 z 5 + 10 z 6 + 20 z 7 + 35 z 8 + 56 z 9 + 80 z 10 + 104 z 11 + 125 z 12 + 140 z 13 + 146 z 14
+140 z 15 + 125 z 16 + 104 z 17 + 80 z 18 + 56 z 19 + 35 z 20 + 20 z 21 + 10 z 22 + 4 z 23 + z 24
and we find that there is one way of obtaining a sum score of 4; there are four
ways of getting a sum score of 5; ten ways of getting a sum score of 6 and so on.
Thus, the number of favorable sums are the coefficients of z 15 , z 16 , z 17 , respectively.
That is, Nf = 140 + 125 + 104 = 369 and the number possible sums Np = 1296.
Therefore, the probability required is
Nf 369 41
P(14 < s < 18) = = =
Np 1296 144
2

Also solved by José Gibergans-Báguena, BARCELONA TECH, Barcelona,

Spain, Eric Milesi Vidal, BARCELONA TECH, Barcelona, Spain, and
Oscar Rivero Salgado, BARCELONA TECH, Barcelona, Spain.

34. Let P be an interior point of triangle ABC and let HA , HB , HC be the or-
thocenters of triangles P BC, P AC, and P AB, respectively. Prove that triangles
HA HB HC and ABC have the same area.

(Catalonian Mathematical Olympiad 2012)

Solution by José Luis Dı́az-Barrero BARCELONA TECH, Barcelona,

Spain. First, we compute the distance from vertex A to the orthocenter of 4ABC.
That is, |AH| = dist(A, H). Since 4BCC 0 ∼ 4AHC 0 , then
AH AC 0 AH AC 0 AC 0
= ⇔ = ⇒ AH = a
CB CC 0 a CC 0 CC 0
Furthermore, in triangle ACC we have AC = b cos A and CC 0 = b sin A. Substi-
0 0
 133

A H

C
C
c A
b H
c
b

A a a
C B C A B

tuting in the preceding expression yields

AH b cos A
= = | cot A|
a b sin A
from which follows AH = a | cot A|. Notice that if 4ABC is a right triangle with
Ab = 90◦ the expression is also valid, but in this particular case is A = H and
AH = a cot 90◦ = 0. If A b > 90◦ , then point H lie on the exterior of 4ABC and we
have AC = b cos(180 − A) and CC 0 = b sin(180◦ − A). So, AH = a cot A and in
0 ◦

this case is cot A < 0. Likewise, we can obtain the distances from H to the acute
vertices (angles) of a right or obtuse triangle ABC.

Joining the arbitrary point with the vertices A, B, C we get the triangles P AB, P BC,
and P CA. Let α = ∠BP C, β = ∠AP B, γ = ∠AP C. Obviously α + β + γ = 360◦ .
Al least two of these three angles are obtuse and the other is obtuse, right or acute.
So we will examine
Cuando these
unimos el punto three cases
arbitrario P conseparately.
los vértices A, B y C del triángulo obtenemos
los tres triángulos P AB, P BC y P CA.
Sean α = ∠BP C, β = ∠AP C, γ = ∠AP B. Evidentemente, α + β + γ = 360 . De estos ◦
(i) Suppose that the three angles are obtuse. Then, we have P HA = −a cot α
tres ángulos, com mı́nimo dos son obtusos. El otro puede ser obtuso, recto o agudo.
C = −c cot
and P HEstudiaremos los γ
treson account
casos of the preceding. Notice that the angle y =
per separado.
∠HA P H1)C = 180 ◦
− ∠AP C = 180 ◦
− B because the sides HA P dicho
Supongamos que los tres ángulos son obtusos (Figura 1). Por lo que hemos
and HC P are,
respectively, perpendicular
al principio tenemos P HA to = −athe sides
cot α y P HBC and
C = −c cot γ.AB and one
Fijémonos que elisángulo
obtuse and the
y = ∠H
other acute. AP H
The C = 180
area A(P−H∠AP
◦
C = 180◦ − B ya que los lados H P y HC P son,
A HC ) of triangle P HA HC is A
respectivamente, perpendiculares a los lados, BC y AB y un es obtuso y el otro es
agudo. P H A P H C sin y ac cot α cot γ sin B
A(P H HCA(P
ElAárea ) =HAHC ) del triángulo P HA=HC es, obviamente, = A(ABC) cot α cot γ
2 2
P H A P H C sin y ac cot α cot γ sin B
A(P HA HC ) = = = A(ABC) cot α cot γ.
2 2
HA
A

z
γ
y Figura 1
β
αP
x
HC

B C

HB
Sumando, pues, las áreas de los tres triángulos P HA HB , P HB HC y P HC HA obtenemos
A(HA HB HC ) = A(P HA HB ) + A(P HB HC ) + A(P HC HA ) i
Adding up the areas of the three triangles

P HA HB , P HB HC y P HC HA we obtain
A(H H H ) = A(ABC) cot α cot β + cot β cot γ + cot γ cot α .
A(HA HB HAC B) =C A(P HA HB ) + A(P HB HC ) + A(P HC HA ) and
Como que α + β = 360 − γ, tenemos que cot(α + β) = − cot γ o, equivalentemente,
1 − cot α cot β
cot γ = − cot(α + β) =
cot α + cot β
o bien, cot α cot β + cot β cot γ + cot γ cot α = 1. (*)
De aquı́ resulta A(HA HB HC ) = A(ABC).

2) Supongamos que uno de los ángulos es recto, per ejemplo β = 90◦ (Figura 2).
Entonces HB = P y
P H A P H C sin y ac cot α cot γ sin B
134

 
A(HA HB HC ) = lA(ABC) cot α cot β + cot β cot γ + cot γ cot α
Since α + β = 360 − γ, then we have cot(α + β) = − cot γ or, equivalently,
1 − cot α cot β
cot γ = − cot(α + β) =
cot α + cot β
or
cot α cot β + cot β cot γ + cot γ cot α = 1 (1)
from which follows A(HA HB HC ) = A(ABC).

(ii) Assume that one of the angles is right, say β = 90◦ (Figura 2). Then HB = P
and
P H A P H C sin y ac cot α cot γ sin B
A(HA HB H = A(HAahora
3) C )Supongamos P HCque )= uno de los ángulos α, β, γ = 
es agudo, por ejemplo, AP = A(ABC) cot α cot γ
C=
2 2
β < 90◦ (Figura 3). El punto P es exterior al triángulo H H H y tenemos A(H H H ) =
A B C A B C
A(P HA HC ) − A(P HA HB ) − A(P HC HB ).
HA

Figura 2
HA
A
A Figura 3

HB
P = HB
P
B C B C
HC

HC

Pero en este caso tenemos P HB = b cot β, P HA = −a cot α i P HC = −c cot γ y, por lo

On account of (1), if cot β = 0 then cot α cot γ = 1, and the results follows in this
tanto,
case.
A(HA HB HC ) = A(P HA HC ) − A(P HA HB ) − A(P HC HB ) =
 
(iii) Suppose that one ofαthe
= cot cot γ angles
− (− cot αα,
cotβ,
β) γ is cot
− (− acute,
γ cot β)say [
AP
A(ABC)C= = β < 90◦ (Figura
3). Point P is exterior to= (cot
theα triangle
cot β + cot βHcot γH+ cot γ cot α)A(ABC) = A(ABC).
A B HC and we have A(HA HB HC ) =
A(P HA HC ) − A(P HA HB ) − A(P HC HB ). But in this case we have P HB = b cot β,
P HA = −a cot α and P HC = −c cot γ. So,
A(HA HB HC ) = A(P HA HC ) − A(P HA HB ) − A(P HC HB )


= cot α cot γ − (− cot α cot β) − (− cot γ cot β) A(ABC)
= (cot α cot β + cot β cot γ + cot γ cot α)A(ABC) = A(ABC)
2
Also solved by José Gibergans-Báguena, BARCELONA TECH, Barcelona,
Spain, Eric Milesi Vidal, BARCELONA TECH, Barcelona, Spain, Os-
car Rivero Salgado, BARCELONA TECH, Barcelona, Spain, and Ioan
Viorel Codreanu, Satulung, Maramure, Romania.

35. Compute
Z n
1 n2 + x2
lim dx
n→∞ n3 0 5−x + 7
 135

(Training Sessions of Spanish Math Team for IMC 2012)

Solution by José Gibergans-Báguena, BARCELONA TECH, Barcelona,

Spain. We begin with a Lemma.
Lemma. Let f : (0, +∞] → R and g : [0, 1] → R be continuous functions. If
lim f (x) = `, then
x→∞
Z x Z 1
1 n
lim f (x)g dx = ` g(x) dx
n→∞ n 0 n 0

Proof. Since the function h(x) = f (x) − ` tends to zero when x → ∞, then
Z x Z x Z
1 n 1 n ` n x
f (x)g dx = h(x)g dx + g dx (2)
n 0 n n 0 n n 0 n
Since g(x) is continuous in [0, 1], then |g(x)| ≤ M for some M ∈ R and
Z x Z
1 n M n
h(x)g dx ≤ |h(x)| dx (3)
n 0 n n 0
If H(x) is a primitive of |h(x)| then, applying L’Hospital’s rule, we have
H(n)
lim = lim |h(x)| = 0
n→∞ n x→∞
and the RHS of (3) tends to zero when n → ∞. Now, setting x/n = t, we have
Z Z 1
` n x
g dx = ` g(t) dt
n 0 n 0
and from (1) the proof immediately follows.
1
Setting f (x) = −x and g(x) = 1 + x2 into the preceding Lemma and taking
5 +7
1 1
into account that lim −x = , we get
x→∞ 5 +7 7
Z n 2 Z
1 n + x2 1 1 1 x3 1 4
lim 3 −x
dx = g(x) dx = +x =
n→∞ n 0 5 +7 7 0 7 3 0 21

and we are done. 2

Also solved by José Luis Dı́az-Barrero, BARCELONA TECH, Barcelona,
Spain, Anastasios Kotronis, Athens, Greece, Ioan Viorel Codreanu, Sat-
ulung, Maramure, Romania, Paolo Perfetti, Department of Mathemat-
ics, Tor Vergata, University, Rome, Italy and Omran Kouba, Higher
Institute for Applied Sciences and Technology, Damascus, Syria.

Editors Comment. Problem 31 was retracted by the proposer.

136

MATHNOTES SECTION

A mean-value formula for the floor

function on integers
Wang Xingbo

Abstract. The paper first proves several inequalities related with the floor func-
tion, and then deduces and proves a mean-value formula for the floor function with
an integer variable. The inequalities and the formulae are useful in some aspects
related to analysis and computation of the floor function.

1. Introduction
The floor function and the ceiling function are two special functions that have
integer values, and they are widely applied in number theory, discrete mathematics,
calculus and computer science. A general introduction to the two functions can be
seen in detail in Graham’s book [1]. Readers can see that the properties of the
two functions are miraculous and fascinating. However, it also can be seen that
there still remain quite a lot of problems for us to study. Due to having discrete
characteristics of integers, the two functions are still lack of analytic tools like the
mean-value theorem in calculus to analyze their intermediary status. Hence it is
worth to draw a mean-value theorem for either of the two functions.
In a study on problems of binary trees, I found and proved the following formula
  j  
α+δ α k δ−1 1
= + +
2z(α) 2z(α) 2z(α) 2
Since the previous expression is quite similar to the formula f (x0 + ∆x) = f (x0 ) +
f 0 (ξ)∆x in calculus, I called it a mean-value formula of the floor function. This
paper mainly presents the way to obtain and to prove it.

2. Preliminaries
This section presents some necessary preliminaries for later sections.
2.1. Definition and Symbols. The floor function of a real number x is denoted
by bxc and it satisfies bxc ≤ x < bxc + 1; the fractional part of x is denoted by
{x} and it satisfies x = bxc + {x}; and the ceiling function of x is denoted by dxe
verifying x ≤ dxe < x + 1. The function z(α), which is defined in [2], represents
the position of the first 0-bit that occurs from the least significant bit (lsb) of α’s
binary representation, e.g., z(0) = z((00000000)2 ) = 1, z(1) = z((00000001)2 ) =
2, z(83) = z((01010011)2 ) = 3.
2.2. Lemmas. The following lemmas are found respectively in Graham’s book [1],
and Wang’s paper[2].
Lemma 1. For arbitrary real numbers x, y and an integer n, it holds
(P1): bxc + byc ≤ bx + yc ≤ bxc + byc + 1
 137

(P2): bxc − byc − 1 ≤ bx − yc ≤ bxc − byc < bxc − byc + 1

(P3): bn + xc = n + bxc
(P4): n bxc ≤ bnxc ; n bxc = bnxc ⇔ n{x} < 1  
(P5): bnxc
 n  = bxc+  x + n1 +...+ x + n−1
n ,particularly,bxc+ x + 12 = b2xc
n−1
(P6): m = m + 1, where m > 0 is an integer.
Lemma 2. For integer α ≥ 0, z(α) is the smallest positive solution of the following
modulo-inequality with unknown x.
0 ≤ α mod 2x < 2x−1

3. Main Results
In the following we present our main results. We begin with (see [4])
Theorem 1. For arbitrary real numbers x, y, it holds
i. bxc > byc ⇒ x > y; bxc < byc ⇒ x < y
ii. x ≤ y ⇒ bxc ≤ byc ; x ≥ y ⇒ bxc ≥ byc
The following result shows the case bxc = byc.
Theorem 2. For a real number ξ and a real number δ > 0,if bξ + δc = bξc, then
for an arbitrary real number ω such that 0 ≤ ω ≤ δ, it holds
bξ + ωc = bξc
For a real number ρ < 0, if bξ + ρc = bξc, then for an arbitrary real number η such
that ρ ≤ η ≤ 0, it holds
bξ + ηc = bξc
Proof. Since 0 ≤ ω ≤ δ, then
bξc ≤ ξ ≤ ξ + ω ≤ ξ + δ < bξ + δc + 1
and the condition bξ + δc = bξc becomes
bξc ≤ ξ + ω < bξc + 1
That is the definition of the floor function. Next, we prove the second conclusion.
Indeed, by the condition ρ < 0 and ρ ≤ η ≤ 0, yields
bξ + ρc ≤ ξ + ρ ≤ ξ + η ≤ ξ < bξc + 1
Considering bξ + δc = bξc, it immediately leads to

bξc ≤ ξ + η < bξc + 1

Hence the theorem holds.
Theorem 3. For an arbitrary integer α ≥ 0, it holds
j α k j α k
2 z(α) = z(α)−1 (4)
2 2
Proof. For convenience, let I = z(α). By lemma 2, I satisfies 0 ≤ α mod 2I < 2I−1
2I
which is equal to 0 ≤ 2( α mod
2I
) < 1. Note that
α mod 2I α
0 ≤ 2( ) < 1 ⇔ 2{ I } < 1
2I 2
138

By the property (P4) of lemma 1, it yields

α mod 2I α jαk j α k
0 ≤ α mod 2I < 2I−1 ⇔ 0 ≤ 2( ) < 1 ⇔ 2{ } < 1 ⇔ 2 = I−1
2I 2I 2I 2
which proves the statement.
Theorem 4. For an arbitrary integer α ≥ 0, it holds
    j
α−1 1 α 1 α k
− = − = −1
2z(α) 2 2z(α) 2 2z(α)
  j
α+1 1 α k
z(α)
+ = z(α) + 1
2 2 2
  j k      
α+1 1 α α−1 1 α+1 α 1
− = z(α) = + = = z(α) +
2z(α) 2 2 2z(α) 2 2z(α) 2 2
Proof. Also use the symbol I = z(α) for convenience. In the case I = 1, it is
obvious that  α−1 1
 2I − 2 < 2αI − 12 = α−1 2I
< α+1
2I
− 12 = 2αI
(5)
 α α−1 1 α+1 α 1 α+1 1
2I = 2 I + 2 < 2 I = 2I + 2 < 2I + 2
With a direct computation and by the property (P3) of lemma 1, it holds
  j k jαk   j k
α−1 1 α α+1 1 α
− = − 1 = − 1; + = I +1
2I 2 2I 2I 2I 2 2
By using (4), it yields
      jαk
α 1 α−1 α 1
− = = + − 1 = − 1;
2I 2 2I 2I 2 2I
  j k      
α+1 1 α α−1 1 α+1 α 1
− = = + = = + ;
2I 2 2I 2I 2 2I 2I 2
Hence the theorem holds in the case I = 1.
When I > 1, it knows
 α−1 1
 2I − 2 < 2αI − 12 < α+1
2I
− 12 ≤ α−1
2I
< 2αI
(6)
 α α+1 α−1 1 α 1 α+1 1
2I
< 2I
≤ 2I
+ 2 < 2I
+ 2 < 2I
+ 2
First by using (4), it yields
  j
α 1 α k jαk jαk
I
+ = I−1 − I = I
2 2 2 2 2
Referring to the theorem 2 and the formula (6), it immediately obtains
j α k α + 1 α − 1 1
 
α 1

= = + = I + (7)
2I 2I 2I 2 2 2
Now, we prove   j k
α+1 1 α
− = I (8)
2I 2 2
In fact, when I > 1, referring to the property (P1) of lemma 1, it yields
    j k   j k
α−1 1 α 1 1 α 1 1 α
+ = I + − I ≥ I + − = I
2I 2 2 2 2 2 2 2I 2
 139

By the properties (P2) and (P3) of lemma 1 and referring to (7), it holds
          jαk
α−1 1 α 1 1 α 1 1 α 1
+ = + − < + − +1 = + +1 = +1
2I 2 2I 2 2I 2I 2 2I 2I 2 2I
Thus the formula (8) holds, and it leads to
      jαk
α+1 1 α+1 1 α+1 1
I
+ = I
− + 1 = I
− +1= I +1 (9)
2 2 2 2 2 2 2
Furthermore, it holds
    jαk jαk
α+1 1 α+1 1
I
− = I
+ −1= I +1−1= I (10)
2 2 2 2 2 2
The formulas (7) (8) (9) (10) show that the theorem 4 holds in the case I > 1.
Hence theorem is proven.  
By theorem 4, we know that the expression 2α−1
z(α) satisfies
  j k 
α−1 α −1, z(α) = 1
= z(α) +
2 z(α) 2 0, z(α) > 1
Theorem 5. Let α ≥ 0, δ and χ be integers such that χ = ..., −3, −2, −1, 0, 1, 2, 3, ...,
z(α) z(α)
(χ − 21 ) × 2 < δ ≤ (χ + 12 ) × 2 ; then it holds
   
α+δ α
= +χ (11)
2z(α) 2z(α)
 
Proof. For convenience,we use the symbol v(α, δ) to denote 2α+δ
z(α) , use the symbol
z(α) z(α)
Iχ to denote the interval (χ − 12 ) × 2 < δ ≤ (χ + 21 ) × 2 and keep using
I = z(α). Note that the condition of the theorem 5 is the same as that of the
theorem 4; hence the conclusions drawn in the theorem 4 can be directly adopted
here. According to thetheorem 2, it is merely necessary to prove that v(α, δ) takes
an identical value 2αI + χ, which merely depends upon α and χ, on the whole
interval Iχ . Also by the theorem 2, this is a job to compute the values of v(α, δ) at
five points δ = (χ − 12 ) × 2I , δ = (χ − 12 ) × 2I + 1, δ = χ × 2I , δ = (χ + 21 ) × 2I and
δ = (χ + 12 ) × 2I + 1, as follows.
(i). When δ = (χ − 12 ) × 2I , it yields
  j α+(χ− 12 )×2I k
v(α, δ) = α+δ 2I = 2I
(12)
     
= 2αI + χ − 21 = χ + 2αI − 21 = χ − 1 + 2αI
(ii). When δ = (χ − 12 ) × 2I + 1, it yields
 α+δ  j k
α+(χ− 12 )×2I +1
v(α, δ) = 2I
= 2I
(13)
 α+1 1
  α+1 1
 α
= 2I
+χ− 2 =χ+ 2I
− 2 =χ+ 2I
I
(iii). When δ = χ × 2 , it yields
    j k jαk
α+δ α + χ × 2I α
v(α, δ) = = = + χ = χ + (14)
2I 2I 2I 2I
140

(iv). When δ = (χ + 12 ) × 2I , it yields

 α+δ  j k
α+(χ+ 12 )×2I
v(α, δ) = 2I
= 2I
(15)
α     
= 2I
+ 1
2 + χ = χ + 2αI + 12 = χ + 2αI
(v). When δ = (χ + 12 ) × 2I + 1, it yields
 α+δ  j k
α+(χ+ 12 )×2I +1
v(α, δ) = 2I
= 2I
(16)
 α+1 1
 α
=χ+1+ 2I
− 2 =χ+1+ 2I

Obviously,
j the
k results in (13),(14) and (15) show that v(α, δ) does take an identical
value of 2αI + χ on the whole interval Iχ , and the value merely depends upon α
and χ since I = z(α) depends on α . Note that the results
j k(12) and (16) also imply
that, for a fixed χ , v(α, δ) takes an identical value of 2αI + χ − 1 on the interval
j k
left to Iχ , and takes an identical value 2αI + χ + 1 on the interval right to Iχ .
j k
Hence, as χ changes, it is true that v(α, δ) does take the value of 2αI + χ on every
interval Iχ . This ends the proof of the theorem 5.
By the theorem 5, the relationship between the function v(α, δ) and the variable δ
can be illustrated by figure 1. Through the figure, it gets to know that the length
of the interval Iχ is 2I . By expressing the interval Iχ in its equivalent form
Iχ = (−2I−1 + χ × 2I , 2I−1 + χ × 2I ]
it shows that the function v(α, δ) changes with the interval I0 = (−2I−1 , 2I−1 ] to
be a basic unit. We call the interval I0 a principal interval since the property of
the function v(α, δ) on Iχ can be translated from the property on it.

Figure 1. Relationship between the variable δ and the function v(α, δ).

Finally, we state and prove the following

Theorem 6. For arbitrary integers α ≥ 0 and δ , it holds
 141

  j α k δ − 1 1
α+δ
= z(α) + z(α) + (17)
2z(α) 2 2 2
Proof. Keep using the symbol I = z(α) and suppose δ ∈ Iχ without loss of
generality; then by the theorem 5, it yields
1 I 1 I
(χ − ) × 2 < δ ≤ (χ + ) × 2 (18)
2 2
Performing a simple transformation on (18) leads to
δ 1 δ 1
I
− ≤χ< I +
2 2 2 2
That is
δ 1 δ 1
− ≤χ< I − +1
2I 2 2 2
Referring to definition of the ceiling function yields
 
δ 1
χ= I −
2 2
By the properties (P6) and (P3) of the lemma 1, it follows
         
δ 1 δ − 2I−1 δ − 2I−1 − 1 δ−1 1 δ−1 1
− = = + 1 = − + 1 = +
2I 2 2I 2I 2I 2 2I 2
δ  l I−1
m j I−1 I
k
2Ij
− 12 = δ−22I = δ−2 2I+2 −1
I−1
k  
= δ−2 2I −1 + 1 = δ−1 2I
+ 12
Hence the theorem 6 holds.

4. Computer Test
Computations related to this paper can be tested on personal computers. The
C-language program is as follows.
\#include"math.h"

int GetI(int a) /* Find the smallest I that fits $0\leq a mod 2^i < 2^{(i-1)}$ */

\{
int i,I;

int \_2i\_1, \_2i;

for(i=1;;i++)
\{

\_2i\_1 = 1$\ll$(i-1);\quad /* Compute $2^{(i-1)}$*/

\_2i=\_2i\_1$\ll${1}; \quad /* Compute $2^i$ */

if(a\%\_2i$<$\_2i\_1)
\{
142

I=i; \quad\quad\quad \quad/* I is obtained */

break; \}

\}

return I;
\}

void Test(double a, double delta, int I) /*Perform Test*/

\{
int U,X,Y;

int \_2i;

\_2i=1$\ll$I; \quad /* COmpute $2^I$*/

U=(int)floor((a+delta)/\_2i);

X=(int)floor(delta/\_2i);

Y=(int)floor((delta-1)/\_2i+0.5);

X+=Y;

if(U!=X) printf("Err:\%d \%d \%d \%d \%d $\backslash{n}$",a,delta,I,U,X);

\}

void main()

\{
int I,\_2i,i;

for(i=10000;i<20000;i++)

\{I=GetI(i);

for(int j=-i;j$<=$i;j++)Test(i,j,I);

\}

printf("Finished!");

getchar();
\}
\fi
 143

Acknowledgments. The research work is supported by Department of Guang-

dong Sci. & Tech. under project 2012B010600018, Foshan Bureau of Sci. & and
Tech. under projects 2011AA100021, 2010C012 and Chancheng Government under
projects 2011GY006, 2011B1023. The author sincerely present thanks to them all.

References
[1] R. L. Graham, D. E. Knuth and O. Patashnik.Integer Functions. Ch.3 in Concrete Math-
ematics: A Foundation for Computer Science, 2nd ed. Reading, MA: Addison-Wesley, pp.
67-101, 1994, ISBN 0-201-55802-5.
[2] Wang Xingbo.Functions Related to Binary Representation of Integers. Journal of Inequalities
and Special Functions, vol.2, issue 2,(2011), pp.8-12.
[3] D. Shanks, Solved and Unsolved Problems in Number Theory, 4th ed. New York: Chelsea,
1993.

Department of Mechatronic Engineering, Foshan University, Guang-

dong Province, PR China,528000
E-mail address: [email protected]