Você está na página 1de 134

Monograph:

Numerous approximations of
Riemann-Stieltjes double integrals

Mohammad W. Alomari

2009

CONTENTS

CHAPTER I

PREFACE

CHAPTER II
LITERATURE REVIEW AND BACKGROUND
2.1 Introduction
2.2 Functions of bounded variation and bivariation
2.3 Inequalities for mappings of one variable
2.3.1 Ostrowski type Inequalities
2.3.2 Simpsons Type Inequalities
2.4 Inequalities for mappings of two variables
2.4.1 Ostrowski and Gruss type inequalities
2.4.2 Simpsons Type Inequalities

3
3
3
10
10
18
19
19
26

CHAPTER III
OSTROWSKIS TYPE INEQUALITIES
3.1 Introduction
3.2 Preliminaries and Lemmas
3.3 Ostrowski inequality for mappings bounded bivariation
3.4 Quadrature Rules For RSdouble integral
3.5 Approximations via BeesackDarstPollard inequality
3.6 Applications to Quadrature Formula

29
29
29
34
48
56
59

CHAPTER IV
ON AN OSTROWSKI TYPE FUNCTIONAL
4.1 Introduction
4.2 A functional Related to the Ostrowski inequality
4.3 integral representation of error
4.4 Error bounds
4.5 A numerical quadrature formula for the RSintegral

62
62
62
68
73
95

CHAPTER V
SOME RELATED INEQUALITIES
5.1 Introduction
5.2 Gruss type inequalities
5.3 Approximating real functions of two variables which possess n-th derivatives
of bounded variation
5.4 Trapezoid-type rules for RS-Double integrals

97
97
97
108
125

BIBLIOGRAPHY

129

CHAPTER I

PREFACE

The approximation problem of the RiemannStieltjes integral in terms of RiemannStieltjes


sums is very interesting problem. The most significant way to study this problem may be
done through the wellknown Ostrowski inequality. In recent years, several authors have
studied the wellknown Ostrowski inequality in one variable for various types of mappings
such as: absolutely continuous, Lipschitzian and n-differentiable mappings as well as mappings
of bounded variation. However, a small attention and a few works have been considered
for mappings of two variables. Among others, Dragomir and his group have studied a very
interesting inequalities for mapping of two independent variables.
The concept of Riemann-Stieltjes integral

Rb
a

f (t) du (t); where f is called the integrand,

u is called the integrator, plays an important role in Mathematics. The approximation problem
Rb
of the RiemannStieltjes integral a f (t) du (t) in terms of the RiemannStieltjes sums have

been considered recently by many authors. However, a small attention and a few works

have been considered for mappings of two variables; i.e., The approximation problem of the
RbRd
RiemannStieltjes double integral a c f (t, s) ds dt u (t, s) in terms of the RiemannStieltjes
RbRd
double sums. This study is devoted to obtain several bounds for a c f (t, s) ds dt u (t, s)
under various assumptions on the integrand f and the integrator u. Mainly, the concepts
of bounded variation and bi-variation are used at large in the context. Several proposed
cubature formula are introduced to approximate such double integrals. For mappings of two
variables several inequalities of Trapezoid, Gruss and Ostrowski type for mappings of bounded
variation, bounded bi-variation, Lipschitzian and monotonic are introduced and discussed.
Namely, Trapezoid-type rules for RS-Double integrals are proved, and therefore the classical
HermiteHadamard inequality for mappings of two variables is established. A Korkine type

identity is used to obtain several Gruss type inequalities for integrable functions. Finally,
approximating real functions of two variables which possess n-th partial derivatives of bounded
bi-variation, Lipschitzian and absolutely continuous are established and investigated.
The main concern in this monograph is to study the approximation problem of the
RiemannStieltjes double integral in terms of the RiemannStieltjes double sums. In fact,
the Ostrowski inequality for mappings of two independent variables which are of bounded
bi-variation, Holder continuous and absolutely continuous are used to discuss this problem.
In this way, an interesting study of the approximation problem of the RiemannStieltjes double
integral is presented and therefore several proposed cubature rules for mappings of two variables
are given.
The organization of this monograph is given as follows. The first chapter gives a general
introduction of the research work where the motivation and objectives are defined. In chapter
II, some basic concepts of bounded variation, Riemann-Stieltjes integral including some of
its properties are given. Some known inequalities of Ostrowskis type with some related
refinements and generalizations in one and two variables are given. In chapter III, several
new inequalities of Ostrowskis type are introduced. Trapezoid and Midpoint type rules for
double RiemannStieltjes double integral are proved. A generalization of the well known
BeesackDarstPollard inequality for double RSdouble integrals is also considered. Finally,
as applications, two cubature formulae are proposed. In chapter IV, in order to approximate the
RSdouble integrals some functionals are introduced and therefore several representations of
the errors are established. Finally, as application, a cubature formulae is proposed. In chapter
V, some related inequalities are proved. Namely, Gruss type inequalities are proved as well as
an approximation of a real function of two variables which possess n-th partial derivatives of
bounded bivariation are established. Finally Trapezoid-type rules for RSdouble integrals are
provided are given.
M. W. Alomari
July 4, 2009

CHAPTER II

LITERATURE REVIEW AND BACKGROUND

2.1

INTRODUCTION

This chapter shall be considered as a review of some famous, fundamental and basic concepts
of mappings of bounded variation and bounded bi-variation in two variables with some of their
properties. As well as, several inequalities of Ostrowskis and Simpsons type in one and two
variables are reviewed.
2.2

FUNCTIONS OF BOUNDED VARIATION AND BIVARIATION

Let D R2 and let f : D R be any function. We say that f is bounded above on D if


there is R such that f (x, y) for all (x, y) D; in this case, we say that f attains
its upper bound on D if there is (x0 , y0 ) D such that sup f (x, y) : (x, y) D = f (x0 , y0 ).
Likewise, we say that f is bounded below on D if there is R such that f (x, y) for all
(x, y) D; in this case, we say that f attains its lower bound on D if there is (x0 , y0 ) D such
that inf f (x, y) : (x, y) D = f (x0 , y0 ). Finally, we say that f is bounded on D if it is bounded
above on D as well as bounded below on D; in this case, we say that f attains its bounds on D
if it attains its upper bound on D and also attains its lower bound on D.
In general, we remark that for the order pairs (x1 , y1 ) (x2 , y2 ) D, we write (x1 , y1 )
(x2 , y2 ) if x1 x2 and y1 y2 . Now, let I and J be intervals in R such that I J D. We
say that

1. f is monotonically increasing on I J if for all (x1 , y1 ), (x2 , y2 ) in I J, we have


(x1 , y1 ) (x2 , y2 ) = f (x1 , y1 ) f (x2 , y2 )

2. f is monotonically decreasing on I J if for all (x1 , y1 ), (x2 , y2 ) in I J, we have


(x1 , y1 ) (x2 , y2 ) = f (x1 , y1 ) f (x2 , y2 )
3. f is bimonotonically increasing on I J if for all (x1 , y1 ), (x2 , y2 ) in I J, we have
(x1 , y1 ) (x2 , y2 ) = f (x1 , y2 ) + f (x2 , y1 ) f (x1 , y1 ) + f (x2 , y2 )
4. f is bimonotonically decreasing on I J if for all (x1 , y1 ), (x2 , y2 ) in I J, we have
(x1 , y1 ) (x2 , y2 ) = f (x1 , y2 ) + f (x2 , y1 ) f (x1 , y1 ) + f (x2 , y2 )

It may be noted that f is monotonically increasing on I J if and only if it is (monotonically)


increasing in each of the two variables. The following result gives conditions under which an
increasing function in the variable x and an increasing function in the variable y can be added
or multiplied to obtain a monotonic and/or bimonotonic function of two variables.
Proposition 2.2.1. (Ghorpade & Limaye 2009) Let I, J be nonempty intervals in R. Given
any : I R and : J R, consider f : I J R and g : I J R defined
by f (x, y) = (x) + (y) and g (x, y) = (x) (y) for (x, y) I J. Then we have the
following

1. f is monotonically increasing on I J if and only if is increasing on I and is


increasing on J.
2. Assume that (x) 0 and (y) 0 for all x I, y J, and also that (x0 ) > 0 and
(y0 ) > 0 for some x0 I and some y0 J. Then g is monotonically increasing on
I J if and only if is increasing on I and is increasing on J.
3. f is always bimonotonically increasing and also bimonotonically decreasing on I J.
4. If is monotonic on I and is monotonic on J, then g is bimonotonic on I J. More
specifically, if and are both increasing or both decreasing, then g is bimonotonically
increasing, whereas if is increasing and is decreasing, or vice-versa, then g is
bimonotonically decreasing.

We recall that for I interval in R. A function f : I R is said to be convex if for all


x, y I and for all [0, 1], the inequality
f (x + (1 ) y) f (x) + (1 ) f (y)

(2.2.1)

holds. If (2.2.1) is strictly for all x 6= y and (0, 1), then f is said to be strictly convex. If the
inequality in (2.2.1) is reversed, then f is said to be concave. If (2.2.1) is strictly for all x 6= y
and (0, 1), then f is said to be strictly concave (see (Pecaric et al. 1992)).
The above proposition as well as the one below can be used to generate several examples
of monotonic and bimonotonic functions.
Proposition 2.2.2. (Ghorpade & Limaye 2009) Let I, J be nonempty intervals in R. The set
I + J := {x + y|x I, y J} R. Further, consider : I + J R be any function and
consider f : I J R defined by f (x, y) = (x + y), for (x, y) I J. Then we have the
following

1. is increasing on I + J = f is monotonically increasing on I J.


2. is decreasing on I + J = f is monotonically decreasing on I J.
3. is convex on I + J = f is bimonotonically increasing on I J.
4. is concave on I + J = f is bimonotonically decreasing on I J.
Example 2.2.3. (Ghorpade & Limaye 2009)

1. Consider f : [1, 1]2 R defined by


(
(x + 1) (y + 1) ,
f (x, y) =
(x + 2) (y + 2) ,

x+y <0
.
x+y 0

If we fix y0 [1, 1] and consider the function : [1, 1] R defined by


(
(y0 + 1) (x + 1) ,
x+y <0
(x) =
(y0 + 2) (x + 2) ,
x+y 0

then it is easy to see that is increasing on [1, 1]. Similarly, if we fix x0 [1, 1] and
consider the function : [1, 1] R defined by
(
(x0 + 1) (y + 1) ,
(y) =
(x0 + 2) (y + 2) ,

x+y <0
x+y 0

then it is easy to see that is increasing on [1, 1]. It follows that f is monotonically
increasing on [1, 1]2 . However, f is not bimonotonic on [1, 1]2 . To see this note that
(0, 0) (1, 1) and f (0, 1) + f (1, 0) = 6 + 6 < 4 + 9 = f (0, 0) + f (1, 1), whereas
(1, 0) (0, 1) and f (1, 1) + f (0, 0) = 3 + 4 > 0 + 6 = f (1, 0) + f (0, 1).
2. Consider f : R2 R defined by f (x, y) := cos x + sin y. Using Proposition 2.2.1, we
readily see that f is bimonotonic, but not monotonic.

A functions of bounded variation is an interesting class of functions that is very closely


related to monotonic functions. Let us recall some facts about functions of bounded variation if
[a, b] is a compact interval , a set of points P := {x0 , x1 , , xn }, satisfying the inequalities
a = x0 < x1 < < xn1 < xn = b,
is called a partition of [a, b]. The interval [xk1 , xk ] is called kth subinterval of P and we write
P
xk = xk xk1 , so that nk=1 xk = b a. The collection of all possible partitions of [a, b]
will be denoted by P[a, b].

Definition 2.2.4. (Apostol 1974) Let f be defined on [a, b]. If P := {x0 , x1 , , xn } is a


partition of [a, b], write fk = f (xk ) f (xk1 ), for k = 1, 2, , n. If there exists a positive
n
P
number M such that
|fk | M for all partition of [a, b], then f is said to be of bounded
k=1
P
variation on [a, b]. Moreover, if f is of bounded variation on [a, b], and (P ) denote the sum
n
P
|fk | corresponding to the partition P = {x0 , x1 , , xn } of [a, b]. The number
k=1

b
_
a

(f ) = sup

nX

o
(P ) : P P[a, b] ,

is called the total variation of f on the interval [a, b].

In two variables or more, the concept of bounded variation is quite different. According
to Clarkson and Adams (1933), several definitions have been given of conditions under which

a function of two or more independent variables shall be said to be of bounded variation. Of


these definitions six are usually associated with the names of Vitali, Hardy, Arzel`a, Pierpont,
Frechet, and Tonelli respectively. A seventh has been formulated by Hahn and attributed by him
to Pierpont; which are equivalent, and the proof of this fact was presented in the same paper. In
general, some relations between these classes are discussed and investigated in the interesting
paper Clarkson & Adams (1933).
In this work, we are interested in two of the above senses, which are; bounded variation in
Arzel`a and Vitali senses.
The monotonic mappings plays a main role in studying mappings of bounded variation,
in two variables, the sum of two monotonic functions need not be monotonic. For example,
f : [0, 1]2 R defined by f (x, y) := x y is a sum of monotonic functions (given by
(x, y) 7 x and (x, y) 7 y, but it is neither increasing nor decreasing. On the other hand,
since a monotonic function on a (closed) rectangle is bounded (f : [a, b][c, d] R monotonic
= the values of f lie between f (a, c) and f (b, d)), sums of monotonic functions are bounded.
In fact, they satisfy a stronger property defined below.
For a, b, c, d R, we consider the subset Q := Qb,d
a,c = {(x, y) : a x b, c y d}
of R2 .
Definition 2.2.5. (Clarkson & Adams 1933) If
P := {(xi , yi ) : xi1 x xi ; yi1 y yi ; i = 1, . . . , n}
is a partition of Q, write
f (xi , yi ) = f (xi , yi ) f (xi1 , yi1 )
for i = 1, 2, , n. The function f (x, y) is said to be of bounded variation in the Arzel`a sense
(or simply bounded variation) if there exists a positive quantity M such that for every partition
n
P
on Q we have
|f (xi , yi )| M .
i=1

Therefore, one can define the concept of total variation of a function of two variables, as
follows:

Let f be of bounded variation on Q, and let


corresponding to the partition P of Q. The number
_
Q

(f ) :=

d _
b
_
c

(f ) := sup

nX

(P ) denote the sum

n
P

|f (xi , yi )|

i=1

o
P : P P (Q) ,

is called the total variation of f on Q.


In the following, we point out some elementary properties of functions of bounded
variation.
Proposition 2.2.6. (Ghorpade & Limaye 2009) Let f, g : Q R and r R. Then

1. f is bounded variation = f is bounded.


2. f is monotonic = f is of bounded variation.
3. f , g are of bounded variation = f + g, rf , f g are of bounded variation.
Proposition 2.2.7. (Ghorpade & Limaye 2009) Let I, J be nonempty intervals in R. Given
any : I R and : J R, consider f : I J R and g : I J R defined
by f (x, y) = (x) + (y) and g (x, y) = (x) (y) for (x, y) I J. Then we have the
following:

1. f is of bounded variation on [a, b] [c, d] if and only if is of bounded variation on [a, b]


and is of bounded variation on [c, d].
2. Assume that and are not identically zero, then g is of bounded variation on [a, b][c, d]
if and only if is of bounded variation on [a, b] and is of bounded variation on [c, d].
Definition 2.2.8. (Clarkson & Adams 1933) If
P := {(xi , yj ) : xi1 x xi ; yj1 y yj ; i = 1, . . . , n ; j = 1, . . . , m}
is a partition of Q, write
11 f (xi , yj ) = f (xi1 , yj1 ) f (xi1 , yj ) f (xi , yj1 ) + f (xi , yj )

for i = 1, 2, , n and j = 1, 2, , m. The function f (x, y) is said to be of bounded variation


in the Vitali sense (or simply bounded bivariation) if there exists a positive quantity M such
n P
m
P
that for every partition on Q we have
|11 f (xi , yj )| M .
i=1 j=1

Therefore, one can define the concept of total bivariation of a function of two variables,
as follows:
Let f be of bounded bivariation on Q, and let
corresponding to the partition P of Q. The number
_
Q

(f ) :=

d _
b
_
c

(f ) := sup

(P ) denote the sum

nX

n P
m
P

|11 f (xi , yj )|

i=1 j=1

o
P : P P (Q) ,

is called the total bivariation of f on Q.


In the following, we point out some elementary properties of functions of bounded
bivariation.
Proposition 2.2.9. (Ghorpade & Limaye 2009) Let f, g : Q R and r R. Then
1. If f is of bounded bivariation and, in addition, f is bounded on any two adjacent sides of
the rectangle [a, b] [c, d], then f is bounded.
2. If f is bimonotonic, then f is of bounded bivariation
3. If f and g are of bounded bivariation, then so are f + g and rf .
Proposition 2.2.10. (Ghorpade & Limaye 2009) Let I, J be nonempty intervals in R. Given
any : I R and : J R, consider f : I J R and g : I J R defined
by f (x, y) = (x) + (y) and g (x, y) = (x) (y) for (x, y) I J. Then we have the
following:

1. f is always of bounded bivariation on [a, b] [c, d].


2. Assume that and are not constant functions, then g is of bounded bivariation on
[a, b][c, d] if and only if is of bounded variation on [a, b] and is of bounded variation
on [c, d].

10

For further properties of mappings of bounded variation and bivariation we refer the
reader to the comprehensive book Ghorpade and Limaye (2009) .
2.3

INEQUALITIES FOR MAPPINGS OF ONE VARIABLE

2.3.1

Ostrowski type Inequalities

For a differentiable mapping f defined [a, b] and f 0 be integrable on [a, b], then the Montgomery
identity
1
f (x) =
ba

f (t) dt +

holds, where P (x, t) is the Peano kernel,


(
P (x, t) =

P (x, t) f 0 (t) dt

(2.3.1)

ta
,
ba
tb
,
ba

a t x,
x < t b.

In 1938, Ostrowski established a very interesting inequality for differentiable mappings


with bounded derivatives, as follows:
Theorem 2.3.1. (Ostrowski 1938) Let f : I R R be a differentiable mapping on I , the
interior of the interval I, such that f 0 L[a, b], where a, b I with a < b. If |f 0 (x)| M ,
then the following inequality,


f (x)

1
ba

"
 #

a+b 2

x

1
2
f (u) du M (b a)
+
4
(b a)2

(2.3.2)

holds for all x [a, b]. The constant 14 is the best possible in the sense that it cannot be replaced
by a smaller constant.

In 1992, Fink and earlier in 1976, Milovanovic and Pecaric have obtained some
interesting generalizations of (2.3.2) in the form


!
Z b
n1
1

X


1


f (x) +
Fk (x)
f (t) dt C (n, p, x) f (n)

n

ba a
k=1

where,

n k f (k1) (a) (x a)k f (k1) (b) (x b)k


,
Fk (x) =
n!
ba

(2.3.3)

11

and, kkr , 1 r are the usual Lebesgue norms on Lr [a, b], i.e.,
kf k := ess sup |f (t)| ,
t[a,b]

and
kf kr :=

Z

1/r
|f (t)| dt
, 1 r < .
r

In fact, Milovanovic and Pecaric (see also Mitrinovic et al. (1994)) have proved that
C (n, , x) =

(x a)n+1 + (b x)n+1
,
(b a) n (n + 1)!

while Fink proved that the inequality (2.3.3) holds provided f (n1) is absolutely continuous on
[a, b] and f (n) Lp [a, b], with

1/q
(x a)nq+1 + (b x)nq+1
C (n, p, x) =
1/q ((n 1) q + 1, q + 1) ,
(b a) n!
for 1 < p , is the beta function, and
(n 1)n1
C (n, 1, x) =
max {(x a)n , (b x)n } .
n
(b a) n n!
In 2001, Dragomir proved the following Ostrowskis inequality for mappings of bounded
variation:
Theorem 2.3.2. (Dragomir 2001b) Let f : [a, b] R be a mapping of bounded variation on
[a, b]. Then we have the inequalities:

# b
"

Z b

a+b
_


x

2
1 +
f (x) 1
f
(t)
dt

(f ) ,
(2.3.4)



ba a
2 ba
a
W
for any x [a, b] , where ba (f ) denotes the total variation of f on [a, b]. The constant 21 is best

possible.

The following trapezoid type inequality for mappings of bounded variation holds:
Theorem 2.3.3. (Cerone & Dragomir 2000) Let f : [a, b] R, be a mapping of bounded
variation on [a, b], Then

Z b
b
_
1

f
(a)
+
f
(b)


f (t) dt (b a)
2 (b a) (f ) .

2
a
a
The constant

1
2

is the best possible.

(2.3.5)

12

A generalization of (2.3.5) and 2.3.4 for mappings of bounded variation, was considered
by Cerone et al. (2000), as follows:

 _


Z b
b




a
+
b
b

a
x
(f )
(b x) f (b) + (x a) f (a)

+
f
(t)
dt



2
2
a
a

for all x [a, b]. The constant

1
2

(2.3.6)

is the best possible.

In the same way, the following midpoint type inequality for mappings of bounded
variation was proved in Cerone and Dragomir (2000):
Theorem 2.3.4. (Cerone & Dragomir 2000) Let f : [a, b] R, be a mapping of bounded
variation on [a, b], Then

The constant

1
2



 Z b

b
_
1


(b a) f a + b
f (t) dt (b a) (f ) .

2
2
a
a

(2.3.7)

is the best possible.

In the recent paper Tseng et al. (2008), the authors have proved the following weighted
Ostrowski inequality for mappings of bounded variation, as follows:
Theorem 2.3.5. (Tseng et al. 2008) Let 0 1, g : [a, b] [0, ) continuous and
positive on (a, b) and let h : [a, b] R be differentiable such that h0 (t) = g(t) on [a, b]. Let




c = h1 1 2 h (a) + 2 h (b) and d = h1 2 h (a) + 1 2 h (b) . Suppose that f is of

bounded variation on [a, b], then for all x [c, d], we have

Z b


Z b
b
_


f (a) + f (b)


f (t) g (t) dt (1 ) f (x) +
g (t) dt K
(f ) (2.3.8)

2
a

where,

K=

1
2

max

Rb
2

and

Wb

Rb



g (t) dt + h (x) +

1
2

Rb
a

g (t) dt,

h(a)+h(b)
,
2



g (t) dt + h (x) +

h(a)+h(b)
, 2
2

Rb
a

o
g (t) dt ,

(f ) is the total variation of f over [a, b]. The constant

constant

for

2
3

1 are the best possible.

1
2

0
1
2

<<
2
3

1
2

2
3

for 0

1
2

and the

13

Another new generalization of weighted Ostrowski type inequality for mappings of


bounded variation has been obtained by Liu (2012), as follows:
Theorem 2.3.6. (Liu 2012) Let f : [a, b] R be a mapping of bounded variation, g : [a, b]
[0, ) continuous and positive on (a, b). Then for any x [a, b] and [0, 1], we have
Z b



Z b
Z x
Z b




f
(t)
g
(t)
dt

(1

)
f
(x)
g
(t)
dt
+

f
(a)
g
(t)
dt
+
f
(b)
g
(t)
dt


a


  Z b
Z x
 _
Z
b
1


1 b
1 1



+
g (t) dt +
g (t) dt
g (t) dt (f ) (2.3.9)

2 2
2 a
2 a
a
a


where,

Wb

(f ) denotes to the total variation of f over [a, b]. The constant

best possible.

1
2



+ 12 is the

In 2002, Guessab and Schmeisser, incorporate the mid-point and the trapezoid inequality
together, and they have proved the following companion of Ostrowskis inequality:
Theorem 2.3.7. (Guessab & Schmeisser 2002) Assume that the function f : [a, b] R is of
r-HHolder type, where r (0, 1] and H > 0 are given, i.e.,
|f (t) f (s)| H |t s|r ,
], one has the inequality
for any t, s [a, b]. Then, for each x [a, a+b
2


Z b
f (x) + f (a + b x)

1

f
(t)
dt


2
ba a
"

#
2r+1 (x a)r+1 + (a + b 2x)r+1
H
. (2.3.10)
2r (r + 1) (b a)

This inequality is sharp for each admissible x. Equality is obtained if and only if f = Hf +c,
with c R and

(x t)r ,

f (t) =
(t x)r ,

f (a + b x) ,

atx
xt
a+b
2

a+b
2

tb

Dragomir (2002), has proved the following companion of the Ostrowski inequality for
mappings of bounded variation:

14

Theorem 2.3.8. (Dragomir 2002) Let f : [a, b] R be a mapping of bounded variation on


[a, b]. Then we have the inequalities:

# b
"

Z b

3a+b
_

f (x) + f (a + b x)
x

1
4
1 +

f
(t)
dt

(f ) ,



2
ba a
4 ba
a

for any x [a, a+b


] , where
2
best possible.

Wb

a (f )

(2.3.11)

denotes the total variation of f on [a, b]. The constant 1/4 is

Dragomir (2000), has introduced an Ostrowski type integral inequality for the
Riemann-Stieltjes integral, as follows:
Theorem 2.3.9. (Dragomir 2000) Let f : [a, b] R be a function of bounded variation and
u : [a, b] R a function of r-H-Holder type, i.e.,
|u (x) u (y)| H |x y|r ,

x, y [a, b] ,

where r (0, 1] and H > 0 are given. Then, for any x [a, b],


Z b


[u (b) u (a)] f (x)
f (t) du (t)

a
"
#
x
b
_
_
H (x a)r
(f ) + (b x)r
(f )
a

(2.3.12)

x

 b
b
W

W
W

r
r
1
1

(f
)
+
(f
)

(f
)
[(x

a)
+
(b

x)
]

2
2

 x
ap  b x p 1/p

W
W
qr
qr 1/q
H
(f )
, p > 1, p1 +
[(x

a)
+
(b

x)
]
(f
)
+

a
x

b

r W
1

a+b

(f )
2 (b a) + x 2

1
q

=1

where,

d
W

(f ) denotes the total variation of f on the interval [c, d].

For other results concerning inequalities for Stieltjes integrals, see Liu (2004) and
Cerone and Dragomir (2002). In 2007, Cerone et al. established some Ostrowski type
inequalities for the Stieltjes integral where the integrand is absolutely continuous while the
integrator is of bounded variation. Also, the case when |f 0 | is convex was explored.
Recently, Dragomir (2008) provided an approximation for the function f which
possesses continuous derivatives up to the order n 1 (n 1) and has the n-th derivative

15

of bounded variation, in terms of the chord that connects its end points A = (a, f (a)) and
B = (b, f (b)) and some more terms which depend on the values of the k derivatives of the
function taken at the end points a and b, where k is between 1 and n.
As pointed out by Dragomir, if f : [a, b] R is assumed to be bounded on [a, b]. The chord
that connects its end points A = (a, f (a)) and B = (b, f (b)) has the equation
df : [a, b] R,

df (x) =

1
[f (a) (b x) + f (b) (x a)] .
ba

Before that in (2007) Dragomir was introduced the error in approximating the value of the
function f (x) by df (x) with x [a, b] by f (x), i.e., f (x) is defined by:
f (x) :=

bx
xa
f (a) +
f (b) f (x) .
ba
ba

The following bounds for f (x) holds :


Theorem 2.3.10. (Dragomir 2007) If f : [a, b] R is of bounded variation, then
 _

 _
x
b
xa
bx

(f ) +
(f )
|f (x)|
ba
ba
a
x
a+b 

x
Wb

1
2

a (f )
2
ba

W
q i 1


  1 h Wx
q
q
b
bx p
xa p p
+ ba
( a (f )) +
;

x (f )
ba

p > 1, p1 + 1q = 1

1 Wb (f ) + 1 Wx (f ) Wb (f )
a
a
x
2
2


The first inequality in (2.3.13) is sharp. The constant

1
2

(2.3.13)

is best possible in the first and third

branches.

Therefore, a generalization Theorem 2.3.10, was considered as follows:


Theorem 2.3.11. (Dragomir 2008) Let I be a closed subinterval on R, let a, b I with a < b
and let n be a nonnegative integer. If f : I R is such that the n-th derivative f (n) is of

16

bounded variation on the interval [a, b], then for any x [a, b] we have the representation
1
[(b x) f (a) + (x a) f (b)]
ba
n
i
(b x) (x a) X 1 h
+

(x a)k1 f (k) (a) + (1)k (b x)k1 f (k) (b)


ba
k!
k=1
Z b

1
+
Sn (x, t) d f (n) (t) , (2.3.14)
ba a

f (x) =

where the kernel Sn : [a, b]2 R is given by

atx
(x t) (b x) ,
Sn (x, t) =

(1)n+1 (t x)n (x a) , a t x

and the integral in the remainder is taken in the RiemannStieltjes sense.

After that, on utilizing the notations


1
[(b x) f (a) + (x a) f (b)]
ba
n
i
(b x) (x a) X 1 h
+

(x a)k1 f (k) (a) + (1)k (b x)k1 f (k) (b)


ba
k!
k=1

Dn (f ; x, a, b) :=

(2.3.15)

and
1
En (f ; x, a, b) :=
ba


Sn (x, t) d f (n) (t)

(2.3.16)

under the assumptions of Theorem 2.3.11, Dragomir approximated the function f utilizing
the polynomials Dn (f ; x, a, b) with the error En (f ; x, a, b). In other words, we have f (x) =
Dn (f ; x, a, b) + En (f ; x, a, b) for any x [a, b].
More recently, Dragomir (2009) introduced an approximation for the RiemannStieltjes
Rb
integral a f (t) du (t) by the use of some generalized trapezoid-type rules. To be more specific,
Rb
we investigate the error bounds in approximating a f (t) du (t) by the simpler quantities:


1
f (b)
ba



u (t) dt u (a) + f (a) u (b)

1
ba


u (t) dt

(2.3.17)

17

and

[u (b) u (a)] f (b) + f (a)
provided the Riemann integral

Rb
a

1
ba


f (t) dt

(2.3.18)

f (t) dt exists and can be either computed exactly or can be

accurately approximated by the use of various classical quadrature rules.


For a function g : [a, b] R, Dragomir defined g : [a, b] R by
(t a) g (a) + (b t) g (b)
ba

g (t) := g (t)
and he gave the following representation

Theorem 2.3.12. (Dragomir 2009) If f, u : [a, b] R are bounded on [a, b] and such that the
Rb
Rb
RiemannStieltjes integral a f (t) du (t) and the Riemann integral a f (t) dt exist, then
Z






Z b
Z b
1
1
u (t) dt u (a) + f (a) u (b)
u (t) dt
f (t) du (t) f (b)
ba a
ba a
Z b
=
f (t) du (t) (2.3.19)
a

and

[u (b) u (a)] f (b) + f (a)

1
ba

 Z b
f (t) dt
f (t) du (t)
a
Z b
=
u (t) df (t) (2.3.20)
a

Therefore, several error inequalities of approximating the RiemannStieltjes integral

Rb
a

f (t) du (t)

by the generalized trapezoid formulae (2.3.17) and (2.3.18) under various assumptions were
obtained in the same paper.
Mercer (2008) has introduced a midpoint and a trapezoid type rules for the
RiemannStieltjes integral which engender a natural generalization of Hadamards integral
inequality. Error terms are then obtained for this Riemann-Stieltjes Trapezoid Rule and other
related quadrature rules.
Theorem 2.3.13. (Mercer 2008) Let g be continuous and increasing, let c satisfy
Z b
g (t) dt = (c a) g (a) + (b c) g (b) ,
a

18

and let
1
G=
ba
If f 00 0, then we have
f (c) [g (b) g (a)]

g (t) dt

f dg [G g (a)] f (a) + [g (b) G] f (b)

(2.3.21)

2.3.2

Simpsons Type Inequalities

The Simpsons inequality was known in the literature, as follows:


Theorem 2.3.14. (Davis & Rabinowitz 1976) Suppose f : [a, b] R is four times continuously




differentiable mapping on (a, b) and f (4) := sup f (4) (x) < . The following
x(a,b)

inequality






Zb
(b a)4 (4)
1 f (a) + f (b)
a
+
b
1
f

+ 2f

f (x) dx
3

2
2
b

a
2880

(2.3.22)

holds.

In 1999, Dragomir proved the Simpsons inequality for functions of bounded variation,
as follows:
Theorem 2.3.15. (Dragomir 1999) Let f : [a, b] R be a mapping of bounded variation on
[a, b]. Then we have the inequality:

b
Z



b
_


f (x) dx (b a) f (a) + f (b) + 2f a + b (b a)
(f ) ,


3
2
2
3


a

(2.3.23)

where

Wb

(f ) denotes the total variation of f on the interval [a, b]. The constant

1
3

is the best

possible.

In 2000, Pecaric and Varosanec, obtained some inequalities of Simpsons type for
functions whose n-th derivative, n {0, 1, 2, 3} is of bounded variation, as follow:

19

Theorem 2.3.16. (Pecaric & Varosanec 2000) Let n {0, 1, 2, 3}. Let f be a real function on
[a, b] such that f (n) is function of bounded variation. Then

b
Z






f (x) dx (b a) f (a) + 4f a + b + f (b)


6
2


a

Cn (b a)n+1

b
_
a


f (n) , (2.3.24)

where,
1
1
1
1
, C3 =
,
C0 = , C1 = , C2 =
3
24
324
1152
and

Wb


f (n) is the total variation of f (n) on the interval [a, b].
In recent years, many authors have considered Simpsons like inequalities and therefore

several bounds are introduced, for details see Dedic et al. (2000), Dedic et al. (2001), Dedic
et al. (2001), Pecaric and Varosanec (2001), Dedic et al. (2005), Pecaric and Franjic (2006) and
Franjic et al. (2006)
2.4
2.4.1

INEQUALITIES FOR MAPPINGS OF TWO VARIABLES


type inequalities
Ostrowski and Gruss

In the recent papers Barnett and Dragomir (2001) and Dragomir et al. (2003), the authors have
proved the following inequality of Ostrowski type for double integrals:
Theorem 2.4.1. Let f : [a, b] [c, d] R be continuous on [a, b] [c, d]. Then for all

20

(x, y) [a, b] [c, d], we have the inequality



Z bZ

1

(b a) (d c)

f (t, s) dsdt + f (x, y)



Z d
Z b

1
1

f (x, s) ds
f (t, y) dt
(d c) c
(b a) a
"
 a+b 2 # "
 c+d 2 #
2

x
y
f
2f
1
1

2
2

+
+
(b

a)
(d

c)
,
if
L (Q) ;

4
ba
4
dc
ts
ts


h (ba)(dc) i 1q h xa q+1
 i 1 h yc q+1
 i1
dy q+1 q 2 f
bx q+1 q
+ ba
+ dc
ts ,
,

ba
dc
(1+q)2
p

f
1
1

if ts Lp (Q) , p > 1, p + q = 1;


a+b  



x
1 y c+d

2f
2f
1
2
2
2 + ba 2 + dc
L1 (Q)
ts , if ts
a

(2.4.1)

for all (x, y) Q, where,


2
2

f
f (t, s)




ts := sup ts
(t,s)Q

and


2
Z b Z d 2
 p1
f (t, s) p
f




, p 1.
ts dsdt
ts :=
a
c
p
The best inequality we can get from (2.4.1) is the one for which x =

a+b
2

and y =

c+d
.
2

For some applications of the above results in numerical integration for cubature formulae see
Barnett and Dragomir (2001) and Dragomir et al. (2003).
In order to approximate the double integral
introduced the following representation:

RbRd
a

f (t, s) dsdt, Dragomir et al. (2000)

Theorem 2.4.2. (Dragomir et al. 2000) Let f : [a, b] [c, d] R be such that the partial
derivatives

f (t,s) f (t,s) 2 f (t,s)


, s , ts
t

exist and are continuous on [a, b] [c, d]. Then for all (x, y)

21

[a, b] [c, d]. Then, we have the representation


Z bZ d
1
f (x, y) =
f (t, s) dsdt
(b a) (d c) a c
Z bZ d
1
f
+
p (x, t)
(t, s) dsdt
(b a) (d c) a c
t
Z bZ d
1
f
+
q (y, s)
(t, s) dsdt
(b a) (d c) a c
s
Z bZ d
2f
1
p (x, t) q (y, s)
(t, s) dsdt, (2.4.2)
+
(b a) (d c) a c
ts
where, p : [a, b]2 R and q : [c, d]2 R and are given by

t a, t [a, x]
p (x, t) =

t b, t (x, b]
and

q (y, s) =

s c, s [c, y]

s d, s (y, d]

An interesting particular case for which x =

a+b
2

and y =

c+d
2

may be deduced to get a

midpoint representation. Also, by letting (x, y) = (a, c) , (a, d) , (b, c) and (b, d) in (2.4.1), then
summing the obtained identities and do the required computations we obtain successively, we
obtain the following a trapezoid type identity
Z b Z d
1
f (a, c) + f (a, d) + f (b, c) + f (b, d)
=
f (t, s) dsdt
4
(b a) (d c) a c


Z bZ d
Z bZ d
c + d f
a + b f
+
t
(t, s) dsdt +
s
(t, s) dsdt
2
t
2
s
a
c
a
c



Z bZ d
a+b
c + d 2f
+
t
s
(t, s) dsdt (2.4.3)
2
2
ts
a
c
After that Dragomir et al. pointed out an inequality of Ostrowski type for mapping of two
independent variables, as follows:
Theorem 2.4.3. (Dragomir et al. 2000) Let f : [a, b] [c, d] R be such that the partial
derivatives

f (t,s) f (t,s) 2 f (t,s)


, s , ts
t

exist and are continuous on [a, b] [c, d]. Then for all (x, y)

[a, b] [c, d], we have




Z dZ b


1
M1 (x) + M2 (y) + M3 (x, y) ,
f (x, y)
f
(t,
s)
dtds


(b a) (d c) c a

(2.4.4)

22

where,

M1 (x) =

2
f
(x a+b
2 )
,
ba
t

1
(ba)2 +
4

(bx)q1 +1 +(xa)q1 +1
q1 +1

1
(ba)(dc) p1

 1
q1

a+b
1

[ 2 (ba)+|x 2 |]
f
(ba)(dc)

M2 (y) =

f (t,s)
t

f
,
t p1

1
(dc)2 +
4

(dy)q2 +1 +(yc)q2 +1
q2 +1

f (t,s)
t

1
(dc)(ba) p2

 1
q2

f (t,s)
s

f
,
s p2

c+d
1

[ 2 (dc)+|y 2 |]
f ,
(ba)(dc)
s 1

and

M3 (x, y) =

ih
i
2
14 (ba)2 +(x a+b
2 )

1
(ba)2 +
4

(bx)q3 +1 +(xa)q3 +1
q3 +1

(x a+b
2 )

(ba)(dc)

Lp1 (Q) ,

p1 > 1, p11 +

2
f
(y c+d
2 )
,
dc
t

f (t,s)
t

L (Q) ;

f (t,s)
s

L1 (Q) .

Lp2 (Q) ,

p2 > 1, p12 +

(ba)(dc)

a+b
c+d
1
1

f
[ 2 (ba)+|x 2 |][ 2 (dc)+|y 2 |]
ts
,
(ba)(dc)
1

2
f
ts ,
p3

1
q2

= 1;

L1 (Q) .

2
f
ts ,

= 1;

L (Q) ;

f (t,s)
s

 1 
 1
q3 (dy)q3 +1 +(yc)q3 +1 q3

q +1

1
q1

2 f (t,s)
ts

L (Q) ;

2 f (t,s)
ts

Lp3 (Q) , .

p3 > 1, p13 +
2 f (t,s)
ts

1
q3

= 1;

L1 (Q) .

Therefore, several special cases, e.g., midpoint and trapezoid type inequalities were
obtained in Dragomir et al. (2000), as well as applications to cubature formulae are considered.
Hanna et al. (2002a) have obtained some generalizations of an Ostrowski type inequality
in two dimensions for n-time differentiable mappings
Theorem 2.4.4. (Hanna et al. 2002a) Let f : [a, b] [c, d] R be continuous on [a, b] [c, d]
, and assume that

n+m f
tn sm

exist on (a, b) (c, d). Further, consider Kn : [a, b]2 R and

23

Sm : [c, d]2 R given by


(ta)n

n! , t [a, x]
Kn (x, t) =
,

(tb)n
, t (x, b]
n!

Sm (y, s) =

Then we have the inequality

(sc)m
,
m!

s [c, y]
.

(sd)m
,
m!

s (y, d]

Z Z
n1 m1
b d
X
X
k+l f (x, y)

f (t, s) dsdt
Xk (x) Yl (y)

a c
tk sl
k=0 l=0
Z d
n1
X
k+m f (x, s)
m
(1)
Xk (x)
Sm (y, s)
ds
tk sm
c
k=0

Z d
m1
X
n+l f (t, y)
n
dt
(1)
Yl (y)
Kn (x, t)

tn sl
c
l=0


n+1
n+1  
m+1
m+1  n+m f
1

(x

a)
+
(b

x)
(y

c)
+
(d

y)

(n+1)!(m+1)!
tn sm

n+m

if t
n sm L (Q)


h
i1 h
i1

1 (xa)nq+1 +(bx)nq+1 q (dc)mq+1 +(dy)mq+1 q n+m f

,

n
m
n!m!
nq+1
mq+1
t s

p
n+m

f
1
1

if t
n sm Lp (Q) , p > 1, p + q = 1;

[(x a)n + (b x)n + |(x a)n (b x)n |]

4n!m!

n+m
n+m

[(y c)m + (d y)m + |(y c)m (d y)m |]


f ; if f L (Q)
tn sm

tn sm

(2.4.5)

for all (x, y) Q, where,


n+m
n+m





f
(t,
s)
f

:= sup

tn sm
tn sm ,
(t,s)Q

n+m
p
Z b Z d n+m
 p1




f
f (t, s)


, p 1,
tn sm :=
tn sm dsdt
a
c
p

and
Xk (x) =

(b x)k+1 + (1)k (x a)k+1


,
(k + 1)!

Yl (y) =

(d y)l+1 + (1)l (y c)l+1


.
(l + 1)!

24

Keeping in mind that x and y are free parameters, then one can produce mid-point and
boundary-point type results by choosing appropriate values for x and y. In addition choosing
values for n and m will re-capture the earlier results of Hanna et al. (2000) and Dragomir et al.
(2000).
In order to compare the integral mean of the product with the product of the integral

means, Gruss (1935) have considered the Ceby


sev functional defined by
Z b
Z b
Z b
1
1
1
f (t) g (t) dt
f (t) dt
g (t) dt,
C (f, g) =
ba a
ba a
ba a

(2.4.6)

and he has proved that for two integrable mappings f, g such that f (x) and
f (x) , the inequality
|C (f, g)|
holds, and the constant

1
4

1
( ) ( )
4

(2.4.7)

is the best possible.

The proof of (2.4.7) may be done by applying the CauchyBunyakovskySchwarz


integral inequality for double integrals, on the right hand side of the well-known Korkine
identity,
(b a)

f (x) g (x) dx

f (x) dx
g (x) dx
a
Z Z
1 b b
[f (x) f (y)] [g (x) g (y)] dxdy (2.4.8)
=
2 a a

After that, and in order to represent the remainder of the Taylor formula in an integral
form which will allow a better estimation using the Gruss type inequalities, Hanna et al.
(2002b), generalized the above Korkine identity (2.4.8), for double integrals and therefore Gruss
type inequalities were proved. Namely, they proved
Theorem 2.4.5. (Hanna et al. 2002b) We assume that
|f (x, y) f (u, v)| M1 |x u|1 + M2 |x u|2
where, M1 , M2 > 0, 1 , 2 (0, 1] and
|f (x, y) f (u, v)| N1 |x u|1 + N2 |x u|2

25

where, N1 , N2 > 0, 1 , 2 (0, 1] for all (x, y), (u, v) [a, b] [c, d], then we have the
following inequality:

Z bZ

1

(b a) (d c)
a

(b a) (d c)

f (x, y) g (x, y) dydx


Z bZ d
f (x, y) dydx
a

c
1 +1

1
(b a) (d c)

Z bZ
a

c
1



g (x, y) dydx

2 (b a) (d c)2
(b a)
+ M1 N 2
M1 N 1
(1 + 1 + 1) (1 + 1 + 2)
(1 + 1) (1 + 2) (2 + 1) (2 + 2)
+ M2 N 1

2 (b a)2 (d c)1
(b a)2 +2
+ M2 N 2
(2.4.9)
(2 + 1) (2 + 2) (1 + 1) (1 + 2)
(2 + 2 + 1) (2 + 2 + 2)

Corollary 2.4.6. When 1 = 2 = 1 and 1 = 2 = 1, then


|f (x, y) f (u, v)| L1 |x u| + L2 |x u| ,

|f (x, y) f (u, v)| K1 |x u| + K2 |x u|


where, L1 , L2 , K1 , K2 > 0, 1 , 2 (0, 1] and then (2.4.9) becomes

Z bZ

1

(b a) (d c)
a

(b a) (d c)
2

L1 K 1

f (x, y) g (x, y) dydx


Z bZ d
f (x, y) dydx
a

1
(b a) (d c)

Z bZ
a



g (x, y) dydx

(b a)
(b a) (d c)
(b a) (d c)
(d c)2
+ L1 K 2
+ L2 K 1
+ L2 K 2
. (2.4.10)
12
18
18
12

For approximation of multiple integrals see the classical book Stroud (1971).

Several

multivariate and univariate higher order Ostrowski type inequalities over Euclidean domains
as well as multivariate Fink type identity were recently proved in Anastassiou (1995),
Anastassiou (1997), Anastassiou (2002), Pachpatte (2002b), Pachpatte (2001), Pachpatte
(2002a) Anastassiou (2007) and Anastassiou and Goldstein (2007) and (2008). For recent
comprehensive list of refinements, counterparts and generalizations of Ostrowski integral
inequality see, Mitrinovic et al. (1993) and Dragomir and Rassias (2002), as well as the recent
PhD study, Hanna (2009) where the author introduced new multivariate approximations from a
generalized Taylor perspective in terms of Appell type polynomials.

26

2.4.2

Simpsons Type Inequalities

Zhongxue (2008), has proved the following Simpson type inequality for mappings of two
independent variables:
Theorem 2.4.7. (Zhongxue 2008) Let f : [a, c] [b, d] R be an absolutely continuous
function, whose partial derivative of order 2 is f 00 L2 ([a, c] [b, d]). Then

f a, b+d  + f a+c , b + f a+c , d + f c, b+d  + 4f a+c , b+d 

2
2
2
2
2
2


9


Rc
b+d
f
(s,
b)
+
4f
s,
+
f
(s,
d)
ds
f (a, b) + f (a, d) + f (c, b) + f (c, d)
2
+
a
36
6 (c a)



Rd
R
R
c
d

a+c
f
(a,
t)
+
4f
,
t
+
f
(c,
t)
dt
f
(s,
t)
dsdt

2
b
+ a b

6 (d b)
(c a) (d b)

[(c a) (d b)]1/2 p

(f 00 ), (2.4.11)
36

where () is defined by
(f ) =

kf k22

(c a) (d b)

Z c Z
a

2
f (s, t) dsdt ,

(2.4.12)

and
kf k22

Z c Z
a

1/2
|f (s, t)| dsdt
.
2

The inequality (2.4.11) is sharp in the sense that the constant 1/36 cannot be replaced by a
smaller one.
Another interesting result was considered by Zhongxue (2008), as follows:
Theorem 2.4.8. (Zhongxue 2008) Under the assumptions of Theorem 1, for any (x, y) [a, c]
[b, d], we have

Z c
Z d

(c a) (d b) f (x, y) (d b)
f (s, y) ds (c a)
f (x, t) dt

a
b



b+d
a+c
y
[f (c, d) f (a, d) f (c, b) + f (a, b)]
x
2
2
3

Z cZ d
[7 (c a) (d b)] 2 p
+
f (s, t) dsdt
(fst00 ), (2.4.13)
12
a
b

where (fst00 ) is defined above. The Inequality (2.4.13) is sharp in the sense that the constant

7 7
12

cannot be replaced by a smaller one.

27

In his recent work, Liu (2010) has derived a new sharp inequality with a parameter for
the absolutely continuous function f : [a, c] [b, d] R whose partial derivative of order
2 is fst00 L2 ([a, c] [b, d]) via the new sharp bound (2.4.12), which will not only provide a
generalization of inequalities (2.4.11) and (2.4.13), but also gives some other interesting sharp
inequalities as special cases.
A generalization of (2.4.11) is considered recently by Liu (2010), as follows:
Theorem 2.4.9. (Liu 2010) Let the assumptions of Theorem 1 hold. Then for any [0, 1] and
(x, y) [a, c] [b, d], we have



(c a) (d b) (1 )2 f (x, y) + (1 ) [f (a, y) + f (c, y) + f (x, b) + f (x, d)]

2

2
+ [f (a, b) + f (a, d) + f (c, b) + f (c, d)]
4



b+d
a+c
2
y
[f (c, d) f (a, d) f (c, b) + f (a, b)]
(1 ) x
2
2
Z
db c

[f (s, b) + 2 (1 ) f (s, y) + f (s, d)] ds


2
a

Z
Z cZ d

ca d

[f (a, t) + 2 (1 ) f (x, t) + f (c, t)] ds +


f (s, t) dsdt
2
b
a
b

a+c
(2 ) (1 ) (c a) (d b) x
2
2

2 
2
b+d
y
2

(1 ) (1 3 32 )
+
(c a) (d b)
12
"

2

2 #
a+c
b+d
2
2
(d b) x
+ (c a) y
2
2
)1
2
2 p
(1 3 32 )
(c a)3 (d b)3
+
(fst00 )
144

(2.4.14)

where (fst00 ) is defined above. The inequality (2.4.14) is sharp in the sense that the coefficient
constant 1 of the right-hand side cannot be replaced by a smaller one.

28

In special case, if we set x =



f a, b+d  + f

2

a+c
,b
2

a+c
2

+f

and y =

a+c
,d
2

b+d
2

with = 31 , we get


+ f c, b+d
+ 2f
2

a+c b+d
, 2
2



Rc
f (s, b) + 2f s, b+d
+ f (s, d) ds
f (a, b) + f (a, d) + f (c, b) + f (c, d)
2
a
+

16
4 (c a)



Rd
RcRd

a+c
f
(a,
t)
+
2f
,
t
+
f
(c,
t)
dt
f
(s,
t)
dsdt

2
b
a b
+


4 (d b)
(c a) (d b)

[(c a) (d b)]1/2 p
(f 00 ), (2.4.15)
48

CHAPTER III

OSTROWSKIS TYPE INEQUALITIES

3.1

INTRODUCTION

In this chapter, several new inequalities of Ostrowskis type are introduced. Trapezoid and
Midpoint type rules for double RSdouble integral are proved. A generalization of the well
known BeesackDarstPollard inequality for double RSdouble integrals is also considered.
Finally, as applications, two cubature formulae are proposed.
3.2

PRELIMINARIES AND LEMMAS

In this section, we introduce some fundamental inequalities concerning RiemannStieltjes


double integrals. Namely, we first prove integration by parts formula for the RiemannStieltjes
double integral and then using the concept of bounded bi-variation, bi-monotonic and Lipschitz
mappings to generalize some basic and wellknown inequalities for double integrals in the
RiemannStieltjes sense.
Frechet (1910) has given the following characterization for the double RiemannStieltjes
integral. Assume that f (x, y) and (x, y) are defined over the rectangle
Q : (a x b ; c y d);
let R be the divided into rectangular subdivisions, or cells, by the net of straight lines x = xi ,
y = yj ,
a = x0 < x1 < < xn = b,

and

c = y0 < y1 < < ym = d;

let i , j be any numbers satisfying the inequalities xi1 i xi , yj1 j yj , (i =


1, 2, , n; j = 1, 2, , m); and for all i, j let
11 (xi , yj ) = (xi1 , yj1 ) (xi1 , yj ) (xi , yj1 ) + (xi , yj ) .

30

Then if the sum

n X
m
X

S=

f (i , j ) 11 (xi , yj )

i=1 j=1

tends to a finite limit as the norm of the subdivisions approaches zero, the integral of f with
respect to is said to exist. We call this limit the restricted integral, and designate it by the
symbol

Zb Zd
a

f (x, y) dx dy (x, y).

(3.2.1)

If in the above formulation S is replaced by the sum


S =

n X
m
X

f (ij , ij ) 11 (xi , yj ),

i=1 j=1

where ij , ij are any numbers satisfying the inequalities xi1 ij xi , yj1 ij yj , we


call the limit, when it exists, the unrestricted integral, and designate it by the symbol
Zb Zd
a

f (x, y) dx dy (x, y).

(3.2.2)

The existence of (3.2.2) implies both the existence of (3.2.1) and its equality to (3.2.2). On the
other hand, Clarkson (1933) has shown that the existence of (3.2.1) does not imply the existence
of (3.2.2) (see Clarkson (1933)).
Lemma 3.2.1. (Integration by parts) If f RS() on Q, then RS(f ) on Q, and we have
Z dZ b
Z dZ b
(t, s) dt ds f (t, s)
f (t, s) dt ds (t, s) +
c

= f (b, d) (b, d) f (b, c) (b, c) f (a, d) (a, d) + f (a, c) (a, c) .

Proof. Let  > 0 be given. Since

RdRb
c

(3.2.3)

f (t, s) dt ds (t, s) exists, there is a partition P of Q

such that for every P 0 finer than P , we have




Z dZ b


0
< .
S (P , f, )
f
(t,
s)
d
d

(t,
s)
t
s


c

(3.2.4)

31

Consider an arbitrary RiemannStieltjes sum for the integral (t, s) dt ds f (t, s), say
S (P, f, ) =
=

m X
n
X

j=1 i=1
m X
n
X

(ti , sj ) 11 f (xi , yj )
(ti , sj ) f (xi1 , yj1 )

j=1 i=1
m X
n
X

m X
n
X

(ti , sj ) f (xi1 , yj )

j=1 i=1
m X
n
X

(ti , sj ) f (xi , yj1 ) +

j=1 i=1

(ti , sj ) f (xi , yj ),

j=1 i=1

where P finer than P . Writing


A = f (b, d) (b, d) f (b, c) (b, c) f (a, d) (a, d) + f (a, c) (a, c) ,
we have the identity
A=

m X
n
X

f (xi1 , yj1 ) (xi1 , yj1 )

j=1 i=1
m X
n
X

f (xi , yj1 ) (xi , yj1 ) +

j=1 i=1

m X
n
X

j=1 i=1
m X
n
X

f (xi1 , yj ) (xi1 , yj )
f (xi , yj ) (xi , yj ).

j=1 i=1

Subtracting the last two displayed equations, we find


A S (P, f, ) =

m X
n
X

f (xi1 , yj1 ) [ (xi1 , yj1 ) (ti , sj )]

j=1 i=1
m X
n
X

+
+

j=1 i=1
m X
n
X

j=1 i=1
m X
n
X

f (xi1 , yj ) [ (ti , sj ) (xi1 , yj )]


f (xi , yj1 ) [ (ti , sj ) (xi , yj1 )]
f (xi , yj ) [ (xi , yj ) (ti , sj )].

j=1 i=1

The sums on the right can be combined into a single sum of the form S (P 0 , f, ), where P 0 is
that partition of Q obtained by taking the points (ti , sj ), (xi , yj ) together. Then P 0 is finer than
P and hence finer than P . Therefore the inequality (3.2.4) is valid and this means that we have


Z dZ b


A S (P, f, )
f (t, s) dt ds (t, s) < ,

c

whenever P is finer than P . But this is exactly the statement


RdRb
and equals A c a f (t, s) dt ds (t, s).

RdRb
c

(t, s) dt ds f (t, s) exists

32

Lemma 3.2.2. If f is continuous on Q and if is of bounded bivariation on Q, then f RS().

Proof. First of all, we note that, by Lemma 3.2.1, a second sufficient condition can be obtained
by interchanging f and in the hypothesis. It suffices to prove the theorem when is
bi-monotonically increasing with (a, ) (b, ), (, c) (, d) and (t, s) (x, y),
for all t < x and s < y in Q. Since f is continuous on Q then f is uniformly continuous on Q,
i.e.,  > 0 there exists > 0, such that
|f (x, y) f (t, s)| <


A

whenever

k(x t, y s)k < ,

where A = 4 [ (b, d) (b, c) (a, d) + (a, c)]. If P is a partition of Q with kP k < ,
then for P finer than P we must have Mij (f ) mij (f ) /A, where
Mij (f ) mij (f ) = sup {f (x, y) f (x, s) f (t, y) + f (t, s) : (x, y) , (t, s) Q} .
Multiplying the inequality by 11 and summing, we find
m


 XX
11 = < .
U (P, f, ) L (P, f, )
A j=1 i=1
4
Hence, f RS() on Q.
Lemma 3.2.3. Assume that g RS() on Q and is of bounded bivariation on Q, then
Z d Z b

_



sup |g (x, y)|
g
(x,
y)
d
d

(x,
y)
() .
(3.2.5)
x
y


c

Proof. The existence of

(x,y)Q

RdRb
c

g (x, y) dx dy (x, y) follows from Lemma 3.2.4. Let n := a =

x0 < x1 < < xn = b and m := c = y0 < y1 < < ym = d be a partitions of [a, b] and
[c, d]; respectively. Let
n,m :=

(x0 , y0 ) , , (x0 , ym ) , (x1 , y0 ) , , (x1 , ym ) , , (xn , y0 ) , , (xn , ym )

33

be a partition of Q and l (n,m ) := max {xi+1 xi , yj+1 yj } be the length of Q, therefore,


i,j



m1
n1 


XX


(n)
(m)

g (x, y) dx dy (x, y) = lim
g i , j
11 (xi , yj )
l(n,m )0

j=0 i=0

lim

l(n,m )0

m1
n1
XX
j=0 i=0




(n) (m)
g i , j
|11 (xi , yj )|

sup |g (x, y)| sup

n,m

(x,y)Q

= sup |g (x, y)|


(x,y)Q

m1
n1
XX

|11 (xi , yj )|

j=0 i=0

() ,

which is required.
Lemma 3.2.4. Let g be a continuous mapping on Qb,d
a,c and is bi-monotonic non-decreasing
on Qb,d
a,c , then
Z


Proof. Let
nm :=

Z

g (x, y) dx dy (x, y)

|g (x, y)| dx dy (x, y)

(3.2.6)

(x0 , y0 ) , , (x0 , ym ) , (x1 , y0 ) , , (x1 , ym ) , , (xn , y0 ) , , (xn , ym )

be a partition of Q and l (nm ) := max {xi+1 xi , yj+1 yj } be the length of Q, therefore,


i,j



m1
n1 


XX


(n) (m)

g (x, y) dx dy (x, y) = lim
g i , j
11 (xi , yj )
l(n,m )0

j=0 i=0

lim

l(n,m )0

m1
n1
XX
j=0 i=0

which is required.




(n) (m)
g i , j
|11 (xi , yj )|

Lemma 3.2.5. Let g, : Q R, be such that g is LLipschitz on Q and is


RdRb
Riemann-integrable on Q then the RiemannStieltjes integral c a g (x, y) dx dy (x, y) exists
and the inequality

holds


Z

g (x, y) dx dy (x, y) L

|g (x, y)| dxdy,

(3.2.7)

34

Proof. The existence of

RdRb
c

g (x, y) dx dy (x, y) follows from Lemma 3.2.2. Let n := a =

x0 < x1 < < xn = b and m := c = y0 < y1 < < ym = d be a partitions of [a, b] and
[c, d]; respectively. Let
nn := {(x0 , y0 ) , (x1 , y1 ) , , (xn , y0 ) , , (xn , yn )}
be a partition of Q and l (nn ) := max {xi+1 xi , yj+1 yj } be the length of Q, therefore,
i,j


Z d Z b

n1 X
n1 


X



(n) (n)


g
(x,
y)
d
d

(x,
y)
=
lim
g

[
(x
,
y
)

(x
,
y
)]

x y
i i
i1 i1
i
j

l(nn )0

c
a
j=0 i=0

lim

l(nn )0

L
=L

n1 X
n1
X
j=0 i=0

lim

l(nn )0




(n) (n)
g i , j | (xi , yi ) (xi1 , yi1 )|

n1 X
n1 

X

(n) (n)
g i , j |(xi , yi ) (xi1 , yi1 )|
j=0 i=0

|g (x, y)| dxdy,

which is required.
3.3

OSTROWSKI INEQUALITY FOR MAPPINGS BOUNDED BIVARIATION

In this section and in order to approximate the RiemannStieltjes double integral, some of
Ostrowski, trapezoid and Simpson type inequalities are proved.
We begin with the following generalization of (2.3.4):
Theorem 3.3.1. Let f : Q R be a mapping of bounded bivariation on Q. Then for all
(x, y) Q, we have the inequality


Z dZ b


(b a) (d c) f (x, y)

f
(t,
s)
dtds


 


 c a
a + b
d c
c + d _
b a
+ x

+ y

(f ) , (3.3.1)

2
2
2
2
Q

where

(f ) denotes the total (double) bivariation of f on Q.

Proof. From Lemma 3.2.1, we have


Z yZ x
Z
(t a) (s c) dt ds f (t, s) = (x a) (y c) f (x, y)
c

f (t, s) dtds,

35

(t a) (s d) dt ds f (t, s) = (x a) (d y) f (x, y)

(t b) (s c) dt ds f (t, s) = (b x) (y c) f (x, y)

f (t, s) dtds,

f (t, s) dtds,

and

(t b) (s d) dt ds f (t, s) = (x b) (y d) f (x, y)

Adding the above equalities, we get


Z dZ b
Z
P (x, t; y, s) dt ds f (t, s) = (b a) (d c) f (x, y)
c

f (t, s) dtds.

f (t, s) dtds

where,

P (x, t; y, s) =

(t a) (s c) ,

(t a) (s d) ,

(t b) (s c) ,

(t b) (s d) ,

for all (t, s) Q.

(x, y) [a, x] [c, y]


(x, y) [a, x] (y, d]
(x, y) (x, b] [c, y]
(x, y) (x, b] (y, d]

Now, applying Lemma 3.2.3, by letting g = P and = f , we get


Z d Z b





P
(x,
t;
y,
s)
d
d
f
(t,
s)
t
s


c

sup |P (x, t; y, s)|


(x,y)Q

(f )

= max {(x a) (y c) , (x a) (d y) , (b x) (y c) , (b x) (d y)}


x,y

(f ) ,

but,
M = max {(x a) (y c) , (x a) (d y) , (b x) (y c) , (b x) (d y)}
x,y


= max max {(x a) (y c) , (x a) (d y) , (b x) (y c) , (b x) (d y)} ,
x

36

and since max is independent of x, we have


y


M = max (x a) max {(y c) , (d y)} , (b x) max {(y c) , (d y)}
x

max {(x a) , (b x)} max {(y c) , (d y)}


x
y

 



b a
a + b
d c
c + d
=
+ x

+ y
,
2
2
2
2

it follows that,
Z d Z b




P (x, t; y, s) dt ds f (t, s)

c
a

 



a + b
d c
c + d _
b a
+ x

+ y
(f ) ,

2
2
2
2
Q

which completes the proof.

Corollary 3.3.2. In Theorem 3.3.1. Let x = a+b


and y = c+d
, then we have
2
2



 Z dZ b

(b a) (d c) _
c
+
d
a
+
b
(b a) (d c) f

,

f
(t,
s)
dtds

(f ) ,


2
2
4
c
a
Q

Remark 3.3.3. Similar inequalities can be found if we assume that u is monotonous on Q, we


left the details to the interested reader.
Corollary 3.3.4. In Theorem 3.3.1. Assume [a, b] = [c, d], we get


Z bZ b


(b a)2 f (x, y)
f (t, s) dtds

a
a

 



a + b
b a
a + b _
b a
+ x

+ y
(f ) .

2
2
2
2
Q

A generalization of the trapezoid inequality (2.3.5) for mappings of two variables may

be stated as follows:
Theorem 3.3.5. Let f : Q R be a mapping of bounded bivariation on Q. Then for all
(x, y) Q, we have the inequality


Z dZ b
(b a) (d c)


[f
(b,
d)

f
(b,
c)

f
(a,
d)
+
f
(a,
c)]

f
(t,
s)
dtds


4
c
a
(b a) (d c) _

(f ) , (3.3.2)

4
Q
The constant

1
4

is best possible value.

37

Proof. From Lemma 3.2.1, we have


d

R (t, s) dt ds f (t, s)

(b a) (d c)
[f (b, d) f (b, c) f (a, d) + f (a, c)]
=
4


c+d
where, R (t, s) = t a+b
s

, a t b ; c s d.
2
2

f (t, s) dtds,

Now, applying Lemma 3.2.3, by letting g = R and = f , we get


d _
b
_

(b a) (d c) _
(f ) ,
R (t, s) dt ds f (t, s) sup |R (t, s)|
(f ) =
4
(t,s)Q

which is required.

The following theorem generalize the inequality (3.3.2).


Theorem 3.3.6. Let f : Q R be a mapping of bounded bivariation on Q. Then for all
(x, y) Q, we have the inequality

Z

(f ) (b, d) (f ) (b, c) (f ) (a, d) + (f ) (a, c)



f (t, s) dt ds (t, s)
c
a
_
sup | (t, s)|
(f ) . (3.3.3)
Z

(t,s)Q

Proof. From Lemma 3.2.1, we have


Z

(t, s) dt ds f (t, s) +

f (t, s) dt ds (t, s)

= f (b, d) (b, d) f (b, c) (b, c) f (a, d) (a, d) + f (a, c) (a, c) .


Now, applying Lemma 3.2.3, by letting g = R and = f , we get
Z d Z b

_



(t,
s)
d
d
f
(t,
s)

sup
|
(t,
s)|

(f ) ,
t s


c

which is required.

The following result holds

(t,s)Q

38

Theorem 3.3.7. Let u : Q R be a function of bounded bivariation and f : Q R a function


such that there exists the constants m, M R with m f (t, s) M , for each (t, s) Q, and
RdRb
the Stieltjes integral c a f (t, s) dt ds u (t, s) exists. Then, by defining the error functional
(f, u, m, M ; Q)
Z dZ b
m+M
:=
f (t, s) dt ds u (t, s)
[u (b, d) u (b, c) u (a, d) + u (a, c)] (3.3.4)
2
c
a
we have the bound
d

__
1
(u)
(f, u, m, M ; Q) (M m)
2
c a

(3.3.5)

Proof. Since, obviously, the function f m+M


satisfies the inequality
2



1
m
+
M
f (t, s)
(M m) , (t, s) Q

2 2
and the Stieltjes integral
Z


Z b
a

RdRb
c

m+M
f (t, s)
2

f (t, s)

m+M
2



dt ds u (t, s)

dt ds u (t, s) exists, then

 _

d _
b
d _
b
_
1
m+M

(u) (M m)
(u)
sup f (t, s)
2
2
(t,s)Q
c a
c a

and the inequality (3.3.5) is proved.

Now, we consider an Ostrowski type inequality for (1 , 2 )Holder type mapping on the
co-ordinate.
Definition 3.3.8. A function f : Qb,d
older type mapping on the
a,c R is to be of (1 , 2 )H
co-ordinate, if for all (t1 , s1 ) , (t1 , s1 ) Qb,d
a,c , there exist H1 , H2 > 0 and 1 , 2 > 0 such that
|f (t1 , s1 ) f (t2 , s2 )| H1 |t1 t2 |1 + H2 |s1 s2 |2 .
If 1 = 2 = 1, then f is called (L1 , L2 )Lipschitz on the co-ordinate, i.e.,
|f (t1 , s1 ) f (t2 , s2 )| L1 |t1 t2 | + L2 |s1 s2 | .

39

Theorem 3.3.9. Let f : Qb,d


older type mapping on the co-ordinate, i.e.,
a,c R be a (1 , 2 )H
for all (t1 , s1 ) , (t1 , s1 ) Qb,d
a,c , there exist H1 , H2 > 0 and 1 , 2 > 0 such that
|f (t1 , s1 ) f (t2 , s2 )| H1 |t1 t2 |1 + H2 |s1 s2 |2 ,
b,d
b,d
and u : Qb,d
a,c R be a mapping of bounded bivariation on Qa,c . Then for all (x, y) Qa,c , we

have the inequality




Z dZ b


[u (b, d) u (b, c) u (a, d) + u (a, c)] f (x, y)

f
(t,
s)
d
d
u
(t,
s)
t
s


c
a

1

2 ! _


a + b
d c
c + d
b a
+ x
+
H
+
y

(u). (3.3.6)
H1
2

2
2
2
2
b,d
Qa,c

Proof. From Lemma 3.2.3, we have




Z dZ b


f (x, y) (u (b, d) u (b, c) u (a, d) + u (a, c))

f
(t,
s)
d
d
u
(t,
s)
t
s


c
a
Z d Z b



=
(f (x, y) f (t, s)) dt ds u (t, s)
c

sup

|f (x, y) f (t, s)|

(t,s)Qb,d
a,c

(u) .

Qb,d
a,c

Now, since f is a (1 , 2 )Holder on the co-ordinate, we have


sup

|f (x, y) f (t, s)|

(t,s)Qb,d
a,c



1
2
H1 |x t| + H2 |y s|

sup
(t,s)Qb,d
a,c

= H1 sup |x t|1 + H2 sup |y s|2


t[a,b]

which follows that,

s[c,d]
1

= H1 max (x a) , (b x)
+ H2 max (y c) , (d y)

1




a + b
d c
c + d 2
b a
+ x
+ H2
+ y
= H1
2
2
2
2



Z dZ b


[u (b, d) u (b, c) u (a, d) + u (a, c)] f (x, y)
f (t, s) dt ds u (t, s)

c
a

1

 !


a + b
d c
c + d 2 _
b a
+ x
+ H2
+ y
(u),
H1
2
2
2
2
b,d
Qa,c

which completes the proof.

40

Corollary 3.3.10. Let u as in Theorem 3.3.13 and let f : Qb,d


a,c R be an LLipschitzian
b,d
mapping on the co-ordinate on Qb,d
a,c , i.e., for all (t1 , s1 ) , (t1 , s1 ) Qa,c , there exist L1 , L2 > 0

such that
|f (t1 , s1 ) f (t2 , s2 )| L1 |t1 t2 | + L2 |s1 s2 | .
Then for all (x, y) Qb,d
a,c , we have the inequality


Z dZ b


[u (b, d) u (b, c) u (a, d) + u (a, c)] f (x, y)
f (t, s) dt ds u (t, s)


c a 




a + b
d c
c + d
ba
+ x
+ L2
+ y
(3.3.7)
L1
2
2
2
2

b,d
Theorem 3.3.11. Let f : Qb,d
a,c R be a mapping of bounded bivariation on Qa,c and u :

Qb,d
older mapping on the co-ordinate. Then for all (x, y) Qb,d
a,c R be a (1 , 2 )H
a,c , we
have the inequality
h

| (f, u; x, a, b; y, c, d)| H1 (x a)

y x
i _
_
+ H2 (y c)
(f )
2

+ H1 (b x)

y b
i _
_
+ H2 (y c)
(f )
2

h
i
+ H1 (x a)1 + H2 (d y)2
h

+ H1 (b x)1

c x
d _
x
_
y

(f )

d _
b
i _
2
+ H2 (d y)
(f ) ,
y

(3.3.8)

where,
(f, u; x, a, b; y, c, d)
:= [u (b, d) u (b, c) u (a, d) + u (a, c)] f (x, y)

f (t, s) dt ds u (t, s)

is the Ostrowskis functional associated to f and u as above.


Proof. As u is continuous and f is of bounded bivariation on Qb,d
a,c , the following double
RiemannStieltjes integrals exist and, by the integration by parts formula, we can state that
Z yZ x
I1 :=
(u (t, s) u (a, c)) dt ds f (t, s)
c
a
Z yZ x
= [u (x, y) u (x, c) u (a, y) + u (a, c)] f (x, y)
f (t, s) dt ds u (t, s),
c

41

I2 :=

(u (t, s) u (b, c)) dt ds f (t, s)

= [u (b, y) u (b, c) u (x, y) + u (x, c)] f (x, y)

I3 :=

f (t, s) dt ds u (t, s),

(u (t, s) u (a, d)) dt ds f (t, s)

= [u (x, d) u (x, y) u (a, d) + u (a, y)] f (x, y)

f (t, s) dt ds u (t, s),

and
I4 :=

(u (t, s) u (b, d)) dt ds f (t, s)

= [u (b, d) u (b, y) u (x, d) + u (x, y)] f (x, y)

f (t, s) dt ds u (t, s).

If we add the above identities, we obtain


Z yZ x
(f, u; x, a, b; y, c, d) =
(u (t, s) u (a, c)) dt ds f (t, s)
c
a
Z yZ b
+
(u (t, s) u (b, c)) dt ds f (t, s)
c
x
Z dZ x
+
(u (t, s) u (a, d)) dt ds f (t, s)
y

(u (t, s) u (b, d)) dt ds f (t, s)

42

Now, using the properties of modulus, we have:


Z y Z x



| (f, u; x, a, b; y, c, d)|
(u (t, s) u (a, c)) dt ds f (t, s)
c
Za y Z b




+
(u (t, s) u (b, c)) dt ds f (t, s)
Zc d Zx x



+
(u (t, s) u (a, d)) dt ds f (t, s)
y
a
Z d Z b




+
(u (t, s) u (b, d)) dt ds f (t, s)
y

sup

(t,s)Qx,y
a,c

|u (t, s) u (a, c)|

sup

y x
_
_
c

|u (t, s) u (b, c)|

(t,s)Qb,y
x,c

sup

sup

y b
_
_
c

|u (t, s) u (a, d)|

(t,s)Qx,d
a,y

(f )

(t,s)Qb,d
x,y

d _
x
_
y

|u (t, s) u (b, d)|

(f )

d _
b
_
y

(f )

(f ) .

However,
|u (t, s) u (a, c)| H1 |t a|1 + H2 |s c|2 ,
so that,
sup

(t,s)Qx,y
a,c

|u (t, s) u (a, c)|

sup

(t,s)Qx,y
a,c

H1 |t a|1 + H2 |s c|2

= H1 (x a)1 + H2 (y c)2 .
Similarly, for
sup

|u (t, s) u (b, c)| H1 (b x)1 + H2 (y c)2 ,

(t,s)Qb,y
x,c

sup

|u (t, s) u (a, d)| H1 (x a)1 + H2 (d y)2

(t,s)Qx,d
a,y

and
sup

|u (t, s) u (b, d)| H1 (b x)1 + H2 (d y)2 ,

(t,s)Qb,d
x,y

in obtaining the above inequalities we get the required result.

43

Remark 3.3.12. In the above results, if one chooses x =

a+b
2

and y =

c+d
,
2

we get inequalities

of midpoint type for mappings of two independent variables.

In the following, we generalize the inequality (2.3.11) which is a companion of


Ostrowskis inequality for mappings of two variables:
b,d
Theorem 3.3.13. Let f : Qb,d
a,c R be a mapping of bounded bivariation on Qa,c . Then for all
a+b c+d
,
2

(x, y) Qa,c2

, we have the inequality


(b a) (d c)

[f (x, y) + f (a + b x, y) + f (x, c + d y) + f (a + b x, c + d y)]

4

Z dZ b

f (t, s) dtds
c
a

 

 d b

b a
3a + b
d c
3c + d _ _

+ x

+ y

(f ) . (3.3.9)
4
4
4
4
c a

Proof. From Lemma 3.2.1, we have


Z yZ x
(t a) (s c) dt ds f (t, s) = (x a) (y c) f (x, y)
c

a+bx


a+b
(s c) dt ds f (t, s)
t
2




a+b
a+b
=
x (y c) f (a + b x, y) x
(y c) f (x, y) (3.3.11)
2
2

(3.3.10)

c+dy

(t b) (s c) dt ds f (t, s) = (x a) (y c) f (a + b x, y)

(3.3.12)

a+bx

c+d
(t a) s
2

dt ds f (t, s)




c+d
c+d
= f (x, c + d y) (x a)
y + f (x, y) (x a)
y
(3.3.13)
2
2
a

44

c+dy

c+dy

a+bx



c+d
s
dt ds f (t, s)
2



a+b
c+d
= f (a + b x, c + d y)
x
y
2
2



c+d
a+b
x
y
+ f (a + b x, y)
2
2



a+b
c+d
+ f (x, c + d y) x
y
2
2



c+d
a+b
y
(3.3.14)
+ f (x, y) x
2
2

a+b
t
2

c+d
(t b) s
2
a+bx

dt ds f (t, s)




c+d
c+d
= f (a + b x, c + d y) (x a)
y + f (a + b x, y) (a x) y
2
2
y

(3.3.15)
d

c+dy

c+dy

(t a) (s d) dt ds f (t, s) = (x a) (y c) f (x, c + d y)

a+bx

(3.3.16)

a+b
t
2

(s d) dt ds f (t, s)




a+b
a+b
x (d y) f (a + b x, c + d y) + x
(c y) f (x, c + d y)
2
2
(3.3.17)

c+dy

(t b) (s d) dt ds f (t, s) = f (a + b x, c + d y) (x a) (d y) (3.3.18)

a+bx

Adding the above equalities, we get


(b a) (d c)
[f (x, y) + f (a + b x, y) + f (x, c + d y) + f (a + b x, c + d y)]
4
Z dZ b

f (t, s) dtds
c
a
Z dZ b
=
K1 (x, t) K2 (y, s) dt ds f (t, s),
c

where,

t a,
K1 (x, t) =
,
t a+b
2

t b,

t [a, x]
t (x, a + b x] ,
t (a + b x, b]

45

and

s c,
K2 (y, s) =
s c+d
,
2

s d,

s [a, y]
s (y, c + d y]
s (c + d y, d]

Now, by Lemma 3.2.2



(b a) (d c)

[f (x, y) + f (a + b x, y) + f (x, c + d y) + f (a + b x, c + d y)]

4

Z dZ b

f (t, s) dtds
c
a
Z d Z b

_


(f ) .
=
K1 (x, t) K2 (y, s) dt ds f (t, s) sup |K1 (x, t) K2 (y, s)|
c

(x,y)Qb,d
a,c

Qb,d
a,c

Since







c+d
a+b
sup |K1 (x, t) K2 (y, s)| = max (x a) ,
x
max (y c) ,
y
y
x
2
2
(x,y)Qb,d
a,c

 



3a + b
d c
3c + d
b a
+ x

+ y
.
=
4
4
4
4

Combining the above identities we get the required results.


Corollary 3.3.14. In the above theorem, choose

1. x = a and y = c, we get


Z dZ b
(b a) (d c)


[f (a, c) + f (b, c) + f (a, d) + f (b, d)]
f (t, s) dtds

4
c
a
(b a) (d c) _

(f ) . (3.3.19)

4
b,d
Qa,c

2. x =

3a+b
4

and y =

3c+d
,
4

we get


 





(b a) (d c)
3a
+
b
3c
+
d
a
+
3b
3c
+
d
3a
+
b
c
+
3d

f
,
+f
,
+f
,

4
4
4
4
4
4
4

 Z d Z b

(b a) (d c) _
a + 3b c + 3d
,

f (t, s) dtds

(f ) . (3.3.20)
+f
4
4
16
c
a
b,d
Qa,c

46

3. x =

a+b
2

and y =

c+d
,
2

we get



 Z

a
+
b
c
+
d
(b a) (d c) f
,


2
2


(b a) (d c) _
f (t, s) dtds

(f ) .
4
b,d
Qa,c

(3.3.21)

A generalization of the Simpson inequality (2.3.23) for mappings of bounded


bivariation, is considered as follows:
Theorem 3.3.15. Let f : Q R be a mapping of bounded bivariation on Q. Then for all
(x, y) Q, we have the inequality

(b a) (d c)

[f (b, d) f (b, c) f (a, d) + f (a, c)]

36
 





a+b
c+d
a+b c+d
(b a) (d c)
+
f
, d + f b,
+ 4f
,
9
2
2
2
2




 Z b Z d

a+b
c+d
f
, c f a,

f (s, t) dtds
2
2
a
c
(b a) (d c) _

(f ) , (3.3.22)
9
Q
where

(f ) denotes the total (double) bivariation of f on Q.

Proof. From Lemma 3.2.1, we have


a+b
2

c+d
2



5c + d
5a + b
s
t
dt ds f (t, s)
6
6
a
c


(b a) (d c)
a+b c+d
=
f
,
9
2
2
 



a+b
c+d
(b a) (d c)
f
, c + f a,

18
2
2
Z a+b Z c+d
2
2
(b a) (d c)
+
f (a, c)
f (s, t) dtds
36
a
c

47

a+b
2



5a + c
c + 5d
s
t
dt ds f (t, s)
c+d
6
6
a
2


(b a) (d c)
a+b c+d
=
f
,
9
2
2
 



(b a) (d c)
c+d
a+b

f a,
f
,d
18
2
2
Z a+b Z d
2
(b a) (d c)
f (s, t) dtds
f (a, d)

c+d
36
a

c+d
2



a + 5b
5c + d
s
t
dt ds f (t, s)
a+b
6
6
c
2


(b a) (d c)
a+b c+d
=
f
,
9
2
2
 



(b a) (d c)
c+d
a+b
+
f b,
f
,c
18
2
2
Z b Z c+d
2
(b a) (d c)

f (b, c)
f (s, t) dtds
a+b
36
c

and
b




c + 5d
a + 5b
t
dt ds f (t, s)
s
a+b
c+d
6
6
2
2


a+b c+d
(b a) (d c)
f
,
=
9
2
2
 



(b a) (d c)
c+d
a+b
+
f b,
+f
,d
18
2
2
Z b Z d
(b a) (d c)
+
f (b, d)
f (s, t) dtds
a+b c+d
36

Adding the above equalities, we get


Z bZ d
K (s, t) dt ds f (s, t)
a

(b a) (d c)
[f (b, d) f (b, c) f (a, d) + f (a, c)]
36
 





(b a) (d c)
a+b
c+d
a+b c+d
+
f
, d + f b,
+ 4f
,
9
2
2
2
2


 Z b Z d

c+d
a+b
, c f a,

f (s, t) dtds
f
2
2
a
c

48

where,

K (s, t) =

for all (t, s) Q.

5a+b
6
5a+b
6
a+5b
6
a+5b
6






5c+d
6

c+5d
6

5c+d
6

c+5d
6






as

a+b
,c
2

as

a+b c+d
, 2
2

a+b
2

< s b, c t

a+b
2

< s b, c+d
< t d.
2

c+d
,
2

< t d,
c+d
,
2

Now, applying Lemma 3.2.3, by letting g = P and = f , we get


Z d Z b

d b
d _
b
_


(b a) (d c) _ _


K (s, t) dt ds f (t, s) sup |K (s, t)|
(f ) =

(f ) ,

9
(x,y)Q
c
a
c a
c a

which completes the proof.


3.4

QUADRATURE RULES FOR RSDOUBLE INTEGRAL

In this section and using Mercer approach to prove Theorem 2.3.13, we introduce the following
quadrature rule for RiemannStieltjes double integral. Looking for a trapezoidal rule for the
double RS-integral, we seek numbers A, B, C and D such that
Zb Zd
a

f (x, y) dy dx g (x, y)
= Af (a, c) + Bf (a, d) + Cf (b, c) + Df (b, d)

is equality for f (x, y) = 1,f (x, y) = x,f (x, y) = y and f (x, y) = xy. That is,
Zb Zd
a

xdy dx g (x, y) = Aa + Ba + Cb + Db,

Zb Zd

ydy dx g (x, y) = Ac + Bd + Cc + Dd,

Zb Zd
a

Zb Zd
a

and

1dy dx g (x, y) = A + B + C + D,

xydy dx g (x, y) = Aac + Bad + Cbc + Dbd.

49

Solving these equations for A, B, C and D, we obtain our double RS-trapezoidal rule:

Zb
Zd
Zb Zd
1
1
g (x, c) dx
g (a, y) dy
f (x, y) dy dx g (x, y) = g (a, c)
ba
dc
a
c
a c

b
d
Z Z
1
+
g (x, y)dydx f (a, c)
(b a) (d c)
a c

Zb
Zd
1
1
g (x, d) dx
g (a, y) dy
g (a, d)
ba
dc
a
c

b
d
Z Z
1
+
g (x, y)dydx f (a, d)
(b a) (d c)
a c

Zb
Zd
1
1
g (b, c)
g (x, c) dx
g (b, y) dy
ba
dc
a
c

b
d
Z Z
1
+
g (x, y)dydx f (b, c)
(b a) (d c)
a c

Zb
Zd
1
1
g (x, d) dx
g (b, y) dy
+ g (b, d)
ba
dc
a
c

b
d
Z Z
1
g (x, y)dydx f (b, d) .
+
(b a) (d c)
a

Looking for a midpoint rule for the double RS-integral, we seek A R and (t, s) [a, b][c, d]
such that

Zb Zd
a

f (x, y) dy dx g (x, y)
= Af (t, s) ,

is equality for f (x, y) = 1, f (x, y) = x, and f (x, y) = y. That is,


Zb Zd
a

1dy dx g (x, y) = A,

Zb Zd
a

xdy dx g (x, y) = tA, and

Zb Zd
a

ydy dx g (x, y) = sA.

Solving these equations for A, t and s, we obtain our double RS-midpoint rule:
Zb Zd
a

f (x, y) dy dx g (x, y)
= [g (b, d) g (a, d) g (b, c) + g (a, c)] f (t, s)

50

where,
b [g (b, d) g (b, c)] a [g (a, d) g (a, c)]

Rb

[g (x, d) g (x, c)] dx

t=

g (b, d) g (a, d) g (b, c) + g (a, c)

and

Rd

d [g (b, d) g (a, d)] c [g (b, c) g (a, c)]

[g (b, y) g (a, y)] dy

s=

g (b, d) g (a, d) g (b, c) + g (a, c)

Theorem 3.4.1. Let g : R2 R be continuous and increasing, let (t, s) [a, b] [c, d] satisfy
Zb

[g (x, d) g (x, c)] dx = (b t) [g (b, d) g (b, c)] + (t a) [g (a, d) g (a, c)] ,

Zd

[g (b, y) g (a, y)] dy = (d s) [g (b, d) g (a, d)] + (s c) [g (b, c) g (a, c)] .

and

If

2f
xy

0 and g has continuous second partial derivatives, then we have


[g (b, d) g (a, d) g (b, c) + g (a, c)] f (t, s)
Zb Zd

f (x, y) dy dx g (x, y)
a

g (a, c)

1
ba

g (a, d)

g (b, c)

Zb

g (x, c) dx

1
dc

where,

(3.4.1)

+ g (b, d)

Zd
c

g (a, y) dy + G f (a, c)

1
ba

Zb

1
ba

Zb
a

Zd

1
ba

Zb

Zd

g (x, d) dx

1
dc

Zd
c

g (x, c) dx

1
dc

g (x, d) dx

1
dc

1
G=
(b a) (d c)

Zb Zd
a

g (a, y) dy + G f (a, d)

g (b, y) dy + G f (b, c)

g (b, y) dy + G f (b, d) .

g (x, y) dydx.

51

Proof. We begin with the right hand inequality. Let


h (u, v) = g (u, v) G1 (u) G2 (v) + G,
where
G1 (u) =

Zd

g (u, y) dy ,

G2 (v) =

Zb

g (x, v) dx,

so that
H (t, s) =

Z t Zs
a

h (u, v) dvdu

satisfies H (a, c) = H (a, d) = H (b, c) = H (b, d) = 0. Therefore,

I =

Zb Zd
a

Zb Zd
a

d

f (u, v) dv du g (u, v) [g (u, v) G1 (u) G2 (v) + G] f (u, v)|ba
c

d

f (u, v) dv du h (u, v) h (u, v) f (u, v)|ba ,
c

therefore using integration by parts twice, then using H (a, c) = H (a, d) = H (b, c) =
H (b, d) = 0, we see that
I =

Zb Zd
a

H (u, v)

2f
(u, v) dvdu.
uv

We claim that H 0. Then by hypothesis

2f
uv

(u, v) 0 and so I 0, which would prove

the right-hand inequality.


To prove the claim, let := (1 , 2 ) be provided by the First Mean Value Theorem
for Double integrals: g(1 , 2 ) = G, where is unique because g is increasing. For (x, y)
[a, 1 ] [c, 2 ] we have
H (x, y) =

Zx Zy
a

h (u, v) dvdu 0,

52

since g is increasing. For (x, y) [1 , b] [2 , d] we have


H (x, y) =

Z1 Z2
a

h (u, v) dvdu +

Zb

h (u, v) dvdu

1 2

Z2

h (u, v) dvdu +

Zx Zy

Zx Zy

h (u, v) dvdu

1 2

Z2

h (u, v) dv +

Zy

h (u, v) dv du

Zb Z2
Zd
Zb
= g (u, v) g (u, y) dy g (x, v) dx + G dv
x

Z2

Zy

Z2
c

Z2
c

Zb

Zb

g (u, y) dy

g (u, v)

c
d
Z

Zb

Zb

g (u, v) du +

Zb
x

Zb Zd

g (u, v) du +

a
b
Z

g (x, v) dx + G du

Zb Zb

g (u, y) dydu +

g (x, v) dxdu

g (x, v) dxdu

Gdudv

Zb
x

Zb
a

Zb
x

Zb Zb

g (u, y) dydu + (b x)

g (x, v) dx + Gdu dv
x

Zb Zd
x

Zb

g (u, y) dydu +

Zb Zd

g (x, v) dx + G dv du

Zd

g (u, v) du +

g (u, y) dy

g (u, y) dy

g (u, v) du +
Zb

Zb
a

Zd

g (u, v)

Zd

b
Z

Zy

Z2

Zy

g (u, v)

Zy

g (u, y) dydu + (b x)

Gdudv

g (x, v) dx (b x) Gdv

Zb
a

g (x, v) dx (b x) Gdv

[ (b x) g (x, v) + (b x) g (x, d) + (b x) g (b, v) (b x) G]dv

Zy

[ (b x) g (x, v) + (b x) g (x, d) + (b x) g (b, v) (b x) G]dv

53

Zd

[ (b x) g (x, v) + (b x) g (x, d) + (b x) g (b, v) (b x) G]dv

Zy

[ (b x) g (x, v) + (b x) g (x, d) + (b x) g (b, v) (b x) G]dv

Zd

[ (b x) g (x, v) + (b x) g (x, d) + (b x) g (b, v) (b x) G]dv

Zd

(b x) g (x, v)dv (b x)

Zd

g (x, d)dv

(b x)

Zd

g (b, v)dv + (d y) (b x) G

(b x) (d y) g (x, y) 2 (b x) (d y) g (b, d) + (d y) (b x) G
= (b x) (d y) [g (x, y) 2g (b, d) + G]
0,
so the claim is proved because g is increasing.
For the left-hand inequality, we begin instead with
(
g (x, d) g (x, c) g (a, d) + g (a, c) , x [a, t]
d
hc (x) =
g (x, d) g (x, c) g (b, d) + g (b, c) , x (t, b]
Here again,
Hcd

(x) =

Zx

hdc (u) du,

clearly

Hcd

(a) = 0. Now, we have


Hcd

(b) =

Zt

[g (x, d) g (x, c) (g (a, d) g (a, c))] dx

Zb

[g (x, d) g (x, c) (g (b, d) g (b, c))] dx

Zb

[g (x, d) g (x, c)] dx (b t) [g (b, d) g (b, c)]

(t a) [g (a, d) g (a, c)] = 0.

54

by our choice of t. Similarly, we have


(
g (b, y) g (a, y) g (b, c) + g (a, c) , y [c, s]
hba (y) =
g (b, y) g (a, y) g (b, d) + g (a, d) , y (s, d]
Here again,
Hab

(y) =

Zy

hba (v) dv,

clearly

Hcd

(c) = 0. Now, we have


Hab (d) =

Zs

[g (b, y) g (a, y) (g (b, c) g (a, c))] dx

Zd

[g (b, y) g (a, y) (g (b, d) g (a, d))] dx

Zd

[g (b, y) g (a, y)] dy (d s) [g (b, d) g (a, d)]

(s c) [g (b, c) g (a, c)] = 0


by our choice of s.
Define h (x, y) = hdc (x) hba (y) on [a, b] [c, d], and therefore
H (x, y) =

Zx Zy
a

Hcd (u) Hab (v) dvdu.

We use integration by parts (twice), and H (a, c) = H (a, d) = H (b, c) = H (b, d) = 0,


to obtain

Zb Zd

f (x, y) dy dx g (x, y) [g (b, d) g (a, d) g (b, c) + g (a, c)] f (t, s)

Zb

Zd

H (x, y)

2f
(x, y) dydx.
xy

We claim that H 0, and then by hypothesis


Zb Zd
a

2f
xy

0 and so

f (x, y) dy dx g (x, y) [g (b, d) g (a, d) g (b, c) + g (a, c)] f (t, s) 0,

55

which would prove the left hand-side inequality.


To prove the claim, that is H 0 iff 0 Hcd (x) Hab (y) which implies either Hcd 0
and Hab 0, or Hcd 0 and Hab 0.
We will show that Hcd 0 and Hab 0 and hence the other case does not hold.
Let x [a, t], since g is increasing, then
Hcd

(x) =

Zx

[g (u, d) g (u, c) g (a, d) + g (a, c)] du.

For x (z, b], we have


Hcd

(x) =

Zt

[g (u, d) g (u, c) g (a, d) + g (a, c)] du

Zx

[g (u, d) g (u, c) g (b, d) + g (b, c)] du

Zx

[g (u, d) g (u, c)] du (x t) [g (b, d) g (b, c)]

(t a) [g (a, d) g (a, c)]


Zb
=
[g (u, d) g (u, c)] du (t a) [g (a, d) g (a, c)]
a

Zb

[g (u, d) g (u, c)] du (x t) [g (b, d) g (b, c)]

= (b t) [g (b, d) g (b, c)]

Zb

[g (u, d) g (u, c)] du

(x t) [g (b, d) g (b, c)]


by our choice of t, which gives
Hcd (x) = (b x) [g (b, d) g (b, c)]

Zb

[g (u, d) g (u, c)] du 0,

again since g is increasing. Similarly one can prove that Hab (y) 0, and therefore our second
claim is proved, which completes the proof of the theorem.

56

The quadrature rule in Theorem 3.4.1 requires knowledge of G. In many applications


this is not an obstacle, however if G is indeed unknown we may use instead its classical
trapezoid rule approximation .
3.5

APPROXIMATIONS VIA BEESACKDARSTPOLLARD INEQUALITY

We start by establishing the BeesackDarstPollard inequality in two real dimensional space


R2 .
b,d
Theorem 3.5.1. Let f, u : Qb,d
a,c R be such that f is of bounded bivariation on Qa,c , u is
RdRb
continuous on Qb,d
a,c and c a f (t, s) dt ds u (t, s) exists. Then, we have
d

f (t, s) dt ds u (t, s) A

inf

(t,s)Qb,d
a,c

 _
f (t, s) + S u; Qb,d
(f )
a,c

(3.5.1)

Qb,d
a,c

where,
A := [u (b, d) u (b, c) u (a, d) + u (a, c)]
and

S u; Qb,d
a,c :=

sup

[u (1 , 2 ) u (1 , 2 ) u (1 , 2 ) + u (1 , 2 )] .

a1 <1 b
c2 <2 d

By replacing u with (u) in (3.5.1), we can also obtain the dual Beesack inequality
Z dZ b
 _
f (t, s) dt ds u (t, s) A inf f (t, s) + s u; Qb,d
(f )
(3.5.2)
a,c
c

(t,s)Qb,d
a,c

Qb,d
a,c

where,

s u; Qb,d
a,c :=

inf

a1 <1 b
c2 <2 d

[u (1 , 2 ) u (1 , 2 ) u (1 , 2 ) + u (1 , 2 )] .

Proof. We observe first that it is enough to prove the inequality in the case inf f = 0, when it
becomes
Z

f (t, s) dt ds u (t, s) S (u; Q)

_
Q

(f ) .

(3.5.3)

57

For the general case can be obtained from (3.5.3) by replacing h in it by f inf f . Clearly we
may also suppose that for some (1 , 2 ), f (1 , 2 ) = 0. Since
Z dZ b
Z d Z 1
f (t, s) dt ds u (t, s) =
f (t, s) ds dt u (t, s)
c
a
c
a
Z d Z 1
+
f (t, s) ds dt [u (t, s)]
c

f (t, s) dt ds u (t, s)
Z 2 Z b
+
f (t, s) dt ds [u (t, s)],
a

and
sup
b1 <1 1



{[u (1 , )] [u (1 , )]} = S u; Qb,d
1 ,c ,

sup

{[u (, 2 )] [u (, 2 )]} = S

d2 <2 2

u; Qb,d
a,2

therefore, we need only to show that, if f 0 and f (b, ) = 0, then (3.5.3) holds.
To observe that let us assume that u (a, ) = 0 = u (, c).

Define (t, s) :=

inf u (1 , 2 ) and

a1 t
c2 s


(t, s) := u (t, s) (t, s) = sup {u (t, s) u (1 , 2 )} S u, Qt,s
a,c .
a1 t
c2 s


Then, is non-increasing, (t, ) = 0 = (, c), and 0 (t, s) S u, Qt,s
a,c . Moreover, we
have

f (t, s) dt ds u (t, s) =

f (t, s) dt ds (t, s) +
c
a
c
Z dZ b
f (t, s) dt ds (t, s)
0+
c
a
Z dZ b
=
(t, s) dt ds f (t, s)
c

kk

f (t, s) dt ds (t, s)

(f )
 _
S u, Qb,d
(f ) ,
a,c

which proves (3.5.1). Similarly, one can obtain (3.5.2) by replacing u with (u) in (3.5.1), and
thus the proof is completely established.

58

As in one variable, a careful examination of the above proof, shows that the continuity
of u was only used at two points of the proof: first, to justify the assumption that f (1 , 2 ) = 0
for some (1 , 2 ) Q in the second reduction step of the proof; finally, to justify the existence
RdRb
of the integral c a f (t, s) dt ds (t, s) (since the continuity of follows from the continuity
of u). In the following we show that the bounds (3.5.1), (3.5.2) remain valid even if u is not
RdRb
continuous on Q, provided only that u is bounded on Q and c a f (t, s) dt ds u (t, s) exists.

We observe first that when u is bounded with u (a, ) = 0 = u (, c), and if (t, s) :=
inf u (1 , 2 ), for all t [a, b] and s [c, d], it follows that, is decreasing on Q. Now, it

a1 t
c2 s

remains to observe that

RdRb
c

f (t, s) dt ds u (t, s) exists. In order to complete the proof of our

assertion, it suffices to rearrange the proof of Theorem 3.5.1 somewhat in order to avoid the
necessity of assuming that f vanishes at some point of Q when m = inf f = 0. As in the proof
of Theorem 3.5.1, the general case of (3.5.1) follows from the case m = 0, so we are to prove
that (3.5.3) holds, when inf f = 0. Given an integer n 1 there exists (1n , 2n ) Q such that
f (1n , 2n ) = 0. Writing
Z dZ b
Z
f (t, s) dt ds u (t, s) =
c

1n

f (t, s) ds dt u (t, s)
Z d Z 1n
+
f (t, s) ds dt [u (t, s)]
c

2n

f (t, s) dt ds u (t, s)
Z 2n Z b
f (t, s) dt ds [u (t, s)],
+
a

and note that f (t, s), f (t, s) and f (t, s) are nonnegative on their respective intervals of
integration and
n

d
_
_1
c

(f (t, s)) =

d _
b
_
c

(f (t, s))

1n

and
2n b
__
d

(f (t, s)) =

d _
b
_
2n

(f (t, s)) ,









n ,d
b,2n
b,d
also, S u (t, s) , Qb,c1
= S u (t, s) , Qb,d
and
S
u
(t,
s)
,
Q
=
S
u
(t,
s)
,
Q
1n ,c
a,d
a,2n ,c .

59

Since u (a, ) = 0 = u (, c), then


Z dZ b
Z dZ b
f (t, s) dt ds u (t, s) =
f (t, s) dt ds (u (t, s) u (a, s) u (t, c)),
c

and defining (t, s) :=


that
Z

2n

1n

inf u (1 , 2 ) as above and (t, s) := u (t, s) (t, s), it follows

a1 t
c2 s

1n ,2n

f (t, s) dt ds u (t, s) f (1n , 2n ) (1n , 2n ) + S u, Qa,c

2 1
_
 _

(f )
c

a
2n 1n

__
n n
n n
1 ,2
1 ,2
f (1n , 2n ) S u, Qa,c
+ S u, Qa,c

(f )
c

1n ,2n

Proceeding in the same way on Qb,d := [b, 1n ] [d, 2n ], we similarly obtain


Z

2n

1n

f (t, s) dt ds u (t, s) f (b, d) (b, d) + S

1n ,2n
u, Qb,d
1n ,2n

f (1n , 2n ) S u (t, s) , Qb,d


+S


1n ,2n
u (t, s) , Qb,d

2n 1n
 _
_

(f )

d b

2n 1n
 _
_
(f )

n n

= S u (t, s) , Qb,d
1n ,2n f (1 , 2 ) +

d _
b
_
2n

1n

(f ) .

It follows that for each n 1,


#
"
Z dZ b
d _
b
_

n n
(f ) ,
f (t, s) dt ds u (t, s) S u, Qb,d
a,c 2f (1 , 2 ) +
c

so that (3.5.1) follows on letting n . Therefore, we just have proved the following fact:

b,d
Theorem 3.5.2. Let f, u : Qb,d
a,c R be such that f is of bounded bivariation on Qa,c , u is
RdRb
bounded on Qb,d
a,c and c a f (t, s) dt ds u (t, s) exists. Then, (3.5.1) and (3.5.2) hold.

In the following, by use of the BeesackDarst-Pollard inequalities (1.1) and (1.3), we


provide other error bounds for the functionals (f, u, m, M ; Q) and ..... .
3.6

APPLICATIONS TO QUADRATURE FORMULA

In this section, we apply some of the above obtained inequalities to give a sample of proposed
quadrature rules for RiemannStieltjes integral. Let us consider the arbitrary division In : a =

60

x0 < x1 < < xn1 < xn = b, and Jm : c = y0 < y1 < < yn1 < yn = d, where
i [xi , xi+1 ] (i = 0, 1, , n 1) and j [yj , yj+1 ] (j = 0, 1, , m 1) are intermediate
points. Consider the Riemann sum
R (f, In , Jm , , ) =

m1
n1
XX

(xi+1 xi ) (yj+1 yj ) f (i , j )

(3.6.1)

j=0 i=0

Using Theorem 3.3.1, we can state the following theorem


Theorem 3.6.1. Let f as in Theorem 3.3.1. Then we have
Z dZ b
f (t, s) dtds = R (f, In , Jm , , ) + E (f, In , Jm , , ) ,
c

(3.6.2)

where R (f, In , Jm , , ) is the Riemann sum defined in (3.6.1) and the remainder the through
the approximation E (f, In , Jm , , ) satisfies the bound
|E (f, In , Jm , , )|
 

 y_
j+1 xi+1
n1 m1
_
X
X  xi+1 xi



x
+
x
y

y
y
+
y
i
i+1
j+1
j
j
j+1


+ i
+ j
(f ).



2
2
2
2
y
x
i=0 j=0
j

(3.6.3)

Proof. Applying Theorem 3.3.1 on the bidimentional interval [xi , xi+1 ] [yj , yj+1 ], we get the
required result.

Similarly, we can give the following estimation for the Simpsons rule for mappings of
bounded variation in two independent variables:
Theorem 3.6.2. Let f as in Theorem 3.3.15. Then we have
Z dZ b
f (t, s) dtds = RS (f, In , Jm , , ) + ES (f, In , Jm , , ) ,
c

(3.6.4)

where RS (f, In , Jm , , ) is the Riemann sum defined such as


RS (f, In , Jm , , )
=

(xi+1 xi ) (yj+1 yj )
[f (xi+1 , yj+1 ) f (xi+1 , yj ) f (xi , yj+1 ) + f (xi , yj )





 36
yj + yj+1
xi + xi+1
xi + xi+1
, yj+1 + 4f xi+1 ,
4f
, yj
+ 4f
2
2
2


 Z yj+1 Z xi+1

xi + xi+1 yj + yj+1
yj + yj+1
+ 16f
,

f (t, s) dtds,
+ 4f xi ,
2
2
2
yj
xi

61

and the remainder the through the approximation ES (f, In , Jm , , ) satisfies the bound
j+1 xi+1
n1 m1
X
X (xi+1 xi ) (yj+1 yj ) y_
_
|ES (f, In , Jm , , )|

(f )
9
y
x
i=0 j=0
j

(3.6.5)

Proof. Applying Theorem 3.3.15 on the bidimentional interval [xi , xi+1 ] [yj , yj+1 ], we get the
required result.

CHAPTER IV

ON AN OSTROWSKI TYPE FUNCTIONAL

4.1

INTRODUCTION

This chapter is devoted to introduce some functionals related with the Ostrowski integral
inequality for mappings of two variables and therefore several representations of the errors
are established. Therefore, inequalities of Trapezoid and Ostrowski type are discussed. Finally,
as application, a cubature formula is given.
4.2

A FUNCTIONAL RELATED TO THE OSTROWSKI INEQUALITY

Theorem 4.2.1. Let f, g : Q R be such that f is (1 , 2 )Holder type mapping, where


H1 , H2 > 0 and 1 , 2 > 0 are given, and g is a mapping of bounded bivariation on Q. Then
we have the inequality
Z



 Z dZ b

g (b, d) g (a, d) g (b, c) + g (a, c)

f (t, s) dtds
f (x, y) dx dy g (x, y)
(b a) (d c)
"
# c a
1
2
_
(b a)
(d c)
+ H2 2 +1

(g) , (4.2.1)
H1 1 +1
2
(1 + 1)
2
(2 + 1)
a,b


Qc,d

where

(g) denotes the total bivariation of g on Q.

63

Proof. As g is of bounded bivariation on Q, by Lemma 3.2.3 we have



Z d Z b
 Z dZ b



g (b, d) g (a, d) g (b, c) + g (a, c)

f (t, s) dtds
f (x, y) dx dy g (x, y)

(b a) (d c)
c
Z ad Z b 
c a

Z dZ b


1
=
f (x, y)
f (t, s) dtds dx dy g (x, y)
(b a) (d c) c a
c
a


Z dZ b
_

1

sup f (x, y)
f (t, s) dtds
(g)
(b a) (d c) c a
(x,y)Qa,b
a,b
Qc,d

c,d

Z

sup
a,b

1
(b a) (d c) (x,y)Q

c,d

dZ

_
b

[f (x, y) f (t, s)] dtds
(g) .

(4.2.2)

Qa,b
c,d

Now, as f is of (1 , 2 )Holder type mapping, then we have


Z d Z b





[f
(x,
y)

f
(t,
s)]
dtds


c
a
Z dZ b

|f (x, y) f (t, s)| dtds


c
a
Z dZ b


H1 |x t|1 + H2 |y s|2 dtds


c
a
Z b
Z d
1
= H1 (d c)
|x t| dt + H2 (b a)
|y s|2 ds
a

(y c)2 +1 + (d y)2 +1
(x a)1 +1 + (b x)1 +1
+ H2 (b a)
= H1 (d c)
1 + 1
2 + 1

and therefore,
Z d Z b




sup
[f (x, y) f (t, s)] dtds
(x,y)Qa,b
c,d

"

(y c)2 +1 + (d y)2 +1
(x a)1 +1 + (b x)1 +1
+ H2 (b a)
sup
H1 (d c)
1 + 1
2 + 1
(x,y)Qa,b
c,d
"
#
(x a)1 +1 + (b x)1 +1
H1 (d c) sup
1 + 1
(x,y)Qa,b
c,d
"
#
(y c)2 +1 + (d y)2 +1
+ H2 (b a) sup
2 + 1
(x,y)Qa,b
c,d

1 +1

= H1 (d c)

(b a)
(d c)2 +1
+
H
(b

a)
.
2
21 +1 (1 + 1)
22 +1 (2 + 1)

64

Using (5.4.2), we get



Z d Z b
 Z dZ b



g (b, d) g (a, d) g (b, c) + g (a, c)

f (t, s) dtds
f (x, y) dx dy g (x, y)

(b a) (d c)
c
a
c
a
Z d Z b
_


1

sup
[f (x, y) f (t, s)] dtds
(g)
(b a) (d c) (x,y)Qa,b c a
a,b
Qc,d

c,d

"

H1

_
(d c)2
(b a)1
+
H

(g) ,
2
21 +1 (1 + 1)
22 +1 (2 + 1)
a,b
Qc,d

as required.
Theorem 4.2.2. Let f, g : Q R be such that f f is (1 , 2 )Holder type mapping, where
H1 , H2 > 0 and 1 , 2 > 0 are given, and g is bimonotonic nondecreasing on Q. Then we have
the inequality
Z



 Z dZ b

g (b, d) g (a, d) g (b, c) + g (a, c)

f (t, s) dtds
f (x, y) dx dy g (x, y)
(b a) (d c)
c
a
a
"
#
2
1
H2 (d c)
H1 (b a)
+
[g (b, d) g (b, c) g (a, d) + g (a, c)] . (4.2.3)

(1 + 1)
(2 + 1)
b

Proof. As g is bimonotonic nondecreasing on Q, by Lemma 3.2.4 we have



Z d Z b
 Z dZ b



g
(b,
d)

g
(a,
d)

g
(b,
c)
+
g
(a,
c)

f
(t,
s)
dtds
f
(x,
y)
d
d
g
(x,
y)

x
y


(b a) (d c)
c
a
c
a
Z d Z b 


Z dZ b


1
=
f (x, y)
f (t, s) dtds dx dy g (x, y)
(b a) (d c) c a
c
a

Z dZ b
Z dZ b


1
f (x, y)
dx dy g (x, y)

f
(t,
s)
dtds


(b a) (d c) c a
c
a

Z d Z b Z d Z b


1

[f (x, y) f (t, s)] dtds dx dy g (x, y)
(4.2.4)
=

(b a) (d c) c a c a

Now, as f is of (1 , 2 )Holder type mapping, then we have


Z d Z b





[f
(x,
y)

f
(t,
s)]
dtds


c

(x a)1 +1 + (b x)1 +1
(y c)2 +1 + (d y)2 +1
H1 (d c)
+ H2 (b a)
1 + 1
2 + 1

65

as we shown in Theorem 4.2.1. Therefore,



Z d Z b
 Z dZ b



g
(b,
d)

g
(a,
d)

g
(b,
c)
+
g
(a,
c)

f (x, y) dx dy g (x, y)

f (t, s) dtds

(b a) (d c)
c
a
c
a

Z d Z b Z d Z b


1

[f (x, y) f (t, s)] dtds dx dy g (x, y)


(b a) (d c) c a c a
Z dZ bh
i
H1

(x a)1 +1 + (b x)1 +1 dx dy g (x, y)


(1 + 1) (b a) c a
Z dZ bh
i
H2
(y c)2 +1 + (d y)2 +1 dx dy g (x, y).
+
(2 + 1) (d c) c a
Using RiemannStieltjes double integral, we may deduce that
Z dZ bh
i
(x a)1 +1 + (b x)1 +1 dx dy g (x, y)
c

= (b a)1 +1 [g (b, d) g (b, c) g (a, d) + g (a, c)]


Z dZ b
h
i

g (x, y) d (x a)1 +1 + (b x)1 +1 dy


c

1 +1

= (b a)

[g (b, d) g (b, c) g (a, d) + g (a, c)]


Z dZ b
(1 + 1)
(x a)1 g (x, y) dxdy
c
a
Z dZ b
+ (1 + 1)
(b x)1 g (x, y) dxdy
c

(4.2.5)

and
Z

Z bh
a

i
(y c)2 +1 + (d y)2 +1 dx dy g (x, y)

= (d c)2 +1 [g (b, d) g (b, c) g (a, d) + g (a, c)]


Z dZ b
h
i

g (x, y) dxd (y c)2 +1 + (d y)2 +1


c

2 +1

= (d c)

[g (b, d) g (b, c) g (a, d) + g (a, c)]


Z dZ b
(2 + 1)
(y c)2 g (x, y) dxdy
c
a
Z dZ b
+ (2 + 1)
(d y)2 g (x, y) dxdy
c

Now, on utilizing the bimonotonicity property of g on Q, we have


Z d
 Z b

Z dZ b
1
1
g (a, y) dy
(x a) dx
(x a) g (x, y) dxdy
c

(d c) g (a, c)

(b a)1 +1
,
1 + 1

(4.2.6)

66

(b x) g (x, y) dxdy

Z

g (b, y) dy

(y c) g (x, y) dxdy

Z

 Z

 Z

(b x) dx

(d c) g (b, d)

 Z

(b a)1 +1
,
1 + 1

g (x, c) dy

(y c) dx

(b a) g (a, c)

(d c)2 +1
,
2 + 1

and
Z

(d y) g (x, y) dxdy

Z

g (x, d) dy

(d y) dx

(d c)2 +1
(b a) g (b, d)
.
2 + 1

Substituting in (4.2.5) and (4.2.6), we get


Z dZ bh
i
1 +1
1 +1
(x a)
+ (b x)
dx dy g (x, y)
c

= (b a)1 +1 [g (b, d) g (b, c) g (a, d) + g (a, c)]


Z dZ b
(1 + 1)
(x a)1 g (x, y) dxdy
c
a
Z dZ b
+ (1 + 1)
(b x)1 g (x, y) dxdy
c

(b a)1 +1 [g (b, d) g (b, c) g (a, d) + g (a, c)]

(4.2.7)

and
Z

Z bh
a

i
(y c)2 +1 + (d y)2 +1 dx dy g (x, y)

= (d c)2 +1 [g (b, d) g (b, c) g (a, d) + g (a, c)]


Z dZ b
(2 + 1)
(y c)2 g (x, y) dxdy
c
a
Z dZ b
+ (2 + 1)
(d y)2 g (x, y) dxdy
c

2 +1

(d c)

[g (b, d) g (b, c) g (a, d) + g (a, c)]

(4.2.8)

67

which gives that



Z d Z b
 Z dZ b



g (b, d) g (a, d) g (b, c) + g (a, c)

f (t, s) dtds
f (x, y) dx dy g (x, y)

(b a) (d c)
c
a
c
a
Z dZ bh
i
H1

(x a)1 +1 + (b x)1 +1 dx dy g (x, y)


(1 + 1) (b a) c a
Z dZ bh
i
H2
(y c)2 +1 + (d y)2 +1 dx dy g (x, y)
+
(2 + 1) (d c) c a
#
"
H1 (b a)1 H2 (d c)2
+
[g (b, d) g (b, c) g (a, d) + g (a, c)] ,

(1 + 1)
(2 + 1)
which completes the proof.
Theorem 4.2.3. Let f, g : Q R be such that f is continuous on Q and f, g are of bounded
bivariation on Q. Then we have the inequality
Z


where


 Z dZ b

g (b, d) g (a, d) g (b, c) + g (a, c)
f (x, y) dx dy g (x, y)

f (t, s) dtds
(b a) (d c)
c
a
_
_

(f ) (g) , (4.2.9)


(f ) denotes the total bivariation of f on Q.

Proof. As g is of bounded bivariation on Q, by Lemma 3.2.3 we have



Z d Z b
 Z dZ b



g (b, d) g (a, d) g (b, c) + g (a, c)

f
(t,
s)
dtds
f
(x,
y)
d
d
g
(x,
y)

x
y


(b

a)
(d

c)
c
Z ad Z b 
c a

Z dZ b


1
=
f (x, y)
f (t, s) dtds dx dy g (x, y)
(b a) (d c) c a
c
a


Z dZ b

_
1

sup f (x, y)
f (t, s) dtds
(g)
(b a) (d c) c a
(x,y)Q
Q
W
W


 


ba
a + b
d c
c + d
Q (f )
Q (g)

sup
+ x

+ y
(b a) (d c) (x,y)Q 2
2
2
2
_
_

(f ) (g) .
Q

Since

 



ba
a + b
d c
c + d


+ x

+ y
= (b a) (d c)
sup
2
2
2
2
(x,y)Q

which completes the proof.

68

4.3

INTEGRAL REPRESENTATION OF ERROR

b,d
For a function g : Qb,d
a,c R, we define g , g : Qa,c R by

g (t, s) : = (t a) [(s c) g (a, c) + (d s) g (a, d)]


+ (b t) [(d s) g (b, d) + (s c) g (b, c)]
and
g (t, s) := g (t, s)

g (t, s)
.
(b a) (d c)

(4.3.1)

We can state the following result


b,d
Theorem 4.3.1. If f, u : Qb,d
a,c R are bounded on Qa,c and such that the RiemannStieltjes
RdRb
RdRb
double integral c a f (t, s) dt ds u (t, s) and the Riemann double integral c a u (t, s) dtds

exist, then
Z dZ b
c

f (t, s) dt ds u (t, s)

 Z dZ b
Z dZ b
f (a, c) f (a, d) f (b, c) + f (b, d)

u (t, s) dtds
u (t, s) dt ds f (t, s)
=
(b a) (d c)
c
a
c
a


(4.3.2)
Proof. By assumptions, we have
Z dZ b
f (t, s) dt ds u (t, s)
c
a
Z dZ b
=
f (t, s) dt ds u (t, s)
c

1
(b a) (d c)

f (t, s) dt ds u (t, s).

Integrating by parts in the RiemannStieltjes double integral (see Lemma 3.2.1), we also have
Z dZ b
f (t, s) dt ds u (t, s)
c

= f (b, d) u (b, d) f (b, c) u (b, c) f (a, d) u (a, d) + f (a, c) u (a, c)


Z dZ b

u (t, s) dt ds f (t, s)
c

= (b a) (d c) [f (a, c) u (b, d) f (a, d) u (b, c) f (b, c) u (a, d) + f (b, d) u (a, c)]


Z dZ b
[f (a, c) f (a, d) f (b, c) + f (b, d)]
u (t, s) dtds,
c

69

which gives that


Z dZ b
f (t, s) dt ds u (t, s)
c
a

 Z dZ b
Z dZ b
f (a, c) f (a, d) f (b, c) + f (b, d)
=
f (t, s) dt ds u (t, s) +

u (t, s) dtds
(b a) (d c)
c
a
c
a
[f (a, c) u (b, d) f (a, d) u (b, c) f (b, c) u (a, d) + f (b, d) u (a, c)]
 Z dZ b
Z dZ b
f (a, c) f (a, d) f (b, c) + f (b, d)

u (t, s) dtds
u (t, s) dt ds f (t, s)
=
(b a) (d c)
c
a
c
a


which completes the proof.


Theorem 4.3.2. Let f, u, f as above. If f is continuous on Qb,d
a,c and u is of bounded bivariation
on Qb,d
a,c . Then we have the inequality
Z d Z b




(t,
s)
d
d
u
(t,
s)
f
t
s


c

max {|f (b, d) f (a, c)| , |f (b, c) f (a, d)| ,


|f (a, d) f (b, d)| , |f (a, c) f (b, c)|}

(u) , (4.3.3)

Qb,d
a,c

where

Qb,d
a,c

(u) denotes the total bivariation of u on Qb,d


a,c .

Proof. As u is of bounded bivariation on Qb,d


a,c , we have
Z d Z b




sup

(t,
s)
d
d
u
(t,
s)
f
t
s


c

(t,s)Qb,d
a,c

|f (t, s)|

(u)

Qb,d
a,c

But

sup

|f (t, s)|

(t,s)Qb,d
a,c

= max {|f (b, d)| , |f (b, c)| , |f (a, d)| , |f (a, c)|}
= max {|f (b, d) f (a, c)| , |f (b, c) f (a, d)| , |f (a, d) f (b, d)| , |f (a, c) f (b, c)|}
which gives
Z d Z b
_



(t,
s)
d
d
u
(t,
s)
(u) max {|f (b, d) f (a, c)| , |f (b, c) f (a, d)| ,
f
t
s


c

Qb,d
a,c

|f (a, d) f (b, d)| , |f (a, c) f (b, c)|}

as required.

70

Remark 4.3.3. With assumptions of Theorem 4.3.2, if we assume in addition

1. f is increasing on Qb,d
a,c , then we have
Z




f (t, s) dt ds u (t, s)

max {|f (b, d) f (a, c)| , |f (a, c) f (b, c)|}

(u) , (4.3.4)

(u) , (4.3.5)

Qb,d
a,c

2. f is decreasing on Qb,d
a,c , then we have
Z




f (t, s) dt ds u (t, s)

max {|f (b, c) f (a, d)| , |f (a, d) f (b, d)|}

Qb,d
a,c

The following result holds:


Theorem 4.3.4. Let f, u : Q R be such that f is (1 , 2 )Holder type mapping, where
H1 , H2 > 0 and 1 , 2 > 0 are given, and u is a mapping of bounded bivariation on Q. Then
we have the inequality
Z


where

#
"
2
1
_

(d

c)
(b

a)
+ H2

(u) ,
f (t, s) dt ds u (t, s) H1
1 + 1
2 + 1
a,b

(4.3.6)

Qc,d

(u) denotes the total bivariation of u on Q.

Proof. The proof may be done by applying Theorems 4.2.1 and 4.3.1 directly.

Now, if we denote the error of approximating the Riemann-Stieltjes double integral



RdRb
f (t, s) dt ds u (t, s) by the representation of error f (t, s) by E f, u; Qb,d
a,c , which is given
c a

71

as follows:
E f, u; Qb,d
a,c

Z dZ b
1
=
(t a) (s c) [f (t, s) f (t, c) f (a, s) + f (a, c)] dt ds u (t, s)
(b a) (d c) c a
Z dZ b
1
+
(t a) (s d) [f (t, d) f (t, s) f (a, d) + f (a, s)] dt ds u (t, s)
(b a) (d c) c a
Z dZ b
1
(t b) (s c) [f (b, s) f (b, c) f (t, s) + f (t, c)] dt ds u (t, s)
+
(b a) (d c) c a
Z dZ b
1
+
(t b) (s d) [f (b, d) f (b, s) f (t, d) + f (t, s)] dt ds u (t, s),
(b a) (d c) c a
then we can state the following result.
Corollary 4.3.5. With the assumptions of Theorem 4.3.2, we have
E f, u; Qb,d
a,c


Z d Z b Z d Z b
1
T (t, r1 , s, r2 ) dr1 dr2 f (r1 , r2 ) dt ds u (t, s)
=
(b a) (d c) c a
c
a

Z d Z b Z d Z b
1
=
T (t, r1 , s, r2 ) dt ds u (t, s) dr1 dr2 f (r1 , r2 ) (4.3.7)
(b a) (d c) c a
c
a
where,

(t a) (s c) ,

(t a) (s d) ,
T (t, r1 , s, r2 ) =

(t b) (s c) ,

(t b) (s d) ,

a r1 t b,
a r1 t b,
a t r1 b,
a t r1 b,

c r2 s d
c s r2 d
c r2 s d
c s r2 d

b,d
Proof. If f is bounded on Qb,d
a,c , then for any (t, s) Qa,c the RiemannStieltjes double integrals
RsRt
RdRt
RsRb
RdRb
d d f (r1 , r2 ), s a dr1 dr2 f (r1 , r2 ), c t dr1 dr2 f (r1 , r2 ) and s t dr1 dr2 f (r1 , r2 ) are
c a r1 r2

exist and

Z sZ

Z sZ

dr1 dr2 f (r1 , r2 ) = f (t, s) f (t, c) f (a, s) + f (a, c) ,

dr1 dr2 f (r1 , r2 ) = f (t, d) f (t, s) f (a, d) + f (a, s) ,

dr1 dr2 f (r1 , r2 ) = f (b, s) f (b, c) f (t, s) + f (t, c) ,

72

and
Z

Therefore,
Z dZ
c

dr1 dr2 f (r1 , r2 ) = f (b, d) f (b, s) f (t, d) + f (t, s) .

T (r1 , r2 ) dr1 dr2 f (r1 , r2 ) = (t a) (s c)

Z sZ

dr1 dr2 f (r1 , r2 )


Z dZ t
+ (t a) (d s)
dr1 dr2 f (r1 , r2 )
s
a
Z sZ b
+ (b t) (s c)
dr1 dr2 f (r1 , r2 )
c
t
Z dZ b
+ (b t) (d s)
dr1 dr2 f (r1 , r2 )

= (b a) (d c) f (t, s) ,
and by (4.3.2) we deduce the first and the second equalities in (4.3.4). The last part follows by
the Fubinitype theorem for the RiemannStieltjes double integral. The details are omitted
Remark 4.3.6. One can obtain another error approximation for the RiemannStieltjes double

integral defined above by considering the dual error of E f, u; Qb,d
a,c i.e., define
F f, u; Qb,d
a,c

Z dZ b
1
(t a) (s c) [u (t, s) u (t, c) u (a, s) + u (a, c)] dt ds f (t, s)
=
(b a) (d c) c a
Z dZ b
1
+
(t a) (d s) [u (t, d) u (t, s) u (a, d) + u (a, s)] dt ds f (t, s)
(b a) (d c) c a
Z dZ b
1
+
(b t) (s c) [u (b, s) u (b, c) u (t, s) + u (t, c)] dt ds f (t, s)
(b a) (d c) c a
Z dZ b
1
+
(b t) (d s) [u (b, d) u (b, s) u (t, d) + u (t, s)] dt ds f (t, s).
(b a) (d c) c a
Therefore, with the assumptions of Theorem 4.3.2, we have
F f, u; Qb,d
a,c


Z d Z b Z d Z b
1
=
T (t, r1 , s, r2 ) dr1 dr2 u (r1 , r2 ) dt ds f (t, s)
(b a) (d c) c a
c
a

Z d Z b Z d Z b
1
T (t, r1 , s, r2 ) dt ds f (t, s) dr1 dr2 u (r1 , r2 ) (4.3.8)
=
(b a) (d c) c a
c
a
where, T (t, r1 , s, r2 ) is defined above.

73

4.4

ERROR BOUNDS

The following result may be stated:


Corollary 4.4.1. Assume that f, u : [a, b] R are bounded.
1. If u (respectively f ) is of bounded bivariation and f (respectively u) is continuous, then
|E (f, u; Q)| (|F (f, u; Q)|)
sup |f (t, s)|
(t,s)Q

(u)

sup |u (t, s)|


(t,s)Q

_
Q

(f )

(4.4.1)

2. If u (respectively f ) is bimonotonic nondecreasing and f (respectively u) is Riemann


integrable on Q, then
|E (f, u; Q)| (|F (f, u; Q)|)
Z
Z dZ b

|f (t, s)| dt ds u (t, s)


c


|u (t, s)| dt ds f (t, s)
(4.4.2)

Proof. The proof follows, using Lemmas 3.2.3 and 3.2.4; respectively.
Theorem 4.4.2. Assume that f, u : [a, b] R are bounded on [a, b] and the RiemannStieltjes
RdRb
(double) integral c a f (t, s) dt ds u (t, s) exists. If < m1 f (t, ) M1 < , for all

t [a, b] and < m2 f (, s) M2 < , for all s [c, d], and u is bimonotonic
nondecreasing on Q. Then, we have
[f (b, d) f (a, c)] [u (b, d) u (a, c)]
[M1 f (a, d) + M2 m] [u (b, d) u (b, c)]
[M2 f (b, c) + M1 m] [u (b, d) u (a, d)]
E (f, u; Q)

(4.4.3)

[f (b, d) f (a, c)] [u (b, d) u (a, c)]


+ [m1 f (a, d) + m2 M ] [u (b, c) u (a, c)]
+ [m2 f (b, c) + m1 M ] [u (a, d) u (a, c)] .
where,
M := max {M1 , M2 },

and m := min {m1 , m2 }.

74

Proof. From the condition < m1 f (t, ) M1 < , for all t [a, b] and < m2
f (, s) M2 < , for all s [c, d]. Setting
M := max {M1 , M2 },

and m := min {m1 , m2 },

then we have m f (t, s) M . Also, we may state that


m f (t, c) f (a, s) + f (a, c) f (t, s) f (t, c) f (a, s) + f (a, c)
M f (t, c) f (a, s) + f (a, c) ,

(4.4.4)

f (t, d) f (a, d) + f (a, s) M f (t, d) f (t, s) f (a, d) + f (a, s)


f (t, d) f (a, d) + f (a, s) m,

(4.4.5)

f (b, s) f (b, c) + f (t, c) M f (b, s) f (b, c) f (t, s) + f (t, c)


f (b, s) f (b, c) + f (t, c) m,

(4.4.6)

and
f (b, d) f (b, s) f (t, d) + m f (b, d) f (b, s) f (t, d) + f (t, s)
f (b, d) f (b, s) f (t, d) + M

(4.4.7)

Therefore, by assumptions the above inequalities become respectively; as follow:


m M1 M2 + f (a, c) f (t, s) f (t, c) f (a, s) + f (a, c)
M m1 m2 + f (a, c) ,

(4.4.8)

m1 f (a, d) + m2 M f (t, d) f (t, s) f (a, d) + f (a, s)


M1 f (a, d) + M2 m,

(4.4.9)

m2 f (b, c) + m1 M f (b, s) f (b, c) f (t, s) + f (t, c)


M2 f (b, c) + M1 m,

(4.4.10)

and
f (b, d) M2 M1 + m f (b, d) f (b, s) f (t, d) + f (t, s)
f (b, d) m2 m1 + M

(4.4.11)

75

If we multiply (4.4.8) by (t a) (s c) 0, (4.4.9) by (t a) (s d) 0, (4.4.10) by


(t b) (s c) 0, and (4.4.11) by (t b) (s d) 0, we obtain
(t a) (s c) [m M1 M2 + f (a, c)]
(t a) (s c) [f (t, s) f (t, c) f (a, s) + f (a, c)]

(4.4.12)

(t a) (s c) [M m1 m2 + f (a, c)] ,

(t a) (s d) [M1 f (a, d) + M2 m]
(t a) (s d) [f (t, d) f (t, s) f (a, d) + f (a, s)]

(4.4.13)

(t a) (s d) [m1 f (a, d) + m2 M ] ,

(t b) (s c) [M2 f (b, c) + M1 m]
(t b) (s c) [f (b, s) f (b, c) f (t, s) + f (t, c)]

(4.4.14)

(t b) (s c) [m2 f (b, c) + m1 M ] ,
and
(t b) (s d) [f (b, d) M2 M1 + m]
(t b) (s d) [f (b, d) f (b, s) f (t, d) + f (t, s)]

(4.4.15)

(t b) (s d) [f (b, d) m2 m1 + M ] .
Summing the above inequalities (4.4.12)(4.4.15), and then integrating over the bimonotonic
nondecreasing function u we get,

76

[m M1 M2 + f (a, c)]

(t a) (s c)dt ds u (t, s)
Z dZ b
+ [M1 f (a, d) + M2 m]
(t a) (s d)dt ds u (t, s)
c
a
Z dZ b
+ [M2 f (b, c) + M1 m]
(t b) (s c)dt ds u (t, s)
c
a
Z dZ b
+ [f (b, d) M2 M1 + m]
(t b) (s d)dt ds u (t, s)
c
a
Z dZ b

(t a) (s c) [f (t, s) f (t, c) f (a, s) + f (a, c)]dt ds u (t, s)


c
a
Z dZ b
+
(t a) (s d) [f (t, d) f (t, s) f (a, d) + f (a, s)]dt ds u (t, s)
c
a
Z dZ b
+
(t b) (s c) [f (b, s) f (b, c) f (t, s) + f (t, c)]dt ds u (t, s)
c
a
Z dZ b
+
(t b) (s d) [f (b, d) f (b, s) f (t, d) + f (t, s)]dt ds u (t, s)
c
a
Z dZ b
[M m1 m2 + f (a, c)]
(t a) (s c)dt ds u (t, s)
c
a
Z dZ b
+ [m1 f (a, d) + m2 M ]
(t a) (s d)dt ds u (t, s)
c
a
Z dZ b
+ [m2 f (b, c) + m1 M ]
(t b) (s c)dt ds u (t, s)
c
a
Z dZ b
+ [f (b, d) m2 m1 + M ]
(t b) (s d)dt ds u (t, s).
c

77

Dividing the above inequalities by (b a) (d c), we deduce



Z dZ b
1
[m M1 M2 + f (a, c)]
(t a) (s c)dt ds u (t, s)
(b a) (d c)
c
a
Z dZ b
+ [M1 f (a, d) + M2 m]
(t a) (s d)dt ds u (t, s)
c
a
Z dZ b
+ [M2 f (b, c) + M1 m]
(t b) (s c)dt ds u (t, s)
c
a

Z dZ b
+ [f (b, d) M2 M1 + m]
(t b) (s d)dt ds u (t, s)
c

E (f, u; Q)

Z dZ b
[M m1 m2 + f (a, c)]
(t a) (s c)dt ds u (t, s)
c
a
Z dZ b
+ [m1 f (a, d) + m2 M ]
(t a) (s d)dt ds u (t, s)
c
a
Z dZ b
+ [m2 f (b, c) + m1 M ]
(t b) (s c)dt ds u (t, s)
c
a

Z dZ b
+ [f (b, d) m2 m1 + M ]
(t b) (s d)dt ds u (t, s) .

(b a) (d c)

However,
Z dZ
c

(t a) (s c) dt ds u (t, s) = (b a) (d c) u (b, d)

(t a) (s d) dt ds u (t, s) = (b a) (d c) u (b, c)

(t b) (s c) dt ds u (t, s) = (b a) (d c) u (a, d)

u (t, s) dtds,

u (t, s) dtds,

u (t, s) dtds,

and

(t b) (s d) dt ds u (t, s) = (b a) (d c) u (a, c)

u (t, s) dtds.

78

Substituting these values in last inequality, we get




Z dZ b
1
u (t, s) dtds
[m M1 M2 + f (a, c)] u (b, d)
(b a) (d c) c a


Z dZ b
1
+ [M1 f (a, d) + M2 m] u (b, c)
u (t, s) dtds
(b a) (d c) c a


Z dZ b
1
+ [M2 f (b, c) + M1 m] u (a, d)
u (t, s) dtds
(b a) (d c) c a


Z dZ b
1
u (t, s) dtds
+ [f (b, d) M2 M1 + m] u (a, c)
(b a) (d c) c a
E (f, u; Q)


Z dZ b
1
[M m1 m2 + f (a, c)] u (b, d)
u (t, s) dtds
(b a) (d c) c a


Z dZ b
1
+ [m1 f (a, d) + m2 M ] u (b, c)
u (t, s) dtds
(b a) (d c) c a


Z dZ b
1
+ [m2 f (b, c) + m1 M ] u (a, d)
u (t, s) dtds
(b a) (d c) c a


Z dZ b
1
+ [f (b, d) m2 m1 + M ] u (a, c)
u (t, s) dtds .
(b a) (d c) c a
Observe that, by the bimonotonicity of u,
1
u (a, c)
(b a) (d c)

u (t, s) dtds u (b, d) ,

and then
[m M1 M2 + f (a, c)] [u (b, d) u (a, c)]
+ [M1 f (a, d) + M2 m] [u (b, c) u (b, d)]
+ [M2 f (b, c) + M1 m] [u (a, d) u (b, d)]
[f (b, d) M2 M1 + m] [u (b, d) u (a, c)]
E (f, u; Q)
[M m1 m2 + f (a, c)] [u (b, d) u (a, c)]
+ [m1 f (a, d) + m2 M ] [u (b, c) u (a, c)]
+ [m2 f (b, c) + m1 M ] [u (a, d) u (a, c)]
+ [f (b, d) m2 m1 + M ] [u (b, d) u (a, c)] .

79

which gives that,


[f (b, d) f (a, c)] [u (b, d) u (a, c)]
[M1 f (a, d) + M2 m] [u (b, d) u (b, c)]
[M2 f (b, c) + M1 m] [u (b, d) u (a, d)]
E (f, u; Q)
[f (b, d) f (a, c)] [u (b, d) u (a, c)]
+ [m1 f (a, d) + m2 M ] [u (b, c) u (a, c)]
+ [m2 f (b, c) + m1 M ] [u (a, d) u (a, c)] .
as required.

A result for F (f, u; Q) is incorporated in the following result:


Corollary 4.4.3. Assume that f, u : [a, b] R are bounded on [a, b] and the RiemannStieltjes
RdRb
(double) integral c a f (t, s) dt ds u (t, s) exists. If < n1 u (t, ) N1 < , for all
t [a, b] and < n2 u (, s) N2 < , for all s [c, d], and f is bimonotonic
nondecreasing on Q. Then, we have
[u (b, d) u (a, c)] [f (b, d) f (a, c)]
[N1 u (a, d) + N2 n] [f (b, d) f (b, c)]
[N2 u (b, c) + N1 n] [f (b, d) f (a, d)]
F (f, u; Q)

(4.4.16)

[u (b, d) u (a, c)] [f (b, d) f (a, c)]


+ [n1 u (a, d) + n2 N ] [f (b, c) f (a, c)]
+ [n2 u (b, c) + n1 N ] [f (a, d) f (a, c)] .
where,
N := max {N1 , N2 },

and n := min {n1 , n2 }.

Proof. The argument is similar to the proof of Theorem 4.4.2 and we shall omit the details

80



b,d
In what follows, we establish some bounds for E f, u; Qb,d
(respectively
F
f,
u;
Q
a,c
a,c )

in the situation where the function from the integrand satisfies the following conditions at the
end points:
|f (t, s) f (a, c)| La (t a)1 + Lc (s c)2 , (t, s) Q.

(4.4.17)

|f (t, s) f (a, d)| La (t a)1 + Ld (d s)2 , (t, s) Q.

(4.4.18)

|f (t, s) f (b, c)| Lb (b t)1 + Lc (s c)2 , (t, s) Q.

(4.4.19)

|f (t, s) f (b, d)| Lb (b t)1 + Ld (d s)2 , (t, s) Q.

(4.4.20)

where, La , Lb , Lc , Ld and 1 , 2 , 1 , 2 > 0 are given.


We notice that if the function f is of (r1 , r2 )-(H1 , H2 )Holder type, then obviously
conditions (4.4.17)(4.4.20) hold with 1 = 1 = r1 , 2 = 2 = r2 and La = Lb = H1 ,
Lc = Ld = H2 . However, 1 , 2 , 1 , 2 can be greater than 1. Indeed, for instance, if we
choose

1. f (t, s) = (t a)1 (s c)2 with 1 , 2 > 0 then f satisfies (4.4.17) with La =


(b a)1 , Lc = (d c)2 .
2. f (t, s) = (t a)1 (d s)2 with 1 , 2 > 0 then f satisfies (4.4.18) with La =
(b a)1 , Ld = (d c)2 .
3. f (t, s) = (b t)1 (s c)2 with 1 , 2 > 0 then f satisfies (4.4.19) with Lb =
(b a)1 , Lc = (d c)2 .
4. f (t, s) = (b t)1 (d s)2 with 1 , 2 > 0 then f satisfies (4.4.20) with Lb =
(b a)1 , Ld = (d c)2 .
Theorem 4.4.4. If f, u : Q R be bounded on Q and such that the RiemannStieltjes double
RdRb
integral c a f (t, s) dt ds (t, s) exists. If f satisfies (4.4.17)(4.4.20) and

81

1. u is of bounded bivariation, then




E f, u; Qb,d

a,c
nh
i h
io _
4 max La (b a)1 + Lb (b a)1 , Lc (d c)2 + Ld (d c)2
(u) .
Q

(4.4.21)
2. u is bimonotonic nondecreasing, then


E f, u; Qb,d

a,c
n
max 2La (b a)1 [u (b, d) u (a, d)] 2Lb (b a)1 [u (b, c) u (a, c)] ,
o
2Lc (d c)2 [u (b, d) u (a, d)] 2Ld (d c)2 [u (b, c) u (a, c)]
(4.4.22)
Proof.

1. If u is of bounded bivariation, then we have




1
E f, u; Qb,d

a,c
(b a) (d c)

 Z d Z b



(t a) (s c) [f (t, s) f (t, c) f (a, s) + f (a, c)] dt ds u (t, s)
Z dc Z ba




+
(t a) (s d) [f (t, d) f (t, s) f (a, d) + f (a, s)] dt ds u (t, s)
Zc d Za b



+
(t b) (s c) [f (b, s) f (b, c) f (t, s) + f (t, c)] dt ds u (t, s)
Zc d Za b




+
(t b) (s d) [f (b, d) f (b, s) f (t, d) + f (t, s)] dt ds u (t, s)
c
a
"
W
(u)
Q
sup |(t a) (s c) [f (t, s) f (t, c) f (a, s) + f (a, c)]|

(b a) (d c) (t,s)Q
+ sup |(t a) (s d) [f (t, d) f (t, s) f (a, d) + f (a, s)]|
(t,s)Q

+ sup |(t b) (s c) [f (b, s) f (b, c) f (t, s) + f (t, c)]|


(t,s)Q

+ sup |(t b) (s d) [f (b, d) f (b, s) f (t, d) + f (t, s)]|


(t,s)Q

82

"
(u)

2 sup |(t a) (s c) Lc (s c)2 |


(b a) (d c)
(t,s)Q



2
+ 2 sup (t a) (s d) Ld (d s)
W

(t,s)Q

+ 2 sup |(t b) (s c) Lc (s c)2 |


(t,s)Q

#




+ 2 sup (t b) (s d) (s d) Ld (d s)2
(t,s)Q

4 Lc (d c)

i _
+ Ld (d c) (u)
2

Similarly, we may observe that


h
i _


1
1
E f, u; Qb,d

4
L
(b

a)
+
L
(b

a)

(u) ,
a
b
a,c
Q

and therefore,



E f, u; Qb,d

a,c
nh
i h
io _
1
1
2
2
4 max La (b a) + Lb (b a) , Lc (d c) + Ld (d c)

(u) .
Q

2. If u is bimonotonic nondecreasing, then we may state that


Z d Z b




(t a) (s c) [f (t, s) f (t, c) f (a, s) + f (a, c)] dt ds u (t, s)

c
a
Z dZ b

(t a) (s c) |f (t, s) f (t, c) f (a, s) + f (a, c)| dt ds u (t, s)


c
a
Z dZ b
2Lc
(t a) (s c)2 +1 dt ds u (t, s)
c
a
Z dZ b
2 +1
2Lc (b a) (d c)
u (b, d) 2 (2 + 1) Lc
(s c)2 u (t, s) dtds,
c

(4.4.23)

and
Z




(t a) (s d) [f (t, d) f (t, s) f (a, d) + f (a, s)] dt ds u (t, s)

a
dZ b

(t a) (d s) |f (t, d) f (t, s) f (a, d) + f (a, s)| dt ds u (t, s)


Z dZ b
2Lc
(t a) (d s)2 +1 dt ds u (t, s)
c
a
Z dZ b
2 +1
2Ld (b a) (d c)
u (b, c) + 2 (2 + 1) Ld
(d s)2 u (t, s) dtds,

(4.4.24)

83

Similarly, we may observe that


Z d Z b




(t b) (s c) [f (b, s) f (b, c) f (t, s) + f (t, c)] dt ds u (t, s)

c
a
Z dZ b
2 +1
2Lc (b a) (d c)
u (a, d) + 2 (2 + 1) Lc
(s c)2 u (t, s) dtds,
c

(4.4.25)

and
d



(t b) (s d) [f (b, d) f (b, s) f (t, d) + f (t, s)] dt ds u (t, s)
c
a
Z dZ b
2 +1
2Ld (b a) (d c)
u (a, c) 2 (2 + 1) Ld
(d s)2 u (t, s) dtds.
Z

(4.4.26)

Adding the above inequalities (4.4.23)(4.4.26), therefore we have




E f, u; Qb,d

a,c

2Lc (d c)2 [u (b, d) u (a, d)] 2Ld (d c)2 [u (b, c) u (a, c)] . (4.4.27)

On the other hand, we may write the above inequalities as follows:


Z d Z b




(t a) (s c) [f (t, s) f (t, c) f (a, s) + f (a, c)] dt ds u (t, s)

c
a
Z dZ b

(t a) (s c) |f (t, s) f (t, c) f (a, s) + f (a, c)| dt ds u (t, s)


c
a
Z dZ b
2La
(t a)1 +1 (s c) dt ds u (t, s)
c
a
Z dZ b
1 +1
2La (b a)
(d c) u (b, d) 2 (1 + 1) La
(t a)2 u (t, s) dtds,
c

(4.4.28)

and
Z




(t a) (s d) [f (t, d) f (t, s) f (a, d) + f (a, s)] dt ds u (t, s)

a
dZ b

(t a) (s d) |f (t, d) f (t, s) f (a, d) + f (a, s)| dt ds u (t, s)


Z dZ b
2Lb
(t a)1 +1 (d s) dt ds u (t, s)
c
a
Z dZ b
1 +1
2Lb (b a)
(d c) u (b, c) 2 (1 + 1) Lb
(b t)1 u (t, s) dtds,

(4.4.29)

84

Similarly, we may observe that


Z d Z b





(t

b)
(s

c)
[f
(b,
s)

f
(b,
c)

f
(t,
s)
+
f
(t,
c)]
d
d
u
(t,
s)
t
s


c
a
Z dZ b
2La (b a)1 +1 (d c) u (a, d) 2 (1 + 1) La
(t a)2 u (t, s) dtds,
c

(4.4.30)

and
Z




(t b) (s d) [f (b, d) f (b, s) f (t, d) + f (t, s)] dt ds u (t, s)
c
a
Z dZ b
1 +1
2Lb (b a)
(d c) u (a, c) 2 (1 + 1) Lb
(b t)1 u (t, s) dtds.
Z

(4.4.31)

Adding the inequalities (4.4.28)(4.4.31), therefore we have




E f, u; Qb,d

a,c

2La (b a)1 [u (b, d) u (a, d)] 2Lb (b a)1 [u (b, c) u (a, c)] . (4.4.32)

Now, using max property, from (4.4.27) and (4.4.32), we get




E f, u; Qb,d

a,c
n
max 2La (b a)1 [u (b, d) u (a, d)] 2Lb (b a)1 [u (b, c) u (a, c)] ,
o
2Lc (d c)2 [u (b, d) u (a, d)] 2Ld (d c)2 [u (b, c) u (a, c)]
(4.4.33)


Now, we may state the following result for F f, u; Qb,d
a,c :

Corollary 4.4.5. With the assumptions of Theorem 4.3.4. If u satisfies (4.4.17)(4.4.20), with
1 = 1 , 2 = 2 , 1 = 1 , 2 = 2 and La = Ha , Lb = Hb , Lc = Hc a, Ld = Hd , then if
1. f is of bounded bivariation, then


F f, u; Qb,d

a,c
nh
i h
io _
4 max Ha (b a)1 + Hb (b a)1 , Hc (d c)2 + Hd (d c)2 (f ) .
Q

(4.4.34)

85

2. f is bimonotonic nondecreasing, then




F f, u; Qb,d

a,c
n
max 2Ha (b a)1 [f (b, d) f (a, d)] 2Hb (b a)1 [f (b, c) f (a, c)] ,
o
2Hc (d c)2 [f (b, d) f (a, d)] 2Hd (d c)2 [f (b, c) f (a, c)]
(4.4.35)

Theorem 4.4.6. If f, u : Q R be bounded on Q and such that the RiemannStieltjes double


RdRb
integral c a f (t, s) dt ds (t, s) exists. If u is of bounded bivariation on Q and
1. f is of bounded bivariation on Q, then we have
_

 _
E f, u; Qb,d

(u)

(f ) .
a,c
Q

(4.4.36)

2. f is bimonotonically nondecreasing on Q, the we have



E f, u; Qb,d

(4.4.37)
a,c
 _

Z dZ b
1
f (r1 , r2 ) dr1 dr2
(u)
f (b, d)
(b a) (d c) c a
r1 ,r2
Qa,c
 _

Z dZ b
1
f (r1 , r2 ) dr1 dr2
(u)
f (b, c)
(b a) (d c) c a
r ,d
1
Qa,r
2


f (a, d)

1
(b a) (d c)


+ f (a, c)

1
(b a) (d c)

f (r1 , r2 ) dr1 dr2

f (r1 , r2 ) dr1 dr2

(u)

(u) .

b,r2
Qr1 ,c

Qb,d
r1 ,r2

Proof. Utilizing the equality between the first and the last terms in (4.3.7) we can write
E f, u; Qb,d
a,c

Z d Z b Z r2 Z r1

1
=
(t a) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )
(b a) (d c) c a
c
a

Z d Z b Z d Z r 2
+
(t a) (s d) dt ds u (t, s) dr1 dr2 f (r1 , r2 )
c
a
r2 a

Z d Z b Z r 2 Z b
+
(t b) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )
c
a
c
r1


Z d Z b Z d Z b
+
(t b) (s d) dt ds u (t, s) dr1 dr2 f (r1 , r2 ) .
c

r2

r1

86

1. If f is of bounded bivariation, then


Z


Z b Z
a

r2

r1




(t a) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )
Z r2 Z r1
_



sup
(t a) (s c) dt ds u (t, s) (f ) ,
(r1 ,r2 )Q

also, since u is of bounded bivariation, then


Z r2 Z r1





(t

a)
(s

c)
d
d
u
(t,
s)
t
s


c

sup (t a) (s c)
(r1 ,r2 )Q

(u)

r1 ,r2
Qa,c

= (b a) (d c)

(u) ,

(4.4.38)

r ,r2

1
Qa,c

which gives that


Z


Z b Z
a

r2

r1




(t a) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

(b a) (d c)

(u)

(u)

(u)

(u)

r1 ,r2
Qa,c

(f ) .

(f ) ,

(f ) ,

(f ) .

Similarly, we may observe that


Z


Z b Z
a

r2

r2




(t a) (s d) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

(b a) (d c)

r ,d

1
Qa,r
2

Z b Z
a

r2

r1




(t b) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

(b a) (d c)

b,r2
Qr1 ,c

and
Z


Z b Z
a

r2

r1




(t b) (s d) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

(b a) (d c)

Qb,d
r1 ,r2

87

Therefore,


E f, u; Qb,d

a,c

_
_
_
_
_

(u) +
(u) +
(u) +
(u)
(f )
r ,r2

r ,d

1
Qa,c

b,r

1
Qa,r
2

(u)

2
Qr1 ,c

Qb,d
r1 ,r2

(f )

2. If f is bimonotonic nondecreasing, then


Z


Z b Z
a

r2

r1




(t a) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

Z d Z b Z r2 Z r1



dr1 dr2 f (r1 , r2 )

(t

a)
(s

c)
d
d
u
(t,
s)
t
s


c

and by (4.4.38), we have


Z r2 Z r1

_



(r1 a) (r2 c)
(t

a)
(s

c)
d
d
u
(t,
s)
(u) ,
t
s


c

r ,r2

1
Qa,c

it follows that
Z


Z b Z
a

r2

r1




(t a) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )
Z dZ b
_

(r1 a) (r2 c) dr1 dr2 f (r1 , r2 )


(u) . (4.4.39)
c

r ,r2

1
Qa,c

Similarly, we may observe that


Z


Z b Z
a

r2

r2




(t a) (s d) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

Z b Z
a

r2

r1

(r1 a) (d r2 ) dr1 dr2 f (r1 , r2 )

(u) , (4.4.40)

(u) , (4.4.41)

r ,d

1
Qa,r
2




(t b) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )
Z

(b r1 ) (r2 c) dr1 dr2 f (r1 , r2 )

b,r

2
Qr1 ,c

88

and
Z


Z b Z
a

r2

r1




(t b) (s d) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

(b r1 ) (d r2 ) dr1 dr2 f (r1 , r2 )

(u) . (4.4.42)

Qb,d
r1 ,r2

Now, using RiemannStieltjes integral, then by (4.4.39)(4.4.42), we get


Z dZ b
(r1 a) (r2 c) dr1 dr2 f (r1 , r2 )
c
a
Z dZ b
= (b a) (d c) f (b, d)
f (r1 , r2 ) dr1 dr2 , (4.4.43)
c

(r1 a) (d r2 ) dr1 dr2 f (r1 , r2 )

= (b a) (d c) f (b, c) +

f (r1 , r2 ) dr1 dr2 , (4.4.44)

(b r1 ) (r2 c) dr1 dr2 f (r1 , r2 )

= (b a) (d c) f (a, d) +
and
Z

f (r1 , r2 ) dr1 dr2 , (4.4.45)

(b r1 ) (d r2 ) dr1 dr2 f (r1 , r2 )

= (b a) (d c) f (a, c)

f (r1 , r2 ) dr1 dr2 . (4.4.46)

Substituting (4.4.43)(4.4.46) in (4.4.39)(4.4.42), respectively; we get


Z d Z b Z r 2 Z r 1






(t

a)
(s

c)
d
d
u
(t,
s)
f
(r
,
r
)
dr
dr
t
s
1
2
1
2


c
a
c
a
 _

Z dZ b
f (r1 , r2 ) dr1 dr2
(b a) (d c) f (b, d)
(u) . (4.4.47)
c

r ,r2

1
Qa,c

Similarly, we may observe that


Z d Z b Z d Z r 2






(t

a)
(s

d)
d
d
u
(t,
s)
d
d
f
(r
,
r
)
t
s
r
r
1
2
1
2


c
a
r2 a
Z d Z b
 _

f (r1 , r2 ) dr1 dr2 (b a) (d c) f (b, c)


(u) , (4.4.48)
c

r ,d

1
Qa,r
2

89

Z b Z
a

r2




(t b) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

Z

r1
dZ b

 _
f (r1 , r2 ) dr1 dr2 (b a) (d c) f (a, d)
(u) , (4.4.49)
b,r

2
Qr1 ,c

and
Z


Z b Z
a

r2

r1




(t b) (s d) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

(b a) (d c) f (a, c)

f (r1 , r2 ) dr1 dr2

(u) . (4.4.50)

Qb,d
r1 ,r2

Adding the inequalities to each other and then dividing by (b a) (d c), we obtain


E f, u; Qb,d

a,c
 _

Z dZ b
1
f (r1 , r2 ) dr1 dr2
(u)
f (b, d)
(b a) (d c) c a
r1 ,r2
Qa,c

 _
Z dZ b
1
f (b, c)
f (r1 , r2 ) dr1 dr2
(u)
(b a) (d c) c a
r ,d
1
Qa,r
2


f (a, d)

+ f (a, c)

1
(b a) (d c)

dZ

1
(b a) (d c)

dZ

f (r1 , r2 ) dr1 dr2

f (r1 , r2 ) dr1 dr2

(u)

(u)

b,r

2
Qr1 ,c

Qb,d
r1 ,r2

Corollary 4.4.7. If f, u : Q R be bounded on Q and such that the RiemannStieltjes double


RdRb
integral c a f (t, s) dt ds (t, s) exists. If f is of bounded bivariation on Q and
1. u is of bounded bivariation on Q, then we have
_

 _
F f, u; Qb,d

(f
)

(u) .
a,c
Q

(4.4.51)

90

2. u is bimonotonically nondecreasing on Q, the we have




F f, u; Qb,d

(4.4.52)
a,c

 _
Z dZ b
1
u (b, d)
u (r1 , r2 ) dr1 dr2
(f )
(b a) (d c) c a
r1 ,r2
Qa,c
 _

Z dZ b
1
u (b, c)
u (r1 , r2 ) dr1 dr2
(f )
(b a) (d c) c a
r ,d
1
Qa,r
2


u (a, d)

+ u (a, c)

1
(b a) (d c)

dZ

1
(b a) (d c)

u (r1 , r2 ) dr1 dr2

u (r1 , r2 ) dr1 dr2

(f )

(f ) .

b,r2
Qr1 ,c

Qb,d
r1 ,r2

Theorem 4.4.8. If f, u : Q R be bounded on Q and such that the RiemannStieltjes double


RdRb
integral c a f (t, s) dt ds (t, s) exists. If u is bimonotonically nondecreasing on Q and
1. f is of bounded bivariation on Q, then we have
_


E f, u; Qb,d

[u
(b,
d)

u
(b,
c)

u
(a,
d)
+
u
(a,
c)]

(f ) .
a,c

(4.4.53)

2. f is bimonotonically nondecreasing on Q, the we have




Z dZ b


1
b,d
E f, u; Qa,c u (b, d)
u (t, s) dtds f (b, d)
(b a) (d c) c a


Z dZ b
1
u (b, c)
u (t, s) dtds f (b, c)
(b a) (d c) c a


Z dZ b
1
u (a, d)
u (t, s) dtds f (a, d)
(b a) (d c) c a


Z dZ b
1
+ u (a, c)
u (t, s) dtds f (a, c)
(b a) (d c) c a
(4.4.54)

91

Proof. Utilizing the equality between the first and the last terms in (4.3.7) we can write

E f, u; Qb,d
a,c
Z d Z b Z r2 Z r1

1
=
(t a) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )
(b a) (d c) c a
c
a

Z d Z b Z d Z r 1
+
(t a) (s d) dt ds u (t, s) dr1 dr2 f (r1 , r2 )
c
a
r2 a

Z d Z b Z r 2 Z b
+
(t b) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )
c
a
c
r1


Z d Z b Z d Z b
+
(t b) (s d) dt ds u (t, s) dr1 dr2 f (r1 , r2 ) .
c

r2

r1

1. If f is of bounded bivariation, then


Z d Z b Z r 2 Z r 1






(t

a)
(s

c)
d
d
u
(t,
s)
d
d
f
(r
,
r
)
t
s
r
r
1
2
1
2


c
a
c
a
Z r2 Z r1
_



sup
(t a) (s c) dt ds u (t, s) (f ) ,
(r1 ,r2 )Q

also, since u is of bimonotonic non-decreasing, then


Z r2 Z r1





(t

a)
(s

c)
d
d
u
(t,
s)
t
s


c
a
Z r2 Z r1

(t a) (s c) dt ds u (t, s)
c
a
Z r2 Z
= (r1 a) (r2 c) u (r1 , r2 )
c

which gives that


Z d Z b Z r 2 Z



c

r1

u (t, s) dtds,

(4.4.55)

r1




(t a) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )
c
a

 _
Z r2 Z r1
sup
(r1 a) (r2 c) u (r1 , r2 )
u (t, s) dtds
(f )

(r1 ,r2 )Q


Z
(b a) (d c) u (b, d)

dZ

 _
u (t, s) dtds (f ) .
Q

Similarly, we may observe that


Z d Z b Z d Z r 1





(t a) (s d) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

c
a
r2 a
Z d Z r1
 _
sup
u (t, s) dtds (r1 a) (d r2 ) u (r1 , r2 )
(f )
(r1 ,r2 )Q

r2

Z

 _
u (t, s) dtds (b a) (d c) u (b, c) (f ) ,
Q

92

Z b Z
a

r2

r1

sup
(r1 ,r2 )Q




(t b) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

Z

r2

r1

 _
u (t, s) dtds (b r1 ) (r2 c) u (r1 , r2 ) (f )
Q

Z

and
Z


Z b Z
a

 _
u (t, s) dtds (b a) (d c) u (a, d) (f ) ,
Q




(t b) (s d) dt ds u (t, s) dr1 dr2 f (r1 , r2 )
r2 r1

 _
Z dZ b
sup
(b r1 ) (d r2 ) u (r1 , r2 )
u (t, s) dtds (f )
Z

(r1 ,r2 )Q

r2

r1


Z
(b a) (d c) u (a, c)

 _
u (t, s) dtds (f ) .
Q

Therefore,
_


E f, u; Qb,d

[u
(b,
d)

u
(b,
c)

u
(a,
d)
+
u
(a,
c)]

(f ) .
a,c
Q

2. If f is bimonotonic nondecreasing, then


Z d Z b Z r 2 Z r 1





(t a) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

c
a
c
a

Z d Z b Z r2 Z r1



dr1 dr2 f (r1 , r2 )

(t

a)
(s

c)
d
d
u
(t,
s)
t
s


c

and by (4.4.55), we have



Z r2 Z r1




(t

a)
(s

c)
d
d
u
(t,
s)
t
s


c
a
Z r2 Z r1

(t a) (s c) dt ds u (t, s)
c
a
Z r2 Z
= (r1 a) (r2 c) u (r1 , r2 )
c

r1

u (t, s) dtds,

(4.4.56)

which gives that



Z d Z b Z r 2 Z r 1





(t

a)
(s

c)
d
d
u
(t,
s)
d
d
f
(r
,
r
)
t
s
r
r
1
2
1
2


c
a
c
a

Z r2 Z r1
Z dZ b
u (t, s) dtds dr1 dr2 f (r1 , r2 ).
(r1 a) (r2 c) u (r1 , r2 )

(4.4.57)

93

Similarly, we may observe that


Z d Z b Z d Z r 1






(t

a)
(s

d)
d
d
u
(t,
s)
d
d
f
(r
,
r
)
t
s
r
r
1
2
1
2


c
a
r2 a

Z d Z b Z d Z r1

u (t, s) dtds (r1 a) (d r2 ) u (r1 , r2 ) dr1 dr2 f (r1 , r2 ),


c

r2

(4.4.58)

Z b Z
a

r2

1
Z b Z




(t b) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

r2

r1


u (t, s) dtds (b r1 ) (r2 c) u (r1 , r2 ) dr1 dr2 f (r1 , r2 ),
(4.4.59)

and

Z d Z b Z d Z b




(t b) (s d) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

r2 r1
c
a

Z dZ b
Z dZ b

(b r1 ) (d r2 ) u (r1 , r2 )
u (t, s) dtds dr1 dr2 f (r1 , r2 ). (4.4.60)
c

r2

r1

Now, using RiemannStieltjes integral, then by (4.4.57), we get


Z


Z b
Z r2 Z r1
(r1 a) (r2 c) u (r1 , r2 )
u (t, s) dtds dr1 dr2 f (r1 , r2 )
a
c
a
Z dZ b
=
(r1 a) (r2 c) u (r1 , r2 ) dr1 dr2 f (r1 , r2 )
c
a

Z d Z b Z r2 Z r1
u (t, s) dtds dr1 dr2 f (r1 , r2 ), (4.4.61)

therefore,
Z

(r1 a) (r2 c) u (r1 , r2 ) dr1 dr2 f (r1 , r2 )


Z dZ b
f (r1 , r2 ) dr1 dr2 u (r1 , r2 ), (4.4.62)
= (b a) (d c) u (b, d) f (b, d)
c

and
Z

Z b Z
a

r2

r1


u (t, s) dtds dr1 dr2 f (r1 , r2 )
c
a
Z d Z b

Z dZ b
f (r1 , r2 ) dr1 dr2 u (r1 , r2 ), (4.4.63)
=
u (t, s) dtds f (b, d)
Z

94

which gives by (4.4.61)(4.4.63), we get


d


Z b
Z r2 Z r1
(r1 a) (r2 c) u (r1 , r2 )
u (t, s) dtds dr1 dr2 f (r1 , r2 )
a
c
a


Z dZ b
= (b a) (d c) u (b, d)
u (t, s) dtds f (b, d) . (4.4.64)
c

Substituting (4.4.64) in (4.4.57)


Z d Z b Z r 2 Z r 1






(t

a)
(s

c)
d
d
u
(t,
s)
d
d
f
(r
,
r
)
t
s
r
r
1
2
1
2


c
a
c
a


Z dZ b
(b a) (d c) u (b, d)
u (t, s) dtds f (b, d) . (4.4.65)
c

Similarly, we may observe that


Z d Z b Z d Z r 1





(t a) (s d) dt ds u (t, s) dr1 dr2 f (r1 , r2 )

r2 a
c
a

Z d Z b
u (t, s) dtds (b a) (d c) u (b, c) f (b, c) , (4.4.66)

and
Z


Z b Z
a

r2




(t b) (s c) dt ds u (t, s) dr1 dr2 f (r1 , r2 )
r1

Z d Z b
u (t, s) dtds (b a) (d c) u (a, d) f (a, d) , (4.4.67)

Z b Z
a

r2




(t b) (s d) dt ds u (t, s) dr1 dr2 f (r1 , r2 )
r1


Z dZ b
u (t, s) dtds f (a, c) . (4.4.68)
(b a) (d c) u (a, c)

Adding the inequalities (4.4.65)(4.4.68) to each other and then dividing by (b a) (d c),
we obtain


Z dZ b


b,d
E f, u; Qa,c (b a) (d c) u (b, d)
u (t, s) dtds f (b, d)
c
a


Z dZ b
u (t, s) dtds f (b, c)
(b a) (d c) u (b, c)
c
a


Z dZ b
u (t, s) dtds f (a, d)
(b a) (d c) u (a, d)
c
a


Z dZ b
u (t, s) dtds f (a, c)
+ (b a) (d c) u (a, c)
c

95

Corollary 4.4.9. If f, u : Q R be bounded on Q and such that the RiemannStieltjes double


RdRb
integral c a f (t, s) dt ds (t, s) exists. If f is bimonotonically nondecreasing on Q and
1. u is of bounded bivariation on Q, then we have
_


F f, u; Qb,d

[f
(b,
d)

f
(b,
c)

f
(a,
d)
+
f
(a,
c)]

(u) .
a,c

(4.4.69)

2. u is bimonotonically nondecreasing on Q, the we have




Z dZ b


1
b,d
F f, u; Qa,c f (b, d)
f (t, s) dtds u (b, d)
(b a) (d c) c a


Z dZ b
1
f (b, c)
f (t, s) dtds u (b, c)
(b a) (d c) c a


Z dZ b
1
f (a, d)
f (t, s) dtds u (a, d)
(b a) (d c) c a


Z dZ b
1
f (t, s) dtds u (a, c)
+ f (a, c)
(b a) (d c) c a
(4.4.70)
4.5

A NUMERICAL QUADRATURE FORMULA FOR THE RSINTEGRAL

In this section, we apply some of the above obtained inequalities to give a sample of proposed
quadrature rules for RiemannStieltjes integral. Let us consider the arbitrary division In : a =
x0 < x1 < < xn1 < xn = b, and Jm : c = y0 < y1 < < yn1 < yn = d, where
i [xi , xi+1 ] (i = 0, 1, , n 1) and j [yj , yj+1 ] (j = 0, 1, , m 1) are intermediate
points. Consider the Riemann sum
A (f, In , Jm , , )
=

n1 m1
X
X g (xi+1 , yj+1 ) g (xi+1 , yj ) g (xi , yj+1 ) + g (xi , yj ) Z
i=0 j=0

(xi+1 xi ) (yj+1 yj )

xi+1

xi

yj+1

f (t, s) dtds

yj

(4.5.1)
Using Theorem 4.2.1, we can state the following theorem

96

Theorem 4.5.1. Let f as in Theorem 4.2.1. Then we have


Z dZ b
f (t, s) dt ds g (t, s) = R (f, In , Jm , , ) + E (f, In , Jm , , ) ,
c

(4.5.2)

where R (f, In , Jm , , ) is the Riemann sum defined in (4.5.1) and the remainder the through
the approximation E (f, In , Jm , , ) satisfies the bound
# b d
"
__
H2 (d c)2
H1 (b a)1
+ 2 +1

(g) .
|E (f, In , Jm , , )| 1 +1
2
(1 + 1) 2
(2 + 1)
a c

(4.5.3)

Proof. Applying Theorem 4.2.1 on the bidimentional interval [xi , xi+1 ] [yj , yj+1 ], we get that
Z
xi+1 Z yj+1

f (t, s) dt ds g (t, s)

xi
yj


Z
Z

g (xi+1 , yj+1 ) g (xi+1 , yj ) g (xi , yj+1 ) + g (xi , yj ) xi+1 yj+1

f (t, s) dtds

(xi+1 xi ) (yj+1 yj )
xi
yj
"
# xi+1 yj+1
_ _
H1 (xi+1 xi )1 H2 (yj+1 yj )2

(g) .
21 +1 (1 + 1)
22 +1 (2 + 1)
x
y
i

Summing over i and j such that 0 i n 1 and 0 j m 1 we get


# i+1 yj+1
"
n1 m1
_
X
X H1 (xi+1 xi )1 H2 (yj+1 yj )2 x_
+

(g)
|E (f, In , Jm , , )|
1 +1 ( + 1)
2 +1 ( + 1)
2
2
1
2
x i yj
i=0 j=0


H1
H2
1
2
1 +1
sup (xi+1 xi ) + 2 +1
sup (yj+1 yj )
2
(1 + 1) 1in1
2
(2 + 1) 1jm1

n1 m1
i+1 yj+1
X
X x_
_
i=0 j=0

xi

(g)

yj

"

# b d
__
H2 (d c)2
H1 (b a)1
= 1 +1
+ 2 +1

(g) ,
2
(1 + 1) 2
(2 + 1)
a c

which gives the result.


Remark 4.5.2. Similarly, we can give several estimations for the error E (f, In , Jm , , ) using
the results the previous sections.

CHAPTER V

SOME RELATED INEQUALITIES

5.1

INTRODUCTION

In this chapter, for mappings of two variables several inequalities of Trapezoid, Gruss
and Ostrowski type are discussed. Namely, in the next two sections, by a Korkine type
identity the Gruss type inequality for integrable functions holds. Inequalities for mappings
of bounded variation, bounded bi-variation, Lipschitzian and bimonotonic are also provided.
In the section after, approximating real functions of two variables which possess n-th partial
derivatives of bounded variation, Lipschitzian and absolutely continuous are proved. In the
section 5.5, Trapezoid-type rules for RS-Double integrals are proved, and therefore, the
classical HermiteHadamard inequality for mappings of two variables is hold. As applications
quadrature rules for RSdouble integral are deduced.
5.2

GRUSS
TYPE INEQUALITIES

We start with the following lemma:


Lemma 5.2.1. Let F1 , F2 , G1 , G2 : Q R be a Riemann-integrable mappings on Q. Then the
following identity holds:
Z dZ b
Z dZ b
F1 (x, y) F2 (x, y) dxdy
G1 (x, y) G2 (x, y) dxdy
c
a
c
a
Z dZ b
Z dZ b
F2 (x, y) G1 (x, y) dxdy
F1 (x, y) G2 (x, y) dxdy

c
a
c
a
Z Z Z Z
1 d d b b
{(F1 (x1 , y1 ) G1 (x2 , y2 ) F1 (x2 , y2 ) G1 (x1 , y1 ))
=
2 c c a a
(F2 (x1 , y1 ) G2 (x2 , y2 ) F2 (x2 , y2 ) G2 (x1 , y1 ))} dx1 dx2 dy1 dy2 ,
provided that the above integrals are exist.

(5.2.1)

98

Proof. Simple calculations yield that


Z Z Z Z
1 d d b b
{(F1 (x1 , y1 ) G1 (x2 , y2 ) F1 (x2 , y2 ) G1 (x1 , y1 ))
2 c c a a
(F2 (x1 , y1 ) G2 (x2 , y2 ) F2 (x2 , y2 ) G2 (x1 , y1 ))} dx1 dx2 dy1 dy2
Z dZ dZ bZ b
1
=
(F1 (x1 , y1 ) G1 (x2 , y2 ) F2 (x1 , y1 ) G2 (x2 , y2 )) dx1 dx2 dy1 dy2
2 c c a a
Z Z Z Z
1 d d b b
(F1 (x1 , y1 ) G1 (x2 , y2 ) F2 (x2 , y2 ) G2 (x1 , y1 )) dx1 dx2 dy1 dy2

2 c c a a
Z Z Z Z
1 d d b b

(F1 (x2 , y2 ) G1 (x1 , y1 ) F2 (x1 , y1 ) G2 (x2 , y2 )) dx1 dx2 dy1 dy2


2 c c a a
Z Z Z Z
1 d d b b
+
(F1 (x2 , y2 ) G1 (x1 , y1 ) F2 (x2 , y2 ) G2 (x1 , y1 )) dx1 dx2 dy1 dy2
2 c c a a
Z Z
Z dZ b
1 d b
=
F1 (x1 , y1 ) F2 (x1 , y1 ) dx1 dy1
G1 (x2 , y2 ) G2 (x2 , y2 ) dx2 dy2
2 c a
c
a
Z Z
Z dZ b
1 d b

F1 (x1 , y1 ) G2 (x1 , y1 ) dx1 dy1


F2 (x2 , y2 ) G1 (x2 , y2 ) dx2 dy2
2 c a
c
a
Z Z
Z dZ b
1 d b
F2 (x1 , y1 ) G1 (x1 , y1 ) dx1 dy1
F1 (x2 , y2 ) G2 (x2 , y2 ) dx2 dy2

2 c a
c
a
Z Z
Z dZ b
1 d b
+
G1 (x1 , y1 ) G2 (x1 , y1 ) dx1 dy1
F1 (x2 , y2 ) F2 (x2 , y2 ) dx2 dy2
2 c a
c
a
Z dZ b
Z dZ b
=
F1 (x, y) F2 (x, y) dxdy
G1 (x, y) G2 (x, y) dxdy
c
a
c
a
Z dZ b
Z dZ b

F1 (x, y) G2 (x, y) dxdy


F2 (x, y) G1 (x, y) dxdy,
c

where, the last equality holds by changing of variables and thus the required result holds.

Therefore, we may deduce the following Korkine type identity for mappings of two
variables.
Corollary 5.2.2. Let F1 , F2 , G1 , G2 as in Lemma 5.2.1. Then the following identity holds:
Z d Z b
 Z d Z b

Z dZ b

f (x, y) g (x, y) dxdy


f (x, y) dxdy
g (x, y) dxdy
c
a
c
a
c
a
Z Z Z Z
1 d d b b
=
(f (x1 , y1 ) f (x2 , y2 )) (g (x1 , y1 ) g (x2 , y2 )) dx1 dx2 dy1 dy2 , (5.2.2)
2 c c a a

where, = (b a) (d c).

Proof. In Lemma 5.2.1, choose G1 (x, y) = G2 (x, y) = 1, F1 (x, y) = f (x, y), F2 (x, y) =
g (x, y), then the required result holds.

99

For two measurable functions f, g : Q R, define the functional, which is known in

the literature as Ceby


csevs functional, by
T (f, g) = M (f g) M (f ) M (g) ,

(5.2.3)

where the integral mean is given by


1
M (f ) =
(b a) (d c)

f (x, y) dxdy.

(5.2.4)

The integrals in (5.2.3) are assumed to exist.

Further, the weighted Ceby


csev functional is defined by
= (f, g; p) = < (f, g; p) < (f ; p) < (g; p) ,
where the weighted integral mean is given by
RdRb
p (x, y) f (x, y) dxdy
< (f ; p) = c Ra d R b
.
p
(x,
y)
dxdy
c a

(5.2.5)

(5.2.6)

Here, we note that,

= (f, g; 1) T (f, g) and < (f ; 1) M (f ) .


Let S< (f ) be an operator defined by
S (f ) (x, y) := f (x, y) M (f ) ,

(5.2.7)

which shifts a function by its integral mean, then the following identity holds. Namely,
T (f, g) = T (S (f ) , g) = T (f, S (g)) = T (S (f ) , S (g))

(5.2.8)

T (f, g) = M (S (f ) g) = M (f S (g)) = M (S (f ) S (g)) ,

(5.2.9)

and so

since M (S (f )) = M (S (g)) = 0.
For the last term in (5.2.8) (or 5.2.9) only one of the functions needs to be shifted by its
integral mean. If the other were to be shifted by any other quantity, the identities would still
hold.

100

Using the above Korkine type identity (5.2.2), we give another proof for the well-known
Gruss inequality:
Theorem 5.2.3. Let f, g : Q R be integrable on Q and satisfy
f (x, y) , g (x, y) , (x, y) Q
Then we have the inequality
|T (f, g)|

1
( ) ( )
4

(5.2.10)

where,
T (f, g)
 Z dZ b
 Z dZ b

Z Z
1
1
1 d b
:=
f (x, y) g (x, y) dxdy
f (x, y) dxdy
g (x, y) dxdy
c a
c a
c a
csevs functional, and := (b a) (d c).
is the Ceby

Proof. First of all, we note that


T (f, g)
1
:=
22

Z bZ
a

(f (x1 , y1 ) f (x2 , y2 )) (g (x1 , y1 ) g (x2 , y2 )) dx1 dx2 dy1 dy2

Applying the Cauchy-Bunyakovsky-Schwarz integral inequality on the Korkine identity (5.2.2),


we get
Z d Z



c

Z bZ
a



(f (x1 , y1 ) f (x2 , y2 )) (g (x1 , y1 ) g (x2 , y2 )) dx1 dx2 dy1 dy2
a
Z d Z d Z b Z b
1/2
2

(f (x1 , y1 ) f (x2 , y2 )) dx1 dx2 dy1 dy2


c

Z

Z bZ
a

(g (x1 , y1 ) g (x2 , y2 )) dx1 dx2 dy1 dy2

1/2

. (5.2.11)

2

, (5.2.12)

Now, observe from (5.2.2) that we have


1
2

Z bZ
a

(f (x1 , y1 ) f (x2 , y2 ))2 dx1 dx2 dy1 dy2

b
2

f (x, y) dxdy

Z

f (x, y) dxdy

101

and a similar identity for g, i.e.,


1
2

Z bZ
a

(g (x1 , y1 ) g (x2 , y2 ))2 dx1 dx2 dy1 dy2

b
2

g (x, y) dxdy

Z

g (x, y) dxdy

2

. (5.2.13)

A simple calculation yields that


1

2

f (x, y) dxdy
f (x, y) dxdy
c
a

 Z dZ b

Z Z
1 d b
1
=
f (x, y) dxdy
f (x, y) dxdy
c a
c a
Z Z
1 d b

(f (x, y) ) ( f (x, y)) dxdy.


c a
c

(5.2.14)

and a similar identity for g


1

2

g (x, y) dxdy
g (x, y) dxdy
c
a

 Z dZ b

Z Z
1 d b
1
=
g (x, y) dxdy
g (x, y) dxdy
c a
c a
Z Z
1 d b

(g (x, y) ) ( g (x, y)) dxdy.


c a
c

(5.2.15)

By the assumption we have (f (x, y) ) ( f (x, y)) 0 and (g (x, y) ) ( g (x, y))
0 for all x, y Q, and so
Z

(f (x, y) ) ( f (x, y)) dxdy 0

(g (x, y) ) ( g (x, y)) dxdy 0

which implies that from (5.2.14)


1

b
2

2

f (x, y) dxdy
f (x, y) dxdy
c
a

 Z dZ b

Z Z
1 d b
1
=
f (x, y) dxdy
f (x, y) dxdy
c a
c a
  Z dZ b
2

Z Z
1
1 d b
f (x, y) dxdy +
f (x, y) dxdy

c a
c a
1
(5.2.16)
= ( )2
4
c

102


A+B 2
.
2

where we have used the fact that AB


1

b
2

g (x, y) dxdy

1
( )2 .
4

A similar argument for g, gives


1

g (x, y) dxdy

2
(5.2.17)

Using the inequality (5.2.11) via (5.2.12), (5.2.13) and the estimations (5.2.16) and (5.2.17), we
get

Z dZ dZ bZ b

1


(f
(x
,
y
)

f
(x
,
y
))
(g
(x
,
y
)

g
(x
,
y
))
dx
dx
dy
dy
1
1
2
2
1
1
2
2
1
2
1
2

2
c
c
a
a
1
( ) ( ) (b a) (d c) ,
4
and then, by (5.2.2), we deduce the desired inequality (5.2.10).
Theorem 5.2.4. Let f, g : Q R be L1 , L2 Lipschitzian mappings on Q, so that
|f (x1 , y1 ) f (x2 , y2 )| L1 k(x1 , y1 ) (x2 , y2 )k
and
|g (x1 , y1 ) g (x2 , y2 )| L2 k(x1 , y1 ) (x2 , y2 )k
for all (x1 , y1 ), (x2 , y2 ) Q. We then have the inequality

L1 L2 
(b a)2 + (d c)2 ,
12
p
where, kk is the usual Euclidean norm, i.e., k(x, y)k = x2 + y 2 .
|T (f, g)|

(5.2.18)

Proof. We have the Korkine identity

1
=
2

Z

 Z

f (x, y) g (x, y) dxdy


f (x, y) dxdy
g (x, y) dxdy
c
a
c
a
Z bZ b
(f (x1 , y1 ) f (x2 , y2 )) (g (x1 , y1 ) g (x2 , y2 )) dx1 dx2 dy1 dy2 , (5.2.19)

a
dZ d
c

where, = (b a) (d c).

103

By assumptions we have
|(f (x1 , y1 ) f (x2 , y2 ))| |(g (x1 , y1 ) g (x2 , y2 ))|
[L1 |(x1 , y1 ) (x2 , y2 )|] [L2 |(x1 , y1 ) (x2 , y2 )|]
q
2
2
2
(x1 x2 ) + (y1 y2 )
= L1 L2


= L1 L2 (x1 x2 )2 + (y1 y2 )2

(5.2.20)

for all (x1 , y1 ), (x2 , y2 ) Q.

Integrating (5.2.20) on Q2 , we get


Z dZ dZ bZ b
(f (x1 , y1 ) f (x2 , y2 )) (g (x1 , y1 ) g (x2 , y2 )) dx1 dx2 dy1 dy2
c
c
a
a
Z dZ dZ bZ b


= L1 L2
(x1 x2 )2 + (y1 y2 )2 dx1 dx2 dy1 dy2
c c aZ baZ b

Z dZ d
2
2
2
2
= L1 L2 (d c)
(x1 x2 ) dx1 dx2 + (b a)
(y1 y2 ) dy1 dy2
a
a
c
c
#
"
2
2
(d

c)
(b

a)
+
= L1 L2 (d c)2 (b a)2
6
6
Using (5.2.19), we get (5.2.18).
Theorem 5.2.5. Let f, g : Q R be satisfy
|f (x1 , y1 ) f (x2 , y2 )| L1 |x1 x2 | |y1 y2 |
and
|g (x1 , y1 ) g (x2 , y2 )| L2 |x1 x2 | |y1 y2 |
for all (x1 , y1 ), (x2 , y2 ) Q. We then have the inequality
L1 L2
(d c)4 (b a)4 ,
36
p
where, kk is the usual Euclidean norm, i.e., k(x, y)k = x2 + y 2 .
|T (f, g)|

(5.2.21)

Proof. We have the Korkine identity


Z d Z b
 Z d Z b

Z dZ b

f (x, y) g (x, y) dxdy


f (x, y) dxdy
g (x, y) dxdy
c
a
c
a
c
a
Z Z Z Z
1 d d b b
(f (x1 , y1 ) f (x2 , y2 )) (g (x1 , y1 ) g (x2 , y2 )) dx1 dx2 dy1 dy2 , (5.2.22)
=
2 c c a a

104

where, = (b a) (d c).
By assumptions we have
|(f (x1 , y1 ) f (x2 , y2 ))| |(g (x1 , y1 ) g (x2 , y2 ))|
[L1 |x1 x2 | |y1 y2 |] [L2 |x1 x2 | |y1 y2 |]
= L1 L2 (x1 x2 )2 (y1 y2 )2

(5.2.23)

for all (x1 , y1 ), (x2 , y2 ) Q.


Integrating (5.2.23) on Q2 , we get
Z dZ dZ bZ b
(f (x1 , y1 ) f (x2 , y2 )) (g (x1 , y1 ) g (x2 , y2 )) dx1 dx2 dy1 dy2
c
c
a
a
Z dZ dZ bZ b


= L1 L2
(x1 x2 )2 (y1 y2 )2 dx1 dx2 dy1 dy2
c c aZ baZ b

Z dZ d
2
2
2
2
= L1 L2 (d c)
(x1 x2 ) dx1 dx2 (b a)
(y1 y2 ) dy1 dy2
a
a
c
c
#
"
(d c)2 (b a)4 (b a)2 (d c)4

= L1 L2
6
6
=

1
L1 L2 (d c)6 (b a)6 .
36

Using (5.2.22), we get (5.2.21).

The following result presents an identity for the Ceby


csev functional that involves a
Riemann-Stieltjes integral and provides a Peano kernel representation:
Lemma 5.2.6. Let f, g : Q R where f is of bounded variation and g is continuous on Q,
then
1
T (f, g) = 2

(t, s) dt ds f (t, s),

(5.2.24)

where,
(t, s) : = (s c) (t a) A (t, b; s, d) (s c) (b t) A (a, t; s, d)
(d s) (t a) A (t, b; c, s) + (d s) (b t) A (a, t; c, s)

(5.2.25)

with
A (a, b; c, d) :=

g (x, y) dxdy

(5.2.26)

105

Proof. From (5.2.24), integrating the RiemannStieltjes integral by parts produces


1
2

(t, s) dt ds f (t, s)

1
[ (b, d) f (b, d) (b, c) f (b, c) (a, d) f (a, d) + (a, c) f (a, c)]
2
Z dZ b
2
1
f (t, s)
(t, s) dtds
2
c a
ts

since (t, s) differentiable. Thus, from (5.2.25) we have (b, d) = (b, c) = (a, d) =
(a, c) = 0, and so
Z dZ b
1
(t, s) dt ds f (t, s)
2 c a
Z dZ b
1
= 2
[(b a) (d c) g (t, s) A (a, b; c, d)] f (t, s) dtds
c a
Z Z
1 d b
=
[g (t, s) M (g)] f (t, s) dtds
c a
= M (f S (g)) = T (f, g) .
from which the result (5.2.25) is obtained on noting identity (5.2.24).
Remark 5.2.7. We remark that (t, s) attain its maximum at

, c+d
.
sup (t, s) = a+b
2
2

a+b c+d
, 2
2

and therefore,

(t,s)Q

Theorem 5.2.8. Let f, g : Q R where f is of bounded variation and g is continuous on Q,


then

W W
sup (t, s) dc ba (f )

(t,s)Q
RdRb
2
|T (f, g)|
L
| (t, s)| dtds,

c a

RdRb
| (t, s)| dt ds f (t, s),
c a

f or f L Lipschitzian
f or f bimonotonic nondecreasing
(5.2.27)

Proof. The first part may be done by Lemma, the second by Lemma and the last part by Lemma.
We shall omit the details.

The following result gives an identity for the weighted Ceby


csev functional that involves
a Riemann-Stieltjes double integral.

106

Theorem 5.2.9. Let f, g, p : Q R where f is of bounded variation and g, p are continuous


RdRb
on Q. Further P (b, d) = c a p (t, s) dtds > 0, then
Z dZ b
1
= (f, g; p) = 2
(t, s) dt ds f (t, s)
(5.2.28)
P (b, d) c a
= (f, g; p) is given in (...),
(t, s) = P (t, s) G (t, s) P (t, s) G (t, s)

(5.2.29)

with
P (t, s) =

Z sZ
c

p (x, y) dxdy,

P (t, s) = P (b, d) P (b, s) P (t, d) + P (t, s) (5.2.30)


and
G (t, s) =

Z sZ
c

p (x, y) g (x, y) dxdy,

G (t, s) = G (b, d) G (b, s) g (t, d) + G (t, s) . (5.2.31)

Proof. The proof may be very closely to proof of Lemma 5.2.6. We shall omit the details.
Theorem 5.2.10. Under the assumptions of Theorem 5.2.9, we have

W W
sup (t, s) dc ba (f )

(t,s)Q
RdRb
P 2 (b, d) |= (f, g; p)|
L c a | (t, s)| dtds,
f or f L Lipschitzian

RdRb
| (t, s)| dt ds f (t, s), f or f bimonotonic nondecreasing
c a

(5.2.32)

Proof. The proof uses results 5.2.6 through 5.2.8 and follows closely the proof in procuring the
bounds in (5.2.32).

In the following, we derive a new inequality of Gruuss type for RiemannStieltjes


double integral. In Chapter 4, we have discussed several properties of the functional
(f, g, Q) :=

f (x, y) dx dy g (x, y)
 Z dZ b

g (b, d) g (a, d) g (b, c) + g (a, c)

f (t, s) dtds (5.2.33)

(b a) (d c)
c
a
a

107

which is of Ostrowskis type for RiemannStieltjes double integral.


The following result holds:
Theorem 5.2.11. Let f, g : Q R, be such that g is LLipschitz on Q and f is
Riemann-integrable on Q and there exist real numbers m, M such that m f (x, y) M ,
for all (x, y) Q. Then we have the following inequality:
| (f, g, Q)|

1
(M m) (b a) (d c) .
2

(5.2.34)

Proof. Using Lemma 3.2.5, we have


Z d Z b

| (f, g, Q)| :=
f (x, y) dx dy g (x, y)
c
a


 Z dZ b

g (b, d) g (a, d) g (b, c) + g (a, c)

f (t, s) dtds
(b a) (d c)
c
a
Z d Z b 


Z dZ b


1

=
f (x, y)
f (t, s) dtds dx dy g (x, y)
(b a) (d c) c a
c
a

Z dZ b
Z dZ b


1
dxdy.
f (x, y)
L
f
(t,
s)
dtds


(b a) (d c) c a
c
a

(5.2.35)

Now, define

1
I :=
(b a) (d c)

Z b
a

1
f (x, y)
(b a) (d c)

2
f (t, s) dtds dxdy,

then, we have
1
I :=
(b a) (d c)

Z b
f 2 (x, y) 2f (x, y)

Z dZ b
1
f (t, s) dtds
(b a) (d c) c a
2 #
Z dZ b
f (t, s) dtds
dxdy

1
=
(b a) (d c)

1
(b a) (d c) c a

Z dZ b
2
f (x, y)dxdy
c

1
(b a) (d c)

2
f (t, s) dtds .

On the other hand, we have




Z dZ b
1
I= M
f (t, s) dtds
(b a) (d c) c a


Z dZ b
1
f (t, s) dtds m

(b a) (d c) c a
Z dZ b
1
[M f (t, s)] [f (t, s) m] dtds.

(b a) (d c) c a

108

As m f (x, y) M , for all (x, y) Q, then


Z dZ b
1
[M f (t, s)] [f (t, s) m] dtds 0
(b a) (d c) c a
which implies that
I

1
M
(b a) (d c)

f (t, s) dtds


Z dZ b
1
f (t, s) dtds m .

(b a) (d c) c a

Using the elementary inequality


(M k) (k m)

1
(M m)2
4

which holds for all m, k, M R, we get


I

1
(M m)2 .
4

(5.2.36)

Using Cauchy-Buniakowski-Schwarzs integral inequality we have




1
I
(b a) (d c)
Now, by (5.2.36), we get
Z


2
Z b
Z dZ b


1
f (x, y)
f (t, s) dtds dxdy

(b a) (d c) c a
a


Z dZ b
Z b


1
dxdy
f (x, y)
f
(t,
s)
dtds


(b

a)
(d

c)
c
a
a

1
(M m) (b a) (d c)
2

and then by (5.2.35) we obtain the desired inequality (5.2.34).


5.3

APPROXIMATING REAL FUNCTIONS OF TWO VARIABLES WHICH POSSESS N -TH DERIVATIVES OF BOUNDED VARIATION

Theorem 5.3.1. Let Q := I J be a closed rectangle on R2 , let a, b I with a < b,


c, d J with c < d and let n be a nonnegative integer. If f : Q R is such that the
n-th partial derivatives Dn f is of bounded variation on Q, then, for any (x, y) Q we have the

109

representation
1
[(b x)(d y)f (a, c) + (b x)(y c)f (a, d)
(b a)(d c)
(y c)(d y)
+(x a)(d y)f (b, c) + (x a)(y c)f (b, d)] +
(b a)(d c)
!
n
h
i
X1
n n

(b x) (x a)nj (y c)j1 Dn f (a, c) + (1)j (d y)j1 Dn f (a, d)


j!
j
j=1
h
io
+ (x a) (b x)nj (1)j (y c)j1 Dn f (b, c) + (d y)j1 Dn f (b, d)
Z dZ b
1
Sn (x, t; y, s) dt ds (Dn f (t, s)) (5.3.1)
+
(b a) (d c) c a
f (x, y) =

where, Dn f (t, s) =

nf
tnj sj

(t, s) and

(x t)n (b x) (y s)n (d y) ,

1 (1)n (t x)n (x a) (y s)n (d y) ,


Sn (x, t; y, s) =
n!
(1)n (x t)n (b x) (s y)n (y c) ,

(t x)n (x a) (s y)n (y c) ,

a t x, c s y
x < t b, c s y
a t x, y < s d
x < t b, y < s d

Proof. We utilize the following Taylors representation formula for functions f : Q R2 R


such that the n-th partial derivatives Dn f are of locally bounded variation on Q,
f (x, y) = Pn (x, y) + Rn (x, y)

(5.3.2)

such that,
n
X
1
Pn (x, y) =
j!
j=0

n
j

(x x0 )nj (y y0 )j Dn f (x0 , y0 )

(5.3.3)

and
1
Rn (x, y) =
n!
where, Dn f (t, s) =

nf
tnj sj

y0

(x t)n (y s)n dt ds (Dn f (t, s))

(5.3.4)

x0

(t, s), and (x, y), (x0 , y0 ) are in Q and the double integral in the

remainder is taken in the Riemann-Stieltjes sense.


Choosing x0 = a, y0 = c and then x0 = b, y0 = d in (5.3.1) we can write that
!
n
X
n
1
(x a)nj (y c)j Dn f (a, c)
f (x, y) =
j!
j
j=0
Z Z
1 y x
+
(x t)n (y s)n dt ds (Dn f (t, s)), (5.3.5)
n! c a

110

f (x, y) =

n
X
(1)j
j=0

f (x, y) =

j!

n
X
(1)j
j=0

n
X
1
f (x, y) =
j!
j=0

j!

n
j

n
j

(x a)nj (d y)j Dn f (a, d)

(1)n+1
+
n!
n
j

(x t)n (s y)n dt ds (Dn f (t, s)), (5.3.6)

(b x)nj (y c)j Dn f (b, c)


(1)n+1
+
n!

(t x)n (y s)n dt ds (Dn f (t, s)) (5.3.7)

(b x)nj (d y)j Dn f (b, d)


1
+
n!

(t x)n (s y)n dt ds (Dn f (t, s)) (5.3.8)

for ny (x, y) Q.
Now, by multiplying (5.3.5) with (b x)(d y), (5.3.6) with (b x)(y c), (5.3.7) with
(x a)(d y), (5.3.8) with (x a)(y c), we get
(b x)(d y)f (x, y) = (b x)(d y)f (a, c)
!
n
X
n
1
(x a)nj (y c)j Dn f (a, c)
+ (b x)(d y)
j!
j
j=1
Z yZ x
1
+ (b x)(d y)
(x t)n (y s)n dt ds (Dn f (t, s)), (5.3.9)
n!
c
a
(b x)(y c)f (x, y) = (b x)(y c)f (a, d)
!
n
X
n
(1)j
+ (b x)(y c)
(x a)nj (d y)j Dn f (a, d)
j!
j
j=0
Z
Z
d
x
(1)n+1
+
(b x)(y c)
(x t)n (s y)n dt ds (Dn f (t, s)), (5.3.10)
n!
y
a
(x a)(d y)f (x, y) = (x a)(d y)f (b, c)
!
n
X
n
(1)j
(b x)nj (y c)j Dn f (b, c)
+ (x a)(d y)
j!
j
j=0
Z
yZ b
(1)n+1
+
(x a)(d y)
(t x)n (y s)n dt ds (Dn f (t, s)) (5.3.11)
n!
c
x

111

(x a)(y c)f (x, y) = (x a)(y c)f (b, d)


!
n
X
n
1
+ (x a)(y c)
(b x)nj (d y)j Dn f (b, d)
j!
j
j=0
Z dZ b
1
+ (x a)(y c)
(t x)n (s y)n dt ds (Dn f (t, s)) (5.3.12)
n!
y
x
respectively, for ny (x, y) Q.
Finally, by adding the equalities (5.3.9)(5.3.12) and dividing the sum with (b a) (d c),
we obtain
f (x, y) =

1
[(b x)(d y)f (a, c) + (b x)(y c)f (a, d)
(b a)(d c)
+(x a)(d y)f (b, c) + (x a)(y c)f (b, d)]
!
n
h
n n
(y c)(d y) X 1
(b x) (x a)nj (y c)j1 Dn f (a, c)
+
(b a)(d c) j=1 j!
j
i
h
+ (1)j (d y)j1 Dn f (a, d) + (x a) (b x)nj (1)j (y c)j1 Dn f (b, c)
io
+ (d y)j1 Dn f (b, d)
+
+
+
+

Z Z
1 (b x)(d y) y x
(x t)n (y s)n dt ds (Dn f (t, s))
n! (b a) (d c) c a
Z Z
(1)n+1 (b x)(y c) d x
(x t)n (s y)n dt ds (Dn f (t, s)),
n! (b a) (d c) y a
Z Z
(1)n+1 (x a)(d y) y b
(t x)n (y s)n dt ds (Dn f (t, s))
n!
(b a) (d c) c x
Z Z
1 (x a)(y c) d b
(t x)n (s y)n dt ds (Dn f (t, s))
n! (b a) (d c) y x

which gives the desired representation (5.3.1).


Remark 5.3.2. The case n = 0 provides the representation
f (x, y) =

1
[(b x)(d y)f (a, c) + (b x)(y c)f (a, d)
(b a)(d c)
+(x a)(d y)f (b, c) + (x a)(y c)f (b, d)]
Z dZ b
1
S0 (x, t; y, s) dt ds (f (t, s)) (5.3.13)
+
(b a) (d c) c a

112

for any x Q, where

(b x) (d y) ,

(x a) (d y) ,
S0 (x, t; y, s) =

(b x) (y c) ,

(x a) (y c) ,

a t x,
x < t b,
a t x,
x < t b,

csy
csy
y<sd
y<sd

and f is of bounded variation on Q.

The above representation provides, as a natural consequence, the possibility to compare


, c+d
) with the values of the function and
the value of a function at the mid point ( a+b
2
2
its derivatives at the end points (the corners of the rectangle generated by the end points).
Therefore, we can state the following corollary:
Corollary 5.3.3. With the assumptions of Theorem 5.3.1 for f and Q, we have the identity


f (a, c) + f (a, d) + f (b, c) + f (b, d)
a+b c+d
,
=
f
2
2
4
!
n

n
1 X1
+ n+2
(b a)nj (d c)j Dn f (a, c) + (1)j Dn f (a, d)
2
j!
j
j=1

+ (1)j Dn f (b, c) + Dn f (b, d)
Z dZ b
1
Mn (t, s) dt ds (Dn f (t, s)) (5.3.14)
+
(b a) (d c) c a
where, Dn f (t, s) =

nf
tnj sj

(t, s) and
a+b
n c+d
n

s
,
a t a+b
, c s c+d

2
2
2
2

n c+d
n
a+b
c+d

s
,
<
t

b,
c

(b a) (d c) (1)n t a+b
2 
2
2
2

Mn (t, s) =
n
n a+b
c+d n
a+b
c+d

4n!
(1)
,
a

t
s

,
<
s

2
2

2
 2

a+b n
c+d n
a+b
c+d
t 2
s 2
,
< t b,
<sd
2
2
On utilizing the following notations

1
[(b x)(d y)f (a, c) + (b x)(y c)f (a, d)
(b a)(d c)
(y c)(d y)
+(x a)(d y)f (b, c) + (x a)(y c)f (b, d)] +
(b a)(d c)
!
n
h
i
X1
n n
(b x) (x a)nj (y c)j1 Dn f (a, c) + (1)j (d y)j1 Dn f (a, d)

j!
j
j=1
h
io
+ (x a) (b x)nj (1)j (y c)j1 Dn f (b, c) + (d y)j1 Dn f (b, d)
(5.3.15)
A (f, Q) =

113

and
1
Bn (f, Q) :=
(b a) (d c)

Sn (x, t; y, s) dt ds (Dn f (t, s))

(5.3.16)

under the assumptions of Theorem 5.3.1, we can approximate the function f utilizing the
polynomials An (f, Q) with the error Bn (f, Q) . In other words, we have
f (x, y) = An (f, Q) + Bn (f, Q)
for any (x, y) Q .
It is then natural to ask for a priori error bounds provided that f belongs to different
classes of functions for which the RiemannStieltjes integral defining the expression in (5.3.16)
exists and can be bounded in absolute value.
Theorem 5.3.4. With the assumptions of Theorem 5.3.1 for f and Q, we have
|Bn (f, Q)|

(x a) (b x) (y c) (d y)
n! (b a) (d c)
"
y x
d _
x
_
_
_
n1
n1
n1
n1
n
(x a)
(y c)

(D f ) (x a)
(d y)

(Dn f )
c

+ (b x)n1 (y c)n1

b
__
c

(Dn f ) + (b x)n1 (d y)n1

d _
b
_
y

(Dn f )

(5.3.17)
(x a) (b x) (y c) (d y)
n! (b a) (d c)






1
x a+b n1 1 (d c) + y c+d n1 Wd Wb (Dn f ) ;

(b

a)
+

c
a
2
2
2
2

h
i1/p h
i1/p

p(n1)
p(n1)
p(n1)
p(n1)

(x

a)
+
(b

x)

(y

c)
+
(d

y)

W W
q W W
q W W
q i1/q

h Wy Wx
q
d
x
y
b
d
b

(
(f
))
+
(f
)
+
(f
)
+
(f
)
c
a
y
a
c
x
y
x

p, q > 1, p1 + 1q = 1;

nW W
o

Wd Wx n
W y Wb
Wd Wb

y
x
n
n
n

max

c
a (D f ) , y
a (D f ) ,
c
x (D f ) ,
y
x (D f )

(x a)n1 + (b x)n1  (y c)n1 + (d y)n1  .

(5.3.18)

114

(b a) (d c)
16n!





1
x a+b n1 1 (d c) + y c+d n1 Wd Wb (Dn f ) ;

(b

a)
+

c
a
2
2
2
2

h
i1/p h
i1/p

p(n1)
p(n1)
p(n1)
p(n1)

(x

a)
+
(b

x)

(y

c)
+
(d

y)

W W
q W W
q W W
q i1/q

h Wy Wx
q
d
x
y
b
d
b

(
(f
))
+
(f
)
+
(f
)
+
(f
)
c
a
y
a
c
x
y
x

p, q > 1, p1 + 1q = 1;

o
nW W

Wd Wx n
W y Wb
Wd Wb

y
x
n
n
n

max
(D
f
)
,
(D
f
)
,
(D
f
)
,
(D
f
)

c
a
y
a
c
x
y
x

(x a)n1 + (b x)n1  (y c)n1 + (d y)n1  .

(5.3.19)

Proof. Using the inequality for the RiemannStieltjes integral of continuous integrands and

115

bounded variation integrators, we have



Z y Z x

1

(x t)n (b x) (y s)n (d y)dt ds (Dn f (t, s))
|Bn (f, Q)| =
n! (b a) (d c) c a
Z dZ x
+
(1)n+1 (x t)n (b x) (s y)n (y c)dt ds (Dn f (t, s))
y

(1)n+1 (t x)n (x a) (y s)n (d y)dt ds (Dn f (t, s))


c
x

Z dZ b

n
n
n
+
(t x) (x a) (y s) (d y)dt ds (D f (t, s))
y
x

 Z y Z x


1
n
n
n

(x t) (b x) (y s) (d y)dt ds (D f (t, s))


n! (b a) (d c)
c
a
Z d Z x



n+1
n
n
n
+
(1)
(x t) (b x) (s y) (y c)dt ds (D f (t, s))
y
a
Z y Z b



n+1
n
n
n
+
(1)
(t x) (x a) (y s) (d y)dt ds (D f (t, s))
Z c d Z x b



n
n
n

+
(t x) (x a) (y s) (d y)dt ds (D f (t, s))
y
x

y x
_
_
1
n
n
max {(x t) (b x) (y s) (d y)}
(Dn f )

n! (b a) (d c) t[a,x]
c a
+

s[c,y]

+ max {(x t) (b x) (s y) (y c)}


t[a,x]
s[y,d]

d _
x
_
y

+ max {(t x) (x a) (y s) (d y)}


t[x,b]
s[c,y]

+ max {(t x)n (x a) (s y)n (y c)}


t[x,b]
s[y,d]

y b
_
_
c

(Dn f )

d _
b
_
y

(Dn f )

"

(Dn f )
y

__
1
(x a)n (b x) (y c)n (d y)
(Dn f )

n! (b a) (d c)
c a
n

+ (x a) (b x) (d y) (y c)

d _
x
_
y

(Dn f )

+ (b x) (x a) (y c) (d y)
+ (b x)n (x a) (d y)n (y c)

b
__

c x
d
b
__
y

(Dn f )
#

(Dn f )

116

"
y x
_
_
(x a) (b x) (y c) (d y)
n1
n1
=
(x a)
(y c)

(Dn f )
n! (b a) (d c)
c a
n1

n1

+ (x a)

(d y)

d _
x
_
y

+ (b x)n1 (d y)n1

(D f ) + (b x)

n1

(y c)

d _
b
_
y

n1

y b
_
_
c

(Dn f )

(Dn f )

and the first inequality in (5.3.17) is proved.


However, by Holders discrete inequality we also have
n1

(x a)

n1

(y c)

y x
_
_
c

(D f ) + (x a)

(d y)

d _
x
_
y

b
__
c

n1

+ (b x)n1 (y c)n1

n1

(Dn f )

(Dn f ) + (b x)n1 (d y)n1

d _
b
_
y

(Dn f )


max (x a)n1 (y c)n1
, (x a)n1 (d y)n1 , (b x)n1 (y c)n1 ,
h

Wy Wx n
Wd Wx n
Wy Wb
n
(b x)n1 (d y)n1
c
a (D f ) +
y
a (D f ) +
c
x (D f )
i
W W
+ dy bx (Dn f ) ;
h

(x a)p(n1) (y c)p(n1) + (x a)p(n1) (d y)p(n1)


p(n1)

p(n1)

p(n1)

p(n1)

i1/p

+ (b x)
(y c)
+ (b x)
(d y)
hW W
W W
q W W
q W W
q i1/q

q
y
x
d
x
y
b
d
b

( c a (f )) +
+
+
;

y
a (f )
c
x (f )
y
x (f )

p, q > 1, p1 + 1q = 1

nW W
o

W d Wx n
W y Wb
W d Wb
y
x

n
n
n

(D
f
)
,
(D
f
)
,
(D
f
)
max
(D
f
)
,

c
a
y
a
c
x
y
x

n1
n1
n1
n1

(x a)
(y c)
+ (x a)
(d y)
+ (b x)n1 (y c)n1

+ (b x)n1 (d y)n1 ;

117



n1
n1
n1
n1 Wd Wb

max
(x

a)
,
(b

x)

max
(y

c)
,
(d

y)
c a (Dn f ) ;

h
i1/p h
i1/p

p(n1)
p(n1)
p(n1)
p(n1)

(x a)
+ (b x)
(y c)
+ (d y)

h




W W
q i1/q

q
q
Wy Wx
Wd Wx
Wy Wb

q
d
b
( c a (f )) +
+
+
y
a (f )
c
x (f )
y
x (f )
=

p, q > 1, p1 + 1q = 1;

o
nW W

W d Wx n
Wy Wb
Wd Wb

x
y
n
n
n

max
(D
f
)
,
(D
f
)
,
(D
f
)
,
(D
f
)

c
a
y
a
c
x
y
x

(x a)n1 + (b x)n1  (y c)n1 + (d y)n1  .








1
x a+b n1 1 (d c) + y c+d n1 Wd Wb (Dn f ) ;

(b

a)
+

c
a
2
2
2
2

h
i1/p h
i1/p

p(n1)
p(n1)
p(n1)
p(n1)

(x

a)
+
(b

x)

(y

c)
+
(d

y)

W W
q W W
q W W
q i1/q

h Wy Wx
q
d
x
y
b
d
b

(
(f
))
+
(f
)
+
(f
)
+
(f
)
c
a
y
a
c
x
y
x
=

p, q > 1, p1 + 1q = 1;

n
o

max Wy Wx (Dn f ) , Wd Wx (Dn f ) , Wy Wb (Dn f ) , Wd Wb (Dn f )

a
c
x
y
x
c
a
y

(x a)n1 + (b x)n1  (y c)n1 + (d y)n1  .

which proves the second inequality in (5.3.18).

The last part is obvious by the elementary inequalities


(x a) (b x)

1
(b a)2 , x [a, b]
4

(y c) (d y)

1
(d c)2 , y [c, d].
4

and

The proof is complete.


Remark 5.3.5. Under the assumptions of Theorem 5.3.4 for f and Q, with the case n = 0
provides the following inequality:
"
y x
d _
x
_
_
_
1
(b x) (d y)
|Bn (f, Q)|
(f ) + (b x) (y c)
(f )
(b a) (d c)
c a
y a
#
y b
d
b
__
__
+ (x a) (d y)
(f ) + (x a) (y c)
(f ) (5.3.20)
c

118

Now, if we denote
f (a, c) + f (a, d) + f (b, c) + f (b, d)
!4
n

n
1 X1
+ n+2
(b a)nj (d c)j Dn f (a, c) + (1)j Dn f (a, d)
2
j!
j
j=1

+ (1)j Dn f (b, c) + Dn f (b, d) (5.3.21)

EM (f ; Q) =

and
1
FM (f ; Q) =
(b a) (d c)

Mn (t, s) dt ds (Dn f (t, s))

(5.3.22)

where,
Mn (t, s) =

(b a) (d c)
4n!
a+b
n c+d
n

s
,

2
2

(1)n+1 t a+b n c+d sn ,


2 
2


n
n+1 a+b
c+d n

t
s

,
(1)

2
2




a+b n
c+d n
t 2
s 2
,

a t a+b
,
2
a+b
< t b,
2
a t a+b
,
2
a+b
< t b,
2

c s c+d
2
c+d
cs 2
c+d
<sd
2
c+d
<sd
2

then we can approximate the value of the function at the midpoint in terms of the values of the
function and its partial derivatives taken at the end points with the error FM (f ; Q). Namely, we
have the representation formula
f

a+b c+d
,
2
2

= EM (f ; Q) + FM (f ; Q)

The absolute value of the error can be bounded as follows:


Corollary 5.3.6. With the assumptions of Theorem 5.3.4 for f , Q and n, we have the inequality
|FM (f ; Q)|

d b
(b a)n (d c)n _ _

(f )
22n+2 n!
c a

(5.3.23)

The following result concerning Lipschitz mappings may be stated as well:


Theorem 5.3.7. Let Q := I J be a closed rectangle on R2 , let a, b I with a < b, c, d J
with c < d and let n be a nonnegative integer. If f : Q R is such that the n-th partial

119

derivatives Dn f is L1 Lipschitz on [a, x] [c, y], L2 Lipschitz on [a, x] [y, d], L3 Lipschitz
on [x, b] [c, y] and L4 Lipschitz on [x, b] [y, d], then we have
|Bn (f, Q)|

(b x) (x a) (d y) (y c)
{(x a)n [L1 (y c)n + L2 (d y)n ]
2
n! (n + 1) (b a) (d c)
+ (b x)n [L3 (y c)n + L4 (d y)n ]}
(b a) (d c)
n
n
n
2 {(x a) [L1 (y c) + L2 (d y) ]
n!16 (n + 1)

+ (b x)n [L3 (y c)n + L4 (d y)n ]} (5.3.24)


Proof. Since Dn f is LLipschitz on Q, then by 3.2.5, we have

Z y Z x

1

|Bn (f, Q)| =
(x t)n (b x) (y s)n (d y)dt ds (Dn f (t, s))
n! (b a) (d c) c a
Z dZ x
+
(1)n+1 (x t)n (b x) (s y)n (y c)dt ds (Dn f (t, s))
y

(1)n+1 (t x)n (x a) (y s)n (d y)dt ds (Dn f (t, s))


c
x

Z dZ b

n
n
n
+
(t x) (x a) (y s) (d y)dt ds (D f (t, s))
y
x

 Z y Z x


1
n
n
n

(x t) (b x) (y s) (d y)dt ds (D f (t, s))



n! (b a) (d c)
c
a
Z d Z x



n+1
n
n
n
+
(1)
(x t) (b x) (s y) (y c)dt ds (D f (t, s))
y
a
Z y Z b



n+1
n
n
n
+
(1)
(t x) (x a) (y s) (d y)dt ds (D f (t, s))
Z c d Z x b



n
n
n

+
(t x) (x a) (y s) (d y)dt ds (D f (t, s))
y
x
 Z yZ x
1
L1
|(x t)n (b x) (y s)n (d y)| dtds

n! (b a) (d c)
c
a
Z dZ x


(1)n+1 (x t)n (b x) (s y)n (y c) dtds
+ L2
+

+ L3
+L4

c
d

x
b



(1)n+1 (t x)n (x a) (y s)n (d y) dtds


|(t x) (x a) (y s) (d y)|dtds
n

120


Z y
Z x
1
n
=
L1 (b x) (d y)
(y s) ds
(x t)n dt
n! (b a) (d c)
c
a
Z d
Z x
+ L2 (b x) (y c)
(s y)n ds
(x t)n dt
y

+ L3 (x a) (d y)
(y s)n ds
(t x)n dt
c
x

Z d
Z b
n
n
+L4 (x a) (d y)
(y s) ds
(t x) dt
y

(b x) (x a) (d y) (y c)
{(x a)n [L1 (y c)n + L2 (d y)n ]
n! (n + 1)2 (b a) (d c)
+ (b x)n [L3 (y c)n + L4 (d y)n ]}

which proves the first inequality in (5.3.24). The last part follows by elementary inequalities
(x a) (b x)

1
(b a)2 , x [a, b],
4

(y c) (d y)

1
(d c)2 , y [c, d],
4

and

the details are omitted. The proof is complete.


Remark 5.3.8. If the function Dn f is LLipschitzian on the whole bidimentional interval
[a, b] [c, d], which, in fact, is a more natural assumption, then we get from (5.3.24) that
|Bn (f, Q)| L

(b x) (x a) (d y) (y c)
[(x a)n + (b x)n ]
2
n! (n + 1) (b a) (d c)
[(y c)n + (d y)n ]

(b a) (d c)
n
n
n
n
2 [(x a) + (b x) ] [(y c) + (d y) ] (5.3.25)
n!16 (n + 1)

Corollary 5.3.9. Let Q := I J be a closed rectangle on R2 , let a, b I with a < b,


c, d J with c < d and let n be a nonnegative integer. If f : Q R is such that the n-th
partial derivatives Dn f is L1 Lipschitz on [a, a+b
][c, c+d
], L2 Lipschitz on [a, a+b
][ c+d
, d],
2
2
2
2
L3 Lipschitz on [ a+b
, b] [c, c+d
] and L4 Lipschitz on [ a+b
, b] [ c+d
, d], then we have
2
2
2
2
(d c)n+1 (b a)n+1
|FM (f, Q)|
[L1 + L2 + L3 + L4 ]
n!22n+4 (n + 1)2

(5.3.26)

In particular, if Dn f is LLipschitzian on Q, then


(d c)n+1 (b a)n+1
|FM (f, Q)| L
n!22n+2 (n + 1)2

(5.3.27)

121

Remark 5.3.10. In Section 4.3, we have considered an integral representation of error


for mappings of two variables. Similarly, one may consider another more accurate error
representation using Taylor representation formula with n = 0 by defining f (t, s) such as
b,d
for a function f : Qb,d
a,c R, we define f , f : Qa,c R by

f (t, s) : = [(b t)(d s)f (a, c) + (b t)(s c)f (a, d)


+(t a)(d s)f (b, c) + (t a)(s c)f (b, d)]
and
f (t, s) := f (t, s)

f (t, s)
.
(b a) (d c)

A generalization of this error may be extended to be for n = k, provided that the k-th partial
derivatives Dk f exist, as we shown above.

Finally, the case when Dn f is absolutely continuous on Q produces the following


estimates for the remainder:
Theorem 5.3.11. Let Q := I J be a closed rectangle on R2 , let a, b I with a < b, c, d J
with c < d and let n be a nonnegative integer. If f : Q R is such that the n-th partial
derivatives Dn f is absolutely continuous on Q, then for any (x, y) Q we have

Z yZ x


1
(b x) (d y)
(x t)n (y s)n Dn+1 f (t, s) dtds
|Bn (f, Q)|
n! (b a) (d c)
c
a
Z dZ x


+ (y c) (b x)
(x t)n (s y)n Dn+1 f (t, s) dtds
y



(t x)n (y s)n Dn+1 f (t, s) dtds
c
x

Z dZ b


n
n n+1

+ (d y) (x a)
(t x) (y s) D f (t, s) dtds

+ (d y) (x a)

122

(xa)n+1 (yc)n+1
(n+1)2

(bx)n+1 (dy)n+1
(n+1)2

kDn+1 f k[a,x][c,y], , Dn+1 f L (Qx,y


a,c );

(b x) (d y)
(xa)n+1/q (yc)n+1/q

kDn+1 f k[a,x][c,y],p , Dn+1 f Lp (Qx,y


a,c ),
(nq+1)1/q
n! (b a) (d c)

1
1

p > 1, p + q = 1;

(x a)n (y c)n kDn+1 f k


n+1
f L1 (Qx,y
a,c );
[a,x][c,y],1 , D

n+1
n+1
(xa)
(dy)

kDn+1 f k[a,x][y,d], , Dn+1 f L (Qx,d


2

a,y );
(n+1)

(b x) (y c)
(xa)n+1/q (dy)n+1/q

+
kDn+1 f k[a,x][y,d],p , Dn+1 f Lp (Qx,d
a,y ),
(nq+1)1/q
n! (b a) (d c)

p > 1, p1 + 1q = 1;

(x a)n (d y)n kDn+1 f k


n+1
f L1 (Qx,d
a,y );
[a,x][c,y],1 , D

n+1
n+1
(bx)
(yc)

kDn+1 f k[x,b][c,y], , Dn+1 f L (Qb,y

x,c );
(n+1)2

(x a) (d y)
(bx)n+1/q (yc)n+1/q

+
kDn+1 f k[x,b][c,y],p , Dn+1 f Lp (Qb,y
x,c ),
(nq+1)1/q
n! (b a) (d c)

p > 1, p1 + 1q = 1;

(b x)n (y c)n kDn+1 f k


n+1
f L1 (Qb,y
x,c );
[x,b][c,y],1 , D

kDn+1 f k[x,b][y,d], , Dn+1 f L (Qb,d


x,y );

(x a) (y c)
(bx)n+1/q (dy)n+1/q

kDn+1 f k[x,b][y,d],p , Dn+1 f Lp (Qb,d


x,y ),
(nq+1)1/q
n! (b a) (d c)

p > 1, p1 + 1q = 1;

(b x)n (d y)n kDn+1 f k


n+1
f L1 (Qb,d
x,y );
[x,b][y,d],1 , D

(5.3.28)

Proof. Since Dn f is absolutely continuous on Q then for any (x, y) Q we have the
representation
1
[(b x)(d y)f (a, c) + (b x)(y c)f (a, d)
(b a)(d c)
(y c)(d y)
+(x a)(d y)f (b, c) + (x a)(y c)f (b, d)] +
(b a)(d c)
!
n
h
i
X1
n n

(b x) (x a)nj (y c)j1 Dn f (a, c) + (1)j (d y)j1 Dn f (a, d)


j!
j
j=1
h
io
+ (x a) (b x)nj (1)j (y c)j1 Dn f (b, c) + (d y)j1 Dn f (b, d)
Z dZ b
1
Sn (x, t; y, s) Dn+1 f (t, s) dtds (5.3.29)
+
(b a) (d c) c a
f (x, y) =

123

where the integral is considered in the Lebesgue sense and the kernel Sn (x, t; y, s) is given in
Theorem 5.3.1.
Utilizing the properties of the Stieltjes integral, we have
Z d Z b



1
n+1


|Bn (f ; Q)| =
S
(x,
t;
y,
s)
D
f
(t,
s)
dtds
n

(b a) (d c) c a

Z y Z x

1
(x t)n (b x) (y s)n (d y) Dn+1 f (t, s) dtds
=
n! (b a) (d c) c a
Z dZ x
+
(1)n+1 (x t)n (b x) (s y)n (y c) Dn+1 f (t, s) dtds
y

(1)n+1 (t x)n (x a) (y s)n (d y) Dn+1 f (t, s) dtds


c
x

Z dZ b

n
n
n+1
+
(t x) (x a) (y s) (d y) D f (t, s) dtds
y
x

 Z y Z x


1
n
n
n+1

(x t) (b x) (y s) (d y) D f (t, s) dtds

n! (b a) (d c)
c
a
Z d Z x



n+1
n
n
n+1
+
(1)
(x t) (b x) (s y) (y c) D f (t, s) dtds
y
a
Z y Z b



n+1
n
n
n+1
+
(1)
(t x) (x a) (y s) (d y) D f (t, s) dtds
Z c d Z x b



n
n
n+1

+
(t x) (x a) (y s) (d y) D f (t, s) dtds
+


Z yZ x


1
(b x) (d y)
(x t)n (y s)n Dn+1 f (t, s) dtds

n! (b a) (d c)
c
a
Z dZ x


+ (y c) (b x)
(x t)n (s y)n Dn+1 f (t, s) dtds
y



(t x)n (y s)n Dn+1 f (t, s) dtds
c
x

Z dZ b


n
n n+1

+ (d y) (x a)
(t x) (y s) D f (t, s) dtds

+ (d y) (x a)

and the first part of the inequality (5.3.28) is proved.

124

Utilizing Holder integral inequality for the Lebesgue integral we have


y



(x t)n (y s)n Dn+1 f (t, s) dtds
(xa)n+1 (yc)n+1
(n+1)2

kDn+1 f k[a,x][c,y], ,

(xa)n+1/q (yc)n+1/q
(nq+1)1/q

kDn+1 f k[a,x][c,y],p ,

(x a)n (y c)n kDn+1 f k


[a,x][c,y],1 ,



(x t)n (s y)n Dn+1 f (t, s) dtds
(xa)n+1 (dy)n+1
(n+1)2

kDn+1 f k[a,x][y,d], ,

(xa)n+1/q (dy)n+1/q
(nq+1)1/q

kDn+1 f k[a,x][y,d],p ,

(x a)n (d y)n kDn+1 f k


[a,x][y,d],1 ,



(t x)n (y s)n Dn+1 f (t, s) dtds
x

(bx)n+1 (yc)n+1

kDn+1 f k[x,b][c,y], ,

(n+1)2

(bx)n+1/q (yc)n+1/q
=
kDn+1 f k[x,b][c,y],p ,
(nq+1)1/q

(b x)n (y c)n kDn+1 f k


,

Dn+1 f Lp [a, x] [c, y] ,

(5.3.30)

p > 1, p1 + 1q = 1;
Dn+1 f L1 [a, x] [c, y] ;

Dn+1 f L [a, x] [y, d] ;


Dn+1 f Lp [a, x] [y, d] ,

(5.3.31)

p > 1, p1 + 1q = 1;
Dn+1 f L1 [a, x] [y, d]

[x,b][c,y],1

Dn+1 f L [a, x] [c, y] ;



(t x)n (y s)n Dn+1 f (t, s) dtds
(bx)n+1 (dy)n+1
(n+1)2

kDn+1 f k[x,b][y,d], ,

(bx)n+1/q (dy)n+1/q
(nq+1)1/q

kDn+1 f k[x,b][y,d],p ,

(b x)n (d y)n kDn+1 f k


[x,b][y,d],1 ,

Dn+1 f L [x, b] [c, y] ;


Dn+1 f Lp [x, b] [c, y] ,

(5.3.32)

p > 1, p1 + 1q = 1;
Dn+1 f L1 [x, b] [c, y] ;

Dn+1 f L [x, b] [y, d] ;


Dn+1 f Lp [x, b] [y, d] ,
p > 1, p1 + 1q = 1;
Dn+1 f L1 [x, b] [y, d]

On making use of (5.3.30)(5.3.33) we deduce the second part of (5.3.28).

(5.3.33)

125

5.4

TRAPEZOID-TYPE RULES FOR RS-DOUBLE INTEGRALS

Let f, u : Q R be such that

RdRb
c

f (t, s) dt ds u (t, s) exists. Define

T (f, u; Q)
:=

f (a, c) + f (a, d) + f (b, c) + f (b, d)


[u (b, d) u (b, c) u (a, d) + u (a, c)]
4
Z dZ b

f (t, s) dt ds u (t, s).


c

For integrators of bounded variation, the following result holds:


Theorem 5.4.1. Let f, u : Q R be such that f is (1 , 2 )-(H1 , H2 )Holder type mapping,
where H1 , H2 > 0 and 1 , 2 > 0 are given, and u is a mapping of bounded variation on Q.
Then we have the inequality


 
 
dc 2 _
ba 1
+ H2
(u) .
|T (f, u; Q)| H1
2
2
Q

(5.4.1)

Proof. Using the inequality for the RiemannStieltjes integral of continuous integrands and
bounded variation integrators, we have

Z d Z b 



f (a, c) + f (a, d) + f (b, c) + f (b, d)

f (t, s) dt ds u (t, s)
|T (f, u; Q)| =
4
c
a
_
f (a, c) + f (a, d) + f (b, c) + f (b, d)


sup
(u) .
(5.4.2)
f (t, s)
4
Q
As f is of (1 , 2 )-(H1 , H2 )Holder type mapping, then we have



f (a, c) + f (a, d) + f (b, c) + f (b, d)

f
(t,
s)


4
1
[|f (a, c) f (t, s)| + |f (a, d) f (t, s)| + |f (b, c) f (t, s)| + |f (b, d) f (t, s)|]
4
1
{[H1 (t a)1 + H2 (s c)2 ] + [H1 (t a)1 + H2 (d s)2 ]
4
+ [H1 (b t)1 + H2 (s c)2 ] + [H1 (b t)1 + H2 (d s)2 ]}
=

H1
H2
[(t a)1 + (b t)1 ] +
[(s c)2 + (d s)2 ] ,
2
2

(5.4.3)

for any t [a, b] and s [c, d].


Now, consider the mapping 1 : [a, b] R, given by 1 (t) := (t a)1 + (b t)1 ,
t [a, b], 1 (0, 1]. Then, 10 (t) := 1 (t a)1 1 1 (b t)1 1 iff t =

a+b
,
2

10 (t) > 0 on

126

(a, a+b
) and 10 (t) < 0 on ( a+b
, b), which shows that its maximum is realized at t =
2
2


a+b
max {1 (t)} = 1
= 211 (b a)1 .
t[a,b]
2

a+b
2

and

Similarly, if we consider the mapping 2 : [a, b] R, given by 2 (s) := (s c)2 + (d s)2 ,


s [c, d], 2 (0, 1]. Therefore, its maximum is realized at s = c+d
and
2


c+d
max {2 (s)} = 2
= 212 (d c)2 .
s[c,d]
2
Consequently, by (5.4.3), we have



f (a, c) + f (a, d) + f (b, c) + f (b, d)
f (t, s)
sup
4




ba 1
dc 2
H1
+ H2
. (5.4.4)
2
2

Using (5.4.2) we obtain the desired inequality (5.4.1).


Remark 5.4.2. We notice that, if f is (L1 , L2 )Lipschitzian on Q, then (5.4.1) becomes
_
1
(u) .
(5.4.5)
|T (f, u; Q)| [L1 (b a) + L2 (d c)]
2
Q
Corollary 5.4.3. If we assume that g : Q R, is Lebesgue integrable on Q, then u (x, y) :=
RyRx
RdRb
g
(t,
s)
dtds
is
differentiable
almost
everywhere,
u
(b,
d)
=
g (t, s) dtds, u (b, c) =
c a
c a
RdRb
W
u (a, d) = u (a, c) = 0 and Q (u) = c a |g (t, s)| dtds. Consequently, by (5.4.1) we obtain


Z dZ b
Z dZ b

f (a, c) + f (a, d) + f (b, c) + f (b, d)

g
(t,
s)
dtds

f
(t,
s)
g
(t,
s)
dtds


4
c
a
c
a
 
1

 2  Z d Z b
ba
dc
H1
+ H2

|g (t, s)| dtds. (5.4.6)


2
2
c
a
From (5.4.6) we get a weighted version of the trapezoid inequality,


f (a, c) + f (a, d) + f (b, c) + f (b, d) R d R b f (t, s) g (t, s) dtds


c Ra d R b




4
g (t, s) dtds
c a
 
 1

 
ba
dc 2
H1
+ H2
, (5.4.7)
2
2
RdRb
provided that g(t, s) > 0, for almost every (t, s) Q and c a g (t, s) dtds 6= 0
Remark 5.4.4. We notice that, if f is (L1 , L2 )Lipschitzian on Q, then (5.4.7) becomes


f (a, c) + f (a, d) + f (b, c) + f (b, d) R d R b f (t, s) g (t, s) dtds


c Ra d R b




4
g (t, s) dtds
c

1
[L1 (b a) + L2 (d c)] . (5.4.8)
2

127

For a bimonotonic non-decreasing integrators, the following result holds:


Theorem 5.4.5. Let f, u : Q R be such that f is (1 , 2 )-(H1 , H2 )Holder type mapping,
where H1 , H2 > 0 and 1 , 2 > 0 are given, and u is bimonotonic non-decreasing on Q. Then
we have the inequality
|T (f, u; Q)|


H2
H1
1
2
(b a) +
(d c)
[u (b, d) u (b, c) u (a, d) + u (a, c)]

2
2


H1
H2
1
2
+
(b a) (d c) +
(b a) (d c)
[u (b, d) u (a, c)] .
2
2

(5.4.9)

Proof. Using the inequality for the RiemannStieltjes integral of continuous integrands and a
bimonotonic non-decreasing integrators, we have
Z d Z b 




f
(a,
c)
+
f
(a,
d)
+
f
(b,
c)
+
f
(b,
d)
|T (f, u; Q)| =
f (t, s) dt ds u (t, s)
4
c
a

Z dZ b
f (a, c) + f (a, d) + f (b, c) + f (b, d)


dt ds u (t, s).

f
(t,
s)


4
c
a

(5.4.10)

As f is of (1 , 2 )-(H1 , H2 )Holder type mapping, then by (5.4.3) we have





f (a, c) + f (a, d) + f (b, c) + f (b, d)

f (t, s)

4
H1
H2

[(t a)1 + (b t)1 ] +


[(s c)2 + (d s)2 ] , (5.4.11)
2
2
for any t [a, b] and s [c, d]. Consequently, by (5.4.10) and then using integration by parts,
we have
|T (f, u; Q)|

Z dZ b
H2
H1
1
1
2
2

[(t a) + (b t) ] +
[(s c) + (d s) ] dt ds u (t, s)
2
2
c
a


H1
H2
1
2
(b a) +
(d c)
=
[u (b, d) u (b, c) u (a, d) + u (a, c)]
2
2
Z dZ b


H1

1
(t a)1 1 (b t)1 1 u (t, s) dtds
2
c
a
Z dZ b


H2
2
(s c)2 1 (d s)2 1 u (t, s) dtds.
(5.4.12)

2
c
a

128

Now, on utilizing the bimonotonicity property of u on Q, we have


Z dZ b
Z dZ b
1 1
(t a)
u (t, s) dtds u (a, c)
(t a)1 1 dtds
c

1
= u (a, c) (b a)1 (d c) ,
1

1 1

(b t)

u (t, s) dtds u (b, d)

(b t)1 1 dtds

1
= u (b, d) (b a)1 (d c) ,
1
b

2 1

(s c)

u (t, s) dtds u (a, c)

(5.4.13)

(5.4.14)

(s a)2 1 dtds

1
= u (a, c) (b a) (d c)2 ,
2

(5.4.15)

and
Z

2 1

(d s)

u (t, s) dtds u (b, d)

(d s)2 1 dtds

1
= u (b, d) (b a) (d c)2 .
2

Substituting (5.4.13)(5.4.16), in (5.4.12) we get


|T (f, u; Q)|


H2
H1
1
2
(b a) +
(d c)
[u (b, d) u (b, c) u (a, d) + u (a, c)]

2
2


H2
H1
1
2
+
(b a) (d c) +
(b a) (d c)
[u (b, d) u (a, c)] .
2
2

(5.4.16)

References
Anastassiou, G. A. 1995. Ostrowski type inequalities. Proceedings of the American
Mathematical Society 123: 37753781.
Anastassiou, G. A. 1997. Multivariate Ostrowski type inequalities. Acta Mathematica
Hungarica 76: 267278.
Anastassiou, G. A. 2002. Univariate Ostrowski inequalities, revisited. Monatshefte Fur
Mathematik 135: 175189.
Anastassiou, G. A. 2007. Multivariate Fink type identity and multivariate Ostrowski,
comparison of means and Gruss type inequalities. Mathematical and Computer
Modelling 46: 351374.
Anastassiou, G. A. & Goldstein, J. A. 2007. Ostrowski type inequalities over euclidean domains.
Atti Accad. Naz. Lincei 18: 305310.
Apostol, T. M. 1974. Mathematical Analysis. New York, Addison-Wesley.
Barnett, N. & Dragomir, S. 2001. An Ostrowski type inequality for double integrals and
applications for cubature formulae. Soochow Journal of Mathematics 27: 110.
Cerone, P. & Dragomir, S. S. 2000. Trapezoidal-type rules from an inequalities point of view.
Handbook of Analytic-Computational Methods in Applied Mathematics, New York,
CRC Press.
Cerone, P. & Dragomir, S. S. 2002. New bounds for the three-point rule involving
the Riemann-Stieltjes integrals. World Science Publishing: Advances in Statistics
Combinatorics and Related Areas, pp. 5362.
Cerone, P., Dragomir, S. S. & Pearce, C. E. M. 2000. A generalised trapezoid inequality for
functions of bounded variation. Turkish Journal of Mathematics 24: 147163.
Clarkson, J. 1933. On double RiemannStieltjes integrals. Bulletin of the Australian
Mathematical Society 39: 929936.
Clarkson, J. & Adams, C. R. 1933. On definitions of bounded variation for functions of two
variables. Bulletin of the Australian Mathematical Society 35: 824854.
Davis, P. J. & Rabinowitz, P. 1976. Methods of Numerical Integration. New York : Academic
Press.
Dedic, L., Matic, M. & Pecaric, J. 2000. On generalizations of Ostrowski inequality via some
Euler-type identities. Mathematical Inequalities & Applications 3: 337353.
Dedic, L., Matic, M. & Pecaric, J. 2001. On dual Euler-Simpson formulae. Bulletin of the
Belgian Mathematical Society 8: 479504.

130

Dedic, L., Matic, M. & Pecaric, J. 2005. On Euler midpoint formulae. Australian & New
Zealand Industrial and Applied Mathematics Journal 46: 417438.
Dragomir, S. 2000. On the Osrowskis inequality for RiemannStieltjes integral and
applications. Korean Journal of Computational & Applied Mathematics 7: 611627.
Dragomir, S. 2007. Sharp bounds for the deviation of a function from the chord generated by its
extremities and applications. RGMIA Res. Rep. Coll. 10: 20 pp.
Dragomir, S. 2008. Approximating real functions which possess nth derivatives of bounded
variation and applications. Computer Methods in Applied Mechanics and Engineering
56: 22682278.
Dragomir, S. 2009. Approximating the RiemannStieltjes integral in terms of generalised
trapezoidal rules. Nonlinear Analysis 71: 6272.
Dragomir, S., Barnett, N. & Cerone, P. 2003. An Ostrowski type inequality for double integrals
in terms of Lp -norms and applications in numerical integration. Revue dAnalyse
Numerique et de Theorie de lApproximation 32: 161169.
Dragomir, S., Cerone, P., Barnett, N. & Roumeliotis, J. 2000. An inequality of the Ostrowski
type for double integrals and applications for cubature formulae. Tamsui Oxford Journal
of Mathematical Sciences 16: 116.
Dragomir, S. S. 1999. On Simpsons quadrature formula for mappings of bounded variation and
applications. Tamkang Journal of Mathematics 30: 5358.
Dragomir, S. S. 2001. On the Ostrowskis integral inequality for mappings with bounded
variation and applications. Mathematical Inequalities & Applications 4: 5966.
Dragomir, S. S. 2002. A companion of Ostrowskis inequality for functions of bounded variation
and applications. RGMIA Res. Rep. Coll. 5: 19 pp.
Dragomir, S. S. & Rassias, T. M. 2002. Ostrowski type inequalities and applications in
numerical integration. Boston: Kluwer Academic.
Fink, A. M. 1992. Bounds of the deviation of a function from its avarages. Czechoslovak
Mathematical Journal 42: 289310.
Franjic, I., Peric, I. & Pecaric, J. 2006. General Euler-Ostrowski formulae and applications to
quadratures. Applied Mathematics and Computation 177: 9298.
Frechet, M. 1910. Extension au cas des integrals multiples dune definition de lintegrale due a`
Stieltjes. Nouvelles Annales de Mathematiques 10: 241256.
Ghorpade, S. & Limaye, B. 2009. A Course in Multivariable Calculus and Analysis. Springer:
New York.
Rb
1

f (x) g (x) dx
Gruss, G. 1935. Uber
das maximum des absoluten betrages von ba
a

131

1
(ba)2

Rb
a

f (x) dx

Rb
a

g (x) dx. Mathematische Zeitschrift 39: 215226.

Guessab, A. & Schmeisser, G. 2002. Sharp integral inequalities of the HermiteHadamard type.
Journal of Approximation Theory 115: 260288.
Hanna, G. 2009. Cubature rules from a generalized Taylor prespective. PH.D. Thesis, Victoria
University .
Hanna, G., Cerone, P. & Roumeliotis, J. 2000. An Ostrowski type inequality in two dimensions
using the three point rule. Tamkang Journal of Mathematics 42: 671689.
Hanna, G., Dragomir, S. & Cerone, P. 2002a. A genral Ostrowski type inequality for double
integrals. Tamkang Journal of Mathematics 33: 100115.
Hanna, G., Dragomir, S. & Cerone, P. 2002b. A taylor like formula for mappings of two
variables defined on a rectangle in the plane. Tamsui Oxford Journal of Mathematical
Sciences 18: 116.
Liu, Z. 2004. Refinement of an inequality of Gruss type for Riemann-Stieltjes integral. Soochow
Journal of Mathematics 30: 483489.
Liu, Z. 2010. A generalization of two sharp inequalities in two independent variables. Computer
Methods in Applied Mechanics and Engineering 59: 28092814.
Liu, Z. 2012. Another generalization of weighted Ostrowski type inequality for mappings
of bounded variation. Computer Methods in Applied Mechanics and Engineering 25:
393397.
Mercer, P. 2008. Hadamards inequality and trapezoid rules for the RiemannStieltjes integral.
Journal of Mathematical Analysis and Applications 344: 921926.
Milovanovic, G. V. & Pecaric, J. E. 1976. On generalizations of the inequality of A. Ostrowski
and related applications. Univ. Beograd. Publ. Elektrotehn. Fak., Ser. Mat. Fiz. 544-576:
155576.
Mitrinovic, D. S., Pecaric, J. E. & Fink, A. M. 1994. Inequalities for functions and their
integrals and derivatives. Dordrecht, Kluwer Academic Publishers.

Ostrowski, A. 1938. Uber


die absolutabweichung einer differentiebaren funktion von ihren
integralmittelwert. Commentarii Mathematici Helvetici 10: 226227.
Pachpatte, B. 2001. An inequality of Ostrowski type in n independent variables. Ser. Math.
Inform 16: 2124.
Pachpatte, B. 2002a. On multidimensional gruss type inequalities. Journal of Inequalities in
Pure and Applied Mathematics 3: Article 27.
Pachpatte, B. G. 2002b. On multivariate Ostrowski type inequalities. Journal of Inequalities in
Pure and Applied Mathematics 3: Article 58.

132

Pecaric, J. & Franjic, I. 2006. Generalization of corrected Simpsons formula. Australian & New
Zealand Industrial and Applied Mathematics Journal 47: 367385.
Pecaric, J. & Varosanec, S. 2000. A note on Simpsons inequality for functions of bounded
variation. Tamkang Journal of Mathematics 3: 239242.
Pecaric, J. & Varosanec, S. 2001. Simpsons formula for functions whose derivatives belong to
Lp spaces. Applied Mathematics Letters 14: 131135.
Pecaric, J. E., Proschan, F. & Tong, Y. L. 1992. Convex functions, partial orderings, and
statistical applications. London, Academic Press.
Stroud, A. H. 1971. Approximate Calculation of Multiple Integrals. New Jersey: Prentice Hall.
Tseng, K. L., Hwang, S. R. & Dragomir, S. S. 2008. Generalizations of weighted Ostrowski
type inequalities for mappings of bounded variation and their applications. Computer
Methods in Applied Mechanics and Engineering 55: 17851793.
Zhongxue, L. 2008. On sharp inequalities of Simpson type and Ostrowski type in two
independent variable. Computer Methods in Applied Mechanics and Engineering 56:
20432047.