A Survey On Cauchy-Bunyakovsky-Schwarz Type Discrete Inequalities

Journal of Inequalities in Pure and
Applied Mathematics
http://jipam.vu.edu.au/
Volume 4, Issue 3, Article 63, 2003
A SURVEY ON CAUCHY-BUNYAKOVSKY-SCHWARZ TYPE DISCRETE
INEQUALITIES
S.S. DRAGOMIR
SCHOOL OF COMPUTER SCIENCE AND MATHEMATICS
VICTORIA UNIVERSITY, PO BOX 14428,
MCMC 8001, MELBOURNE,
VICTORIA, AUSTRALIA.
sever.dragomir@vu.edu.au
Received 22 January, 2003; accepted 14 May, 2003
Communicated by P.S. Bullen
ABSTRACT. The main purpose of this survey is to identify and highlight the discrete inequalities
that are connected with (CBS) inequality and provide renements and reverse results as well
as to study some functional properties of certain mappings that can be naturally associated with
this inequality such as superadditivity, supermultiplicity, the strong versions of these and the
corresponding monotonicity properties. Many companion, reverse and related results both for
real and complex numbers are also presented.
Key words and phrases: Cauchy-Bunyakovsky-Schwarz inequality, Discrete inequalities.
2000 Mathematics Subject Classication. 26D15, 26D10.
CONTENTS
1. Introduction 4
2. (CBS) Type Inequalities 5
2.1. (CBS) Inequality for Real Numbers 5
2.2. (CBS) Inequality for Complex Numbers 6
2.3. An Additive Generalisation 7
2.4. A Related Additive Inequality 8
2.5. A Parameter Additive Inequality 10
2.6. A Generalisation Provided by Youngs Inequality 11
2.7. Further Generalisations via Youngs Inequality 12
2.8. A Generalisation Involving JConvex Functions 16
2.9. A Functional Generalisation 18
2.10. A Generalisation for Power Series 19
2.11. A Generalisation of Callebauts Inequality 21
2.12. Wagners Inequality for Real Numbers 22
2.13. Wagners inequality for Complex Numbers 24
ISSN (electronic): 1443-5756
c 2003 Victoria University. All rights reserved.
010-03
2 S.S. DRAGOMIR
References 26
3. Renements of the (CBS) Inequality 28
3.1. A Renement in Terms of Moduli 28
3.2. A Renement for a Sequence Whose Norm is One 29
3.3. A Second Renement in Terms of Moduli 31
3.4. A Renement for a Sequence Less than the Weights 33
3.5. A Conditional Inequality Providing a Renement 35
3.6. A Renement for Non-Constant Sequences 37
3.7. De Bruijns Inequality 40
3.8. McLaughlins Inequality 41
3.9. A Renement due to Daykin-Eliezer-Carlitz 42
3.10. A Renement via Dunkl-Williams Inequality 43
3.11. Some Renements due to Alzer and Zheng 45
References 50
4. Functional Properties 51
4.1. A Monotonicity Property 51
4.2. A Superadditivity Property in Terms of Weights 52
4.3. The Superadditivity as an Index Set Mapping 54
4.4. Strong Superadditivity in Terms of Weights 55
4.5. Strong Superadditivity as an Index Set Mapping 57
4.6. Another Superadditivity Property 59
4.7. The Case of Index Set Mapping 61
4.8. Supermultiplicity in Terms of Weights 63
4.9. Supermultiplicity as an Index Set Mapping 67
References 71
5. Reverse Inequalities 72
5.1. The Cassels Inequality 72
5.2. The Plya-Szeg Inequality 73
5.3. The Greub-Rheinboldt Inequality 75
5.4. A Cassels Type Inequality for Complex Numbers 75
5.5. A Reverse Inequality for Real Numbers 77
5.6. A Reverse Inequality for Complex Numbers 79
5.7. Shisha-Mond Type Inequalities 82
5.8. Zagier Type Inequalities 83
5.9. A Reverse Inequality in Terms of the supNorm 85
5.10. A Reverse Inequality in Terms of the 1Norm 87
5.11. A Reverse Inequality in Terms of the pNorm 90
5.12. A Reverse Inequality Via an Andrica-Badea Result 92
5.13. A Renement of Cassels Inequality 94
5.14. Two Reverse Results Via Diaz-Metcalf Results 97
5.15. Some Reverse Results Via the

Cebyev Functional 99
5.16. Another Reverse Result via a Grss Type Result 105
References 107
6. Related Inequalities 109
6.1. Ostrowskis Inequality for Real Sequences 109
6.2. Ostrowskis Inequality for Complex Sequences 110
6.3. Another Ostrowskis Inequality 111
6.4. Fan and Todd Inequalities 113
6.5. Some Results for Asynchronous Sequences 114
6.6. An Inequality via AGH Mean Inequality 115
6.7. A Related Result via Jensens Inequality for Power Functions 116
6.8. Inequalities Derived from the Double Sums Case 117
J. Inequal. Pure and Appl. Math., 4(3) Art. 63, 2003 http://jipam.vu.edu.au/
A SURVEY ON CAUCHY-BUNYAKOVSKY-SCHWARZ TYPE DISCRETE INEQUALITIES 3
6.9. A Functional Generalisation for Double Sums 118
6.10. A (CBS) Type Result for Lipschitzian Functions 119
6.11. An Inequality via Jensens Discrete Inequality 121
6.12. An Inequality via Lah-Ribari c Inequality 122
6.13. An Inequality via Dragomir-Ionescu Inequality 123
6.14. An Inequality via a Renement of Jensens Inequality 124
6.15. Another Renement via Jensens Inequality 127
6.16. An Inequality via Slaters Result 129
6.17. An Inequality via an Andrica-Ra sa Result 131
6.18. An Inequality via Jensens Result for Double Sums 132
6.19. Some Inequalities for the

6.20. Other Inequalities for the

6.21. Bounds for the

References 140
Index 141
4 S.S. DRAGOMIR
1. INTRODUCTION
The Cauchy-Bunyakovsky-Schwarz inequality, or for short, the (CBS)inequality, plays an
important role in different branches of Modern Mathematics including Hilbert Spaces Theory,
Probability & Statistics, Classical Real and Complex Analysis, Numerical Analysis, Qualitative
Theory of Differential Equations and their applications.
The main purpose of this survey is to identify and highlight the discrete inequalities that
are connected with (CBS) inequality and provide renements and reverse results as well as
to study some functional properties of certain mappings that can be naturally associated with
this inequality such as superadditivity, supermultiplicity, the strong versions of these and the
corresponding monotonicity properties. Many companions and related results both for real and
complex numbers are also presented.
The rst section is devoted to a number of (CBS) type inequalities that provides not only
natural generalizations but also several extensions for different classes of analytic functions of
a real variable. A generalization of the Wagner inequality for complex numbers is obtained.
Several results discovered by the author in the late eighties and published in different journals
of lesser circulation are also surveyed.
The second section contains different renements of the (CBS) inequality including de
Bruijns inequality, McLaughlins inequality, the Daykin-Eliezer-Carlitz result in the version
presented by Mitrinovi c-Pe cari c and Fink as well as the renements of a particular version
obtained by Alzer and Zheng. A number of new results obtained by the author, which are
connected with the above ones, are also presented.
Section 4 is devoted to the study of functional properties of different mappings naturally
associated to the (CBS)inequality. Properties such as superadditivity, strong superadditivity,
monotonicity and supermultiplicity and the corresponding inequalities are mentioned.
In the next section, Section 5, reverse results for the (CBS) inequality are surveyed. The
results of Cassels, Plya-Szeg, Greub-Rheinbold, Shisha-Mond and Zagier are presented with
their original proofs. New results and versions for complex numbers are also obtained. Reverse
results in terms of pnorms of the forward difference recently discovered by the author and
some renements of Cassels and Plya-Szeg results obtained via Andrica-Badea inequality
are mentioned. Some new facts derived from Grss type inequalities are also pointed out.
Section 6 is devoted to various inequalities related to the (CBS) inequality. The two in-
equalities obtained by Ostrowski and Fan-Todd results are presented. New inequalities obtained
via Jensen type inequality for convex functions are derived, some inequalities for the

Ceby sev
functionals are pointed out. Versions for complex numbers that generalize Ostrowski results are
also emphasised.
It was one of the main aims of the survey to provide complete proofs for the results consid-
ered. We also note that in most cases only the original references are mentioned. Each section
concludes with a list of the references utilized and thus may be read independently.
Being self contained, the survey may be used by both postgraduate students and researchers
interested in Theory of Inequalities & Applications as well as by Mathematicians and other
Scientists dealing with numerical computations, bounds and estimates where the (CBS) in-
equality may be used as a powerful tool.
The author intends to continue this survey with another one devoted to the functional and
integral versions of the (CBS)inequality. The corresponding results holding in inner-product
and normed spaces will be considered as well.
2. (CBS) TYPE INEQUALITIES
2.1. (CBS) Inequality for Real Numbers. The following inequality is known in the lit-
erature as Cauchys or Cauchy-Schwarzs or Cauchy-Bunyakovsky-Schwarzs inequality. For
simplicity, we shall refer to it throughout this work as the (CBS) inequality.
Theorem 2.1. If a = (a
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
) are sequences of real numbers, then
(2.1)
_
n
k=1
a
k
b
k
_
2
k=1
a
2
k
n
k=1
b
2
k
with equality if and only if the sequences a and

b are proportional, i.e., there is a r R such
that a
k
= rb
k
for each k {1, . . . , n} .
Proof. (1) Consider the quadratic polynomial P : R R,
(2.2) P (t) =
n
k=1
(a
k
t b
k
)
2
.
It is obvious that
P (t) =
_
n
k=1
a
2
k
_
t
2
2
_
n
k=1
a
k
b
k
_
t +
n
k=1
b
2
k
for any t R.
Since P (t) 0 for any t R it follows that the discriminant of P is negative, i.e.,
0
1
4
=
_
n
k=1
a
k
b
k
_
2
k=1
a
2
k
n
k=1
b
2
k
and the inequality (2.1) is proved.
(2) If we use Lagranges identity
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
=
1
2
n
i,j=1
(a
i
b
j
a
j
b
i
)
2
(2.3)
=
1i<jn
(a
i
b
j
a
j
b
i
)
2
then (2.1) obviously holds.
The equality holds in (2.1) iff
(a
i
b
j
a
j
b
i
)
2
= 0
for any i, j {1, . . . , n} which is equivalent with the fact that a and

b are proportional.
Remark 2.2. The inequality (2.1) apparently was rstly mentioned in the work [2] of A.L.
Cauchy in 1821. The integral form was obtained in 1859 by V.Y. Bunyakovsky [1]. The cor-
responding version for inner-product spaces obtained by H.A. Schwartz is mainly known as
Schwarzs inequality. For a short history of this inequality see [3]. In what follows we use the
spelling adopted in the paper [3]. For other spellings of Bunyakovskys name, see MathSciNet.
6 S.S. DRAGOMIR
2.2. (CBS) Inequality for Complex Numbers. The following version of the (CBS) inequality
for complex numbers holds [4, p. 84].
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
) are sequences of complex numbers,
then
(2.4)
k=1
a
k
b
k
k=1
|a
k
|
2
n
k=1
|b
k
|
2
,
with equality if and only if there is a complex number c C such that a
k
= c
b
k
for any
k {1, . . . , n} .
Proof. (1) For any complex number C one has the equality
n
k=1
a
k

b
k
2
=
n
k=1
_
a
k

b
k
_ _
a
k

b
k
_
(2.5)
=
n
k=1
|a
k
|
2
+||
2
n
k=1
|b
k
|
2
2 Re
_
k=1
a
k
b
k
_
.
If in (2.5) we choose
0
C,
0
:=
n
k=1
a
k
b
k
n
k=1
|b
k
|
2
,

b = 0
then we get the identity
(2.6) 0
n
k=1
a
k

0
b
k
2
=
n
k=1
|a
k
|
2
n
k=1
a
k
b
k
|
2
n
k=1
|b
k
|
2
,
which proves (2.4).
By virtue of (2.6), we conclude that equality holds in (2.4) if and only if a
k
=
0
b
k
for any k {1, . . . , n} .
(2) Using Binet-Cauchys identity for complex numbers
n
i=1
x
i
y
i
n
i=1
z
i
t
i
i=1
x
i
t
i
n
i=1
z
i
y
i
(2.7)
=
1
2
n
i,j=1
(x
i
z
j
x
j
z
i
) (y
i
t
j
y
j
t
i
)
=
1i<jn
(x
i
z
j
x
j
z
i
) (y
i
t
j
y
j
t
i
)
for the choices x
i
= a
i
, z
i
= b
i
, y
i
= a
i
, t
i
=

b
i
, i = {1, . . . , n}, we get
n
i=1
|a
i
|
2
n
i=1
|b
i
|
2
i=1
a
i
b
i
2
=
1
2
n
i,j=1
| a
i
b
j
a
j
b
i
|
2
(2.8)
=
1i<jn
| a
i
b
j
a
j
b
i
|
2
.
Now the inequality (2.4) is a simple consequence of (2.8).
The case of equality is obvious by the identity (2.8) as well.
Remark 2.4. By the (CBS) inequality for real numbers and the generalised triangle inequal-
ity for complex numbers
n
i=1
|z
i
|
i=1
z
i
, z
i
C, i {1, . . . , n}
we also have

k=1
a
k
b
k
_
n
k=1
|a
k
b
k
|
_
2
k=1
|a
k
|
2
n
k=1
|b
k
|
2
.
Remark 2.5. The Lagrange identity for complex numbers stated in [4, p. 85] is wrong. It
should be corrected as in (2.8).
2.3. An Additive Generalisation. The following generalisation of the (CBS) inequality
was obtained in [5, p. 5].
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) , c = (c
1
, . . . , c
n
) and

d = (d
1
, . . . , d
n
)
are sequences of real numbers and p = (p
1
, . . . , p
n
) , q = (q
1
, . . . , q
n
) are nonnegative, then
(2.9)
n
i=1
p
i
a
2
i
n
i=1
q
i
b
2
i
+
n
i=1
p
i
c
2
i
n
i=1
q
i
d
2
i
2
n
i=1
p
i
a
i
c
i
n
i=1
q
i
b
i
d
i
.
If p and q are sequences of positive numbers, then the equality holds in (2.9) iff a
i
b
j
= c
i
d
j
for
any i, j {1, . . . , n} .
Proof. We will follow the proof from [5].
From the elementary inequality
(2.10) a
2
+b
2
2ab for any a, b R
with equality iff a = b, we have
(2.11) a
2
i
b
2
j
+c
2
i
d
2
j
2a
i
c
i
b
j
d
j
for any i, j {1, . . . , n} .
Multiplying (2.11) by p
i
q
j
0, i, j {1, . . . , n} and summing over i and j from 1 to n, we
deduce (2.9).
If p
i
, q
j
> 0 (i = 1, . . . , n) , then the equality holds in (2.9) iff a
i
b
j
= c
i
d
j
for any i, j
{1, . . . , n} .
Remark 2.7. The condition a
i
b
j
= c
i
d
j
for c
i
= 0, b
j
= 0 (i, j = 1, . . . , n) is equivalent with
a
i
c
i
=
d
j
b
j
(i, j = 1, . . . , n) , i.e., a, c and

b,
d are proportional with the same constant k.

Remark 2.8. If in (2.9) we choose p
i
= q
i
= 1 (i = 1, . . . , n) , c
i
= b
i
, and d
i
= a
i
(i = 1, . . . , n) , then we recapture the (CBS) inequality.
The following corollary holds [5, p. 6].
Corollary 2.9. If a,

b, c and

d are nonnegative, then
(2.12)
1
2
_
n
i=1
a
3
i
c
i
n
i=1
b
3
i
d
i
+
n
i=1
c
3
i
a
i
n
i=1
d
3
i
b
i
_
i=1
a
2
i
c
2
i
n
i=1
b
2
i
d
2
i
,
(2.13)
1
2
_
n
i=1
a
2
i
b
i
d
i
i=1
b
2
i
a
i
c
i
+
n
i=1
c
2
i
b
i
d
i
i=1
d
2
i
a
i
c
i
_
_
n
i=1
a
i
b
i
c
i
d
i
_
2
.
Another result is embodied in the following corollary [5, p. 6].
8 S.S. DRAGOMIR
Corollary 2.10. If a,

b, c and

d are sequences of positive and real numbers, then:
(2.14)
1
2
_
n
i=1
a
3
i
c
i
n
i=1
b
3
i
d
i
+
n
i=1
a
i
c
i
n
i=1
b
i
d
i
_
i=1
a
2
i
n
i=1
b
2
i
,
(2.15)
1
2
_
n
i=1
a
2
i
b
i
c
i
n
i=1
b
2
i
a
i
d
i
+
n
i=1
b
i
c
i
n
i=1
a
i
d
i
_
_
n
i=1
a
i
b
i
_
2
.
Finally, we also have [5, p. 6].
Corollary 2.11. If a, and

b are positive, then
1
2
_
_
n
i=1
a
3
i
b
i
n
i=1
b
3
i
a
i
_
n
i=1
a
i
b
i
_
2
_
_
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
0.
The following version for complex numbers also holds.
Theorem 2.12. Let a = (a
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) , c = (c
1
, . . . , c
n
) and

d = (d
1
, . . . , d
n
)
be sequences of complex numbers and p = (p
1
, . . . , p
n
) , q = (q
1
, . . . , q
n
) are nonnegative.
Then one has the inequality
(2.16)
n
i=1
p
i
|a
i
|
2
n
i=1
q
i
|b
i
|
2
+
n
i=1
p
i
|c
i
|
2
n
i=1
q
i
|d
i
|
2
2 Re
_
n
i=1
p
i
a
i
c
i
n
i=1
q
i
b
i

d
i
_
.
The case of equality for p, q positive holds iff a
i
b
j
= c
i
d
j
for any i, j {1, . . . , n} .
Proof. From the elementary inequality for complex numbers
|a|
2
+|b|
2
2 Re
_
a
, a, b C,
with equality iff a = b, we have
(2.17) |a
i
|
2
|b
j
|
2
+|c
i
|
2
|d
j
|
2
2 Re
_
a
i
c
i
b
j

dj
for any i, j {1, . . . , n} . Multiplying (2.17) by p

i
q
j
0 and summing over i and j from 1 to
n, we deduce (2.16).
The case of equality is obvious and we omit the details.
Remark 2.13. Similar particular cases may be stated but we omit the details.
2.4. A Related Additive Inequality. The following inequality was obtained in [5, Theorem
1.1].
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) are sequences of real numbers and
c = (c
1
, . . . , c
n
) ,

d = (d
1
, . . . , d
n
) are nonnegative, then
(2.18)
n
i=1
d
i
n
i=1
c
i
a
2
i
+
n
i=1
c
i
n
i=1
d
i
b
2
i
2
n
i=1
c
i
a
i
n
i=1
d
i
b
i
.
If c
i
and d
i
(i = 1, . . . , n) are positive, then equality holds in (2.18) iff a =

b =

k where
k = (k, k, . . . , k) is a constant sequence.

From the elementary inequality
(2.19) a
2
+b
2
2ab for any a, b R
with equality iff a = b; we have
(2.20) a
2
i
+b
2
j
2a
i
b
j
for any i, j {1, . . . , n} .
Multiplying (2.20) by c
i
d
j
0, i, j {1, . . . , n} and summing over i from 1 to n and over j
from 1 to n, we deduce (2.18).
If c
i
, d
j
> 0 (i = 1, . . . , n) , then the equality holds in (2.18) iff a
i
= b
j
for any i, j
{1, . . . , n} which is equivalent with the fact that a
i
= b
i
= k for any i {1, . . . , n} .
The following corollary holds [5, p. 4].
Corollary 2.15. If a and

b are nonnegative sequences, then
(2.21)
1
2
_
n
i=1
a
3
i
n
i=1
b
i
+
n
i=1
a
i
n
i=1
b
3
i
_
i=1
a
2
i
n
i=1
b
2
i
;
(2.22)
1
2
_
n
i=1
a
i
n
i=1
a
2
i
b
i
+
n
i=1
b
i
n
i=1
b
2
i
a
i
_
_
n
i=1
a
i
b
i
_
2
.
Another corollary that may be obtained is [5, p. 4 5].

b are sequences of positive real numbers, then
(2.23)
n
i=1
a
2
i
+b
2
i
2a
i
b
i
n
i=1
1
a
i
n
i=1
1
b
i
n
i=1
1
a
i
b
i
,
(2.24)
n
i=1
a
i
n
i=1
1
b
i
+
n
i=1
1
a
i
n
i=1
b
i
2n
2
,
and
(2.25) n
n
i=1
a
2
i
+b
2
i
2a
2
i
b
2
i
i=1
1
a
i
n
i=1
1
b
i
.
The following version for complex numbers also holds.
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) are sequences of complex numbers, then
for p = (p
1
, . . . , p
n
) and q = (q
1
, . . . , q
n
) two sequences of nonnegative real numbers, one has
the inequality
(2.26)
n
i=1
q
i
n
i=1
p
i
|a
i
|
2
+
n
i=1
p
i
n
i=1
q
i
|b
i
|
2
2 Re
_
n
i=1
p
i
a
i
n
i=1
q
i
b
i
_
.
For p, q positive sequences, the equality holds in (2.26) iff a =

b =

k = (k, . . . , k) .
The proof goes in a similar way with the one in Theorem 2.14 on making use of the following
elementary inequality holding for complex numbers
(2.27) |a|
2
+|b|
2
2 Re
_
a
, a, b C;
with equality iff a = b.
10 S.S. DRAGOMIR
2.5. AParameter Additive Inequality. The following inequality was obtained in [5, Theorem
4.1].
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) be sequences of real numbers and
c = (c
1
, . . . , c
n
) ,

d = (d
1
, . . . , d
n
) be nonnegative. If , > 0 and R such that
2
,
then
(2.28)
n
i=1
d
i
n
i=1
a
2
i
c
i
+
n
i=1
c
i
n
i=1
b
2
i
d
i
2
n
i=1
c
i
a
i
n
i=1
d
i
b
i
.
Since , > 0 and
2
, it follows that for any x, y R one has
(2.29) x
2
+y
2
2xy.
Choosing in (2.29) x = a
i
, y = b
j
(i, j = 1, . . . , n) , we get
(2.30) a
2
i
+b
2
j
2a
i
b
j
for any i, j {1, . . . , n} .
If we multiply (2.30) by c
i
d
j
0 and sum over i and j from 1 to n, we deduce the desired
inequality (2.28).
The following corollary holds.

b are nonnegative sequences and , , are as in Theorem 2.18, then
(2.31)
n
i=1
b
i
n
i=1
a
3
i
+
n
i=1
a
i
n
i=1
b
3
i
2
n
i=1
a
2
i
n
i=1
b
2
i
,
(2.32)
n
i=1
a
i
n
i=1
a
2
i
b
i
+
n
i=1
b
i
n
i=1
b
2
i
a
i
2
_
n
i=1
a
i
b
i
_
2
.
The following particular case is important [5, p. 8].
Theorem 2.20. Let a,

b be sequences of real numbers. If p is a sequence of nonnegative real
numbers with
n
i=1
p
i
> 0, then:
(2.33)
n
i=1
p
i
a
2
i
n
i=1
p
i
b
2
i

n
i=1
p
i
a
i
b
i
n
i=1
p
i
a
i
n
i=1
p
i
b
i
n
i=1
p
i
.
In particular,
(2.34)
n
i=1
a
2
i
n
i=1
b
2
i

1
n
n
i=1
a
i
b
i
n
i=1
a
i
n
i=1
b
i
.
Proof. We will follow the proof from [5, p. 8].
If we choose in Theorem 2.18, c
i
= d
i
= p
i
(i = 1, . . . , n) and =
n
i=1
p
i
b
2
i
, =
n
i=1
p
i
a
2
i
, =
n
i=1
p
i
a
i
b
i
, we observe, by the (CBS) inequality with the weights p
i
(i = 1, . . . , n) one has
2
, and then by (2.28) we deduce (2.33).
Remark 2.21. If we assume that a and

b are asynchronous, i.e.,
(a
i
a
j
) (b
i
b
j
) 0 for any i, j {1, . . . , n} ,
then by

Cebyevs inequality
(2.35)
n
i=1
p
i
a
i
n
i=1
p
i
b
i

n
i=1
p
i
n
i=1
p
i
a
i
b
i
respectively
(2.36)
n
i=1
a
i
n
i=1
b
i
n
n
i=1
a
i
b
i
,
we have the following renements of the (CBS) inequality
n
i=1
p
i
a
2
i
n
i=1
p
i
b
2
i

n
i=1
p
i
a
i
b
i
n
i=1
p
i
a
i
n
i=1
p
i
b
i
n
i=1
p
i
(2.37)
_
n
i=1
p
i
a
i
b
i
_
2
provided
n
i=1
p
i
a
i
b
i
0, respectively
(2.38)
n
i=1
a
2
i
n
i=1
b
2
i

1
n
n
i=1
a
i
b
i
n
i=1
a
i
n
i=1
b
i

_
n
i=1
a
i
b
i
_
2
provided
n
i=1
a
i
b
i
0.
2.6. A Generalisation Provided by Youngs Inequality. The following result was obtained
in [5, Theorem 5.1].
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) , p = (p
1
, . . . , p
n
) and q = (q
1
, . . . , q
n
)
be sequences of nonnegative real numbers and , > 1 with
1
+
1
= 1. Then one has the

inequality
(2.39)
n
i=1
q
i
n
i=1
p
i
b
i
+
n
i=1
p
i
n
i=1
q
i
a
i

n
i=1
p
i
b
i
n
i=1
q
i
a
i
.
If p and q are sequences of positive real numbers, then the equality holds in (2.39) iff there
exists a constant k 0 such that a
i
= b
i
= k for each i {1, . . . , n} .
Proof. It is, by the Arithmetic-Geometric inequality [6, p. 15], well known that
(2.40)
1
x +
1
y x
1
y
1
for x, y 0,
1
+
1
= 1, , > 1
with equality iff x = y.
Applying (2.40) for x = a
i
, y = b
j
(i, j = 1, . . . , n) we have
(2.41) b
j
+a
i
a
i
b
j
for any i, j {1, . . . , n}
with equality iff a
i
= b
j
for any i, j {1, . . . , n} .
If we multiply (2.41) by q
i
p
j
0 (i, j {1, . . . , n}) and sum over i and j from 1 to n we
deduce (2.39).
The case of equality is obvious by the above considerations.
The following corollary is a natural consequence of the above theorem.
Corollary 2.23. Let a,

b, and be as in Theorem 2.22. Then
(2.42)
1
i=1
b
i
n
i=1
a
+1
i
+
1
i=1
a
i
n
i=1
b
+1
i

n
i=1
a
2
i
n
i=1
b
2
i
;
12 S.S. DRAGOMIR
(2.43)
1
i=1
a
i
n
i=1
b
i
a
i
+
1
i=1
b
i
n
i=1
a
i
b
i

_
n
i=1
a
i
b
i
_
2
.
The following result which provides a generalisation of the (CBS) inequality may be ob-
tained by Theorem 2.22 as well [5, Theorem 5.2].
Theorem 2.24. Let x and y be sequences of positive real numbers. If , are as above, then
(2.44)
_
1
i=1
x
i
y
2
i
+
1
i=1
x
i
y
2
i
_
i=1
y
2
i

_
n
i=1
x
i
y
i
_
2
.
The equality holds iff x and y are proportional.
Proof. Follows by Theorem 2.22 on choosing p
i
= q
i
= y
2
i
, a
i
=
x
i
y
i
, b
i
=
x
i
y
i
, i {1, . . . , n} .
Remark 2.25. For = = 2, we recapture the (CBS) inequality.

Remark 2.26. For a
i
= |z
i
| , b
i
= |w
i
| , with z
i
, w
i
C; i = 1, . . . , n, we may obtain similar
inequalities for complex numbers. We omit the details.
2.7. Further Generalisations via Youngs Inequality. The following inequality is known in
the literature as Youngs inequality
(2.45) px
q
+qy
p
pqxy, x, y 0 and
1
p
+
1
q
= 1, p > 1
with equality iff x
q
= y
p
.
The following result generalising the (CBS) inequality was obtained in [7, Theorem 2.1]
(see also [8, Theorem 1]).
Theorem 2.27. Let x = (x
1
, . . . , x
n
) , y = (y
1
, . . . , y
n
) be sequences of complex numbers and
p = (p
1
, . . . , p
n
) , q = (q
1
, . . . , q
n
) be two sequences of nonnegative real numbers. If p > 1,
1
p
+
1
q
= 1, then
(2.46)
1
p
n
k=1
p
k
|x
k
|
p
n
k=1
q
k
|y
k
|
p
+
1
q
n
k=1
q
k
|x
k
|
q
n
k=1
p
k
|y
k
|
q
k=1
p
k
|x
k
y
k
|
n
k=1
q
k
|x
k
y
k
| .
Proof. We shall follow the proof in [7].
Choosing x = |x
j
| |y
i
| , y = |x
i
| |y
j
| , i, j {1, . . . , n} , we get from (2.45)
(2.47) q |x
i
|
p
|y
j
|
p
+p |x
j
|
q
|y
i
|
q
pq |x
i
y
i
| |x
j
y
j
|
for any i, j {1, . . . , n} .
Multiplying with p
i
q
j
0 and summing over i and j from 1 to n, we deduce the desired
result (2.46).
The following corollary is a natural consequence of the above theorem [7, Corollary 2.2] (see
also [8, p. 105]).
Corollary 2.28. If x and y are as in Theorem 2.27 and m = (m
1
, . . . , m
n
) is a sequence of
nonnegative real numbers, then
(2.48)
1
p
n
k=1
m
k
|x
k
|
p
n
k=1
m
k
|y
k
|
p
+
1
q
n
k=1
m
k
|x
k
|
q
n
k=1
m
k
|y
k
|
q
_
n
k=1
m
k
|x
k
y
k
|
_
2
,
where p > 1,
1
p
+
1
q
= 1.
Remark 2.29. If in (2.48) we assume that m
k
= 1, k {1, . . . , n} , then we obtain [7, p. 7]
(see also [8, p. 105])
(2.49)
1
p
n
k=1
|x
k
|
p
n
k=1
|y
k
|
p
+
1
q
n
k=1
|x
k
|
q
n
k=1
|y
k
|
q
_
n
k=1
|x
k
y
k
|
_
2
,
which, in the particular case p = q = 2 will provide the (CBS) inequality.
The second generalisation of the (CBS) inequality via Youngs inequality is incorporated
in the following theorem [7, Theorem 2.4] (see also [8, Theorem 2]).
Theorem 2.30. Let x, y, p, q and p, q be as in Theorem 2.27. Then one has the inequality
(2.50)
1
p
n
k=1
p
k
|x
k
|
p
n
k=1
q
k
|y
k
|
q
+
1
q
n
k=1
q
k
|x
k
|
q
n
k=1
p
k
|y
k
|
p
k=1
p
k
|x
k
| |y
k
|
p1
n
k=1
q
k
|x
k
| |y
k
|
q1
.
Choosing in (2.45), x =
|x
j
|
|y
j
|
, y =
|x
i
|
|y
i
|
, we get
(2.51) p
_
|x
j
|
|y
j
|
_
q
+q
_
|x
i
|
|y
i
|
_
p
pq
|x
i
| |x
j
|
|y
i
| |y
j
|
for any y
i
= 0, i, j {1, . . . , n} .
It is easy to see that (2.51) is equivalent to
(2.52) q |x
i
|
p
|y
j
|
q
+p |y
i
|
p
|x
j
|
q
pq |x
i
| |y
i
|
p1
|x
j
| |y
j
|
q1
for any i, j {1, . . . , n} .
i
q
j
0 (i, j {1, . . . , n}) and summing over i and j from 1 to n, we
deduce the desired inequlality (2.50).
The following corollary holds [7, Corollary 2.5] (see also [8, p. 106]).
Corollary 2.31. Let x, y, m and p, q be as in Corollary 2.28. Then
(2.53)
1
p
n
k=1
m
k
|x
k
|
p
n
k=1
m
k
|y
k
|
q
+
1
q
n
k=1
m
k
|x
k
|
q
n
k=1
m
k
|y
k
|
p
k=1
m
k
|x
k
| |y
k
|
p1
n
k=1
m
k
|x
k
| |y
k
|
q1
.
k
= 1, k {1, . . . , n} , then we obtain [7, p. 8]
(see also [8, p. 106])
(2.54)
1
p
n
k=1
|x
k
|
p
n
k=1
|y
k
|
q
+
1
q
n
k=1
|x
k
|
q
n
k=1
|y
k
|
p
k=1
|x
k
| |y
k
|
p1
n
k=1
|x
k
| |y
k
|
q1
,
The third result is embodied in the following theorem [7, Theorem 2.7] (see also [8, Theorem
3]).
14 S.S. DRAGOMIR
(2.55)
1
p
n
k=1
p
k
|x
k
|
p
n
k=1
q
k
|y
k
|
q
+
1
q
n
k=1
q
k
|x
k
|
p
n
k=1
p
k
|y
k
|
q
k=1
p
k
|x
k
y
k
|
n
k=1
p
k
|x
k
|
p1
|y
k
|
q1
.
If we choose x =
|y
i
|
|y
j
|
and y =
|x
i
|
|x
j
|
in (2.45) we get
p
_
|y
i
|
|y
j
|
_
q
+q
_
|x
i
|
|x
j
|
_
p
pq
|x
i
| |y
i
|
|x
j
| |y
j
|
,
for any x
i
, y
j
= 0, i, j {1, . . . , n} , giving
(2.56) q |x
i
|
p
|y
j
|
q
+p |y
i
|
q
|x
j
|
p
pq |x
i
y
i
| |x
j
|
p1
|y
j
|
q1
for any i, j {1, . . . , n} .
i
q
j
deduce the desired inequality (2.55).
The following corollary is a natural consequence of the above theorem [8, p. 106].
Corollary 2.34. Let x, y, m and p, q be as in Corollary 2.28. Then one has the inequality:
(2.57)
n
k=1
m
k
|x
k
|
p
n
k=1
m
k
|y
k
|
q
k=1
m
k
|x
k
y
k
|
n
k=1
m
k
|x
k
|
p1
|y
k
|
q1
.
k
= 1, k = {1, . . . , n} , then we obtain [7, p. 8]
(see also [8, p. 10])
(2.58)
n
k=1
|x
k
|
p
n
k=1
|y
k
|
q
k=1
|x
k
y
k
|
n
k=1
|x
k
|
p1
|y
k
|
q1
,
The fourth generalisation of the (CBS) inequality is embodied in the following theorem
[7, Theorem 2.9] (see also [8, Theorem 4]).
(2.59)
1
q
n
k=1
p
k
|x
k
|
2
n
k=1
q
k
|y
k
|
q
+
1
p
n
k=1
p
k
|y
k
|
2
n
k=1
q
k
|x
k
|
p
k=1
q
k
|x
k
y
k
|
n
k=1
p
k
|x
k
|
2
q
|y
k
|
2
p
.
Choosing in (2.45), x = |x
i
|
2
q
|y
j
| , y = |x
j
| |y
i
|
2
p
, we get
(2.60) p |x
i
|
2
|y
j
|
q
+q |x
j
|
p
|y
i
|
2
pq |x
i
|
2
q
|y
i
|
2
p
|x
j
y
j
|
for any i, j {1, . . . , n} .
Multiply (2.60) by p
i
q
j
(2.61)
1
q
n
k=1
m
k
|x
k
|
2
n
k=1
m
k
|y
k
|
q
+
1
p
n
k=1
m
k
|y
k
|
2
n
k=1
m
k
|x
k
|
p
k=1
m
k
|x
k
y
k
|
n
k=1
m
k
|x
k
|
2
q
|y
k
|
2
p
.
Remark 2.38. If in (2.61) we take m
k
= 1, k {1, . . . , n} , then we get
(2.62)
1
q
n
k=1
|x
k
|
2
n
k=1
|y
k
|
q
+
1
p
n
k=1
|y
k
|
2
n
k=1
|x
k
|
p
k=1
|x
k
y
k
|
n
k=1
|x
k
|
2
q
|y
k
|
2
p
,
The fth result generalising the (CBS) inequality is embodied in the following theorem [7,
Theorem 2.12] (see also [8, Theorem 5]).
(2.63)
1
p
n
k=1
p
k
|x
k
|
2
n
k=1
q
k
|y
k
|
q
+
1
q
n
k=1
p
k
|y
k
|
2
n
k=1
q
k
|x
k
|
p
k=1
p
k
|x
k
|
2
p
|y
k
|
2
q
n
k=1
q
k
|x
k
|
p1
|y
k
|
q1
.
Proof. We will follow the proof in [7].
Choosing in (2.45), x =
|y
i
|
2
q
|y
j
|
, y =
|x
i
|
2
p
|x
j
|
, y
i
, x
j
= 0, i, j {1, . . . , n} , we may write
p
_
|y
i
|
2
q
|y
j
|
_
q
+q
_
|x
i
|
2
p
|x
j
|
_
p
pq
|y
i
|
2
q
|x
i
|
2
p
|x
j
| |y
j
|
,
from where results
(2.64) p |y
i
|
2
|x
j
|
p
+q |x
i
|
2
|y
j
|
q
pq |x
i
|
2
p
|y
i
|
2
q
|x
j
|
p1
|y
j
|
q1
for any i, j {1, . . . , n} .
i
q
j
0 (i, j {1, . . . , n}) and summing over i and j from 1 to n,
we deduce the desired inequality (2.63).
(2.65)
1
p
n
k=1
m
k
|x
k
|
2
n
k=1
m
k
|y
k
|
q
+
1
q
n
k=1
m
k
|y
k
|
2
n
k=1
m
k
|x
k
|
p
k=1
m
k
|x
k
|
2
p
|y
k
|
2
q
n
k=1
m
k
|x
k
|
p1
|y
k
|
q1
.
16 S.S. DRAGOMIR
Remark 2.41. If in (2.46) we choose m
k
= 1, k {1, . . . , n} , then we get [7, p. 10] (see also
[8, p. 108])
(2.66)
1
p
n
k=1
|x
k
|
2
n
k=1
|y
k
|
q
+
1
q
n
k=1
|y
k
|
2
n
k=1
|x
k
|
p
k=1
|x
k
|
2
p
|y
k
|
2
q
n
k=1
|x
k
|
p1
|y
k
|
q1
,
which in the particular case p = q = 2 will provide the (CBS) inequality.
Finally, the following result generalising the (CBS) inequality holds [7, Theorem 2.15]
(see also [8, Theorem 6]).
Theorem 2.42. Let x, y, p, q and p, q be as in Theorem 2.27. Then one has the inequality:
(2.67)
1
p
n
k=1
p
k
|x
k
|
2
n
k=1
q
k
|y
k
|
p
+
1
q
n
k=1
q
k
|y
k
|
2
n
k=1
p
k
|x
k
|
q
k=1
p
k
|x
k
|
2
p
|y
k
|
n
k=1
q
k
|x
k
|
2
q
|y
k
| .
From (2.45) one has the inequality
(2.68) q
_
|x
i
|
2
p
|y
j
|
_
p
+p
_
|x
j
|
2
q
|y
i
|
_
q
pq |x
i
|
2
p
|y
i
| |x
j
|
2
q
|y
j
|
for any i, j {1, . . . , n} .
i
q
j
The following corollary also holds [7, Corollary 2.16] (see also [8, p. 108]).
Corollary 2.43. With the assumptions in Corollary 2.28, one has the inequality
(2.69)
n
k=1
m
k
|x
k
|
2
n
k=1
m
k
_
1
p
|y
k
|
p
+
1
q
|y
k
|
q
_
k=1
m
k
|x
k
|
2
p
|y
k
|
n
k=1
m
k
|x
k
|
2
q
|y
k
| .
Remark 2.44. If in (2.69) we choose m
k
= 1 (k {1, . . . , n}) , then we get
(2.70)
n
k=1
|x
k
|
2
n
k=1
_
1
p
|y
k
|
p
+
1
q
|y
k
|
q
_
k=1
|x
k
|
2
p
|y
k
|
n
k=1
|x
k
|
2
q
|y
k
| ,
which, in the particular case p = q = 2, provides the (CBS) inequality.
2.8. A Generalisation Involving JConvex Functions. For a > 1, we denote by exp
a
the
function
(2.71) exp
a
: R (0, ) , exp
a
(x) = a
x
.
Denition 2.1. A function f : I R R is said to be Jconvex on an interval I if
(2.72) f
_
x +y
2
_
f (x) +f (y)
2
for any x, y I.
It is obvious that any convex function on I is a J convex function on I, but the converse does
not generally hold.
The following lemma holds (see [7, Lemma 4.3]).
Lemma 2.45. Let f : I R R be a Jconvex function on I, a > 1 and x, y R\ {0} with
log
a
x
2
, log
a
y
2
I. Then log
a
|xy| I and
(2.73) {exp
b
[f (log
a
|xy|)]}
2
exp
b
_
f
_
log
a
x
2
_
exp
b
_
f
_
log
a
y
2
_
for any b > 1.
Proof. I, being an interval, is a convex set in R and thus
log
a
|xy| =
1
2
_
log
a
x
2
+ log
a
y
2
I.
Since f is Jconvex, one has
f (log
a
|xy|) = f
_
1
2
_
log
a
x
2
+ log
a
y
2
_
_
(2.74)
f (log
a
x
2
) +f (log
a
y
2
)
2
.
Taking the exp
b
in both parts, we deduce
exp
b
[f (log
a
|xy|)] exp
b
_
f (log
a
x
2
) +f (log
a
y
2
)
2
_
=
_
exp
b
_
f
_
log
a
x
2
_
exp
b
_
f
_
log
a
y
2
__1
2
,
which is equivalent to (2.73).
The following generalisation of the (CBS) inequality in terms of a J convex function
holds [7, Theorem 4.4].
Theorem 2.46. Let f : I R R be a Jconvex function on I, a, b > 1 and a =
(a
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) sequences of nonzero real numbers. If log
a
a
2
k
, log
a
b
2
k
I
for all k {1, . . . , n} , then one has the inequality:
(2.75)
_
n
k=1
exp
b
[f (log
a
|a
k
b
k
|)]
_
2
k=1
exp
b
_
f
_
log
a
a
2
k
_
n
k=1
exp
b
_
f
_
log
a
b
2
k
_
.
Proof. Using Lemma 2.45 and the (CBS) inequality one has
n
k=1
exp
b
[f (log
a
|a
k
b
k
|)]
k=1
_
exp
b
_
f
_
log
a
a
2
k
_
exp
b
_
f
_
log
a
b
2
k
_1
2
_
n
k=1
_
_
exp
b
_
f
_
log
a
a
2
k
_1
2
_
2
n
k=1
_
_
exp
b
_
f
_
log
a
b
2
k
_1
2
_
2
_1
2
which is clearly equivalent to (2.75).
Remark 2.47. If in (2.75) we choose a = b > 1 and f (x) = x, x R, then we recapture the
(CBS) inequality.
18 S.S. DRAGOMIR
2.9. A Functional Generalisation. The following result was proved in [10, Theorem 2].
Theorem 2.48. Let A be a subset of real numbers R, f : A R and a = (a
1
, . . . , a
n
) ,
b = (b
1
, . . . , b
n
) sequences of real numbers with the properties that
(i) a
i
b
i
, a
2
i
, b
2
i
A for any i {1, . . . , n} ,
(ii) f (a
2
i
) , f (b
2
i
) 0 for any i {1, . . . , n} ,
(iii) f
2
(a
i
b
i
) f (a
2
i
) f (b
2
i
) for any i {1, . . . , n} .
Then one has the inequality:
(2.76)
_
n
i=1
f (a
i
b
i
)
_
2
i=1
f
_
a
2
i
_
n
i=1
f
_
b
2
i
_
.
Proof. We give here a simpler proof than that found in [10].
We have
i=1
f (a
i
b
i
)
i=1
|f (a
i
b
i
)|
i=1
_
f
_
a
2
i
_1
2
_
f
_
b
2
i
_1
2
_
n
i=1
_
_
f
_
a
2
i
_1
2
_
2
n
i=1
_
_
f
_
b
2
i
_1
2
_
2
_1
2
(by the (CBS)-inequality)
=
_
n
i=1
f
_
a
2
i
_
n
i=1
f
_
b
2
i
_
_1
2
Remark 2.49. It is obvious that for A = Rand f (x) = x, we recapture the (CBS) inequality.
Assume that : N N is Eulers indicator. In 1940, T. Popoviciu [11] proved the following
inequality for
(2.77) [(ab)]
2

_
a
2
_
_
b
2
_
for any natural number a, b;
with equality iff a and b have the same prime factors.
A simple proof of this fact may be done by using the representation
(n) = n
_
1
1
p
1
_

_
1
1
p
k
_
,
where n = p
1
1
p
2
2
p
k
k
[9, p. 109].
The following generalisation of Popovicius result holds [10, Theorem 1].
Theorem 2.50. Let a
i
, b
i
N (i = 1, . . . , n) . Then one has the inequality
(2.78)
_
n
i=1
(a
i
b
i
)
_
2
i=1
_
a
2
i
_
n
i=1
_
b
2
i
_
.
Proof. Follows by Theorem 2.48 on taking into account that, by (2.77),
[(a
i
b
i
)]
2

_
a
2
i
_
_
b
2
i
_
for any i {1, . . . , n} .
Further, let us denote by s (n) the sum of all relatively prime numbers with n and less than
n. Then the following result also holds [10, Theorem 1].
Theorem 2.51. Let a
i
, b
i
N (i = 1, . . . , n) . Then one has the inequality
(2.79)
_
n
i=1
s (a
i
b
i
)
_
2
i=1
s
_
a
2
i
_
n
i=1
s
_
b
2
i
_
.
Proof. It is known (see for example [9, p. 109]) that for any n N one has
(2.80) s (n) =
1
2
n(n) .
Thus
(2.81) [s (a
i
b
i
)]
2
=
1
4
a
2
i
b
2
i
2
(a
i
b
i
)
1
4
a
2
i
b
2
i
_
a
2
i
_
_
b
2
i
_
= s
_
a
2
i
_
s
_
b
2
i
_
for each i {1, . . . , n} .
Using Theorem 2.48 we then deduce the desired inequality (2.79).
The following corollaries of Theorem 2.48 are also natural to be considered [10, p. 126].
Corollary 2.52. Let a
i
, b
i
R (i = 1, . . . , n) and a > 1. Denote exp
a
x = a
x
, x R. Then
one has the inequality
(2.82)
_
n
i=1
exp
a
(a
i
b
i
)
_
2
i=1
exp
a
_
a
2
i
_
n
i=1
exp
a
_
b
2
i
_
.
Corollary 2.53. Let a
i
, b
i
(1, 1) (i = 1, . . . , n) and m > 0. Then one has the inequality:
(2.83)
_
n
i=1
1
(1 a
i
b
i
)
m
_
2
i=1
1
(1 a
2
i
)
m
n
i=1
1
(1 b
2
i
)
m
.
2.10. A Generalisation for Power Series. The following result holds [12, Remark 2].
Theorem 2.54. Let F : (r, r) R, F (x) =
k=0
k
x
k
with
k
0, k N. If a =
(a
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) are sequences of real numbers such that
(2.84) a
i
b
i
, a
2
i
, b
2
i
(r, r) for any i {1, . . . , n} ,
then one has the inequality:
(2.85)
n
i=1
F
_
a
2
i
_
n
i=1
F
_
b
2
i
_
_
n
i=1
F (a
i
b
i
)
_
2
.
Proof. Firstly, let us observe that if x, y R such that xy, x
2
, y
2
(r, r) , then one has the
inequality
(2.86) [F (xy)]
2
F
_
x
2
_
F
_
y
2
_
.
Indeed, by the (CBS) inequality, we have
(2.87)
_
n
k=0
k
x
k
y
k
_
2
k=0
k
x
2k
n
k=0
k
y
2k
, n 0.
Taking the limit as n in (2.87), we deduce (2.86).
20 S.S. DRAGOMIR
Using the (CBS) inequality and (2.86) we have
i=1
F (a
i
b
i
)
i=1
|F (a
i
b
i
)|
i=1
_
F
_
a
2
i
_1
2
_
F
_
b
2
i
_1
2
_
n
i=1
_
_
F
_
a
2
i
_1
2
_
2
n
i=1
_
_
F
_
b
2
i
_1
2
_
2
_1
2
=
_
n
i=1
F
_
a
2
i
_
n
i=1
F
_
b
2
i
_
_1
2
,
The following particular inequalities of (CBS) type hold [12, p. 164].
(1) If a,

b are sequences of real numbers, then one has the inequality
n
k=1
exp
_
a
2
k
_
n
k=1
exp
_
b
2
k
_
_
n
k=1
exp (a
k
b
k
)
_
2
; (2.88)
n
k=1
sinh
_
a
2
k
_
n
k=1
sinh
_
b
2
k
_
_
n
k=1
sinh (a
k
b
k
)
_
2
; (2.89)
n
k=1
cosh
_
a
2
k
_
n
k=1
cosh
_
b
2
k
_
_
n
k=1
cosh (a
k
b
k
)
_
2
. (2.90)
(2) If a,

b are such that a
i
, b
i
(1, 1) , i {1, . . . , n} , then one has the inequalities
n
k=1
tan
_
a
2
k
_
n
k=1
tan
_
b
2
k
_
_
n
k=1
tan (a
k
b
k
)
_
2
; (2.91)
n
k=1
arcsin
_
a
2
k
_
n
k=1
arcsin
_
b
2
k
_
_
n
k=1
arcsin (a
k
b
k
)
_
2
; (2.92)
(2.93) ln
_
n
k=1
_
1 +a
2
k
1 a
2
k
_
_
ln
_
n
k=1
_
1 +b
2
k
1 b
2
k
_
_
_
ln
_
n
k=1
_
1 +a
k
b
k
1 a
k
b
k
_
__
2
;
(2.94) ln
_
n
k=1
_
1
1 a
2
k
_
_
ln
_
n
k=1
_
1
1 b
2
k
_
_
_
ln
_
n
k=1
_
1
1 a
k
b
k
_
__
2
;
(2.95)
n
k=1
1
(1 a
2
k
)
m
n
k=1
1
(1 b
2
k
)
m

_
n
k=1
1
(1 a
k
b
k
)
m
_
2
, m > 0.
2.11. A Generalisation of Callebauts Inequality. The following result holds (see also [12,
Theorem 2] for a generalisation for positive linear functionals).
Theorem 2.55. Let F : (r, r) R, F (x) =
k=0
k
x
k
with
k
0, k N. If a =
(a
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) are sequences of nonnegative real numbers such that
(2.96) a
i
b
i
, a
i
b
2
i
, a
2
i
b
i
(0, r) for any i {1, . . . , n} ; [0, 2] ,
then one has the inequality
(2.97)
_
n
i=1
F (a
i
b
i
)
_
2
i=1
F
_
a
i
b
2
i
_
n
i=1
F
_
a
2
i
b
i
_
.
Proof. Firstly, we note that for any x, y > 0 such that xy, x
y
2
, x
2
y
(0, r) one has

(2.98) [F (xy)]
2
F
_
x
y
2
_
F
_
x
2
y
_
.
Indeed, using Callebauts inequality, i.e., we recall it [4]
(2.99)
_
m
i=1
i
x
i
y
i
_
2
i=1
i
x
i
y
2
i
m
i=1
i
x
2
i
y
i
,
we may write, for m 0, that
(2.100)
_
m
i=0
i
x
i
y
i
_
2
i=0
i
_
x
y
2
_
i
m
i=0
i
_
x
2
y
_
i
.
Taking the limit as m , we deduce (2.98).
Using the (CBS) inequality and (2.98) we may write:
i=1
F (a
i
b
i
)
i=1
|F (a
i
b
i
)|
i=1
_
F
_
a
i
b
2
i
_1
2
_
F
_
a
2
i
b
i
_1
2
_
n
i=1
_
_
F
_
a
i
b
2
i
_1
2
_
2
n
i=1
_
_
F
_
a
2
i
b
i
_1
2
_
2
_1
2
=
_
n
i=1
F
_
a
i
b
2
i
_
n
i=1
F
_
a
2
i
b
i
_
_1
2
The following particular inequalities also hold [12, pp. 165-166].
22 S.S. DRAGOMIR
(1) Let a and

b be sequences of nonnegative real numbers. Then one has the inequalities
_
n
k=1
exp (a
k
b
k
)
_
2
k=1
exp
_
a
k
b
2
k
_
n
k=1
exp
_
a
2
k
b
k
_
; (2.101)
_
n
k=1
sinh (a
k
b
k
)
_
2
k=1
sinh
_
a
k
b
2
k
_
n
k=1
sinh
_
a
2
k
b
k
_
; (2.102)
_
n
k=1
cosh (a
k
b
k
)
_
2
k=1
cosh
_
a
k
b
2
k
_
n
k=1
cosh
_
a
2
k
b
k
_
. (2.103)
(2) Let a and

b be such that a
k
, b
k
(0, 1) for any k {1, . . . , n} . Then one has the
inequalities:
(2.104)
_
n
k=1
tan (a
k
b
k
)
_
2
k=1
tan
_
a
k
b
2
k
_
n
k=1
tan
_
a
2
k
b
k
_
;
(2.105)
_
n
k=1
arcsin (a
k
b
k
)
_
2
k=1
arcsin
_
a
k
b
2
k
_
n
k=1
arcsin
_
a
2
k
b
k
_
;
(2.106)
_
ln
_
n
k=1
_
1 +a
k
b
k
1 a
k
b
k
_
__
2
ln
_
n
k=1
_
1 +a
k
b
2
k
1 a
k
b
2
k
_
_
ln
_
n
k=1
_
1 +a
2
k
b
k
1 a
2
k
b
k
_
_
;
(2.107)
_
ln
_
n
k=1
_
1
1 a
k
b
k
_
__
2
ln
_
n
k=1
_
1
1 a
k
b
2
k
_
_
ln
_
n
k=1
_
1
1 a
2
k
b
k
_
_
.
2.12. Wagners Inequality for Real Numbers. The following generalisation of the (CBS)
inequality for sequences of real numbers is known in the literature as Wagners inequality [15],
or [14] (see also [4, p. 85]).
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
) be sequences of real numbers. If
0 x 1, then one has the inequality
(2.108)
_
n
k=1
a
k
b
k
+x
1i=jn
a
i
b
j
_
2
_
n
k=1
a
2
k
+ 2x
1i<jn
a
i
a
j
__
n
k=1
b
2
k
+ 2x
1i<jn
b
i
b
j
_
.
Proof. We shall follow the proof in [13] (see also [4, p. 85]).
For any x [0, 1] , consider the quadratic polynomial in y
P (y) := (1 x)
n
k=1
(a
k
y b
k
)
2
+x
_
n
k=1
(a
k
y b
k
)
_
2
= (1 x)
_
y
2
n
k=1
a
2
k
2y
n
k=1
a
k
b
k
+
n
k=1
b
2
k
_
+x
_
_
y
2
_
n
k=1
a
k
_
2
2y
_
n
k=1
a
k
__
n
k=1
b
k
_
+
_
n
k=1
b
k
_
2
_
_
=
_
_
(1 x)
n
k=1
a
2
k
+x
_
n
k=1
a
k
_
2
_
_
y
2
2y
_
(1 x)
n
k=1
a
k
b
k
+x
n
k=1
a
k
n
k=1
b
k
_
+ (1 x)
n
k=1
b
2
k
+x
_
n
k=1
b
k
_
2
=
_
_
_
n
k=1
a
2
k
+x
_
_
_
n
k=1
a
k
_
2
k=1
a
2
k
_
_
_
_
_
y
2
2y
_
n
k=1
a
k
b
k
+x
_
n
k=1
a
k
n
k=1
b
k

n
k=1
a
k
b
k
__
+
n
k=1
b
2
k
+x
_
_
_
n
k=1
b
k
_
2
k=1
b
2
k
_
_
.
Since, it is obvious that:
_
n
k=1
a
k
_
2
k=1
a
2
k
= 2
1i<jn
a
i
a
j
,
n
k=1
a
k
n
k=1
b
k

n
k=1
a
k
b
k
=
1i=jn
a
i
b
j
and
_
n
k=1
b
k
_
2
k=1
b
2
k
= 2
1i<jn
b
i
b
j
,
we get
P (y) =
_
n
k=1
a
2
k
+ 2x
1i<jn
a
i
a
j
_
y
2
2y
_
n
k=1
a
k
b
k
+x
1i=jn
a
i
b
j
_
+
n
k=1
b
2
k
+ 2x
1i<jn
b
i
b
j
.
24 S.S. DRAGOMIR
Taking into consideration, by the denition of P, that P (y) 0 for any y R, it follows that
the discriminant 0, i.e.,
0
1
4
=
_
n
k=1
a
k
b
k
+x
1i=jn
a
i
b
j
_
2
_
n
k=1
a
2
k
+ 2x
1i<jn
a
i
a
j
__
n
k=1
b
2
k
+ 2x
1i<jn
b
i
b
j
_
Remark 2.57. If x = 0, then from (2.108) we recapture the (CBS) inequality for real num-
bers.
2.13. Wagners inequality for Complex Numbers. The following inequality which provides
a version for complex numbers of Wagners result holds [16].
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
) be sequences of complex numbers.
Then for any x [0, 1] one has the inequality
(2.109)
_
n
k=1
Re
_
a
k
b
k
_
+x
1i=jn
Re
_
a
i
b
j
_
_
2
_
n
k=1
|a
k
|
2
+ 2x
1i<jn
Re (a
i
a
j
)
__
n
k=1
|b
k
|
2
+ 2x
1i<jn
Re
_
b
i
b
j
_
_
.
Proof. Start with the function f : R R,
(2.110) f (t) = (1 x)
n
k=1
|ta
k
b
k
|
2
+x
k=1
(ta
k
b
k
)
2
.
We have
f (t) = (1 x)
n
k=1
(ta
k
b
k
)
_
t a
k

b
k
_
(2.111)
+x
_
t
n
k=1
a
k

n
k=1
b
k
__
t
n
k=1
a
k

n
k=1
b
k
_
= (1 x)
_
t
2
n
k=1
|a
k
|
2
t
n
k=1
b
k
a
k
t
n
k=1
a
k
b
k
+
n
k=1
|b
k
|
2
_
+x
_
t
2
n
k=1
|a
k
|
2
t
n
k=1
b
k
n
k=1
a
k
t
n
k=1
a
k
n
k=1
b
k
+
n
k=1
|b
k
|
2
_
=
_
_
(1 x)
n
k=1
|a
k
|
2
+x
k=1
a
k
2
_
_
t
2
+ 2
_
(1 x)
n
k=1
Re
_
a
k
b
k
_
+x Re
_
n
k=1
a
k
n
k=1
b
k
__
t
+ (1 x)
n
k=1
|b
k
|
2
+x
k=1
b
k
2
.
Observe that
k=1
a
k
2
=
n
i,j=1
a
i
a
j
(2.112)
=
n
i=1
|a
i
|
2
+
1i=jn
a
i
a
j
=
n
i=1
|a
i
|
2
+
1i<jn
a
i
a
j
+
1j<in
a
i
a
j
=
n
i=1
|a
i
|
2
+ 2
1i<jn
Re (a
i
a
j
)
and, similarly,
(2.113)
k=1
b
k
2
=
n
i=1
|b
i
|
2
+ 2
1i<jn
Re
_
b
i
b
j
_
.
Also
n
k=1
a
k
n
k=1
b
k
=
n
i=1
a
i
b
i
+
1i=jn
a
i
b
j
and thus
(2.114) Re
_
n
k=1
a
k
n
k=1
b
k
_
=
n
i=1
Re
_
a
i
b
i
_
+
1i=jn
Re
_
a
i
b
j
_
.
Utilising (2.112) (2.114), by (2.111), we deduce
(2.115) f (t) =
_
n
k=1
|a
k
|
2
+ 2x
1i<jn
Re (a
i
a
j
)
_
t
2
+ 2
_
n
k=1
Re
_
a
k
b
k
_
+x
1i=jn
Re
_
a
i
b
j
_
_
t +
n
k=1
|b
k
|
2
+ 2x
1i<jn
Re
_
b
i
b
j
_
.
26 S.S. DRAGOMIR
Since, by (2.110), f (t) 0 for any t R, it follows that the discriminant of the quadratic
function given by (2.115) is negative, i.e.,
0
1
4
=
_
n
k=1
Re
_
a
k
b
k
_
+x
1i=jn
Re
_
a
i
b
j
_
_
2
_
n
k=1
|a
k
|
2
+ 2x
1i<jn
Re (a
i
a
j
)
__
n
k=1
|b
k
|
2
+ 2x
1i<jn
Re
_
b
i
b
j
_
_
Remark 2.59. If x = 0, then we get the (CBS) inequality
(2.116)
_
n
k=1
Re
_
a
k
b
k
_
_
2
k=1
|a
k
|
2
n
k=1
|b
k
|
2
.
REFERENCES
[1] V.Y. BUNIAKOWSKI, Sur quelques ingalits concernant les intgrales aux differences nies,
Mem. Acad. St. Petersburg, (7) 1 (1859), No. 9, 1-18.
[2] A.L. CAUCHY, Cours dAnalyse de lcole Royale Polytechnique, I
re
Partie, Analyse Algbrique,
Paris, 1821.
[3] P. SCHREIDER, The Cauchy-Bunyakovsky-Schwarz inequality, Hermann Gramann (Lieschow,
1994), 6470.
[4] D.S. MITRINOVI
C, J.E. PE
CARI
C AND A.M. FINK, Classical and New Inequalities in Analysis,

Kluwer Academic Publishers, Dordrecht/Boston/London, 1993.
[5] S.S. DRAGOMIR, On some inequalities (Romanian), Caiete Metodico Stiintice, No. 13, 1984,
pp. 20. Faculty of Mathematics, Timi soara University, Romania.
[6] E.F. BECKENBACH AND R. BELLMAN, Inequalities, Springer-Verlag, Berlin-Gttingen-
Heidelberg, 1961.
[7] S.S. DRAGOMIR, On Cauchy-Buniakowski-Schwartzs Inequality for Real Numbers (Romanian),
Caiete Metodico- Stiintice, No. 57, pp. 24, 1989. Faculty of Mathematics, Timi soara University,
Romania.
[8] S.S. DRAGOMIR AND J. SNDOR, Some generalisations of Cauchy-Buniakowski-Schwartzs
inequality (Romanian), Gaz. Mat. Metod. (Bucharest), 11 (1990), 104109.
[9] I. CUCUREZEANU, Problems on Number Theory (Romanian), Ed. Technic a, Bucharest, 1976.
[10] S.S. DRAGOMIR, On an inequality of Tiberiu Popoviciu (Romanian), Gaz. Mat. Metod.,
(Bucharest) 8 (1987), 124128. ZBL No. 712:110A.
[11] T. POPOVICIU, Gazeta Matematic a, 16 (1940), p. 334.
[12] S.S. DRAGOMIR, Inequalities of Cauchy-Buniakowski-Schwartzs type for positive linear func-
tionals (Romanian), Gaz. Mat. Metod. (Bucharest), 9 (1988), 162164.
[13] T. ANDRESCU, D. ANDRICA AND M.O. DRMBE, The trinomial principle in obtaining inequal-
ities (Romanian), Gaz. Mat. (Bucharest), 90 (1985), 332338.
[14] P. FLOR, ber eine Unglichung von S.S. Wagner, Elemente Math., 20 (1965), 136.
[15] S.S. WAGNER, Amer. Math. Soc., Notices, 12 (1965), 220.
[16] S.S. DRAGOMIR, A version of Wagners inequality for complex numbers, submitted.
28 S.S. DRAGOMIR
3. REFINEMENTS OF THE (CBS) INEQUALITY
3.1. A Renement in Terms of Moduli. The following result was proved in [1].
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
) be sequences of real numbers. Then
(3.1)
n
k=1
a
2
k
n
k=1
b
2
k

_
n
k=1
a
k
b
k
_
2
k=1
a
k
|a
k
|
n
k=1
b
k
|b
k
|
n
k=1
a
k
|b
k
|
n
k=1
|a
k
| b
k
0.
For any i, j {1, . . . , n} the next elementary inequality is true:
(3.2) |a
i
b
j
a
j
b
i
| ||a
i
b
j
| |a
j
b
i
|| .
By multiplying this inequality with |a
i
b
j
a
j
b
i
| 0 we get
(a
i
b
j
a
j
b
i
)
2
|(a
i
b
j
a
j
b
i
) (|a
i
| |b
j
| |a
j
| |b
i
|)| (3.3)
= |a
i
|a
i
| b
j
|b
j
| +b
i
|b
i
| a
j
|a
j
| |a
i
| b
i
a
j
|b
j
| a
i
b
j
|a
j
| |b
i
|| .
Summing (3.3) over i and j from 1 to n, we deduce
n
i,j=1
(a
i
b
j
a
j
b
i
)
2
i,j=1
a
i
|a
i
| b
j
|b
j
| +b
i
|b
i
| a
j
|a
j
| |a
i
| b
i
a
j
|b
j
| a
i
b
j
|a
j
| |b
i
|
i,j=1
(a
i
|a
i
| b
j
|b
j
| +b
i
|b
i
| a
j
|a
j
| |a
i
| b
i
a
j
|b
j
| a
i
b
j
|a
j
| |b
i
|)
,
giving the desired inequality (3.1).
The following corollary is a natural consequence of (3.1) [1, Corollary 4].
Corollary 3.2. Let a be a sequence of real numbers. Then
(3.4)
1
n
n
k=1
a
2
k

_
1
n
n
k=1
a
k
_
2
1
n
n
k=1
a
k
|a
k
|
1
n
n
k=1
a
k

1
n
n
k=1
|a
k
|
0.
There are some particular inequalities that may also be deduced from the above Theorem 3.1
(see [1, p. 80]).
(1) Suppose that for a and

b sequences of real numbers, one has sgn (a
k
) = sgn (b
k
) =
e
k
{1, 1} . Then one has the inequality
(3.5)
n
k=1
a
2
k
n
k=1
b
2
k

_
n
k=1
a
k
b
k
_
2
k=1
e
k
a
2
k
n
k=1
e
k
b
2
k

_
n
k=1
e
k
a
k
b
k
_
2
0.
(2) If a = (a
1
, . . . , a
2n
) , then we have the inequality
(3.6) 2n
2n
k=1
a
2
k

_
2n
k=1
(1)
k
a
k
_
2
2n
k=1
a
k
2n
k=1
(1)
k
|a
k
|
0.
(3) If a = (a
1
, . . . , a
2n+1
) , then we have the inequality
(3.7) (2n + 1)
2n+1
k=1
a
2
k

_
2n+1
k=1
(1)
k
a
k
_
2
2n+1
k=1
a
k
2n+1
k=1
(1)
k
|a
k
|
0.
The following version for complex numbers is valid as well.
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
) be sequences of complex numbers.
(3.8)
n
i=1
|a
i
|
2
n
i=1
|b
i
|
2
i=1
a
i
b
i
i=1
|a
i
| a
i
n
i=1
|b
i
| b
i
i=1
|a
i
| b
i
n
i=1
|b
i
| a
i
0.
Proof. We have for any i, j {1, . . . , n} that
| a
i
b
j
a
j
b
i
| ||a
i
| |b
j
| |a
j
| |b
i
|| .
Multiplying by | a
i
b
j
a
j
b
i
| 0, we get
| a
i
b
j
a
j
b
i
|
2
||a
i
| a
i
|b
j
| b
j
+|a
j
| a
j
|b
i
| b
i
|a
i
| b
i
|b
j
| a
j
|b
i
| a
i
|a
j
| b
j
| .
Summing over i and j from 1 to n and using the Lagranges identity for complex numbers:
n
i=1
|a
i
|
2
n
i=1
|b
i
|
2
i=1
a
i
b
i
2
=
1
2
n
i,j=1
| a
i
b
j
a
j
b
i
|
2
we deduce the desired inequality (3.8).
Remark 3.4. Similar particular inequalities may be stated, but we omit the details.
3.2. A Renement for a Sequence Whose Norm is One. The following result holds [1, The-
orem 6].
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
) be sequences of real numbers and e =
(e
1
, . . . , e
n
) be such that
n
i=1
e
2
i
= 1. Then the following inequality holds
n
i=1
a
2
i
n
i=1
b
2
i

_
k=1
a
k
b
k

n
k=1
e
k
a
k
n
k=1
e
k
b
k
k=1
e
k
a
k
n
k=1
e
k
b
k
_
2
(3.9)
_
n
k=1
a
k
b
k
_
2
.
From the (CBS) inequality, one has
(3.10)
n
k=1
_
a
k

_
n
i=1
e
i
a
i
_
e
k
_
2
n
k=1
_
b
k

_
n
i=1
e
i
b
i
_
e
k
_
2
_
n
k=1
_
a
k

_
n
i=1
e
i
a
i
_
e
k
__
b
k

_
n
i=1
e
i
b
i
_
e
k
__
2
.
30 S.S. DRAGOMIR
Since
n
k=1
e
2
k
= 1, a simple calculation shows that
n
k=1
_
a
k

_
n
i=1
e
i
a
i
_
e
k
_
2
=
n
k=1
a
2
k

_
n
k=1
e
k
a
k
_
2
,
n
k=1
_
b
k

_
n
i=1
e
i
b
i
_
e
k
_
2
=
n
k=1
b
2
k

_
n
k=1
e
k
b
k
_
2
,
and
n
k=1
_
a
k

_
n
i=1
e
i
a
i
_
e
k
__
b
k

_
n
i=1
e
i
b
i
_
e
k
_
=
n
k=1
a
k
b
k

n
k=1
e
k
a
k
n
k=1
e
k
b
k
and then the inequality (3.10) becomes
(3.11)
_
_
n
k=1
a
2
k

_
n
k=1
e
k
a
k
_
2
_
_
_
_
n
k=1
b
2
k

_
n
k=1
e
k
b
k
_
2
_
_
_
n
k=1
a
k
b
k

n
k=1
e
k
a
k
n
k=1
e
k
b
k
_
2
0.
Using the elementary inequality
_
m
2
l
2
_ _
p
2
q
2
_
(mp lq)
2
, m, l, p, q R
for the choices
m =
_
n
k=1
a
2
k
_1
2
, l =
k=1
e
k
a
k
, p =
_
n
k=1
b
2
k
_1
2
and q =
k=1
e
k
b
k
the above inequality (3.11) provides the following result

(3.12)
_
_
_
n
k=1
a
2
k
_1
2
_
n
k=1
b
2
k
_1
2
k=1
e
k
a
k
n
k=1
e
k
b
k
_
_
2
k=1
a
k
b
k

n
k=1
e
k
a
k
n
k=1
e
k
b
k
2
.
Since
_
n
k=1
a
2
k
_1
2
_
n
k=1
b
2
k
_1
2
k=1
e
k
a
k
n
k=1
e
k
b
k
then, by taking the square root in (3.12) we deduce the rst part of (3.9).
The second part is obvious, and the theorem is proved.
The following corollary is a natural consequence of the above theorem [1, Corollary 7].
b, e be as in Theorem 3.5. If
n
k=1
a
k
b
k
= 0, then one has the inequality:
(3.13)
n
k=1
a
2
k
n
k=1
b
2
k
4
_
n
k=1
e
k
a
k
_
2
_
n
k=1
e
k
b
k
_
2
.
The following inequalities are interesting as well [1, p. 81].
(1) For any a,
b one has the inequality

n
k=1
a
2
k
n
k=1
b
2
k

_
k=1
a
k
b
k

1
n
n
k=1
a
k
n
k=1
b
k
+
1
n
k=1
a
k
n
k=1
b
k
_
2
(3.14)
_
n
k=1
a
k
b
k
_
2
.
(2) If
n
k=1
a
k
b
k
= 0, then
(3.15)
n
k=1
a
2
k
n
k=1
b
2
k

4
n
2
_
n
k=1
a
k
_
2
_
n
k=1
b
k
_
2
.
In a similar manner, we may state and prove the following result for complex numbers.
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
) be sequences of complex numbers and
e = (e
1
, . . . , e
n
) a sequence of complex numbers satisfying the condition
n
i=1
|e
i
|
2
= 1. Then
the following renement of the (CBS) inequality holds
n
i=1
|a
i
|
2
n
i=1
|b
i
|
2
k=1
a
k
b
k

n
k=1
a
k
e
k

n
k=1
e
k
b
k
k=1
a
k
e
k

n
k=1
e
k
b
k
_
2
(3.16)
k=1
a
k
b
k
2
.
The proof is similar to the one in Theorem3.5 on using the corresponding (CBS) inequality
for complex numbers.
Remark 3.8. Similar particular inequalities may be stated, but we omit the details.
3.3. A Second Renement in Terms of Moduli. The following lemma holds.
Lemma 3.9. Let a = (a
1
, . . . , a
n
) be a sequence of real numbers and p = (p
1
, . . . , p
n
) a
sequence of positive real numbers with
n
i=1
p
i
= 1. Then one has the inequality:
(3.17)
n
i=1
p
i
a
2
i

_
n
i=1
p
i
a
i
_
2
i=1
p
i
|a
i
| a
i
i=1
p
i
|a
i
|
n
i=1
p
i
a
i
.
Proof. By the properties of moduli we have
(a
i
a
j
)
2
= |(a
i
a
j
) (a
i
a
j
)| |(|a
i
| |a
j
|) (a
i
a
j
)|
for any i, j {1, . . . , n} . This is equivalent to
(3.18) a
2
i
2a
i
a
j
+a
2
j
||a
i
| a
i
+|a
j
| a
j
|a
i
| a
j
|a
j
| a
i
|
for any i, j {1, . . . , n} .
32 S.S. DRAGOMIR
If we multiply (3.18) by p
i
p
j
0 and sum over i and j from 1 to n we deduce
n
j=1
p
j
n
i=1
p
i
a
2
i
2
n
i=1
p
i
a
i
n
j=1
p
j
a
j
+
n
i=1
p
i
n
j=1
p
j
a
2
j
i,j=1
p
i
p
j
|a
i
| a
i
+|a
j
| a
j
|a
i
| a
j
|a
j
| a
i
i,j=1
p
i
p
j
(|a
i
| a
i
+|a
j
| a
j
|a
i
| a
j
|a
j
| a
i
)
,
Using the above lemma, we may prove the following renement of the (CBS) -inequality.
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
) be two sequences of real numbers.
(3.19)
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
i=1
a
2
i

n
i=1
sgn(a
i
) |b
i
| b
i
i=1
|a
i
b
i
|
n
i=1
a
i
b
i
0.
Proof. If we choose (for a
i
= 0, i {1, . . . , n}) in (3.17), that
p
i
:=
a
2
i
n
k=1
a
2
k
, x
i
=
b
i
a
i
, i {1, . . . , n} ,
we get
n
i=1
a
2
i
n
k=1
a
2
k
_
b
i
a
i
_
2
_
n
i=1
a
2
i
n
k=1
a
2
k
b
i
a
i
_
2
i=1
a
2
i
n
k=1
a
2
k
b
i
a
i
b
i
a
i
i=1
a
2
i
n
k=1
a
2
k
b
i
a
i
i=1
a
2
i
n
k=1
a
2
k
b
i
a
i
from where we get

n
i=1
b
2
i
n
k=1
a
2
k
n
i=1
a
i
b
i
)
2
(
n
k=1
a
2
k
)
2

n
i=1
|a
i
|
a
i
|b
i
| b
i
n
k=1
a
2
k
n
i=1
|a
i
b
i
|
n
i=1
a
i
b
i
(
n
k=1
a
2
k
)
2

The case for complex numbers is as follows.
Lemma 3.11. Let z = (z
1
, . . . , z
n
) be a sequence of complex numbers and p = (p
1
, . . . , p
n
) a
sequence of positive real numbers with
n
i=1
p
i
= 1. Then one has the inequality:
(3.20)
n
i=1
p
i
|z
i
|
2
i=1
p
i
z
i
i=1
p
i
|z
i
| z
i
i=1
p
i
|z
i
|
n
i=1
p
i
z
i
.
Proof. By the properties of moduli for complex numbers we have
|z
i
z
j
|
2
|(|z
i
| |z
j
|) (z
i
z
j
)|
for any i, j {1, . . . , n} , which is clearly equivalent to
|z
i
|
2
2 Re (z
i
z
j
) +|z
j
|
2
||z
i
| z
i
+|z
j
| z
j
z
i
|z
j
| |z
i
| z
j
|
for any i, j {1, . . . , n} .
If we multiply with p
i
p
j
0 and sum over i and j from 1 to n, we deduce the desired
inequality (3.20).
Now, in a similar manner to the one in Theorem 3.10, we may state the following result for
complex numbers.
1
, . . . , a
n
) (a
i
= 0, i = 1, . . . , n) and

b = (b
1
, . . . , b
n
) be two se-
quences of complex numbers. Then one has the inequality:
(3.21)
n
i=1
|a
i
|
2
n
i=1
|b
i
|
2
i=1
a
i
b
i
i=1
|a
i
|
a
i
|b
i
| b
i
i=1
|a
i
| b
i
n
i=1
a
i
b
i
0.
3.4. A Renement for a Sequence Less than the Weights. The following result was obtained
in [1, Theorem 9] (see also [2, Theorem 3.10]).
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
) be sequences of real numbers and
p = (p
1
, . . . , p
n
), q = (q
1
, . . . , q
n
) be sequences of nonnegative real numbers such that p
k
q
k
for any k {1, . . . , n} . Then we have the inequality
n
k=1
p
k
a
2
k
n
k=1
p
k
b
2
k

_
_
k=1
(p
k
q
k
) a
k
b
k
+
_
n
k=1
q
k
a
2
k
n
k=1
q
k
b
2
k
_1
2
_
_
2
(3.22)
k=1
(p
k
q
k
) a
k
b
k
k=1
q
k
a
k
b
k
_
2
_
n
k=1
p
k
a
k
b
k
_
2
.
Since p
k
q
k
0, then the (CBS) inequality for the weights r
k
:= p
k
q
k
will give
(3.23)
_
n
k=1
p
k
a
2
k

n
k=1
q
k
a
2
k
__
n
k=1
p
k
b
2
k

n
k=1
q
k
b
2
k
_
_
n
k=1
(p
k
q
k
) a
k
b
k
_
2
.
(ac bd)
2
_
a
2
b
2
_ _
c
2
d
2
_
, a, b, c, d R
for the choices
a =
_
n
k=1
p
k
a
2
k
_1
2
, b =
_
n
k=1
q
k
a
2
k
_1
2
, c =
_
n
k=1
p
k
b
2
k
_1
2
and
d =
_
n
k=1
q
k
b
2
k
_1
2
34 S.S. DRAGOMIR
we deduce by (3.23) that
(3.24)
_
_
_
n
k=1
p
k
a
2
k
_1
2
_
n
k=1
p
k
b
2
k
_1
2
_
n
k=1
q
k
a
2
k
_1
2
_
n
k=1
q
k
b
2
k
_1
2
_
_
2
_
n
k=1
(p
k
q
k
) a
k
b
k
_
2
.
Since, obviously,
_
n
k=1
p
k
a
2
k
_1
2
_
n
k=1
p
k
b
2
k
_1
2
_
n
k=1
q
k
a
2
k
_1
2
_
n
k=1
q
k
b
2
k
_1
2
then, by (3.24), on taking the square root, we would get
_
n
k=1
p
k
a
2
k
_1
2
_
n
k=1
p
k
b
2
k
_1
2
_
n
k=1
q
k
a
2
k
_1
2
_
n
k=1
q
k
b
2
k
_1
2
+
k=1
(p
k
q
k
) a
k
b
k
,
which provides the rst inequality in (3.22).
The other inequalities are obvious and we omit the details.
The following corollary is a natural consequence of the above theorem [2, Corollary 3.11].

b be sequences of real numbers and s = (s
1
, . . . , s
n
) be such that
0 s
k
1 for any k {1, . . . , n} . Then one has the inequalities
n
k=1
a
2
k
n
k=1
b
2
k

_
_
k=1
(1 s
k
) a
k
b
k
+
_
n
k=1
s
k
a
2
k
n
k=1
s
k
b
2
k
_1
2
_
_
2
(3.25)
k=1
(1 s
k
) a
k
b
k
k=1
s
k
a
k
b
k
_
2
_
n
k=1
a
k
b
k
_
2
.
Remark 3.15. Assume that a,

b and s are as in Corollary 3.14. The following inequalities hold
(see [2, p. 15]).
a) If
n
k=1
a
k
b
k
= 0, then
(3.26)
n
k=1
a
2
k
n
k=1
b
2
k
4
_
n
k=1
s
k
a
k
b
k
_
2
.
b) If
n
k=1
s
k
a
k
b
k
= 0, then
(3.27)
n
k=1
a
2
k
n
k=1
b
2
k

_
_
k=1
a
k
b
k
+
_
n
k=1
k
a
2
k
n
k=1
k
b
2
k
_1
2
_
_
2
.
In particular, we may obtain the following particular inequalities involving trigonometric
functions (see [2, p. 15])
n
k=1
a
2
k
n
k=1
b
2
k

_
_
k=1
a
k
b
k
cos
2
+
_
n
k=1
a
2
k
sin
2
k
n
k=1
b
2
k
sin
2
k
_1
2
_
_
2
(3.28)
k=1
a
k
b
k
cos
2
k=1
a
k
b
k
sin
2
_
2
_
n
k=1
a
k
b
k
_
2
,
where a
k
, b
k
,
k
R, k = 1, . . . , n.
If one would assume that
n
k=1
a
k
b
k
= 0, then
(3.29)
n
k=1
a
2
k
n
k=1
b
2
k
4
_
n
k=1
a
k
b
k
sin
2
k
_
2
.
If
n
k=1
a
k
b
k
sin
2
k
= 0, then
(3.30)
n
k=1
a
2
k
n
k=1
b
2
k

_
_
k=1
a
k
b
k
+
_
n
k=1
a
2
k
sin
2
k
n
k=1
b
2
k
sin
2
k
_1
2
_
_
2
.
3.5. AConditional Inequality Providing a Renement. The following lemma holds [2, Lemma
4.1].
Lemma 3.16. Consider the sequences of real numbers x = (x
1
, . . . , x
n
) , y = (y
1
, . . . , y
n
) and
z = (z
1
, . . . , z
n
) . If
(3.31) y
2
k
|x
k
z
k
| for any k {1, . . . , n} ,
then one has the inequality:
(3.32)
_
n
k=1
|y
k
|
_
2
k=1
|x
k
|
n
k=1
|z
k
| .
Proof. We will follow the proof in [2]. Using the condition (3.31) and the (CBS) inequality,
we have
n
k=1
|y
k
|
n
k=1
|x
k
|
1
2
|z
k
|
1
2
_
n
k=1
_
|x
k
|
1
2
_
2
n
k=1
_
|z
k
|
1
2
_
2
_1
2
=
_
n
k=1
|x
k
|
n
k=1
|z
k
|
_1
2
The following result holds [2, Theorem 4.6].
36 S.S. DRAGOMIR
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
) and c = (c
1
, . . . , c
n
) be sequences of
real numbers such that
(i) |b
k
| +|c
k
| = 0 (k {1, . . . , n});
(ii) |a
k
|
2|b
k
c
k
|
|b
k
|+|c
k
|
for any k {1, . . . , n} .
(3.33)
n
k=1
|a
k
|
2
n
k=1
|b
k
|
n
k=1
|c
k
|
n
k=1
(|b
k
| +|c
k
|)
.
Proof. We will follow the proof in [2]. By (ii) we observe that
|a
k
|
2 |b
k
c
k
|
|b
k
| +|c
k
|

_
_
_
2 |b
k
|
for any k {1, . . . , n}
2 |c
k
|
and thus
x
k
:= 2 |b
k
| |a
k
| 0 and (3.34)
z
k
:= 2 |c
k
| |a
k
| 0 for any k {1, . . . , n} .
A simple calculation also shows that the relation (ii) is equivalent to
(3.35) a
2
k
(2 |b
k
| |a
k
|) (2 |c
k
| |a
k
|) for any k {1, . . . , n} .
If we consider y
k
:= a
k
and take x
k
, z
k
(k = 1, . . . , n) as dened by (3.34), then we get
y
2
k
x
k
z
k
(with x
k
, z
k
0) for any k {1, . . . , n}. Applying Lemma 3.16 we deduce
(3.36)
_
n
k=1
|a
k
|
_
2
_
2
n
k=1
|b
k
|
n
k=1
|a
k
|
__
2
n
k=1
|c
k
|
n
k=1
|a
k
|
_
Corollary 3.18. For any sequence x and y of real numbers, with |x
k
|+|y
k
| = 0 (k = 1, . . . , n) ,
one has:
(3.37)
n
k=1
|x
k
y
k
|
|x
k
| +|y
k
|

2
n
k=1
|x
k
|
n
k=1
|y
k
|
n
k=1
(|x
k
| +|y
k
|)
.
For two positive real numbers, let us recall the following means
A(a, b) :=
a +b
2
(the arithmetic mean)
G(a, b) :=
ab (the geometric mean)

and
H (a, b) :=
2
1
a
+
1
b
(the harmonic mean).
We remark that if a = (a
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
obviously
(3.38)
n
i=1
A(a
i
, b
i
) = A
_
n
i=1
a
i
,
n
i=1
b
i
_
,
and, by the (CBS) inequality,
(3.39)
n
i=1
G(a
i
, b
i
) G
_
n
i=1
a
i
,
n
i=1
b
i
_
.
The following similar result for harmonic means also holds [2, p. 19].
Theorem 3.19. For any two sequences of positive real numbers a and

b we have the property:
(3.40)
n
i=1
H (a
i
, b
i
) H
_
n
i=1
a
i
,
n
i=1
b
i
_
.
Proof. Follows by Corollary 3.18 on choosing x
k
= a
k
, y
k
= b
k
and multiplying the inequality
(3.37) with 2.
The following renement of the (CBS) inequality holds [2, Corollary 4.9]. This result is
known in the literature as Milnes inequality [8].
Theorem 3.20. For any two sequences of real numbers p = (p
1
, . . . , p
n
), q = (q
1
, . . . , q
n
) with
|p
k
| +|q
k
| = 0 (k = 1, . . . , n) , one has the inequality:
(3.41)
_
n
k=1
p
k
q
k
_
2
k=1
_
p
2
k
+q
2
k
_
n
k=1
p
2
k
q
2
k
p
2
k
+q
2
k
k=1
p
2
k
n
k=1
q
2
k
.
Proof. We shall follow the proof in [2]. The rst inequality is obvious by Lemma 3.16 on
choosing y
k
= p
k
q
k
, x
k
= p
2
k
+q
2
k
and z
k
=
p
2
k
q
2
k
p
2
k
+q
2
k
(k = 1, . . . , n) .
The second inequality follows by Corollary 3.18 on choosing x
k
= p
2
k
and y = q
2
k
(k = 1,
. . . , n).
Remark 3.21. The following particular inequality is obvious by (3.41)
_
n
i=1
sin
k
cos
k
_
2
n
n
i=1
sin
2
k
cos
2
k
(3.42)
i=1
sin
2
k
n
i=1
cos
2
k
;
for any
k
R, k {1, . . . , n} .
3.6. A Renement for Non-Constant Sequences. The following result was proved in [3, The-
orem 1].
i
)
iN
,

b = (b
i
)
iN
, p = (p
i
)
iN
be sequences of real numbers such
that
(i) a
i
= a
j
and b
i
= b
j
for i = j, i, j N;
(ii) p
i
> 0 for all i N.
Then for any H a nite part of N one has the inequality:
(3.43)
iH
p
i
a
2
i
iH
p
i
b
2
i

_
iH
p
i
a
i
b
i
_
2
max {A, B} 0,
where
(3.44) A := max
JH
J=
_
iH
p
i
a
i
b
i
jJ
p
j
a
j

iH
p
i
a
2
i
jJ
p
j
b
j
_
2
P
J
iH
p
i
a
2
i

_
iJ
p
i
a
i
_
2
38 S.S. DRAGOMIR
and
(3.45) B := max
JH
J=
_
iH
p
i
a
i
b
i
jJ
p
j
b
j

iH
p
i
b
2
i
jJ
p
j
a
j
_
2
P
J
iH
p
i
b
2
i

_
iJ
p
i
b
i
_
2
and P
J
:=
jJ
p
j
.
Let J be a part of H. Dene the mapping f
J
: R R given by
f
J
(t) =
iH
p
i
a
2
i
_
_

iH\J
p
i
b
2
i
+
iJ
p
i
(b
i
+t)
2
_
_
_
_

iH\J
p
i
a
i
b
i
+
iJ
p
i
a
i
(b
i
+t)
_
_
2
.
Then by the (CBS) inequality we have that f
J
(t) 0 for all t R.
On the other hand we have
f
J
(t) =
iH
p
i
a
2
i
_
iH
p
i
b
2
i
+ 2t
iH
p
i
b
i
+t
2
P
J
_
iH
p
i
a
i
b
i
+t
iJ
p
i
a
i
_
2
= t
2
_
_
P
J
iH
p
i
a
2
i

_
iJ
p
i
a
i
_
2
_
_
+ 2t
_
iH
p
i
a
2
i
iJ
p
i
b
i
iH
p
i
a
i
b
i
iJ
p
i
a
i
_
+
_
_
iH
p
i
a
2
i
iH
p
i
b
2
i

_
iH
p
i
a
i
b
i
_
2
_
_
for all t R.
Since
P
J
iH
p
i
a
2
i

_
iJ
p
i
a
i
_
2
P
J
iJ
p
i
a
2
i

_
iJ
p
i
a
i
_
2
> 0
as a
i
= a
j
for all i, j {1, . . . , n} with i = j, then, by the inequality f
J
(t) 0 for any t R
we get that
0
1
4
=
_
iH
p
i
a
i
b
i
jJ
p
j
a
j

iH
p
i
a
2
i
jJ
p
j
b
j
_
2
_
_
P
J
iH
p
i
a
2
i

_
iJ
p
i
a
i
_
2
_
_
_
_
iH
p
i
a
2
i
iH
p
i
b
2
i

_
iH
p
i
a
i
b
i
_
2
_
_
from where results the inequality
iH
p
i
a
2
i
iH
p
i
b
2
i

_
iH
p
i
a
i
b
i
_
2
A.
The second part of the proof goes likewise for the mapping g
J
: R R given by
g
J
(t) =
_
_

iH\J
p
i
a
2
i
+
iJ
p
i
(a
i
+t)
2
_
_
iH
p
i
b
2
i

_
_

iH\J
p
i
a
i
b
i
+
iJ
p
i
b
i
(a
i
+t)
_
_
2
and we omit the details.
The following corollary also holds [3, Corollary 1].
Corollary 3.23. With the assumptions of Theorem 3.22 and if
(3.46) C :=
_
iH
p
i
a
i
b
i
iH
p
i
a
i
iH
p
i
a
2
i
iH
p
i
b
i
2
P
H
iH
p
i
a
2
i

_
iH
p
i
a
i
_
2
,
(3.47) D :=
_
iH
p
i
a
i
b
i
iH
p
i
b
i
iH
p
i
b
2
i
iH
p
i
a
i
2
P
H
iH
p
i
b
2
i

_
iH
p
i
b
i
_
2
,
(3.48)
iH
p
i
a
2
i
iH
p
i
b
2
i

_
iH
p
i
a
i
b
i
_
2
max {C, D} 0.
The following corollary also holds [3, Corollary 2].
Corollary 3.24. If a
i
, b
i
= 0 for i N and H is a nite part of N, then one has the inequality
(3.49)
iH
p
i
a
2
i
iH
p
i
b
2
i

_
iH
p
i
a
i
b
i
_
2
1
card (H) 1
max
_
jH
p
j
c
2
j
iH
p
i
a
2
i
,
jH
p
j
d
2
j
iH
p
i
b
2
i
_
0,
where
(3.50) c
j
:= a
j
iH
p
i
a
i
b
i
b
j
iH
p
i
a
2
i
, j H
and
(3.51) d
j
:= a
j
iH
p
i
b
2
i
b
j
iH
p
i
a
i
b
i
, j H.
Proof. Choosing in Theorem 3.22, J = {j} , we get the inequality
iH
p
i
a
2
i
iH
p
i
b
2
i

_
iH
p
i
a
i
b
i
_
2
p
2
j
c
2
j
p
j
iH
p
i
a
2
i
p
2
j
a
2
j
, j H
from where we obtain
_
iH
p
i
a
2
i
p
j
a
2
j
_
_
_
iH
p
i
a
2
i
iH
p
i
b
2
i

_
iH
p
i
a
i
b
i
_
2
_
_
p
j
c
2
j
for any j H.
Summing these inequalities over j H, we get
[card (H) 1]
iH
p
i
a
2
i
_
_
iH
p
i
a
2
i
iH
p
i
b
2
i

_
iH
p
i
a
i
b
i
_
2
_
_
jH
p
j
c
2
j
40 S.S. DRAGOMIR
from where we get the rst part of (3.49).
The second part goes likewise and we omit the details.
Remark 3.25. The following particular inequalities provide renement for the (CBS) inequality
[3, p. 60 p. 61].
(1) Assume that a = (a
1
, . . . , a
n
), b = (b
1
, . . . , b
n
) are nonconstant sequences of real
numbers. Then
(3.52)
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
max
_
_
[
n
i=1
a
2
i
n
i=1
b
i
n
i=1
a
i
n
i=1
a
i
b
i
]
2
n
n
i=1
a
2
i
(
n
i=1
a
i
)
2
,
[
n
i=1
b
i
n
i=1
a
i
b
i
n
i=1
a
i
n
i=1
b
2
i
]
2
n
n
i=1
b
2
i
(
n
i=1
b
i
)
2
_
_
.
(2) Assume that a and

b are sequences of real numbers with not all elements equal to zero,
then
(3.53)
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
1
n 1
max
_
_
n
j=1
_
a
j
n
i=1
a
i
b
i
b
j
n
i=1
a
2
i
_
2
n
i=1
a
2
i
,
n
j=1
_
a
j
n
i=1
a
2
i
b
j
n
i=1
a
i
b
i
_
2
n
i=1
b
2
i
_
_
.
3.7. De Bruijns Inequality. The following renement of the (CBS) inequality was proved
by N.G. de Bruijn in 1960, [4] (see also [5, p. 89]).
1
, . . . , a
n
) is a sequence of real numbers and z = (z
1
, . . . , z
n
) is a
sequence of complex numbers, then
(3.54)
k=1
a
k
z
k
1
2
n
k=1
a
2
k
_
n
k=1
|z
k
|
2
+
k=1
z
2
k
_
.
Equality holds in (3.54) if and only if for k {1, . . . , n} , a
k
= Re (z
k
) , where is a complex
number such that
n
k=1
2
z
2
k
is a nonnegative real number.
Proof. We shall follow the proof in [5, p. 89 p. 90].
By a simultaneous rotation of all the z
k
s about the origin, we get
n
k=1
a
k
z
k
0.
This rotation does not affect the moduli
k=1
a
k
z
k
k=1
z
2
k
and |z
k
| for k {1, . . . , n} .
Hence, it is sufcient to prove inequality (3.54) for the case where
n
k=1
a
k
z
k
0.
If we put z
k
= x
k
+iy
k
(k {1, . . . , n}) , then, by the (CBS) inequality for real numbers,
we have
(3.55)
k=1
a
k
z
k
2
=
_
n
k=1
a
k
z
k
_
2
k=1
a
2
k
n
k=1
x
2
k
.
Since
2x
2
k
= |z
k
|
2
+ Re z
2
k
for any k {1, . . . , n}
we obtain, by (3.55), that
(3.56)
k=1
a
k
z
k
1
2
n
k=1
a
2
k
_
n
k=1
|z
k
|
2
+
n
k=1
Re z
2
k
_
.
As
n
k=1
Re z
2
k
= Re
_
n
k=1
z
2
k
_
k=1
z
2
k
,
then by (3.56) we deduce the desired inequality (3.54).
3.8. McLaughlins Inequality. The following renement of the (CBS) inequality for se-
quences of real numbers was obtained in 1966 by H.W. McLaughlin [7, p. 66].
1
, . . . , a
2n
),

b = (b
1
, . . . , b
2n
(3.57)
_
2n
i=1
a
i
b
i
_
2
+
_
n
i=1
(a
i
b
n+i
a
n+i
b
i
)
_
2
2n
i=1
a
2
i
2n
i=1
b
2
i
with equality if and only if for any i, j {1, . . . , n}
(3.58) a
i
b
j
a
j
b
i
a
n+i
b
n+j
+a
n+j
b
n+i
= 0
and
(3.59) a
i
b
n+j
a
j
b
n+i
+a
n+i
b
j
a
n+j
b
i
= 0.
Proof. We shall follow the proof in [6] by M.O. Drmbe.
The following identity may be obtained by direct computation
(3.60)
2n
i=1
a
2
i
2n
i=1
b
2
i

_
2n
i=1
a
i
b
i
_
2
_
n
i=1
(a
i
b
n+i
a
n+i
b
i
)
_
2
=
1i<jn
(a
i
b
j
a
j
b
i
a
n+i
b
n+j
+a
n+j
b
n+i
)
2
+
1i<jn
(a
i
b
n+j
a
j
b
n+i
+a
n+i
b
j
a
n+j
b
i
)
2
.
42 S.S. DRAGOMIR
It is obvious that (3.57) is a simple consequence of the identity (3.60). The case of equality is
also obvious.
Remark 3.28. For other similar (CBS) type inequalties see the survey paper [7]. An anal-
ogous inequality to (3.57) for sequences a and

b having 4n terms each may be found in [7, p.
70].
3.9. A Renement due to Daykin-Eliezer-Carlitz. We will present now the version due to
Mitrinovi c, Pe cari c and Fink [5, p. 87] of Daykin-Eliezer-Carlitzs renement of the discrete
(CBS) inequality [8].
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
) be two sequences of positive num-
bers. The ineuality
(3.61)
_
n
i=1
a
i
b
i
_
2
i=1
f (a
i
b
i
)
n
i=1
g (a
i
b
i
)
n
i=1
a
2
i
n
i=1
b
2
i
holds if and only if
f (a, b) g (a, b) = a
2
b
2
, (3.62)
f (ka, kb) = k
2
f (a, b) , (3.63)
bf (a, 1)
af (b, 1)
+
af (b, 1)
bf (a, 1)

a
b
+
b
a
(3.64)
for any a, b, k > 0.
Necessity. Indeed, for n = 1, the inequality (3.61) becomes
(ab)
2
f (a, b) g (a, b) a
2
b
2
, a, b > 0
which gives the condition (3.62).
For n = 2 in (3.61), using (3.62), we get
2a
1
b
1
a
2
b
2
f (a
1
, b
1
) g (a
2
, b
2
) +f (a
2
, b
2
) g (a
1
, b
1
) a
2
1
b
2
2
+a
2
2
b
2
1
.
By eliminating g, we get
(3.65) 2
f (a
1
, b
1
)
f (a
2
, b
2
)

a
2
b
2
a
1
b
1
+
f (a
2
, b
2
)
f (a
1
, b
1
)

a
1
b
1
a
2
b
2
a
1
b
2
a
2
b
1
+
a
2
b
1
a
1
b
2
.
By substituting in (3.65) a, b for a
1
, b
1
and ka, kb for a
2
, b
2
(k > 0), we get
2
f (a, b)
f (ka, kb)
k
2
+
f (ka, kb)
f (a, b)
k
2
2
and this is valid only if k
2
f (a, b) (f (ka, kb)) = 1, i.e., the condition (3.63) holds.
Using (3.65), for a
1
= a, b
1
= b, a
2
= b, b
2
= 1, we have
(3.66) 2
f(a,1)
a
f(b,1)
b
+
f(b,1)
b
f(a,1)
a
a
b
+
b
a
.
The rst inequality in (3.66) is always satised while the second inequality is equivalent to
(3.64).
Sufciency. Suppose that (3.62) holds. Then inequality (3.61) can be written in the form
2
1i<jn
a
i
b
i
a
j
b
j

1i<jn
[f (a
i
, b
i
) g (a
j
, b
j
) +f (a
j
, b
j
) g (a
i
, b
i
)]
1i<jn
_
a
2
i
b
2
j
+a
2
j
b
2
i
_
.
Therefore, it is enough to prove
2a
i
b
i
a
j
b
j
f (a
i
, b
i
) g (a
j
, b
j
) +f (a
j
, b
j
) g (a
i
, b
i
) (3.67)
a
2
i
b
2
j
+a
2
j
b
2
i
.
Suppose that (3.64) holds. Then (3.66) holds and putting a =
a
i
b
i
, b =
a
j
b
j
in (3.66) and using
(3.63), we get
2
f (a
i
, b
i
)
f (a
j
, b
j
)

a
j
b
j
a
i
b
i
+
f (a
j
, b
j
)
f (a
i
, b
i
)

a
i
b
i
a
j
b
j
a
i
b
j
a
j
b
i
+
a
j
b
i
a
i
b
j
.
Multiplying the last inequality by a
i
b
i
a
j
b
j
and using (3.62), we obtain (3.67).
Remark 3.30. In [8] (see [5, p. 89]) the condition (3.64) is given as
(3.68) f (b, 1) f (a, 1) ,
f (a, 1)
a
2

f (b, 1)
b
2
for a b > 0.
Remark 3.31. O.E. Daykin, C.J. Eliezer and C. Carlitz [8] stated that examples for f, g satis-
fying (3.62) (3.64) were obtained in the literature. The choice f (x, y) = x
2
+ y
2
, g (x, y) =
x
2
y
2
x
2
+y
2
will give the Milnes inequality
(3.69)
_
n
i=1
a
i
b
i
_
2
i=1
_
a
2
i
+b
2
i
_
i=1
a
2
i
b
2
i
a
2
i
+b
2
i
i=1
a
2
i

n
i=1
b
2
i
.
For a different proof of this fact, see Section 3.5.
The choice f (x, y) = x
1+
y
1
, g (x, y) = x
1
y
1+
( [0, 1]) will give the Callebaut
inequality
(3.70)
_
n
i=1
a
i
b
i
_
2
i=1
a
1+
i
b
1
i
n
i=1
a
1
i
b
1+
i

n
i=1
a
2
i

n
i=1
b
2
i
.
3.10. A Renement via Dunkl-Williams Inequality. We will use the following version of
Dunkl-Williams inequality established in 1964 in inner product spaces [9].
Lemma 3.32. Let a, b be two non-null complex numbers. Then
(3.71) |a b|
1
2
(|a| +|b|)
a
|a|

b
|b|
.
44 S.S. DRAGOMIR
Proof. We start with the identity (see also [5, pp. 515 516])
a
|a|

b
|b|
2
=
_
a
|a|

b
|b|
__
a
|a|

b
|b|
_
= 2 2 Re
_
a
|a|

b
|b|
_
=
1
|a| |b|
_
2 |a| |b| 2 Re
_
a
b
__
=
1
|a| |b|
_
2 |a| |b|
_
|a|
2
+|b|
2
|a b|
2
_
=
1
|a| |b|
_
|a b|
2
(|a| |b|)
2
.
Hence
|a b|
2
_
1
2
(|a| +|b|)
_
2
a
|a|

b
|b|
2
=
(|a| |b|)
2
4 |a| |b|
_
(|a| +|b|)
2
|a b|
2
and (3.71) is proved.

Using the above result, we may prove the following renement of the (CBS) inequality for
complex numbers.
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
) are two sequences of nonzero complex
numbers, then
(3.72)
n
k=1
|a
k
|
2
n
k=1
|b
k
|
2
k=1
a
k
b
k
1
8
n
i,j=1
a
i
b
j
a
j
b
i
+
|b
i
|
|a
i
|
a
i
|a
j
|
|b
j
|
b
j

|a
i
|
|b
i
|
b
i
|b
j
|
|a
j
|
a
j
2
0.
Proof. The inequality (3.71) is clearly equivalent to
(3.73) |a b|
2
1
4
a b +
|b|
|a|
a
|a|
|b|
b
2
for any a, b C, a, b = 0.
We know the Lagranges identity for sequences of complex numbers
(3.74)
n
k=1
|a
k
|
2
n
k=1
|b
k
|
2
k=1
a
k
b
k
2
=
1
2
n
i,j=1
| a
i
b
j
a
j
b
i
|
2
.
By (3.73), we have
| a
i
b
j
a
j
b
i
|
2
1
4
a
i
b
j
a
j
b
i
+
|a
j
| |b
i
|
|a
i
| |b
j
|
a
i
b
j

|a
i
| |b
j
|
|a
j
| |b
i
|
a
j
b
i
2
.
Summing over i, j from 1 to n and using the (CBS) inequality for double sums, we deduce
(3.72).
3.11. Some Renements due to Alzer and Zheng. In 1992, H. Alzer [10] presented the fol-
lowing renement of the Cauchy-Schwarz inequality written in the form
(3.75)
_
n
k=1
x
k
y
k
_
2
k=1
y
k
n
k=1
x
2
k
y
k
.
Theorem 3.34. Let x
k
and y
k
(k = 1, . . . , n) be real numbers satisfying 0 = x
0
< x
1

x
2
2

x
n
n
and 0 < y
n
y
n1
y
1
. Then
(3.76)
_
n
k=1
x
k
y
k
_
2
k=1
y
k
n
k=1
_
x
2
k

1
4
x
k1
x
k
_
y
k
,
with equality holding if and only if x
k
= kx
1
(k = 1, . . . , n) and y
1
= = y
n
.
In 1998, Liu Zheng [11] pointed out an error in the proof given in [10], which can be corrected
as shown in [11]. Moreover, Liu Zheng established the following result which sharpens (3.76).
Theorem 3.35. Let x
k
and y
k
(k = 1, . . . , n) be real numbers satisfying 0 < x
1

x
2
2

x
n
n
and 0 < y
n
y
n1
y
1
. Then
(3.77)
_
n
k=1
x
k
y
k
_
2
k=1
y
k
n
k=1
k
y
k
,
with
(3.78)
1
= x
2
1
and
k
=
7k + 1
8k
x
2
k

k
8 (k 1)
x
2
k1
(k 2) .
Equality holds in (3.77) if and only if x
k
= kx
1
(k = 1, . . . , n) and y
1
= = y
n
.
In 1999, H. Alzer improved the above results as follows.
To present his results, we will follow [12].
In order to prove the main result, we need some technical lemmas.
Lemma 3.36. Let x
k
(k = 1, . . . , n) be real numbers such that
0 < x
1

x
2
2

x
n
n
.
Then
(3.79) 2
n
k=1
x
k
(n + 1) x
n
,
k
= kx
1
(k = 1, . . . , n) .
A proof of Lemma 3.36 is given in [10].
Lemma 3.37. Let x
k
0 < x
1

x
2
2

x
n
n
.
Then
(3.80)
_
n
k=1
x
k
_
2
n
n
k=1
3k + 1
4k
x
2
k
,
k
= kx
1
(k = 1, . . . , n) .
46 S.S. DRAGOMIR
Proof. Let
S
n
= S
n
(x
1
, . . . , x
n
) = n
n
k=1
3k + 1
4k
x
2
k

_
n
k=1
x
k
_
2
.
Then we have for n 2 :
S
n
S
n1
=
n1
k=1
3k + 1
4k
x
2
k
2x
n
n1
k=1
x
k
+
3 (n 1)
4
x
2
n
(3.81)
= f (x
n
) , say.
We differentiate with respect to x
n
and use (3.79) and x
n

n
n1
x
n1
. This yields
f
(x
n
) =
3 (n 1)
2
x
n
2
n1
k=1
x
k

n
2
x
n1
> 0
and
f (x
n
) f
_
n
n 1
x
n1
_
(3.82)
=
n
k=1
3k + 1
4k
x
2
k

2n
n 1
x
n1
n1
k=1
x
k
+
3n
2
4 (n 1)
x
2
n1
= T
n1
(x
1
, . . . , x
n1
) , say.
We use induction on n to establish that T
n1
(x
1
, . . . , x
n1
) 0 for n 2. We have T
1
(x
1
) =
0. Let n 3; applying (3.79) we obtain
x
n1
T
n1
(x
1
, . . . , x
n1
) =
3n + 2
2
x
n1
2n
n 1
n1
k=1
x
k
(n 2) (n + 1)
2 (n 1)
x
n1
> 0
and
(3.83) T
n1
(x
1
, . . . , x
n1
) T
n1
_
x
1
, . . . , x
n2
,
n 1
n 2
x
n2
_
.
Using the induction hypothesis T
n2
(x
1
, . . . , x
n2
) 0 and (3.79), we get
(3.84) T
n1
_
x
1
, . . . , x
n2
,
n 1
n 2
x
n2
_
x
n2
n 2
_
(n 1) x
n2
2
n2
k=1
x
k
_
.
From (3.83) and (3.84) we conclude T
n1
(x
1
, . . . , x
n1
) 0 for n 2, so that (3.81) and
(3.82) imply
(3.85) S
n
S
n1
S
2
S
1
= 0.
This proves inequality (3.80). We discuss the cases of equality. A simple calculation reveals
that S
n
(x
1
, 2x
1
, . . . , nx
1
) = 0. We use induction on n to prove the implication
(3.86) S
n
(x
1
, . . . , x
n
) = 0 =x
k
= kx
1
for k = 1, . . . , n.
If n = 1, then (3.86) is obviously true. Next, we assume that (3.86) holds with n 1 instead
of n. Let n 2 and S
n
(x
1
, . . . , x
n
) = 0. Then (3.85) leads to S
n1
(x
1
, . . . , x
n1
) = 0 which
implies x
k
= kx
1
for k = 1, . . . , n 1. Thus, we have S
n
(x
1
, 2x
1
, . . . , (n 1) x
1
, x
n
) = 0
which is equivalent to (x
n
nx
1
) (3x
n
nx
1
) = 0. Since 3x
n
> nx
1
, we get x
n
= nx
1
.
Lemma 3.38. Let x
k
0 < x
1

x
2
2

x
n
n
.
If the natural numbers n and q satisfy n q + 1, then
(3.87) 0 <
_
q
k=1
x
k
_
2
2qx
n
q
k=1
x
k
+
(3n + 1) q
2
4n
x
2
n
.
Proof. We denote the expression on the right-hand side of (3.87) by u(x
n
) . Then we differen-
tiate with respect to x
n
and apply (3.79), x
n

_
n
q
_
x
q
and n q + 1. This yields
1
q
u
(x
n
) =
(3n + 1) q
2n
x
n
2q
q
k=1
x
k
(3n + 1) q
2n
x
n
(q + 1) x
q
3n 2q 1
2
x
q
> 0.
Hence, we get
(3.88) u(x
n
) u
_
n
q
x
q
_
=
(3n + 1) n
4
x
2
q
2nx
q
q
k=1
x
k
+
_
q
k=1
x
k
_
2
.
Let
v (t) =
(3t + 1) t
4
x
q
2t
q
k=1
x
k
and t q + 1;
from (3.79) we conclude that
v
(t) =
6t + 1
4
x
q
2
q
k=1
x
k

2q + 3
4
x
q
> 0.
This implies that the expression on the right-hand side of (3.88) is increasing on [q +1, ) with
respect to n. Since n q + 1, we get from (3.88):
u(x
n
)
(3q + 4) (q + 1)
4
x
2
q
2 (q + 1) x
q
q
k=1
x
k
+
_
q
k=1
x
k
_
2
(3.89)
= P
q
(x
1
, . . . , x
q
) , say.
We use induction on q to show that P
q
(x
1
, . . . , x
q
) > 0 for q 1. We have P
1
(x
1
) =
1
2
x
2
1
. If
P
q1
(x
1
, . . . , x
q1
) > 0, then we obtain for q 2 :
(3.90) P
q
(x
1
, . . . , x
q
) > 2q (x
q1
x
q
)
q1
k=1
x
k

(3q + 1) q
4
x
2
q1
+
q (3q 1)
4
x
2
q
= w(x
q
) , say.
48 S.S. DRAGOMIR
We differentiate with respect to x
q
and use (3.79) and x
q

_
q
q1
_
x
q1
. Then we get
w
(x
q
) = q
_
3q 1
2
x
q
2
q1
k=1
x
k
_
q
2
(q + 1)
2 (q 1)
x
q1
> 0
and
w(x
q
) w
_
q
q 1
x
q1
_
(3.91)
=
q
q 1
x
q1
_
4q
2
q 1
4 (q 1)
x
q1
2
q1
k=1
x
k
_
(3q 1) q
4 (q 1)
x
2
q1
> 0.
From (3.89), (3.90) and (3.91), we obtain u(x
n
) > 0.
We are now in a position to prove the following companion of inequalities (3.76) and (3.77)
(see [12]).
Theorem 3.39. The inequality
(3.92)
_
n
k=1
x
k
y
k
_
2
k=1
y
k
n
k=1
_
+

k
_
y
k
,
holds for all natural numbers n and for all real numbers x
k
and y
k
(k = 1, . . . , n) with
(3.93) 0 < x
1

x
2
2

x
n
n
and 0 < y
n
y
n1
y
1
,
if and only if

3
4
and 1 .
Proof. First, we assume that (3.92) is valid for all n 1 and for all real numbers x
k
and y
k
(k = 1, . . . , n) which satisfy (3.93). We set x
k
= k and y
k
= 1 (k = 1, . . . , n). Then (3.92)
leads to
(3.94) 0
_

3
4
_
2n + + 3
3
2
(n 1) .
This implies
3
4
. And, (3.94) with n = 1 yields + 1.
Now, we suppose that
3
4
and 1 . Then we obtain for k 1 :
+

k
+
1
k

3
4
+
1
4k
,
so that is sufces to show that (3.92) holds with =
3
4
and =
1
4
. Let
F (y
1
, . . . , y
n
) =
n
k=1
y
k
n
k=1
3k + 1
4k
x
2
k
y
k

_
n
k=1
x
k
y
k
_
2
and
F
q
(y) = F (y, . . . , y, y
q+1
, . . . , y
n
) (1 q n 1) .
We shall prove that F
q
is strictly increasing on [y
q+1
, ). Since y
q+1
y
q
, we obtain
(3.95) F
q
(y
q
) F
q
(y
q+1
) = F
q+1
(y
q+1
) (1 q n 1) ,
and Lemma 3.37 imply
F (y
1
, . . . , y
n
) = F
1
(y
1
) F
1
(y
2
) = F
2
(y
2
) F
2
(y
3
)
F
n1
(y
n1
) F
n1
(y)
= y
2
n
_
_
n
n
k=1
3k + 1
4k
x
2
k

_
n
k=1
x
k
_
2
_
_
0.
If F (y
1
, . . . , y
n
) = 0, then we conclude from the strict monotonicity of F
q
and from Lemma
3.37 that y
1
= = y
n
and x
k
= kx
1
(k = 1, . . . , n) .
It remains to show that F
q
is strictly increasing on [y
q+1
, ). Let y y
q+1
; we differentiate
F
q
and apply Lemma 3.37. This yields
F
q
(y) = 2y
_
_
q
q
k=1
3k + 1
4k
x
2
k

_
q
k=1
x
k
_
2
_
_
+q
n
k=q+1
3k + 1
4k
x
2
k
y
k
+
n
k=q+1
y
k
q
k=1
3k + 1
4k
x
2
k
2
n
k=q+1
x
k
y
k
q
k=1
x
k
and
1
2
F
q
(y) = q
q
k=1
3k + 1
4k
x
2
k

_
q
k=1
x
k
_
2
0.
Hence, we have
F
q
(y) F
q
(y
q+1
) (3.96)
=
_
2qy
q+1
+
n
k=q+1
y
k
_
_
_
q
k=1
3k + 1
4k
x
2
k

1
q
_
q
k=1
x
k
_
2
_
_
+
1
q
n
k=q+1
y
k
_
_
_
(3k + 1) q
2
4k
x
2
k
2qx
k
q
i=1
x
i
+
_
q
i=1
x
i
_
2
_
_
_
.
From Lemma 3.37 and Lemma 3.38 we obtain F
q
(y
q+1
) > 0, so that (3.96) implies F
q
(y) > 0
for y y
q+1
. This completes the proof of the theorem.
Remark 3.40. The proof of the theorem reveals that the sign of equality holds in (3.92) (with
=
3
4
and =
1
4
) if and only if x
k
= kx
1
(k = 1, . . . , n) and y
1
= = y
n
.
Remark 3.41. If
k
is dened by (3.78), then we have for k 2 :
k

_
3
4
+
1
4k
_
x
2
k
=
k (k 1)
8
_
_
x
k
k
_
2
_
x
k1
k 1
_
2
_
,
which implies that inequality (3.92) (with =
3
4
and =
1
4
) sharpens (3.77).
50 S.S. DRAGOMIR
Remark 3.42. It is shown in [10] that if a sequence (x
k
) satises x
0
= 0 and 2x
k
x
k1
+x
k+1
(k 1) , then
_
x
k
k
_
is increasing. Hence, inequality (3.92) is valid for all sequences (x
k
) which
are positive and convex.
REFERENCES
[1] S.S. DRAGOMIR AND N.M. IONESCU, Some renements of Cauchy-Buniakowski-Schwartz in-
equality for sequences, Proc. of the Third Symp. of Math. and Its Appl., 3-4 Nov. 1989, Polytechni-
cal Institute of Timi soara, Romania, 7882.
[2] S.S. DRAGOMIR, On Cauchy-Buniakowski-Schwartzs inequality for real numbers (Romanian),
Caiete Metodico Stiintice, No. 57, 1989, pp. 24. Faculty of Mathematics, Timi soara University,
Romania.
[3] S.S. DRAGOMIR, .Z. ARSLANAGI
C AND D.M. MILOEVI
C, On Cauchy-Buniakowski-
Schwartzs inequality for real numbers, Mat. Bilten, 18 (1994), 5762.
[4] N.G. de BRUIJN, Problem 12, Wisk. Opgaven, 21 (1960), 1214.
[5] D.S. MITRINOVI
C, J.E. PE
CARI

[6] M.O. DRMBE, A generalisation of Lagranges identity (Romanian), Gaz. Mat. (Bucharest), 89
(1984), 139-141.
[7] H.W. McLAUGHLIN, Inequalities complementary to the Cauchy-Schwartz inequality for nite
sums of quaternions, renements of the Cauchy-Schwartz inequality for nite sums of real num-
bers; inequalities concerning ratios and differences of generalised means of different order, Mary-
land University, Technical Note BN-454, (1966), 129 pp.
[8] D.E. DAYKIN, C.J. ELIEZER AND C. CARLITZ, Problem 5563, Amer. Math. Monthly, 75 (1968),
p. 198 and 76 (1969), 98100.
[9] C.F. DUNKL AND K.S. WILLIAMS, A simple inequality, Amer. Math. Monthly, 71 (1964), 5354.
[10] H. ALZER, A renement of the Cauchy-Schwartz inequality, J. Math. Anal. Appl., 168 (1992),
596604.
[11] L. ZHENG, Remark on a renement of the Cauchy-Schwartz inequality, J. Math. Anal. Appl., 218
(1998), 1321.
[12] H. ALZER, On the Cauchy-Schwartz inequality, J. Math. Anal. Appl., 234 (1999), 614.
4. FUNCTIONAL PROPERTIES
4.1. A Monotonicity Property. The following result was obtained in [1, Theorem].
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
) be sequences of real numbers and p =
(p
1
, . . . , p
n
) , q = (q
1
, . . . , q
n
) be sequences of nonnegative real numbers such that p
k
q
k
for
any k {1, . . . , n} . Then one has the inequality
(4.1)
_
n
i=1
p
i
a
2
i
_1
2
_
n
i=1
p
i
b
2
i
_1
2
i=1
p
i
a
i
b
i
_
n
i=1
q
i
a
2
i
_1
2
_
n
i=1
q
i
b
2
i
_1
2
i=1
q
i
a
i
b
i
0.
Since p
k
q
k
0, then the (CBS) inequality for the weights r
k
= p
k
q
k
(k {1, . . . , n})
will produce
(4.2)
_
n
k=1
p
k
a
2
k
k=1
q
k
a
2
k
__
n
k=1
p
k
b
2
k
k=1
q
k
b
2
k
_
_
n
k=1
(p
k
q
k
) a
k
b
k
_
2
.
(ac bd)
2
_
a
2
b
2
_ _
c
2
d
2
_
, a, b, c, d R
for the choices
a =
_
n
k=1
p
k
a
2
k
_1
2
, b =
_
n
k=1
q
k
a
2
k
_1
2
, c =
_
n
k=1
p
k
b
2
k
_1
2
and
d =
_
n
k=1
q
k
b
2
k
_1
2
we deduce by (4.2), that
_
n
k=1
p
k
a
2
k
_1
2
_
n
k=1
p
k
b
2
k
_1
2
_
n
k=1
q
k
a
2
k
_1
2
_
n
k=1
q
k
b
2
k
_1
2
k=1
p
k
a
k
b
k

n
k=1
q
k
a
k
b
k
k=1
p
k
a
k
b
k
k=1
q
k
a
k
b
k
proving the desired inequality (4.1).

The following corollary holds [1, Corollary 1].
Corollary 4.2. Let a and

b be as in Theorem 4.1. Denote
S
n
(1) := { x = (x
1
, . . . , x
n
) |0 x
i
1, i {1, . . . , n}} .
52 S.S. DRAGOMIR
Then
(4.3)
_
n
i=1
a
2
i
_1
2
_
n
i=1
b
2
i
_1
2
i=1
a
i
b
i
= sup
xS
n
(1)
_
_
_
n
i=1
x
i
a
2
i
_1
2
_
n
i=1
x
i
b
2
i
_1
2
i=1
x
i
a
i
b
i
_
_
0.
Remark 4.3. The following inequality is a natural particular case that may be obtained from
(4.1) [1, p. 79]
(4.4)
_
n
i=1
a
2
i
_1
2
_
n
i=1
b
2
i
_1
2
i=1
a
i
b
i
_
n
i=1
a
2
i
trig
2
(
i
)
_1
2
_
n
i=1
b
2
i
trig
2
(
i
)
_1
2
i=1
a
i
b
i
trig
2
(
i
)
0,
where trig (x) = sin x or cos x, x R and = (
1
, . . . ,
n
) is a sequence of real numbers.
4.2. A Superadditivity Property in Terms of Weights. Let P
f
(N) be the family of nite
parts of the set of natural numbers N, S (K) the linear space of real or complex numbers, i.e.,
S (K) :=
_
x|x = (x
i
)
iN
, x
i
K, i N
_
and S
+
(R) the family of nonnegative real sequences. Dene the mapping
(4.5) S (p, I, x, y) :=
_
iI
p
i
|x
i
|
2
iI
p
i
|y
i
|
2
_1
2
iI
p
i
x
i
y
i
,
where p S
+
(R) , I P
f
(N) and x, y S (K) .
The following superadditivity property in terms of weights holds [2, p. 16].
Theorem 4.4. For any p, q S
+
(R) , I P
f
(N) and x, y S (K) we have
(4.6) S (p +q, I, x, y) S (p, I, x, y) +S (q, I, x, y) 0.
Proof. Using the (CBS) inequality for real numbers
(4.7)
_
a
2
+b
2
_1
2
_
c
2
+d
2
_1
2
ac +bd; a, b, c, d 0,
we have
S (p, I, x, y) =
_
iI
p
i
|x
i
|
2
+
iI
q
i
|x
i
|
2
_1
2
_
iI
p
i
|y
i
|
2
+
iI
q
i
|y
i
|
2
_1
2
iI
p
i
x
i
y
i
+
iI
q
i
x
i
y
i
iI
p
i
|x
i
|
2
_1
2
_
iI
p
i
|y
i
|
2
_1
2
+
_
iI
q
i
|x
i
|
2
_1
2
_
iI
q
i
|y
i
|
2
_1
2
iI
p
i
x
i
y
i
iI
q
i
x
i
y
i
= S (p, I, x, y) +S (q, I, x, y) ,
The following corollary concerning the monotonicity of S (, I, x, y) also holds [2, p. 16].
Corollary 4.5. For any p, q S
+
(R) with p q and I P
f
(N) , x, y S (K) one has the
inequality:
(4.8) S (p, I, x, y) S (q, I, x, y) 0.
Proof. Using Theorem 4.4, we have
S (p, I, x, y) = S ((p q) +q, I, x, y) S (p q, I, x, y) +S (q, I, x, y)
giving
S (p, I, x, y) S (q, I, x, y) S (p q, I, x, y) 0
Remark 4.6. The following inequalities follow by the above results [2, p. 17].
(1) Let
i
R (i {1, . . . , n}) and x
i
, y
i
K(i {1, . . . , n}) . Then one has the inequal-
ity:
(4.9)
_
n
i=1
|x
i
|
2
n
i=1
|y
i
|
2
_1
2
i=1
x
i
y
i
_
n
i=1
|x
i
|
2
sin
2
i
n
i=1
|y
i
|
2
sin
2
i
_1
2
i=1
x
i
y
i
sin
2
+
_
n
i=1
|x
i
|
2
cos
2
i
n
i=1
|y
i
|
2
cos
2
i
_1
2
i=1
x
i
y
i
cos
2
0.
54 S.S. DRAGOMIR
(2) Denote S
n
(1) := {p S
+
(R) |p
i
1 for all i {1, . . . , n}} . Then for all x, y
S (K) one has the bound (see also Corollary 4.2):
0
_
n
i=1
|x
i
|
2
n
i=1
|y
i
|
2
_1
2
i=1
x
i
y
i
(4.10)
= sup
pS
n
(1)
_
_
_
_
n
i=1
p
i
|x
i
|
2
n
i=1
p
i
|y
i
|
2
_1
2
i=1
p
i
x
i
y
i
_
_
_
.
4.3. The Superadditivity as an Index Set Mapping. We assume that we are under the hy-
pothesis and notations in Section 4.2. Reconsider the functional S (, , , ) : S
+
(R) P
f
(N)
S (K) S (K) R,
(4.11) S (p, I, x, y) :=
_
iI
p
i
|x
i
|
2
iI
p
i
|y
i
|
2
_1
2
iI
p
i
x
i
y
i
.
The following superadditivity property as an index set mapping holds [2].
Theorem 4.7. For any I, J P
f
(N) \ {} with I J = , one has the inequality
(4.12) S (p, I J, x, y) S (p, I, x, y) +S (p, J, x, y) 0.
Proof. Using the elementary inequality for real numbers
(4.13)
_
a
2
+b
2
_1
2
_
c
2
+d
2
_1
2
ac +bd; a, b, c, d 0,
we have
S (p, I J, x, y) =
_
iI
p
i
|x
i
|
2
+
jJ
p
j
|x
j
|
2
_1
2
_
iI
p
i
|y
i
|
2
+
jJ
p
j
|y
j
|
2
_1
2
iI
p
i
x
i
y
i
+
jJ
p
j
x
j
y
j
iI
p
i
|x
i
|
2
_1
2
_
iI
p
i
|y
i
|
2
_1
2
+
_
jJ
p
j
|x
j
|
2
_1
2
_
jJ
p
j
|y
j
|
2
_1
2
iI
p
i
x
i
y
i
jJ
p
j
x
j
y
j
= S (p, I, x, y) +S (p, J, x, y)
The following corollary concerning the monotonicity of S (p, , x, y) as an index set mapping
also holds [2, p. 16].
Corollary 4.8. For any I, J P
f
(N) with I J = , one has
(4.14) S (p, I, x, y) S (p, J, x, y) 0.
Proof. Using Theorem 4.7, we may write
S (p, I, x, y) = S (p, (I\J) J, x, y) S (p, I\J, x, y) +S (p, J, x, y)
giving
S (p, I, x, y) S (p, J, x, y) S (p, I\J, x, y) 0
which proves the desired inequality (4.14).
Remark 4.9. The following inequalities follow by the above results [2, p. 17].
(1) Let p
i
0 (i {1, . . . , 2n}) and x
i
, y
i
K (i {1, . . . , 2n}) . Then we have the
inequality
_
2n
i=1
p
i
|x
i
|
2
2n
i=1
p
i
|y
i
|
2
_1
2
2n
i=1
p
i
x
i
y
i
(4.15)
_
n
i=1
p
2i
|x
2i
|
2
n
i=1
p
2i
|y
2i
|
2
_1
2
i=1
p
2i
x
i
y
i
+
_
n
i=1
p
2i1
|x
2i1
|
2
n
i=1
p
2i1
|y
2i1
|
2
_1
2
i=1
p
2i1
x
2i1
y
2i1
0.
(2) We have the bound
(4.16)
_
n
i=1
p
i
|x
i
|
2
n
i=1
p
i
|y
i
|
2
_1
2
i=1
p
i
x
i
y
i
= sup
I{1,...,n}
I=
_
_
_
iI
p
i
|x
i
|
2
iI
p
i
|y
i
|
2
_1
2
iI
p
i
x
i
y
i
_
_
0.
(3) Dene the sequence
(4.17) S
n
:=
_
n
i=1
p
i
|x
i
|
2
n
i=1
p
i
|y
i
|
2
_1
2
i=1
p
i
x
i
y
i
0
where p = (p
i
)
iN
S
+
(R) , x = (x
i
)
iN
, y = (y
i
)
iN
S (K) . Then S
n
is monoton-
tic nondecreasing and we have the following lower bound
S
n
max
1i,jn
_
_
p
i
|x
i
|
2
+p
j
|x
j
|
2
_
1
2
_
p
i
|y
i
|
2
+p
j
|y
j
|
2
_
1
2
(4.18)
|p
i
x
i
y
i
+p
j
x
j
y
j
|
_
0.
4.4. Strong Superadditivity in Terms of Weights. With the notations in Section 4.2, dene
the mapping
(4.19)

S (p, I, x, y) :=
iI
p
i
|x
i
|
2
iI
p
i
|y
i
|
2
iI
p
i
x
i
y
i
2
,
where p S
+
(R) , I P
f
(N) and x, y S (K) .
56 S.S. DRAGOMIR
Denote also by
,H
the weighted Euclidean norm
(4.20) x
,H
:=
_
iH
i
|x
i
|
2
_1
2
, S
+
(R) , H P
f
(N) .
The following strong superadditivity property in terms of weights holds [2, p. 18].
+
(R) , I P
f
(N) and x, y S (K) we have
(4.21)

S (p +q, I, x, y)

S (p, I, x, y)

S (q, I, x, y)
_
_
det
_
_
x
p,I
y
p,I
x
q,I
y
q,I
_
_
_
_
2
0.
Proof. We have
S (p +q, I, x, y) =
_
iI
p
i
|x
i
|
2
+
iI
q
i
|x
i
|
2
__
iI
p
i
|y
i
|
2
+
iI
q
i
|y
i
|
2
_
(4.22)
iI
p
i
x
i
y
i
+
iI
q
i
x
i
y
i
iI
p
i
|x
i
|
2
iI
p
i
|y
i
|
2
+
iI
q
i
|x
i
|
2
iI
q
i
|y
i
|
2
iI
p
i
x
i
y
i
iI
q
i
x
i
y
i
_
2
=

S (p, I, x, y) +

S (q, I, x, y) +
iI
p
i
|x
i
|
2
iI
q
i
|y
i
|
2
+
iI
q
i
|x
i
|
2
iI
p
i
|y
i
|
2
2
iI
p
i
x
i
y
i
iI
q
i
x
i
y
i
.
By (CBS) inequality, we have
iI
p
i
x
i
y
i
iI
q
i
x
i
y
i
iI
p
i
|x
i
|
2
iI
p
i
|y
i
|
2
iI
q
i
|x
i
|
2
iI
q
i
|y
i
|
2
_1
2
and thus
(4.23)
iI
p
i
|x
i
|
2
iI
q
i
|y
i
|
2
+
iI
q
i
|x
i
|
2
iI
p
i
|y
i
|
2
2
iI
p
i
x
i
y
i
iI
q
i
x
i
y
i
_
_
_
iI
p
i
|x
i
|
2
_1
2
_
iI
q
i
|y
i
|
2
_1
2
iI
q
i
|x
i
|
2
_1
2
_
iI
p
i
|y
i
|
2
_1
2
_
_
2
.
Utilising (4.22) and (4.23) we deduce the desired inequality (4.21).
The following corollary concerning a strong monotonicity result also holds [2, p. 18].
Corollary 4.11. For any p, q S
+
(R) with p q one has the inequality:
(4.24)

S (p, I, x, y)

S (q, I, x, y)
_
_
det
_
_
x
q,I
y
q,I
x
pq,I
y
pq,I
_
_
_
_
2
0.
Remark 4.12. The following renement of the (CBS) inequality is a natural consequence of
(4.21) [2, p. 19]
(4.25)
iI
|x
i
|
2
iI
|y
i
|
2
iI
x
i
y
i
iI
|x
i
|
2
sin
2
iI
|y
i
|
2
sin
2
iI
x
i
y
i
sin
2
2
+
iI
|x
i
|
2
cos
2
iI
|y
i
|
2
cos
2
iI
x
i
y
i
cos
2
2
+
_
_
_
_
_
_
_
det
_
_
_
iI
|x
i
|
2
sin
2
i
_1
2
_
iI
|y
i
|
2
sin
2
i
_1
2
_
iI
|x
i
|
2
cos
2
i
_1
2
_
iI
|y
i
|
2
cos
2
i
_1
2
_
_
_
_
_
_
_
_
_
2
0.
where
i
R, i I.
4.5. Strong Superadditivity as an Index Set Mapping. We assume that we are under the hy-
pothesis and notations in Section 4.2. Reconsider the functional

S (, , , ) : S
+
(R) P
f
(N)
S (K) S (K) R,
(4.26)

S (p, I, x, y) :=
iI
p
i
|x
i
|
2
iI
p
i
|y
i
|
2
iI
p
i
x
i
y
i
2
.
The following strong supperadditivity property as an index set mapping holds [2, p. 18].
Theorem 4.13. For any p S
+
(R) , I, J P
f
(N) \ {} with I J = and x, y S (K) ,
we have
(4.27)

S (p, I J, x, y)

S (p, I, x, y)

S (p, J, x, y)
_
_
det
_
_
x
p,I
y
p,I
x
p,J
y
p,J
_
_
_
_
2
0.
58 S.S. DRAGOMIR
Proof. We have
S (p, I J, x, y) (4.28)
=
_
iI
p
i
|x
i
|
2
+
jJ
p
j
|x
j
|
2
__
iI
p
i
|y
i
|
2
+
jJ
p
j
|y
j
|
2
_
iI
p
i
x
i
y
i
+
jI
p
j
x
j
y
j
iI
p
i
|x
i
|
2
iI
p
i
|y
i
|
2
+
jJ
p
j
|x
j
|
2
jJ
p
j
|y
j
|
2
+
iI
p
i
|x
i
|
2
jJ
p
j
|y
j
|
2
+
iI
p
i
|y
i
|
2
jJ
p
j
|x
j
|
2
iI
p
i
x
i
y
i
jI
p
j
x
j
y
j
_
2
=

S (p, I, x, y) +

S (p, J, x, y) +
iI
p
i
|x
i
|
2
jJ
p
j
|y
j
|
2
+
iI
p
i
|y
i
|
2
jJ
p
j
|x
j
|
2
2
iI
p
i
x
i
y
i
jI
p
j
x
j
y
j
.
By the (CBS) inequality, we have
iI
p
i
x
i
y
i
jI
p
j
x
j
y
j
iI
p
i
|x
i
|
2
iI
p
i
|y
i
|
2
jJ
p
j
|x
j
|
2
jJ
p
j
|y
j
|
2
_1
2
and thus
(4.29)
iI
p
i
|x
i
|
2
jJ
p
j
|y
j
|
2
+
iI
p
i
|y
i
|
2
jJ
p
j
|x
j
|
2
2
iI
p
i
x
i
y
i
jI
p
j
x
j
y
j
_
_
_
iI
p
i
|x
i
|
2
_1
2
_
jJ
p
j
|y
j
|
2
_1
2
iI
p
i
|y
i
|
2
_1
2
_
jJ
p
j
|x
j
|
2
_1
2
_
_
2
.
If we use now (4.28) and (4.29), we may deduce the desired inequality (4.27).
The following corollary concerning strong monotonicity also holds [2, p. 18].
Corollary 4.14. For any I, J P
f
(N) \ {} with I J one has the inequality
(4.30)

S (p, I, x, y)

S (p, J, x, y)
_
_
det
_
_
x
p,J
y
p,J
x
p,I\J
y
p,I\J
_
_
_
_
2
0.
Remark 4.15. The following renement of the (CBS) inequality is a natural consequence of
(4.27) [2, p. 19].
Suppose p
i
0, i {1, . . . , 2n} and x
i
, y
i
K, i {1, . . . , 2n} . Then we have the
inequality
(4.31)
2n
i=1
p
i
|x
i
|
2
2n
i=1
p
i
|y
i
|
2
2n
i=1
p
i
x
i
y
i
i=1
p
2i
|x
2i
|
2
n
i=1
p
2i
|y
2i
|
2
i=1
p
2i
x
2i
y
2i
2
+
n
i=1
p
2i1
|x
2i1
|
2
n
i=1
p
2i1
|y
2i1
|
2
i=1
p
2i1
x
2i1
y
2i1
2
+
_
_
_
_
_
_
_
det
_
_
_
n
i=1
p
2i
|x
2i
|
2
_1
2
_
n
i=1
p
2i
|y
2i
|
2
_1
2
_
n
i=1
p
2i1
|x
2i1
|
2
_1
2
_
n
i=1
p
2i1
|y
2i1
|
2
_1
2
_
_
_
_
_
_
_
_
_
2
0.
4.6. Another Superadditivity Property. Let P
f
(N) be the family of nite parts of the set of
natural numbers, S (R) the linear space of real sequences and S
+
(R) the family of nonnegative
real sequences.
Consider the mapping C : S
+
(R) P
f
(N) S (R) S (R) R
(4.32) C
_
p, I, a, b
_
:=
iI
p
i
a
2
i
iI
p
i
b
2
i

_
iI
p
i
a
i
b
i
_
2
.
The following identity holds [3, p. 115].
Lemma 4.16. For any p, q S
+
(R) one has
(4.33) C
_
p +q, I, a, b
_
= C
_
p, I, a, b
_
+C
_
q, I, a, b
_
+
(i,j)II
p
i
q
j
(a
i
b
j
a
j
b
i
)
2
.
Proof. Using the well-known Lagranges identity, we have
(4.34) C
_
p, I, a, b
_
=
1
2
(i,j)II
p
i
p
j
(a
i
b
j
a
j
b
i
)
2
.
Thus
C
_
p +q, I, a, b
_
=
1
2
(i,j)II
(p
i
+q
i
) (p
j
+q
j
) (a
i
b
j
a
j
b
i
)
2
=
1
2
(i,j)II
p
i
p
j
(a
i
b
j
a
j
b
i
)
2
+
1
2
(i,j)II
q
i
q
j
(a
i
b
j
a
j
b
i
)
2
+
1
2
(i,j)II
p
i
q
j
(a
i
b
j
a
j
b
i
)
2
+
1
2
(i,j)II
p
j
q
i
(a
i
b
j
a
j
b
i
)
2
= C
_
p, I, a, b
_
+C
_
q, I, a, b
_
+
(i,j)II
p
i
q
j
(a
i
b
j
a
j
b
i
)
2
60 S.S. DRAGOMIR
since, by symmetry,
(i,j)II
p
i
q
j
(a
i
b
j
a
j
b
i
)
2
=
(i,j)II
p
j
q
i
(a
i
b
j
a
j
b
i
)
2
.
Consider the following mapping:

D
_
p, I, a, b
_
:=
_
C
_
p, I, a, b
_
1
2
=
_
_
iI
p
i
a
2
i
iI
p
i
b
2
i

_
iI
p
i
a
i
b
i
_
2
_
_
1
2
.
The following result has been obtained in [4, p. 88] as a particular case of a more general result
holding in inner product spaces.
+
(R) , I P
f
(N) and a, b S (R) , we have the superaddi-
tive property
(4.35) D
_
p +q, I, a, b
_
D
_
p, I, a, b
_
+D
_
q, I, a, b
_
0.
Proof. We will give here an elementary proof following the one in [3, p. 116 p. 117].
By Lemma 4.16, we obviously have
(4.36) D
2
_
p +q, I, a, b
_
= D
2
_
p, I, a, b
_
+D
2
_
q, I, a, b
_
+
(i,j)II
p
i
q
j
(a
i
b
j
a
j
b
i
)
2
.
We claim that
(4.37)
(i,j)II
p
i
q
j
(a
i
b
j
a
j
b
i
)
2
2D
_
p, I, a, b
_
D
_
q, I, a, b
_
.
Taking the square in both sides of (4.37), we must prove that
(4.38)
_
iI
p
i
a
2
i
iI
q
i
b
2
i
+
iI
q
i
a
2
i
iI
p
i
b
2
i
2
iI
p
i
a
i
b
i
iI
q
i
a
i
b
i
_
2
4
_
_
iI
p
i
a
2
i
iI
p
i
b
2
i

_
iI
p
i
a
i
b
i
_
2
_
_
_
_
iI
q
i
a
2
i
iI
q
i
b
2
i

_
iI
q
i
a
i
b
i
_
2
_
_
.
Let us denote
a :=
_
iI
p
i
a
2
i
_1
2
, x :=
_
iI
q
i
a
2
i
_1
2
, b :=
_
iI
p
i
b
2
i
_1
2
,
y :=
_
iI
q
i
b
2
i
_1
2
, c :=
iI
p
i
a
i
b
i
, z :=
iI
q
i
a
i
b
i
.
With these notations (4.38) may be written in the following form
(4.39)
_
a
2
y
2
+b
2
x
2
2cz
_
2
4
_
a
2
b
2
c
2
_ _
x
2
y
2
z
2
_
.
_
m
2
n
2
_ _
p
2
q
2
_
(mp nq)
2
, m, n, p, q R
we may state that
(4.40) 4 (abxy cz)
2
4
_
a
2
b
2
c
2
_ _
x
2
y
2
z
2
_
0.
Since, by the (CBS) inequality, we observe that abxy |cz| |cz| , we can state that
a
2
y
2
+b
2
x
2
2cz 2 (abxy cz) 0
giving
(4.41)
_
a
2
y
2
+b
2
x
2
2cz
_
2
4 (abxy cz)
2
.
Utilizing (4.40) and (4.41) we deduce the inequality (4.39), and (4.37) is proved.
Finally, by (4.36) and (4.37) we have
D
2
_
p +q, I, a, b
_
_
D
_
p, I, a, b
_
+D
_
q, I, a, b
_
2
,
i.e., the superadditivity property (4.35).
Remark 4.18. The following renement of the (CBS) inequality holds [4, p. 89]
(4.42)
_
_
iI
a
2
i
iI
b
2
i

_
iI
a
i
b
i
_
2
_
_
1
2
_
_
iI
a
2
i
sin
2
iI
b
2
i
sin
2
iI
a
i
b
i
sin
2
i
_
2
_
_
1
2
+
_
_
iI
a
2
i
cos
2
iI
b
2
i
cos
2
iI
a
i
b
i
cos
2
i
_
2
_
_
1
2
0
for any
i
R, i {1, . . . , n} .
4.7. The Case of Index Set Mapping. Assume that we are under the hypothesis and notations
in Section 4.6. Reconsider the functional C : S
+
(R) P
f
(N) S (R) S (R) R given by
(4.43) C
_
p, I, a, b
_
:=
iI
p
i
a
2
i
iI
p
i
b
2
i

_
iI
p
i
a
i
b
i
_
2
.
The following identity holds.
Lemma 4.19. For any I, J P
f
(N) \ {} with I J = one has the identity:
(4.44) C
_
p, I J, a, b
_
= C
_
p, I, a, b
_
+C
_
p, J, a, b
_
+
(i,j)IJ
p
i
p
j
(a
i
b
j
a
j
b
i
)
2
.
Proof. Using Lagranges identity [5, p. 84], we may state
(4.45) C
_
p, K, a, b
_
=
1
2
(i,j)KK
p
i
p
j
(a
i
b
j
a
j
b
i
)
2
, K P
f
(N) \ {} .
62 S.S. DRAGOMIR
Thus
C
_
p, I J, a, b
_
=
1
2
(i,j)(IJ)(IJ)
p
i
p
j
(a
i
b
j
a
j
b
i
)
2
=
1
2
(i,j)II
p
i
p
j
(a
i
b
j
a
j
b
i
)
2
+
1
2
(i,j)IJ
p
i
p
j
(a
i
b
j
a
j
b
i
)
2
+
1
2
(i,j)JI
p
i
p
j
(a
i
b
j
a
j
b
i
)
2
+
1
2
(i,j)JJ
p
i
p
j
(a
i
b
j
a
j
b
i
)
2
= C
_
p, I, a, b
_
+C
_
p, J, a, b
_
+
(i,j)IJ
p
i
p
j
(a
i
b
j
a
j
b
i
)
2
since, by symmetry,
(i,j)IJ
p
i
p
j
(a
i
b
j
a
j
b
i
)
2
=
(i,j)JI
p
i
p
j
(a
i
b
j
a
j
b
i
)
2
.
Now, if we consider the mapping

D
_
p, I, a, b
_
:=
_
C
_
p, I, a, b
_
1
2
=
_
_
iI
p
i
a
2
i
iI
p
i
b
2
i

_
iI
p
i
a
i
b
i
_
2
_
_
1
2
,
then the following superadditivity property as an index set mapping holds:
Theorem 4.20. For any I, J P
f
(N) \ {} with I J = one has
(4.46) D
_
p, I J, a, b
_
D
_
p, I, a, b
_
+D
_
p, J, a, b
_
0.
Proof. By Lemma 4.19, we have
(4.47) D
2
_
p, I J, a, b
_
= D
2
_
p, I, a, b
_
+D
2
_
p, J, a, b
_
+
(i,j)IJ
p
i
p
j
(a
i
b
j
a
j
b
i
)
2
To prove (4.46) it is sufcient to show that
(4.48)
(i,j)IJ
p
i
p
j
(a
i
b
j
a
j
b
i
)
2
2D
_
p, I, a, b
_
D
_
p, J, a, b
_
.
Taking the square in (4.48), we must demonstrate that
_
iI
p
i
a
2
i
jJ
p
j
b
2
j
+
jJ
p
j
a
2
j
iI
p
i
b
2
i
2
iI
p
i
a
i
b
i
jJ
p
j
a
j
b
j
_
2
4
_
_
iI
p
i
a
2
i
iI
p
i
b
2
i

_
iI
p
i
a
i
b
i
_
2
_
_
_
_
jJ
p
j
a
2
j
jJ
p
j
b
2
j

_
jJ
p
j
a
j
b
j
_
2
_
_
.
If we denote
a :=
_
iI
p
i
a
2
i
_1
2
, x :=
_
jJ
p
j
a
2
j
_1
2
, b :=
_
iI
p
i
b
2
i
_1
2
,
y :=
_
jJ
p
j
b
2
j
_1
2
, c :=
iI
p
i
a
i
b
i
, z :=
jJ
p
j
a
j
b
j
,
then we need to prove
(4.49)
_
a
2
y
2
+b
2
x
2
2cz
_
2
4
_
a
2
b
2
c
2
_ _
x
2
y
2
z
2
_
,
which has been shown in Section 4.6.
This completes the proof.
Remark 4.21. The following renement of the (CBS) inequality holds
_
_
2n
i=1
p
i
a
2
i
2n
i=1
p
i
b
2
i

_
2n
i=1
p
i
a
i
b
i
_
2
_
_
1
2
_
_
n
i=1
p
2i
a
2
2i
n
i=1
p
2i
b
2
2i
_
n
i=1
p
2i
a
2i
b
2i
_
2
_
_
1
2
+
_
_
n
i=1
p
2i1
a
2
2i1
n
i=1
p
2i1
b
2
2i1
_
n
i=1
p
2i1
a
2i1
b
2i1
_
2
_
_
1
2
0.
4.8. Supermultiplicity in Terms of Weights. Denote by S
+
(R) the set of nonnegative se-
quences. Assume that A : S
+
(R) R is additive on S
+
(R) , i.e.,
(4.50) A(p +q) = A(p) +A(q) , p, q S
+
(R)
and L : S
+
(R) R is superadditive on S
+
(R) , i.e.,
(4.51) L(p +q) L(p) +L(q) , p, q S
+
(R) .
Dene the following associated functionals
(4.52) F (p) :=
L(p)
A(p)
and H (p) := [F (p)]
A(p)
.
The following result holds [3, Theorem 2.1].
Lemma 4.22. With the above assumptions, we have
(4.53) H (p +q) H (p) H (q) ;
for any p, q S
+
(R) , i.e., H () is supermultiplicative on S
+
(R) .
Using the well-known arithmetic mean-geometric mean inequality for real numbers
(4.54)
x +y
+
x
+
y
+
64 S.S. DRAGOMIR
for any x, y 0 and , 0 with + > 0, we have successively
F (p +q) =
L(p +q)
A(p +q)
(4.55)
=
L(p +q)
A(p) +A(q)
L(p) +L(q)
A(p) +A(q)
=
A(p)
L(p)
A(p)
+A(q)
L(q)
A(q)
A(p) +A(q)
=
A(p) F (p) +A(q) F (q)
A(p) +A(q)
[F (p)]
A(p)
A(p)+A(q)
[F (q)]
A(q)
A(p)+A(q)
for all p, q S
+
(R) . However, A(p) + A(q) = A(p +q) , and thus (4.55) implies the
desired inequality (4.53).
We are now able to point out the following inequality related to the (CBS) inequality.
The rst result is incorporated in the following theorem [3, p. 115].
+
(R) , and a, b S (R) , one has the inequality
(4.56)
_
1
P
I
+Q
I
_
iI
(p
i
+q
i
) a
2
i
iI
(p
i
+q
i
) b
2
i

_
iI
(p
i
+q
i
) a
i
b
i
_
2
_
_
_
_
_
P
I
+Q
I
_
_
_
1
P
I
_
_
iI
p
i
a
2
i
iI
p
i
b
2
i

_
iI
p
i
a
i
b
i
_
2
_
_
_
_
_
P
I
_
_
_
1
Q
I
_
_
iI
q
i
a
2
i
iI
q
i
b
2
i

_
iI
q
i
a
i
b
i
_
2
_
_
_
_
_
Q
I
> 0,
where P
I
:=
iI
p
i
> 0, Q
I
:=
iI
q
i
> 0.
Proof. Consider the functionals
A(p) :=
iI
p
i
= P
I
;
C (p) :=
iI
p
i
a
2
i
iI
p
i
b
2
i

_
iI
p
i
a
i
b
i
_
2
.
Then A() is additive and C () is superadditive (see for example Lemma 4.16) on S
+
(R) .
Applying Lemma 4.22 we deduce the desired inequality (4.56).
The following renement of the (CBS) inequality holds.
Corollary 4.24. For any a, b, S (R) , one has the inequality
(4.57)
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
1
_
1
n
n
i=1
sin
2
i
_ 1
n
n
i=1
sin
2
i
_
1
n
n
i=1
cos
2
i
_ 1
n
n
i=1
cos
2
_
_
n
i=1
a
2
i
sin
2
i
n
i=1
b
2
i
sin
2
_
n
i=1
a
i
b
i
sin
2
i
_
2
_
_
1
n
n
i=1
sin
2
_
_
n
i=1
a
2
i
cos
2
i
n
i=1
b
2
i
cos
2
_
n
i=1
a
i
b
i
cos
2
i
_
2
_
_
1
n
n
i=1
cos
2
i
0.
The following result holds [3, p. 116].
+
(4.58)
_
_
_
_
1
P
I
+Q
I
iI
(p
i
+q
i
) a
2
i
_1
2
_
1
P
I
+Q
I
iI
(p
i
+q
i
) b
2
i
_1
2
1
P
I
+Q
I
iI
(p
i
+q
i
) a
i
b
i
_
P
I
+Q
I
_
_
_
1
P
I
iI
p
i
a
2
i
_1
2
_
1
P
I
iI
p
i
b
2
i
_1
2
1
P
I
iI
p
i
a
i
b
i
_
_
P
I
_
_
_
1
Q
I
iI
q
i
a
2
i
_1
2
_
1
Q
I
iI
q
i
b
2
i
_1
2
1
Q
I
iI
q
i
a
i
b
i
_
_
Q
I
0.
Proof. Follows by Lemma 4.22 on taking into account that the functional
B(p) :=
_
iI
p
i
a
2
i
_1
2
_
iI
p
i
b
2
i
_1
2
iI
p
i
a
i
b
i
is superadditive on S
+
(R) (see Section 4.2).
66 S.S. DRAGOMIR
(4.59)
_
n
i=1
a
2
i
_1
2
_
n
i=1
b
2
i
_1
2
i=1
a
i
b
i
1
_
1
n
n
i=1
sin
2
i
_ 1
n
n
i=1
sin
2
1
_
1
n
n
i=1
cos
2
i
_ 1
n
n
i=1
cos
2
_
_
_
n
i=1
a
2
i
sin
2
i
_1
2
_
n
i=1
b
2
i
sin
2
i
_1
2
i=1
a
i
b
i
sin
2
_
_
1
n
n
i=1
sin
2
_
_
_
n
i=1
a
2
i
cos
2
i
_1
2
_
n
i=1
b
2
i
cos
2
i
_1
2
i=1
a
i
b
i
cos
2
_
_
1
n
n
i=1
cos
2
i
0.
Finally, we may also state [3, p. 117].
+
(4.60)
_
1
P
I
+Q
I
iI
(p
i
+q
i
) a
2
i

1
P
I
+Q
I
iI
(p
i
+q
i
) b
2
i
_
1
P
I
+Q
I
iI
(p
i
+q
i
) a
i
b
i
_
2
_
_
P
I
+Q
I
2
_
_
1
P
I
iI
p
i
a
2
i

1
P
I
iI
p
i
b
2
i

_
1
P
I
iI
p
i
a
i
b
i
_
2
_
_
P
I
2
_
_
1
Q
I
iI
q
i
a
2
i

1
Q
I
iI
q
i
b
2
i

_
1
Q
I
iI
q
i
a
i
b
i
_
2
_
_
Q
I
2
.
D(p) :=
_
_
n
i=1
p
i
a
2
i
n
i=1
p
i
b
2
i

_
n
i=1
p
i
a
i
b
i
_
2
_
_
1
2
is superadditive on S
+
(R) (see Section 4.6).
The following corollary also holds.
(4.61)
_
_
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
_
_
1
2
1
_
1
n
n
i=1
sin
2
i
_ 1
n
n
i=1
sin
2
1
_
1
n
n
i=1
cos
2
i
_ 1
n
n
i=1
cos
2
_
_
n
i=1
a
2
i
sin
2
i
n
i=1
b
2
i
sin
2
_
n
i=1
a
i
b
i
sin
2
i
_
2
_
_
1
2n
n
i=1
sin
2
_
_
n
i=1
a
2
i
cos
2
i
n
i=1
b
2
i
cos
2
_
n
i=1
a
i
b
i
cos
2
i
_
2
_
_
1
2n
n
i=1
cos
2
i
0.
4.9. Supermultiplicity as an Index Set Mapping. Denote by P
f
(N) the set of all nite parts
of the natural number set N and assume that B : P
f
(N) R is set-additive on P
f
(N) , i.e.,
(4.62) B(I J) = B(I) +B(J) for any I, J P
f
(N) , I J = ,
and G : P
f
(N) R is set-superadditive on P
f
(N) , i.e.,
(4.63) G(I J) G(I) +G(J) for any I, J P
f
(N) , I J = .
We may dene the following associated functionals
(4.64) M (I) :=
G(I)
B(I)
and N (I) := [M (I)]
A(I)
.
With these notations we may prove the following lemma that is interesting in itself as well.
Lemma 4.29. Under the above assumptions one has
(4.65) N (I J) N (I) N (J)
for any I, J P
f
(N) \ {} with I J = , i.e., N () is set-supermultiplicative on P
f
(N) .
Proof. Using the arithmetic mean geometric mean inequality
(4.66)
x +y
+
x
+
y
+
68 S.S. DRAGOMIR
for any x, y 0 and , 0 with + > 0, we have successively for I, J P
f
(N) \ {}
with I J = that
M (I J) =
G(I J)
B(I J)
(4.67)
=
G(I J)
B(I) +B(J)
G(I) +G(J)
B(I) +B(J)
=
B(I)
G(I)
B(I)
+B(J)
G(J)
B(J)
B(I) +B(J)
=
B(I) M (I) +B(J) M (J)
B(I) +B(J)
(M (I))
B(I)
B(I)+B(J)
(M (J))
B(J)
B(I)+B(J)
.
Since B(I) +B(J) = B(I J) , we deduce by (4.67) the desired inequality (4.65).
Now, we are able to point out some set-superadditivity properties for some functionals asso-
ciates to the (CBS) inequality.
The rst result is embodied in the following theorem.
Theorem 4.30. If a, b S (R) , p S
+
(R) and I, J P
f
(N) \ {} so that I J = , then
(4.68)
_
_
_
1
P
IJ
_
_

kIJ
p
k
a
2
k
kIJ
p
k
b
2
k

_

kIJ
p
k
a
k
b
k
_
2
_
_
_
_
_
P
IJ
_
_
_
1
P
I
_
_
iI
p
i
a
2
i
iI
p
i
b
2
i

_
iI
p
i
a
i
b
i
_
2
_
_
_
_
_
P
I
_
_
_
1
P
J
_
_
jJ
p
j
a
2
j
jJ
p
j
b
2
j

_
jJ
p
j
a
j
b
j
_
2
_
_
_
_
_
P
J
,
when P
J
:=
jJ
p
j
.
Proof. Consider the functionals
B(I) :=
iI
p
i
;
G(I) :=
iI
p
i
a
2
i
iI
p
i
b
2
i

_
iI
p
i
a
i
b
i
_
2
.
The functional B() is obviously set-additive and (see Section 4.7) the functional G() is set-
superadditive. Applying Lemma 4.29 we then deduce the desired inequality (4.68).
The following corollary is a natural application.
Corollary 4.31. If a, b S (R) and p S
+
(R) , then for any n 1 one has the inequality
(4.69)
_
_
_
1
P
2n
_
_
2n
i=1
p
i
a
2
i
2n
i=1
p
i
b
2
i

_
2n
i=1
p
i
a
i
b
i
_
2
_
_
_
_
_
P
2n
_
_
_
1
n
i=1
p
2i
_
_
n
i=1
p
2i
a
2
2i
n
i=1
p
2i
b
2
2i
_
n
i=1
p
2i
a
2i
b
2i
_
2
_
_
_
_
_
n
i=1
p
2i
_
1
n
i=1
p
2i1
_
n
i=1
p
2i1
a
2
2i1
n
i=1
p
2i1
b
2
2i1
_
n
i=1
p
2i1
a
2i1
b
2i1
_
2
_
_
_
_
_
n
i=1
p
2i1
.
The following result also holds.
+
(R) and I, J P
f
(4.70)
_
_
_
_
1
P
IJ
kIJ
p
k
a
2
k
_1
2
_
1
P
IJ
kIJ
p
k
b
2
k
_1
2
1
P
IJ
kIJ
p
k
a
k
b
k
_
P
IJ
_
_
_
_
1
P
I
iI
p
i
a
2
i
_1
2
_
1
P
I
iI
p
i
b
2
i
_1
2
1
P
I
iI
p
i
a
i
b
i
_
_
_
P
I
_
_
_
_
1
P
J
jJ
p
j
a
2
j
_1
2
_
1
P
J
jJ
p
j
b
2
j
_1
2
1
P
J
jJ
p
j
a
j
b
j
_
_
_
P
J
.
G(I) :=
_
iI
p
i
a
2
i
_1
2
_
iI
p
i
b
2
i
_1
2
iI
p
i
a
i
b
i
is set-superadditive on P
f
(N) .
The following corollary is a natural application.
+
(4.71)
_
_
_
1
P
2n
2n
i=1
p
i
a
2
i
_1
2
_
1
P
2n
2n
i=1
p
i
b
2
i
_1
2
1
P
2n
2n
i=1
p
i
a
i
b
i
_
_
P
2n
70 S.S. DRAGOMIR
_
_
_
1
n
i=1
p
2i
n
i=1
p
2i
a
2
2i
_1
2
_
1
n
i=1
p
2i
n
i=1
p
2i
b
2
2i
_1
2
n
i=1
p
2i
n
i=1
p
2i
a
2i
b
2i
n
i=1
p
2i
_
_
_
1
n
i=1
p
2i1
n
i=1
p
2i1
a
2
2i1
_1
2
_
1
n
i=1
p
2i1
n
i=1
p
2i1
b
2
2i1
_1
2
n
i=1
p
2i1
n
i=1
p
2i1
a
2i1
b
2i1
n
i=1
p
2i1
.
Finally, we may also state:
+
(R) and I, J P
f
(4.72)
_
_
1
P
IJ
kIJ
p
k
a
2
k

1
P
IJ
kIJ
p
k
b
2
k

_
1
P
IJ
kIJ
p
k
a
k
b
k
_
2
_
_
P
IJ
2
_
_
1
P
I
iI
p
i
a
2
i

1
P
I
iI
p
i
b
2
i

_
1
P
I
iI
p
i
a
i
b
i
_
2
_
_
P
I
2
_
_
1
P
J
jJ
p
j
a
2
j

1
P
J
jJ
p
j
b
2
j

_
1
P
J
jJ
p
j
a
j
b
j
_
2
_
_
P
J
2
.
Q(I) :=
_
_
iI
p
i
a
2
i
iI
p
i
b
2
i

_
iI
p
i
a
i
b
i
_
2
_
_
1
2
is set-superadditive on P
f
(N) (see Section 4.7).
The following corollary holds as well.
+
(4.73)
_
_
1
P
2n
2n
i=1
p
i
a
2
i

1
P
2n
2n
i=1
p
i
b
2
i

_
1
P
2n
2n
i=1
p
i
a
i
b
i
_
2
_
_
P
2n
2
__
1
n
i=1
p
2i
n
i=1
p
2i
a
2
2i
__
1
n
i=1
p
2i
n
i=1
p
2i
b
2
2i
_

_
1
n
i=1
p
2i
n
i=1
p
2i
a
2i
b
2i
_
2
_
_
1
2
n
i=1
p
2i
__
1
n
i=1
p
2i1
n
i=1
p
2i1
a
2
2i1
_ _
1
n
i=1
p
2i1
n
i=1
p
2i1
b
2
2i1
_
_
1
n
i=1
p
2i1
n
i=1
p
2i1
a
2i1
b
2i1
_
2
_
_
1
2
n
i=1
p
2i1
.
REFERENCES
[1] S.S. DRAGOMIR AND .Z. ARSLANAGI
C, The improvement of Cauchy-Buniakowski-Schwartzs

inequality, Mat. Bilten, 16 (1992), 7780.
[2] S.S. DRAGOMIR AND B. MOND, On the superadditivity and monotonicity of Schwartzs inequality
in inner product spaces, Contributions, Ser. Math. Sci. Macedonian Acad. Sci., 15(1) (1994), 522.
[3] R.P. AGARWAL AND S.S. DRAGOMIR, The property of supermultiplicity for some classical in-
equalities and applications, Computer Math. Applic., 35(6) (1998), 105118.
[4] S.S. DRAGOMIR AND B. MOND, On the superadditivity and monotonicity of Grams inequality
and related results, Acta Math. Hungarica, 71(1-2) (1996), 7590.
[5] D.S. MITRINOVI
C, J.E. PE
CARI

72 S.S. DRAGOMIR
5. REVERSE INEQUALITIES
5.1. The Cassels Inequality. The following result was proved by J.W.S. Cassels in 1951 (see
Appendix 1 of [2] or Appendix of [3]):
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) be sequences of positive real numbers
and w = (w
1
, . . . , w
n
) a sequence of nonnegative real numbers. Suppose that
(5.1) m = min
i=1,n
_
a
i
b
i
_
and M = max
i=1,n
_
a
i
b
i
_
.
(5.2)
n
i=1
w
i
a
2
i
n
i=1
w
i
b
2
i
(
n
i=1
w
i
a
i
b
i
)
2

(m+M)
2
4mM
.
The equality holds in (5.2) when w
1
=
1
a
1
b
1
, w
n
=
1
a
n
b
n
, w
2
= = w
n1
= 0, m =
a
n
b
1
and
M =
a
1
b
n
.
Proof. 1. The original proof by Cassels (1951) is of interest. We shall follow the paper [5] in
sketching this proof.
We begin with the assertion that
(5.3)
(1 +kw) (1 +k
1
w)
(1 +w)
2

(1 +k) (1 +k
1
)
4
, k > 0, w 0
which, being an equivalent form of (5.2) for n = 2, shows that it holds for n = 2.
To prove that the maximum of (5.2) is obtained when we have more than two w
i
s being
nonzero, Cassels then notes that if for example, w
1
, w
2
, w
3
= 0 lead to an extremum M of
XY
2
2
,
then we would have the linear equations
a
2
n
X +b
2
n
Y 2Ma
n
b
n
Z = 0, k = 1, 2, 3.
Nontrivial solutions exist if and only if the three vectors [a
2
n
, b
2
n
, a
n
b
n
] are linearly dependent.
But this will be so only if, for some i = j (i, j = 1, 2, 3) a
i
= a
j
, b
i
= b
j
. And if that were
true, we could, for example, drop the a
i
, b
i
terms and so deal with the same problem with one
less variable. If only one w
i
= 0, then M = 1, the lower bound. So we need only examine all
pairs w
i
= 0, w
j
= 0. The result (5.2) then quickly follows.
2. We will now use the barycentric method of Frucht [1] and Watson [4]. We will follow the
paper [5].
We substitute w
i
=
u
i
b
2
i
in the left hand side of (5.2), which may then be expressed as the ratio
N
D
2
where
N =
n
i=1
_
a
i
b
i
_
2
u
i
and D =
n
i=1
_
a
i
b
i
_
u
i
,
assuming without loss of generality, that
n
i=1
a
i
= 1. But the point with co-ordinates (D, N)
must lie within the convex closure of the n points
_
a
i
b
i
,
a
2
i
b
2
i
_
. The value of
N
D
2
at points on the
parabola is one unit. If m = min
i=1,n
_
a
i
b
i
_
and M = max
i=1,n
_
a
i
b
i
_
, then the minimum must lie on
the chord joining the point (m, m
2
) and (M, M
2
) . Some easy calculus then leads to (5.2).
The following unweighted Cassels inequality holds.

b satisfy the assumptions in Theorem 5.1, one has the inequality
(5.4)
n
i=1
a
2
i
n
i=1
b
2
i
(
n
i=1
a
i
b
i
)
2

(m+M)
2
4mM
.
The following two additive versions of Cassels inequality hold.
Corollary 5.3. With the assumptions of Theorem 5.1, one has
0
_
n
i=1
w
i
a
2
i
n
i=1
w
i
b
2
i
_1
2
i=1
w
i
a
i
b
i
(5.5)
m
_
2
2
mM
n
i=1
w
i
a
i
b
i
.
and
0
n
i=1
w
i
a
2
i
n
i=1
w
i
b
2
i

_
n
i=1
w
i
a
i
b
i
_
2
(5.6)
(M m)
2
4mM
_
n
i=1
w
i
a
i
b
i
_
2
.
Proof. Taking the square root in (5.2) we get
1
(
n
i=1
w
i
a
2
i
n
i=1
w
i
b
2
i
)
1
2
n
i=1
w
i
a
i
b
i
M +m
2
mM
.
Subtracting 1 on both sides, a simple calculation will lead to (5.5).
The second inequality follows by (5.2) on subtracting 1 and appropriate computation.
The following additive version of unweighted Cassels inequality also holds.
Corollary 5.4. With the assumption of Theorem 5.1 for a and

b one has the inequalities
(5.7) 0
_
n
i=1
a
2
i
n
i=1
b
2
i
_1
2
i=1
a
i
b
i

_
m
_
2
2
mM
n
i=1
a
i
b
i
and
(5.8) 0
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
(M m)
2
4mM
_
n
i=1
a
i
b
i
_
2
.
5.2. The Plya-Szeg Inequality. The following inequality was proved in 1925 by Plya and
Szeg [6, pp. 57, 213 214], [7, pp. 71 72, 253 255].
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
) be two sequences of positive real
numbers. If
(5.9) 0 < a a
i
A < , 0 < b b
i
B < for each i {1, . . . , n} ,
(5.10)
n
i=1
a
2
i
n
i=1
b
2
i
(
n
i=1
a
i
b
i
)
2

(ab +AB)
2
4abAB
.
74 S.S. DRAGOMIR
The equality holds in (5.10) if and only if
p = n
A
a
__
A
a
+
B
b
_
and q = n
B
b
__
A
a
+
B
b
_
are integers and if p of the numbers a
1
, . . . , a
n
are equal to a and q of these numbers are equal
to A, and if the corresponding numbers b
i
are equal to B and b respectively.
Proof. Following [5], we shall present here the original proof of Plya and Szeg.
We may, without loss of generality, suppose that a
1
a
n
, then to maximise the left-
hand side of (5.10) we must have that the critical b
i
s be reversely ordered (for if b
k
> b
m
with
k < m, then we can interchange b
k
and b
m
such that b
2
k
+ b
2
m
= b
2
m
+ b
2
k
and a
k
b
k
+ a
m
b
m

a
k
b
m
+a
m
b
k
), i.e., that b
1
b
n
.
Plya and Szeg then continue by dening nonnegative numbers u
i
and v
i
for i = 1, . . . , n1
and n > 2 such that
(5.11) a
2
i
= u
i
a
2
1
+v
i
a
2
n
and b
2
i
= u
i
b
2
1
+v
i
b
2
n
.
Since a
i
b
i
> u
i
a
1
b
1
+v
i
a
n
b
n
the left hand side of (5.10),
n
i=1
a
2
i
n
i=1
b
2
i
(
n
i=1
a
i
b
i
)
2

(Ua
2
1
+V a
2
n
) (Ub
2
1
+V b
2
n
)
(Ua
1
b
1
+V a
n
b
n
)
2
,
where U =
n
i=1
u
i
and V =
n
i=1
v
i
.
This reduces the problem to that with n = 2, which is solvable by elementary methods,
leading to
(5.12)
n
i=1
a
2
i
n
i=1
b
2
i
(
n
i=1
a
i
b
i
)
2

(a
1
b
1
+a
n
b
n
)
2
4a
1
a
n
b
1
b
n
,
where, since the a
i
s and b
i
s here are reversely ordered,
(5.13) a
1
= max
i=1,n
{a
i
} , a
n
= min
i=1,n
{a
i
} , b
1
= min
i=1,n
{b
i
} , b
n
= max
i=1,n
{b
i
} .
If we now assume, as in (5.9), that
0 < a a
i
A, 0 < b b
i
B, i = (1, . . . , n) ,
then
(a
1
b
1
+a
n
b
n
)
2
4a
1
a
n
b
1
b
n
(ab +AB)
2
4abAB
(because
(k+1)
2
4k

(+1)
2
4
for k ), and the inequality (5.10) is proved.
Remark 5.6. The inequality (5.10) may also be obtained from the unweighted Cassels in-
equality
(5.14)
n
i=1
a
2
i
n
i=1
b
2
i
(
n
i=1
a
i
b
i
)
2

(m+M)
2
4mM
,
where 0 < m
a
i
b
i
M for each i {1, . . . , n} .
The following additive versions of the Plya-Szeg inequality also hold.
Corollary 5.7. With the assumptions in Theorem 5.5, one has the inequality
(5.15) 0
_
n
i=1
a
2
i
n
i=1
b
2
i
_1
2
i=1
a
i
b
i

_
AB
ab
_
2
2
abAB
n
i=1
a
i
b
i
and
(5.16) 0
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
(AB ab)
2
4abAB
_
n
i=1
a
i
b
i
_
2
.
5.3. The Greub-Rheinboldt Inequality. The following weighted version of the Plya-Szeg
inequality was obtained by Greub and Rheinboldt in 1959, [6].
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
numbers and w = (w
1
, . . . , w
n
) a sequence of nonnegative real numbers. Suppose that
(5.17) 0 < a a
i
A < , 0 < b b
i
B < (i = 1, . . . , n) .
(5.18)
n
i=1
w
i
a
2
i
n
i=1
w
i
b
2
i
(
n
i=1
w
i
a
i
b
i
)
2

(ab +AB)
2
4abAB
.
Equality holds in (5.18) when w
i
=
1
a
1
b
1
, w
n
=
1
a
n
b
n
, w
2
= = w
n1
= 0, m =
a
n
b
1
, M =
a
1
b
n
with a
1
= A, a
n
= a, b
1
= b and b
n
= b.
Remark 5.9. This inequality follows by Cassels result which states that
(5.19)
n
i=1
w
i
a
2
i
n
i=1
w
i
b
2
i
(
n
i=1
w
i
a
i
b
i
)
2

(m+M)
2
4mM
,
provided 0 < m
a
i
b
i
M < for each i {1, . . . , n} .
The following additive versions of Greub-Rheinboldt also hold.
Corollary 5.10. With the assumptions in Theorem 5.8, one has the inequalities
0
_
n
i=1
w
i
a
2
i
n
i=1
w
i
b
2
i
_1
2
i=1
w
i
a
i
b
i
(5.20)
AB
ab
_
2
2
abAB
n
i=1
w
i
a
i
b
i
and
0
n
i=1
w
i
a
2
i
n
i=1
w
i
b
2
i

_
n
i=1
w
i
a
i
b
i
_
2
(5.21)
(AB ab)
2
4abAB
_
n
i=1
w
i
a
i
b
i
_
2
.
5.4. A Cassels Type Inequality for Complex Numbers. The following reverse inequality for
the (CBS) inequality holds [9].
Theorem 5.11. Let a, A K (K = C, R) such that Re ( aA) > 0.
If x = (x
1
, . . . , x
n
) , y = (y
1
, . . . , y
n
) are sequences of complex numbers and w = (w
1
, . . . , w
n
)
is a sequence of nonnegative real numbers with the property that
(5.22)
n
i=1
w
i
Re [(Ay
i
x
i
) ( x
i
a y
i
)] 0,
76 S.S. DRAGOMIR
_
n
i=1
w
i
|x
i
|
2
n
i=1
w
i
|y
i
|
2
_1
2
1
2

n
i=1
w
i
Re [A x
i
y
i
+ ax
i
y
i
]
[Re ( aA)]
1
2
(5.23)
1
2

|A| +|a|
[Re ( aA)]
1
2
i=1
w
i
x
i
y
i
.
The constant
1
2
is sharp in the sense that it cannot be replaced by a smaller one.
Proof. We have, obviously, that
:=
n
i=1
w
i
Re [(Ay
i
x
i
) ( x
i
a y
i
)]
=
n
i=1
w
i
Re [A x
i
y
i
+ ax
i
y
i
]
n
i=1
w
i
|x
i
|
2
Re ( aA)
n
i=1
w
i
|y
i
|
2
and then, by (5.22), one has
n
i=1
w
i
|x
i
|
2
+ Re ( aA)
n
i=1
w
i
|y
i
|
2
i=1
w
i
Re [A x
i
y
i
+ ax
i
y
i
]
giving
(5.24)
1
[Re ( aA)]
1
2
n
i=1
w
i
|x
i
|
2
+ [Re ( aA)]
1
2
n
i=1
w
i
|y
i
|
2
n
i=1
w
i
Re [A x
i
y
i
+ ax
i
y
i
]
[Re ( aA)]
1
2
.
On the other hand, by the elementary inequality
p
2
+
1
q
2
2pq
holding for any p, q 0 and > 0, we deduce
(5.25) 2
_
n
i=1
w
i
|x
i
|
2
n
i=1
w
i
|y
i
|
2
_1
2
1
[Re ( aA)]
1
2
n
i=1
w
i
|x
i
|
2
+ [Re ( aA)]
1
2
n
i=1
w
i
|y
i
|
2
.
Utilising (5.24) and (5.25), we deduce the rst part of (5.23).
The second part is obvious by the fact that for z C, |Re (z)| |z| .
Now, assume that the rst inequality in (5.23) holds with a constant c > 0, i.e.,
(5.26)
n
i=1
w
i
|x
i
|
2
n
i=1
w
i
|y
i
|
2
c
n
i=1
w
i
Re [A x
i
y
i
+ ax
i
y
i
]
[Re ( aA)]
1
2
,
where a, A, x, y satisfy (5.22).
If we choose a = A = 1, y = x = 0, then obviously (5.23) holds and from (5.26) we may
get
n
i=1
w
i
|x
i
|
2
2c
n
i=1
w
i
|x
i
|
2
,
giving c
1
2
.
The theorem is completely proved.
Corollary 5.12. Let m, M > 0 and x, y, w be as in Theorem 5.11 and with the property that
(5.27)
n
i=1
w
i
Re [(My
i
x
i
) ( x
i
m y
i
)] 0,
_
n
i=1
w
i
|x
i
|
2
n
i=1
w
i
|y
i
|
2
_1
2
1
2

M +m
mM
n
i=1
w
i
Re (x
i
y
i
) (5.28)
1
2

M +m
mM
i=1
w
i
x
i
y
i
.
The following corollary also holds.
Corollary 5.13. With the assumptions in Corollary 5.12, then one has the following inequality:
0
_
n
i=1
w
i
|x
i
|
2
n
i=1
w
i
|y
i
|
2
_1
2
i=1
w
i
x
i
y
i
(5.29)
_
n
i=1
w
i
|x
i
|
2
n
i=1
w
i
|y
i
|
2
_1
2
i=1
w
i
Re (x
i
y
i
)
m
_
2
2
mM
n
i=1
w
i
Re (x
i
y
i
)
m
_
2
2
mM
i=1
w
i
x
i
y
i
and
0
n
i=1
w
i
|x
i
|
2
n
i=1
w
i
|y
i
|
2
i=1
w
i
x
i
y
i
2
(5.30)
i=1
w
i
|x
i
|
2
n
i=1
w
i
|y
i
|
2
_
n
i=1
w
i
Re (x
i
y
i
)
_
2
(M m)
2
4mM
_
n
i=1
w
i
Re (x
i
y
i
)
_
2
(M m)
2
4mM
i=1
w
i
x
i
y
i
2
.
5.5. A Reverse Inequality for Real Numbers. The following result holds [10, Proposition
5.1].
Theorem 5.14. Let a, A R and x = (x
1
, . . . , x
n
) , y = (y
1
, . . . , y
n
) be two sequences with
the property that:
(5.31) ay
i
x
i
Ay
i
for each i {1, . . . , n} .
78 S.S. DRAGOMIR
Then for any w = (w
1
, . . . , w
n
) a sequence of positive real numbers, one has the inequality
0
n
i=1
w
i
x
2
i
n
i=1
w
i
y
2
i

_
n
i=1
w
i
x
i
y
i
_
2
(5.32)
1
4
(A a)
2
_
n
i=1
w
i
y
2
i
_
2
.
The constant
1
4
is sharp in (5.32).
Proof. Let us dene
I
1
:=
_
A
n
i=1
w
i
y
2
i

n
i=1
w
i
x
i
y
i
__
n
i=1
w
i
x
i
y
i
a
n
i=1
w
i
y
2
i
_
and
I
2
:=
_
n
i=1
w
i
y
2
i
_
n
i=1
(Ay
i
x
i
) (x
i
ay
i
) w
i
.
Then
I
1
= (a +A)
n
i=1
w
i
y
2
i
n
i=1
w
i
x
i
y
i
_
n
i=1
w
i
x
i
y
i
_
2
aA
_
n
i=1
w
i
y
2
i
_
2
and
I
2
= (a +A)
n
i=1
w
i
y
2
i
n
i=1
w
i
x
i
y
i
i=1
w
i
x
2
i
n
i=1
w
i
y
2
i
aA
_
n
i=1
w
i
y
2
i
_
2
giving
(5.33) I
1
I
2
=
n
i=1
w
i
x
2
i
n
i=1
w
i
y
2
i

_
n
i=1
w
i
y
2
i
_
2
.
If (5.31) holds, then (Ay
i
x
i
) (x
i
ay
i
) 0 for each i {1, . . . , n} and thus I
2
0 giving
(5.34)
n
i=1
w
i
x
2
i
n
i=1
w
i
y
2
i

_
n
i=1
w
i
y
2
i
_
2
__
A
n
i=1
w
i
y
2
i

n
i=1
w
i
x
i
y
i
__
n
i=1
w
i
x
i
y
i
a
n
i=1
w
i
y
2
i
__
.
If we use the elementary inequality for real numbers u, v R
(5.35) uv
1
4
(u +v)
2
,
then we have for
u := A
n
i=1
w
i
y
2
i

n
i=1
w
i
x
i
y
i
, v :=
n
i=1
w
i
x
i
y
i
a
n
i=1
w
i
y
2
i
that
_
A
n
i=1
w
i
y
2
i

n
i=1
w
i
x
i
y
i
__
n
i=1
w
i
x
i
y
i
a
n
i=1
w
i
y
2
i
_
1
4
(A a)
2
_
n
i=1
w
i
y
2
i
_
2
Now, assume that (5.32) holds with a constant c > 0, i.e.,
(5.36)
n
i=1
w
i
x
2
i
n
i=1
w
i
y
2
i

_
n
i=1
w
i
x
i
y
i
_
2
c (A a)
2
_
n
i=1
w
i
y
2
i
_
2
,
where a, A, x, y satisfy (5.31).
We choose n = 2, w
1
= w
2
= 1 and let a, A, y
1
, y
2
, x, R such that
ay
1
< x
1
= Ay
1
,
ay
2
= x
2
< Ay
2
.
With these choices, we get from (5.36) that
_
a
2
y
2
1
+a
2
y
2
2
_ _
y
2
1
+y
2
2
_
_
A
2
y
2
1
+a
2
y
2
2
_
2
c (A a)
2
_
y
2
1
+y
2
2
_
2
,
which is equivalent to
(A a)
2
y
2
1
y
2
2
c (A a)
2
_
y
2
1
+y
2
2
_
2
.
Since we may choose a = A, we deduce
y
2
1
y
2
2
c
_
y
2
1
+y
2
2
_
2
,
giving, for y
1
= y
2
= 1, c
1
4
.
The following corollary is obvious.
Corollary 5.15. With the above assumptions for a, A, x and y, we have the inequality
(5.37) 0
n
i=1
x
2
i
n
i=1
y
2
i

_
n
i=1
x
i
y
i
_
2
1
4
(A a)
2
_
n
i=1
y
2
i
_
2
.
Remark 5.16. Condition (5.31) may be replaced by the weaker condition
(5.38)
n
i=1
w
i
(Ay
i
x
i
) (x
i
ay
i
) 0
and the conclusion in Theorem 5.14 will still be valid, i.e., the inequality (5.32) holds.
For (5.37) to be true it sufces that
(5.39)
n
i=1
(Ay
i
x
i
) (x
i
ay
i
) 0
holds true.
5.6. A Reverse Inequality for Complex Numbers. The following result holds [10, Proposi-
tion 5.1].
Theorem5.17. Let a, A Cand x = (x
1
, . . . , x
n
) , y = (y
1
, . . . , y
n
) C
n
, w = (w
1
, . . . , w
n
)
R
n
+
. If
(5.40)
n
i=1
w
i
Re [(Ay
i
x
i
) ( x
i
a y
i
)] 0,
80 S.S. DRAGOMIR
0
n
i=1
w
i
|x
i
|
2
n
i=1
w
i
|y
i
|
2
i=1
w
i
x
i
y
i
2
(5.41)
1
4
|A a|
2
_
n
i=1
w
i
|y
i
|
2
_
2
.
The constant
1
4
is sharp in (5.41).
Proof. Consider
A
1
:= Re
__
A
n
i=1
w
i
|y
i
|
2
i=1
w
i
x
i
y
i
__
n
i=1
w
i
x
i
y
i
a
n
i=1
w
i
|y
i
|
2
__
and
A
2
:=
n
i=1
w
i
|y
i
|
2
Re
_
n
i=1
w
i
(Ay
i
x
i
) ( x
i
a y
i
)
_
.
Then
A
1
=
n
i=1
w
i
|y
i
|
2
Re
_
A
n
i=1
w
i
x
i
y
i
+ a
n
i=1
w
i
x
i
y
i
_
i=1
w
i
x
i
y
i
2
Re ( aA)
_
n
i=1
w
i
|y
i
|
2
_
2
and
A
2
=
n
i=1
w
i
|y
i
|
2
Re
_
A
n
i=1
w
i
x
i
y
i
+ a
n
i=1
w
i
x
i
y
i
_
i=1
w
i
|x
i
|
2
n
i=1
w
i
|y
i
|
2
Re ( aA)
_
n
i=1
w
i
|y
i
|
2
_
2
giving
(5.42) A
1
A
2
=
n
i=1
w
i
|x
i
|
2
n
i=1
w
i
|y
i
|
2
i=1
w
i
x
i
y
i
2
.
If (5.40) holds, then A
2
0 and thus
(5.43)
n
i=1
w
i
|x
i
|
2
n
i=1
w
i
|y
i
|
2
i=1
w
i
x
i
y
i
2
Re
__
A
n
i=1
w
i
|y
i
|
2
i=1
w
i
x
i
y
i
_ _
n
i=1
w
i
x
i
y
i
a
n
i=1
w
i
|y
i
|
2
__
.
If we use the elementary inequality for complex numbers z, t C
(5.44) Re [z
t]
1
4
|z t|
2
,
then we have for
z := A
n
i=1
w
i
|y
i
|
2
i=1
w
i
x
i
y
i
,
t :=
n
i=1
w
i
x
i
y
i
a
n
i=1
w
i
|y
i
|
2
that
(5.45) Re
__
A
n
i=1
w
i
|y
i
|
2
i=1
w
i
x
i
y
i
_ _
n
i=1
w
i
x
i
y
i
a
n
i=1
w
i
|y
i
|
2
__
1
4
|A a|
2
_
n
i=1
w
i
|y
i
|
2
_
2
Now, assume that (5.41) holds with a constant c > 0, i.e.,
(5.46)
n
i=1
w
i
|x
i
|
2
n
i=1
w
i
|y
i
|
2
i=1
w
i
x
i
y
i
2
c |A a|
2
_
n
i=1
w
i
|y
i
|
2
_
2
,
where x, y, a, A satisfy (5.40).
Consider y C
n
,
n
i=1
|y
i
|
2
w
i
= 1, a = A, m C
n
,
n
i=1
w
i
|m
i
|
2
= 1 with
n
i=1
w
i
y
i
m
i
=
0. Dene
x
i
:=
A +a
2
y
i
+
A +a
2
m
i
, i {1, . . . , n} .
Then
n
i=1
w
i
(Ay
i
x
i
) ( x
i
ay
i
) =
A a
2
2 n
i=1
w
i
(y
i
m
i
) ( y
i
m
i
) = 0
and thus the condition (5.40) is fullled.
From (5.46) we deduce
n
i=1
A +a
2
y
i
+
A a
2
m
i
2
w
i
i=1
_
A +a
2
y
i
+
A a
2
m
i
_
y
i
w
i
2
c |A a|
2
and since
n
i=1
w
i
_
A +a
2
y
i
+
A a
2
m
i
_
2
=
A +a
2
A a
2
2
and
i=1
_
A +a
2
y
i
+
A a
2
m
i
_
y
i
w
i
2
=
A +a
2
2
then by (5.46) we get
|A a|
2
4
c |A a|
2
giving c
1
4
and the theorem is completely proved.
82 S.S. DRAGOMIR
Corollary 5.18. Let a, A C and x = (x
1
, . . . , x
n
) , y = (y
1
, . . . , y
n
) C
n
be with the
property that
(5.47)
n
i=1
Re [(Ay
i
x
i
) ( x
i
a y
i
)] 0,
(5.48) 0
n
i=1
|x
i
|
2
n
i=1
|y
i
|
2
i=1
x
i
y
i
1
4
|A a|
2
_
n
i=1
|y
i
|
2
_
2
.
The constant
1
4
is best in (5.48).
Remark 5.19. A sufcient condition for both (5.40) and (5.47) to hold is
(5.49) Re [(Ay
i
x
i
) ( x
i
a y
i
)] 0
for any i {1, . . . , n} .
5.7. Shisha-Mond Type Inequalities. As some particular case for bounds on differences of
means, O. Shisha and B. Mond obtained in 1967 (see [23]) the following reverse of (CBS)
inequality:
Theorem 5.20. Assume that a = (a
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
) are such that there exists
a, A, b, B > 0 with the property that:
(5.50) a a
j
A and b b
j
B for any j {1, . . . , n}
then we have the inequality
(5.51)
n
j=1
a
2
j
n
j=1
b
2
j

_
n
j=1
a
j
b
j
_
2
_
_
A
b

_
a
B
_
2
n
j=1
a
j
b
j
n
j=1
b
2
j
.
The equality holds in (5.51) if and only if there exists a subsequence (k
1
, . . . , k
p
) of (1, 2, . . . , n)
such that
n
p
= 1 +
_
A
a
_1
2
_
B
b
_3
2
,
a
k
= A, b
k
= b ( = 1, . . . , p) and a
k
= a, b
k
= B for every k distinct from all k
.
Using another result stated for weighted means in [23], we may prove the following reverse
of the (CBS) inequality.
Theorem 5.21. Assume that a,

b are positive sequences and there exists , > 0 with the
property that
(5.52) 0 <
a
i
b
i
< for any i {1, . . . , n} .
Then we have the inequality
(5.53) 0
_
n
i=1
a
2
i
n
i=1
b
2
i
_1
2
i=1
a
i
b
i

( )
2
4 ( + )
n
j=1
b
2
j
.
The equality holds in (5.53) if and only if there exists a subsequence (k
1
, . . . , k
p
) of (1, 2, . . . , n)
such that
p
m=1
b
2
k
m
=
+ 3
4 ( + )
n
j=1
b
2
j
,
a
k
m
b
k
m
= (m = 1, . . . , p) and
a
k
b
k
=
for every k distinct from all k
m
.
Proof. In [23, p. 301], Shisha and Mond have proved the following weighted inequality
(5.54) 0
_
n
j=1
q
j
x
2
j
_1
2
j=1
q
j
x
j

(C c)
2
4 (c +C)
,
provided q
j
0 (j = 1, . . . , n) with

n
j=1
q
j
= 1 and 0 < c x
j
< C < for any
j {1, . . . , n} .
Equality holds in (5.54) if and only if there exists a subsequence (k
1
, . . . , k
p
) of (1, 2, . . . , n)
such that
(5.55)
p
m=1
q
k
m
=
C + 3c
4 (c +C)
,
x
k
m
= C (m = 1, 2, . . . , p) and x
k
= c for every k distinct from all k
m
.
x
j
=
a
j
b
j
, q
j
=
b
2
j
n
k=1
b
2
k
, j {1, . . . , n} ;
then we get
_
n
j=1
a
2
j
n
k=1
b
2
k
_1
2
n
j=1
a
j
b
j
n
k=1
b
2
k
( )
2
4 ( + )
,
The case of equality follows by the similar case in (5.54) and we omit the details.
5.8. Zagier Type Inequalities. The following result was obtained by D. Zagier in 1995, [24].
Lemma 5.22. Let f, g : [0, ) R be monotone decreasing nonnegative functions on [0, ).
Then
(5.56)
_

0
f (x) g (x) dx
_
0
f (x) F (x) dx
_
0
g (x) G(x) dx
max
__
0
F (x) dx,
_
0
G(x) dx
_ ,
for any integrable functions F, G : [0, ) [0, 1] .
For all x 0 we have
_

0
f (t) F (t) dt = f (x)
_

0
F (t) dt +
_

0
[f (t) f (x)] F (t) dt
f (x)
_

0
F (t) dt +
_
x
0
[f (t) f (x)] dt
and hence, since
_
x
0
G(t) dt is bounded from above by both x and
_
0
G(t) dt,
_

0
f (t) F (t) dt
_
x
0
G(t) dt
xf (x)
_

0
F (t) dt +
_

0
G(t) dt
_
x
0
[f (t) f (x)] dt
max
__

0
F (t) dt,
_

0
G(t) dt
_
_
x
0
f (t) dt.
84 S.S. DRAGOMIR
Now, we multiply by dg (x) and integrate by parts from 0 to . The left hand side gives
_
f (t) F (t) dt
_
g (t) G(t) dt, the right hand side gives

max
__

0
F (t) dt,
_

0
G(t) dt
_
_

0
f (t) g (t) dt,
and the inequality remains true because the measure dg (x) is nonnegative.
The following particular case is a reverse of the (CBS) integral inequality obtained by D.
Zagier in 1977, [25].
Corollary 5.23. If f, g : [0, ) [0, ) are decreasing function on [0, ), then
(5.57) max
_
f (0)
_

0
g (t) dt, g (0)
_

0
f (t) dt
_
_

0
f (t) g (t) dt
_

0
f
2
(t) dt
_

0
g
2
(t) dt.
Remark 5.24. The following weighted version of (5.56) may be proved in a similar way, as
noted by D. Zagier in [25]
(5.58)
_

0
w(t) f (t) g (t) dt
_
0
w(t) f (t) F (t) dt
_
0
w(t) f (t) G(t) dt
max
__
0
w(t) F (t) dt,
_
0
w(t) G(t) dt
_ ,
provided w(t) > 0 on [0, ), f, g : [0, ) [0, ) are monotonic decreasing and F, G :
[0, ) [0, 1] are integrable on [0, ).
We may state and prove the following discrete inequality.
Theorem 5.25. Consider the sequences of real numbers a = (a
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
) ,
p = (p
1
, . . . , p
n
), q = (q
1
, . . . , q
n
) and w = (w
1
, . . . , w
n
) .
If
(i) a and

b are decreasing and nonnegative;
(ii) p
i
, q
i
[0, 1] and w
i
0 for any i {1, . . . , n} ,
(5.59)
n
i=1
w
i
a
i
b
i

n
i=1
w
i
p
i
a
i
n
i=1
w
i
q
i
b
i
max {
n
i=1
w
i
p
i
,
n
i=1
w
i
q
i
}
.
Proof. Consider the functions f, g, F, G, W : [0, ) R given by
f (t) =
_
_
a
1
, t [0, 1)
a
2
, t [1, 2)
.
.
.
a
n
, t [n 1, n)
0 t [n, )
, g (t) =
_
_
b
1
, t [0, 1)
b
2
, t [1, 2)
.
.
.
b
n
, t [n 1, n)
0 t [n, )
,
F (t) =
_
_
p
1
, t [0, 1)
p
2
, t [1, 2)
.
.
.
p
n
, t [n 1, n)
0 t [n, )
, G(t) =
_
_
q
1
, t [0, 1)
q
2
, t [1, 2)
.
.
.
q
n
, t [n 1, n)
0 t [n, )
,
and
W (t) =
_
_
w
1
, t [0, 1)
w
2
, t [1, 2)
.
.
.
w
n
, t [n 1, n)
0 t [n, )
.
We observe that, the above functions satisfy the hypothesis of Remark 5.24 and since, for ex-
ample,
_

0
w(t) f (t) g (t) dt =
n
i=1
_
i
i1
w(t) f (t) g (t) dt +
_

n
w(t) f (t) g (t) dt
=
n
k=1
w
k
a
k
b
k
,
then by (5.58) we deduce the desired inequality (5.59).
Remark 5.26. A similar inequality for sequences under some monotonicity assumptions for p
and q was obtained in 1995 by J. Pe cari c in [26].
The following reverse of the (CBS) discrete inequality holds.
Theorem 5.27. Assume that a,

b are decreasing nonnegative sequences with a
1
, b
1
= 0 and w
a nonnegative sequence. Then
(5.60)
n
i=1
w
i
a
2
i
n
i=1
w
i
b
2
i
max
_
b
1
n
i=1
w
i
a
i
, a
1
n
i=1
w
i
b
i
_
n
i=1
w
i
a
i
b
i
.
The proof follows by Theorem 5.25 on choosing p
i
=
a
i
a
1
[0, 1] , q
i
=
b
i
b
1
[0, 1] , i
{1, . . . , n} . We omit the details.
Remark 5.28. When w
i
= 1, we recapture Alzers result from 1992, [27].
5.9. A Reverse Inequality in Terms of the sup Norm. The following result has been proved
in [11].
Lemma 5.29. Let = (
1
, . . . ,
n
) and x = (x
1
, . . . , x
n
) be sequences of complex numbers
and p = (p
1
, . . . , p
n
) a sequence of nonnegative real numbers such that
n
i=1
p
i
= 1. Then one
has the inequality
(5.61)
i=1
p
i
i
x
i
i=1
p
i
i
n
i=1
p
i
x
i
max
i=1,n1
|
i
| max
i=1,n1
|x
i
|
_
_
n
i=1
i
2
p
i
_
n
i=1
ip
i
_
2
_
_
,
where
i
is the forward difference, i.e.,
i
:=
i+1
i
.
Inequality (5.61) is sharp in the sense that the constant C = 1 in the right membership cannot
be replaced be a smaller one.
86 S.S. DRAGOMIR
Proof. We shall follow the proof in [11]. We start with the following identity
n
i=1
p
i
i
x
i
i=1
p
i
i
n
i=1
p
i
x
i
=
1
2
n
i,j=1
p
i
p
j
(
i
j
) (x
i
x
j
)
=
1i<jn
p
i
p
j
(
i
j
) (x
i
x
j
) .
As i < j, we can write that
j

i
=
j1
k=i
k
and
x
j
x
i
=
j1
k=i
x
k
.
Using the generalised triangle inequality, we have successively
i=1
p
i
i
x
i
i=1
p
i
i
n
i=1
p
i
x
i
1i<jn
p
i
p
j
j1
k=i
k
j1
k=i
x
k
1i<jn
p
i
p
j
j1
k=i
j1
k=i
x
k
1i<jn
p
i
p
j
j1
k=i
|
k
|
j1
k=i
|x
k
|
:= A.
Note that
|
k
| max
1sn1
|
s
|
and
|x
k
| max
1sn1
|x
s
|
for all k = i, . . . , j 1 and then by summation
j1
k=i
|
k
| (j i) max
1sn1
|
s
|
and
j1
k=i
|x
k
| (j i) max
1sn1
|x
s
| .
Taking into account the above estimations, we can write
A
_

1i<jn
p
i
p
j
(j i)
2
_
max
1sn1
|
s
| max
1sn1
|x
s
| .
As a simple calculation shows that
1i<jn
p
i
p
j
(j i)
2
=
n
i=1
i
2
p
i
_
n
i=1
ip
i
_
2
,
inequality (5.61) is proved.
To prove the sharpness of the constant, let us assume that (5.61) holds with a constant C > 0,
i.e.,
(5.62)
i=1
p
i
i
x
i
i=1
p
i
i
n
i=1
p
i
x
i
C max
i=1,n1
|
i
| max
i=1,n1
|x
i
|
_
_
n
i=1
i
2
p
i
_
n
i=1
ip
i
_
2
_
_
.
Now, choose the sequences
k
= + k ( = 0) and x
k
= x + ky (y = 0) , k {1, . . . , n}
to get
i=1
p
i
i
x
i
i=1
p
i
i
n
i=1
p
i
x
i
=
1
2
i,j=1
p
i
p
j
(i j) y
= || |y|
_
_
n
i=1
i
2
p
i
_
n
i=1
ip
i
_
2
_
_
and
max
i=1,n1
|
i
| max
i=1,n1
|x
i
|
_
_
n
i=1
i
2
p
i
_
n
i=1
ip
i
_
2
_
_
= || |y|
_
_
n
i=1
i
2
p
i
_
n
i=1
ip
i
_
2
_
_
and then, by (5.62), we get C 1.
The following reverse of the (CBS) inequality holds [12].
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
) be two sequences of real numbers
with a
i
= 0, (i = 1, . . . , n) . Then one has the inequality
0
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
max
k=1,n1
_
_
b
k
a
k
__
2
_
_
n
i=1
a
2
i
n
i=1
i
2
a
2
i

_
n
i=1
ia
2
i
_
2
_
_
.
The constant C = 1 is sharp in the sense that it cannot be replaced by a smaller constant.
Proof. Follows by Lemma 5.29 on choosing
p
i
=
a
2
i
n
k=1
a
2
k
,
i
=
b
i
a
i
, x
i
=
b
i
a
i
, i {1, . . . , n}
and performing some elementary calculations.
We omit the details.
5.10. A Reverse Inequality in Terms of the 1Norm. The following result has been obtained
in [13].
Lemma 5.31. Let = (
1
, . . . ,
n
) and x = (x
1
, . . . , x
n
and p = (p
1
, . . . , p
n
n
i=1
p
i
= 1. Then one
88 S.S. DRAGOMIR
has the inequality
(5.63)
i=1
p
i
i
x
i
i=1
p
i
i
n
i=1
p
i
x
i
1
2
n
i=1
p
i
(1 p
i
)
n1
i=1
|
i
|
n1
i=1
|x
i
| ,
where
i
:=
i+1
i
is the forward difference.
The constant
1
2
is sharp in the sense that it cannot be replaced by a smaller constant.
As in the proof of Lemma 5.29 in Section 5.9, we have
(5.64)
i=1
p
i
i
x
i
i=1
p
i
i
n
i=1
p
i
x
i
1i<jn
p
i
p
j
j1
k=i
|
k
|
j1
l=i
|x
l
| := A.
It is obvious that for all 1 i < j n 1, we have that
j1
k=i
|
k
|
n1
k=1
|
k
|
and
j1
l=i
|x
l
|
n1
l=1
|x
l
| .
Utilising these and the denition of A, we conclude that
(5.65) A
n1
k=1
|
k
|
n1
l=1
|x
l
|
1i<jn
p
i
p
j
.
Now, let us observe that
1i<jn
p
i
p
j
=
1
2
_
n
i,j=1
p
i
p
j

i=j
p
i
p
j
_
(5.66)
=
1
2
_
n
i=1
p
i
n
j=1
p
j

n
i=1
p
2
i
_
=
1
2
n
i=1
p
i
(1 p
i
) .
Making use of (5.64) (5.66), we deduce the desired inequality (5.63).
To prove the sharpness of the constant
1
2
, let us assume that (5.63) holds with a constant
C > 0. That is
(5.67)
i=1
p
i
i
x
i
i=1
p
i
i
n
i=1
p
i
x
i
C
n
i=1
p
i
(1 p
i
)
n1
i=1
|
i
|
n1
i=1
|x
i
|
for all
i
, x
i
, p
i
(i = 1, . . . , n) as above and n 1.
Choose in (5.63) n = 2 and compute
2
i=1
p
i
i
x
i
i=1
p
i
i
2
i=1
p
i
x
i
=
1
2
2
i,j=1
p
i
p
j
(
i
j
) (x
i
x
j
)
=
1i<j2
p
i
p
j
(
i
j
) (x
i
x
j
)
= p
1
p
2
(
1
2
) (x
1
x
2
) .
Also
2
i=1
p
i
(1 p
i
)
2
i=1
|
i
|
2
i=1
|x
i
| = (p
1
p
2
+p
1
p
2
) |
1
2
| |x
1
x
2
| .
Substituting in (5.67), we obtain
p
1
p
2
|
1
2
| |x
1
x
2
| 2Cp
1
p
2
|
1
2
| |x
1
x
2
| .
If we assume that p
1
, p
2
> 0,
1
=
2
, x
1
= x
2
, then we obtain C
1
2
, which proves the
sharpness of the constant
1
2
.
We are now able to state the following reverse of the (CBS) inequality [12].
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
with a
i
= 0 (i = 1, . . . , n) . Then one has the inequality
0
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
(5.68)
_
n1
k=1
_
b
k
a
k
_
_
2
1i<jn
a
2
i
a
2
j
.
The constant C = 1 is sharp in (5.68), in the sense that it cannot be replaced by a smaller
constant.
Proof. We choose
p
i
=
a
2
i
n
k=1
a
2
k
,
i
= x
i
=
b
i
a
i
, i {1, . . . , n}
in (5.63) to get
0
n
i=1
b
2
i
n
k=1
a
2
k
n
i=1
a
i
b
i
)
2
(
n
k=1
a
2
k
)
2
1
2

n
i=1
a
2
i
_
1
a
2
i
n
k=1
a
2
k
_
n
k=1
a
2
k
_
n1
j=1
_
b
j
a
j
_
_
2
=
1
2

n
i=1
a
2
i
(
n
k=1
a
2
k
a
2
i
)
(
n
k=1
a
2
k
)
2
_
n1
j=1
_
b
j
a
j
_
_
2
90 S.S. DRAGOMIR
which is clearly equivalent to
0
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
1
2
_
_
_
n
k=1
a
2
k
_
2
i=1
a
4
i
_
_
_
n
j=1
_
b
j
a
j
_
_
2
.
Since
1
2
_
_
_
n
k=1
a
2
k
_
2
i=1
a
4
i
_
_
=
1i<jn
a
2
i
a
2
j
the inequality (5.68) is thus proved.
5.11. A Reverse Inequality in Terms of the pNorm. The following result has been obtained
in [14].
Lemma 5.33. Let = (
1
, . . . ,
n
) and x = (x
1
, . . . , x
n
and p = (p
1
, . . . , p
n
n
i=1
p
i
= 1. Then one
has the inequality
(5.69)
i=1
p
i
i
x
i
i=1
p
i
i
n
i=1
p
i
x
i
1j<in
(i j) p
i
p
j
_
n1
k=1
|
k
|
p
_1
p
_
n1
k=1
|x
k
|
q
_1
q
,
where p > 1,
1
p
+
1
q
= 1.
The constant C = 1 in the right hand side of (5.69) is sharp in the sense that it cannot be
replaced by a smaller constant.
As in the proof of Lemma 5.29 in Section 5.9, we have
(5.70)
i=1
p
i
i
x
i
i=1
p
i
i
n
i=1
p
i
x
i
1j<in
p
i
p
j
i1
k=j
|
k
|
i1
l=j
|x
l
| := A.
Using Hlders discrete inequality, we can state that
i1
k=j
|
k
| (i j)
1
q
_
i1
k=j
|
k
|
p
_
1
p
and
i1
l=j
|x
l
| (i j)
1
p
_
i1
l=j
|x
l
|
q
_
1
q
,
where p > 1,
1
p
+
1
q
= 1, and then we get
(5.71) A
1j<in
p
i
p
j
(i j)
_
i1
k=j
|
k
|
p
_
1
p
_
i1
k=j
|x
k
|
q
_
1
q
.
Since
i1
k=j
|
k
|
p
n1
k=1
|
k
|
p
and
i1
k=j
|x
k
|
q
n1
k=1
|x
k
|
q
,
for all 1 j < i n, then by (5.70) and (5.71) we deduce the desired inequality (5.69).
To prove the sharpness of the constant, let us assume that (5.69) holds with a constant C > 0.
That is,
(5.72)
i=1
p
i
i
x
i
i=1
p
i
i
n
i=1
p
i
x
i
1j<in
(i j) p
i
p
j
_
n1
k=1
|
k
|
p
_1
p
_
n1
k=1
|x
k
|
q
_1
q
.
Note that, for n = 2, we have
i=1
p
i
i
x
i
i=1
p
i
i
2
i=1
p
i
x
i
= p
1
p
2
|
1
2
| |x
1
x
2
|
and
1j<i2
(i j) p
i
p
j
_
1
k=1
|
k
|
p
_1
p
_
1
k=1
|x
k
|
q
_1
q
= p
1
p
2
|
1
2
| |x
1
x
2
| .
Therefore, from (5.72), we obtain
p
1
p
2
|
1
2
| |x
1
x
2
| Cp
1
p
2
|
1
2
| |x
1
x
2
|
for all
1
=
2
, x
1
= x
2
, p
1
p
2
> 0, giving C 1.
We are able now to state the following reverse of the (CBS) inequality.
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) be two sequences of real numbers with
a
i
= 0, (i = 1, . . . , n) . Then one has the inequality
0
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
_
n1
k=1
_
b
k
a
k
_
p
_1
p
_
n1
k=1
_
b
k
a
k
_
q
_1
q
1j<in
(i j) a
2
i
a
2
j
,
where p > 1,
1
p
+
1
q
= 1.
The constant C = 1 is sharp in the above sense.
Proof. Follows by Lemma 5.33 for
p
i
=
a
2
i
n
k=1
a
2
k
,
i
= x
i
=
b
i
a
i
, i {1, . . . , n} .
The following corollary is a natural consequence of Theorem 5.34 for p = q = 2.

92 S.S. DRAGOMIR
Corollary 5.35. With the assumptions of Theorem 5.34 for a and

b, we have
0
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
n1
k=1
_
b
k
a
k
_
1j<in
(i j) a
2
i
a
2
j
.
5.12. A Reverse Inequality Via an Andrica-Badea Result. The following result is due to
Andrica and Badea [15, p. 16].
Lemma 5.36. Let x = (x
1
, . . . , x
n
) I
n
= [m, M]
n
be a sequence of real numbers and let S
be the subset of {1, . . . , n} that minimises the expression
(5.73)
iS
p
i
1
2
P
n
,
where P
n
:=
n
i=1
p
i
> 0, p = (p
1
, . . . , p
n
) is a sequence of nonnegative real numbers. Then
(5.74) max
xI
n
_
_
1
P
n
n
i=1
p
i
x
2
i

_
1
P
n
n
i=1
p
i
x
i
_
2
_
_
=
iS
p
i
_
P
n
iS
p
i
_
(M m)
2
P
2
n
.
Proof. We shall follow the proof in [15, p. 161].
Dene
D
n
( x, p) :=
1
P
n
n
i=1
p
i
x
2
i

_
1
P
n
n
i=1
p
i
x
i
_
2
=
1
P
n
1i<jn
p
i
p
j
(x
i
x
j
)
2
.
Keeping in mind the convexity of the quadratic function, we have
D
n
( x + (1 ) y, p)
=
1
P
2
n
1i<jn
p
i
p
j
[x
i
+ (1 ) y
i
x
j
(1 ) y
j
]
2
=
1
P
2
n
1i<jn
p
i
p
j
[(x
i
x
j
) + (1 ) (y
i
y
j
)]
2
1
P
2
n
1i<jn
p
i
p
j
_
(x
i
x
j
)
2
+ (1 ) (y
i
y
j
)
2
= D
n
( x, p) + (1 ) D
n
( y, p) ,
hence D
n
(, p) is a convex function on I
n
.
Using a well known theorem (see for instance [16, p. 124]), we get that the maximum of
D
n
(, p) is attained on the boundary of I
n
.
Let
_
S,

S
_
be the partition of {1, . . . , n} such that the maximum of D
n
(, p) is obtained for
x
0
= (x
0
1
, . . . , x
0
n
) , where x
0
i
= m if i

S and x
0
i
= M if i S. In this case we have
D
n
( x
0
, p) =
1
P
2
n
1i<jn
p
i
p
j
(x
i
x
j
)
2
(5.75)
=
(M m)
2
P
2
n
iS
p
i
_
P
n
iS
p
i
_
.
The expression
iS
p
i
_
P
n
iS
p
i
_
is a maximum when the set S minimises the expression
iS
p
i
1
2
P
n
.
From (5.75) it follows that D
n
( x, p) is also a maximum and the proof of the above lemma is
complete.
The following reverse result of the (CBS) inequality holds.
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
with a
i
= 0 (i = 1, . . . , n) and
(5.76) < m
b
i
a
i
M < for each i {1, . . . , n} .
Let S be the subset of {1, . . . , n} that minimizes the expression
(5.77)
iS
a
2
i

1
2
n
i=1
a
2
i
,
and denote

S := {1, . . . , n} \ S. Then we have the inequality
0
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
(5.78)
(M m)
2
iS
a
2
i
S
a
2
i
1
4
(M m)
2
_
n
i=1
a
2
i
_
2
.
Proof. The proof of the second inequality in (5.78) follows by Lemma 5.36 on choosing p
i
=
a
2
i
, x
i
=
b
i
a
i
, i {1, . . . , n} .
94 S.S. DRAGOMIR
The third inequality is obvious as
iS
a
2
i
S
a
2
i
=
iS
a
2
i
_
n
j=1
a
2
j

iS
a
2
i
_
1
4
_
iS
a
2
i
+
n
j=1
a
2
j

iS
a
2
i
_
2
=
1
4
_
n
j=1
a
2
j
_
2
.
5.13. A Renement of Cassels Inequality. In 1914, P. Schweitzer [18] proved the following
result.
Theorem 5.38. If a =(a
1
, . . . , a
n
) is a sequence of real numbers such that 0 < m a
i

M < (i {1, . . . , n}) , then
(5.79)
_
1
n
n
i=1
a
i
__
1
n
n
i=1
1
a
i
_
(M +m)
2
4mM
.
In 1972, A. Lupa s [17] proved the following renement of Schweitzers result which gives
the best bound for n odd as well.
Theorem 5.39. With the assumptions in Theorem 5.38, one has
(5.80)
n
i=1
a
i
n
i=1
1
a
i
__
n
2
M +
_
n+1
2
m
_ __
n+1
2
M +
_
n
2
m
_
Mm
,
where [] is the integer part.
In 1988, Andrica and Badea [15] established a weighted version of Schweitzer and Lupa s
inequalities via the use of the following weighted version of the Grss inequality [15, Theorem
2].
Theorem 5.40. If m
1
a
i
M
1
, m
2
b
i
M
2
(i {1, . . . , n}) and S is the subset of
{1, . . . , n} which minimises the expression
(5.81)
iS
p
i
1
2
P
n
,
where P
n
:=
n
i=1
p
i
> 0, then
P
n
n
i=1
p
i
a
i
b
i
i=1
p
i
a
i
i=1
p
i
b
i
(5.82)
(M
1
m
1
) (M
2
m
2
)
iS
p
i
_
P
n
iS
p
i
_
.
1
4
P
2
n
(M
1
m
1
) (M
2
m
2
) .
Proof. Using the result in Lemma 5.36, Section 5.12, we have
(5.83)
1
P
n
n
i=1
p
i
a
2
i

_
1
P
n
n
i=1
p
i
a
i
_
2
(M
1
m
1
)
2
P
2
n
iS
p
i
_
P
n
iS
p
i
_
and
(5.84)
1
P
n
n
i=1
p
i
b
2
i

_
1
P
n
n
i=1
p
i
b
i
_
2
(M
2
m
2
)
2
P
2
n
iS
p
i
_
P
n
iS
p
i
_
and since
(5.85)
_
1
P
n
n
i=1
p
i
a
i
b
i
1
P
n
n
i=1
p
i
a
i
1
P
n
n
i=1
p
i
b
i
_
2
_
_
1
P
n
n
i=1
p
i
a
2
i

_
1
P
n
n
i=1
p
i
a
i
_
2
_
_
_
_
1
P
n
n
i=1
p
i
b
2
i

_
1
P
n
n
i=1
p
i
b
i
_
2
_
_
,
the rst part of (5.82) holds true.
The second part follows by the elementary inequality
ab
1
4
(a +b)
2
, a, b R
for the choices a :=
iS
p
i
, b := P
n
iS
p
i
.
We are now able to state and prove the result of Andrica and Badea [15, Theorem 4], which
is related to Schweitzers inequality.
Theorem 5.41. If 0 < m a
i
M < , i {1, . . . , n} and S is a subset of {1, . . . , n} that
minimises the expression
iS
p
i
P
n
2
,
_
n
i=1
p
i
a
i
__
n
i=1
p
i
a
i
_
P
2
n
+
(M m)
2
Mm
iS
p
i
_
P
n
iS
p
i
_
(5.86)
(M +m)
2
4Mm
P
2
n
.
Proof. We shall follow the proof in [15]. We obtain from Theorem 5.39 with b
i
=
1
a
i
, m
1
= m,
M
1
= m, m
2
=
1
M
, M
2
=
1
m
, the following estimate
P
2
n

n
i=1
p
i
a
i
n
i=1
p
i
a
i
(M m)
_
1
m

1
M
_
iS
p
i
_
P
n
iS
p
i
_
,
that leads, in a simple manner, to (5.86).
We may now prove the following reverse result for the weighted (CBS) inequality that
improves the additive version of Cassels inequality.
96 S.S. DRAGOMIR
Theorem 5.42. Let a =(a
1
, . . . , a
n
) ,

b =(b
1
, . . . , b
n
) be two sequences of positive real num-
bers with the property that
(5.87) 0 < m
b
i
a
i
M < for each i {1, . . . , n} ,
and p =(p
1
, . . . , p
n
) a sequence of nonnegative real numbers such that P
n
:=
n
i=1
p
i
> 0. If
S is a subset of {1, . . . , n} that minimises the expression
(5.88)
iS
p
i
a
i
b
i
1
2
n
i=1
p
i
a
i
b
i

n
i=1
p
i
a
2
i
n
i=1
p
i
b
2
i

_
n
i=1
p
i
a
i
b
i
_
2
(5.89)
(M m)
2
Mm
iS
p
i
a
i
b
i
_
n
i=1
p
i
a
i
b
i
iS
p
i
a
i
b
i
_
(M m)
2
4Mm
_
n
i=1
p
i
a
i
b
i
_
2
.
Proof. Applying Theorem 5.41 for a
i
= x
i
, p
i
= q
i
x
i
we may deduce the inequality
(5.90)
n
i=1
q
i
x
2
i
n
i=1
q
i
_
n
i=1
q
i
x
i
_
2
(M m)
2
Mm
iS
q
i
x
i
_
n
i=1
q
i
x
i
iS
q
i
x
i
_
,
provided q
i
0,
n
i=1
q
i
> 0, 0 < m x
i
M < , for i {1, . . . , n} and S is a subset of
{1, . . . , n} that minimises the expression
(5.91)
iS
q
i
x
i
1
2
n
i=1
q
i
x
i
.
Now, if in (5.90) we choose q
i
= p
i
a
2
i
, x
i
=
b
i
a
i
[m, M] for i {1, . . . , n} , we deduce the
desired result (5.89).
The following corollary provides a renement of Cassels inequality.
Corollary 5.43. With the assumptions of Theorem 5.42, we have the inequality
1
n
i=1
p
i
a
2
i
n
i=1
p
i
b
2
i
(
n
i=1
p
i
a
i
b
i
)
2
(5.92)
1 +
(M m)
2
Mm

iS
p
i
a
i
b
i
n
i=1
p
i
a
i
b
i
_
1
iS
p
i
a
i
b
i
n
i=1
p
i
a
i
b
i
_
(M +m)
2
4Mm
.
The case of the unweighted Cassels inequality is embodied in the following corollary as
well.
Corollary 5.44. Assume that a and

b satisfy (5.88). If S is a subset of {1, . . . , n} that minimises
the expression
(5.93)
iS
a
i
b
i
1
2
n
i=1
a
i
b
i

1
n
i=1
a
2
i
n
i=1
b
2
i
(
n
i=1
a
i
b
i
)
2
(5.94)
1 +
(M m)
2
Mm

iS
a
i
b
i
n
i=1
a
i
b
i
_
1
iS
a
i
b
i
n
i=1
a
i
b
i
_
(M +m)
2
4Mm
.
In particular, we may obtain the following renement of the Plya-Szegs inequality.
Corollary 5.45. Assume that
(5.95) 0 < a a
i
A < , 0 < b b
i
B < for i {1, . . . , n} .
If S is a subset of {1, . . . , n} that minimises the expression (5.93), then one has the inequality
1
n
i=1
a
2
i
n
i=1
b
2
i
(
n
i=1
a
i
b
i
)
2
(5.96)
1 +
(AB ab)
2
abAB

iS
a
i
b
i
n
i=1
a
i
b
i
_
1
iS
a
i
b
i
n
i=1
a
i
b
i
_
(AB +ab)
2
4abAB
.
5.14. Two Reverse Results Via Diaz-Metcalf Results. In [19], J.B. Diaz and F.T. Metcalf
proved the following inequality for sequences of complex numbers.
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
such that a
k
= 0, k {1, . . . , n} and
(5.97) m Re
_
b
k
a
k
_
+ Im
_
b
k
a
k
_
M, m Re
_
b
k
a
k
_
Im
_
b
k
a
k
_
M,
where m, M R and k {1, . . . , n} . Then one has the inequality
n
k=1
|b
k
|
2
+mM
n
k=1
|a
k
|
2
(m+M) Re
_
n
k=1
a
k
b
k
_
(5.98)
|M +m|
k=1
a
k
b
k
.
Using the above result we may state and prove the following reverse inequality.
98 S.S. DRAGOMIR
Theorem 5.47. If a and

b are as in (5.97) and m, M > 0, then one has the inequality
n
k=1
|a
k
|
2
n
k=1
|b
k
|
2
(M +m)
2
4mM
_
Re
n
k=1
a
k
b
k
_
2
(5.99)
(M +m)
2
4mM
k=1
a
k
b
k
2
.
Proof. Using the elementary inequality
p
2
+
1
q
2
2pq, > 0, p, q 0
we have
(5.100)

mM
n
k=1
|a
k
|
2
+
1
mM
n
k=1
|b
k
|
2
2
_
n
k=1
|a
k
|
2
n
k=1
|b
k
|
2
_1
2
.
On the other hand, by (5.98), we have
1
mM
n
k=1
|b
k
|
2
+
mM
n
k=1
|a
k
|
2
(M +m)
mM
Re
_
n
k=1
a
k
b
k
_
(5.101)
M +m
mM
k=1
a
k
b
k
.
Combining (5.100) and (5.101), we deduce the desired result (5.99).
The following corollary is a natural consequence of the above lemma.

b and m, M satisfy the hypothesis of Theorem 5.47, then
0
_
n
i=1
|a
i
|
2
n
i=1
|b
i
|
2
_1
2
i=1
a
i
b
i
(5.102)
_
n
i=1
|a
i
|
2
n
i=1
|b
i
|
2
_1
2
Re
_
n
k=1
a
k
b
k
_
m
_
2
2
mM
Re
_
n
k=1
a
k
b
k
_
m
_
2
2
mM
k=1
a
k
b
k
and
0
n
i=1
|a
i
|
2
n
i=1
|b
i
|
2
i=1
a
i
b
i
2
(5.103)
i=1
|a
i
|
2
n
i=1
|b
i
|
2
Re
_
n
i=1
a
i
b
i
_
2
(5.104)
(M m)
2
4mM
Re
_
n
i=1
a
i
b
i
_
2
(5.105)
(M m)
2
4mM
i=1
a
i
b
i
2
. (5.106)
Another result obtained by Diaz and Metcalf in [19] is the following one.
Lemma 5.49. Let a,

b, m and M be complex numbers such that
Re (m) + Im(m) Re
_
b
k
a
k
_
+ Im
_
b
k
a
k
_
Re (M) + Im(M) ; (5.107)
Re (m) Im(m) Re
_
b
k
a
k
_
Im
_
b
k
a
k
_
Re (M) Im(M) ;
for each k {1, . . . , n} . Then
n
k=1
|b
k
|
2
+ Re
_
m

M
_
n
k=1
|a
k
|
2
Re
_
(M +m)
n
k=1
a
k
b
k
_
(5.108)
|M +m|
k=1
a
k
b
k
.
The following reverse result for the (CBS) inequality may be stated as well.
Theorem 5.50. With the assumptions in Lemma 5.49, and if Re
_
m

M
_
> 0, then we have the
inequality:
_
n
k=1
|a
k
|
2
n
k=1
|b
k
|
2
_1
2
Re
_
(M +m)
n
k=1
a
k
b
k
2
_
Re
_
m

M
_1
2
(5.109)
|M +m|
n
k=1
a
k
b
k
2
_
Re
_
m

M
_1
2
.
The proof is similar to the one in Theorem 5.47 and we omit the details.
Remark 5.51. Similar additive versions may be stated. They are left as an exercise for the
interested reader.
5.15. Some Reverse Results Via the

Cebyev Functional. For x= (x
1
, . . . , x
n
) , y= (y
1
, . . . , y
n
)
two sequences of real numbers and p = (p
1
, . . . , p
n
) a sequence of nonnegative real numbers
with
n
i=1
p
i
= 1, dene the

Cebyev functional
(5.110) T
n
( p; x, y) :=
n
i=1
p
i
x
i
y
i
i=1
p
i
x
i
n
i=1
p
i
y
i
.
100 S.S. DRAGOMIR
For x and p as above consider the norms:
x
:= max
i=1,n
|x
i
|
x
p,
:=
_
n
i=1
p
i
|x
i
|
_1
, [1, ).
The following result holds [20].
Theorem 5.52. Let x, y, p be as above and c= (c, . . . , c) a constant sequence with c R. Then
one has the inequalities
0 |T
n
( p; x, y)| (5.111)
_
y y
,p
p,1
inf
cR
x c
;
y y
,p
p,
inf
cR
x c
p,
, > 1,
1
+
1
= 1;
y y
,p
inf
cR
x c
p,1
;
_
y y
,p
p,1
min
_
x
, x x
,p
_
;
y y
,p
p,
min
_
x
p,
, x x
,p
p,
_
,
> 1,
1
+
1
= 1;
y y
,p
min
_
x
p,1
, x x
,p
p,1
_
;
where
x
,p
:=
n
i=1
p
i
x
i
, y
,p
:=
n
i=1
p
i
y
i
and x
,p
, y
,p
are the sequences with all components equal to x
,p
, y
,p
.
Proof. Firstly, let us observe that for any c R, one has Sonins identity
T
n
( p; x, y) = T
n
( p; x c, y y
,p
) (5.112)
=
n
i=1
p
i
(x
i
c)
_
y
i
j=1
p
j
y
j
_
.
Taking the modulus and using Hlders inequality, we have
|T
n
( p; x, y)|
n
i=1
p
i
|x
i
c| |y
i
y
,p
| (5.113)
_
max
i=1,n
|x
i
c|
n
i=1
p
i
|y
i
y
,p
|
(
n
i=1
p
i
|x
i
c|
)
1
n
i=1
p
i
|y
i
y
,p
|
_1
,
> 1,
1
+
1
= 1;
n
i=1
p
i
|x
i
c| max
i=1,n
|y
i
y
,p
|
=
_
_
x c
y y
,p
p,1
;
x c
p,
y y
,p
p,
, > 1,
1
+
1
= 1;
x c
p,1
y y
,p
.
Taking the inf over c R in (5.113), we deduce the second inequality in (5.111).
Since
inf
cR
x c
p,

_
_
_
x
p,
,
x x
,p
p,
for any [1, ]
the last part of (5.110) is also proved.
For p and x as above, dene
T
n
( p; x) :=
n
i=1
p
i
x
2
i

_
n
i=1
p
i
x
i
_
2
.
The following corollary holds [20].
Corollary 5.53. With the above assumptions we have
0 |T
n
( p; x)| (5.114)
_
x x
,p
p,1
inf
cR
x c
;
x x
,p
p,
inf
cR
x c
p,
, > 1,
1
+
1
= 1;
x x
,p
inf
cR
x c
p,1
;
_
x x
,p
p,1
min
_
x
, x x
,p
_
;
x x
,p
p,
min
_
x
p,
, x x
,p
p,
_
,
> 1,
1
+
1
= 1;
x x
,p
min
_
x
p,1
, x x
,p
p,1
_
.
Remark 5.54. If p
i
:=
1
n
, i = 1, . . . , n, then from Theorem 5.52 and Corollary 5.53 we recap-
ture the results in [22].
The following reverse of the (CBS) inequality holds [20].
102 S.S. DRAGOMIR
Theorem 5.55. Let a,

b be two sequences of real numbers with a
i
= 0, i {1, . . . , n} . Then
0
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
(5.115)
inf
cR
_
max
i=1,n
b
i
a
i
c
_
n
i=1
_
|a
i
|
k=1
a
k
a
k
a
i
b
k
b
i
i=1
_
|a
i
|
k=1
a
k
a
k
a
i
b
k
b
i
_
max
i=1,n
b
i
a
i
max
i=1,n
b
i
a
i
n
k=1
a
k
b
k
n
k=1
a
2
k
.
Proof. By Corollary 5.53, we may state that
0 T
n
( p; x) (5.116)
x x
,p
p,1
inf
cR
x c
x x
,p
p,1
_
_
_
x
,
x x
,p
.
For the choices
p
i
=
a
2
i
n
k=1
a
2
k
, x
i
=
b
i
a
i
, i = 1, . . . , n;
we get
T
n
( p; x) =
n
i=1
a
2
i
n
i=1
b
2
i
(
n
i=1
a
i
b
i
)
2
(
n
k=1
a
2
k
)
2
,
x x
,p
p,1
=
n
i=1
p
i
x
i
j=1
p
j
x
j
=
1
n
k=1
a
2
k
n
i=1
a
2
i
b
i
a
i
n
k=1
a
2
k
n
j=1
a
j
b
j
=
1
(
n
k=1
a
2
k
)
2
n
i=1
a
i
b
i
n
k=1
a
2
k
a
2
i
n
j=1
a
j
b
j
=
1
(
n
k=1
a
2
k
)
2
n
i=1
|a
i
|
k=1
a
k
(a
k
b
i
a
i
b
k
)
=
1
(
n
k=1
a
2
k
)
2
n
i=1
|a
i
|
k=1
a
k
a
k
a
i
b
k
b
i
,
x c
= max
i=1,n
b
i
a
i
c
, x
= max
i=1,n
b
i
a
i
and
x x
,p
= max
i=1,n
b
i
a
i
n
j=1
a
j
b
j
n
k=1
a
2
k
.
Utilising the inequality (5.116) we deduce the desired result (5.115).
The following result also holds [20].
Theorem 5.56. With the assumption in Theorem 5.55 and if > 1,
1
+
1
= 1, then we have
the inequality:
0
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
(5.117)
_
_
n
i=1
_
_
|a
i
|
2
k=1
a
k
a
k
a
i
b
k
b
i
_
_
_
_
1
inf
cR
_
n
i=1
|a
i
|
2
|b
i
ca
i
|
_1
_
_
n
i=1
_
_
|a
i
|
2
k=1
a
k
a
k
a
i
b
k
b
i
_
_
_
_
1
_
_
n
i=1
|a
i
|
2
|b
i
|
_1
n
k=1
a
2
k
_
n
i=1
|a
i
|
2
n
k=1
a
k
a
k
a
i
b
k
b
i
_1
.
0 T
n
( p; x) (5.118)
x x
,p
p,
inf
cR
x c
p,
x x
,p
p,

_
_
_
x
p,
,
x x
,p
p,
,
for > 1,
1
+
1
= 1.
For the choices
p
i
=
a
2
i
n
k=1
a
2
k
, x
i
=
b
i
a
i
, i = 1, . . . , n;
we get
x x
,p
p,
=
_
_
n
i=1
p
i
x
i
j=1
p
j
x
j
_
_
1
=
_
_
n
i=1
a
2
i
n
k=1
a
2
k
b
i
n
k=1
a
2
k
a
i
n
j=1
a
j
b
j
a
i
n
k=1
a
2
k
_
_
1
=
1
(
n
k=1
a
2
k
)
1+
1
_
_
n
i=1
|a
i
|
2
k=1
a
k
a
k
a
i
b
k
b
i
_
_
1
,
x c
p,
=
_
n
i=1
p
i
|x
i
c|
_1
=
1
(
n
k=1
a
2
k
)
1
_
n
i=1
|a
i
|
2
|b
i
ca
i
|
_1
,
104 S.S. DRAGOMIR
x
p,
=
1
(
n
k=1
a
2
k
)
1
_
n
i=1
|a
i
|
2
|b
i
|
_1
and
x x
,p
p,
=
1
(
n
k=1
a
2
k
)
1+
1
_
n
i=1
|a
i
|
2
k=1
a
k
a
k
a
i
b
k
b
i
_1
.
Utilising the inequality (5.118), we deduce the desired result (5.117).
Finally, the following result also holds [20].
Theorem 5.57. With the assumptions in Theorem 5.55 we have the following reverse of the
(CBS) inequality:
0
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
(5.119)
max
i=1,n
b
i
a
i
n
k=1
a
2
k

n
j=1
a
j
b
j
inf
cR
_
n
i=1
|a
i
| |b
i
ca
i
|
_
max
i=1,n
b
i
a
i
n
k=1
a
2
k

n
j=1
a
j
b
j
_
n
i=1
|a
i
b
i
|
1
n
k=1
a
2
k
n
i=1
|a
i
|
k=1
a
k
a
k
a
i
b
k
b
i
.
0 T
n
( p; x) (5.120)
x x
,p
inf
cR
x c
p,1
x x
,p
_
_
_
x
p,1
,
x x
,p
p,1
.
For the choices
p
i
=
a
2
i
n
k=1
a
2
k
, x
i
=
b
i
a
i
, i = 1, . . . , n;
we get
x x
,p
= max
i=1,n
x
i
j=1
p
j
x
j
= max
i=1,n
b
i
a
i
n
j=1
a
j
b
j
n
k=1
a
2
k
=
1
n
k=1
a
2
k
max
i=1,n
b
i
a
i
n
k=1
a
2
k

n
j=1
a
j
b
j
,
x c
p,1
=
n
i=1
p
i
|x
i
c|
=
n
i=1
a
2
i
n
k=1
a
2
k
b
i
a
i
c
=
1
n
k=1
a
2
k
n
i=1
|a
i
| |b
i
ca
i
| ,
x
p,1
=
n
i=1
p
i
|x
i
| =
n
i=1
a
2
i
n
k=1
a
2
k
b
i
a
i
=
1
n
k=1
a
2
k
n
i=1
|a
i
b
i
|
and
x x
,p
p,1
=
1
(
n
k=1
a
2
k
)
2
n
i=1
|a
i
|
k=1
a
k
a
k
a
i
b
k
b
i
.
Utilising the inequality (5.120) we deduce (5.119).
5.16. Another Reverse Result via a Grss Type Result. The following Grss type inequality
has been obtained in [21].
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
) be two sequences of real numbers and
assume that there are , R such that
(5.121) < a
i
< for each i {1, . . . , n} .
Then for any p = (p
1
, . . . , p
n
) a nonnegative sequence with the property that
n
i=1
p
i
= 1, one
has the inequality
(5.122)
i=1
p
i
a
i
b
i
i=1
p
i
a
i
n
i=1
p
i
b
i
1
2
( )
n
i=1
p
i
b
i
k=1
p
k
b
k
.
The constant
1
2
is sharp in the sense that it cannot be replaced by a smaller constant.
Proof. We will give here a simpler direct proof based on Sonins identity. A simple calculation
shows that:
(5.123)
n
i=1
p
i
a
i
b
i
i=1
p
i
a
i
n
i=1
p
i
b
i
=
n
i=1
p
i
_
a
i
+
2
_
_
b
i
k=1
p
k
b
k
_
.
By (5.121) we have
a
i
+
2

2
for all i {1, . . . , n}
and thus, by (5.123), on taking the modulus, we get
i=1
p
i
a
i
b
i
i=1
p
i
a
i
n
i=1
p
i
b
i
i=1
p
i
a
i
+
2
b
i
k=1
p
k
b
k
1
2
( )
n
i=1
p
i
b
i
k=1
p
k
b
k
.
106 S.S. DRAGOMIR
To prove the sharpness of the constant
1
2
, let us assume that (5.122) holds with a constant c > 0,
i.e.,
(5.124)
i=1
p
i
a
i
b
i
i=1
p
i
a
i
n
i=1
p
i
b
i
c ( )
n
i=1
p
i
b
i
k=1
p
k
b
k
.
provided a
i
satises (5.121).
If we choose n = 2 in (5.124) and take into account that
2
i=1
p
i
a
i
b
i
i=1
p
i
a
i
2
i=1
p
i
b
i
= p
1
p
2
(a
1
a
2
) (b
1
b
2
)
provided p
1
+p
2
= 1, p
1
, p
2
[0, 1] , and since
2
i=1
p
i
b
i
k=1
p
k
b
k
= p
1
|(p
1
+p
2
) b
1
p
1
b
1
p
2
b
2
| +p
2
|(p
1
+p
2
) b
2
p
1
b
1
p
2
b
2
|
= 2p
1
p
2
|b
1
b
2
|
we deduce by (5.124)
(5.125) p
1
p
2
|a
1
a
2
| |b
1
b
2
| 2c ( ) |b
1
b
2
| p
1
p
2
.
If we assume that p
1
, p
2
= 0, b
1
= b
2
and a
1
= , a
2
= , then by (5.125) we deduce c
1
2
,
which proves the sharpness of the constant
1
2
.
The following corollary is a natural consequence of the above lemma.
Corollary 5.59. Assume that a = (a
1
, . . . , a
n
) satises the assumption (5.121) and p is a
probability sequence. Then
(5.126) 0
n
i=1
p
i
a
2
i

_
n
i=1
p
i
a
i
_
2
1
2
( )
n
i=1
p
i
a
i
k=1
p
k
a
k
.
The constant
1
2
is best possible in the sense mentioned above.
The following reverse of the (CBS) inequality may be stated.
Theorem 5.60. Assume that x = (x
1
, . . . , x
n
) and y = (y
1
, . . . , y
n
) are sequences of real
numbers with y
i
= 0 (i = 1, . . . , n) . If there exists the real numbers m, M such that
(5.127) m
x
i
y
i
M for each i {1, . . . , n} ,
0
n
i=1
x
2
i
n
i=1
y
2
i

_
n
i=1
x
i
y
i
_
2
(5.128)
1
2
(M m)
n
i=1
|y
i
|
k=1
y
k

x
i
y
i
x
k
y
k
.
Proof. If we choose p
i
=
y
2
i
n
k=1
y
2
k
, a
i
=
x
i
y
i
for i = 1, . . . , n and = m, = M in (5.126), we
deduce
n
i=1
x
2
i
n
k=1
y
2
k
_
1
n
k=1
y
2
k
n
i=1
x
i
y
i
_
2
1
2
(M m)
1
n
k=1
y
2
k
n
i=1
y
2
i
x
i
y
i
n
k=1
y
2
k
n
k=1
x
k
y
k
=
1
2
(M m)
1
(
n
k=1
y
2
k
)
2
n
i=1
|y
i
|
x
i
n
k=1
y
2
k
y
i
n
k=1
x
k
y
k
=
1
2
(M m)
1
(
n
k=1
y
2
k
)
2
n
i=1
|y
i
|
k=1
y
k

x
i
y
i
x
k
y
k
.
REFERENCES
[1] R. FRUCHT, Sobre algunas disigualdades: Observacin relative a la solucin del Problema No. 21,
indicada par el Ing. Ernesto M. Saleme (in Spanish), Math. Notae, 3 (1943), 4146.
[2] G.S. WATSON, Serial Correlation in Regression Analysis, Ph.D. Thesis, Dept. of Experimental
Statistics, North Carolina State College, Raleigh; Univ. of North Carolina, Mimograph Ser., No.
49, 1951.
[3] G.S. WATSON, Serial correlation in regression analysis, I, Biometrika, 42 (1955), 327341.
[4] G.S. WATSON, A method for discovering Kantorovich-type inequalities and a probabilistic inter-
pretation, Linear Algebra Appl., 97 (1987), 211217.
[5] G.S. WATSON, G. ALPARGU and G.P.H. STYAN, Some comments on six inequalities associated
with the inefciency of ordinary least squares with one regressor, Linear Algebra and its Appl., 264
(1997), 1354.
[6] G. PLYA and G. SZEG, Aufgaben und Lehrstze ans der Analysis, Band I: Reihen, Integral-
rechnung, Funktiontheorie (in German), 4th Ed., Springer Verlag, Berlin, 1970 (Original version:
Julius Springer, Berlin, 1925).
[7] G. PLYA and G. SZEG, Problems and Theorems in Analysis, Volume 1: Series, Integral Calcu-
lus, Theory of Functions (in English), translated from german by D. Aeppli, corrected printing of
the revised translation of the fourth German edition, Springer Verlag, New York, 1972.
[8] W. GREUB AND W. RHEINBOLDT, On a generalisation of an inequality of L.V. Kantorovich,
Proc. Amer. Math. Soc., 10 (1959), 407415.
[9] S.S. DRAGOMIR, A generalisation of Cassels and Greub-Reinboldts inequalities in inner product
spaces, RGMIA Res. Rep. Coll., 6 (2003), Supplement. [ONLINE: http://rgmia.vu.edu.
au/v6(E).html]
[10] S.S. DRAGOMIR, A counterpart of Schwartzs inequality in inner product spaces, RGMIA Res.
Rep. Coll., 6 (2003), Supplement, Article 18. [ONLINE: http://rgmia.vu.edu.au/v6(E)
.html]
[11] S.S. DRAGOMIR AND G.L. BOOTH, Grss-Lupa s type inequality and its applications for the
estimation of p-moments of guessing mappings, Math. Comm., 5 (2000), 117126.
108 S.S. DRAGOMIR
[12] S.S. DRAGOMIR, Some counterpart inequalities in terms of forward difference for
(CBS) inequality, submitted.
[13] S.S. DRAGOMIR, A Grss type inequality for sequences of vectors in normed linear spaces and
applications, submitted.
[14] S.S. DRAGOMIR, Another Grss type inequality for sequences of vectors in normed linear spaces
and applications, J. Comp. Anal. & Applic., 4(2) (2002), 155172.
[15] D. ANDRICA AND C. BADEA, Grss inequality for positive linear functionals, Periodica Mat.
Hungarica, 19(2) (1988), 155167.
[16] A.W. ROBERTS AND D.E. VARBERG, Convex Functions, Academic Press, New York, 1973.
[17] A. LUPA S, A remark on the Schweitzer and Kantorovich inequalities, Publ. Elek. Fak. Univ.
Beograde, Ser. Mat. i Fiz., 383 (1972), 1314.
[18] P. SCHWEITZER, An inequality about the arithmetic mean (Hungarian), Math. Phys. Lapok (Bu-
dapest), 23 (1914), 257261.
[19] J.B. DIAZ AND F.T. METCALF, Stronger forms of a class of inequalities of G. Plya G. Szeg
and L.V. Kantorovich, Bull. Amer. Math. Soc., 69 (1963), 415418.
[20] P. CERONE AND S.S. DRAGOMIR, New inequalities for the

Cebyev functional involving two
ntuples of real numbers and applications, RGMIA Res. Rep. Coll., 5(3) (2002), Article 4. [ON-
LINE: http://rgmia.vu.edu.au/v5n3.html]
[21] P. CERONE AND S.S. DRAGOMIR, Arenement of the Grss inequality and applications, RGMIA
Res. Rep. Coll., 5(2) (2002), Article 2. [ONLINE: http://rgmia.vu.edu.au/v5n2.html]
[22] S.S. DRAGOMIR, New inequalities for the weighted

Cebyev functionals and applications for the
(CBS) inequality, in preparation.
[23] O. SHISHA AND B. MOND, Bounds on differences of means, Inequalities, Academic Press Inc.,
New York, 1967, 293308.
[24] D. ZAGIER, A converse to Cauchys inequality, Amer. Math. Month., 102(10) (1995), 919920.
[25] D. ZAGIER, An inequality converse to that of Cauchy (Dutch), Indag. Math., 39(4) (1997), 349
351.
[26] J. PE
CARI
C, The inequalities of Zagier and

Cebyev, Arch. Math., (Basel), 64(5) (1995), 415417.
[27] H. ALZER, On a converse Cauchy inequality of D. Zagier, Arch. Math., (Basel), 58(2) (1995),
157159.
6. RELATED INEQUALITIES
6.1. Ostrowskis Inequality for Real Sequences. In 1951, A.M. Ostrowski [2, p. 289] gave
the following result related to the (CBS) inequality for real sequences (see also [1, p. 92]).
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
) be two non-proportional sequences
of real numbers. Let x = (x
1
, . . . , x
n
) be a sequence of real numbers such that
(6.1)
n
i=1
a
i
x
i
= 0 and
n
i=1
b
i
x
i
= 1.
Then
(6.2)
n
i=1
x
2
i

n
i=1
a
2
i
n
i=1
a
2
i
n
i=1
b
2
i
(
n
i=1
a
i
b
i
)
2
with equality if and only if
(6.3) x
k
=
b
k
n
i=1
a
2
i
a
k
n
i=1
a
i
b
i
n
i=1
a
2
i
n
i=1
b
2
i
(
n
i=1
a
i
b
i
)
2
for any k {1, . . . , n} .
Let
(6.4) A =
n
i=1
a
2
i
, B =
n
i=1
b
2
i
, C =
n
i=1
a
i
b
i
and
(6.5) y
i
=
Ab
i
Ca
i
AB C
2
for any i {1, . . . , n} .
It is easy to see that the sequence y = (y
1
, . . . , y
n
) as dened by (6.5) satises (6.1).
Any sequence x = (x
1
, . . . , x
n
) that satises (6.1) fullls the equality
n
i=1
x
i
y
i
=
n
i=1
x
i
(Ab
i
Ca
i
)
AB C
2
=
A
AB C
2
;
so, in particular
n
i=1
y
2
i
=
A
AB C
2
.
Any sequence x = (x
1
, . . . , x
n
) that satises (6.1) therefore satises
(6.6)
n
i=1
x
2
i

n
i=1
y
2
i
=
n
i=1
(x
i
y
i
)
2
0,
and thus
n
i=1
x
2
i

n
i=1
y
2
i
=
A
AB C
2
From (6.6) it follows that equality holds in (6.1) iff x
i
= y
i
for each i {1, . . . , n}, and the
theorem is completely proved.
110 S.S. DRAGOMIR
6.2. Ostrowskis Inequality for Complex Sequences. The following result that points out a
natural generalisation of Ostrowskis inequality for complex numbers holds [3].
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) and x = (x
1
, . . . , x
n
) be sequences of
complex numbers. If a and

b, where

b =
_
b
1
, . . . ,
b
n
_
, are not proportional and
n
i=1
x
i
a
i
= 0; (6.7)
i=1
x
i
b
i
= 1, (6.8)
(6.9)
n
i=1
|x
i
|
2
n
i=1
|a
i
|
2
n
i=1
|a
i
|
2
n
i=1
|b
i
|
2
n
i=1
a
i
b
i
2
with equality iff
(6.10) x
i
=
_
b
i
n
k=1
b
k
a
k
n
k=1
|a
k
|
2
a
i
_
, i {1, . . . , n}
and C with
(6.11) || =
n
k=1
|a
k
|
2
n
k=1
|a
k
|
2
n
k=1
|b
k
|
2
n
k=1
a
k
b
k
2
.
Proof. Recall the (CBS) inequality for complex sequences
(6.12)
n
k=1
|u
k
|
2
n
k=1
|v
k
|
2
k=1
u
k
v
k
2
with equality iff there is a complex number C such that
(6.13) u
k
= v
k
, k = 1, . . . , n.
If we apply (6.12) for
u
k
= z
k

n
i=1
z
i
c
i
n
i=1
|c
i
|
2
c
k
,
v
k
= d
k

n
i=1
d
i
c
i
n
i=1
|c
i
|
2
c
k
, where c = 0 and c,
d, z C
n
,
we have
(6.14)
n
k=1
z
k

n
i=1
z
i
c
i
n
i=1
|c
i
|
2
c
k
2
n
k=1
d
k

n
i=1
d
i
c
i
n
i=1
|c
i
|
2
c
k
k=1
_
z
k

n
i=1
z
i
c
i
n
i=1
|c
i
|
2
c
k
__
d
k

n
i=1
d
i
c
i
n
i=1
|c
i
|
2
c
k
_
2
with equality iff there is a C such that
(6.15) z
k
=
n
i=1
z
i
c
i
n
i=1
|c
i
|
2
c
k
+
_
d
k

n
i=1
d
i
c
i
n
i=1
|c
i
|
2
c
k
_
.
Since a simple calculation shows that
n
k=1
z
k

n
i=1
z
i
c
i
n
i=1
|c
i
|
2
c
k
2
=
n
k=1
|z
k
|
2
n
k=1
|c
k
|
2
|
n
k=1
z
k
c
k
|
2
_
n
k=1
|c
k
|
2
_
2
,
n
k=1
d
k

n
i=1
d
i
c
i
n
i=1
|c
i
|
2
c
k
2
=
n
k=1
|d
k
|
2
n
k=1
|c
k
|
2
|
n
k=1
d
k
c
k
|
2
_
n
k=1
|c
k
|
2
_
2
and
n
k=1
_
z
k

n
i=1
z
i
c
i
n
i=1
|c
i
|
2
c
k
__
d
k

n
i=1
d
i
c
i
n
i=1
|c
i
|
2
c
k
_
=
n
k=1
z
k

d
k

n
k=1
|c
k
|
2
n
k=1
z
k
c
k

n
k=1
c
k

d
k
_
n
i=1
|c
i
|
2
_
2
then by (6.12) we deduce
(6.16)
_
_
n
k=1
|z
k
|
2
n
k=1
|c
k
|
2
k=1
z
k
c
k
2
_
_
_
_
n
k=1
|d
k
|
2
n
k=1
|c
k
|
2
k=1
d
k
c
k
2
_
_
k=1
z
k

d
k

n
k=1
|c
k
|
2
k=1
z
k
c
k
n
k=1
c
k

d
k
2
with equality iff there is a C such that (6.15) holds.
If a, x,
b satisfy (6.7) and (6.8), then by (6.16) and (6.15) for the choices z = x, c = a and
d =

b, we deduce (6.9) with equality iff there is a C such that
x
k
=
_
b
k

n
i=1
a
i
b
i
n
i=1
|a
i
|
2
a
k
_
,
and, by (6.8),
(6.17)
k=1
_
b
k

n
i=1
a
i
b
i
n
i=1
|a
i
|
2
a
k
_
b
k
= 1.
Since (6.17) is clearly equivalent to (6.15), the theorem is completely proved.
6.3. Another Ostrowskis Inequality. In his book from 1951, [2, p. 130], A.M. Ostrowski
proved the following inequality as well (see also [1, p. 94]).
Theorem 6.3. Let a,
b, x be sequences of real numbers so that a = 0 and

n
k=1
x
2
k
= 1 (6.18)
n
k=1
a
k
x
k
= 0. (6.19)
Then
(6.20)
n
k=1
a
2
k
n
k=1
b
2
n
(
n
k=1
a
k
b
k
)
2
n
k=1
a
2
k
_
n
k=1
b
k
x
k
_
2
.
112 S.S. DRAGOMIR
If a and

b are non-proportional, then equality holds in (6.20) iff
(6.21) x
k
= q
b
k
n
i=1
a
2
i
a
k
n
i=1
a
i
b
i
(
n
k=1
a
2
k
)
1
2
_
n
i=1
a
2
i
n
i=1
b
2
i
(
n
i=1
a
i
b
i
)
2
_1
2
,
k {1, . . . , n} , q {1, 1} .
We may extend this result for sequences of complex numbers as follows [4].
Theorem 6.4. Let a,
b, x be sequences of complex numbers so that a = 0, a,

b are not propor-
tional, and
n
k=1
|x
k
|
2
= 1 (6.22)
n
k=1
x
k
a
k
= 0. (6.23)
Then
(6.24)
n
k=1
|a
k
|
2
n
k=1
|b
k
|
2
n
k=1
a
k
b
k
n
k=1
|a
k
|
2

k=1
x
k
b
k
2
.
The equality holds in (6.24) iff
(6.25) x
k
=
_
b
k

n
i=1
b
i
a
i
n
i=1
|a
i
|
2
a
k
_
, k {1, . . . , n} ;
where C is such that
(6.26) || =
_
n
k=1
|a
k
|
2
_
1
2
_
n
k=1
|a
k
|
2
n
k=1
|b
k
|
2
n
k=1
a
k
b
k
2
_1
2
.
Proof. In Subsection 6.2, we proved the following inequality:
(6.27)
_
_
n
k=1
|z
k
|
2
n
k=1
|c
k
|
2
k=1
z
k
c
k
2
_
_
_
_
n
k=1
|d
k
|
2
n
k=1
|c
k
|
2
k=1
d
k
c
k
2
_
_
k=1
z
k

d
k

n
k=1
|c
k
|
2
k=1
z
k
c
k
n
k=1
c
k

d
k
2
for any z, c,
d sequences of complex numbers, with equality iff there is a C such that

(6.28) z
k
=
n
i=1
z
i
c
i
n
i=1
|c
i
|
2
c
k
+
_
d
k

n
i=1
d
i
c
i
n
i=1
|c
i
|
2
c
k
_
for each k {1, . . . , n} .
If in (6.27) we choose z = x, c = a and

d =

b and take into consideration that (6.22) and
(6.23) hold, then we get
n
k=1
|x
k
|
2
n
k=1
|a
k
|
2
_
_
n
k=1
|b
k
|
2
n
k=1
|a
k
|
2
k=1
b
k
a
k
2
_
_
k=1
x
k
b
k
2
_
n
k=1
|a
k
|
2
_
2
By (6.28) the equality holds in (6.24) iff
x
k
=
_
b
k

n
i=1
b
i
a
i
n
i=1
|a
i
|
2
a
k
_
, k {1, . . . , n} .
Since x should satisfy (6.22), we get
1 =
n
k=1
|x
k
|
2
= ||
2
n
k=1
b
k

n
i=1
b
i
a
i
n
i=1
|a
i
|
2
a
k
2
= ||
2
_
n
k=1
|b
k
|
2
n
k=1
b
k
a
k
|
2
n
k=1
|a
k
|
2
_
from where we deduce that satises (6.26).
6.4. Fan and Todd Inequalities. In 1955, K. Fan and J. Todd [5] proved the following in-
equality (see also [1, p. 94]).
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
such that a
i
b
j
= a
j
b
i
for i = j. Then
(6.29)
n
i=1
a
2
i
(
n
i=1
a
2
i
) (
n
i=1
b
2
i
) (
n
i=1
a
i
b
i
)
2

_
n
2
_
n
i=1
_
_
_
n
j=1
j=i
a
j
a
j
b
i
a
i
b
j
_
_
_
2
.
Dene
x
i
:=
_
n
2
_
1
j=i
a
j
a
j
b
i
a
i
b
j
(1 i n) .
The terms in the sum on the right-hand side
n
i=1
x
i
a
i
=
_
n
2
_
1 n
i=1
_
_
_
n
j=1
j=i
a
i
a
j
a
j
b
i
a
i
b
j
_
_
_
can be grouped in pairs of the form
_
n
2
_
1
_
a
i
a
j
a
j
b
i
a
i
b
j
+
a
j
a
i
a
i
b
j
a
j
b
i
_
(i = j)
and the sum of each such pair vanishes.
Hence, we deduce
n
i=1
a
i
x
i
= 0 and
n
i=1
b
i
x
i
= 1.
Applying Ostrowskis inequality (see Section 6.1) we deduce the desired result (6.29).
A weighted version of the result is also due to K. Fan and J. Todd [5] (see also [1, p. 95]).
We may state the result as follows.
114 S.S. DRAGOMIR
Theorem 6.6. Let p
ij
(i, j {1, . . . , n} , i = j) be real numbers such that
(6.30) p
ij
= p
ji
, for any i, j {1, . . . , n} with i = j.
Denote P :=
1i<jn
p
ij
and assume that P = 0. Then for any two sequences of real numbers
a = (a
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
) satisfying a
i
b
j
= a
j
b
i
(i = j) , we have
(6.31)
n
i=1
a
2
i
n
i=1
a
2
i
n
i=1
b
2
i
(
n
i=1
a
i
b
i
)
2

1
P
2
n
i=1
_
_
_
n
j=1
j=i
p
ij
a
j
a
j
b
i
a
i
b
j
_
_
_
2
.
6.5. Some Results for Asynchronous Sequences. If S (R) is the linear space of real se-
quences, S
+
(R) is the subset of nonnegative sequences and P
f
(N) denotes the set of nite
parts of N, then for the functional T : P
f
(N) S
+
(R) S
2
(R) R,
(6.32) T
_
I, p, a,
b
_
:=
_
iI
p
i
a
2
i
iI
p
i
b
2
i
_1
2
iI
p
i
a
i
b
i
we may state the following result [6, Theorem 3].

Theorem 6.7. If |a| = (|a
i
|)
iN
and
= (|b
i
|)
iN
are asynchronous, i.e.,
(|a
i
| |a
j
|) (|b
i
| |b
j
|) 0
for all i, j N, then
(6.33) T
_
I, p, a,
b
_
iI
p
i
|a
i
|
iI
p
i
|b
i
|
iI
p
i
iI
p
i
|a
i
b
i
| 0.
Consider the inequalities
_
iI
p
i
iI
p
i
a
2
i
_1
2
iI
p
i
|a
i
|
and
_
iI
p
i
iI
p
i
b
2
i
_1
2
iI
p
i
|b
i
|
which by multiplication give
_
iI
p
i
a
2
i
iI
p
i
b
2
i
_1
2
iI
p
i
|a
i
|
iI
p
i
|b
i
|
iI
p
i
.
Now, by the denition of T and by

Cebyevs inequality for asynchronous sequences, we have
T
_
I, p, a,
b
_
iI
p
i
|a
i
|
iI
p
i
|b
i
|
iI
p
i
iI
p
i
a
i
b
i
iI
p
i
|a
i
|
iI
p
i
|b
i
|
iI
p
i
iI
p
i
|a
i
| |b
i
|
0
and the theorem is proved.
The following result also holds [6, Theorem 4].
Theorem 6.8. If |a| and
are synchronous, i.e., (|a

i
| |a
j
|) (|b
i
| |b
j
|) 0 for all i, j N,
(6.34) 0 T
_
I, p, a,
b
_
T
_
I, p, a
b, 1
_
,
where 1 = (e
i
)
iN
, e
i
= 1, i N.
Proof. We have, by

Cebyevs inequality for the synchronous sequences a
2
= (a
2
i
)
iN
and
b
2
= (b
2
i
)
iN
, that
T
_
I, p, a,
b
_
=
_
iI
p
i
a
2
i
iI
p
i
b
2
i
_1
2
iI
p
i
a
i
b
i
iI
p
i
a
2
i
b
2
i
iI
p
i
_1
2
iI
p
i
a
i
b
i
= T
_
I, p, a
b, 1
_
and the theorem is proved.
6.6. An Inequality via AGH Mean Inequality. The following result holds [6, Theorem
5].
Theorem 6.9. Let a and

b be sequences of positive real numbers. Dene
(6.35)
i
=
a
2
i
b
2
i
iI
a
2
i
iI
b
2
i
where i I and I is a nite part of N. Then one has the inequality

_
iI
a
i
b
i
_
2
iI
a
2
i
iI
b
2
i
iI
_
a
i
b
i
_
i
_ 1
iI
a
2
i
iI
b
2
i
(6.36)
iI
a
2
i
iI
b
2
i
iI
a
3
i
b
i
iI
b
3
i
a
i
.
The equality holds in all the inequalities from (6.36) iff there exists a positive number k > 0
such that a
i
= kb
i
for all i I.
We will use the AGHinequality
(6.37)
1
P
I
iI
p
i
x
i

_
iI
x
p
i
i
_ 1
P
I
P
I
iI
p
i
x
i
,
where p
i
> 0, x
i
0 for all i I, where P
I
:=
iI
p
i
> 0.
We remark that the equality holds in (6.37) iff x
i
= x
j
for each i, j I.
Choosing p
i
= a
2
i
and x
i
=
b
i
a
i
(i I) in (6.37), then we get
(6.38)
iI
a
i
b
i
iI
a
2
i
iI
_
b
i
a
i
_
a
2
i
iI
a
2
i
iI
a
2
i
iI
a
3
i
b
i
116 S.S. DRAGOMIR
and by p
i
= b
2
i
and x
i
=
a
i
b
i
, we also have
(6.39)
iI
a
i
b
i
iI
b
2
i
iI
_
a
i
b
i
_
b
2
i
iI
b
2
i
iI
b
2
i
iI
b
3
i
a
i
.
If we multiply (6.38) with (6.39) we easily deduce the desired inequality (6.36).
The case of equality follows by the same case in the arithmetic mean geometric mean
harmonic mean inequality. We omit the details.
The following corollary holds [6, Corollary 5.1].
Corollary 6.10. With a and

b as above, one has the inequality
(6.40)
_
_
iI
a
3
i
b
i
iI
b
3
i
a
i
_
iI
a
i
b
i
_
2
_
_
1
2
iI
a
2
i
iI
b
2
i
_
iI
a
i
b
i
_
2
.
The equality holds in (6.40) iff there is a k > 0 such that a
i
= kb
i
, i {1, . . . , n} .
6.7. A Related Result via Jensens Inequality for Power Functions. The following result
also holds [6, Theorem 6].
Theorem 6.11. Let a and

b be sequences of positive real numbers and p 1. If I is a nite
part of N, then one has the inequality
(6.41)
_
iI
a
i
b
i
_
2
iI
a
2
i
iI
b
2
i
iI
a
2p
i
b
p
i
iI
a
p
i
b
2p
i
iI
a
2
i
iI
b
2
i
_1
p
.
The equality holds in (6.41) if and only if there exists a k > 0 such that a
i
= kb
i
for all i I.
If p (0, 1) , the inequality in (6.41) reverses.
By Jensens inequality for the convex mapping f : R
+
R,
f (x) = x
p
, p 1
one has
(6.42)
_
iI
p
i
x
i
P
I
_
p
iI
p
i
x
p
i
P
I
,
where P
I
:=
iI
p
i
, p
i
> 0, x
i
0, i I. The equality holds in (6.42) iff x
i
= x
j
for all
i, j I.
Now, choosing in (6.42) p
i
= a
2
i
, x
i
=
b
i
a
i
, we get
(6.43)
iI
a
i
b
i
iI
a
2
i
iI
a
2p
i
b
p
i
iI
a
2
i
_1
p
and for p
i
= b
2
i
, x
i
=
a
i
b
i
, the inequality (6.42) also gives
(6.44)
iI
a
i
b
i
iI
b
2
i
iI
a
p
i
b
2p
i
iI
b
2
i
_1
p
.
By multiplying the inequalities (6.43) and (6.44), we deduce the desired result from (6.42).
The case of equality follows by the fact that in (6.42) the equality holds iff (x
i
)
iI
is constant.
If p (0, 1) , then a reverse inequality holds in (6.42) giving the corresponding result in
(6.41).
Remark 6.12. If p = 2, then (6.41) becomes the (CBS) inequality.
6.8. Inequalities Derived fromthe Double Sums Case. Let A = (a
ij
)
i,j=1,n
and B = (b
ij
)
i,j=1,n
be two matrices of real numbers. The following inequality is known as the (CBS) inequality
for double sums
(6.45)
_
n
i,j=1
a
ij
b
ij
_
2
i,j=1
a
2
ij
n
i,j=1
b
2
ij
with equality iff there is a real number r such that a
ij
= rb
ij
for any i, j {1, . . . , n} .
The following inequality holds [7, Theorem 5.2].
1
, . . . , a
n
) and

b = (b
1
, . . . , b
n
) be sequences of real numbers. Then
(6.46)
_
n
k=1
a
k
_
2
+
_
n
k=1
b
k
_
2
2n
n
k=1
a
k
b
k
n
n
k=1
_
a
2
k
+b
2
k
_
2
n
k=1
a
k
n
k=1
b
k
.
Proof. We shall follow the proof from [7].
Applying (6.45) for a
ij
= a
i
b
j
, b
ij
= b
i
a
j
and taking into account that
n
i,j=1
(a
i
b
j
) (b
i
a
j
) = 2n
n
k=1
a
k
b
k

_
n
k=1
a
k
_
2
_
n
k=1
b
k
_
2
n
i,j=1
(a
i
b
j
)
2
= n
n
k=1
_
a
2
k
+b
2
k
_
2
n
k=1
a
k
n
k=1
b
k
and
n
i,j=1
(b
i
a
j
)
2
= n
n
k=1
_
a
2
k
+b
2
k
_
2
n
k=1
a
k
n
k=1
b
k
,
we may deduce the desired inequality (6.46).
The following result also holds [7, Theorem 5.3].
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
) , c = (c
1
, . . . , c
n
) and

d = (d
1
, . . . , d
n
)
be sequences of real numbers. Then one has the inequality:
(6.47)
_
_
det
_
_
n
i=1
a
i
c
i
n
i=1
a
i
d
i
n
i=1
b
i
c
i
n
i=1
b
i
d
i
_
_
_
_
2
det
_
_
n
i=1
a
2
i
n
i=1
a
i
b
i
n
i=1
a
i
b
i
n
i=1
b
2
i
_
_
det
_
_
n
i=1
c
2
i
n
i=1
c
i
d
i
n
i=1
c
i
d
i
n
i=1
d
2
i
_
_
.
Applying (6.45) for a
ij
= a
i
b
j
a
j
b
i
, b
ij
= c
i
d
j
c
j
d
i
and using Cauchy-Binets identity [1,
p. 85]
(6.48)
1
2
n
i,j=1
(a
i
b
j
a
j
b
i
) (c
i
d
j
c
j
d
i
) =
n
i=1
a
i
c
i
n
i=1
b
i
d
i
_
n
i=1
a
i
d
i
__
n
i=1
b
i
c
i
_
118 S.S. DRAGOMIR
and Lagranges identity [1, p. 84]
(6.49)
1
2
n
i,j=1
(a
i
b
j
a
j
b
i
)
2
=
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
,
we deduce the desired result (6.47).
6.9. A Functional Generalisation for Double Sums. The following result holds [7, Theorem
5.5].
Theorem 6.15. Let A be a subset of real numbers R, f : A R and a = (a
1
, . . . , a
n
),
b = (b
1
, . . . , b
n
) sequences of real numbers with the property that
(i) a
k
b
i
, a
2
i
, b
2
k
A for any i, k {1, . . . , n} ;
(ii) f (a
2
k
) , f (b
2
k
) 0 for any k {1, . . . , n} ;
(iii) f
2
(a
k
b
i
) f (a
2
k
) f (b
2
i
) for any i, k {1, . . . , n} .
(6.50)
_
n
k,i=1
f (a
k
b
i
)
_
2
n
2
n
k=1
f
_
a
2
k
_
n
k=1
f
_
b
2
k
_
.
Using the assumption (iii) and the (CBS) inequality for double sums, we have
k,i=1
f (a
k
b
i
)
k,i=1
|f (a
k
b
i
)| (6.51)
k,i=1
_
f
_
a
2
k
_
f
_
b
2
i
_1
2
_
_
_
_
n
k,i=1
_
f
_
a
2
k
_1
2
_
2
_
n
k,i=1
_
f
_
b
2
i
_1
2
_
2
_
_
_
1
2
=
_
n
k,i=1
f
_
a
2
k
_
n
k,i=1
f
_
b
2
i
_
_1
2
= n
_
n
k=1
f
_
a
2
k
_
_1
2
_
n
k=1
f
_
b
2
k
_
_1
2
Corollary 6.16. Let A, f and a be as above. If
(i) a
k
a
i
A for any i, k {1, . . . , n} ;
(ii) f (a
2
k
) 0 for any k {1, . . . , n} ;
(iii) f
2
(a
k
a
i
) f (a
2
k
) f (a
2
i
) for any i, k {1, . . . , n} ,
(6.52)
k,i=1
f (a
k
a
i
)
n
n
k=1
f
_
a
2
k
_
.
The following particular inequalities also hold [7, p. 23].
(1) If : N N is Eulers indicator and s (n) denotes the sum of all relatively prime
numbers including and less than n, then for any a = (a
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
)
sequences of natural numbers, one has the inequalities
_
n
k,i=1
(a
k
b
i
)
_
2
n
2
n
k=1
_
a
2
k
_
n
k=1
_
b
2
k
_
; (6.53)
n
k,i=1
(a
k
a
i
) n
n
k=1
_
a
2
k
_
; (6.54)
_
n
k,i=1
s (a
k
b
i
)
_
2
n
2
n
k=1
s
_
a
2
k
_
n
k=1
s
_
b
2
k
_
; (6.55)
n
k,i=1
s (a
k
a
i
) n
n
k=1
s
_
a
2
k
_
. (6.56)
(2) If a > 1 and a = (a
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
_
n
k,i=1
exp
a
(a
k
b
i
)
_
2
n
2
n
k=1
exp
a
_
a
2
k
_
n
k=1
exp
a
_
b
2
k
_
; (6.57)
n
k,i=1
exp
a
(a
k
a
i
) n
n
k=1
exp
a
_
a
2
k
_
; (6.58)
(3) If a and

b are sequences of real numbers such that a
k
, b
k
(1, 1) (k {1, . . . , n}) ,
then one has the inequalities:
_
n
k,i=1
1
(1 a
k
b
i
)
m
_
2
n
2
n
k=1
1
(1 a
2
k
)
m
n
k=1
1
(1 b
2
k
)
m
, (6.59)
n
k,i=1
1
(1 a
k
a
i
)
m
n
n
k=1
1
(1 a
2
k
)
m
, (6.60)
where m > 0.
6.10. A (CBS) Type Result for Lipschitzian Functions. The following result was obtained
in [8, Theorem].
Theorem 6.17. Let f : I R R be a Lipschitzian function with the constant M, i.e., it
satises the condition
(6.61) |f (x) f (y)| M|x y| for any x, y I.
120 S.S. DRAGOMIR
If a = (a
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
) are sequences of real numbers with a
i
b
j
I for any
i, j {1, . . . , n} , then
0
i,j=1
f (a
i
b
j
) |f (a
i
b
j
)|
n
i,j=1
|f (a
j
b
i
)| f (a
i
b
j
)
(6.62)
i,j=1
f
2
(a
j
b
i
)
n
i,j=1
f (a
i
b
j
) f (a
j
b
i
)
M
2
_
_
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
_
_
.
Since f is Lipschitzian with the constant M, we have
(6.63) 0 ||f (a
i
b
j
)| |f (a
j
b
i
)|| |f (a
i
b
j
) f (a
j
b
i
)| M|a
i
b
j
a
j
b
i
|
for any i, j {1, . . . , n} , giving
0 |(|f (a
i
b
j
)| |f (a
j
b
i
)|) (f (a
i
b
j
) f (a
j
b
i
))| (6.64)
(f (a
i
b
j
) f (a
j
b
i
))
2
M
2
(a
i
b
j
a
j
b
i
)
2
for any i, j {1, . . . , n} .
The inequality (6.64) is obviously equivalent to
|f (a
i
b
j
)| f (a
i
b
j
) +|f (a
j
b
i
)| f (a
j
b
i
) |f (a
i
b
j
)| f (a
i
b
j
) |f (a
j
b
i
)| f (a
j
b
i
)
(6.65)
f
2
(a
i
b
j
) 2f (a
i
b
j
) f (a
j
b
i
) +f
2
(a
j
b
i
)
M
2
_
a
2
i
b
2
j
2a
i
b
i
a
j
b
j
+a
2
j
b
2
i
_
for any i, j {1, . . . , n} . Summing over i and j from 1 to n in (6.65) and taking into account
that:
n
i,j=1
|f (a
i
b
j
)| f (a
i
b
j
) =
n
i,j=1
|f (a
j
b
i
)| f (a
i
b
j
) ,
n
i,j=1
|f (a
i
b
j
)| f (a
j
b
i
) =
n
i,j=1
|f (a
j
b
i
)| f (a
i
b
j
) ,
n
i,j=1
f
2
(a
i
b
j
) =
n
i,j=1
f
2
(a
j
b
i
) ,
we deduce the desired inequality.
The following particular inequalities hold [8, p. 27 p. 28].
(1) Let x = (x
1
, . . . , x
n
) , y = (y
1
, . . . , y
n
) be sequences of real numbers such that 0
|x
i
| M
1
, 0 |y
i
| M
2
, i {1, . . . , n} . Then for any r 1 one has
0
n
i=1
x
2r
i
n
i=1
y
2r
i

_
n
i=1
|x
i
y
i
|
r
_
2
(6.66)
r
2
(M
1
M
2
)
2(r1)
_
_
n
i=1
x
2
i
n
i=1
y
2
i

_
n
i=1
|x
i
y
i
|
_
2
_
_
.
(2) If 0 < m
1
|x
i
| , 0 < m
2
|y
i
| , i {1, . . . , n} and r (0, 1) , then
0
n
i=1
x
2r
i
n
i=1
y
2r
i

_
n
i=1
|x
i
y
i
|
r
_
2
(6.67)
r
2
(m
1
m
2
)
2(r1)
_
_
n
i=1
x
2
i
n
i=1
y
2
i

_
n
i=1
|x
i
y
i
|
_
2
_
_
.
(3) If 0 |x
i
| M
1
, 0 |y
i
| M
2
, i {1, . . . , n} , then for any natural number k one
has
0
i=1
x
2k+1
i
|x
i
|
2k+1
n
i=1
y
2k+1
i
|y
i
|
2k+1
(6.68)
i=1
x
2k+1
i
|y
i
|
2k+1
n
i=1
y
2k+1
i
|x
i
|
2k+1
i=1
x
2(2k+1)
i
n
i=1
y
2(2k+1)
i

_
n
i=1
x
2k+1
i
y
2k+1
i
_
2
(2k + 1)
2
(M
1
M
2
)
4k
_
_
n
i=1
x
2
i
n
i=1
y
2
i

_
n
i=1
|x
i
y
i
|
_
2
_
_
.
(4) If 0 < m
1
x
i
, 0 < m
2
y
i
, for any i {1, . . . , n} , then one has the inequality
0 n
n
i=1
_
ln
_
x
i
y
i
__
2
_
n
i=1
ln
_
x
i
y
i
_
_
2
(6.69)
1
(m
1
m
2
)
2
_
_
n
i=1
x
2
i
n
i=1
y
2
i

_
n
i=1
x
i
y
i
_
2
_
_
.
6.11. An Inequality via Jensens Discrete Inequality. The following result holds [9].
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
a
i
= 0, i {1, . . . , n} . If f : I R R is a convex (concave) function on I and
b
i
a
i
I for
each i {1, . . . , n} , and w = (w
1
, . . . , w
n
) is a sequence of nonnegative real numbers, then
(6.70) f
_
n
i=1
w
i
a
i
b
i
n
i=1
w
i
a
2
i
_
()
n
i=1
w
i
a
2
i
f
_
b
i
a
i
_
n
i=1
w
i
a
2
i
.
Proof. We shall use Jensens discrete inequality for convex (concave) functions
(6.71) f
_
1
P
n
n
i=1
p
i
x
i
_
()
1
P
n
n
i=1
p
i
f (x
i
) ,
where p
i
0 with P
n
:=
n
i=1
p
i
> 0 and x
i
I for each i {1, . . . , n} .
If in (6.71) we choose x
i
=
b
i
a
i
and p
i
= w
i
a
2
i
, then by (6.71) we deduce the desired result
(6.70).
122 S.S. DRAGOMIR
Corollary 6.19. Let a and

b be sequences of positive real numbers and assume that w is as
above. If p (, 0) [1, ) (p (0, 1)) , then one has the inequality
(6.72)
_
n
i=1
w
i
a
i
b
i
_
p
()
_
n
i=1
w
i
a
2
i
_
p1
n
i=1
w
i
a
2p
i
b
p
i
.
Proof. Follows by Theorem 6.18 applied for convex (concave) function f : [0, ) R,
f (x) = x
p
, p (, 0) [1, ) (p (0, 1)) .
Remark 6.20. If p = 2, then by (6.72) we deduce the (CBS) inequality.
6.12. An Inequality via Lah-Ribari c Inequality. The following reverse of Jensens discrete
inequality was obtained in 1973 by Lah and Ribari c [10].
Lemma 6.21. Let f : I R be a convex function, x
i
[m, M] I for each i {1, . . . , n}
and p = (p
1
, . . . , p
n
) be a positive ntuple. Then
(6.73)
1
P
n
n
i=1
p
i
f (x
i
)
M
1
P
n
n
i=1
p
i
x
i
M m
f (m) +
1
P
n
n
i=1
p
i
x
i
m
M m
f (M) .
Proof. We observe for each i {1, . . . , n} , that
(6.74) x
i
=
(M x
i
) m+ (x
i
m) M
M m
.
If in the denition of convexity, i.e., , 0, + > 0
(6.75) f
_
a +b
+
_
f (a) +f (b)
+
we choose = M x
i
, = x
i
m, a = m and b = M, we deduce, by (6.75), that
f (x
i
) = f
_
(M x
i
) m+ (x
i
m) M
M m
_
(6.76)
(M x
i
) f (m) + (x
i
m) f (M)
M m
for each i {1, . . . , n} .
i
> 0 and sum over i from 1 to n, we deduce (6.73).
The following result holds.
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
a
i
= 0, i {1, . . . , n} . If I R Ris a convex (concave) function on I and
b
i
a
i
[m, M] I
for each i {1, . . . , n} and w = (w
1
, . . . , w
n
) is a sequence of nonnegative real numbers, then
(6.77)
n
i=1
w
i
a
2
i
f
_
b
i
a
i
_
n
i=1
w
i
a
2
i
()
M
n
i=1
w
i
a
i
b
i
n
i=1
w
i
a
2
i
M m
f (m) +
n
i=1
w
i
a
i
b
i
n
i=1
w
i
a
2
i
m
M m
f (M) .
Proof. Follows by Lemma 6.21 for the choices p
i
= w
i
a
2
i
, x
i
=
b
i
a
i
, i {1, . . . , n} .
Corollary 6.23. Let a = (a
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
numbers and such that
(6.78) 0 < m
b
i
a
i
M < for each i {1, . . . , n} .
If w = (w
1
, . . . , w
n
) is a sequence of nonnegative real numbers and p (, 0) [1, )
(p (0, 1)) , then one has the inequality
(6.79)
n
i=1
w
i
a
2p
i
b
p
i
+
Mm(M
p1
m
p1
)
M m
n
i=1
w
i
a
2
i
()
M
p
m
p
M m
n
i=1
w
i
a
i
b
i
.
Proof. If we write the inequality (6.77) for the convex (concave) function f (x) = x
p
, p
(, 0) [1, ) (p (0, 1)) , we get
n
i=1
w
i
a
2p
i
b
p
i
n
i=1
w
i
a
2
i
()
M
n
i=1
w
i
a
i
b
i
n
i=1
w
i
a
2
i
M m
m
p
+
n
i=1
w
i
a
i
b
i
n
i=1
w
i
a
2
i
m
M m
M
p
,
which, after elementary calculations, is equivalent to (6.79).
Remark 6.24. For p = 2, we get
(6.80)
n
i=1
w
i
b
2
i
+Mm
n
i=1
w
i
a
2
i
(M +m)
n
i=1
w
i
a
i
b
i
,
which is the well known Diaz-Metcalf inequality [11].
6.13. An Inequality via Dragomir-Ionescu Inequality. The following reverse of Jensens
inequality was proved in 1994 by S.S. Dragomir and N.M. Ionescu [12].
Lemma 6.25. Let f : I R Rbe a differentiable convex function on

I, x
i

I (i {1, . . . , n})
and p
i
0 (i {1, . . . , n}) such that P
n
:=
n
i=1
p
i
> 0. Then one has the inequality
0
1
P
n
n
i=1
p
i
f (x
i
) f
_
1
P
n
n
i=1
p
i
x
i
_
(6.81)
1
P
n
n
i=1
p
i
x
i
f
(x
i
)
1
P
n
n
i=1
p
i
x
i
1
P
n
n
i=1
p
i
f
(x
i
) .
Proof. Since f is differentiable convex on

I, one has
(6.82) f (x) f (y) (x y) f
(y) ,
for any x, y
I.
If we choose x =
1
P
n
n
i=1
p
i
x
i
and y = y
k
, k {1, . . . , n} , we get
(6.83) f
_
1
P
n
n
i=1
p
i
x
i
_
f (y
k
)
_
1
P
n
n
i=1
p
i
x
i
y
k
_
f
(y
k
) .
k
0 and summing over k from 1 to n, we deduce the desired result
(6.81).
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
a
i
= 0, i {1, . . . , n} . If f : I R R is a differentiable convex (concave) function on

I
and
b
i
a
i
I for each i {1, . . . , n} , and w = (w

1
, . . . , w
n
) is a sequence of nonnegative real
124 S.S. DRAGOMIR
numbers, then
0
n
i=1
w
i
a
2
i
n
i=1
w
i
a
2
i
f
_
b
i
a
i
_
_
n
i=1
w
i
a
2
i
_
2
f
_
n
i=1
w
i
a
i
b
i
n
i=1
w
i
a
2
i
_
(6.84)
i=1
w
i
a
2
i
n
i=1
w
i
a
i
b
i
f
_
b
i
a
i
_
i=1
w
i
a
i
b
i
n
i=1
w
i
a
2
i
f
_
b
i
a
i
_
.
Proof. Follows from Lemma 6.25 on choosing p
i
= w
i
a
2
i
, x
i
=
b
i
a
i
, i {1, . . . , n} .
Corollary 6.27. Let a = (a
1
, . . . , a
n
),

b = (b
1
, . . . , b
n
numbers with a
i
= 0, i {1, . . . , n} . If p [1, ), then one has the inequality
0
n
i=1
w
i
a
2
i
n
i=1
w
i
a
2p
i
b
p
i

_
n
i=1
w
i
a
2
i
_
2p
_
n
i=1
w
i
a
i
b
i
_
p
(6.85)
p
_
n
i=1
w
i
a
2
i
n
i=1
w
i
a
2p
i
b
i
i=1
w
i
a
i
b
i
n
i=1
w
i
a
3p
i
b
p1
i
_
.
If p (0, 1) , then
0
_
n
i=1
w
i
a
2
i
_
2p
_
n
i=1
w
i
a
i
b
i
_
i=1
w
i
a
2
i
n
i=1
w
i
a
2p
i
b
p
i
(6.86)
p
_
n
i=1
w
i
a
i
b
i
n
i=1
w
i
a
3p
i
b
p
i

n
i=1
w
i
a
2
i
n
i=1
w
i
a
2p
i
b
i
_
.
6.14. An Inequality via a Renement of Jensens Inequality. We will use the following
lemma which contains a renement of Jensens inequality obtained in [13].
Lemma 6.28. Let f : I R R be a convex function on the interval I and x
i
I, p
i
0
with P
n
:=
n
i=1
p
i
> 0. Then the following inequality holds:
f
_
1
P
n
n
i=1
p
i
x
i
_
1
P
k+1
n
n
i
1
,...,i
k+1
=1
p
i
1
p
i
k
f
_
x
i
1
+ +x
i
k+1
k + 1
_
(6.87)
1
P
k
n
n
i
1
,...,i
k
=1
p
i
1
p
i
k
f
_
x
i
1
+ +x
i
k
k
_

1
P
n
n
i=1
p
i
f (x
i
) ,
where k 1, k N.
The rst inequality follows by Jensens inequality for multiple sums
f
_
_
n
i
1
,...,i
k+1
=1
p
i
1
p
i
k+1
_
x
i
1
++x
i
k+1
k+1
_
n
i
1
,...,i
k+1
=1
p
i
1
p
i
k+1
_
_
=
n
i
1
,...,i
k+1
=1
p
i
1
p
i
k+1
f
_
x
i
1
++x
i
k+1
k+1
_
n
i
1
,...,i
k+1
=1
p
i
1
p
i
k+1
since
n
i
1
,...,i
k+1
=1
p
i
1
p
i
k+1
_
x
i
1
++x
i
k+1
k+1
_
n
i
1
,...,i
k+1
=1
p
i
1
p
i
k+1
= P
k
n
n
i=1
p
i
x
i
and
n
i
1
,...,i
k+1
=1
p
i
1
p
i
k+1
= P
k+1
n
.
Now, applying Jensens inequality for
y
1
=
x
i
1
+x
i
2
+x
i
k1
+x
i
k
k
, y
2
=
x
i
2
+x
i
3
+x
i
k
+x
i
k+1
k
,
, y
k+1
=
x
i
k+1
+x
i
1
+x
i
2
+ +x
i
k1
k
we have
f
_
y
1
+y
2
+ +y
k
+y
k+1
k + 1
_
f (y
1
) +f (y
2
) + +f (y
k
) +f (y
k+1
)
k + 1
,
which is equivalent to
(6.88) f
_
x
i
1
+ +x
i
k+1
k + 1
_
f
_
x
i
1
+x
i
2
+x
i
k1
+x
i
k
k
_
+ +f
_
x
i
k+1
+x
i
1
+x
i
2
++x
i
k1
k
_
k + 1
.
Multiplying (6.88) with the nonnegative real numbers p
i
1
, . . . , p
i
k+1
and summing over
i
1
, . . . , i
k+1
from 1 to n we deduce
n
i
1
,...,i
k+1
=1
p
i
1
p
i
k+1
f
_
x
i
1
+ +x
i
k+1
k + 1
_
(6.89)
1
k + 1
_
_
n
i
1
,...,i
k+1
=1
p
i
1
p
i
k+1
f
_
x
i
1
+ +x
i
k
k
_
+ +
n
i
1
,...,i
k+1
=1
p
i
1
p
i
k+1
f
_
x
i
k+1
+x
i
1
+x
i
2
+ +x
i
k1
k
_
_
_
= P
n
n
i
1
,...,i
k
=1
p
i
1
p
i
k
f
_
x
i
1
+ +x
i
k
k
_
which proves the second part of (6.87).
The following result holds.
126 S.S. DRAGOMIR
Theorem 6.29. Let f : I R R be a convex function on the interval I, a = (a
1
, . . . , a
n
),
b = (b
1
, . . . , b
n
) real numbers such that a
i
= 0, i {1, . . . , n} and
b
i
a
i
I, i {1, . . . , n} . If
w = (w
1
, . . . , w
n
) are positive real numbers, then
f
_
n
i=1
w
i
a
i
b
i
n
i=1
w
i
a
2
i
_
(6.90)
1
(
n
i=1
w
i
a
2
i
)
k+1
n
i
1
,...,i
k+1
=1
w
i
1
w
i
k+1
a
2
i
1
a
2
i
k+1
f
_
_
b
i
1
a
i
1
+ +
b
i
k+1
a
i
k+1
k + 1
_
_
1
(
n
i=1
w
i
a
2
i
)
k
n
i
1
,...,i
k
=1
w
i
1
w
i
k
a
2
i
1
a
2
i
k
f
_
_
b
i
1
a
i
1
+ +
b
i
k
a
i
k
k
_
_

1
n
i=1
w
i
a
2
i
n
i=1
w
i
a
2
i
f
_
b
i
a
i
_
.
The proof is obvious by Lemma 6.28 applied for p
i
= w
i
a
2
i
, x
i
=
b
i
a
i
, i {1, . . . , n} .

b and w be sequences of positive real numbers. If p (, 0) [1, )
(p (0, 1)) , then one has the inequalities
_
n
i=1
w
i
a
i
b
i
_
p
()
(
n
i=1
w
i
a
2
i
)
pk1
(k + 1)
p
n
i
1
,...,i
k+1
=1
w
i
1
w
i
k+1
a
2
i
1
a
2
i
k+1
_
b
i
1
a
i
1
+ +
b
i
k+1
a
i
k+1
_
p
()
(
n
i=1
w
i
a
2
i
)
pk
k
p
n
i
1
,...,i
k
=1
w
i
1
w
i
k
a
2
i
1
a
2
i
k
_
b
i
1
a
i
1
+ +
b
i
k
a
i
k
_
p
()

()
_
n
i=1
w
i
a
2
i
_
p1
n
i=1
w
i
a
2p
i
b
p
i
.
Remark 6.31. If p = 2, then we deduce the following renement of the (CBS) inequality
_
n
i=1
w
i
a
i
b
i
_
2
n
i=1
w
i
a
2
i
)
1k
(k + 1)
2
n
i
1
,...,i
k+1
=1
w
i
1
w
i
k+1
_
_
_
k+1
=1
b
i
k+1
j=1
j=
a
ij
_
_
_
2
n
i=1
w
i
a
2
i
)
2k
k
2
n
i
1
,...,i
k
=1
w
i
1
w
i
k
_
_
_
k
=1
b
i
j=1
j=
a
ij
_
_
_
2

1
4
n
i,j=1
w
i
w
j
(b
i
a
j
+a
i
b
j
)
2
i=1
w
i
a
2
i
n
i=1
w
i
b
2
i
.
6.15. Another Renement via Jensens Inequality. The following renement of Jensens
inequality holds (see [15]).
Lemma 6.32. Let f : [a, b] R be a differentiable convex function on (a, b) and x
i
(a, b) ,
p
i
0 with P
n
:=
n
i=1
p
i
> 0. Then one has the inequality
(6.91)
1
P
n
n
i=1
p
i
f (x
i
) f
_
1
P
n
n
i=1
p
i
x
i
_
1
P
n
n
i=1
p
i
f (x
i
) f
_
1
P
n
n
j=1
p
j
x
j
_
_
1
P
n
n
j=1
p
j
x
j
_
1
P
n
n
i=1
p
i
x
i
1
P
n
n
j=1
p
j
x
j
0.
Proof. Since f is differentiable convex on (a, b) , then for each x, y (a, b), one has the in-
equality
(6.92) f (x) f (y) (x y) f
(y) .
Using the properties of the modulus, we have
f (x) f (y) (x y) f
(y) = |f (x) f (y) (x y) f
(y)| (6.93)
||f (x) f (y)| |x y| |f
(y)||
for each x, y (a, b) .
If we choose y =
1
P
n
n
j=1
p
j
x
j
and x = x
i
, i {1, . . . , n} , then we have
(6.94) f (x
i
) f
_
1
P
n
n
j=1
p
j
x
j
_
_
x
i
1
P
n
n
j=1
p
j
x
j
_
f
_
1
P
n
n
j=1
p
j
x
j
_
f (x
i
) f
_
1
P
n
n
j=1
p
j
x
j
_
x
i
1
P
n
n
j=1
p
j
x
j
_
1
P
n
n
j=1
p
j
x
j
_
for any i {1, . . . , n} .

i
0, sum over i from 1 to n, and divide by P
n
> 0, we deduce
1
P
n
n
i=1
p
i
f (x
i
) f
_
1
P
n
n
j=1
p
j
x
j
_
1
P
n
n
i=1
p
i
_
x
i
1
P
n
n
j=1
p
j
x
j
_
f
_
1
P
n
n
j=1
p
j
x
j
_
1
P
n
n
i=1
p
i
f (x
i
) f
_
1
P
n
n
j=1
p
j
x
j
_
x
i
1
P
n
n
j=1
p
j
x
j
_
1
P
n
n
j=1
p
j
x
j
_
1
P
n
n
i=1
p
i
f (x
i
) f
_
1
P
n
n
j=1
p
j
x
j
_
1
P
n
n
i=1
p
i
x
i
1
P
n
n
j=1
p
j
x
j
_
1
P
n
n
j=1
p
j
x
j
_
.
128 S.S. DRAGOMIR
Since
1
P
n
n
i=1
p
i
_
x
i
1
P
n
n
j=1
p
j
x
j
_
= 0,
the inequality (6.91) is proved.
In particular, we have the following result for unweighted means.
Corollary 6.33. With the above assumptions for f and x
i
, one has the inequality
(6.95)
f (x
1
) + +f (x
n
)
n
f
_
x
1
+ +x
n
n
_
1
n
n
i=1
x
i
f
_
x
1
+ +x
n
n
_
_
x
1
+ +x
n
n
_
1
n
n
i=1
x
i
1
n
n
j=1
x
j
0.
Theorem 6.34. If a
i
, b
i
R, i {1, . . . , n} , then one has the inequality;
(6.96)
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
n
i=1
b
2
i
i=1
a
2
i
b
2
i
_
n
j=1
a
j
b
j
_
2
_
n
j=1
b
2
j
_
2
k=1
a
k
b
k
i=1
|b
i
|
j=1
b
j
a
i
b
i
a
j
b
j
0.
Proof. If we apply Lemma 6.32 for f (x) = x
2
, we get
(6.97)
1
P
n
n
i=1
p
i
x
2
i

_
1
P
n
n
i=1
p
i
x
i
_
2
1
P
n
n
i=1
p
i
x
2
i

_
1
P
n
n
j=1
p
j
x
j
_
2
1
P
n
n
k=1
p
k
x
k
1
P
n
n
i=1
p
i
x
i
1
P
n
n
j=1
p
j
x
j
0.
If in (6.97), we choose p
i
= b
2
i
, x
i
=
a
i
b
i
, i {1, . . . , n} , we get
(6.98)
n
i=1
a
2
i
n
i=1
b
2
i
n
i=1
a
i
b
i
)
2
(
n
i=1
b
2
i
)
2
n
i=1
b
2
i
n
i=1
b
2
i

a
2
i
b
2
i
n
j=1
a
j
b
j
n
j=1
b
2
j
_
2
n
k=1
a
k
b
k
n
i=1
b
2
i
n
i=1
b
2
i
a
i
b
i

n
j=1
a
j
b
j
_
n
j=1
b
2
j
n
i=1
b
2
i
,
6.16. An Inequality via Slaters Result. Suppose that I is an interval of real numbers with
interior

I
and f : I R is a convex function on I. Then f is continuous on

I
and has nite
left and right derivatives at each point of

I
. Moreover, if x, y
I
and x < y, then D
f (x)
D
+
f (x) D
f (y) D
+
f (y) which shows that both D
f and D
+
f are nondecreasing
functions on

I
. It is also known that a convex function must be differentiable except for at most
countably many points.
For a convex function f : I R, the subdifferential of f denoted by f is the set of all
functions : I [, ] such that
_
I
_
R and
(6.99) f (x) f (a) + (x a) (a) for any x, a I.
It is also well known that if f is convex on I, then f is nonempty, D
+
f, D
f f and if
f, then
(6.100) D
f (x) (x) D
+
f (x)
for every x
I
. In particular, is a nondecreasing function.
If f is differentiable convex on

I
, then f = {f
} .
The following inequality is well known in the literature as Slaters inequality [16].
Lemma 6.35. Let f : I R be a nondecreasing (nonincreasing) convex function, x
i
I,
p
i
0 with P
n
:=
n
i=1
p
i
> 0 and

n
i=1
p
i
(x
i
) = 0 where f. Then one has the
inequality:
(6.101)
1
P
n
n
i=1
p
i
f (x
i
) f
_
n
i=1
p
i
x
i
(x
i
)
n
i=1
p
i
(x
i
)
_
.
Proof. Firstly, observe that since, for example, f is nondecreasing, then (x) 0 for any x I
and thus
(6.102)
n
i=1
p
i
x
i
(x
i
)
n
i=1
p
i
(x
i
)
I,
since it is a convex combination of x
i
with the positive weights
x
i
(x
i
)
n
i=1
p
i
(x
i
)
, i = 1, . . . , n.
A similar argument applies if f is nonincreasing.
130 S.S. DRAGOMIR
Now, if we use the inequality (6.99), we deduce
(6.103) f (x) f (x
i
) (x x
i
) (x
i
) for any x, x
i
I, i = 1, . . . , n.
i
0 and summing over i from 1 to n, we deduce
(6.104) f (x)
1
P
n
n
i=1
p
i
f (x
i
) x
1
P
n
n
i=1
p
i
(x
i
)
1
P
n
n
i=1
p
i
x
i
(x
i
)
for any x I.
x =
n
i=1
p
i
x
i
(x
i
)
n
i=1
p
i
(x
i
)
,
which, we have proved that it belongs to I, we deduce the desired inequality (6.101).
If one would like to drop the assumption of monotonicity for the function f, then one can
state and prove in a similar way the following result.
Lemma 6.36. Let f : I R be a convex function, x
i
I, p
i
0 with P
n
> 0 and
n
i=1
p
i
(x
i
) = 0, where f. If
(6.105)
n
i=1
p
i
x
i
(x
i
)
n
i=1
p
i
(x
i
)
I,
then the inequality (6.101) holds.
The following result in connection to the (CBS) inequality holds.
Theorem 6.37. Assume that f : R
+
R is a convex function on R
+
:= [0, ), a
i
, b
i
0 with
a
i
= 0, i {1, . . . , n} and
n
i=1
a
2
i
_
b
i
a
i
_
= 0 where f.
(i) If f is monotonic nondecreasing (nonincreasing) in [0, ) then
(6.106)
n
i=1
a
2
i
_
b
i
a
i
_
i=1
a
2
i
f
_
_
n
i=1
a
i
b
i
_
b
i
a
i
_
n
i=1
a
2
i
_
b
i
a
i
_
_
_
.
(ii) If
(6.107)
n
i=1
a
i
b
i
_
b
i
a
i
_
n
i=1
a
2
i
_
b
i
a
i
_ 0,
then (6.106) also holds.
Remark 6.38. Consider the function f : [0, ) R, f (x) = x
p
, p 1. Then f is convex and
monotonic nondecreasing and (x) = px
p1
. Applying (6.106), we may deduce the following
inequality:
(6.108) p
_
n
i=1
a
3p
i
b
p1
i
_
p+1
i=1
a
2
i
_
n
i=1
a
2p
i
b
p
i
_
p
for p 1, a
i
, b
i
0, i = 1, . . . , n.
6.17. An Inequality via an Andrica-Ra sa Result. The following Jensen type inequality has
been obtained in [17] by Andrica and Ra sa.
Lemma 6.39. Let f : [a, b] R be a twice differentiable function and assume that
m = inf
t(a,b)
f
(t) > and M = sup

t(a,b)
f
(t) < .
If x
i
[a, b] and p
i
0 (i = 1, . . . , n) with
n
i=1
p
i
= 1, then one has the inequalities:
1
2
m
_
_
n
i=1
p
i
x
2
i

_
n
i=1
p
i
x
i
_
2
_
_
i=1
p
i
f (x
i
) f
_
n
i=1
p
i
x
i
_
(6.109)
1
2
M
_
_
n
i=1
p
i
x
2
i

_
n
i=1
p
i
x
i
_
2
_
_
.
Proof. Consider the auxiliary function f
m
(t) := f (t)
1
2
mt
2
. This function is twice differen-
tiable and f
m
(t) 0, t (a, b) , showing that f
m
is convex.
Applying Jensens inequality for f
m
, i.e.,
n
i=1
p
i
f
m
(x
i
) f
m
_
n
i=1
p
i
x
i
_
,
we deduce, by a simple calculation, the rst inequality in (6.109).
The second inequality follows in a similar way for the auxiliary function f
M
(t) =
1
2
Mt
2
f (t) . We omit the details.

The above result may be naturally used to obtain the following inequality related to the
(CBS) inequality.
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
the property that there exists , R such that
(6.110)
a
i
b
i
, for each i {1, . . . , n} ,
and b
i
= 0, i = 1, . . . , n. If f : (, ) R is twice differentiable and
m = inf
t(,)
f
(t) and M = sup

t(,)
f
(t) ,
1
2
m
_
_
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
_
_
(6.111)
i=1
b
2
i
n
i=1
b
2
i
f
_
a
i
b
i
_
_
n
i=1
b
2
i
_
2
f
_
n
i=1
a
i
b
i
n
i=1
b
2
i
_
1
2
M
_
_
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
_
_
.
132 S.S. DRAGOMIR
Proof. We may apply Lemma 6.39 for the choices p
i
=
b
2
i
n
k=1
b
2
k
and x
i
=
a
i
b
i
to get
1
2
m
_
n
i=1
a
2
i
n
k=1
b
2
k
_
n
i=1
a
i
b
i
n
k=1
b
2
k
_
2
_
n
i=1
b
2
i
f
_
a
i
b
i
_
n
k=1
b
2
k
f
_
n
i=1
a
i
b
i
n
k=1
b
2
k
_
1
2
M
_
n
i=1
a
2
i
n
k=1
b
2
k
_
n
i=1
a
i
b
i
n
k=1
b
2
k
_
2
_
giving the desired result (6.111).
Corollary 6.41. Assume that a,

b are sequences of nonnegative real numbers and
(6.112) 0 <
a
i
b
i
< for each i {1, . . . , n} .
Then for any p [1, ) one has the inequalities
1
2
p (p 1)
p2
_
_
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
_
_
(6.113)
i=1
b
2
i
_
n
i=1
a
p
i
b
2p
i
_
p
_
n
i=1
b
2
i
_
2p
_
n
i=1
a
i
b
i
_
p
1
2
p (p 1)
p2
_
_
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
_
_
if p [2, ) and
1
2
p (p 1)
p2
_
_
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
_
_
(6.114)
i=1
b
2
i
_
n
i=1
a
p
i
b
2p
i
_
p
_
n
i=1
b
2
i
_
2p
_
n
i=1
a
i
b
i
_
p
1
2
p (p 1)
p2
_
_
n
i=1
a
2
i
n
i=1
b
2
i

_
n
i=1
a
i
b
i
_
2
_
_
if p [1, 2] .
6.18. An Inequality via Jensens Result for Double Sums. The following result for convex
functions via Jensens inequality also holds [18].
Lemma 6.42. Let f : R R be a convex (concave) function and x = (x
1
, . . . , x
n
) , p =
(p
1
, . . . , p
n
) real sequences with the property that p
i
0 (i = 1, . . . , n) and
n
i=1
p
i
= 1. Then
one has the inequality:
(6.115) f
_
n
i=1
p
i
x
2
i
(
n
i=1
p
i
x
i
)
2
n
i=1
i
2
p
i
(
n
i=1
ip
i
)
2
_
()
1i<jn
p
i
p
j
_
j1
k,l=i
f (x
k
x
l
)
_
n
i=1
i
2
p
i
(
n
i=1
ip
i
)
2
,
where x
k
:= x
k+1
x
k
(k = 1, . . . , n 1) is the forward difference.
Proof. We have, by Jensens inequality for multiple sums that
f
_
n
i=1
p
i
x
2
i
(
n
i=1
p
i
x
i
)
2
n
i=1
i
2
p
i
(
n
i=1
ip
i
)
2
_
= f
_
1i<jn
p
i
p
j
(x
i
x
j
)
2
1i<jn
p
i
p
j
(j i)
2
_
(6.116)
= f
_
_
1i<jn
p
i
p
j
(j i)
2 (x
j
x
i
)
2
(ji)
2
1i<jn
p
i
p
j
(j i)
2
_
_
1i<jn
p
i
p
j
(j i)
2
f
_
(x
j
x
i
)
2
(ji)
2
_
1i<jn
p
i
p
j
(j i)
2
=: I.
On the other hand, for j > i, one has
(6.117) x
j
x
i
=
j1
k=i
(x
k+1
x
k
) =
j1
k=i
x
k
and thus
(x
j
x
i
)
2
=
_
j1
k=i
x
k
_
2
=
j1
k,l=i
x
k
x
l
.
Applying once more the Jensen inequality for multiple sums, we deduce
(6.118) f
_
(x
j
x
i
)
2
(j i)
2
_
= f
_
j1
k,l=i
x
k
x
l
(j i)
2
_
()
j1
k,l=i
f (x
k
x
l
)
(j i)
2
and thus, by (6.118), we deduce
I ()
1i<jn
p
i
p
j
(j i)
2
j1
k,l=i
f(x
k
x
l
)
(ji)
2
1i<jn
p
i
p
j
(j i)
2
(6.119)
=
1i<jn
p
i
p
j
_
j1
k,l=i
f (x
k
x
l
)
_
n
i=1
i
2
p
i
(
n
i=1
ip
i
)
2
,
and then, by (6.116) and (6.119), we deduce the desired inequality (6.115).
The following inequality connected with the (CBS) inequality may be stated.
Theorem 6.43. Let f : R R be a convex (concave) function and a = (a
1
, . . . , a
n
) ,

b =
(b
1
, . . . , b
n
) , w = (w
1
, . . . , w
n
) sequences of real numbers such that b
i
= 0, w
i
0 (i = 1, . . . , n)
and not all w
i
are zero. Then one has the inequality
(6.120) f
_
n
i=1
w
i
a
2
i
n
i=1
w
i
b
2
i
(
n
i=1
w
i
a
i
b
i
)
2
n
i=1
w
i
b
2
i
n
i=1
i
2
w
i
b
2
i
(
n
i=1
iw
i
b
i
)
2
_
()
1i<jn
w
i
w
j
b
2
i
b
2
j
_
j1
k,l=i
f
_
_
a
k
b
k
_

_
a
l
b
l
___
n
i=1
w
i
b
2
i
n
i=1
i
2
w
i
b
2
i
(
n
i=1
iw
i
b
i
)
2
.
Proof. Follows by Lemma 6.42 on choosing p
i
= w
i
b
2
i
and x
i
=
a
i
b
i
, i = 1, . . . , n. We omit the
details.
134 S.S. DRAGOMIR
6.19. Some Inequalities for the

Cebyev Functional. For two sequences of real numbers
a = (a
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) and p = (p
1
, . . . , p
n
) with p
i
0 (i {1, . . . , n}) and
n
i=1
p
i
= 1, consider the

Cebyev functional
(6.121) T
_
p, a,
b
_
:=
n
i=1
p
i
a
i
b
i
i=1
p
i
a
i
n
i=1
p
i
b
i
.
By Korkines identity [1, p. 242] one has the representation
T
_
p, a,
b
_
=
1
2
n
i,j=1
p
i
p
j
(a
i
a
j
) (b
i
b
j
) (6.122)
=
1i<jn
p
i
p
j
(a
j
a
i
) (b
j
b
i
) .
Using the (CBS) inequality for double sums one may state the following result
(6.123)
_
T
_
p, a,
b
_
2
T ( p, a, a) T
_
p,
b,
b
_
,
where, obviously
T ( p, a, a) =
1
2
n
i,j=1
p
i
p
j
(a
i
a
j
)
2
(6.124)
=
1i<jn
p
i
p
j
(a
j
a
i
)
2
.
Lemma 6.44. Assume that a = (a
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) are real numbers such that for
each i, j {1, . . . , n} , i < j, one has
(6.125) m(b
j
b
i
) a
j
a
i
M (b
j
b
i
) ,
where m, M are given real numbers.
If p = (p
1
, . . . , p
n
) is a nonnegative sequence with
n
i=1
p
i
= 1, then one has the inequality
(6.126) (m+M) T
_
p, a,
b
_
T ( p, a, a) +mMT
_
p,
b,
b
_
.
Proof. If we use the condition (6.125), we get
(6.127) [M (b
i
b
j
) (a
i
a
j
)] [(a
i
a
j
) m(b
i
b
j
)] 0
for i, j {1, . . . , n} , i < j.
If we multiply in (6.127), then, obviously, for any i, j {1, . . . , n} , i < j we have
(6.128) (a
j
a
i
)
2
+mM (b
j
b
i
)
2
(m+M) (a
j
a
i
) (b
j
b
i
) .
i
p
j
0, i, j {1, . . . , n} , i < j, summing over i and j, i < j from 1
to n and using the identities (6.122) and (6.124), we deduce the required inequality (6.125).
Theorem 6.45. If a,

b, p are as in Lemma 6.44 and M m > 0, then one has the inequality
providing a reverse for (6.123)
(6.129)
_
T
_
p, a,
b
_
2
4mM
(m+M)
2
T ( p, a, a) T
_
p,
b,
b
_
.
Proof. We use the following elementary inequality
(6.130) x
2
+
1
y
2
2xy, x, y 0, > 0
to get, for the choices
=
mM > 0, x =
_
T
_
p,
b,
b
_1
2
0, y = [T ( p, a, a)]
1
2
0
the following inequality:
(6.131)

mMT
_
p,
b,
b
_
+
1
mM
T ( p, a, a) 2
_
T
_
p,
b,
b
_1
2
[T ( p, a, a)]
1
2
.
Using (6.130) and (6.131), we deduce
(m+M)
2
mM
T
_
p, a,
b
_
_
T
_
p,
b,
b
_1
2
[T ( p, a, a)]
1
2
The following corollary also holds [14].
Corollary 6.46. With the assumptions of Theorem 6.45, we have:
0
_
T
_
p,
b,
b
_1
2
[T ( p, a, a)]
1
2
T
_
p, a,
b
_
(6.132)
m
_
2
2
mM
T
_
p, a,
b
_
and
0 T ( p, a, a) T
_
p,
b,
b
_
_
T
_
p, a,
b
_
2
(6.133)
(M m)
2
4mM
_
T
_
p, a,
b
_
2
.
The following result is useful in practical applications [14].
Theorem 6.47. Let f, g : [, ] R be continuous on [, ] and differentiable on (, ) with
g
(x) = 0 for x (, ) . Assume

(6.134) < = inf
x(,)
f
(x)
g
(x)
, sup
x(,)
f
(x)
g
(x)
= < .
If x is a real sequence with x
i
[, ] and x
i
= x
j
for i = j and if we denote by f ( x) :=
(f (x
1
) , . . . , f (x
n
)) , then we have the inequality:
(6.135) ( + ) T ( p, f ( x) , g ( x)) T ( p, f ( x) , f ( x)) +T ( p, g ( x) , g ( x))
for any p with p
i
0 (i {1, . . . , n}) ,
n
i=1
p
i
= 1.
Proof. Applying the Cauchy Mean-Value Theorem, there exists
ij
(, ) (i < j) such that
(6.136)
f (x
j
) f (x
i
)
g (x
j
) g (x
i
)
=
f
(
ij
)
g
(
ij
)
[, ]
for i, j {1, . . . , n} , i < j. Then
(6.137)
_

f (x
j
) f (x
i
)
g (x
j
) g (x
i
)
_ _
f (x
j
) f (x
i
)
g (x
j
) g (x
i
)

_
0, 1 i < j n,
which, by a similar argument to the one in Lemma 6.44 will give the desired result (6.135).
The following corollary is natural [14].
136 S.S. DRAGOMIR
Corollary 6.48. With the assumptions in Theorem 6.47 and if > 0, then one has the
inequalities:
(6.138) [T ( p, f ( x) , g ( x))]
2
4
( + )
2
T ( p, f ( x) , f ( x)) T ( p, g ( x) , g ( x)) ,
0 [T ( p, f ( x) , f ( x))]
1
2
[T ( p, g ( x) , g ( x))]
1
2
T ( p, f ( x) , g ( x)) (6.139)
_
2
2
T ( p, f ( x) , g ( x))
and
0 T ( p, f ( x) , f ( x)) T ( p, g ( x) , g ( x)) T
2
( p, f ( x) , g ( x)) (6.140)
( )
2
4
T
2
( p, f ( x) , g ( x)) .
6.20. Other Inequalities for the

Cebyev Functional. For two sequences of real numbers
a = (a
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) and p = (p
1
, . . . , p
n
) with p
i
0 (i {1, . . . , n}) and
n
i=1
p
i
= 1, consider the

Cebyev functional
(6.141) T
_
p, a,
b
_
=
n
i=1
p
i
a
i
b
i
i=1
p
i
a
i
n
i=1
p
i
b
i
.
By Sonins identity [1, p. 246] one has the representation
(6.142) T
_
p, a,
b
_
=
n
i=1
p
i
(a
i
A
n
( p, a))
_
b
i
A
n
_
p,
b
__
,
where
A
n
( p, a) :=
n
j=1
p
j
a
j
, A
n
_
p,
b
_
:=
n
j=1
p
j
b
j
.
Using the (CBS) inequality for weighted sums, we may state the following result
(6.143)
_
T
_
p, a,
b
_
2
T ( p, a, a) T
_
p,
b,
b
_
,
where, obviously
(6.144) T ( p, a, a) =
n
i=1
p
i
(a
i
A
n
( p, a))
2
.
Lemma 6.49. Assume that a = (a
1
, . . . , a
n
) ,
b = (b
1
, . . . , b
n
) are real numbers, p = (p
1
, . . . , p
n
)
are nonnegative numbers with
n
i=1
p
i
= 1 and b
i
= A
n
_
p,
b
_
for each i {1, . . . , n} . If
(6.145) < l
a
i
A
n
( p, a)
b
i
A
n
_
p,
b
_ L < for all i {1, . . . , n} ,
where l, L are given real numbers, then one has the inequality
(6.146) (l +L) T
_
p, a,
b
_
T ( p, a, a) +LlT
_
p,
b,
b
_
.
Proof. Using (6.145) we have
(6.147)
_
L
a
i
A
n
( p, a)
b
i
A
n
_
p,
b
_
__
a
i
A
n
( p, a)
b
i
A
n
_
p,
b
_ l
_
0
for each i {1, . . . , n} .
If we multiply (6.147) by
_
b
i
A
n
_
p,
b
__
2
0, we get
(6.148) (a
i
A
n
( p, a))
2
+Ll
_
b
i
A
n
_
p,
b
__
2
(L +l) (a
i
A
n
( p, a))
_
b
i
A
n
_
p,
b
__
for each i {1, . . . , n} .
Finally, if we multiply (6.148) by p
i
0, sum over i from 1 to n and use the identity (6.142)
and (6.144), we obtain (6.146).
Using Lemma 6.49 and a similar argument to that in the previous section, we may state the
following theorem [14].
Theorem 6.50. With the assumption of Lemma 6.49 and if L l > 0, then one has the
inequality
(6.149)
_
T
_
p, a,
b
_
2
4lL
(L +l)
2
T ( p, a, a) T
_
p,
b,
b
_
.
The following corollary is natural [14].
Corollary 6.51. With the assumptions in Theorem 6.50 one has
0 [T ( p, a, a)]
1
2
_
T
_
p,
b,
b
_1
2
T
_
p, a,
b
_
(6.150)
l
_
2
2
lL
T
_
p, a,
b
_
,
and
0 T ( p, a, a) T
_
p,
b,
b
_
_
T
_
p, a,
b
_
2
(6.151)
(L l)
2
4lL
_
T
_
p, a,
b
_
2
.
6.21. Bounds for the

Cebyev Functional. The following result holds.
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) (with b
i
= b
j
for i = j) be two
sequences of real numbers with the property that there exists the real constants m, M such that
for any 1 i < j n one has
(6.152) m
a
j
a
i
b
j
b
i
M.
Then we have the inequality
(6.153) mT
_
p,
b,
b
_
T
_
p, a,
b
_
MT
_
p,
b,
b
_
,
for any nonnegative sequence p = (p
1
, . . . , p
n
) with
n
i=1
p
i
= 1.
138 S.S. DRAGOMIR
Proof. From (6.152), by multiplying with (b
j
b
i
)
2
> 0, one has
m(b
j
b
i
)
2
(a
j
a
i
) (b
j
b
i
) M (b
j
b
i
)
2
for any 1 i < j n, giving by multiplication with p
i
p
j
0, that
m
1i<jn
p
i
p
j
(b
i
b
j
)
2
1i<jn
p
i
p
j
(a
j
a
i
) (b
j
b
i
)
M
1i<jn
p
i
p
j
(b
i
b
j
)
2
.
Using Korkines identity (see for example Subsection 6.19), we deduce the desired result (6.153).
The following corollary is natural.

Corollary 6.53. Assume that the sequence

b in Theorem 6.52 is strictly increasing and there
exists m, M such that
(6.154) m
a
k
b
k
M, k = 1, . . . , n 1;
where a
k
:= a
k+1
a
k
is the forward difference, then (6.153) holds true.
Proof. Follows from Theorem 6.52 on taking into account that for j > i and from (6.154) one
has
m
j1
k=i
b
k

j1
k=i
a
k
M
j1
k=i
b
k
,
giving m(b
j
b
i
) a
j
a
i
M (b
j
b
i
) .
Another possibility is to use functions that generate similar inequalities.
Theorem 6.54. Let f, g : [, ] R be continuous on [, ] and differentiable on (, ) with
g
(x) = 0 for x (, ) . Assume that

< m = inf
x(,)
f
(x)
g
(x)
, sup
x(,)
f
(x)
g
(x)
= M < .
If x = (x
1
, . . . , x
n
) is a real sequence with x
i
[, ] and x
i
= x
j
for i = j and if we denote
f ( x) := (f (x
1
) , . . . , f (x
n
)) , then we have the inequality
(6.155) mT ( p, g ( x) , g ( x)) T ( p, f ( x) , g ( x)) MT ( p, g ( x) , g ( x)) .
Proof. Applying the Cauchy Mean-Value Theorem, for any j > i there exists
ij
(, ) such
that
f (x
j
) f (x
i
)
g (x
j
) g (x
i
)
=
f
(
ij
)
g
(
ij
)
[m, M] .
Then, by Theorem 6.52 applied for a
i
= f (x
i
) , b
i
= g (x
i
) , we deduce the desired inequality
(6.155).
The following inequality related to the (CBS) inequality holds.
Theorem 6.55. Let a, x, y be sequences of real numbers such that x
i
= 0 and
y
i
x
i
=
y
j
x
j
for
i = j, (i, j {1, . . . , n}) . If there exist the real numbers , such that
(6.156)
a
j
a
i
y
j
x
j
y
i
x
i
for 1 i < j n,
_
_
n
i=1
x
2
i
n
i=1
y
2
i

_
n
i=1
x
i
y
i
_
2
_
_
i=1
x
2
i
n
i=1
a
i
x
i
y
i
i=1
a
i
x
2
i
n
i=1
x
i
y
i
(6.157)

_
_
n
i=1
x
2
i
n
i=1
y
2
i

_
n
i=1
x
i
y
i
_
2
_
_
.
i
=
x
2
i
n
k=1
x
2
k
, b
i
=
y
i
x
i
, m = and M = . We
omit the details.
The following different approach may be considered as well.
Theorem 6.56. Assume that a = (a
1
, . . . , a
n
) ,

b = (b
1
, . . . , b
n
) are sequences of real numbers,
p = (p
1
, . . . , p
n
) is a sequence of nonnegative real numbers with

n
i=1
p
i
= 1 and b
i
=
A
n
_
p,
b
_
:=
n
i=1
p
i
b
i
. If
(6.158) < l
a
i
A
n
( p, a)
b
i
A
n
_
p,
b
_ L < for each i {1, . . . , n} ,
where l, L are given real numbers, then one has the inequality
(6.159) lT
_
p,
b,
b
_
T
_
p, a,
b
_
LT
_
p,
b,
b
_
.
Proof. From (6.158), by multiplying with
_
b
i
A
n
_
p,
b
__
2
> 0, we deduce
l
_
b
i
A
n
_
p,
b
__
2
(a
i
A
n
( p, a))
_
b
i
A
n
_
p,
b
__
L
_
b
i
A
n
_
p,
b
__
2
,
for any i {1, . . . , n} .
By multiplying with p
i
0, and summing over i from 1 to n, we deduce
l
n
i=1
p
i
_
b
i
A
n
_
p,
b
__
2
i=1
p
i
(a
i
A
n
( p, a))
_
b
i
A
n
_
p,
b
__
L
n
i=1
p
i
_
b
i
A
n
_
p,
b
__
2
.
Using Sonins identity (see for example Section 6.20), we deduce the desired result (6.159).
The following result in connection with the (CBS) inequality may be stated.
Theorem 6.57. Let a, x,
b be sequences of real numbers such that x

i
= 0 and
y
i
x
i
=
1
n
i=1
x
2
i
A
n
_
x
2
,
y
x
_
for i {1, . . . , n} . If there exists the real numbers , such that
(6.160)
a
i
n
i=1
x
2
i
A
n
_
x
2
, a
_
y
i
x
i
n
i=1
x
2
i
A
n
_
x
2
,
y
x
_ ,
where x
2
= (x
2
1
, . . . , x
2
n
) and
y
x
=
_
y
1
x
1
, . . . ,
y
n
x
n
_
, then one has the inequality (6.157).
140 S.S. DRAGOMIR
i
=
x
2
i
n
i=1
x
2
i
, b
i
=
y
i
x
i
, l = and L = . We omit
the details.
REFERENCES
[1] D.S. MITRINOVI
C, J.E. PE
CARI

Kluwer Academic Publishers, 1993.
[2] A.M. OSTROWSKI, Vorlesungen ber Differential und Integralrechnung II, Birkhuser, Basel,
1951.
[3] S.S. DRAGOMIR, Ostrowskis inequality in complex inner product spaces, submitted.
[4] S.S. DRAGOMIR, Another Ostrowski type inequality, submitted.
[5] K. FAN AND J. TODD, A determinantal inequality, J. London Math. Soc., 30 (1955), 5864.
[6] S.S. DRAGOMIR AND V.M. BOLDEA, Some new remarks on Cauchy-Buniakowski-Schwartzs
inequality, Coll. Sci. Pap. Faculty of Science Kragujevac, 21 (1999), 3340.
[7] S.S. DRAGOMIR, On Cauchy-Buniakowski-Schwartzs Inequality for Real Numbers (Romanian),
Caiete Metodico- Stiintice, No. 57, 1989, pp. 24, Faculty of Mathematics, University of Tim-
i soara, Romania.
[8] S.S. DRAGOMIR, D.M. MILOEVI
C AND .Z. ARSLANAGI
C, A Cauchy-Buniakowski-
Schwartz inequality for Lipschitzian functions, Coll. Sci. Pap. of the Faculty of Science, Kragu-
jevac, 14 (1991), 2528.
[9] S.S. DRAGOMIR, Generalisation of the (CBS) inequality via Jensen type inequalities, submit-
ted.
[10] P. LAH AND M. RIBARI
C, Converse of Jensens inequality for convex functions, Univ. Beograd

Publ. Elek. Fak. Ser. Math, Fiz., No. 412-460, (1973), 201205.
[11] J.B. DIAZ AND F.T. METCALF, Stronger forms of a class of inequalities of G. Plya-G. Szeg
and L.V. Kantorovich, Bull. Amer. Math. Soc., 69 (1963), 415419.
[12] S.S. DRAGOMIR AND N.M. IONESCU, Some converse of Jensens inequaliy and applications,
Anal. Num. Theor. Approx., 23 (1994), 7178.
[13] J.E. PE
CARI
C AND S.S. DRAGOMIR, A renement of Jensens inequality and applications, Stu-

dia Univ. Babe s-Bolyai, Mathematica, 34(1) (1989), 1519.
[14] S.S. DRAGOMIR, Some inequalities for the

Cebyev functional, The Australian Math. Soc.
Gazette., 29(3) (2002), 164168.
[15] S.S. DRAGOMIR AND F.P. SCARMOZZINO, A renement of Jensens discrete inequality for dif-
ferentiable convex functions, Preprint, RGMIA Res. Rep. Coll., 5(4) (2002), Article 12. [ONLINE:
http://rgmia.vu.edu.au/v5n4.html]
[16] M.S. SLATER, A companion inequality to Jensens inequality, J. Approx. Theory, 32 (1981), 160
166.
[17] D. ANDRICA AND I. RA SA, The Jensen inequality: renements and applications, Anal. Num.
Theor. Approx., (Romania), 14 (1985), 155167.
[18] S.S. DRAGOMIR, On an inequality for convex functions, submitted.
INDEX
Jconvex function, 16, 17
Cebyev functional, 99, 134, 136
Cebyevs inequality, 10, 114, 115

1-norm, 87
additive, 10, 63, 64, 7375
Alzer, 45, 85
Andrica-Badea inequality, 92, 94, 95
arithmetic mean, 36, 63, 67, 116
Arithmetic-Geometric inequality, 11
asynchronous, 10, 114
sequence, 114
barycentric method, 72
Binet-Cauchys identity, 6
Callebauts inequality, 21, 43
Cassels inequality, 72, 75
additive version, 73
renement, 94, 96
unweighted, 7274, 96
Cauchy, 5
Cauchy-Bunyakovsky-Schwarz, 5
Cauchy-Schwarz, 45
Cauchy Mean-Value Theorem, 135
Cauchy-Bunyakovsky-Schwarz, 5
Cauchy-Schwarz, 45
CBS-inequality, 5, 1219, 26, 35, 44, 117, 122
complex numbers, 6
complex sequences, 110
double sums, 117, 118
for Lipschitzian functions, 119
for weights, 33, 51
functionals associated to, 68
generalisation, 7, 12, 1417, 22
generalisation via Youngs inequality, 13
inequality related to, 64
real numbers, 5, 7, 24, 41, 52
real sequence, 109
renement, 11, 32, 37, 40, 41, 57, 58, 61, 6365,
126, 128
discrete, 42
reverse, 75, 82, 84, 85, 87, 89, 91, 93, 95, 99, 101,
104
type, 20
weights, 10
concave function, 121123
convex mapping, 116
Daykin-Eliezer-Carlitz, 42, 43
De Bruijns inequality, 40
Diaz-Metcalf inequality, 97, 99, 123
discriminant, 5, 24, 26
double sums, 117, 118, 134
Dragomir-Ionescu inequality, 123
Dunkl-Williams inequality, 43
Eulers indicator, 18, 119
Fan and Todd inequalities, 113
forward difference, 85, 88
geometric mean, 36, 63, 67, 116
Grss inequality, 94
Greub-Rheinboldt inequality, 75
harmonic mean, 36, 37, 116
identity, 6, 41, 42, 44, 59, 61, 86, 100, 137
Binet-Cauchy, 6
Cauchy-Binet, 117
Korkines, 134
Lagrange, 5, 7, 44, 59, 61, 118
Sonin, 136
index set mapping, 54, 57, 61, 62, 67
inner product spaces, 5, 43, 60
Jensens inequality, 116, 125
discrete, 121
reverse, 122
renement, 124, 127
reverse, 123
Korkines identity, 134
Lagranges identity, 5, 7, 44, 59, 61, 118
Lah-Ribari c inequality, 122
Lipschitzian functions, 119
Lupa s inequality, 94
matrices, 117
maximum, 72, 92, 93
McLaughlins inequality, 41
Milnes inequality, 43
minimum, 72
monotonicity, 49, 51, 53, 54, 56, 58, 85
multiple sums, 125
n-tuple, 122
non-proportional, 109, 110, 112
Ostrowskis inequality, 109111
p-norm, 90
Plya-Szeg, 74
Plya-Szegs inequality, 7375, 97
parameter, 10
Popovicius inequality, 18
positive real numbers, 9, 11, 12, 31, 36, 37, 72, 73,
75, 78, 96, 115, 116, 122, 124, 126
Power functions, 116
power series, 19
proportional, 5, 7, 12
quadratic polynomial, 5, 23
141
142 S.S. DRAGOMIR
renement, 35
Cassels inequality, 94, 96
CBS inequality, 11, 32, 37, 40, 41, 44, 57, 58, 61,
6365, 126, 128
due to Alzer and Zheng, 45
due to Daykin-Eliezer-Carlitz, 42
in terms of moduli, 28, 31
Jensens inequality, 124, 127
non-constant sequences, 37
Plya-Szeg inequality, 97
Schwitzers result, 94
sequence less than the weights, 33
sequence whose norm is 1, 29
via Dunkl-Williams Inequality, 43
reverse inequality, 75, 82, 84, 85, 87, 8991, 93, 97,
99, 101, 104
complex numbers, 79
Jensen, 122, 123
real numbers, 77
via

Cebyev Functional, 99
via Andrica-Badea, 92
via Diaz-Metcalf, 97
rotation, 41
Schweitzer inequality, 94
sequence, 36, 42, 50, 55, 77, 85, 87, 100, 109
asynchronous, 114
complex, 110
complex numbers, 6, 12, 24, 44, 75, 85, 87, 90,
97, 110, 112
decreasing nonnegative, 85
natural numbers, 119
nonnegative, 9, 10, 63, 85, 114, 134
nonnegative numbers, 11
nonnegative real numbers, 12, 21, 22, 33, 51, 52,
72, 75, 85, 87, 90, 92, 96, 121124
nonzero complex numbers, 44
nonzero real numbers, 17
positive, 82
positive and real numbers, 8
positive numbers, 7, 42
positive real numbers, 9, 11, 12, 31, 37, 72, 73, 75,
78, 96, 115, 116, 122, 124, 126
real, 109, 114, 135
real numbers, 7, 8, 10, 1820, 22, 28, 29, 3137,
40, 41, 51, 52, 59, 84, 87, 89, 9194, 99, 102,
105, 109, 111, 113, 114, 117123, 134, 136
synchronous, 115
set-additive, 67, 68
set-superadditive, 6770
set-superadditivity, 68
set-supermultiplicative, 67
Shisha-Mond inequalities, 82
Sonins identity, 136
strong monotonicity, 56, 58
strong superadditivity, 5557
sup-norm, 85
superadditivity, 52, 5457, 59, 61, 62
supermultiplicative, 63
supermultiplicity, 63, 67
synchronous, 115
sequence, 115
triangle inequality, 7, 86
Wagners inequality, 22, 24
weights, 10, 33, 51, 52, 55, 56, 63
Youngs Inequality, 1113
Zagier inequalities, 83, 84
Zheng, 45

A Survey On Cauchy-Bunyakovsky-Schwarz Type Discrete Inequalities

Uploaded by

Copyright:

Available Formats

You might also like

A Survey On Cauchy-Bunyakovsky-Schwarz Type Discrete Inequalities

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Survey On Cauchy-Bunyakovsky-Schwarz Type Discrete Inequalities

Uploaded by

Copyright:

Available Formats

Journal of Inequalities in Pure and

d are proportional with the same constant k.

for any i, j {1, . . . , n} . Multiplying (2.17) by p

k = (k, k, . . . , k) is a constant sequence.

= 1. Then one has the

Remark 2.25. For = = 2, we recapture the (CBS) inequality.

(0, r) one has

C AND A.M. FINK, Classical and New Inequalities in Analysis,

the above inequality (3.11) provides the following result

b one has the inequality

from where we get

which is clearly equivalent to (3.19).

ab (the geometric mean)

and (3.71) is proved.

C AND D.M. MILOEVI

C AND A.M. FINK, Classical and New Inequalities in Analysis,

proving the desired inequality (4.1).

Consider the following mapping:

Now, if we consider the mapping

C, The improvement of Cauchy-Buniakowski-Schwartzs

C AND A.M. FINK, Classical and New Inequalities in Analysis,

g (t) G(t) dt, the right hand side gives

The following corollary is a natural consequence of Theorem 5.34 for p = q = 2.

then one has the inequality

then one has the inequality

C, The inequalities of Zagier and

b, x be sequences of real numbers so that a = 0 and

b, x be sequences of complex numbers so that a = 0, a,

d sequences of complex numbers, with equality iff there is a C such that

we may state the following result [6, Theorem 3].

are synchronous, i.e., (|a

where i I and I is a nite part of N. Then one has the inequality

I for each i {1, . . . , n} , and w = (w

(y) = |f (x) f (y) (x y) f

for any i {1, . . . , n} .

(t) > and M = sup

f (t) . We omit the details.

(t) and M = sup

(x) = 0 for x (, ) . Assume

The following corollary is natural.

(x) = 0 for x (, ) . Assume that

b be sequences of real numbers such that x

C AND A.M. FINK, Classical and New Inequalities in Analysis,

C AND .Z. ARSLANAGI

C, Converse of Jensens inequality for convex functions, Univ. Beograd

C AND S.S. DRAGOMIR, A renement of Jensens inequality and applications, Stu-

Cebyev functional, 99, 134, 136

Cebyevs inequality, 10, 114, 115

You might also like